xref: /OK3568_Linux_fs/kernel/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c (revision 4882a59341e53eb6f0b4789bf948001014eff981)
1*4882a593Smuzhiyun // SPDX-License-Identifier: GPL-2.0-only
2*4882a593Smuzhiyun /*
3*4882a593Smuzhiyun  * intel_pt_decoder.c: Intel Processor Trace support
4*4882a593Smuzhiyun  * Copyright (c) 2013-2014, Intel Corporation.
5*4882a593Smuzhiyun  */
6*4882a593Smuzhiyun 
7*4882a593Smuzhiyun #ifndef _GNU_SOURCE
8*4882a593Smuzhiyun #define _GNU_SOURCE
9*4882a593Smuzhiyun #endif
10*4882a593Smuzhiyun #include <stdlib.h>
11*4882a593Smuzhiyun #include <stdbool.h>
12*4882a593Smuzhiyun #include <string.h>
13*4882a593Smuzhiyun #include <errno.h>
14*4882a593Smuzhiyun #include <stdint.h>
15*4882a593Smuzhiyun #include <inttypes.h>
16*4882a593Smuzhiyun #include <linux/compiler.h>
17*4882a593Smuzhiyun #include <linux/string.h>
18*4882a593Smuzhiyun #include <linux/zalloc.h>
19*4882a593Smuzhiyun 
20*4882a593Smuzhiyun #include "../auxtrace.h"
21*4882a593Smuzhiyun 
22*4882a593Smuzhiyun #include "intel-pt-insn-decoder.h"
23*4882a593Smuzhiyun #include "intel-pt-pkt-decoder.h"
24*4882a593Smuzhiyun #include "intel-pt-decoder.h"
25*4882a593Smuzhiyun #include "intel-pt-log.h"
26*4882a593Smuzhiyun 
27*4882a593Smuzhiyun #define BITULL(x) (1ULL << (x))
28*4882a593Smuzhiyun 
29*4882a593Smuzhiyun /* IA32_RTIT_CTL MSR bits */
30*4882a593Smuzhiyun #define INTEL_PT_CYC_ENABLE		BITULL(1)
31*4882a593Smuzhiyun #define INTEL_PT_CYC_THRESHOLD		(BITULL(22) | BITULL(21) | BITULL(20) | BITULL(19))
32*4882a593Smuzhiyun #define INTEL_PT_CYC_THRESHOLD_SHIFT	19
33*4882a593Smuzhiyun 
34*4882a593Smuzhiyun #define INTEL_PT_BLK_SIZE 1024
35*4882a593Smuzhiyun 
36*4882a593Smuzhiyun #define BIT63 (((uint64_t)1 << 63))
37*4882a593Smuzhiyun 
38*4882a593Smuzhiyun #define INTEL_PT_RETURN 1
39*4882a593Smuzhiyun 
40*4882a593Smuzhiyun /* Maximum number of loops with no packets consumed i.e. stuck in a loop */
41*4882a593Smuzhiyun #define INTEL_PT_MAX_LOOPS 10000
42*4882a593Smuzhiyun 
43*4882a593Smuzhiyun struct intel_pt_blk {
44*4882a593Smuzhiyun 	struct intel_pt_blk *prev;
45*4882a593Smuzhiyun 	uint64_t ip[INTEL_PT_BLK_SIZE];
46*4882a593Smuzhiyun };
47*4882a593Smuzhiyun 
48*4882a593Smuzhiyun struct intel_pt_stack {
49*4882a593Smuzhiyun 	struct intel_pt_blk *blk;
50*4882a593Smuzhiyun 	struct intel_pt_blk *spare;
51*4882a593Smuzhiyun 	int pos;
52*4882a593Smuzhiyun };
53*4882a593Smuzhiyun 
54*4882a593Smuzhiyun enum intel_pt_pkt_state {
55*4882a593Smuzhiyun 	INTEL_PT_STATE_NO_PSB,
56*4882a593Smuzhiyun 	INTEL_PT_STATE_NO_IP,
57*4882a593Smuzhiyun 	INTEL_PT_STATE_ERR_RESYNC,
58*4882a593Smuzhiyun 	INTEL_PT_STATE_IN_SYNC,
59*4882a593Smuzhiyun 	INTEL_PT_STATE_TNT_CONT,
60*4882a593Smuzhiyun 	INTEL_PT_STATE_TNT,
61*4882a593Smuzhiyun 	INTEL_PT_STATE_TIP,
62*4882a593Smuzhiyun 	INTEL_PT_STATE_TIP_PGD,
63*4882a593Smuzhiyun 	INTEL_PT_STATE_FUP,
64*4882a593Smuzhiyun 	INTEL_PT_STATE_FUP_NO_TIP,
65*4882a593Smuzhiyun 	INTEL_PT_STATE_RESAMPLE,
66*4882a593Smuzhiyun };
67*4882a593Smuzhiyun 
intel_pt_sample_time(enum intel_pt_pkt_state pkt_state)68*4882a593Smuzhiyun static inline bool intel_pt_sample_time(enum intel_pt_pkt_state pkt_state)
69*4882a593Smuzhiyun {
70*4882a593Smuzhiyun 	switch (pkt_state) {
71*4882a593Smuzhiyun 	case INTEL_PT_STATE_NO_PSB:
72*4882a593Smuzhiyun 	case INTEL_PT_STATE_NO_IP:
73*4882a593Smuzhiyun 	case INTEL_PT_STATE_ERR_RESYNC:
74*4882a593Smuzhiyun 	case INTEL_PT_STATE_IN_SYNC:
75*4882a593Smuzhiyun 	case INTEL_PT_STATE_TNT_CONT:
76*4882a593Smuzhiyun 	case INTEL_PT_STATE_RESAMPLE:
77*4882a593Smuzhiyun 		return true;
78*4882a593Smuzhiyun 	case INTEL_PT_STATE_TNT:
79*4882a593Smuzhiyun 	case INTEL_PT_STATE_TIP:
80*4882a593Smuzhiyun 	case INTEL_PT_STATE_TIP_PGD:
81*4882a593Smuzhiyun 	case INTEL_PT_STATE_FUP:
82*4882a593Smuzhiyun 	case INTEL_PT_STATE_FUP_NO_TIP:
83*4882a593Smuzhiyun 		return false;
84*4882a593Smuzhiyun 	default:
85*4882a593Smuzhiyun 		return true;
86*4882a593Smuzhiyun 	};
87*4882a593Smuzhiyun }
88*4882a593Smuzhiyun 
89*4882a593Smuzhiyun #ifdef INTEL_PT_STRICT
90*4882a593Smuzhiyun #define INTEL_PT_STATE_ERR1	INTEL_PT_STATE_NO_PSB
91*4882a593Smuzhiyun #define INTEL_PT_STATE_ERR2	INTEL_PT_STATE_NO_PSB
92*4882a593Smuzhiyun #define INTEL_PT_STATE_ERR3	INTEL_PT_STATE_NO_PSB
93*4882a593Smuzhiyun #define INTEL_PT_STATE_ERR4	INTEL_PT_STATE_NO_PSB
94*4882a593Smuzhiyun #else
95*4882a593Smuzhiyun #define INTEL_PT_STATE_ERR1	(decoder->pkt_state)
96*4882a593Smuzhiyun #define INTEL_PT_STATE_ERR2	INTEL_PT_STATE_NO_IP
97*4882a593Smuzhiyun #define INTEL_PT_STATE_ERR3	INTEL_PT_STATE_ERR_RESYNC
98*4882a593Smuzhiyun #define INTEL_PT_STATE_ERR4	INTEL_PT_STATE_IN_SYNC
99*4882a593Smuzhiyun #endif
100*4882a593Smuzhiyun 
101*4882a593Smuzhiyun struct intel_pt_decoder {
102*4882a593Smuzhiyun 	int (*get_trace)(struct intel_pt_buffer *buffer, void *data);
103*4882a593Smuzhiyun 	int (*walk_insn)(struct intel_pt_insn *intel_pt_insn,
104*4882a593Smuzhiyun 			 uint64_t *insn_cnt_ptr, uint64_t *ip, uint64_t to_ip,
105*4882a593Smuzhiyun 			 uint64_t max_insn_cnt, void *data);
106*4882a593Smuzhiyun 	bool (*pgd_ip)(uint64_t ip, void *data);
107*4882a593Smuzhiyun 	int (*lookahead)(void *data, intel_pt_lookahead_cb_t cb, void *cb_data);
108*4882a593Smuzhiyun 	void *data;
109*4882a593Smuzhiyun 	struct intel_pt_state state;
110*4882a593Smuzhiyun 	const unsigned char *buf;
111*4882a593Smuzhiyun 	size_t len;
112*4882a593Smuzhiyun 	bool return_compression;
113*4882a593Smuzhiyun 	bool branch_enable;
114*4882a593Smuzhiyun 	bool mtc_insn;
115*4882a593Smuzhiyun 	bool pge;
116*4882a593Smuzhiyun 	bool have_tma;
117*4882a593Smuzhiyun 	bool have_cyc;
118*4882a593Smuzhiyun 	bool fixup_last_mtc;
119*4882a593Smuzhiyun 	bool have_last_ip;
120*4882a593Smuzhiyun 	bool in_psb;
121*4882a593Smuzhiyun 	bool hop;
122*4882a593Smuzhiyun 	bool hop_psb_fup;
123*4882a593Smuzhiyun 	bool leap;
124*4882a593Smuzhiyun 	enum intel_pt_param_flags flags;
125*4882a593Smuzhiyun 	uint64_t pos;
126*4882a593Smuzhiyun 	uint64_t last_ip;
127*4882a593Smuzhiyun 	uint64_t ip;
128*4882a593Smuzhiyun 	uint64_t cr3;
129*4882a593Smuzhiyun 	uint64_t timestamp;
130*4882a593Smuzhiyun 	uint64_t tsc_timestamp;
131*4882a593Smuzhiyun 	uint64_t ref_timestamp;
132*4882a593Smuzhiyun 	uint64_t buf_timestamp;
133*4882a593Smuzhiyun 	uint64_t sample_timestamp;
134*4882a593Smuzhiyun 	uint64_t ret_addr;
135*4882a593Smuzhiyun 	uint64_t ctc_timestamp;
136*4882a593Smuzhiyun 	uint64_t ctc_delta;
137*4882a593Smuzhiyun 	uint64_t cycle_cnt;
138*4882a593Smuzhiyun 	uint64_t cyc_ref_timestamp;
139*4882a593Smuzhiyun 	uint32_t last_mtc;
140*4882a593Smuzhiyun 	uint32_t tsc_ctc_ratio_n;
141*4882a593Smuzhiyun 	uint32_t tsc_ctc_ratio_d;
142*4882a593Smuzhiyun 	uint32_t tsc_ctc_mult;
143*4882a593Smuzhiyun 	uint32_t tsc_slip;
144*4882a593Smuzhiyun 	uint32_t ctc_rem_mask;
145*4882a593Smuzhiyun 	int mtc_shift;
146*4882a593Smuzhiyun 	struct intel_pt_stack stack;
147*4882a593Smuzhiyun 	enum intel_pt_pkt_state pkt_state;
148*4882a593Smuzhiyun 	enum intel_pt_pkt_ctx pkt_ctx;
149*4882a593Smuzhiyun 	enum intel_pt_pkt_ctx prev_pkt_ctx;
150*4882a593Smuzhiyun 	enum intel_pt_blk_type blk_type;
151*4882a593Smuzhiyun 	int blk_type_pos;
152*4882a593Smuzhiyun 	struct intel_pt_pkt packet;
153*4882a593Smuzhiyun 	struct intel_pt_pkt tnt;
154*4882a593Smuzhiyun 	int pkt_step;
155*4882a593Smuzhiyun 	int pkt_len;
156*4882a593Smuzhiyun 	int last_packet_type;
157*4882a593Smuzhiyun 	unsigned int cbr;
158*4882a593Smuzhiyun 	unsigned int cbr_seen;
159*4882a593Smuzhiyun 	unsigned int max_non_turbo_ratio;
160*4882a593Smuzhiyun 	double max_non_turbo_ratio_fp;
161*4882a593Smuzhiyun 	double cbr_cyc_to_tsc;
162*4882a593Smuzhiyun 	double calc_cyc_to_tsc;
163*4882a593Smuzhiyun 	bool have_calc_cyc_to_tsc;
164*4882a593Smuzhiyun 	int exec_mode;
165*4882a593Smuzhiyun 	unsigned int insn_bytes;
166*4882a593Smuzhiyun 	uint64_t period;
167*4882a593Smuzhiyun 	enum intel_pt_period_type period_type;
168*4882a593Smuzhiyun 	uint64_t tot_insn_cnt;
169*4882a593Smuzhiyun 	uint64_t period_insn_cnt;
170*4882a593Smuzhiyun 	uint64_t period_mask;
171*4882a593Smuzhiyun 	uint64_t period_ticks;
172*4882a593Smuzhiyun 	uint64_t last_masked_timestamp;
173*4882a593Smuzhiyun 	uint64_t tot_cyc_cnt;
174*4882a593Smuzhiyun 	uint64_t sample_tot_cyc_cnt;
175*4882a593Smuzhiyun 	uint64_t base_cyc_cnt;
176*4882a593Smuzhiyun 	uint64_t cyc_cnt_timestamp;
177*4882a593Smuzhiyun 	uint64_t ctl;
178*4882a593Smuzhiyun 	uint64_t cyc_threshold;
179*4882a593Smuzhiyun 	double tsc_to_cyc;
180*4882a593Smuzhiyun 	bool continuous_period;
181*4882a593Smuzhiyun 	bool overflow;
182*4882a593Smuzhiyun 	bool set_fup_tx_flags;
183*4882a593Smuzhiyun 	bool set_fup_ptw;
184*4882a593Smuzhiyun 	bool set_fup_mwait;
185*4882a593Smuzhiyun 	bool set_fup_pwre;
186*4882a593Smuzhiyun 	bool set_fup_exstop;
187*4882a593Smuzhiyun 	bool set_fup_bep;
188*4882a593Smuzhiyun 	bool sample_cyc;
189*4882a593Smuzhiyun 	unsigned int fup_tx_flags;
190*4882a593Smuzhiyun 	unsigned int tx_flags;
191*4882a593Smuzhiyun 	uint64_t fup_ptw_payload;
192*4882a593Smuzhiyun 	uint64_t fup_mwait_payload;
193*4882a593Smuzhiyun 	uint64_t fup_pwre_payload;
194*4882a593Smuzhiyun 	uint64_t cbr_payload;
195*4882a593Smuzhiyun 	uint64_t timestamp_insn_cnt;
196*4882a593Smuzhiyun 	uint64_t sample_insn_cnt;
197*4882a593Smuzhiyun 	uint64_t stuck_ip;
198*4882a593Smuzhiyun 	int no_progress;
199*4882a593Smuzhiyun 	int stuck_ip_prd;
200*4882a593Smuzhiyun 	int stuck_ip_cnt;
201*4882a593Smuzhiyun 	const unsigned char *next_buf;
202*4882a593Smuzhiyun 	size_t next_len;
203*4882a593Smuzhiyun 	unsigned char temp_buf[INTEL_PT_PKT_MAX_SZ];
204*4882a593Smuzhiyun };
205*4882a593Smuzhiyun 
intel_pt_lower_power_of_2(uint64_t x)206*4882a593Smuzhiyun static uint64_t intel_pt_lower_power_of_2(uint64_t x)
207*4882a593Smuzhiyun {
208*4882a593Smuzhiyun 	int i;
209*4882a593Smuzhiyun 
210*4882a593Smuzhiyun 	for (i = 0; x != 1; i++)
211*4882a593Smuzhiyun 		x >>= 1;
212*4882a593Smuzhiyun 
213*4882a593Smuzhiyun 	return x << i;
214*4882a593Smuzhiyun }
215*4882a593Smuzhiyun 
intel_pt_cyc_threshold(uint64_t ctl)216*4882a593Smuzhiyun static uint64_t intel_pt_cyc_threshold(uint64_t ctl)
217*4882a593Smuzhiyun {
218*4882a593Smuzhiyun 	if (!(ctl & INTEL_PT_CYC_ENABLE))
219*4882a593Smuzhiyun 		return 0;
220*4882a593Smuzhiyun 
221*4882a593Smuzhiyun 	return (ctl & INTEL_PT_CYC_THRESHOLD) >> INTEL_PT_CYC_THRESHOLD_SHIFT;
222*4882a593Smuzhiyun }
223*4882a593Smuzhiyun 
intel_pt_setup_period(struct intel_pt_decoder * decoder)224*4882a593Smuzhiyun static void intel_pt_setup_period(struct intel_pt_decoder *decoder)
225*4882a593Smuzhiyun {
226*4882a593Smuzhiyun 	if (decoder->period_type == INTEL_PT_PERIOD_TICKS) {
227*4882a593Smuzhiyun 		uint64_t period;
228*4882a593Smuzhiyun 
229*4882a593Smuzhiyun 		period = intel_pt_lower_power_of_2(decoder->period);
230*4882a593Smuzhiyun 		decoder->period_mask  = ~(period - 1);
231*4882a593Smuzhiyun 		decoder->period_ticks = period;
232*4882a593Smuzhiyun 	}
233*4882a593Smuzhiyun }
234*4882a593Smuzhiyun 
multdiv(uint64_t t,uint32_t n,uint32_t d)235*4882a593Smuzhiyun static uint64_t multdiv(uint64_t t, uint32_t n, uint32_t d)
236*4882a593Smuzhiyun {
237*4882a593Smuzhiyun 	if (!d)
238*4882a593Smuzhiyun 		return 0;
239*4882a593Smuzhiyun 	return (t / d) * n + ((t % d) * n) / d;
240*4882a593Smuzhiyun }
241*4882a593Smuzhiyun 
intel_pt_decoder_new(struct intel_pt_params * params)242*4882a593Smuzhiyun struct intel_pt_decoder *intel_pt_decoder_new(struct intel_pt_params *params)
243*4882a593Smuzhiyun {
244*4882a593Smuzhiyun 	struct intel_pt_decoder *decoder;
245*4882a593Smuzhiyun 
246*4882a593Smuzhiyun 	if (!params->get_trace || !params->walk_insn)
247*4882a593Smuzhiyun 		return NULL;
248*4882a593Smuzhiyun 
249*4882a593Smuzhiyun 	decoder = zalloc(sizeof(struct intel_pt_decoder));
250*4882a593Smuzhiyun 	if (!decoder)
251*4882a593Smuzhiyun 		return NULL;
252*4882a593Smuzhiyun 
253*4882a593Smuzhiyun 	decoder->get_trace          = params->get_trace;
254*4882a593Smuzhiyun 	decoder->walk_insn          = params->walk_insn;
255*4882a593Smuzhiyun 	decoder->pgd_ip             = params->pgd_ip;
256*4882a593Smuzhiyun 	decoder->lookahead          = params->lookahead;
257*4882a593Smuzhiyun 	decoder->data               = params->data;
258*4882a593Smuzhiyun 	decoder->return_compression = params->return_compression;
259*4882a593Smuzhiyun 	decoder->branch_enable      = params->branch_enable;
260*4882a593Smuzhiyun 	decoder->hop                = params->quick >= 1;
261*4882a593Smuzhiyun 	decoder->leap               = params->quick >= 2;
262*4882a593Smuzhiyun 
263*4882a593Smuzhiyun 	decoder->flags              = params->flags;
264*4882a593Smuzhiyun 
265*4882a593Smuzhiyun 	decoder->ctl                = params->ctl;
266*4882a593Smuzhiyun 	decoder->period             = params->period;
267*4882a593Smuzhiyun 	decoder->period_type        = params->period_type;
268*4882a593Smuzhiyun 
269*4882a593Smuzhiyun 	decoder->max_non_turbo_ratio    = params->max_non_turbo_ratio;
270*4882a593Smuzhiyun 	decoder->max_non_turbo_ratio_fp = params->max_non_turbo_ratio;
271*4882a593Smuzhiyun 
272*4882a593Smuzhiyun 	decoder->cyc_threshold = intel_pt_cyc_threshold(decoder->ctl);
273*4882a593Smuzhiyun 
274*4882a593Smuzhiyun 	intel_pt_setup_period(decoder);
275*4882a593Smuzhiyun 
276*4882a593Smuzhiyun 	decoder->mtc_shift = params->mtc_period;
277*4882a593Smuzhiyun 	decoder->ctc_rem_mask = (1 << decoder->mtc_shift) - 1;
278*4882a593Smuzhiyun 
279*4882a593Smuzhiyun 	decoder->tsc_ctc_ratio_n = params->tsc_ctc_ratio_n;
280*4882a593Smuzhiyun 	decoder->tsc_ctc_ratio_d = params->tsc_ctc_ratio_d;
281*4882a593Smuzhiyun 
282*4882a593Smuzhiyun 	if (!decoder->tsc_ctc_ratio_n)
283*4882a593Smuzhiyun 		decoder->tsc_ctc_ratio_d = 0;
284*4882a593Smuzhiyun 
285*4882a593Smuzhiyun 	if (decoder->tsc_ctc_ratio_d) {
286*4882a593Smuzhiyun 		if (!(decoder->tsc_ctc_ratio_n % decoder->tsc_ctc_ratio_d))
287*4882a593Smuzhiyun 			decoder->tsc_ctc_mult = decoder->tsc_ctc_ratio_n /
288*4882a593Smuzhiyun 						decoder->tsc_ctc_ratio_d;
289*4882a593Smuzhiyun 	}
290*4882a593Smuzhiyun 
291*4882a593Smuzhiyun 	/*
292*4882a593Smuzhiyun 	 * A TSC packet can slip past MTC packets so that the timestamp appears
293*4882a593Smuzhiyun 	 * to go backwards. One estimate is that can be up to about 40 CPU
294*4882a593Smuzhiyun 	 * cycles, which is certainly less than 0x1000 TSC ticks, but accept
295*4882a593Smuzhiyun 	 * slippage an order of magnitude more to be on the safe side.
