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>
30 #include "intel-pt-insn-decoder.h"
31 #include "intel-pt-pkt-decoder.h"
32 #include "intel-pt-decoder.h"
33 #include "intel-pt-log.h"
35 #define INTEL_PT_BLK_SIZE 1024
37 #define BIT63 (((uint64_t)1 << 63))
39 #define INTEL_PT_RETURN 1
41 /* Maximum number of loops with no packets consumed i.e. stuck in a loop */
42 #define INTEL_PT_MAX_LOOPS 10000
45 struct intel_pt_blk *prev;
46 uint64_t ip[INTEL_PT_BLK_SIZE];
49 struct intel_pt_stack {
50 struct intel_pt_blk *blk;
51 struct intel_pt_blk *spare;
55 enum intel_pt_pkt_state {
56 INTEL_PT_STATE_NO_PSB,
58 INTEL_PT_STATE_ERR_RESYNC,
59 INTEL_PT_STATE_IN_SYNC,
62 INTEL_PT_STATE_TIP_PGD,
64 INTEL_PT_STATE_FUP_NO_TIP,
67 static inline bool intel_pt_sample_time(enum intel_pt_pkt_state pkt_state)
70 case INTEL_PT_STATE_NO_PSB:
71 case INTEL_PT_STATE_NO_IP:
72 case INTEL_PT_STATE_ERR_RESYNC:
73 case INTEL_PT_STATE_IN_SYNC:
74 case INTEL_PT_STATE_TNT:
76 case INTEL_PT_STATE_TIP:
77 case INTEL_PT_STATE_TIP_PGD:
78 case INTEL_PT_STATE_FUP:
79 case INTEL_PT_STATE_FUP_NO_TIP:
86 #ifdef INTEL_PT_STRICT
87 #define INTEL_PT_STATE_ERR1 INTEL_PT_STATE_NO_PSB
88 #define INTEL_PT_STATE_ERR2 INTEL_PT_STATE_NO_PSB
89 #define INTEL_PT_STATE_ERR3 INTEL_PT_STATE_NO_PSB
90 #define INTEL_PT_STATE_ERR4 INTEL_PT_STATE_NO_PSB
92 #define INTEL_PT_STATE_ERR1 (decoder->pkt_state)
93 #define INTEL_PT_STATE_ERR2 INTEL_PT_STATE_NO_IP
94 #define INTEL_PT_STATE_ERR3 INTEL_PT_STATE_ERR_RESYNC
95 #define INTEL_PT_STATE_ERR4 INTEL_PT_STATE_IN_SYNC
98 struct intel_pt_decoder {
99 int (*get_trace)(struct intel_pt_buffer *buffer, void *data);
100 int (*walk_insn)(struct intel_pt_insn *intel_pt_insn,
101 uint64_t *insn_cnt_ptr, uint64_t *ip, uint64_t to_ip,
102 uint64_t max_insn_cnt, void *data);
103 bool (*pgd_ip)(uint64_t ip, void *data);
105 struct intel_pt_state state;
106 const unsigned char *buf;
108 bool return_compression;
120 uint64_t tsc_timestamp;
121 uint64_t ref_timestamp;
122 uint64_t sample_timestamp;
124 uint64_t ctc_timestamp;
127 uint64_t cyc_ref_timestamp;
129 uint32_t tsc_ctc_ratio_n;
130 uint32_t tsc_ctc_ratio_d;
131 uint32_t tsc_ctc_mult;
133 uint32_t ctc_rem_mask;
135 struct intel_pt_stack stack;
136 enum intel_pt_pkt_state pkt_state;
137 struct intel_pt_pkt packet;
138 struct intel_pt_pkt tnt;
141 int last_packet_type;
143 unsigned int max_non_turbo_ratio;
144 double max_non_turbo_ratio_fp;
145 double cbr_cyc_to_tsc;
146 double calc_cyc_to_tsc;
147 bool have_calc_cyc_to_tsc;
149 unsigned int insn_bytes;
151 enum intel_pt_period_type period_type;
152 uint64_t tot_insn_cnt;
153 uint64_t period_insn_cnt;
154 uint64_t period_mask;
155 uint64_t period_ticks;
156 uint64_t last_masked_timestamp;
157 bool continuous_period;
159 bool set_fup_tx_flags;
160 unsigned int fup_tx_flags;
161 unsigned int tx_flags;
162 uint64_t timestamp_insn_cnt;
163 uint64_t sample_insn_cnt;
168 const unsigned char *next_buf;
170 unsigned char temp_buf[INTEL_PT_PKT_MAX_SZ];
173 static uint64_t intel_pt_lower_power_of_2(uint64_t x)
177 for (i = 0; x != 1; i++)
183 static void intel_pt_setup_period(struct intel_pt_decoder *decoder)
185 if (decoder->period_type == INTEL_PT_PERIOD_TICKS) {
188 period = intel_pt_lower_power_of_2(decoder->period);
189 decoder->period_mask = ~(period - 1);
190 decoder->period_ticks = period;
194 static uint64_t multdiv(uint64_t t, uint32_t n, uint32_t d)
198 return (t / d) * n + ((t % d) * n) / d;
201 struct intel_pt_decoder *intel_pt_decoder_new(struct intel_pt_params *params)
203 struct intel_pt_decoder *decoder;
205 if (!params->get_trace || !params->walk_insn)
208 decoder = zalloc(sizeof(struct intel_pt_decoder));
212 decoder->get_trace = params->get_trace;
213 decoder->walk_insn = params->walk_insn;
214 decoder->pgd_ip = params->pgd_ip;
215 decoder->data = params->data;
216 decoder->return_compression = params->return_compression;
218 decoder->period = params->period;
219 decoder->period_type = params->period_type;
221 decoder->max_non_turbo_ratio = params->max_non_turbo_ratio;
222 decoder->max_non_turbo_ratio_fp = params->max_non_turbo_ratio;
224 intel_pt_setup_period(decoder);
226 decoder->mtc_shift = params->mtc_period;
227 decoder->ctc_rem_mask = (1 << decoder->mtc_shift) - 1;
229 decoder->tsc_ctc_ratio_n = params->tsc_ctc_ratio_n;
230 decoder->tsc_ctc_ratio_d = params->tsc_ctc_ratio_d;
232 if (!decoder->tsc_ctc_ratio_n)
233 decoder->tsc_ctc_ratio_d = 0;
235 if (decoder->tsc_ctc_ratio_d) {
236 if (!(decoder->tsc_ctc_ratio_n % decoder->tsc_ctc_ratio_d))
237 decoder->tsc_ctc_mult = decoder->tsc_ctc_ratio_n /
238 decoder->tsc_ctc_ratio_d;
241 * Allow for timestamps appearing to backwards because a TSC
242 * packet has slipped past a MTC packet, so allow 2 MTC ticks
245 decoder->tsc_slip = multdiv(2 << decoder->mtc_shift,
246 decoder->tsc_ctc_ratio_n,
247 decoder->tsc_ctc_ratio_d);
249 /* ... or 0x100 paranoia */
250 if (decoder->tsc_slip < 0x100)
251 decoder->tsc_slip = 0x100;
253 intel_pt_log("timestamp: mtc_shift %u\n", decoder->mtc_shift);
254 intel_pt_log("timestamp: tsc_ctc_ratio_n %u\n", decoder->tsc_ctc_ratio_n);
255 intel_pt_log("timestamp: tsc_ctc_ratio_d %u\n", decoder->tsc_ctc_ratio_d);
256 intel_pt_log("timestamp: tsc_ctc_mult %u\n", decoder->tsc_ctc_mult);
257 intel_pt_log("timestamp: tsc_slip %#x\n", decoder->tsc_slip);
262 static void intel_pt_pop_blk(struct intel_pt_stack *stack)
264 struct intel_pt_blk *blk = stack->blk;
266 stack->blk = blk->prev;
273 static uint64_t intel_pt_pop(struct intel_pt_stack *stack)
278 intel_pt_pop_blk(stack);
281 stack->pos = INTEL_PT_BLK_SIZE;
283 return stack->blk->ip[--stack->pos];
286 static int intel_pt_alloc_blk(struct intel_pt_stack *stack)
288 struct intel_pt_blk *blk;
294 blk = malloc(sizeof(struct intel_pt_blk));
299 blk->prev = stack->blk;
305 static int intel_pt_push(struct intel_pt_stack *stack, uint64_t ip)
309 if (!stack->blk || stack->pos == INTEL_PT_BLK_SIZE) {
310 err = intel_pt_alloc_blk(stack);
315 stack->blk->ip[stack->pos++] = ip;
319 static void intel_pt_clear_stack(struct intel_pt_stack *stack)
322 intel_pt_pop_blk(stack);
326 static void intel_pt_free_stack(struct intel_pt_stack *stack)
328 intel_pt_clear_stack(stack);