296*4882a593Smuzhiyun 	 */
297*4882a593Smuzhiyun 	decoder->tsc_slip = 0x10000;
298*4882a593Smuzhiyun 
299*4882a593Smuzhiyun 	intel_pt_log("timestamp: mtc_shift %u\n", decoder->mtc_shift);
300*4882a593Smuzhiyun 	intel_pt_log("timestamp: tsc_ctc_ratio_n %u\n", decoder->tsc_ctc_ratio_n);
301*4882a593Smuzhiyun 	intel_pt_log("timestamp: tsc_ctc_ratio_d %u\n", decoder->tsc_ctc_ratio_d);
302*4882a593Smuzhiyun 	intel_pt_log("timestamp: tsc_ctc_mult %u\n", decoder->tsc_ctc_mult);
303*4882a593Smuzhiyun 	intel_pt_log("timestamp: tsc_slip %#x\n", decoder->tsc_slip);
304*4882a593Smuzhiyun 
305*4882a593Smuzhiyun 	if (decoder->hop)
306*4882a593Smuzhiyun 		intel_pt_log("Hop mode: decoding FUP and TIPs, but not TNT\n");
307*4882a593Smuzhiyun 
308*4882a593Smuzhiyun 	return decoder;
309*4882a593Smuzhiyun }
310*4882a593Smuzhiyun 
intel_pt_pop_blk(struct intel_pt_stack * stack)311*4882a593Smuzhiyun static void intel_pt_pop_blk(struct intel_pt_stack *stack)
312*4882a593Smuzhiyun {
313*4882a593Smuzhiyun 	struct intel_pt_blk *blk = stack->blk;
314*4882a593Smuzhiyun 
315*4882a593Smuzhiyun 	stack->blk = blk->prev;
316*4882a593Smuzhiyun 	if (!stack->spare)
317*4882a593Smuzhiyun 		stack->spare = blk;
318*4882a593Smuzhiyun 	else
319*4882a593Smuzhiyun 		free(blk);
320*4882a593Smuzhiyun }
321*4882a593Smuzhiyun 
intel_pt_pop(struct intel_pt_stack * stack)322*4882a593Smuzhiyun static uint64_t intel_pt_pop(struct intel_pt_stack *stack)
323*4882a593Smuzhiyun {
324*4882a593Smuzhiyun 	if (!stack->pos) {
325*4882a593Smuzhiyun 		if (!stack->blk)
326*4882a593Smuzhiyun 			return 0;
327*4882a593Smuzhiyun 		intel_pt_pop_blk(stack);
328*4882a593Smuzhiyun 		if (!stack->blk)
329*4882a593Smuzhiyun 			return 0;
330*4882a593Smuzhiyun 		stack->pos = INTEL_PT_BLK_SIZE;
331*4882a593Smuzhiyun 	}
332*4882a593Smuzhiyun 	return stack->blk->ip[--stack->pos];
333*4882a593Smuzhiyun }
334*4882a593Smuzhiyun 
intel_pt_alloc_blk(struct intel_pt_stack * stack)335*4882a593Smuzhiyun static int intel_pt_alloc_blk(struct intel_pt_stack *stack)
336*4882a593Smuzhiyun {
337*4882a593Smuzhiyun 	struct intel_pt_blk *blk;
338*4882a593Smuzhiyun 
339*4882a593Smuzhiyun 	if (stack->spare) {
340*4882a593Smuzhiyun 		blk = stack->spare;
341*4882a593Smuzhiyun 		stack->spare = NULL;
342*4882a593Smuzhiyun 	} else {
343*4882a593Smuzhiyun 		blk = malloc(sizeof(struct intel_pt_blk));
344*4882a593Smuzhiyun 		if (!blk)
345*4882a593Smuzhiyun 			return -ENOMEM;
346*4882a593Smuzhiyun 	}
347*4882a593Smuzhiyun 
348*4882a593Smuzhiyun 	blk->prev = stack->blk;
349*4882a593Smuzhiyun 	stack->blk = blk;
350*4882a593Smuzhiyun 	stack->pos = 0;
351*4882a593Smuzhiyun 	return 0;
352*4882a593Smuzhiyun }
353*4882a593Smuzhiyun 
intel_pt_push(struct intel_pt_stack * stack,uint64_t ip)354*4882a593Smuzhiyun static int intel_pt_push(struct intel_pt_stack *stack, uint64_t ip)
355*4882a593Smuzhiyun {
356*4882a593Smuzhiyun 	int err;
357*4882a593Smuzhiyun 
358*4882a593Smuzhiyun 	if (!stack->blk || stack->pos == INTEL_PT_BLK_SIZE) {
359*4882a593Smuzhiyun 		err = intel_pt_alloc_blk(stack);
360*4882a593Smuzhiyun 		if (err)
361*4882a593Smuzhiyun 			return err;
362*4882a593Smuzhiyun 	}
363*4882a593Smuzhiyun 
364*4882a593Smuzhiyun 	stack->blk->ip[stack->pos++] = ip;
365*4882a593Smuzhiyun 	return 0;
366*4882a593Smuzhiyun }
367*4882a593Smuzhiyun 
intel_pt_clear_stack(struct intel_pt_stack * stack)368*4882a593Smuzhiyun static void intel_pt_clear_stack(struct intel_pt_stack *stack)
369*4882a593Smuzhiyun {
370*4882a593Smuzhiyun 	while (stack->blk)
371*4882a593Smuzhiyun 		intel_pt_pop_blk(stack);
372*4882a593Smuzhiyun 	stack->pos = 0;
373*4882a593Smuzhiyun }
374*4882a593Smuzhiyun 
intel_pt_free_stack(struct intel_pt_stack * stack)375*4882a593Smuzhiyun static void intel_pt_free_stack(struct intel_pt_stack *stack)
376*4882a593Smuzhiyun {
377*4882a593Smuzhiyun 	intel_pt_clear_stack(stack);
378*4882a593Smuzhiyun 	zfree(&stack->blk);
379*4882a593Smuzhiyun 	zfree(&stack->spare);
380*4882a593Smuzhiyun }
381*4882a593Smuzhiyun 
intel_pt_decoder_free(struct intel_pt_decoder * decoder)382*4882a593Smuzhiyun void intel_pt_decoder_free(struct intel_pt_decoder *decoder)
383*4882a593Smuzhiyun {
384*4882a593Smuzhiyun 	intel_pt_free_stack(&decoder->stack);
385*4882a593Smuzhiyun 	free(decoder);
386*4882a593Smuzhiyun }
387*4882a593Smuzhiyun 
intel_pt_ext_err(int code)388*4882a593Smuzhiyun static int intel_pt_ext_err(int code)
389*4882a593Smuzhiyun {
390*4882a593Smuzhiyun 	switch (code) {
391*4882a593Smuzhiyun 	case -ENOMEM:
392*4882a593Smuzhiyun 		return INTEL_PT_ERR_NOMEM;
393*4882a593Smuzhiyun 	case -ENOSYS:
394*4882a593Smuzhiyun 		return INTEL_PT_ERR_INTERN;
395*4882a593Smuzhiyun 	case -EBADMSG:
396*4882a593Smuzhiyun 		return INTEL_PT_ERR_BADPKT;
397*4882a593Smuzhiyun 	case -ENODATA:
398*4882a593Smuzhiyun 		return INTEL_PT_ERR_NODATA;
399*4882a593Smuzhiyun 	case -EILSEQ:
400*4882a593Smuzhiyun 		return INTEL_PT_ERR_NOINSN;
401*4882a593Smuzhiyun 	case -ENOENT:
402*4882a593Smuzhiyun 		return INTEL_PT_ERR_MISMAT;
403*4882a593Smuzhiyun 	case -EOVERFLOW:
404*4882a593Smuzhiyun 		return INTEL_PT_ERR_OVR;
405*4882a593Smuzhiyun 	case -ENOSPC:
406*4882a593Smuzhiyun 		return INTEL_PT_ERR_LOST;
407*4882a593Smuzhiyun 	case -ELOOP:
408*4882a593Smuzhiyun 		return INTEL_PT_ERR_NELOOP;
409*4882a593Smuzhiyun 	default:
410*4882a593Smuzhiyun 		return INTEL_PT_ERR_UNK;
411*4882a593Smuzhiyun 	}
412*4882a593Smuzhiyun }
413*4882a593Smuzhiyun 
414*4882a593Smuzhiyun static const char *intel_pt_err_msgs[] = {
415*4882a593Smuzhiyun 	[INTEL_PT_ERR_NOMEM]  = "Memory allocation failed",
416*4882a593Smuzhiyun 	[INTEL_PT_ERR_INTERN] = "Internal error",
417*4882a593Smuzhiyun 	[INTEL_PT_ERR_BADPKT] = "Bad packet",
418*4882a593Smuzhiyun 	[INTEL_PT_ERR_NODATA] = "No more data",
419*4882a593Smuzhiyun 	[INTEL_PT_ERR_NOINSN] = "Failed to get instruction",
420*4882a593Smuzhiyun 	[INTEL_PT_ERR_MISMAT] = "Trace doesn't match instruction",
421*4882a593Smuzhiyun 	[INTEL_PT_ERR_OVR]    = "Overflow packet",
422*4882a593Smuzhiyun 	[INTEL_PT_ERR_LOST]   = "Lost trace data",
423*4882a593Smuzhiyun 	[INTEL_PT_ERR_UNK]    = "Unknown error!",
424*4882a593Smuzhiyun 	[INTEL_PT_ERR_NELOOP] = "Never-ending loop",
425*4882a593Smuzhiyun };
426*4882a593Smuzhiyun 
intel_pt__strerror(int code,char * buf,size_t buflen)427*4882a593Smuzhiyun int intel_pt__strerror(int code, char *buf, size_t buflen)
428*4882a593Smuzhiyun {
429*4882a593Smuzhiyun 	if (code < 1 || code >= INTEL_PT_ERR_MAX)
430*4882a593Smuzhiyun 		code = INTEL_PT_ERR_UNK;
431*4882a593Smuzhiyun 	strlcpy(buf, intel_pt_err_msgs[code], buflen);
432*4882a593Smuzhiyun 	return 0;
433*4882a593Smuzhiyun }
434*4882a593Smuzhiyun 
intel_pt_calc_ip(const struct intel_pt_pkt * packet,uint64_t last_ip)435*4882a593Smuzhiyun static uint64_t intel_pt_calc_ip(const struct intel_pt_pkt *packet,
436*4882a593Smuzhiyun 				 uint64_t last_ip)
437*4882a593Smuzhiyun {
438*4882a593Smuzhiyun 	uint64_t ip;
439*4882a593Smuzhiyun 
440*4882a593Smuzhiyun 	switch (packet->count) {
441*4882a593Smuzhiyun 	case 1:
442*4882a593Smuzhiyun 		ip = (last_ip & (uint64_t)0xffffffffffff0000ULL) |
443*4882a593Smuzhiyun 		     packet->payload;
444*4882a593Smuzhiyun 		break;
445*4882a593Smuzhiyun 	case 2:
446*4882a593Smuzhiyun 		ip = (last_ip & (uint64_t)0xffffffff00000000ULL) |
447*4882a593Smuzhiyun 		     packet->payload;
448*4882a593Smuzhiyun 		break;
449*4882a593Smuzhiyun 	case 3:
450*4882a593Smuzhiyun 		ip = packet->payload;
451*4882a593Smuzhiyun 		/* Sign-extend 6-byte ip */
452*4882a593Smuzhiyun 		if (ip & (uint64_t)0x800000000000ULL)
453*4882a593Smuzhiyun 			ip |= (uint64_t)0xffff000000000000ULL;
454*4882a593Smuzhiyun 		break;
455*4882a593Smuzhiyun 	case 4:
456*4882a593Smuzhiyun 		ip = (last_ip & (uint64_t)0xffff000000000000ULL) |
457*4882a593Smuzhiyun 		     packet->payload;
458*4882a593Smuzhiyun 		break;
459*4882a593Smuzhiyun 	case 6:
460*4882a593Smuzhiyun 		ip = packet->payload;
461*4882a593Smuzhiyun 		break;
462*4882a593Smuzhiyun 	default:
463*4882a593Smuzhiyun 		return 0;
464*4882a593Smuzhiyun 	}
465*4882a593Smuzhiyun 
466*4882a593Smuzhiyun 	return ip;
467*4882a593Smuzhiyun }
468*4882a593Smuzhiyun 
intel_pt_set_last_ip(struct intel_pt_decoder * decoder)469*4882a593Smuzhiyun static inline void intel_pt_set_last_ip(struct intel_pt_decoder *decoder)
470*4882a593Smuzhiyun {
471*4882a593Smuzhiyun 	decoder->last_ip = intel_pt_calc_ip(&decoder->packet, decoder->last_ip);
472*4882a593Smuzhiyun 	decoder->have_last_ip = true;
473*4882a593Smuzhiyun }
474*4882a593Smuzhiyun 
intel_pt_set_ip(struct intel_pt_decoder * decoder)475*4882a593Smuzhiyun static inline void intel_pt_set_ip(struct intel_pt_decoder *decoder)
476*4882a593Smuzhiyun {
477*4882a593Smuzhiyun 	intel_pt_set_last_ip(decoder);
478*4882a593Smuzhiyun 	decoder->ip = decoder->last_ip;
479*4882a593Smuzhiyun }
480*4882a593Smuzhiyun 
intel_pt_decoder_log_packet(struct intel_pt_decoder * decoder)481*4882a593Smuzhiyun static void intel_pt_decoder_log_packet(struct intel_pt_decoder *decoder)
482*4882a593Smuzhiyun {
483*4882a593Smuzhiyun 	intel_pt_log_packet(&decoder->packet, decoder->pkt_len, decoder->pos,
484*4882a593Smuzhiyun 			    decoder->buf);
485*4882a593Smuzhiyun }
486*4882a593Smuzhiyun 
intel_pt_bug(struct intel_pt_decoder * decoder)487*4882a593Smuzhiyun static int intel_pt_bug(struct intel_pt_decoder *decoder)
488*4882a593Smuzhiyun {
489*4882a593Smuzhiyun 	intel_pt_log("ERROR: Internal error\n");
490*4882a593Smuzhiyun 	decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
491*4882a593Smuzhiyun 	return -ENOSYS;
492*4882a593Smuzhiyun }
493*4882a593Smuzhiyun 
intel_pt_clear_tx_flags(struct intel_pt_decoder * decoder)494*4882a593Smuzhiyun static inline void intel_pt_clear_tx_flags(struct intel_pt_decoder *decoder)
495*4882a593Smuzhiyun {
496*4882a593Smuzhiyun 	decoder->tx_flags = 0;
497*4882a593Smuzhiyun }
498*4882a593Smuzhiyun 
intel_pt_update_in_tx(struct intel_pt_decoder * decoder)499*4882a593Smuzhiyun static inline void intel_pt_update_in_tx(struct intel_pt_decoder *decoder)
500*4882a593Smuzhiyun {
501*4882a593Smuzhiyun 	decoder->tx_flags = decoder->packet.payload & INTEL_PT_IN_TX;
502*4882a593Smuzhiyun }
503*4882a593Smuzhiyun 
intel_pt_bad_packet(struct intel_pt_decoder * decoder)504*4882a593Smuzhiyun static int intel_pt_bad_packet(struct intel_pt_decoder *decoder)
505*4882a593Smuzhiyun {
506*4882a593Smuzhiyun 	intel_pt_clear_tx_flags(decoder);
507*4882a593Smuzhiyun 	decoder->have_tma = false;
508*4882a593Smuzhiyun 	decoder->pkt_len = 1;
509*4882a593Smuzhiyun 	decoder->pkt_step = 1;
510*4882a593Smuzhiyun 	intel_pt_decoder_log_packet(decoder);
511*4882a593Smuzhiyun 	if (decoder->pkt_state != INTEL_PT_STATE_NO_PSB) {
512*4882a593Smuzhiyun 		intel_pt_log("ERROR: Bad packet\n");
513*4882a593Smuzhiyun 		decoder->pkt_state = INTEL_PT_STATE_ERR1;
514*4882a593Smuzhiyun 	}
515*4882a593Smuzhiyun 	return -EBADMSG;
516*4882a593Smuzhiyun }
517*4882a593Smuzhiyun 
intel_pt_update_sample_time(struct intel_pt_decoder * decoder)518*4882a593Smuzhiyun static inline void intel_pt_update_sample_time(struct intel_pt_decoder *decoder)
519*4882a593Smuzhiyun {
520*4882a593Smuzhiyun 	decoder->sample_timestamp = decoder->timestamp;
521*4882a593Smuzhiyun 	decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
522*4882a593Smuzhiyun }
523*4882a593Smuzhiyun 
intel_pt_reposition(struct intel_pt_decoder * decoder)524*4882a593Smuzhiyun static void intel_pt_reposition(struct intel_pt_decoder *decoder)
525*4882a593Smuzhiyun {
526*4882a593Smuzhiyun 	decoder->ip = 0;
527*4882a593Smuzhiyun 	decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
528*4882a593Smuzhiyun 	decoder->timestamp = 0;
529*4882a593Smuzhiyun 	decoder->have_tma = false;
530*4882a593Smuzhiyun }
531*4882a593Smuzhiyun 
intel_pt_get_data(struct intel_pt_decoder * decoder,bool reposition)532*4882a593Smuzhiyun static int intel_pt_get_data(struct intel_pt_decoder *decoder, bool reposition)
533*4882a593Smuzhiyun {
534*4882a593Smuzhiyun 	struct intel_pt_buffer buffer = { .buf = 0, };
535*4882a593Smuzhiyun 	int ret;
536*4882a593Smuzhiyun 
537*4882a593Smuzhiyun 	decoder->pkt_step = 0;
538*4882a593Smuzhiyun 
539*4882a593Smuzhiyun 	intel_pt_log("Getting more data\n");
540*4882a593Smuzhiyun 	ret = decoder->get_trace(&buffer, decoder->data);
541*4882a593Smuzhiyun 	if (ret)
542*4882a593Smuzhiyun 		return ret;
543*4882a593Smuzhiyun 	decoder->buf = buffer.buf;
544*4882a593Smuzhiyun 	decoder->len = buffer.len;
545*4882a593Smuzhiyun 	if (!decoder->len) {
546*4882a593Smuzhiyun 		intel_pt_log("No more data\n");
547*4882a593Smuzhiyun 		return -ENODATA;
548*4882a593Smuzhiyun 	}
549*4882a593Smuzhiyun 	decoder->buf_timestamp = buffer.ref_timestamp;
550*4882a593Smuzhiyun 	if (!buffer.consecutive || reposition) {
551*4882a593Smuzhiyun 		intel_pt_reposition(decoder);
552*4882a593Smuzhiyun 		decoder->ref_timestamp = buffer.ref_timestamp;
553*4882a593Smuzhiyun 		decoder->state.trace_nr = buffer.trace_nr;
554*4882a593Smuzhiyun 		intel_pt_log("Reference timestamp 0x%" PRIx64 "\n",
555*4882a593Smuzhiyun 			     decoder->ref_timestamp);
556*4882a593Smuzhiyun 		return -ENOLINK;
557*4882a593Smuzhiyun 	}
558*4882a593Smuzhiyun 
559*4882a593Smuzhiyun 	return 0;
560*4882a593Smuzhiyun }
561*4882a593Smuzhiyun 
intel_pt_get_next_data(struct intel_pt_decoder * decoder,bool reposition)562*4882a593Smuzhiyun static int intel_pt_get_next_data(struct intel_pt_decoder *decoder,
563*4882a593Smuzhiyun 				  bool reposition)
564*4882a593Smuzhiyun {
565*4882a593Smuzhiyun 	if (!decoder->next_buf)
566*4882a593Smuzhiyun 		return intel_pt_get_data(decoder, reposition);
567*4882a593Smuzhiyun 
568*4882a593Smuzhiyun 	decoder->buf = decoder->next_buf;
569*4882a593Smuzhiyun 	decoder->len = decoder->next_len;
570*4882a593Smuzhiyun 	decoder->next_buf = 0;
571*4882a593Smuzhiyun 	decoder->next_len = 0;
572*4882a593Smuzhiyun 	return 0;
573*4882a593Smuzhiyun }
574*4882a593Smuzhiyun 
intel_pt_get_split_packet(struct intel_pt_decoder * decoder)575*4882a593Smuzhiyun static int intel_pt_get_split_packet(struct intel_pt_decoder *decoder)
576*4882a593Smuzhiyun {
577*4882a593Smuzhiyun 	unsigned char *buf = decoder->temp_buf;
578*4882a593Smuzhiyun 	size_t old_len, len, n;
579*4882a593Smuzhiyun 	int ret;
580*4882a593Smuzhiyun 
581*4882a593Smuzhiyun 	old_len = decoder->len;
582*4882a593Smuzhiyun 	len = decoder->len;
583*4882a593Smuzhiyun 	memcpy(buf, decoder->buf, len);
584*4882a593Smuzhiyun 
585*4882a593Smuzhiyun 	ret = intel_pt_get_data(decoder, false);
586*4882a593Smuzhiyun 	if (ret) {
587*4882a593Smuzhiyun 		decoder->pos += old_len;
588*4882a593Smuzhiyun 		return ret < 0 ? ret : -EINVAL;
589*4882a593Smuzhiyun 	}
590*4882a593Smuzhiyun 
591*4882a593Smuzhiyun 	n = INTEL_PT_PKT_MAX_SZ - len;
592*4882a593Smuzhiyun 	if (n > decoder->len)
593*4882a593Smuzhiyun 		n = decoder->len;
594*4882a593Smuzhiyun 	memcpy(buf + len, decoder->buf, n);
595*4882a593Smuzhiyun 	len += n;
596*4882a593Smuzhiyun 
597*4882a593Smuzhiyun 	decoder->prev_pkt_ctx = decoder->pkt_ctx;
598*4882a593Smuzhiyun 	ret = intel_pt_get_packet(buf, len, &decoder->packet, &decoder->pkt_ctx);
599*4882a593Smuzhiyun 	if (ret < (int)old_len) {
600*4882a593Smuzhiyun 		decoder->next_buf = decoder->buf;
601*4882a593Smuzhiyun 		decoder->next_len = decoder->len;
602*4882a593Smuzhiyun 		decoder->buf = buf;
603*4882a593Smuzhiyun 		decoder->len = old_len;
604*4882a593Smuzhiyun 		return intel_pt_bad_packet(decoder);
605*4882a593Smuzhiyun 	}
606*4882a593Smuzhiyun 
607*4882a593Smuzhiyun 	decoder->next_buf = decoder->buf + (ret - old_len);
608*4882a593Smuzhiyun 	decoder->next_len = decoder->len - (ret - old_len);
609*4882a593Smuzhiyun 
610*4882a593Smuzhiyun 	decoder->buf = buf;
611*4882a593Smuzhiyun 	decoder->len = ret;
612*4882a593Smuzhiyun 
613*4882a593Smuzhiyun 	return ret;
614*4882a593Smuzhiyun }
615*4882a593Smuzhiyun 
616*4882a593Smuzhiyun struct intel_pt_pkt_info {
617*4882a593Smuzhiyun 	struct intel_pt_decoder	  *decoder;
618*4882a593Smuzhiyun 	struct intel_pt_pkt       packet;
619*4882a593Smuzhiyun 	uint64_t                  pos;
620*4882a593Smuzhiyun 	int                       pkt_len;
621*4882a593Smuzhiyun 	int                       last_packet_type;
622*4882a593Smuzhiyun 	void                      *data;
623*4882a593Smuzhiyun };
624*4882a593Smuzhiyun 
625*4882a593Smuzhiyun typedef int (*intel_pt_pkt_cb_t)(struct intel_pt_pkt_info *pkt_info);
626*4882a593Smuzhiyun 
627*4882a593Smuzhiyun /* Lookahead packets in current buffer */
intel_pt_pkt_lookahead(struct intel_pt_decoder * decoder,intel_pt_pkt_cb_t cb,void * data)628*4882a593Smuzhiyun static int intel_pt_pkt_lookahead(struct intel_pt_decoder *decoder,
629*4882a593Smuzhiyun 				  intel_pt_pkt_cb_t cb, void *data)
630*4882a593Smuzhiyun {
631*4882a593Smuzhiyun 	struct intel_pt_pkt_info pkt_info;
632*4882a593Smuzhiyun 	const unsigned char *buf = decoder->buf;
633*4882a593Smuzhiyun 	enum intel_pt_pkt_ctx pkt_ctx = decoder->pkt_ctx;
634*4882a593Smuzhiyun 	size_t len = decoder->len;
635*4882a593Smuzhiyun 	int ret;
636*4882a593Smuzhiyun 
637*4882a593Smuzhiyun 	pkt_info.decoder          = decoder;
638*4882a593Smuzhiyun 	pkt_info.pos              = decoder->pos;
639*4882a593Smuzhiyun 	pkt_info.pkt_len          = decoder->pkt_step;
640*4882a593Smuzhiyun 	pkt_info.last_packet_type = decoder->last_packet_type;
641*4882a593Smuzhiyun 	pkt_info.data             = data;
642*4882a593Smuzhiyun 
643*4882a593Smuzhiyun 	while (1) {
644*4882a593Smuzhiyun 		do {
645*4882a593Smuzhiyun 			pkt_info.pos += pkt_info.pkt_len;
646*4882a593Smuzhiyun 			buf          += pkt_info.pkt_len;
647*4882a593Smuzhiyun 			len          -= pkt_info.pkt_len;
648*4882a593Smuzhiyun 
649*4882a593Smuzhiyun 			if (!len)
650*4882a593Smuzhiyun 				return INTEL_PT_NEED_MORE_BYTES;
651*4882a593Smuzhiyun 
652*4882a593Smuzhiyun 			ret = intel_pt_get_packet(buf, len, &pkt_info.packet,
653*4882a593Smuzhiyun 						  &pkt_ctx);
654*4882a593Smuzhiyun 			if (!ret)
655*4882a593Smuzhiyun 				return INTEL_PT_NEED_MORE_BYTES;
656*4882a593Smuzhiyun 			if (ret < 0)
657*4882a593Smuzhiyun 				return ret;
658*4882a593Smuzhiyun 
659*4882a593Smuzhiyun 			pkt_info.pkt_len = ret;
660*4882a593Smuzhiyun 		} while (pkt_info.packet.type == INTEL_PT_PAD);
661*4882a593Smuzhiyun 
662*4882a593Smuzhiyun 		ret = cb(&pkt_info);
663*4882a593Smuzhiyun 		if (ret)
664*4882a593Smuzhiyun 			return 0;
665*4882a593Smuzhiyun 
666*4882a593Smuzhiyun 		pkt_info.last_packet_type = pkt_info.packet.type;
667*4882a593Smuzhiyun 	}
668*4882a593Smuzhiyun }
669*4882a593Smuzhiyun 
670*4882a593Smuzhiyun struct intel_pt_calc_cyc_to_tsc_info {
671*4882a593Smuzhiyun 	uint64_t        cycle_cnt;
672*4882a593Smuzhiyun 	unsigned int    cbr;
673*4882a593Smuzhiyun 	uint32_t        last_mtc;
674*4882a593Smuzhiyun 	uint64_t        ctc_timestamp;
675*4882a593Smuzhiyun 	uint64_t        ctc_delta;
676*4882a593Smuzhiyun 	uint64_t        tsc_timestamp;
677*4882a593Smuzhiyun 	uint64_t        timestamp;
678*4882a593Smuzhiyun 	bool            have_tma;
679*4882a593Smuzhiyun 	bool            fixup_last_mtc;
680*4882a593Smuzhiyun 	bool            from_mtc;
681*4882a593Smuzhiyun 	double          cbr_cyc_to_tsc;
682*4882a593Smuzhiyun };
683*4882a593Smuzhiyun 
684*4882a593Smuzhiyun /*
685*4882a593Smuzhiyun  * MTC provides a 8-bit slice of CTC but the TMA packet only provides the lower
686*4882a593Smuzhiyun  * 16 bits of CTC. If mtc_shift > 8 then some of the MTC bits are not in the CTC
687*4882a593Smuzhiyun  * provided by the TMA packet. Fix-up the last_mtc calculated from the TMA
688*4882a593Smuzhiyun  * packet by copying the missing bits from the current MTC assuming the least
689*4882a593Smuzhiyun  * difference between the two, and that the current MTC comes after last_mtc.
690*4882a593Smuzhiyun  */
intel_pt_fixup_last_mtc(uint32_t mtc,int mtc_shift,uint32_t * last_mtc)691*4882a593Smuzhiyun static void intel_pt_fixup_last_mtc(uint32_t mtc, int mtc_shift,
692*4882a593Smuzhiyun 				    uint32_t *last_mtc)
693*4882a593Smuzhiyun {
694*4882a593Smuzhiyun 	uint32_t first_missing_bit = 1U << (16 - mtc_shift);
695*4882a593Smuzhiyun 	uint32_t mask = ~(first_missing_bit - 1);
696*4882a593Smuzhiyun 
697*4882a593Smuzhiyun 	*last_mtc |= mtc & mask;
698*4882a593Smuzhiyun 	if (*last_mtc >= mtc) {
699*4882a593Smuzhiyun 		*last_mtc -= first_missing_bit;
700*4882a593Smuzhiyun 		*last_mtc &= 0xff;
701*4882a593Smuzhiyun 	}
702*4882a593Smuzhiyun }
703*4882a593Smuzhiyun 
intel_pt_calc_cyc_cb(struct intel_pt_pkt_info * pkt_info)704*4882a593Smuzhiyun static int intel_pt_calc_cyc_cb(struct intel_pt_pkt_info *pkt_info)
705*4882a593Smuzhiyun {
706*4882a593Smuzhiyun 	struct intel_pt_decoder *decoder = pkt_info->decoder;
707*4882a593Smuzhiyun 	struct intel_pt_calc_cyc_to_tsc_info *data = pkt_info->data;
708*4882a593Smuzhiyun 	uint64_t timestamp;
709*4882a593Smuzhiyun 	double cyc_to_tsc;
710*4882a593Smuzhiyun 	unsigned int cbr;
711*4882a593Smuzhiyun 	uint32_t mtc, mtc_delta, ctc, fc, ctc_rem;
712*4882a593Smuzhiyun 
713*4882a593Smuzhiyun 	switch (pkt_info->packet.type) {
714*4882a593Smuzhiyun 	case INTEL_PT_TNT:
715*4882a593Smuzhiyun 	case INTEL_PT_TIP_PGE:
716*4882a593Smuzhiyun 	case INTEL_PT_TIP:
717*4882a593Smuzhiyun 	case INTEL_PT_FUP:
718*4882a593Smuzhiyun 	case INTEL_PT_PSB:
719*4882a593Smuzhiyun 	case INTEL_PT_PIP:
720*4882a593Smuzhiyun 	case INTEL_PT_MODE_EXEC:
721*4882a593Smuzhiyun 	case INTEL_PT_MODE_TSX:
722*4882a593Smuzhiyun 	case INTEL_PT_PSBEND:
723*4882a593Smuzhiyun 	case INTEL_PT_PAD:
724*4882a593Smuzhiyun 	case INTEL_PT_VMCS:
725*4882a593Smuzhiyun 	case INTEL_PT_MNT:
726*4882a593Smuzhiyun 	case INTEL_PT_PTWRITE:
727*4882a593Smuzhiyun 	case INTEL_PT_PTWRITE_IP:
728*4882a593Smuzhiyun 	case INTEL_PT_BBP:
729*4882a593Smuzhiyun 	case INTEL_PT_BIP:
730*4882a593Smuzhiyun 	case INTEL_PT_BEP:
731*4882a593Smuzhiyun 	case INTEL_PT_BEP_IP:
732*4882a593Smuzhiyun 		return 0;
733*4882a593Smuzhiyun 
734*4882a593Smuzhiyun 	case INTEL_PT_MTC:
735*4882a593Smuzhiyun 		if (!data->have_tma)
736*4882a593Smuzhiyun 			return 0;
737*4882a593Smuzhiyun 
738*4882a593Smuzhiyun 		mtc = pkt_info->packet.payload;
739*4882a593Smuzhiyun 		if (decoder->mtc_shift > 8 && data->fixup_last_mtc) {
740*4882a593Smuzhiyun 			data->fixup_last_mtc = false;
741*4882a593Smuzhiyun 			intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
742*4882a593Smuzhiyun 						&data->last_mtc);
743*4882a593Smuzhiyun 		}
744*4882a593Smuzhiyun 		if (mtc > data->last_mtc)
745*4882a593Smuzhiyun 			mtc_delta = mtc - data->last_mtc;
746*4882a593Smuzhiyun 		else
747*4882a593Smuzhiyun 			mtc_delta = mtc + 256 - data->last_mtc;
748*4882a593Smuzhiyun 		data->ctc_delta += mtc_delta << decoder->mtc_shift;
749*4882a593Smuzhiyun 		data->last_mtc = mtc;
750*4882a593Smuzhiyun 
751*4882a593Smuzhiyun 		if (decoder->tsc_ctc_mult) {
752*4882a593Smuzhiyun 			timestamp = data->ctc_timestamp +
753*4882a593Smuzhiyun 				data->ctc_delta * decoder->tsc_ctc_mult;
754*4882a593Smuzhiyun 		} else {
755*4882a593Smuzhiyun 			timestamp = data->ctc_timestamp +
756*4882a593Smuzhiyun 				multdiv(data->ctc_delta,
757*4882a593Smuzhiyun 					decoder->tsc_ctc_ratio_n,
758*4882a593Smuzhiyun 					decoder->tsc_ctc_ratio_d);
759*4882a593Smuzhiyun 		}
760*4882a593Smuzhiyun 
761*4882a593Smuzhiyun 		if (timestamp < data->timestamp)
762*4882a593Smuzhiyun 			return 1;
763*4882a593Smuzhiyun 
764*4882a593Smuzhiyun 		if (pkt_info->last_packet_type != INTEL_PT_CYC) {
765*4882a593Smuzhiyun 			data->timestamp = timestamp;
766*4882a593Smuzhiyun 			return 0;
767*4882a593Smuzhiyun 		}
768*4882a593Smuzhiyun 
769*4882a593Smuzhiyun 		break;
770*4882a593Smuzhiyun 
771*4882a593Smuzhiyun 	case INTEL_PT_TSC:
772*4882a593Smuzhiyun 		/*
773*4882a593Smuzhiyun 		 * For now, do not support using TSC packets - refer
774*4882a593Smuzhiyun 		 * intel_pt_calc_cyc_to_tsc().