330 zfree(&stack->spare);
333 void intel_pt_decoder_free(struct intel_pt_decoder *decoder)
335 intel_pt_free_stack(&decoder->stack);
339 static int intel_pt_ext_err(int code)
343 return INTEL_PT_ERR_NOMEM;
345 return INTEL_PT_ERR_INTERN;
347 return INTEL_PT_ERR_BADPKT;
349 return INTEL_PT_ERR_NODATA;
351 return INTEL_PT_ERR_NOINSN;
353 return INTEL_PT_ERR_MISMAT;
355 return INTEL_PT_ERR_OVR;
357 return INTEL_PT_ERR_LOST;
359 return INTEL_PT_ERR_NELOOP;
361 return INTEL_PT_ERR_UNK;
365 static const char *intel_pt_err_msgs[] = {
366 [INTEL_PT_ERR_NOMEM] = "Memory allocation failed",
367 [INTEL_PT_ERR_INTERN] = "Internal error",
368 [INTEL_PT_ERR_BADPKT] = "Bad packet",
369 [INTEL_PT_ERR_NODATA] = "No more data",
370 [INTEL_PT_ERR_NOINSN] = "Failed to get instruction",
371 [INTEL_PT_ERR_MISMAT] = "Trace doesn't match instruction",
372 [INTEL_PT_ERR_OVR] = "Overflow packet",
373 [INTEL_PT_ERR_LOST] = "Lost trace data",
374 [INTEL_PT_ERR_UNK] = "Unknown error!",
375 [INTEL_PT_ERR_NELOOP] = "Never-ending loop",
378 int intel_pt__strerror(int code, char *buf, size_t buflen)
380 if (code < 1 || code >= INTEL_PT_ERR_MAX)
381 code = INTEL_PT_ERR_UNK;
382 strlcpy(buf, intel_pt_err_msgs[code], buflen);
386 static uint64_t intel_pt_calc_ip(const struct intel_pt_pkt *packet,
391 switch (packet->count) {
393 ip = (last_ip & (uint64_t)0xffffffffffff0000ULL) |
397 ip = (last_ip & (uint64_t)0xffffffff00000000ULL) |
401 ip = packet->payload;
402 /* Sign-extend 6-byte ip */
403 if (ip & (uint64_t)0x800000000000ULL)
404 ip |= (uint64_t)0xffff000000000000ULL;
407 ip = (last_ip & (uint64_t)0xffff000000000000ULL) |
411 ip = packet->payload;
420 static inline void intel_pt_set_last_ip(struct intel_pt_decoder *decoder)
422 decoder->last_ip = intel_pt_calc_ip(&decoder->packet, decoder->last_ip);
423 decoder->have_last_ip = true;
426 static inline void intel_pt_set_ip(struct intel_pt_decoder *decoder)
428 intel_pt_set_last_ip(decoder);
429 decoder->ip = decoder->last_ip;
432 static void intel_pt_decoder_log_packet(struct intel_pt_decoder *decoder)
434 intel_pt_log_packet(&decoder->packet, decoder->pkt_len, decoder->pos,
438 static int intel_pt_bug(struct intel_pt_decoder *decoder)
440 intel_pt_log("ERROR: Internal error\n");
441 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
445 static inline void intel_pt_clear_tx_flags(struct intel_pt_decoder *decoder)
447 decoder->tx_flags = 0;
450 static inline void intel_pt_update_in_tx(struct intel_pt_decoder *decoder)
452 decoder->tx_flags = decoder->packet.payload & INTEL_PT_IN_TX;
455 static int intel_pt_bad_packet(struct intel_pt_decoder *decoder)
457 intel_pt_clear_tx_flags(decoder);
458 decoder->have_tma = false;
459 decoder->pkt_len = 1;
460 decoder->pkt_step = 1;
461 intel_pt_decoder_log_packet(decoder);
462 if (decoder->pkt_state != INTEL_PT_STATE_NO_PSB) {
463 intel_pt_log("ERROR: Bad packet\n");
464 decoder->pkt_state = INTEL_PT_STATE_ERR1;
469 static int intel_pt_get_data(struct intel_pt_decoder *decoder)
471 struct intel_pt_buffer buffer = { .buf = 0, };
474 decoder->pkt_step = 0;
476 intel_pt_log("Getting more data\n");
477 ret = decoder->get_trace(&buffer, decoder->data);
480 decoder->buf = buffer.buf;
481 decoder->len = buffer.len;
483 intel_pt_log("No more data\n");
486 if (!buffer.consecutive) {
488 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
489 decoder->ref_timestamp = buffer.ref_timestamp;
490 decoder->timestamp = 0;
491 decoder->have_tma = false;
492 decoder->state.trace_nr = buffer.trace_nr;
493 intel_pt_log("Reference timestamp 0x%" PRIx64 "\n",
494 decoder->ref_timestamp);
501 static int intel_pt_get_next_data(struct intel_pt_decoder *decoder)
503 if (!decoder->next_buf)
504 return intel_pt_get_data(decoder);
506 decoder->buf = decoder->next_buf;
507 decoder->len = decoder->next_len;
508 decoder->next_buf = 0;
509 decoder->next_len = 0;
513 static int intel_pt_get_split_packet(struct intel_pt_decoder *decoder)
515 unsigned char *buf = decoder->temp_buf;
516 size_t old_len, len, n;
519 old_len = decoder->len;
521 memcpy(buf, decoder->buf, len);
523 ret = intel_pt_get_data(decoder);
525 decoder->pos += old_len;
526 return ret < 0 ? ret : -EINVAL;
529 n = INTEL_PT_PKT_MAX_SZ - len;
530 if (n > decoder->len)
532 memcpy(buf + len, decoder->buf, n);
535 ret = intel_pt_get_packet(buf, len, &decoder->packet);
536 if (ret < (int)old_len) {
537 decoder->next_buf = decoder->buf;
538 decoder->next_len = decoder->len;
540 decoder->len = old_len;
541 return intel_pt_bad_packet(decoder);
544 decoder->next_buf = decoder->buf + (ret - old_len);
545 decoder->next_len = decoder->len - (ret - old_len);
553 struct intel_pt_pkt_info {
554 struct intel_pt_decoder *decoder;
555 struct intel_pt_pkt packet;
558 int last_packet_type;
562 typedef int (*intel_pt_pkt_cb_t)(struct intel_pt_pkt_info *pkt_info);
564 /* Lookahead packets in current buffer */
565 static int intel_pt_pkt_lookahead(struct intel_pt_decoder *decoder,
566 intel_pt_pkt_cb_t cb, void *data)
568 struct intel_pt_pkt_info pkt_info;
569 const unsigned char *buf = decoder->buf;
570 size_t len = decoder->len;
573 pkt_info.decoder = decoder;
574 pkt_info.pos = decoder->pos;
575 pkt_info.pkt_len = decoder->pkt_step;
576 pkt_info.last_packet_type = decoder->last_packet_type;
577 pkt_info.data = data;
581 pkt_info.pos += pkt_info.pkt_len;
582 buf += pkt_info.pkt_len;
583 len -= pkt_info.pkt_len;
586 return INTEL_PT_NEED_MORE_BYTES;
588 ret = intel_pt_get_packet(buf, len, &pkt_info.packet);
590 return INTEL_PT_NEED_MORE_BYTES;
594 pkt_info.pkt_len = ret;
595 } while (pkt_info.packet.type == INTEL_PT_PAD);
601 pkt_info.last_packet_type = pkt_info.packet.type;
605 struct intel_pt_calc_cyc_to_tsc_info {
609 uint64_t ctc_timestamp;
611 uint64_t tsc_timestamp;
616 double cbr_cyc_to_tsc;
620 * MTC provides a 8-bit slice of CTC but the TMA packet only provides the lower
621 * 16 bits of CTC. If mtc_shift > 8 then some of the MTC bits are not in the CTC
622 * provided by the TMA packet. Fix-up the last_mtc calculated from the TMA
623 * packet by copying the missing bits from the current MTC assuming the least
624 * difference between the two, and that the current MTC comes after last_mtc.