775*4882a593Smuzhiyun 		 */
776*4882a593Smuzhiyun 		if (data->from_mtc)
777*4882a593Smuzhiyun 			return 1;
778*4882a593Smuzhiyun 		timestamp = pkt_info->packet.payload |
779*4882a593Smuzhiyun 			    (data->timestamp & (0xffULL << 56));
780*4882a593Smuzhiyun 		if (data->from_mtc && timestamp < data->timestamp &&
781*4882a593Smuzhiyun 		    data->timestamp - timestamp < decoder->tsc_slip)
782*4882a593Smuzhiyun 			return 1;
783*4882a593Smuzhiyun 		if (timestamp < data->timestamp)
784*4882a593Smuzhiyun 			timestamp += (1ULL << 56);
785*4882a593Smuzhiyun 		if (pkt_info->last_packet_type != INTEL_PT_CYC) {
786*4882a593Smuzhiyun 			if (data->from_mtc)
787*4882a593Smuzhiyun 				return 1;
788*4882a593Smuzhiyun 			data->tsc_timestamp = timestamp;
789*4882a593Smuzhiyun 			data->timestamp = timestamp;
790*4882a593Smuzhiyun 			return 0;
791*4882a593Smuzhiyun 		}
792*4882a593Smuzhiyun 		break;
793*4882a593Smuzhiyun 
794*4882a593Smuzhiyun 	case INTEL_PT_TMA:
795*4882a593Smuzhiyun 		if (data->from_mtc)
796*4882a593Smuzhiyun 			return 1;
797*4882a593Smuzhiyun 
798*4882a593Smuzhiyun 		if (!decoder->tsc_ctc_ratio_d)
799*4882a593Smuzhiyun 			return 0;
800*4882a593Smuzhiyun 
801*4882a593Smuzhiyun 		ctc = pkt_info->packet.payload;
802*4882a593Smuzhiyun 		fc = pkt_info->packet.count;
803*4882a593Smuzhiyun 		ctc_rem = ctc & decoder->ctc_rem_mask;
804*4882a593Smuzhiyun 
805*4882a593Smuzhiyun 		data->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
806*4882a593Smuzhiyun 
807*4882a593Smuzhiyun 		data->ctc_timestamp = data->tsc_timestamp - fc;
808*4882a593Smuzhiyun 		if (decoder->tsc_ctc_mult) {
809*4882a593Smuzhiyun 			data->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
810*4882a593Smuzhiyun 		} else {
811*4882a593Smuzhiyun 			data->ctc_timestamp -=
812*4882a593Smuzhiyun 				multdiv(ctc_rem, decoder->tsc_ctc_ratio_n,
813*4882a593Smuzhiyun 					decoder->tsc_ctc_ratio_d);
814*4882a593Smuzhiyun 		}
815*4882a593Smuzhiyun 
816*4882a593Smuzhiyun 		data->ctc_delta = 0;
817*4882a593Smuzhiyun 		data->have_tma = true;
818*4882a593Smuzhiyun 		data->fixup_last_mtc = true;
819*4882a593Smuzhiyun 
820*4882a593Smuzhiyun 		return 0;
821*4882a593Smuzhiyun 
822*4882a593Smuzhiyun 	case INTEL_PT_CYC:
823*4882a593Smuzhiyun 		data->cycle_cnt += pkt_info->packet.payload;
824*4882a593Smuzhiyun 		return 0;
825*4882a593Smuzhiyun 
826*4882a593Smuzhiyun 	case INTEL_PT_CBR:
827*4882a593Smuzhiyun 		cbr = pkt_info->packet.payload;
828*4882a593Smuzhiyun 		if (data->cbr && data->cbr != cbr)
829*4882a593Smuzhiyun 			return 1;
830*4882a593Smuzhiyun 		data->cbr = cbr;
831*4882a593Smuzhiyun 		data->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
832*4882a593Smuzhiyun 		return 0;
833*4882a593Smuzhiyun 
834*4882a593Smuzhiyun 	case INTEL_PT_TIP_PGD:
835*4882a593Smuzhiyun 	case INTEL_PT_TRACESTOP:
836*4882a593Smuzhiyun 	case INTEL_PT_EXSTOP:
837*4882a593Smuzhiyun 	case INTEL_PT_EXSTOP_IP:
838*4882a593Smuzhiyun 	case INTEL_PT_MWAIT:
839*4882a593Smuzhiyun 	case INTEL_PT_PWRE:
840*4882a593Smuzhiyun 	case INTEL_PT_PWRX:
841*4882a593Smuzhiyun 	case INTEL_PT_OVF:
842*4882a593Smuzhiyun 	case INTEL_PT_BAD: /* Does not happen */
843*4882a593Smuzhiyun 	default:
844*4882a593Smuzhiyun 		return 1;
845*4882a593Smuzhiyun 	}
846*4882a593Smuzhiyun 
847*4882a593Smuzhiyun 	if (!data->cbr && decoder->cbr) {
848*4882a593Smuzhiyun 		data->cbr = decoder->cbr;
849*4882a593Smuzhiyun 		data->cbr_cyc_to_tsc = decoder->cbr_cyc_to_tsc;
850*4882a593Smuzhiyun 	}
851*4882a593Smuzhiyun 
852*4882a593Smuzhiyun 	if (!data->cycle_cnt)
853*4882a593Smuzhiyun 		return 1;
854*4882a593Smuzhiyun 
855*4882a593Smuzhiyun 	cyc_to_tsc = (double)(timestamp - decoder->timestamp) / data->cycle_cnt;
856*4882a593Smuzhiyun 
857*4882a593Smuzhiyun 	if (data->cbr && cyc_to_tsc > data->cbr_cyc_to_tsc &&
858*4882a593Smuzhiyun 	    cyc_to_tsc / data->cbr_cyc_to_tsc > 1.25) {
859*4882a593Smuzhiyun 		intel_pt_log("Timestamp: calculated %g TSC ticks per cycle too big (c.f. CBR-based value %g), pos " x64_fmt "\n",
860*4882a593Smuzhiyun 			     cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
861*4882a593Smuzhiyun 		return 1;
862*4882a593Smuzhiyun 	}
863*4882a593Smuzhiyun 
864*4882a593Smuzhiyun 	decoder->calc_cyc_to_tsc = cyc_to_tsc;
865*4882a593Smuzhiyun 	decoder->have_calc_cyc_to_tsc = true;
866*4882a593Smuzhiyun 
867*4882a593Smuzhiyun 	if (data->cbr) {
868*4882a593Smuzhiyun 		intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. CBR-based value %g, pos " x64_fmt "\n",
869*4882a593Smuzhiyun 			     cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
870*4882a593Smuzhiyun 	} else {
871*4882a593Smuzhiyun 		intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. unknown CBR-based value, pos " x64_fmt "\n",
872*4882a593Smuzhiyun 			     cyc_to_tsc, pkt_info->pos);
873*4882a593Smuzhiyun 	}
874*4882a593Smuzhiyun 
875*4882a593Smuzhiyun 	return 1;
876*4882a593Smuzhiyun }
877*4882a593Smuzhiyun 
intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder * decoder,bool from_mtc)878*4882a593Smuzhiyun static void intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder *decoder,
879*4882a593Smuzhiyun 				     bool from_mtc)
880*4882a593Smuzhiyun {
881*4882a593Smuzhiyun 	struct intel_pt_calc_cyc_to_tsc_info data = {
882*4882a593Smuzhiyun 		.cycle_cnt      = 0,
883*4882a593Smuzhiyun 		.cbr            = 0,
884*4882a593Smuzhiyun 		.last_mtc       = decoder->last_mtc,
885*4882a593Smuzhiyun 		.ctc_timestamp  = decoder->ctc_timestamp,
886*4882a593Smuzhiyun 		.ctc_delta      = decoder->ctc_delta,
887*4882a593Smuzhiyun 		.tsc_timestamp  = decoder->tsc_timestamp,
888*4882a593Smuzhiyun 		.timestamp      = decoder->timestamp,
889*4882a593Smuzhiyun 		.have_tma       = decoder->have_tma,
890*4882a593Smuzhiyun 		.fixup_last_mtc = decoder->fixup_last_mtc,
891*4882a593Smuzhiyun 		.from_mtc       = from_mtc,
892*4882a593Smuzhiyun 		.cbr_cyc_to_tsc = 0,
893*4882a593Smuzhiyun 	};
894*4882a593Smuzhiyun 
895*4882a593Smuzhiyun 	/*
896*4882a593Smuzhiyun 	 * For now, do not support using TSC packets for at least the reasons:
897*4882a593Smuzhiyun 	 * 1) timing might have stopped
898*4882a593Smuzhiyun 	 * 2) TSC packets within PSB+ can slip against CYC packets
899*4882a593Smuzhiyun 	 */
900*4882a593Smuzhiyun 	if (!from_mtc)
901*4882a593Smuzhiyun 		return;
902*4882a593Smuzhiyun 
903*4882a593Smuzhiyun 	intel_pt_pkt_lookahead(decoder, intel_pt_calc_cyc_cb, &data);
904*4882a593Smuzhiyun }
905*4882a593Smuzhiyun 
intel_pt_get_next_packet(struct intel_pt_decoder * decoder)906*4882a593Smuzhiyun static int intel_pt_get_next_packet(struct intel_pt_decoder *decoder)
907*4882a593Smuzhiyun {
908*4882a593Smuzhiyun 	int ret;
909*4882a593Smuzhiyun 
910*4882a593Smuzhiyun 	decoder->last_packet_type = decoder->packet.type;
911*4882a593Smuzhiyun 
912*4882a593Smuzhiyun 	do {
913*4882a593Smuzhiyun 		decoder->pos += decoder->pkt_step;
914*4882a593Smuzhiyun 		decoder->buf += decoder->pkt_step;
915*4882a593Smuzhiyun 		decoder->len -= decoder->pkt_step;
916*4882a593Smuzhiyun 
917*4882a593Smuzhiyun 		if (!decoder->len) {
918*4882a593Smuzhiyun 			ret = intel_pt_get_next_data(decoder, false);
919*4882a593Smuzhiyun 			if (ret)
920*4882a593Smuzhiyun 				return ret;
921*4882a593Smuzhiyun 		}
922*4882a593Smuzhiyun 
923*4882a593Smuzhiyun 		decoder->prev_pkt_ctx = decoder->pkt_ctx;
924*4882a593Smuzhiyun 		ret = intel_pt_get_packet(decoder->buf, decoder->len,
925*4882a593Smuzhiyun 					  &decoder->packet, &decoder->pkt_ctx);
926*4882a593Smuzhiyun 		if (ret == INTEL_PT_NEED_MORE_BYTES && BITS_PER_LONG == 32 &&
927*4882a593Smuzhiyun 		    decoder->len < INTEL_PT_PKT_MAX_SZ && !decoder->next_buf) {
928*4882a593Smuzhiyun 			ret = intel_pt_get_split_packet(decoder);
929*4882a593Smuzhiyun 			if (ret < 0)
930*4882a593Smuzhiyun 				return ret;
931*4882a593Smuzhiyun 		}
932*4882a593Smuzhiyun 		if (ret <= 0)
933*4882a593Smuzhiyun 			return intel_pt_bad_packet(decoder);
934*4882a593Smuzhiyun 
935*4882a593Smuzhiyun 		decoder->pkt_len = ret;
936*4882a593Smuzhiyun 		decoder->pkt_step = ret;
937*4882a593Smuzhiyun 		intel_pt_decoder_log_packet(decoder);
938*4882a593Smuzhiyun 	} while (decoder->packet.type == INTEL_PT_PAD);
939*4882a593Smuzhiyun 
940*4882a593Smuzhiyun 	return 0;
941*4882a593Smuzhiyun }
942*4882a593Smuzhiyun 
intel_pt_next_period(struct intel_pt_decoder * decoder)943*4882a593Smuzhiyun static uint64_t intel_pt_next_period(struct intel_pt_decoder *decoder)
944*4882a593Smuzhiyun {
945*4882a593Smuzhiyun 	uint64_t timestamp, masked_timestamp;
946*4882a593Smuzhiyun 
947*4882a593Smuzhiyun 	timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
948*4882a593Smuzhiyun 	masked_timestamp = timestamp & decoder->period_mask;
949*4882a593Smuzhiyun 	if (decoder->continuous_period) {
950*4882a593Smuzhiyun 		if (masked_timestamp > decoder->last_masked_timestamp)
951*4882a593Smuzhiyun 			return 1;
952*4882a593Smuzhiyun 	} else {
953*4882a593Smuzhiyun 		timestamp += 1;
954*4882a593Smuzhiyun 		masked_timestamp = timestamp & decoder->period_mask;
955*4882a593Smuzhiyun 		if (masked_timestamp > decoder->last_masked_timestamp) {
956*4882a593Smuzhiyun 			decoder->last_masked_timestamp = masked_timestamp;
957*4882a593Smuzhiyun 			decoder->continuous_period = true;
958*4882a593Smuzhiyun 		}
959*4882a593Smuzhiyun 	}
960*4882a593Smuzhiyun 
961*4882a593Smuzhiyun 	if (masked_timestamp < decoder->last_masked_timestamp)
962*4882a593Smuzhiyun 		return decoder->period_ticks;
963*4882a593Smuzhiyun 
964*4882a593Smuzhiyun 	return decoder->period_ticks - (timestamp - masked_timestamp);
965*4882a593Smuzhiyun }
966*4882a593Smuzhiyun 
intel_pt_next_sample(struct intel_pt_decoder * decoder)967*4882a593Smuzhiyun static uint64_t intel_pt_next_sample(struct intel_pt_decoder *decoder)
968*4882a593Smuzhiyun {
969*4882a593Smuzhiyun 	switch (decoder->period_type) {
970*4882a593Smuzhiyun 	case INTEL_PT_PERIOD_INSTRUCTIONS:
971*4882a593Smuzhiyun 		return decoder->period - decoder->period_insn_cnt;
972*4882a593Smuzhiyun 	case INTEL_PT_PERIOD_TICKS:
973*4882a593Smuzhiyun 		return intel_pt_next_period(decoder);
974*4882a593Smuzhiyun 	case INTEL_PT_PERIOD_NONE:
975*4882a593Smuzhiyun 	case INTEL_PT_PERIOD_MTC:
976*4882a593Smuzhiyun 	default:
977*4882a593Smuzhiyun 		return 0;
978*4882a593Smuzhiyun 	}
979*4882a593Smuzhiyun }
980*4882a593Smuzhiyun 
intel_pt_sample_insn(struct intel_pt_decoder * decoder)981*4882a593Smuzhiyun static void intel_pt_sample_insn(struct intel_pt_decoder *decoder)
982*4882a593Smuzhiyun {
983*4882a593Smuzhiyun 	uint64_t timestamp, masked_timestamp;
984*4882a593Smuzhiyun 
985*4882a593Smuzhiyun 	switch (decoder->period_type) {
986*4882a593Smuzhiyun 	case INTEL_PT_PERIOD_INSTRUCTIONS:
987*4882a593Smuzhiyun 		decoder->period_insn_cnt = 0;
988*4882a593Smuzhiyun 		break;
989*4882a593Smuzhiyun 	case INTEL_PT_PERIOD_TICKS:
990*4882a593Smuzhiyun 		timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
991*4882a593Smuzhiyun 		masked_timestamp = timestamp & decoder->period_mask;
992*4882a593Smuzhiyun 		if (masked_timestamp > decoder->last_masked_timestamp)
993*4882a593Smuzhiyun 			decoder->last_masked_timestamp = masked_timestamp;
994*4882a593Smuzhiyun 		else
995*4882a593Smuzhiyun 			decoder->last_masked_timestamp += decoder->period_ticks;
996*4882a593Smuzhiyun 		break;
997*4882a593Smuzhiyun 	case INTEL_PT_PERIOD_NONE:
998*4882a593Smuzhiyun 	case INTEL_PT_PERIOD_MTC:
999*4882a593Smuzhiyun 	default:
1000*4882a593Smuzhiyun 		break;
1001*4882a593Smuzhiyun 	}
1002*4882a593Smuzhiyun 
1003*4882a593Smuzhiyun 	decoder->state.type |= INTEL_PT_INSTRUCTION;
1004*4882a593Smuzhiyun }
1005*4882a593Smuzhiyun 
intel_pt_walk_insn(struct intel_pt_decoder * decoder,struct intel_pt_insn * intel_pt_insn,uint64_t ip)1006*4882a593Smuzhiyun static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
1007*4882a593Smuzhiyun 			      struct intel_pt_insn *intel_pt_insn, uint64_t ip)
1008*4882a593Smuzhiyun {
1009*4882a593Smuzhiyun 	uint64_t max_insn_cnt, insn_cnt = 0;
1010*4882a593Smuzhiyun 	int err;
1011*4882a593Smuzhiyun 
1012*4882a593Smuzhiyun 	if (!decoder->mtc_insn)
1013*4882a593Smuzhiyun 		decoder->mtc_insn = true;
1014*4882a593Smuzhiyun 
1015*4882a593Smuzhiyun 	max_insn_cnt = intel_pt_next_sample(decoder);
1016*4882a593Smuzhiyun 
1017*4882a593Smuzhiyun 	err = decoder->walk_insn(intel_pt_insn, &insn_cnt, &decoder->ip, ip,
1018*4882a593Smuzhiyun 				 max_insn_cnt, decoder->data);
1019*4882a593Smuzhiyun 
1020*4882a593Smuzhiyun 	decoder->tot_insn_cnt += insn_cnt;
1021*4882a593Smuzhiyun 	decoder->timestamp_insn_cnt += insn_cnt;
1022*4882a593Smuzhiyun 	decoder->sample_insn_cnt += insn_cnt;
1023*4882a593Smuzhiyun 	decoder->period_insn_cnt += insn_cnt;
1024*4882a593Smuzhiyun 
1025*4882a593Smuzhiyun 	if (err) {
1026*4882a593Smuzhiyun 		decoder->no_progress = 0;
1027*4882a593Smuzhiyun 		decoder->pkt_state = INTEL_PT_STATE_ERR2;
1028*4882a593Smuzhiyun 		intel_pt_log_at("ERROR: Failed to get instruction",
1029*4882a593Smuzhiyun 				decoder->ip);
1030*4882a593Smuzhiyun 		if (err == -ENOENT)
1031*4882a593Smuzhiyun 			return -ENOLINK;
1032*4882a593Smuzhiyun 		return -EILSEQ;
1033*4882a593Smuzhiyun 	}
1034*4882a593Smuzhiyun 
1035*4882a593Smuzhiyun 	if (ip && decoder->ip == ip) {
1036*4882a593Smuzhiyun 		err = -EAGAIN;
1037*4882a593Smuzhiyun 		goto out;
1038*4882a593Smuzhiyun 	}
1039*4882a593Smuzhiyun 
1040*4882a593Smuzhiyun 	if (max_insn_cnt && insn_cnt >= max_insn_cnt)
1041*4882a593Smuzhiyun 		intel_pt_sample_insn(decoder);
1042*4882a593Smuzhiyun 
1043*4882a593Smuzhiyun 	if (intel_pt_insn->branch == INTEL_PT_BR_NO_BRANCH) {
1044*4882a593Smuzhiyun 		decoder->state.type = INTEL_PT_INSTRUCTION;
1045*4882a593Smuzhiyun 		decoder->state.from_ip = decoder->ip;
1046*4882a593Smuzhiyun 		decoder->state.to_ip = 0;
1047*4882a593Smuzhiyun 		decoder->ip += intel_pt_insn->length;
1048*4882a593Smuzhiyun 		err = INTEL_PT_RETURN;
1049*4882a593Smuzhiyun 		goto out;
1050*4882a593Smuzhiyun 	}
1051*4882a593Smuzhiyun 
1052*4882a593Smuzhiyun 	if (intel_pt_insn->op == INTEL_PT_OP_CALL) {
1053*4882a593Smuzhiyun 		/* Zero-length calls are excluded */
1054*4882a593Smuzhiyun 		if (intel_pt_insn->branch != INTEL_PT_BR_UNCONDITIONAL ||
1055*4882a593Smuzhiyun 		    intel_pt_insn->rel) {
1056*4882a593Smuzhiyun 			err = intel_pt_push(&decoder->stack, decoder->ip +
1057*4882a593Smuzhiyun 					    intel_pt_insn->length);
1058*4882a593Smuzhiyun 			if (err)
1059*4882a593Smuzhiyun 				goto out;
1060*4882a593Smuzhiyun 		}
1061*4882a593Smuzhiyun 	} else if (intel_pt_insn->op == INTEL_PT_OP_RET) {
1062*4882a593Smuzhiyun 		decoder->ret_addr = intel_pt_pop(&decoder->stack);
1063*4882a593Smuzhiyun 	}
1064*4882a593Smuzhiyun 
1065*4882a593Smuzhiyun 	if (intel_pt_insn->branch == INTEL_PT_BR_UNCONDITIONAL) {
1066*4882a593Smuzhiyun 		int cnt = decoder->no_progress++;
1067*4882a593Smuzhiyun 
1068*4882a593Smuzhiyun 		decoder->state.from_ip = decoder->ip;
1069*4882a593Smuzhiyun 		decoder->ip += intel_pt_insn->length +
1070*4882a593Smuzhiyun 				intel_pt_insn->rel;
1071*4882a593Smuzhiyun 		decoder->state.to_ip = decoder->ip;
1072*4882a593Smuzhiyun 		err = INTEL_PT_RETURN;
1073*4882a593Smuzhiyun 
1074*4882a593Smuzhiyun 		/*
1075*4882a593Smuzhiyun 		 * Check for being stuck in a loop.  This can happen if a
1076*4882a593Smuzhiyun 		 * decoder error results in the decoder erroneously setting the
1077*4882a593Smuzhiyun 		 * ip to an address that is itself in an infinite loop that
1078*4882a593Smuzhiyun 		 * consumes no packets.  When that happens, there must be an
1079*4882a593Smuzhiyun 		 * unconditional branch.
1080*4882a593Smuzhiyun 		 */
1081*4882a593Smuzhiyun 		if (cnt) {
1082*4882a593Smuzhiyun 			if (cnt == 1) {
1083*4882a593Smuzhiyun 				decoder->stuck_ip = decoder->state.to_ip;
1084*4882a593Smuzhiyun 				decoder->stuck_ip_prd = 1;
1085*4882a593Smuzhiyun 				decoder->stuck_ip_cnt = 1;
1086*4882a593Smuzhiyun 			} else if (cnt > INTEL_PT_MAX_LOOPS ||
1087*4882a593Smuzhiyun 				   decoder->state.to_ip == decoder->stuck_ip) {
1088*4882a593Smuzhiyun 				intel_pt_log_at("ERROR: Never-ending loop",
1089*4882a593Smuzhiyun 						decoder->state.to_ip);
1090*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1091*4882a593Smuzhiyun 				err = -ELOOP;
1092*4882a593Smuzhiyun 				goto out;
1093*4882a593Smuzhiyun 			} else if (!--decoder->stuck_ip_cnt) {
1094*4882a593Smuzhiyun 				decoder->stuck_ip_prd += 1;
1095*4882a593Smuzhiyun 				decoder->stuck_ip_cnt = decoder->stuck_ip_prd;
1096*4882a593Smuzhiyun 				decoder->stuck_ip = decoder->state.to_ip;
1097*4882a593Smuzhiyun 			}
1098*4882a593Smuzhiyun 		}
1099*4882a593Smuzhiyun 		goto out_no_progress;
1100*4882a593Smuzhiyun 	}
1101*4882a593Smuzhiyun out:
1102*4882a593Smuzhiyun 	decoder->no_progress = 0;
1103*4882a593Smuzhiyun out_no_progress:
1104*4882a593Smuzhiyun 	decoder->state.insn_op = intel_pt_insn->op;
1105*4882a593Smuzhiyun 	decoder->state.insn_len = intel_pt_insn->length;
1106*4882a593Smuzhiyun 	memcpy(decoder->state.insn, intel_pt_insn->buf,
1107*4882a593Smuzhiyun 	       INTEL_PT_INSN_BUF_SZ);
1108*4882a593Smuzhiyun 
1109*4882a593Smuzhiyun 	if (decoder->tx_flags & INTEL_PT_IN_TX)
1110*4882a593Smuzhiyun 		decoder->state.flags |= INTEL_PT_IN_TX;
1111*4882a593Smuzhiyun 
1112*4882a593Smuzhiyun 	return err;
1113*4882a593Smuzhiyun }
1114*4882a593Smuzhiyun 
intel_pt_fup_event(struct intel_pt_decoder * decoder)1115*4882a593Smuzhiyun static bool intel_pt_fup_event(struct intel_pt_decoder *decoder)
1116*4882a593Smuzhiyun {
1117*4882a593Smuzhiyun 	enum intel_pt_sample_type type = decoder->state.type;
1118*4882a593Smuzhiyun 	bool ret = false;
1119*4882a593Smuzhiyun 
1120*4882a593Smuzhiyun 	decoder->state.type &= ~INTEL_PT_BRANCH;
1121*4882a593Smuzhiyun 
1122*4882a593Smuzhiyun 	if (decoder->set_fup_tx_flags) {
1123*4882a593Smuzhiyun 		decoder->set_fup_tx_flags = false;
1124*4882a593Smuzhiyun 		decoder->tx_flags = decoder->fup_tx_flags;
1125*4882a593Smuzhiyun 		decoder->state.type |= INTEL_PT_TRANSACTION;
1126*4882a593Smuzhiyun 		if (decoder->fup_tx_flags & INTEL_PT_ABORT_TX)
1127*4882a593Smuzhiyun 			decoder->state.type |= INTEL_PT_BRANCH;
1128*4882a593Smuzhiyun 		decoder->state.flags = decoder->fup_tx_flags;
1129*4882a593Smuzhiyun 		ret = true;
1130*4882a593Smuzhiyun 	}
1131*4882a593Smuzhiyun 	if (decoder->set_fup_ptw) {
1132*4882a593Smuzhiyun 		decoder->set_fup_ptw = false;
1133*4882a593Smuzhiyun 		decoder->state.type |= INTEL_PT_PTW;
1134*4882a593Smuzhiyun 		decoder->state.flags |= INTEL_PT_FUP_IP;
1135*4882a593Smuzhiyun 		decoder->state.ptw_payload = decoder->fup_ptw_payload;
1136*4882a593Smuzhiyun 		ret = true;
1137*4882a593Smuzhiyun 	}
1138*4882a593Smuzhiyun 	if (decoder->set_fup_mwait) {
1139*4882a593Smuzhiyun 		decoder->set_fup_mwait = false;
1140*4882a593Smuzhiyun 		decoder->state.type |= INTEL_PT_MWAIT_OP;
1141*4882a593Smuzhiyun 		decoder->state.mwait_payload = decoder->fup_mwait_payload;
1142*4882a593Smuzhiyun 		ret = true;
1143*4882a593Smuzhiyun 	}
1144*4882a593Smuzhiyun 	if (decoder->set_fup_pwre) {
1145*4882a593Smuzhiyun 		decoder->set_fup_pwre = false;
1146*4882a593Smuzhiyun 		decoder->state.type |= INTEL_PT_PWR_ENTRY;
1147*4882a593Smuzhiyun 		decoder->state.pwre_payload = decoder->fup_pwre_payload;
1148*4882a593Smuzhiyun 		ret = true;
1149*4882a593Smuzhiyun 	}
1150*4882a593Smuzhiyun 	if (decoder->set_fup_exstop) {
1151*4882a593Smuzhiyun 		decoder->set_fup_exstop = false;
1152*4882a593Smuzhiyun 		decoder->state.type |= INTEL_PT_EX_STOP;
1153*4882a593Smuzhiyun 		decoder->state.flags |= INTEL_PT_FUP_IP;
1154*4882a593Smuzhiyun 		ret = true;
1155*4882a593Smuzhiyun 	}
1156*4882a593Smuzhiyun 	if (decoder->set_fup_bep) {
1157*4882a593Smuzhiyun 		decoder->set_fup_bep = false;
1158*4882a593Smuzhiyun 		decoder->state.type |= INTEL_PT_BLK_ITEMS;
1159*4882a593Smuzhiyun 		ret = true;
1160*4882a593Smuzhiyun 	}
1161*4882a593Smuzhiyun 	if (decoder->overflow) {
1162*4882a593Smuzhiyun 		decoder->overflow = false;
1163*4882a593Smuzhiyun 		if (!ret && !decoder->pge) {
1164*4882a593Smuzhiyun 			if (decoder->hop) {
1165*4882a593Smuzhiyun 				decoder->state.type = 0;
1166*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
1167*4882a593Smuzhiyun 			}
1168*4882a593Smuzhiyun 			decoder->pge = true;
1169*4882a593Smuzhiyun 			decoder->state.type |= INTEL_PT_BRANCH | INTEL_PT_TRACE_BEGIN;
1170*4882a593Smuzhiyun 			decoder->state.from_ip = 0;
1171*4882a593Smuzhiyun 			decoder->state.to_ip = decoder->ip;
1172*4882a593Smuzhiyun 			return true;
1173*4882a593Smuzhiyun 		}
1174*4882a593Smuzhiyun 	}
1175*4882a593Smuzhiyun 	if (ret) {
1176*4882a593Smuzhiyun 		decoder->state.from_ip = decoder->ip;
1177*4882a593Smuzhiyun 		decoder->state.to_ip = 0;
1178*4882a593Smuzhiyun 	} else {
1179*4882a593Smuzhiyun 		decoder->state.type = type;
1180*4882a593Smuzhiyun 	}
1181*4882a593Smuzhiyun 	return ret;
1182*4882a593Smuzhiyun }
1183*4882a593Smuzhiyun 
intel_pt_fup_with_nlip(struct intel_pt_decoder * decoder,struct intel_pt_insn * intel_pt_insn,uint64_t ip,int err)1184*4882a593Smuzhiyun static inline bool intel_pt_fup_with_nlip(struct intel_pt_decoder *decoder,
1185*4882a593Smuzhiyun 					  struct intel_pt_insn *intel_pt_insn,
1186*4882a593Smuzhiyun 					  uint64_t ip, int err)
1187*4882a593Smuzhiyun {
1188*4882a593Smuzhiyun 	return decoder->flags & INTEL_PT_FUP_WITH_NLIP && !err &&
1189*4882a593Smuzhiyun 	       intel_pt_insn->branch == INTEL_PT_BR_INDIRECT &&
1190*4882a593Smuzhiyun 	       ip == decoder->ip + intel_pt_insn->length;
1191*4882a593Smuzhiyun }
1192*4882a593Smuzhiyun 
intel_pt_walk_fup(struct intel_pt_decoder * decoder)1193*4882a593Smuzhiyun static int intel_pt_walk_fup(struct intel_pt_decoder *decoder)
1194*4882a593Smuzhiyun {
1195*4882a593Smuzhiyun 	struct intel_pt_insn intel_pt_insn;
1196*4882a593Smuzhiyun 	uint64_t ip;
1197*4882a593Smuzhiyun 	int err;
1198*4882a593Smuzhiyun 
1199*4882a593Smuzhiyun 	ip = decoder->last_ip;
1200*4882a593Smuzhiyun 
1201*4882a593Smuzhiyun 	while (1) {
1202*4882a593Smuzhiyun 		err = intel_pt_walk_insn(decoder, &intel_pt_insn, ip);
1203*4882a593Smuzhiyun 		if (err == INTEL_PT_RETURN)
1204*4882a593Smuzhiyun 			return 0;
1205*4882a593Smuzhiyun 		if (err == -EAGAIN ||
1206*4882a593Smuzhiyun 		    intel_pt_fup_with_nlip(decoder, &intel_pt_insn, ip, err)) {
1207*4882a593Smuzhiyun 			bool no_tip = decoder->pkt_state != INTEL_PT_STATE_FUP;
1208*4882a593Smuzhiyun 
1209*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1210*4882a593Smuzhiyun 			if (intel_pt_fup_event(decoder) && no_tip)
1211*4882a593Smuzhiyun 				return 0;
1212*4882a593Smuzhiyun 			return -EAGAIN;
1213*4882a593Smuzhiyun 		}
1214*4882a593Smuzhiyun 		decoder->set_fup_tx_flags = false;
1215*4882a593Smuzhiyun 		if (err)
1216*4882a593Smuzhiyun 			return err;
1217*4882a593Smuzhiyun 
1218*4882a593Smuzhiyun 		if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1219*4882a593Smuzhiyun 			intel_pt_log_at("ERROR: Unexpected indirect branch",
1220*4882a593Smuzhiyun 					decoder->ip);
1221*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1222*4882a593Smuzhiyun 			return -ENOENT;
1223*4882a593Smuzhiyun 		}
1224*4882a593Smuzhiyun 
1225*4882a593Smuzhiyun 		if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1226*4882a593Smuzhiyun 			intel_pt_log_at("ERROR: Unexpected conditional branch",
1227*4882a593Smuzhiyun 					decoder->ip);
1228*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1229*4882a593Smuzhiyun 			return -ENOENT;
1230*4882a593Smuzhiyun 		}
1231*4882a593Smuzhiyun 
1232*4882a593Smuzhiyun 		intel_pt_bug(decoder);
1233*4882a593Smuzhiyun 	}
1234*4882a593Smuzhiyun }
1235*4882a593Smuzhiyun 
intel_pt_walk_tip(struct intel_pt_decoder * decoder)1236*4882a593Smuzhiyun static int intel_pt_walk_tip(struct intel_pt_decoder *decoder)
1237*4882a593Smuzhiyun {
1238*4882a593Smuzhiyun 	struct intel_pt_insn intel_pt_insn;
1239*4882a593Smuzhiyun 	int err;
1240*4882a593Smuzhiyun 
1241*4882a593Smuzhiyun 	err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1242*4882a593Smuzhiyun 	if (err == INTEL_PT_RETURN &&
1243*4882a593Smuzhiyun 	    decoder->pgd_ip &&
1244*4882a593Smuzhiyun 	    decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1245*4882a593Smuzhiyun 	    (decoder->state.type & INTEL_PT_BRANCH) &&
1246*4882a593Smuzhiyun 	    decoder->pgd_ip(decoder->state.to_ip, decoder->data)) {
1247*4882a593Smuzhiyun 		/* Unconditional branch leaving filter region */
1248*4882a593Smuzhiyun 		decoder->no_progress = 0;
1249*4882a593Smuzhiyun 		decoder->pge = false;
1250*4882a593Smuzhiyun 		decoder->continuous_period = false;
1251*4882a593Smuzhiyun 		decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1252*4882a593Smuzhiyun 		decoder->state.type |= INTEL_PT_TRACE_END;
1253*4882a593Smuzhiyun 		return 0;
1254*4882a593Smuzhiyun 	}
1255*4882a593Smuzhiyun 	if (err == INTEL_PT_RETURN)
1256*4882a593Smuzhiyun 		return 0;
1257*4882a593Smuzhiyun 	if (err)
1258*4882a593Smuzhiyun 		return err;
1259*4882a593Smuzhiyun 
1260*4882a593Smuzhiyun 	if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1261*4882a593Smuzhiyun 		if (decoder->pkt_state == INTEL_PT_STATE_TIP_PGD) {
1262*4882a593Smuzhiyun 			decoder->pge = false;
1263*4882a593Smuzhiyun 			decoder->continuous_period = false;
1264*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1265*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
1266*4882a593Smuzhiyun 			if (decoder->packet.count == 0) {
1267*4882a593Smuzhiyun 				decoder->state.to_ip = 0;
1268*4882a593Smuzhiyun 			} else {
1269*4882a593Smuzhiyun 				decoder->state.to_ip = decoder->last_ip;
1270*4882a593Smuzhiyun 				decoder->ip = decoder->last_ip;
1271*4882a593Smuzhiyun 			}
1272*4882a593Smuzhiyun 			decoder->state.type |= INTEL_PT_TRACE_END;
1273*4882a593Smuzhiyun 		} else {
1274*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1275*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
1276*4882a593Smuzhiyun 			if (decoder->packet.count == 0) {
1277*4882a593Smuzhiyun 				decoder->state.to_ip = 0;
1278*4882a593Smuzhiyun 			} else {
1279*4882a593Smuzhiyun 				decoder->state.to_ip = decoder->last_ip;
1280*4882a593Smuzhiyun 				decoder->ip = decoder->last_ip;
1281*4882a593Smuzhiyun 			}
1282*4882a593Smuzhiyun 		}
1283*4882a593Smuzhiyun 		return 0;
1284*4882a593Smuzhiyun 	}
1285*4882a593Smuzhiyun 
1286*4882a593Smuzhiyun 	if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1287*4882a593Smuzhiyun 		uint64_t to_ip = decoder->ip + intel_pt_insn.length +
1288*4882a593Smuzhiyun 				 intel_pt_insn.rel;
1289*4882a593Smuzhiyun 
1290*4882a593Smuzhiyun 		if (decoder->pgd_ip &&
1291*4882a593Smuzhiyun 		    decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1292*4882a593Smuzhiyun 		    decoder->pgd_ip(to_ip, decoder->data)) {
1293*4882a593Smuzhiyun 			/* Conditional branch leaving filter region */
1294*4882a593Smuzhiyun 			decoder->pge = false;
1295*4882a593Smuzhiyun 			decoder->continuous_period = false;
1296*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1297*4882a593Smuzhiyun 			decoder->ip = to_ip;
1298*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
1299*4882a593Smuzhiyun 			decoder->state.to_ip = to_ip;
1300*4882a593Smuzhiyun 			decoder->state.type |= INTEL_PT_TRACE_END;
1301*4882a593Smuzhiyun 			return 0;
1302*4882a593Smuzhiyun 		}
1303*4882a593Smuzhiyun 		intel_pt_log_at("ERROR: Conditional branch when expecting indirect branch",
1304*4882a593Smuzhiyun 				decoder->ip);
1305*4882a593Smuzhiyun 		decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1306*4882a593Smuzhiyun 		return -ENOENT;
1307*4882a593Smuzhiyun 	}
1308*4882a593Smuzhiyun 
1309*4882a593Smuzhiyun 	return intel_pt_bug(decoder);
1310*4882a593Smuzhiyun }
1311*4882a593Smuzhiyun 
intel_pt_walk_tnt(struct intel_pt_decoder * decoder)1312*4882a593Smuzhiyun static int intel_pt_walk_tnt(struct intel_pt_decoder *decoder)
1313*4882a593Smuzhiyun {
1314*4882a593Smuzhiyun 	struct intel_pt_insn intel_pt_insn;
1315*4882a593Smuzhiyun 	int err;
1316*4882a593Smuzhiyun 
1317*4882a593Smuzhiyun 	while (1) {
1318*4882a593Smuzhiyun 		err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1319*4882a593Smuzhiyun 		if (err == INTEL_PT_RETURN)
1320*4882a593Smuzhiyun 			return 0;
1321*4882a593Smuzhiyun 		if (err)
1322*4882a593Smuzhiyun 			return err;
1323*4882a593Smuzhiyun 
1324*4882a593Smuzhiyun 		if (intel_pt_insn.op == INTEL_PT_OP_RET) {
1325*4882a593Smuzhiyun 			if (!decoder->return_compression) {
1326*4882a593Smuzhiyun 				intel_pt_log_at("ERROR: RET when expecting conditional branch",
1327*4882a593Smuzhiyun 						decoder->ip);
1328*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_ERR3;