626 static void intel_pt_fixup_last_mtc(uint32_t mtc, int mtc_shift,
629 uint32_t first_missing_bit = 1U << (16 - mtc_shift);
630 uint32_t mask = ~(first_missing_bit - 1);
632 *last_mtc |= mtc & mask;
633 if (*last_mtc >= mtc) {
634 *last_mtc -= first_missing_bit;
639 static int intel_pt_calc_cyc_cb(struct intel_pt_pkt_info *pkt_info)
641 struct intel_pt_decoder *decoder = pkt_info->decoder;
642 struct intel_pt_calc_cyc_to_tsc_info *data = pkt_info->data;
646 uint32_t mtc, mtc_delta, ctc, fc, ctc_rem;
648 switch (pkt_info->packet.type) {
650 case INTEL_PT_TIP_PGE:
655 case INTEL_PT_MODE_EXEC:
656 case INTEL_PT_MODE_TSX:
657 case INTEL_PT_PSBEND:
667 mtc = pkt_info->packet.payload;
668 if (decoder->mtc_shift > 8 && data->fixup_last_mtc) {
669 data->fixup_last_mtc = false;
670 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
673 if (mtc > data->last_mtc)
674 mtc_delta = mtc - data->last_mtc;
676 mtc_delta = mtc + 256 - data->last_mtc;
677 data->ctc_delta += mtc_delta << decoder->mtc_shift;
678 data->last_mtc = mtc;
680 if (decoder->tsc_ctc_mult) {
681 timestamp = data->ctc_timestamp +
682 data->ctc_delta * decoder->tsc_ctc_mult;
684 timestamp = data->ctc_timestamp +
685 multdiv(data->ctc_delta,
686 decoder->tsc_ctc_ratio_n,
687 decoder->tsc_ctc_ratio_d);
690 if (timestamp < data->timestamp)
693 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
694 data->timestamp = timestamp;
701 timestamp = pkt_info->packet.payload |
702 (data->timestamp & (0xffULL << 56));
703 if (data->from_mtc && timestamp < data->timestamp &&
704 data->timestamp - timestamp < decoder->tsc_slip)
706 if (timestamp < data->timestamp)
707 timestamp += (1ULL << 56);
708 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
711 data->tsc_timestamp = timestamp;
712 data->timestamp = timestamp;
721 if (!decoder->tsc_ctc_ratio_d)
724 ctc = pkt_info->packet.payload;
725 fc = pkt_info->packet.count;
726 ctc_rem = ctc & decoder->ctc_rem_mask;
728 data->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
730 data->ctc_timestamp = data->tsc_timestamp - fc;
731 if (decoder->tsc_ctc_mult) {
732 data->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
734 data->ctc_timestamp -=
735 multdiv(ctc_rem, decoder->tsc_ctc_ratio_n,
736 decoder->tsc_ctc_ratio_d);
740 data->have_tma = true;
741 data->fixup_last_mtc = true;
746 data->cycle_cnt += pkt_info->packet.payload;
750 cbr = pkt_info->packet.payload;
751 if (data->cbr && data->cbr != cbr)
754 data->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
757 case INTEL_PT_TIP_PGD:
758 case INTEL_PT_TRACESTOP:
760 case INTEL_PT_BAD: /* Does not happen */
765 if (!data->cbr && decoder->cbr) {
766 data->cbr = decoder->cbr;
767 data->cbr_cyc_to_tsc = decoder->cbr_cyc_to_tsc;
770 if (!data->cycle_cnt)
773 cyc_to_tsc = (double)(timestamp - decoder->timestamp) / data->cycle_cnt;
775 if (data->cbr && cyc_to_tsc > data->cbr_cyc_to_tsc &&
776 cyc_to_tsc / data->cbr_cyc_to_tsc > 1.25) {
777 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle too big (c.f. CBR-based value %g), pos " x64_fmt "\n",
778 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
782 decoder->calc_cyc_to_tsc = cyc_to_tsc;
783 decoder->have_calc_cyc_to_tsc = true;
786 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. CBR-based value %g, pos " x64_fmt "\n",
787 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
789 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. unknown CBR-based value, pos " x64_fmt "\n",
790 cyc_to_tsc, pkt_info->pos);
796 static void intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder *decoder,
799 struct intel_pt_calc_cyc_to_tsc_info data = {
802 .last_mtc = decoder->last_mtc,
803 .ctc_timestamp = decoder->ctc_timestamp,
804 .ctc_delta = decoder->ctc_delta,
805 .tsc_timestamp = decoder->tsc_timestamp,
806 .timestamp = decoder->timestamp,
807 .have_tma = decoder->have_tma,
808 .fixup_last_mtc = decoder->fixup_last_mtc,
809 .from_mtc = from_mtc,
813 intel_pt_pkt_lookahead(decoder, intel_pt_calc_cyc_cb, &data);
816 static int intel_pt_get_next_packet(struct intel_pt_decoder *decoder)
820 decoder->last_packet_type = decoder->packet.type;
823 decoder->pos += decoder->pkt_step;
824 decoder->buf += decoder->pkt_step;
825 decoder->len -= decoder->pkt_step;
828 ret = intel_pt_get_next_data(decoder);
833 ret = intel_pt_get_packet(decoder->buf, decoder->len,
835 if (ret == INTEL_PT_NEED_MORE_BYTES &&
836 decoder->len < INTEL_PT_PKT_MAX_SZ && !decoder->next_buf) {
837 ret = intel_pt_get_split_packet(decoder);
842 return intel_pt_bad_packet(decoder);
844 decoder->pkt_len = ret;
845 decoder->pkt_step = ret;
846 intel_pt_decoder_log_packet(decoder);
847 } while (decoder->packet.type == INTEL_PT_PAD);
852 static uint64_t intel_pt_next_period(struct intel_pt_decoder *decoder)
854 uint64_t timestamp, masked_timestamp;
856 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
857 masked_timestamp = timestamp & decoder->period_mask;
858 if (decoder->continuous_period) {
859 if (masked_timestamp != decoder->last_masked_timestamp)
863 masked_timestamp = timestamp & decoder->period_mask;
864 if (masked_timestamp != decoder->last_masked_timestamp) {
865 decoder->last_masked_timestamp = masked_timestamp;
866 decoder->continuous_period = true;
869 return decoder->period_ticks - (timestamp - masked_timestamp);
872 static uint64_t intel_pt_next_sample(struct intel_pt_decoder *decoder)
874 switch (decoder->period_type) {
875 case INTEL_PT_PERIOD_INSTRUCTIONS:
876 return decoder->period - decoder->period_insn_cnt;
877 case INTEL_PT_PERIOD_TICKS:
878 return intel_pt_next_period(decoder);
879 case INTEL_PT_PERIOD_NONE:
880 case INTEL_PT_PERIOD_MTC:
886 static void intel_pt_sample_insn(struct intel_pt_decoder *decoder)
888 uint64_t timestamp, masked_timestamp;
890 switch (decoder->period_type) {
891 case INTEL_PT_PERIOD_INSTRUCTIONS:
892 decoder->period_insn_cnt = 0;
894 case INTEL_PT_PERIOD_TICKS:
895 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
896 masked_timestamp = timestamp & decoder->period_mask;
897 decoder->last_masked_timestamp = masked_timestamp;
899 case INTEL_PT_PERIOD_NONE:
900 case INTEL_PT_PERIOD_MTC:
905 decoder->state.type |= INTEL_PT_INSTRUCTION;
908 static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
909 struct intel_pt_insn *intel_pt_insn, uint64_t ip)
911 uint64_t max_insn_cnt, insn_cnt = 0;
914 if (!decoder->mtc_insn)
915 decoder->mtc_insn = true;
917 max_insn_cnt = intel_pt_next_sample(decoder);
919 err = decoder->walk_insn(intel_pt_insn, &insn_cnt, &decoder->ip, ip,
920 max_insn_cnt, decoder->data);
922 decoder->tot_insn_cnt += insn_cnt;
923 decoder->timestamp_insn_cnt += insn_cnt;
924 decoder->sample_insn_cnt += insn_cnt;
925 decoder->period_insn_cnt += insn_cnt;
928 decoder->no_progress = 0;
929 decoder->pkt_state = INTEL_PT_STATE_ERR2;
930 intel_pt_log_at("ERROR: Failed to get instruction",
937 if (ip && decoder->ip == ip) {
942 if (max_insn_cnt && insn_cnt >= max_insn_cnt)
943 intel_pt_sample_insn(decoder);
945 if (intel_pt_insn->branch == INTEL_PT_BR_NO_BRANCH) {
946 decoder->state.type = INTEL_PT_INSTRUCTION;
947 decoder->state.from_ip = decoder->ip;
948 decoder->state.to_ip = 0;
949 decoder->ip += intel_pt_insn->length;
950 err = INTEL_PT_RETURN;
954 if (intel_pt_insn->op == INTEL_PT_OP_CALL) {
955 /* Zero-length calls are excluded */
956 if (intel_pt_insn->branch != INTEL_PT_BR_UNCONDITIONAL ||
957 intel_pt_insn->rel) {
958 err = intel_pt_push(&decoder->stack, decoder->ip +
959 intel_pt_insn->length);
963 } else if (intel_pt_insn->op == INTEL_PT_OP_RET) {
964 decoder->ret_addr = intel_pt_pop(&decoder->stack);
967 if (intel_pt_insn->branch == INTEL_PT_BR_UNCONDITIONAL) {
968 int cnt = decoder->no_progress++;
970 decoder->state.from_ip = decoder->ip;
971 decoder->ip += intel_pt_insn->length +
973 decoder->state.to_ip = decoder->ip;
974 err = INTEL_PT_RETURN;
977 * Check for being stuck in a loop. This can happen if a
978 * decoder error results in the decoder erroneously setting the
979 * ip to an address that is itself in an infinite loop that
980 * consumes no packets. When that happens, there must be an
981 * unconditional branch.