1329*4882a593Smuzhiyun 				return -ENOENT;
1330*4882a593Smuzhiyun 			}
1331*4882a593Smuzhiyun 			if (!decoder->ret_addr) {
1332*4882a593Smuzhiyun 				intel_pt_log_at("ERROR: Bad RET compression (stack empty)",
1333*4882a593Smuzhiyun 						decoder->ip);
1334*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_ERR3;
1335*4882a593Smuzhiyun 				return -ENOENT;
1336*4882a593Smuzhiyun 			}
1337*4882a593Smuzhiyun 			if (!(decoder->tnt.payload & BIT63)) {
1338*4882a593Smuzhiyun 				intel_pt_log_at("ERROR: Bad RET compression (TNT=N)",
1339*4882a593Smuzhiyun 						decoder->ip);
1340*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_ERR3;
1341*4882a593Smuzhiyun 				return -ENOENT;
1342*4882a593Smuzhiyun 			}
1343*4882a593Smuzhiyun 			decoder->tnt.count -= 1;
1344*4882a593Smuzhiyun 			if (decoder->tnt.count)
1345*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_TNT_CONT;
1346*4882a593Smuzhiyun 			else
1347*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1348*4882a593Smuzhiyun 			decoder->tnt.payload <<= 1;
1349*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
1350*4882a593Smuzhiyun 			decoder->ip = decoder->ret_addr;
1351*4882a593Smuzhiyun 			decoder->state.to_ip = decoder->ip;
1352*4882a593Smuzhiyun 			return 0;
1353*4882a593Smuzhiyun 		}
1354*4882a593Smuzhiyun 
1355*4882a593Smuzhiyun 		if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1356*4882a593Smuzhiyun 			/* Handle deferred TIPs */
1357*4882a593Smuzhiyun 			err = intel_pt_get_next_packet(decoder);
1358*4882a593Smuzhiyun 			if (err)
1359*4882a593Smuzhiyun 				return err;
1360*4882a593Smuzhiyun 			if (decoder->packet.type != INTEL_PT_TIP ||
1361*4882a593Smuzhiyun 			    decoder->packet.count == 0) {
1362*4882a593Smuzhiyun 				intel_pt_log_at("ERROR: Missing deferred TIP for indirect branch",
1363*4882a593Smuzhiyun 						decoder->ip);
1364*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_ERR3;
1365*4882a593Smuzhiyun 				decoder->pkt_step = 0;
1366*4882a593Smuzhiyun 				return -ENOENT;
1367*4882a593Smuzhiyun 			}
1368*4882a593Smuzhiyun 			intel_pt_set_last_ip(decoder);
1369*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
1370*4882a593Smuzhiyun 			decoder->state.to_ip = decoder->last_ip;
1371*4882a593Smuzhiyun 			decoder->ip = decoder->last_ip;
1372*4882a593Smuzhiyun 			return 0;
1373*4882a593Smuzhiyun 		}
1374*4882a593Smuzhiyun 
1375*4882a593Smuzhiyun 		if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1376*4882a593Smuzhiyun 			decoder->tnt.count -= 1;
1377*4882a593Smuzhiyun 			if (decoder->tnt.count)
1378*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_TNT_CONT;
1379*4882a593Smuzhiyun 			else
1380*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1381*4882a593Smuzhiyun 			if (decoder->tnt.payload & BIT63) {
1382*4882a593Smuzhiyun 				decoder->tnt.payload <<= 1;
1383*4882a593Smuzhiyun 				decoder->state.from_ip = decoder->ip;
1384*4882a593Smuzhiyun 				decoder->ip += intel_pt_insn.length +
1385*4882a593Smuzhiyun 					       intel_pt_insn.rel;
1386*4882a593Smuzhiyun 				decoder->state.to_ip = decoder->ip;
1387*4882a593Smuzhiyun 				return 0;
1388*4882a593Smuzhiyun 			}
1389*4882a593Smuzhiyun 			/* Instruction sample for a non-taken branch */
1390*4882a593Smuzhiyun 			if (decoder->state.type & INTEL_PT_INSTRUCTION) {
1391*4882a593Smuzhiyun 				decoder->tnt.payload <<= 1;
1392*4882a593Smuzhiyun 				decoder->state.type = INTEL_PT_INSTRUCTION;
1393*4882a593Smuzhiyun 				decoder->state.from_ip = decoder->ip;
1394*4882a593Smuzhiyun 				decoder->state.to_ip = 0;
1395*4882a593Smuzhiyun 				decoder->ip += intel_pt_insn.length;
1396*4882a593Smuzhiyun 				return 0;
1397*4882a593Smuzhiyun 			}
1398*4882a593Smuzhiyun 			decoder->sample_cyc = false;
1399*4882a593Smuzhiyun 			decoder->ip += intel_pt_insn.length;
1400*4882a593Smuzhiyun 			if (!decoder->tnt.count) {
1401*4882a593Smuzhiyun 				intel_pt_update_sample_time(decoder);
1402*4882a593Smuzhiyun 				return -EAGAIN;
1403*4882a593Smuzhiyun 			}
1404*4882a593Smuzhiyun 			decoder->tnt.payload <<= 1;
1405*4882a593Smuzhiyun 			continue;
1406*4882a593Smuzhiyun 		}
1407*4882a593Smuzhiyun 
1408*4882a593Smuzhiyun 		return intel_pt_bug(decoder);
1409*4882a593Smuzhiyun 	}
1410*4882a593Smuzhiyun }
1411*4882a593Smuzhiyun 
intel_pt_mode_tsx(struct intel_pt_decoder * decoder,bool * no_tip)1412*4882a593Smuzhiyun static int intel_pt_mode_tsx(struct intel_pt_decoder *decoder, bool *no_tip)
1413*4882a593Smuzhiyun {
1414*4882a593Smuzhiyun 	unsigned int fup_tx_flags;
1415*4882a593Smuzhiyun 	int err;
1416*4882a593Smuzhiyun 
1417*4882a593Smuzhiyun 	fup_tx_flags = decoder->packet.payload &
1418*4882a593Smuzhiyun 		       (INTEL_PT_IN_TX | INTEL_PT_ABORT_TX);
1419*4882a593Smuzhiyun 	err = intel_pt_get_next_packet(decoder);
1420*4882a593Smuzhiyun 	if (err)
1421*4882a593Smuzhiyun 		return err;
1422*4882a593Smuzhiyun 	if (decoder->packet.type == INTEL_PT_FUP) {
1423*4882a593Smuzhiyun 		decoder->fup_tx_flags = fup_tx_flags;
1424*4882a593Smuzhiyun 		decoder->set_fup_tx_flags = true;
1425*4882a593Smuzhiyun 		if (!(decoder->fup_tx_flags & INTEL_PT_ABORT_TX))
1426*4882a593Smuzhiyun 			*no_tip = true;
1427*4882a593Smuzhiyun 	} else {
1428*4882a593Smuzhiyun 		intel_pt_log_at("ERROR: Missing FUP after MODE.TSX",
1429*4882a593Smuzhiyun 				decoder->pos);
1430*4882a593Smuzhiyun 		intel_pt_update_in_tx(decoder);
1431*4882a593Smuzhiyun 	}
1432*4882a593Smuzhiyun 	return 0;
1433*4882a593Smuzhiyun }
1434*4882a593Smuzhiyun 
intel_pt_8b_tsc(uint64_t timestamp,uint64_t ref_timestamp)1435*4882a593Smuzhiyun static uint64_t intel_pt_8b_tsc(uint64_t timestamp, uint64_t ref_timestamp)
1436*4882a593Smuzhiyun {
1437*4882a593Smuzhiyun 	timestamp |= (ref_timestamp & (0xffULL << 56));
1438*4882a593Smuzhiyun 
1439*4882a593Smuzhiyun 	if (timestamp < ref_timestamp) {
1440*4882a593Smuzhiyun 		if (ref_timestamp - timestamp > (1ULL << 55))
1441*4882a593Smuzhiyun 			timestamp += (1ULL << 56);
1442*4882a593Smuzhiyun 	} else {
1443*4882a593Smuzhiyun 		if (timestamp - ref_timestamp > (1ULL << 55))
1444*4882a593Smuzhiyun 			timestamp -= (1ULL << 56);
1445*4882a593Smuzhiyun 	}
1446*4882a593Smuzhiyun 
1447*4882a593Smuzhiyun 	return timestamp;
1448*4882a593Smuzhiyun }
1449*4882a593Smuzhiyun 
intel_pt_calc_tsc_timestamp(struct intel_pt_decoder * decoder)1450*4882a593Smuzhiyun static void intel_pt_calc_tsc_timestamp(struct intel_pt_decoder *decoder)
1451*4882a593Smuzhiyun {
1452*4882a593Smuzhiyun 	uint64_t timestamp;
1453*4882a593Smuzhiyun 
1454*4882a593Smuzhiyun 	decoder->have_tma = false;
1455*4882a593Smuzhiyun 
1456*4882a593Smuzhiyun 	if (decoder->ref_timestamp) {
1457*4882a593Smuzhiyun 		timestamp = intel_pt_8b_tsc(decoder->packet.payload,
1458*4882a593Smuzhiyun 					    decoder->ref_timestamp);
1459*4882a593Smuzhiyun 		decoder->tsc_timestamp = timestamp;
1460*4882a593Smuzhiyun 		decoder->timestamp = timestamp;
1461*4882a593Smuzhiyun 		decoder->ref_timestamp = 0;
1462*4882a593Smuzhiyun 		decoder->timestamp_insn_cnt = 0;
1463*4882a593Smuzhiyun 	} else if (decoder->timestamp) {
1464*4882a593Smuzhiyun 		timestamp = decoder->packet.payload |
1465*4882a593Smuzhiyun 			    (decoder->timestamp & (0xffULL << 56));
1466*4882a593Smuzhiyun 		decoder->tsc_timestamp = timestamp;
1467*4882a593Smuzhiyun 		if (timestamp < decoder->timestamp &&
1468*4882a593Smuzhiyun 		    decoder->timestamp - timestamp < decoder->tsc_slip) {
1469*4882a593Smuzhiyun 			intel_pt_log_to("Suppressing backwards timestamp",
1470*4882a593Smuzhiyun 					timestamp);
1471*4882a593Smuzhiyun 			timestamp = decoder->timestamp;
1472*4882a593Smuzhiyun 		}
1473*4882a593Smuzhiyun 		if (timestamp < decoder->timestamp) {
1474*4882a593Smuzhiyun 			intel_pt_log_to("Wraparound timestamp", timestamp);
1475*4882a593Smuzhiyun 			timestamp += (1ULL << 56);
1476*4882a593Smuzhiyun 			decoder->tsc_timestamp = timestamp;
1477*4882a593Smuzhiyun 		}
1478*4882a593Smuzhiyun 		decoder->timestamp = timestamp;
1479*4882a593Smuzhiyun 		decoder->timestamp_insn_cnt = 0;
1480*4882a593Smuzhiyun 	}
1481*4882a593Smuzhiyun 
1482*4882a593Smuzhiyun 	if (decoder->last_packet_type == INTEL_PT_CYC) {
1483*4882a593Smuzhiyun 		decoder->cyc_ref_timestamp = decoder->timestamp;
1484*4882a593Smuzhiyun 		decoder->cycle_cnt = 0;
1485*4882a593Smuzhiyun 		decoder->have_calc_cyc_to_tsc = false;
1486*4882a593Smuzhiyun 		intel_pt_calc_cyc_to_tsc(decoder, false);
1487*4882a593Smuzhiyun 	}
1488*4882a593Smuzhiyun 
1489*4882a593Smuzhiyun 	intel_pt_log_to("Setting timestamp", decoder->timestamp);
1490*4882a593Smuzhiyun }
1491*4882a593Smuzhiyun 
intel_pt_overflow(struct intel_pt_decoder * decoder)1492*4882a593Smuzhiyun static int intel_pt_overflow(struct intel_pt_decoder *decoder)
1493*4882a593Smuzhiyun {
1494*4882a593Smuzhiyun 	intel_pt_log("ERROR: Buffer overflow\n");
1495*4882a593Smuzhiyun 	intel_pt_clear_tx_flags(decoder);
1496*4882a593Smuzhiyun 	decoder->timestamp_insn_cnt = 0;
1497*4882a593Smuzhiyun 	decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1498*4882a593Smuzhiyun 	decoder->state.from_ip = decoder->ip;
1499*4882a593Smuzhiyun 	decoder->ip = 0;
1500*4882a593Smuzhiyun 	decoder->pge = false;
1501*4882a593Smuzhiyun 	decoder->set_fup_tx_flags = false;
1502*4882a593Smuzhiyun 	decoder->set_fup_ptw = false;
1503*4882a593Smuzhiyun 	decoder->set_fup_mwait = false;
1504*4882a593Smuzhiyun 	decoder->set_fup_pwre = false;
1505*4882a593Smuzhiyun 	decoder->set_fup_exstop = false;
1506*4882a593Smuzhiyun 	decoder->set_fup_bep = false;
1507*4882a593Smuzhiyun 	decoder->overflow = true;
1508*4882a593Smuzhiyun 	return -EOVERFLOW;
1509*4882a593Smuzhiyun }
1510*4882a593Smuzhiyun 
intel_pt_mtc_cyc_cnt_pge(struct intel_pt_decoder * decoder)1511*4882a593Smuzhiyun static inline void intel_pt_mtc_cyc_cnt_pge(struct intel_pt_decoder *decoder)
1512*4882a593Smuzhiyun {
1513*4882a593Smuzhiyun 	if (decoder->have_cyc)
1514*4882a593Smuzhiyun 		return;
1515*4882a593Smuzhiyun 
1516*4882a593Smuzhiyun 	decoder->cyc_cnt_timestamp = decoder->timestamp;
1517*4882a593Smuzhiyun 	decoder->base_cyc_cnt = decoder->tot_cyc_cnt;
1518*4882a593Smuzhiyun }
1519*4882a593Smuzhiyun 
intel_pt_mtc_cyc_cnt_cbr(struct intel_pt_decoder * decoder)1520*4882a593Smuzhiyun static inline void intel_pt_mtc_cyc_cnt_cbr(struct intel_pt_decoder *decoder)
1521*4882a593Smuzhiyun {
1522*4882a593Smuzhiyun 	decoder->tsc_to_cyc = decoder->cbr / decoder->max_non_turbo_ratio_fp;
1523*4882a593Smuzhiyun 
1524*4882a593Smuzhiyun 	if (decoder->pge)
1525*4882a593Smuzhiyun 		intel_pt_mtc_cyc_cnt_pge(decoder);
1526*4882a593Smuzhiyun }
1527*4882a593Smuzhiyun 
intel_pt_mtc_cyc_cnt_upd(struct intel_pt_decoder * decoder)1528*4882a593Smuzhiyun static inline void intel_pt_mtc_cyc_cnt_upd(struct intel_pt_decoder *decoder)
1529*4882a593Smuzhiyun {
1530*4882a593Smuzhiyun 	uint64_t tot_cyc_cnt, tsc_delta;
1531*4882a593Smuzhiyun 
1532*4882a593Smuzhiyun 	if (decoder->have_cyc)
1533*4882a593Smuzhiyun 		return;
1534*4882a593Smuzhiyun 
1535*4882a593Smuzhiyun 	decoder->sample_cyc = true;
1536*4882a593Smuzhiyun 
1537*4882a593Smuzhiyun 	if (!decoder->pge || decoder->timestamp <= decoder->cyc_cnt_timestamp)
1538*4882a593Smuzhiyun 		return;
1539*4882a593Smuzhiyun 
1540*4882a593Smuzhiyun 	tsc_delta = decoder->timestamp - decoder->cyc_cnt_timestamp;
1541*4882a593Smuzhiyun 	tot_cyc_cnt = tsc_delta * decoder->tsc_to_cyc + decoder->base_cyc_cnt;
1542*4882a593Smuzhiyun 
1543*4882a593Smuzhiyun 	if (tot_cyc_cnt > decoder->tot_cyc_cnt)
1544*4882a593Smuzhiyun 		decoder->tot_cyc_cnt = tot_cyc_cnt;
1545*4882a593Smuzhiyun }
1546*4882a593Smuzhiyun 
intel_pt_calc_tma(struct intel_pt_decoder * decoder)1547*4882a593Smuzhiyun static void intel_pt_calc_tma(struct intel_pt_decoder *decoder)
1548*4882a593Smuzhiyun {
1549*4882a593Smuzhiyun 	uint32_t ctc = decoder->packet.payload;
1550*4882a593Smuzhiyun 	uint32_t fc = decoder->packet.count;
1551*4882a593Smuzhiyun 	uint32_t ctc_rem = ctc & decoder->ctc_rem_mask;
1552*4882a593Smuzhiyun 
1553*4882a593Smuzhiyun 	if (!decoder->tsc_ctc_ratio_d)
1554*4882a593Smuzhiyun 		return;
1555*4882a593Smuzhiyun 
1556*4882a593Smuzhiyun 	if (decoder->pge && !decoder->in_psb)
1557*4882a593Smuzhiyun 		intel_pt_mtc_cyc_cnt_pge(decoder);
1558*4882a593Smuzhiyun 	else
1559*4882a593Smuzhiyun 		intel_pt_mtc_cyc_cnt_upd(decoder);
1560*4882a593Smuzhiyun 
1561*4882a593Smuzhiyun 	decoder->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
1562*4882a593Smuzhiyun 	decoder->ctc_timestamp = decoder->tsc_timestamp - fc;
1563*4882a593Smuzhiyun 	if (decoder->tsc_ctc_mult) {
1564*4882a593Smuzhiyun 		decoder->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
1565*4882a593Smuzhiyun 	} else {
1566*4882a593Smuzhiyun 		decoder->ctc_timestamp -= multdiv(ctc_rem,
1567*4882a593Smuzhiyun 						  decoder->tsc_ctc_ratio_n,
1568*4882a593Smuzhiyun 						  decoder->tsc_ctc_ratio_d);
1569*4882a593Smuzhiyun 	}
1570*4882a593Smuzhiyun 	decoder->ctc_delta = 0;
1571*4882a593Smuzhiyun 	decoder->have_tma = true;
1572*4882a593Smuzhiyun 	decoder->fixup_last_mtc = true;