985 decoder->stuck_ip = decoder->state.to_ip;
986 decoder->stuck_ip_prd = 1;
987 decoder->stuck_ip_cnt = 1;
988 } else if (cnt > INTEL_PT_MAX_LOOPS ||
989 decoder->state.to_ip == decoder->stuck_ip) {
990 intel_pt_log_at("ERROR: Never-ending loop",
991 decoder->state.to_ip);
992 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
995 } else if (!--decoder->stuck_ip_cnt) {
996 decoder->stuck_ip_prd += 1;
997 decoder->stuck_ip_cnt = decoder->stuck_ip_prd;
998 decoder->stuck_ip = decoder->state.to_ip;
1001 goto out_no_progress;
1004 decoder->no_progress = 0;
1006 decoder->state.insn_op = intel_pt_insn->op;
1007 decoder->state.insn_len = intel_pt_insn->length;
1009 if (decoder->tx_flags & INTEL_PT_IN_TX)
1010 decoder->state.flags |= INTEL_PT_IN_TX;
1015 static int intel_pt_walk_fup(struct intel_pt_decoder *decoder)
1017 struct intel_pt_insn intel_pt_insn;
1021 ip = decoder->last_ip;
1024 err = intel_pt_walk_insn(decoder, &intel_pt_insn, ip);
1025 if (err == INTEL_PT_RETURN)
1027 if (err == -EAGAIN) {
1028 if (decoder->set_fup_tx_flags) {
1029 decoder->set_fup_tx_flags = false;
1030 decoder->tx_flags = decoder->fup_tx_flags;
1031 decoder->state.type = INTEL_PT_TRANSACTION;
1032 decoder->state.from_ip = decoder->ip;
1033 decoder->state.to_ip = 0;
1034 decoder->state.flags = decoder->fup_tx_flags;
1039 decoder->set_fup_tx_flags = false;
1043 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1044 intel_pt_log_at("ERROR: Unexpected indirect branch",
1046 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1050 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1051 intel_pt_log_at("ERROR: Unexpected conditional branch",
1053 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1057 intel_pt_bug(decoder);
1061 static int intel_pt_walk_tip(struct intel_pt_decoder *decoder)
1063 struct intel_pt_insn intel_pt_insn;
1066 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1067 if (err == INTEL_PT_RETURN &&
1069 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1070 (decoder->state.type & INTEL_PT_BRANCH) &&
1071 decoder->pgd_ip(decoder->state.to_ip, decoder->data)) {
1072 /* Unconditional branch leaving filter region */
1073 decoder->no_progress = 0;
1074 decoder->pge = false;
1075 decoder->continuous_period = false;
1076 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1077 decoder->state.to_ip = 0;
1080 if (err == INTEL_PT_RETURN)
1085 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1086 if (decoder->pkt_state == INTEL_PT_STATE_TIP_PGD) {
1087 decoder->pge = false;
1088 decoder->continuous_period = false;
1089 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1090 decoder->state.from_ip = decoder->ip;
1091 decoder->state.to_ip = 0;
1092 if (decoder->packet.count != 0)
1093 decoder->ip = decoder->last_ip;
1095 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1096 decoder->state.from_ip = decoder->ip;
1097 if (decoder->packet.count == 0) {
1098 decoder->state.to_ip = 0;
1100 decoder->state.to_ip = decoder->last_ip;
1101 decoder->ip = decoder->last_ip;
1107 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1108 uint64_t to_ip = decoder->ip + intel_pt_insn.length +
1111 if (decoder->pgd_ip &&
1112 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1113 decoder->pgd_ip(to_ip, decoder->data)) {
1114 /* Conditional branch leaving filter region */
1115 decoder->pge = false;
1116 decoder->continuous_period = false;
1117 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1118 decoder->ip = to_ip;
1119 decoder->state.from_ip = decoder->ip;
1120 decoder->state.to_ip = 0;
1123 intel_pt_log_at("ERROR: Conditional branch when expecting indirect branch",
1125 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1129 return intel_pt_bug(decoder);
1132 static int intel_pt_walk_tnt(struct intel_pt_decoder *decoder)
1134 struct intel_pt_insn intel_pt_insn;
1138 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1139 if (err == INTEL_PT_RETURN)
1144 if (intel_pt_insn.op == INTEL_PT_OP_RET) {
1145 if (!decoder->return_compression) {
1146 intel_pt_log_at("ERROR: RET when expecting conditional branch",
1148 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1151 if (!decoder->ret_addr) {
1152 intel_pt_log_at("ERROR: Bad RET compression (stack empty)",
1154 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1157 if (!(decoder->tnt.payload & BIT63)) {
1158 intel_pt_log_at("ERROR: Bad RET compression (TNT=N)",
1160 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1163 decoder->tnt.count -= 1;
1164 if (!decoder->tnt.count)
1165 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1166 decoder->tnt.payload <<= 1;
1167 decoder->state.from_ip = decoder->ip;
1168 decoder->ip = decoder->ret_addr;
1169 decoder->state.to_ip = decoder->ip;
1173 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1174 /* Handle deferred TIPs */
1175 err = intel_pt_get_next_packet(decoder);
1178 if (decoder->packet.type != INTEL_PT_TIP ||
1179 decoder->packet.count == 0) {
1180 intel_pt_log_at("ERROR: Missing deferred TIP for indirect branch",
1182 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1183 decoder->pkt_step = 0;
1186 intel_pt_set_last_ip(decoder);
1187 decoder->state.from_ip = decoder->ip;
1188 decoder->state.to_ip = decoder->last_ip;
1189 decoder->ip = decoder->last_ip;
1193 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1194 decoder->tnt.count -= 1;
1195 if (!decoder->tnt.count)
1196 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1197 if (decoder->tnt.payload & BIT63) {
1198 decoder->tnt.payload <<= 1;
1199 decoder->state.from_ip = decoder->ip;
1200 decoder->ip += intel_pt_insn.length +
1202 decoder->state.to_ip = decoder->ip;
1205 /* Instruction sample for a non-taken branch */
1206 if (decoder->state.type & INTEL_PT_INSTRUCTION) {
1207 decoder->tnt.payload <<= 1;
1208 decoder->state.type = INTEL_PT_INSTRUCTION;
1209 decoder->state.from_ip = decoder->ip;
1210 decoder->state.to_ip = 0;
1211 decoder->ip += intel_pt_insn.length;
1214 decoder->ip += intel_pt_insn.length;
1215 if (!decoder->tnt.count)
1217 decoder->tnt.payload <<= 1;
1221 return intel_pt_bug(decoder);
1225 static int intel_pt_mode_tsx(struct intel_pt_decoder *decoder, bool *no_tip)
1227 unsigned int fup_tx_flags;
1230 fup_tx_flags = decoder->packet.payload &
1231 (INTEL_PT_IN_TX | INTEL_PT_ABORT_TX);
1232 err = intel_pt_get_next_packet(decoder);
1235 if (decoder->packet.type == INTEL_PT_FUP) {
1236 decoder->fup_tx_flags = fup_tx_flags;
1237 decoder->set_fup_tx_flags = true;
1238 if (!(decoder->fup_tx_flags & INTEL_PT_ABORT_TX))
1241 intel_pt_log_at("ERROR: Missing FUP after MODE.TSX",
1243 intel_pt_update_in_tx(decoder);
1248 static void intel_pt_calc_tsc_timestamp(struct intel_pt_decoder *decoder)
1252 decoder->have_tma = false;
1254 if (decoder->ref_timestamp) {
1255 timestamp = decoder->packet.payload |
1256 (decoder->ref_timestamp & (0xffULL << 56));
1257 if (timestamp < decoder->ref_timestamp) {
1258 if (decoder->ref_timestamp - timestamp > (1ULL << 55))
1259 timestamp += (1ULL << 56);
1261 if (timestamp - decoder->ref_timestamp > (1ULL << 55))
1262 timestamp -= (1ULL << 56);
1264 decoder->tsc_timestamp = timestamp;
1265 decoder->timestamp = timestamp;
1266 decoder->ref_timestamp = 0;
1267 decoder->timestamp_insn_cnt = 0;
1268 } else if (decoder->timestamp) {
1269 timestamp = decoder->packet.payload |
1270 (decoder->timestamp & (0xffULL << 56));
1271 decoder->tsc_timestamp = timestamp;
1272 if (timestamp < decoder->timestamp &&
1273 decoder->timestamp - timestamp < decoder->tsc_slip) {
1274 intel_pt_log_to("Suppressing backwards timestamp",
1276 timestamp = decoder->timestamp;
1278 if (timestamp < decoder->timestamp) {
1279 intel_pt_log_to("Wraparound timestamp", timestamp);
1280 timestamp += (1ULL << 56);
1281 decoder->tsc_timestamp = timestamp;
1283 decoder->timestamp = timestamp;
1284 decoder->timestamp_insn_cnt = 0;
1287 if (decoder->last_packet_type == INTEL_PT_CYC) {
1288 decoder->cyc_ref_timestamp = decoder->timestamp;
1289 decoder->cycle_cnt = 0;
1290 decoder->have_calc_cyc_to_tsc = false;
1291 intel_pt_calc_cyc_to_tsc(decoder, false);
1294 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1297 static int intel_pt_overflow(struct intel_pt_decoder *decoder)
1299 intel_pt_log("ERROR: Buffer overflow\n");
1300 intel_pt_clear_tx_flags(decoder);
1301 decoder->have_tma = false;
1303 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1304 decoder->overflow = true;
1308 static void intel_pt_calc_tma(struct intel_pt_decoder *decoder)
1310 uint32_t ctc = decoder->packet.payload;
1311 uint32_t fc = decoder->packet.count;
1312 uint32_t ctc_rem = ctc & decoder->ctc_rem_mask;
1314 if (!decoder->tsc_ctc_ratio_d)
1317 decoder->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
1318 decoder->ctc_timestamp = decoder->tsc_timestamp - fc;
1319 if (decoder->tsc_ctc_mult) {
1320 decoder->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
1322 decoder->ctc_timestamp -= multdiv(ctc_rem,
1323 decoder->tsc_ctc_ratio_n,
1324 decoder->tsc_ctc_ratio_d);
1326 decoder->ctc_delta = 0;
1327 decoder->have_tma = true;
1328 decoder->fixup_last_mtc = true;
1329 intel_pt_log("CTC timestamp " x64_fmt " last MTC %#x CTC rem %#x\n",
1330 decoder->ctc_timestamp, decoder->last_mtc, ctc_rem);
1333 static void intel_pt_calc_mtc_timestamp(struct intel_pt_decoder *decoder)
1336 uint32_t mtc, mtc_delta;
1338 if (!decoder->have_tma)
1341 mtc = decoder->packet.payload;
1343 if (decoder->mtc_shift > 8 && decoder->fixup_last_mtc) {
1344 decoder->fixup_last_mtc = false;
1345 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
1346 &decoder->last_mtc);
1349 if (mtc > decoder->last_mtc)
1350 mtc_delta = mtc - decoder->last_mtc;
1352 mtc_delta = mtc + 256 - decoder->last_mtc;
1354 decoder->ctc_delta += mtc_delta << decoder->mtc_shift;
1356 if (decoder->tsc_ctc_mult) {
1357 timestamp = decoder->ctc_timestamp +
1358 decoder->ctc_delta * decoder->tsc_ctc_mult;
1360 timestamp = decoder->ctc_timestamp +
1361 multdiv(decoder->ctc_delta,
1362 decoder->tsc_ctc_ratio_n,
1363 decoder->tsc_ctc_ratio_d);
1366 if (timestamp < decoder->timestamp)
1367 intel_pt_log("Suppressing MTC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1368 timestamp, decoder->timestamp);
1370 decoder->timestamp = timestamp;
1372 decoder->timestamp_insn_cnt = 0;
1373 decoder->last_mtc = mtc;
1375 if (decoder->last_packet_type == INTEL_PT_CYC) {
1376 decoder->cyc_ref_timestamp = decoder->timestamp;
1377 decoder->cycle_cnt = 0;
1378 decoder->have_calc_cyc_to_tsc = false;
1379 intel_pt_calc_cyc_to_tsc(decoder, true);
1383 static void intel_pt_calc_cbr(struct intel_pt_decoder *decoder)
1385 unsigned int cbr = decoder->packet.payload;
1387 if (decoder->cbr == cbr)
1391 decoder->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
1394 static void intel_pt_calc_cyc_timestamp(struct intel_pt_decoder *decoder)
1396 uint64_t timestamp = decoder->cyc_ref_timestamp;
1398 decoder->have_cyc = true;
1400 decoder->cycle_cnt += decoder->packet.payload;
1402 if (!decoder->cyc_ref_timestamp)
1405 if (decoder->have_calc_cyc_to_tsc)
1406 timestamp += decoder->cycle_cnt * decoder->calc_cyc_to_tsc;
1407 else if (decoder->cbr)
1408 timestamp += decoder->cycle_cnt * decoder->cbr_cyc_to_tsc;
1412 if (timestamp < decoder->timestamp)
1413 intel_pt_log("Suppressing CYC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1414 timestamp, decoder->timestamp);
1416 decoder->timestamp = timestamp;
1418 decoder->timestamp_insn_cnt = 0;
1421 /* Walk PSB+ packets when already in sync. */
1422 static int intel_pt_walk_psbend(struct intel_pt_decoder *decoder)
1427 err = intel_pt_get_next_packet(decoder);
1431 switch (decoder->packet.type) {
1432 case INTEL_PT_PSBEND:
1435 case INTEL_PT_TIP_PGD:
1436 case INTEL_PT_TIP_PGE:
1439 case INTEL_PT_TRACESTOP:
1442 decoder->have_tma = false;
1443 intel_pt_log("ERROR: Unexpected packet\n");
1447 return intel_pt_overflow(decoder);
1450 intel_pt_calc_tsc_timestamp(decoder);
1454 intel_pt_calc_tma(decoder);
1458 intel_pt_calc_cbr(decoder);
1461 case INTEL_PT_MODE_EXEC:
1462 decoder->exec_mode = decoder->packet.payload;
1466 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1470 decoder->pge = true;
1471 if (decoder->packet.count)
1472 intel_pt_set_last_ip(decoder);
1475 case INTEL_PT_MODE_TSX:
1476 intel_pt_update_in_tx(decoder);
1480 intel_pt_calc_mtc_timestamp(decoder);