1573*4882a593Smuzhiyun 	intel_pt_log("CTC timestamp " x64_fmt " last MTC %#x  CTC rem %#x\n",
1574*4882a593Smuzhiyun 		     decoder->ctc_timestamp, decoder->last_mtc, ctc_rem);
1575*4882a593Smuzhiyun }
1576*4882a593Smuzhiyun 
intel_pt_calc_mtc_timestamp(struct intel_pt_decoder * decoder)1577*4882a593Smuzhiyun static void intel_pt_calc_mtc_timestamp(struct intel_pt_decoder *decoder)
1578*4882a593Smuzhiyun {
1579*4882a593Smuzhiyun 	uint64_t timestamp;
1580*4882a593Smuzhiyun 	uint32_t mtc, mtc_delta;
1581*4882a593Smuzhiyun 
1582*4882a593Smuzhiyun 	if (!decoder->have_tma)
1583*4882a593Smuzhiyun 		return;
1584*4882a593Smuzhiyun 
1585*4882a593Smuzhiyun 	mtc = decoder->packet.payload;
1586*4882a593Smuzhiyun 
1587*4882a593Smuzhiyun 	if (decoder->mtc_shift > 8 && decoder->fixup_last_mtc) {
1588*4882a593Smuzhiyun 		decoder->fixup_last_mtc = false;
1589*4882a593Smuzhiyun 		intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
1590*4882a593Smuzhiyun 					&decoder->last_mtc);
1591*4882a593Smuzhiyun 	}
1592*4882a593Smuzhiyun 
1593*4882a593Smuzhiyun 	if (mtc > decoder->last_mtc)
1594*4882a593Smuzhiyun 		mtc_delta = mtc - decoder->last_mtc;
1595*4882a593Smuzhiyun 	else
1596*4882a593Smuzhiyun 		mtc_delta = mtc + 256 - decoder->last_mtc;
1597*4882a593Smuzhiyun 
1598*4882a593Smuzhiyun 	decoder->ctc_delta += mtc_delta << decoder->mtc_shift;
1599*4882a593Smuzhiyun 
1600*4882a593Smuzhiyun 	if (decoder->tsc_ctc_mult) {
1601*4882a593Smuzhiyun 		timestamp = decoder->ctc_timestamp +
1602*4882a593Smuzhiyun 			    decoder->ctc_delta * decoder->tsc_ctc_mult;
1603*4882a593Smuzhiyun 	} else {
1604*4882a593Smuzhiyun 		timestamp = decoder->ctc_timestamp +
1605*4882a593Smuzhiyun 			    multdiv(decoder->ctc_delta,
1606*4882a593Smuzhiyun 				    decoder->tsc_ctc_ratio_n,
1607*4882a593Smuzhiyun 				    decoder->tsc_ctc_ratio_d);
1608*4882a593Smuzhiyun 	}
1609*4882a593Smuzhiyun 
1610*4882a593Smuzhiyun 	if (timestamp < decoder->timestamp)
1611*4882a593Smuzhiyun 		intel_pt_log("Suppressing MTC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1612*4882a593Smuzhiyun 			     timestamp, decoder->timestamp);
1613*4882a593Smuzhiyun 	else
1614*4882a593Smuzhiyun 		decoder->timestamp = timestamp;
1615*4882a593Smuzhiyun 
1616*4882a593Smuzhiyun 	intel_pt_mtc_cyc_cnt_upd(decoder);
1617*4882a593Smuzhiyun 
1618*4882a593Smuzhiyun 	decoder->timestamp_insn_cnt = 0;
1619*4882a593Smuzhiyun 	decoder->last_mtc = mtc;
1620*4882a593Smuzhiyun 
1621*4882a593Smuzhiyun 	if (decoder->last_packet_type == INTEL_PT_CYC) {
1622*4882a593Smuzhiyun 		decoder->cyc_ref_timestamp = decoder->timestamp;
1623*4882a593Smuzhiyun 		decoder->cycle_cnt = 0;
1624*4882a593Smuzhiyun 		decoder->have_calc_cyc_to_tsc = false;
1625*4882a593Smuzhiyun 		intel_pt_calc_cyc_to_tsc(decoder, true);
1626*4882a593Smuzhiyun 	}
1627*4882a593Smuzhiyun 
1628*4882a593Smuzhiyun 	intel_pt_log_to("Setting timestamp", decoder->timestamp);
1629*4882a593Smuzhiyun }
1630*4882a593Smuzhiyun 
intel_pt_calc_cbr(struct intel_pt_decoder * decoder)1631*4882a593Smuzhiyun static void intel_pt_calc_cbr(struct intel_pt_decoder *decoder)
1632*4882a593Smuzhiyun {
1633*4882a593Smuzhiyun 	unsigned int cbr = decoder->packet.payload & 0xff;
1634*4882a593Smuzhiyun 
1635*4882a593Smuzhiyun 	decoder->cbr_payload = decoder->packet.payload;
1636*4882a593Smuzhiyun 
1637*4882a593Smuzhiyun 	if (decoder->cbr == cbr)
1638*4882a593Smuzhiyun 		return;
1639*4882a593Smuzhiyun 
1640*4882a593Smuzhiyun 	decoder->cbr = cbr;
1641*4882a593Smuzhiyun 	decoder->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
1642*4882a593Smuzhiyun 
1643*4882a593Smuzhiyun 	intel_pt_mtc_cyc_cnt_cbr(decoder);
1644*4882a593Smuzhiyun }
1645*4882a593Smuzhiyun 
intel_pt_calc_cyc_timestamp(struct intel_pt_decoder * decoder)1646*4882a593Smuzhiyun static void intel_pt_calc_cyc_timestamp(struct intel_pt_decoder *decoder)
1647*4882a593Smuzhiyun {
1648*4882a593Smuzhiyun 	uint64_t timestamp = decoder->cyc_ref_timestamp;
1649*4882a593Smuzhiyun 
1650*4882a593Smuzhiyun 	decoder->have_cyc = true;
1651*4882a593Smuzhiyun 
1652*4882a593Smuzhiyun 	decoder->cycle_cnt += decoder->packet.payload;
1653*4882a593Smuzhiyun 	if (decoder->pge)
1654*4882a593Smuzhiyun 		decoder->tot_cyc_cnt += decoder->packet.payload;
1655*4882a593Smuzhiyun 	decoder->sample_cyc = true;
1656*4882a593Smuzhiyun 
1657*4882a593Smuzhiyun 	if (!decoder->cyc_ref_timestamp)
1658*4882a593Smuzhiyun 		return;
1659*4882a593Smuzhiyun 
1660*4882a593Smuzhiyun 	if (decoder->have_calc_cyc_to_tsc)
1661*4882a593Smuzhiyun 		timestamp += decoder->cycle_cnt * decoder->calc_cyc_to_tsc;
1662*4882a593Smuzhiyun 	else if (decoder->cbr)
1663*4882a593Smuzhiyun 		timestamp += decoder->cycle_cnt * decoder->cbr_cyc_to_tsc;
1664*4882a593Smuzhiyun 	else
1665*4882a593Smuzhiyun 		return;
1666*4882a593Smuzhiyun 
1667*4882a593Smuzhiyun 	if (timestamp < decoder->timestamp)
1668*4882a593Smuzhiyun 		intel_pt_log("Suppressing CYC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1669*4882a593Smuzhiyun 			     timestamp, decoder->timestamp);
1670*4882a593Smuzhiyun 	else
1671*4882a593Smuzhiyun 		decoder->timestamp = timestamp;
1672*4882a593Smuzhiyun 
1673*4882a593Smuzhiyun 	decoder->timestamp_insn_cnt = 0;
1674*4882a593Smuzhiyun 
1675*4882a593Smuzhiyun 	intel_pt_log_to("Setting timestamp", decoder->timestamp);
1676*4882a593Smuzhiyun }
1677*4882a593Smuzhiyun 
intel_pt_bbp(struct intel_pt_decoder * decoder)1678*4882a593Smuzhiyun static void intel_pt_bbp(struct intel_pt_decoder *decoder)
1679*4882a593Smuzhiyun {
1680*4882a593Smuzhiyun 	if (decoder->prev_pkt_ctx == INTEL_PT_NO_CTX) {
1681*4882a593Smuzhiyun 		memset(decoder->state.items.mask, 0, sizeof(decoder->state.items.mask));
1682*4882a593Smuzhiyun 		decoder->state.items.is_32_bit = false;
1683*4882a593Smuzhiyun 	}
1684*4882a593Smuzhiyun 	decoder->blk_type = decoder->packet.payload;
1685*4882a593Smuzhiyun 	decoder->blk_type_pos = intel_pt_blk_type_pos(decoder->blk_type);
1686*4882a593Smuzhiyun 	if (decoder->blk_type == INTEL_PT_GP_REGS)
1687*4882a593Smuzhiyun 		decoder->state.items.is_32_bit = decoder->packet.count;
1688*4882a593Smuzhiyun 	if (decoder->blk_type_pos < 0) {
1689*4882a593Smuzhiyun 		intel_pt_log("WARNING: Unknown block type %u\n",
1690*4882a593Smuzhiyun 			     decoder->blk_type);
1691*4882a593Smuzhiyun 	} else if (decoder->state.items.mask[decoder->blk_type_pos]) {
1692*4882a593Smuzhiyun 		intel_pt_log("WARNING: Duplicate block type %u\n",
1693*4882a593Smuzhiyun 			     decoder->blk_type);
1694*4882a593Smuzhiyun 	}
1695*4882a593Smuzhiyun }
1696*4882a593Smuzhiyun 
intel_pt_bip(struct intel_pt_decoder * decoder)1697*4882a593Smuzhiyun static void intel_pt_bip(struct intel_pt_decoder *decoder)
1698*4882a593Smuzhiyun {
1699*4882a593Smuzhiyun 	uint32_t id = decoder->packet.count;
1700*4882a593Smuzhiyun 	uint32_t bit = 1 << id;
1701*4882a593Smuzhiyun 	int pos = decoder->blk_type_pos;
1702*4882a593Smuzhiyun 
1703*4882a593Smuzhiyun 	if (pos < 0 || id >= INTEL_PT_BLK_ITEM_ID_CNT) {
1704*4882a593Smuzhiyun 		intel_pt_log("WARNING: Unknown block item %u type %d\n",
1705*4882a593Smuzhiyun 			     id, decoder->blk_type);
1706*4882a593Smuzhiyun 		return;
1707*4882a593Smuzhiyun 	}
1708*4882a593Smuzhiyun 
1709*4882a593Smuzhiyun 	if (decoder->state.items.mask[pos] & bit) {
1710*4882a593Smuzhiyun 		intel_pt_log("WARNING: Duplicate block item %u type %d\n",
1711*4882a593Smuzhiyun 			     id, decoder->blk_type);
1712*4882a593Smuzhiyun 	}
1713*4882a593Smuzhiyun 
1714*4882a593Smuzhiyun 	decoder->state.items.mask[pos] |= bit;
1715*4882a593Smuzhiyun 	decoder->state.items.val[pos][id] = decoder->packet.payload;
1716*4882a593Smuzhiyun }
1717*4882a593Smuzhiyun 
1718*4882a593Smuzhiyun /* Walk PSB+ packets when already in sync. */
intel_pt_walk_psbend(struct intel_pt_decoder * decoder)1719*4882a593Smuzhiyun static int intel_pt_walk_psbend(struct intel_pt_decoder *decoder)
1720*4882a593Smuzhiyun {
1721*4882a593Smuzhiyun 	int err;
1722*4882a593Smuzhiyun 
1723*4882a593Smuzhiyun 	decoder->in_psb = true;
1724*4882a593Smuzhiyun 
1725*4882a593Smuzhiyun 	while (1) {
1726*4882a593Smuzhiyun 		err = intel_pt_get_next_packet(decoder);
1727*4882a593Smuzhiyun 		if (err)
1728*4882a593Smuzhiyun 			goto out;
1729*4882a593Smuzhiyun 
1730*4882a593Smuzhiyun 		switch (decoder->packet.type) {
1731*4882a593Smuzhiyun 		case INTEL_PT_PSBEND:
1732*4882a593Smuzhiyun 			err = 0;
1733*4882a593Smuzhiyun 			goto out;
1734*4882a593Smuzhiyun 
1735*4882a593Smuzhiyun 		case INTEL_PT_TIP_PGD:
1736*4882a593Smuzhiyun 		case INTEL_PT_TIP_PGE:
1737*4882a593Smuzhiyun 		case INTEL_PT_TIP:
1738*4882a593Smuzhiyun 		case INTEL_PT_TNT:
1739*4882a593Smuzhiyun 		case INTEL_PT_TRACESTOP:
1740*4882a593Smuzhiyun 		case INTEL_PT_BAD:
1741*4882a593Smuzhiyun 		case INTEL_PT_PSB:
1742*4882a593Smuzhiyun 		case INTEL_PT_PTWRITE:
1743*4882a593Smuzhiyun 		case INTEL_PT_PTWRITE_IP:
1744*4882a593Smuzhiyun 		case INTEL_PT_EXSTOP:
1745*4882a593Smuzhiyun 		case INTEL_PT_EXSTOP_IP:
1746*4882a593Smuzhiyun 		case INTEL_PT_MWAIT:
1747*4882a593Smuzhiyun 		case INTEL_PT_PWRE:
1748*4882a593Smuzhiyun 		case INTEL_PT_PWRX:
1749*4882a593Smuzhiyun 		case INTEL_PT_BBP:
1750*4882a593Smuzhiyun 		case INTEL_PT_BIP:
1751*4882a593Smuzhiyun 		case INTEL_PT_BEP:
1752*4882a593Smuzhiyun 		case INTEL_PT_BEP_IP:
1753*4882a593Smuzhiyun 			decoder->have_tma = false;
1754*4882a593Smuzhiyun 			intel_pt_log("ERROR: Unexpected packet\n");
1755*4882a593Smuzhiyun 			err = -EAGAIN;
1756*4882a593Smuzhiyun 			goto out;
1757*4882a593Smuzhiyun 
1758*4882a593Smuzhiyun 		case INTEL_PT_OVF:
1759*4882a593Smuzhiyun 			err = intel_pt_overflow(decoder);
1760*4882a593Smuzhiyun 			goto out;
1761*4882a593Smuzhiyun 
1762*4882a593Smuzhiyun 		case INTEL_PT_TSC:
1763*4882a593Smuzhiyun 			intel_pt_calc_tsc_timestamp(decoder);
1764*4882a593Smuzhiyun 			break;
1765*4882a593Smuzhiyun 
1766*4882a593Smuzhiyun 		case INTEL_PT_TMA:
1767*4882a593Smuzhiyun 			intel_pt_calc_tma(decoder);
1768*4882a593Smuzhiyun 			break;
1769*4882a593Smuzhiyun 
1770*4882a593Smuzhiyun 		case INTEL_PT_CBR:
1771*4882a593Smuzhiyun 			intel_pt_calc_cbr(decoder);
1772*4882a593Smuzhiyun 			break;
1773*4882a593Smuzhiyun 
1774*4882a593Smuzhiyun 		case INTEL_PT_MODE_EXEC:
1775*4882a593Smuzhiyun 			decoder->exec_mode = decoder->packet.payload;
1776*4882a593Smuzhiyun 			break;
1777*4882a593Smuzhiyun 
1778*4882a593Smuzhiyun 		case INTEL_PT_PIP:
1779*4882a593Smuzhiyun 			decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1780*4882a593Smuzhiyun 			break;
1781*4882a593Smuzhiyun 
1782*4882a593Smuzhiyun 		case INTEL_PT_FUP:
1783*4882a593Smuzhiyun 			decoder->pge = true;
1784*4882a593Smuzhiyun 			if (decoder->packet.count) {
1785*4882a593Smuzhiyun 				intel_pt_set_last_ip(decoder);
1786*4882a593Smuzhiyun 				if (decoder->hop) {
1787*4882a593Smuzhiyun 					/* Act on FUP at PSBEND */
1788*4882a593Smuzhiyun 					decoder->ip = decoder->last_ip;
1789*4882a593Smuzhiyun 					decoder->hop_psb_fup = true;
1790*4882a593Smuzhiyun 				}
1791*4882a593Smuzhiyun 			}
1792*4882a593Smuzhiyun 			break;
1793*4882a593Smuzhiyun 
1794*4882a593Smuzhiyun 		case INTEL_PT_MODE_TSX:
1795*4882a593Smuzhiyun 			intel_pt_update_in_tx(decoder);
1796*4882a593Smuzhiyun 			break;
1797*4882a593Smuzhiyun 
1798*4882a593Smuzhiyun 		case INTEL_PT_MTC:
1799*4882a593Smuzhiyun 			intel_pt_calc_mtc_timestamp(decoder);
1800*4882a593Smuzhiyun 			if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1801*4882a593Smuzhiyun 				decoder->state.type |= INTEL_PT_INSTRUCTION;
1802*4882a593Smuzhiyun 			break;
1803*4882a593Smuzhiyun 
1804*4882a593Smuzhiyun 		case INTEL_PT_CYC:
1805*4882a593Smuzhiyun 			intel_pt_calc_cyc_timestamp(decoder);
1806*4882a593Smuzhiyun 			break;
1807*4882a593Smuzhiyun 
1808*4882a593Smuzhiyun 		case INTEL_PT_VMCS:
1809*4882a593Smuzhiyun 		case INTEL_PT_MNT:
1810*4882a593Smuzhiyun 		case INTEL_PT_PAD:
1811*4882a593Smuzhiyun 		default:
1812*4882a593Smuzhiyun 			break;
1813*4882a593Smuzhiyun 		}
1814*4882a593Smuzhiyun 	}
1815*4882a593Smuzhiyun out:
1816*4882a593Smuzhiyun 	decoder->in_psb = false;
1817*4882a593Smuzhiyun 
1818*4882a593Smuzhiyun 	return err;
1819*4882a593Smuzhiyun }
1820*4882a593Smuzhiyun 
intel_pt_walk_fup_tip(struct intel_pt_decoder * decoder)1821*4882a593Smuzhiyun static int intel_pt_walk_fup_tip(struct intel_pt_decoder *decoder)
1822*4882a593Smuzhiyun {
1823*4882a593Smuzhiyun 	int err;
1824*4882a593Smuzhiyun 
1825*4882a593Smuzhiyun 	if (decoder->tx_flags & INTEL_PT_ABORT_TX) {
1826*4882a593Smuzhiyun 		decoder->tx_flags = 0;
1827*4882a593Smuzhiyun 		decoder->state.flags &= ~INTEL_PT_IN_TX;
1828*4882a593Smuzhiyun 		decoder->state.flags |= INTEL_PT_ABORT_TX;
1829*4882a593Smuzhiyun 	} else {
1830*4882a593Smuzhiyun 		decoder->state.flags |= INTEL_PT_ASYNC;
1831*4882a593Smuzhiyun 	}
1832*4882a593Smuzhiyun 
1833*4882a593Smuzhiyun 	while (1) {
1834*4882a593Smuzhiyun 		err = intel_pt_get_next_packet(decoder);
1835*4882a593Smuzhiyun 		if (err)
1836*4882a593Smuzhiyun 			return err;
1837*4882a593Smuzhiyun 