1481 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1482 decoder->state.type |= INTEL_PT_INSTRUCTION;
1495 static int intel_pt_walk_fup_tip(struct intel_pt_decoder *decoder)
1499 if (decoder->tx_flags & INTEL_PT_ABORT_TX) {
1500 decoder->tx_flags = 0;
1501 decoder->state.flags &= ~INTEL_PT_IN_TX;
1502 decoder->state.flags |= INTEL_PT_ABORT_TX;
1504 decoder->state.flags |= INTEL_PT_ASYNC;
1508 err = intel_pt_get_next_packet(decoder);
1512 switch (decoder->packet.type) {
1515 case INTEL_PT_TRACESTOP:
1520 case INTEL_PT_MODE_TSX:
1522 case INTEL_PT_PSBEND:
1523 intel_pt_log("ERROR: Missing TIP after FUP\n");
1524 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1528 return intel_pt_overflow(decoder);
1530 case INTEL_PT_TIP_PGD:
1531 decoder->state.from_ip = decoder->ip;
1532 decoder->state.to_ip = 0;
1533 if (decoder->packet.count != 0) {
1534 intel_pt_set_ip(decoder);
1535 intel_pt_log("Omitting PGD ip " x64_fmt "\n",
1538 decoder->pge = false;
1539 decoder->continuous_period = false;
1542 case INTEL_PT_TIP_PGE:
1543 decoder->pge = true;
1544 intel_pt_log("Omitting PGE ip " x64_fmt "\n",
1546 decoder->state.from_ip = 0;
1547 if (decoder->packet.count == 0) {
1548 decoder->state.to_ip = 0;
1550 intel_pt_set_ip(decoder);
1551 decoder->state.to_ip = decoder->ip;
1556 decoder->state.from_ip = decoder->ip;
1557 if (decoder->packet.count == 0) {
1558 decoder->state.to_ip = 0;
1560 intel_pt_set_ip(decoder);
1561 decoder->state.to_ip = decoder->ip;
1566 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1570 intel_pt_calc_mtc_timestamp(decoder);
1571 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1572 decoder->state.type |= INTEL_PT_INSTRUCTION;
1576 intel_pt_calc_cyc_timestamp(decoder);
1579 case INTEL_PT_MODE_EXEC:
1580 decoder->exec_mode = decoder->packet.payload;
1589 return intel_pt_bug(decoder);
1594 static int intel_pt_walk_trace(struct intel_pt_decoder *decoder)
1596 bool no_tip = false;
1600 err = intel_pt_get_next_packet(decoder);
1604 switch (decoder->packet.type) {
1606 if (!decoder->packet.count)
1608 decoder->tnt = decoder->packet;
1609 decoder->pkt_state = INTEL_PT_STATE_TNT;
1610 err = intel_pt_walk_tnt(decoder);
1615 case INTEL_PT_TIP_PGD:
1616 if (decoder->packet.count != 0)
1617 intel_pt_set_last_ip(decoder);
1618 decoder->pkt_state = INTEL_PT_STATE_TIP_PGD;
1619 return intel_pt_walk_tip(decoder);
1621 case INTEL_PT_TIP_PGE: {
1622 decoder->pge = true;
1623 if (decoder->packet.count == 0) {
1624 intel_pt_log_at("Skipping zero TIP.PGE",
1628 intel_pt_set_ip(decoder);
1629 decoder->state.from_ip = 0;
1630 decoder->state.to_ip = decoder->ip;
1635 return intel_pt_overflow(decoder);
1638 if (decoder->packet.count != 0)
1639 intel_pt_set_last_ip(decoder);
1640 decoder->pkt_state = INTEL_PT_STATE_TIP;
1641 return intel_pt_walk_tip(decoder);
1644 if (decoder->packet.count == 0) {
1645 intel_pt_log_at("Skipping zero FUP",
1650 intel_pt_set_last_ip(decoder);
1651 err = intel_pt_walk_fup(decoder);
1652 if (err != -EAGAIN) {
1656 decoder->pkt_state =
1657 INTEL_PT_STATE_FUP_NO_TIP;
1659 decoder->pkt_state = INTEL_PT_STATE_FUP;
1666 return intel_pt_walk_fup_tip(decoder);
1668 case INTEL_PT_TRACESTOP:
1669 decoder->pge = false;
1670 decoder->continuous_period = false;
1671 intel_pt_clear_tx_flags(decoder);
1672 decoder->have_tma = false;
1676 decoder->last_ip = 0;
1677 decoder->have_last_ip = true;
1678 intel_pt_clear_stack(&decoder->stack);
1679 err = intel_pt_walk_psbend(decoder);
1687 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1691 intel_pt_calc_mtc_timestamp(decoder);
1692 if (decoder->period_type != INTEL_PT_PERIOD_MTC)
1695 * Ensure that there has been an instruction since the
1698 if (!decoder->mtc_insn)
1700 decoder->mtc_insn = false;
1701 /* Ensure that there is a timestamp */
1702 if (!decoder->timestamp)
1704 decoder->state.type = INTEL_PT_INSTRUCTION;
1705 decoder->state.from_ip = decoder->ip;
1706 decoder->state.to_ip = 0;
1707 decoder->mtc_insn = false;
1711 intel_pt_calc_tsc_timestamp(decoder);
1715 intel_pt_calc_tma(decoder);
1719 intel_pt_calc_cyc_timestamp(decoder);
1723 intel_pt_calc_cbr(decoder);
1726 case INTEL_PT_MODE_EXEC:
1727 decoder->exec_mode = decoder->packet.payload;
1730 case INTEL_PT_MODE_TSX:
1731 /* MODE_TSX need not be followed by FUP */
1732 if (!decoder->pge) {
1733 intel_pt_update_in_tx(decoder);
1736 err = intel_pt_mode_tsx(decoder, &no_tip);
1741 case INTEL_PT_BAD: /* Does not happen */
1742 return intel_pt_bug(decoder);
1744 case INTEL_PT_PSBEND:
1751 return intel_pt_bug(decoder);
1756 static inline bool intel_pt_have_ip(struct intel_pt_decoder *decoder)
1758 return decoder->packet.count &&
1759 (decoder->have_last_ip || decoder->packet.count == 3 ||
1760 decoder->packet.count == 6);
1763 /* Walk PSB+ packets to get in sync. */
1764 static int intel_pt_walk_psb(struct intel_pt_decoder *decoder)
1769 err = intel_pt_get_next_packet(decoder);
1773 switch (decoder->packet.type) {
1774 case INTEL_PT_TIP_PGD:
1775 decoder->continuous_period = false;
1777 case INTEL_PT_TIP_PGE:
1779 intel_pt_log("ERROR: Unexpected packet\n");
1783 decoder->pge = true;
1784 if (intel_pt_have_ip(decoder)) {
1785 uint64_t current_ip = decoder->ip;
1787 intel_pt_set_ip(decoder);
1789 intel_pt_log_to("Setting IP",
1795 intel_pt_calc_mtc_timestamp(decoder);
1799 intel_pt_calc_tsc_timestamp(decoder);
1803 intel_pt_calc_tma(decoder);
1807 intel_pt_calc_cyc_timestamp(decoder);
1811 intel_pt_calc_cbr(decoder);
1815 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1818 case INTEL_PT_MODE_EXEC:
1819 decoder->exec_mode = decoder->packet.payload;
1822 case INTEL_PT_MODE_TSX:
1823 intel_pt_update_in_tx(decoder);
1826 case INTEL_PT_TRACESTOP:
1827 decoder->pge = false;
1828 decoder->continuous_period = false;
1829 intel_pt_clear_tx_flags(decoder);
1833 decoder->have_tma = false;
1834 intel_pt_log("ERROR: Unexpected packet\n");
1836 decoder->pkt_state = INTEL_PT_STATE_ERR4;
1838 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1841 case INTEL_PT_BAD: /* Does not happen */
1842 return intel_pt_bug(decoder);
1845 return intel_pt_overflow(decoder);
1847 case INTEL_PT_PSBEND:
1860 static int intel_pt_walk_to_ip(struct intel_pt_decoder *decoder)
1865 err = intel_pt_get_next_packet(decoder);
1869 switch (decoder->packet.type) {
1870 case INTEL_PT_TIP_PGD:
1871 decoder->continuous_period = false;
1873 case INTEL_PT_TIP_PGE:
1875 decoder->pge = decoder->packet.type != INTEL_PT_TIP_PGD;
1876 if (intel_pt_have_ip(decoder))
1877 intel_pt_set_ip(decoder);
1883 if (decoder->overflow) {
1884 if (intel_pt_have_ip(decoder))
1885 intel_pt_set_ip(decoder);
1889 if (decoder->packet.count && decoder->have_last_ip)
1890 intel_pt_set_last_ip(decoder);
1894 intel_pt_calc_mtc_timestamp(decoder);
1898 intel_pt_calc_tsc_timestamp(decoder);
1902 intel_pt_calc_tma(decoder);
1906 intel_pt_calc_cyc_timestamp(decoder);
1910 intel_pt_calc_cbr(decoder);
1914 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1917 case INTEL_PT_MODE_EXEC:
1918 decoder->exec_mode = decoder->packet.payload;
1921 case INTEL_PT_MODE_TSX:
1922 intel_pt_update_in_tx(decoder);
1926 return intel_pt_overflow(decoder);
1928 case INTEL_PT_BAD: /* Does not happen */
1929 return intel_pt_bug(decoder);
1931 case INTEL_PT_TRACESTOP:
1932 decoder->pge = false;
1933 decoder->continuous_period = false;
1934 intel_pt_clear_tx_flags(decoder);
1935 decoder->have_tma = false;
1939 decoder->last_ip = 0;
1940 decoder->have_last_ip = true;
1941 intel_pt_clear_stack(&decoder->stack);
1942 err = intel_pt_walk_psb(decoder);
1946 /* Do not have a sample */
1947 decoder->state.type = 0;
1953 case INTEL_PT_PSBEND:
1963 static int intel_pt_sync_ip(struct intel_pt_decoder *decoder)
1967 intel_pt_log("Scanning for full IP\n");
1968 err = intel_pt_walk_to_ip(decoder);
1972 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1973 decoder->overflow = false;
1975 decoder->state.from_ip = 0;
1976 decoder->state.to_ip = decoder->ip;
1977 intel_pt_log_to("Setting IP", decoder->ip);
1982 static int intel_pt_part_psb(struct intel_pt_decoder *decoder)
1984 const unsigned char *end = decoder->buf + decoder->len;
1987 for (i = INTEL_PT_PSB_LEN - 1; i; i--) {
1988 if (i > decoder->len)
1990 if (!memcmp(end - i, INTEL_PT_PSB_STR, i))
1996 static int intel_pt_rest_psb(struct intel_pt_decoder *decoder, int part_psb)
1998 size_t rest_psb = INTEL_PT_PSB_LEN - part_psb;
1999 const char *psb = INTEL_PT_PSB_STR;
2001 if (rest_psb > decoder->len ||
2002 memcmp(decoder->buf, psb + part_psb, rest_psb))
2008 static int intel_pt_get_split_psb(struct intel_pt_decoder *decoder,
2013 decoder->pos += decoder->len;
2016 ret = intel_pt_get_next_data(decoder);
2020 rest_psb = intel_pt_rest_psb(decoder, part_psb);
2024 decoder->pos -= part_psb;
2025 decoder->next_buf = decoder->buf + rest_psb;
2026 decoder->next_len = decoder->len - rest_psb;
2027 memcpy(decoder->temp_buf, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2028 decoder->buf = decoder->temp_buf;
2029 decoder->len = INTEL_PT_PSB_LEN;
2034 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder)
2036 unsigned char *next;
2039 intel_pt_log("Scanning for PSB\n");
2041 if (!decoder->len) {
2042 ret = intel_pt_get_next_data(decoder);
2047 next = memmem(decoder->buf, decoder->len, INTEL_PT_PSB_STR,
2052 part_psb = intel_pt_part_psb(decoder);
2054 ret = intel_pt_get_split_psb(decoder, part_psb);
2058 decoder->pos += decoder->len;
2064 decoder->pkt_step = next - decoder->buf;
2065 return intel_pt_get_next_packet(decoder);
2069 static int intel_pt_sync(struct intel_pt_decoder *decoder)
2073 decoder->pge = false;
2074 decoder->continuous_period = false;
2075 decoder->have_last_ip = false;
2076 decoder->last_ip = 0;
2078 intel_pt_clear_stack(&decoder->stack);
2080 err = intel_pt_scan_for_psb(decoder);
2084 decoder->have_last_ip = true;
2085 decoder->pkt_state = INTEL_PT_STATE_NO_IP;
2087 err = intel_pt_walk_psb(decoder);
2092 decoder->state.type = 0; /* Do not have a sample */
2093 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2095 return intel_pt_sync_ip(decoder);
2101 static uint64_t intel_pt_est_timestamp(struct intel_pt_decoder *decoder)
2103 uint64_t est = decoder->sample_insn_cnt << 1;
2105 if (!decoder->cbr || !decoder->max_non_turbo_ratio)
2108 est *= decoder->max_non_turbo_ratio;
2109 est /= decoder->cbr;
2111 return decoder->sample_timestamp + est;
2114 const struct intel_pt_state *intel_pt_decode(struct intel_pt_decoder *decoder)
2119 decoder->state.type = INTEL_PT_BRANCH;
2120 decoder->state.flags = 0;
2122 switch (decoder->pkt_state) {
2123 case INTEL_PT_STATE_NO_PSB:
2124 err = intel_pt_sync(decoder);
2126 case INTEL_PT_STATE_NO_IP:
2127 decoder->have_last_ip = false;
2128 decoder->last_ip = 0;
2131 case INTEL_PT_STATE_ERR_RESYNC:
2132 err = intel_pt_sync_ip(decoder);
2134 case INTEL_PT_STATE_IN_SYNC:
2135 err = intel_pt_walk_trace(decoder);
2137 case INTEL_PT_STATE_TNT:
2138 err = intel_pt_walk_tnt(decoder);
2140 err = intel_pt_walk_trace(decoder);
2142 case INTEL_PT_STATE_TIP:
2143 case INTEL_PT_STATE_TIP_PGD:
2144 err = intel_pt_walk_tip(decoder);
2146 case INTEL_PT_STATE_FUP:
2147 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2148 err = intel_pt_walk_fup(decoder);
2150 err = intel_pt_walk_fup_tip(decoder);
2152 decoder->pkt_state = INTEL_PT_STATE_FUP;
2154 case INTEL_PT_STATE_FUP_NO_TIP:
2155 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2156 err = intel_pt_walk_fup(decoder);
2158 err = intel_pt_walk_trace(decoder);
2161 err = intel_pt_bug(decoder);
2164 } while (err == -ENOLINK);
2167 decoder->state.err = intel_pt_ext_err(err);
2168 decoder->state.from_ip = decoder->ip;
2169 decoder->sample_timestamp = decoder->timestamp;
2170 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
2172 decoder->state.err = 0;
2173 if (intel_pt_sample_time(decoder->pkt_state)) {
2174 decoder->sample_timestamp = decoder->timestamp;
2175 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
2179 decoder->state.timestamp = decoder->sample_timestamp;
2180 decoder->state.est_timestamp = intel_pt_est_timestamp(decoder);
2181 decoder->state.cr3 = decoder->cr3;
2182 decoder->state.tot_insn_cnt = decoder->tot_insn_cnt;
2184 return &decoder->state;
2187 static bool intel_pt_at_psb(unsigned char *buf, size_t len)
2189 if (len < INTEL_PT_PSB_LEN)
2191 return memmem(buf, INTEL_PT_PSB_LEN, INTEL_PT_PSB_STR,
2196 * intel_pt_next_psb - move buffer pointer to the start of the next PSB packet.
2197 * @buf: pointer to buffer pointer
2198 * @len: size of buffer
2200 * Updates the buffer pointer to point to the start of the next PSB packet if
2201 * there is one, otherwise the buffer pointer is unchanged. If @buf is updated,
2202 * @len is adjusted accordingly.
2204 * Return: %true if a PSB packet is found, %false otherwise.
2206 static bool intel_pt_next_psb(unsigned char **buf, size_t *len)
2208 unsigned char *next;
2210 next = memmem(*buf, *len, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2212 *len -= next - *buf;
2220 * intel_pt_step_psb - move buffer pointer to the start of the following PSB
2222 * @buf: pointer to buffer pointer
2223 * @len: size of buffer
2225 * Updates the buffer pointer to point to the start of the following PSB packet
2226 * (skipping the PSB at @buf itself) if there is one, otherwise the buffer
2227 * pointer is unchanged. If @buf is updated, @len is adjusted accordingly.
2229 * Return: %true if a PSB packet is found, %false otherwise.
2231 static bool intel_pt_step_psb(unsigned char **buf, size_t *len)
2233 unsigned char *next;
2238 next = memmem(*buf + 1, *len - 1, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2240 *len -= next - *buf;
2248 * intel_pt_last_psb - find the last PSB packet in a buffer.
2250 * @len: size of buffer
2252 * This function finds the last PSB in a buffer.
2254 * Return: A pointer to the last PSB in @buf if found, %NULL otherwise.
2256 static unsigned char *intel_pt_last_psb(unsigned char *buf, size_t len)
2258 const char *n = INTEL_PT_PSB_STR;
2262 if (len < INTEL_PT_PSB_LEN)
2265 k = len - INTEL_PT_PSB_LEN + 1;
2267 p = memrchr(buf, n[0], k);
2270 if (!memcmp(p + 1, n + 1, INTEL_PT_PSB_LEN - 1))
2279 * intel_pt_next_tsc - find and return next TSC.
2281 * @len: size of buffer
2282 * @tsc: TSC value returned
2284 * Find a TSC packet in @buf and return the TSC value. This function assumes
2285 * that @buf starts at a PSB and that PSB+ will contain TSC and so stops if a
2286 * PSBEND packet is found.
2288 * Return: %true if TSC is found, false otherwise.
2290 static bool intel_pt_next_tsc(unsigned char *buf, size_t len, uint64_t *tsc)
2292 struct intel_pt_pkt packet;
2296 ret = intel_pt_get_packet(buf, len, &packet);
2299 if (packet.type == INTEL_PT_TSC) {
2300 *tsc = packet.payload;
2303 if (packet.type == INTEL_PT_PSBEND)
2312 * intel_pt_tsc_cmp - compare 7-byte TSCs.
2313 * @tsc1: first TSC to compare
2314 * @tsc2: second TSC to compare
2316 * This function compares 7-byte TSC values allowing for the possibility that
2317 * TSC wrapped around. Generally it is not possible to know if TSC has wrapped
2318 * around so for that purpose this function assumes the absolute difference is
2319 * less than half the maximum difference.
2321 * Return: %-1 if @tsc1 is before @tsc2, %0 if @tsc1 == @tsc2, %1 if @tsc1 is
2324 static int intel_pt_tsc_cmp(uint64_t tsc1, uint64_t tsc2)
2326 const uint64_t halfway = (1ULL << 55);
2332 if (tsc2 - tsc1 < halfway)
2337 if (tsc1 - tsc2 < halfway)
2345 * intel_pt_find_overlap_tsc - determine start of non-overlapped trace data
2347 * @buf_a: first buffer
2348 * @len_a: size of first buffer
2349 * @buf_b: second buffer
2350 * @len_b: size of second buffer
2352 * If the trace contains TSC we can look at the last TSC of @buf_a and the
2353 * first TSC of @buf_b in order to determine if the buffers overlap, and then
2354 * walk forward in @buf_b until a later TSC is found. A precondition is that
2355 * @buf_a and @buf_b are positioned at a PSB.
2357 * Return: A pointer into @buf_b from where non-overlapped data starts, or
2358 * @buf_b + @len_b if there is no non-overlapped data.
2360 static unsigned char *intel_pt_find_overlap_tsc(unsigned char *buf_a,
2362 unsigned char *buf_b,
2365 uint64_t tsc_a, tsc_b;
2369 p = intel_pt_last_psb(buf_a, len_a);
2371 return buf_b; /* No PSB in buf_a => no overlap */
2373 len = len_a - (p - buf_a);
2374 if (!intel_pt_next_tsc(p, len, &tsc_a)) {
2375 /* The last PSB+ in buf_a is incomplete, so go back one more */
2377 p = intel_pt_last_psb(buf_a, len_a);
2379 return buf_b; /* No full PSB+ => assume no overlap */
2380 len = len_a - (p - buf_a);
2381 if (!intel_pt_next_tsc(p, len, &tsc_a))
2382 return buf_b; /* No TSC in buf_a => assume no overlap */
2386 /* Ignore PSB+ with no TSC */
2387 if (intel_pt_next_tsc(buf_b, len_b, &tsc_b) &&
2388 intel_pt_tsc_cmp(tsc_a, tsc_b) < 0)
2389 return buf_b; /* tsc_a < tsc_b => no overlap */
2391 if (!intel_pt_step_psb(&buf_b, &len_b))
2392 return buf_b + len_b; /* No PSB in buf_b => no data */
2397 * intel_pt_find_overlap - determine start of non-overlapped trace data.
2398 * @buf_a: first buffer
2399 * @len_a: size of first buffer
2400 * @buf_b: second buffer
2401 * @len_b: size of second buffer
2402 * @have_tsc: can use TSC packets to detect overlap
2404 * When trace samples or snapshots are recorded there is the possibility that
2405 * the data overlaps. Note that, for the purposes of decoding, data is only
2406 * useful if it begins with a PSB packet.
2408 * Return: A pointer into @buf_b from where non-overlapped data starts, or
2409 * @buf_b + @len_b if there is no non-overlapped data.
2411 unsigned char *intel_pt_find_overlap(unsigned char *buf_a, size_t len_a,
2412 unsigned char *buf_b, size_t len_b,
2415 unsigned char *found;
2417 /* Buffer 'b' must start at PSB so throw away everything before that */
2418 if (!intel_pt_next_psb(&buf_b, &len_b))
2419 return buf_b + len_b; /* No PSB */
2421 if (!intel_pt_next_psb(&buf_a, &len_a))
2422 return buf_b; /* No overlap */
2425 found = intel_pt_find_overlap_tsc(buf_a, len_a, buf_b, len_b);
2431 * Buffer 'b' cannot end within buffer 'a' so, for comparison purposes,
2432 * we can ignore the first part of buffer 'a'.
2434 while (len_b < len_a) {
2435 if (!intel_pt_step_psb(&buf_a, &len_a))
2436 return buf_b; /* No overlap */
2439 /* Now len_b >= len_a */
2440 if (len_b > len_a) {
2441 /* The leftover buffer 'b' must start at a PSB */
2442 while (!intel_pt_at_psb(buf_b + len_a, len_b - len_a)) {
2443 if (!intel_pt_step_psb(&buf_a, &len_a))
2444 return buf_b; /* No overlap */
2449 /* Potential overlap so check the bytes */
2450 found = memmem(buf_a, len_a, buf_b, len_a);
2452 return buf_b + len_a;
2454 /* Try again at next PSB in buffer 'a' */
2455 if (!intel_pt_step_psb(&buf_a, &len_a))
2456 return buf_b; /* No overlap */
2458 /* The leftover buffer 'b' must start at a PSB */
2459 while (!intel_pt_at_psb(buf_b + len_a, len_b - len_a)) {
2460 if (!intel_pt_step_psb(&buf_a, &len_a))
2461 return buf_b; /* No overlap */