1838*4882a593Smuzhiyun 		switch (decoder->packet.type) {
1839*4882a593Smuzhiyun 		case INTEL_PT_TNT:
1840*4882a593Smuzhiyun 		case INTEL_PT_FUP:
1841*4882a593Smuzhiyun 		case INTEL_PT_TRACESTOP:
1842*4882a593Smuzhiyun 		case INTEL_PT_PSB:
1843*4882a593Smuzhiyun 		case INTEL_PT_TSC:
1844*4882a593Smuzhiyun 		case INTEL_PT_TMA:
1845*4882a593Smuzhiyun 		case INTEL_PT_MODE_TSX:
1846*4882a593Smuzhiyun 		case INTEL_PT_BAD:
1847*4882a593Smuzhiyun 		case INTEL_PT_PSBEND:
1848*4882a593Smuzhiyun 		case INTEL_PT_PTWRITE:
1849*4882a593Smuzhiyun 		case INTEL_PT_PTWRITE_IP:
1850*4882a593Smuzhiyun 		case INTEL_PT_EXSTOP:
1851*4882a593Smuzhiyun 		case INTEL_PT_EXSTOP_IP:
1852*4882a593Smuzhiyun 		case INTEL_PT_MWAIT:
1853*4882a593Smuzhiyun 		case INTEL_PT_PWRE:
1854*4882a593Smuzhiyun 		case INTEL_PT_PWRX:
1855*4882a593Smuzhiyun 		case INTEL_PT_BBP:
1856*4882a593Smuzhiyun 		case INTEL_PT_BIP:
1857*4882a593Smuzhiyun 		case INTEL_PT_BEP:
1858*4882a593Smuzhiyun 		case INTEL_PT_BEP_IP:
1859*4882a593Smuzhiyun 			intel_pt_log("ERROR: Missing TIP after FUP\n");
1860*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_ERR3;
1861*4882a593Smuzhiyun 			decoder->pkt_step = 0;
1862*4882a593Smuzhiyun 			return -ENOENT;
1863*4882a593Smuzhiyun 
1864*4882a593Smuzhiyun 		case INTEL_PT_CBR:
1865*4882a593Smuzhiyun 			intel_pt_calc_cbr(decoder);
1866*4882a593Smuzhiyun 			break;
1867*4882a593Smuzhiyun 
1868*4882a593Smuzhiyun 		case INTEL_PT_OVF:
1869*4882a593Smuzhiyun 			return intel_pt_overflow(decoder);
1870*4882a593Smuzhiyun 
1871*4882a593Smuzhiyun 		case INTEL_PT_TIP_PGD:
1872*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
1873*4882a593Smuzhiyun 			if (decoder->packet.count == 0) {
1874*4882a593Smuzhiyun 				decoder->state.to_ip = 0;
1875*4882a593Smuzhiyun 			} else {
1876*4882a593Smuzhiyun 				intel_pt_set_ip(decoder);
1877*4882a593Smuzhiyun 				decoder->state.to_ip = decoder->ip;
1878*4882a593Smuzhiyun 			}
1879*4882a593Smuzhiyun 			decoder->pge = false;
1880*4882a593Smuzhiyun 			decoder->continuous_period = false;
1881*4882a593Smuzhiyun 			decoder->state.type |= INTEL_PT_TRACE_END;
1882*4882a593Smuzhiyun 			return 0;
1883*4882a593Smuzhiyun 
1884*4882a593Smuzhiyun 		case INTEL_PT_TIP_PGE:
1885*4882a593Smuzhiyun 			decoder->pge = true;
1886*4882a593Smuzhiyun 			intel_pt_log("Omitting PGE ip " x64_fmt "\n",
1887*4882a593Smuzhiyun 				     decoder->ip);
1888*4882a593Smuzhiyun 			decoder->state.from_ip = 0;
1889*4882a593Smuzhiyun 			if (decoder->packet.count == 0) {
1890*4882a593Smuzhiyun 				decoder->state.to_ip = 0;
1891*4882a593Smuzhiyun 			} else {
1892*4882a593Smuzhiyun 				intel_pt_set_ip(decoder);
1893*4882a593Smuzhiyun 				decoder->state.to_ip = decoder->ip;
1894*4882a593Smuzhiyun 			}
1895*4882a593Smuzhiyun 			decoder->state.type |= INTEL_PT_TRACE_BEGIN;
1896*4882a593Smuzhiyun 			intel_pt_mtc_cyc_cnt_pge(decoder);
1897*4882a593Smuzhiyun 			return 0;
1898*4882a593Smuzhiyun 
1899*4882a593Smuzhiyun 		case INTEL_PT_TIP:
1900*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
1901*4882a593Smuzhiyun 			if (decoder->packet.count == 0) {
1902*4882a593Smuzhiyun 				decoder->state.to_ip = 0;
1903*4882a593Smuzhiyun 			} else {
1904*4882a593Smuzhiyun 				intel_pt_set_ip(decoder);
1905*4882a593Smuzhiyun 				decoder->state.to_ip = decoder->ip;
1906*4882a593Smuzhiyun 			}
1907*4882a593Smuzhiyun 			return 0;
1908*4882a593Smuzhiyun 
1909*4882a593Smuzhiyun 		case INTEL_PT_PIP:
1910*4882a593Smuzhiyun 			decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1911*4882a593Smuzhiyun 			break;
1912*4882a593Smuzhiyun 
1913*4882a593Smuzhiyun 		case INTEL_PT_MTC:
1914*4882a593Smuzhiyun 			intel_pt_calc_mtc_timestamp(decoder);
1915*4882a593Smuzhiyun 			if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1916*4882a593Smuzhiyun 				decoder->state.type |= INTEL_PT_INSTRUCTION;
1917*4882a593Smuzhiyun 			break;
1918*4882a593Smuzhiyun 
1919*4882a593Smuzhiyun 		case INTEL_PT_CYC:
1920*4882a593Smuzhiyun 			intel_pt_calc_cyc_timestamp(decoder);
1921*4882a593Smuzhiyun 			break;
1922*4882a593Smuzhiyun 
1923*4882a593Smuzhiyun 		case INTEL_PT_MODE_EXEC:
1924*4882a593Smuzhiyun 			decoder->exec_mode = decoder->packet.payload;
1925*4882a593Smuzhiyun 			break;
1926*4882a593Smuzhiyun 
1927*4882a593Smuzhiyun 		case INTEL_PT_VMCS:
1928*4882a593Smuzhiyun 		case INTEL_PT_MNT:
1929*4882a593Smuzhiyun 		case INTEL_PT_PAD:
1930*4882a593Smuzhiyun 			break;
1931*4882a593Smuzhiyun 
1932*4882a593Smuzhiyun 		default:
1933*4882a593Smuzhiyun 			return intel_pt_bug(decoder);
1934*4882a593Smuzhiyun 		}
1935*4882a593Smuzhiyun 	}
1936*4882a593Smuzhiyun }
1937*4882a593Smuzhiyun 
intel_pt_resample(struct intel_pt_decoder * decoder)1938*4882a593Smuzhiyun static int intel_pt_resample(struct intel_pt_decoder *decoder)
1939*4882a593Smuzhiyun {
1940*4882a593Smuzhiyun 	decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1941*4882a593Smuzhiyun 	decoder->state.type = INTEL_PT_INSTRUCTION;
1942*4882a593Smuzhiyun 	decoder->state.from_ip = decoder->ip;
1943*4882a593Smuzhiyun 	decoder->state.to_ip = 0;
1944*4882a593Smuzhiyun 	return 0;
1945*4882a593Smuzhiyun }
1946*4882a593Smuzhiyun 
1947*4882a593Smuzhiyun #define HOP_PROCESS	0
1948*4882a593Smuzhiyun #define HOP_IGNORE	1
1949*4882a593Smuzhiyun #define HOP_RETURN	2
1950*4882a593Smuzhiyun #define HOP_AGAIN	3
1951*4882a593Smuzhiyun 
1952*4882a593Smuzhiyun static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder);
1953*4882a593Smuzhiyun 
1954*4882a593Smuzhiyun /* Hop mode: Ignore TNT, do not walk code, but get ip from FUPs and TIPs */
intel_pt_hop_trace(struct intel_pt_decoder * decoder,bool * no_tip,int * err)1955*4882a593Smuzhiyun static int intel_pt_hop_trace(struct intel_pt_decoder *decoder, bool *no_tip, int *err)
1956*4882a593Smuzhiyun {
1957*4882a593Smuzhiyun 	*err = 0;
1958*4882a593Smuzhiyun 
1959*4882a593Smuzhiyun 	/* Leap from PSB to PSB, getting ip from FUP within PSB+ */
1960*4882a593Smuzhiyun 	if (decoder->leap && !decoder->in_psb && decoder->packet.type != INTEL_PT_PSB) {
1961*4882a593Smuzhiyun 		*err = intel_pt_scan_for_psb(decoder);
1962*4882a593Smuzhiyun 		if (*err)
1963*4882a593Smuzhiyun 			return HOP_RETURN;
1964*4882a593Smuzhiyun 	}
1965*4882a593Smuzhiyun 
1966*4882a593Smuzhiyun 	switch (decoder->packet.type) {
1967*4882a593Smuzhiyun 	case INTEL_PT_TNT:
1968*4882a593Smuzhiyun 		return HOP_IGNORE;
1969*4882a593Smuzhiyun 
1970*4882a593Smuzhiyun 	case INTEL_PT_TIP_PGD:
1971*4882a593Smuzhiyun 		decoder->pge = false;
1972*4882a593Smuzhiyun 		if (!decoder->packet.count)
1973*4882a593Smuzhiyun 			return HOP_IGNORE;
1974*4882a593Smuzhiyun 		intel_pt_set_ip(decoder);
1975*4882a593Smuzhiyun 		decoder->state.type |= INTEL_PT_TRACE_END;
1976*4882a593Smuzhiyun 		decoder->state.from_ip = 0;
1977*4882a593Smuzhiyun 		decoder->state.to_ip = decoder->ip;
1978*4882a593Smuzhiyun 		return HOP_RETURN;
1979*4882a593Smuzhiyun 
1980*4882a593Smuzhiyun 	case INTEL_PT_TIP:
1981*4882a593Smuzhiyun 		if (!decoder->packet.count)
1982*4882a593Smuzhiyun 			return HOP_IGNORE;
1983*4882a593Smuzhiyun 		intel_pt_set_ip(decoder);
1984*4882a593Smuzhiyun 		decoder->state.type = INTEL_PT_INSTRUCTION;
1985*4882a593Smuzhiyun 		decoder->state.from_ip = decoder->ip;
1986*4882a593Smuzhiyun 		decoder->state.to_ip = 0;
1987*4882a593Smuzhiyun 		return HOP_RETURN;
1988*4882a593Smuzhiyun 
1989*4882a593Smuzhiyun 	case INTEL_PT_FUP:
1990*4882a593Smuzhiyun 		if (!decoder->packet.count)
1991*4882a593Smuzhiyun 			return HOP_IGNORE;
1992*4882a593Smuzhiyun 		intel_pt_set_ip(decoder);
1993*4882a593Smuzhiyun 		if (decoder->set_fup_mwait || decoder->set_fup_pwre)
1994*4882a593Smuzhiyun 			*no_tip = true;
1995*4882a593Smuzhiyun 		if (!decoder->branch_enable || !decoder->pge)
1996*4882a593Smuzhiyun 			*no_tip = true;
1997*4882a593Smuzhiyun 		if (*no_tip) {
1998*4882a593Smuzhiyun 			decoder->state.type = INTEL_PT_INSTRUCTION;
1999*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
2000*4882a593Smuzhiyun 			decoder->state.to_ip = 0;
2001*4882a593Smuzhiyun 			intel_pt_fup_event(decoder);
2002*4882a593Smuzhiyun 			return HOP_RETURN;
2003*4882a593Smuzhiyun 		}
2004*4882a593Smuzhiyun 		intel_pt_fup_event(decoder);
2005*4882a593Smuzhiyun 		decoder->state.type |= INTEL_PT_INSTRUCTION | INTEL_PT_BRANCH;
2006*4882a593Smuzhiyun 		*err = intel_pt_walk_fup_tip(decoder);
2007*4882a593Smuzhiyun 		if (!*err && decoder->state.to_ip)
2008*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
2009*4882a593Smuzhiyun 		return HOP_RETURN;
2010*4882a593Smuzhiyun 
2011*4882a593Smuzhiyun 	case INTEL_PT_PSB:
2012*4882a593Smuzhiyun 		decoder->last_ip = 0;
2013*4882a593Smuzhiyun 		decoder->have_last_ip = true;
2014*4882a593Smuzhiyun 		decoder->hop_psb_fup = false;
2015*4882a593Smuzhiyun 		*err = intel_pt_walk_psbend(decoder);
2016*4882a593Smuzhiyun 		if (*err == -EAGAIN)
2017*4882a593Smuzhiyun 			return HOP_AGAIN;
2018*4882a593Smuzhiyun 		if (*err)
2019*4882a593Smuzhiyun 			return HOP_RETURN;
2020*4882a593Smuzhiyun 		if (decoder->hop_psb_fup) {
2021*4882a593Smuzhiyun 			decoder->hop_psb_fup = false;
2022*4882a593Smuzhiyun 			decoder->state.type = INTEL_PT_INSTRUCTION;
2023*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
2024*4882a593Smuzhiyun 			decoder->state.to_ip = 0;
2025*4882a593Smuzhiyun 			return HOP_RETURN;
2026*4882a593Smuzhiyun 		}
2027*4882a593Smuzhiyun 		if (decoder->cbr != decoder->cbr_seen) {
2028*4882a593Smuzhiyun 			decoder->state.type = 0;
2029*4882a593Smuzhiyun 			return HOP_RETURN;
2030*4882a593Smuzhiyun 		}
2031*4882a593Smuzhiyun 		return HOP_IGNORE;
2032*4882a593Smuzhiyun 
2033*4882a593Smuzhiyun 	case INTEL_PT_BAD:
2034*4882a593Smuzhiyun 	case INTEL_PT_PAD:
2035*4882a593Smuzhiyun 	case INTEL_PT_TIP_PGE:
2036*4882a593Smuzhiyun 	case INTEL_PT_TSC:
2037*4882a593Smuzhiyun 	case INTEL_PT_TMA:
2038*4882a593Smuzhiyun 	case INTEL_PT_MODE_EXEC:
2039*4882a593Smuzhiyun 	case INTEL_PT_MODE_TSX:
2040*4882a593Smuzhiyun 	case INTEL_PT_MTC:
2041*4882a593Smuzhiyun 	case INTEL_PT_CYC:
2042*4882a593Smuzhiyun 	case INTEL_PT_VMCS:
2043*4882a593Smuzhiyun 	case INTEL_PT_PSBEND:
2044*4882a593Smuzhiyun 	case INTEL_PT_CBR:
2045*4882a593Smuzhiyun 	case INTEL_PT_TRACESTOP:
2046*4882a593Smuzhiyun 	case INTEL_PT_PIP:
2047*4882a593Smuzhiyun 	case INTEL_PT_OVF:
2048*4882a593Smuzhiyun 	case INTEL_PT_MNT:
2049*4882a593Smuzhiyun 	case INTEL_PT_PTWRITE:
2050*4882a593Smuzhiyun 	case INTEL_PT_PTWRITE_IP:
2051*4882a593Smuzhiyun 	case INTEL_PT_EXSTOP:
2052*4882a593Smuzhiyun 	case INTEL_PT_EXSTOP_IP:
2053*4882a593Smuzhiyun 	case INTEL_PT_MWAIT:
2054*4882a593Smuzhiyun 	case INTEL_PT_PWRE:
2055*4882a593Smuzhiyun 	case INTEL_PT_PWRX:
2056*4882a593Smuzhiyun 	case INTEL_PT_BBP:
2057*4882a593Smuzhiyun 	case INTEL_PT_BIP:
2058*4882a593Smuzhiyun 	case INTEL_PT_BEP:
2059*4882a593Smuzhiyun 	case INTEL_PT_BEP_IP:
2060*4882a593Smuzhiyun 	default:
2061*4882a593Smuzhiyun 		return HOP_PROCESS;
2062*4882a593Smuzhiyun 	}
2063*4882a593Smuzhiyun }
2064*4882a593Smuzhiyun 
intel_pt_walk_trace(struct intel_pt_decoder * decoder)2065*4882a593Smuzhiyun static int intel_pt_walk_trace(struct intel_pt_decoder *decoder)
2066*4882a593Smuzhiyun {
2067*4882a593Smuzhiyun 	int last_packet_type = INTEL_PT_PAD;
2068*4882a593Smuzhiyun 	bool no_tip = false;
2069*4882a593Smuzhiyun 	int err;
2070*4882a593Smuzhiyun 
2071*4882a593Smuzhiyun 	while (1) {
2072*4882a593Smuzhiyun 		err = intel_pt_get_next_packet(decoder);
2073*4882a593Smuzhiyun 		if (err)
2074*4882a593Smuzhiyun 			return err;
2075*4882a593Smuzhiyun next:
2076*4882a593Smuzhiyun 		err = 0;
2077*4882a593Smuzhiyun 		if (decoder->cyc_threshold) {
2078*4882a593Smuzhiyun 			if (decoder->sample_cyc && last_packet_type != INTEL_PT_CYC)
2079*4882a593Smuzhiyun 				decoder->sample_cyc = false;
2080*4882a593Smuzhiyun 			last_packet_type = decoder->packet.type;
2081*4882a593Smuzhiyun 		}
2082*4882a593Smuzhiyun 
2083*4882a593Smuzhiyun 		if (decoder->hop) {
2084*4882a593Smuzhiyun 			switch (intel_pt_hop_trace(decoder, &no_tip, &err)) {
2085*4882a593Smuzhiyun 			case HOP_IGNORE:
2086*4882a593Smuzhiyun 				continue;
2087*4882a593Smuzhiyun 			case HOP_RETURN:
2088*4882a593Smuzhiyun 				return err;
2089*4882a593Smuzhiyun 			case HOP_AGAIN:
2090*4882a593Smuzhiyun 				goto next;
2091*4882a593Smuzhiyun 			default:
2092*4882a593Smuzhiyun 				break;
2093*4882a593Smuzhiyun 			}
2094*4882a593Smuzhiyun 		}
2095*4882a593Smuzhiyun 
2096*4882a593Smuzhiyun 		switch (decoder->packet.type) {
2097*4882a593Smuzhiyun 		case INTEL_PT_TNT:
2098*4882a593Smuzhiyun 			if (!decoder->packet.count)
2099*4882a593Smuzhiyun 				break;
2100*4882a593Smuzhiyun 			decoder->tnt = decoder->packet;
2101*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_TNT;
2102*4882a593Smuzhiyun 			err = intel_pt_walk_tnt(decoder);
2103*4882a593Smuzhiyun 			if (err == -EAGAIN)
2104*4882a593Smuzhiyun 				break;
2105*4882a593Smuzhiyun 			return err;
2106*4882a593Smuzhiyun 
2107*4882a593Smuzhiyun 		case INTEL_PT_TIP_PGD:
2108*4882a593Smuzhiyun 			if (decoder->packet.count != 0)
2109*4882a593Smuzhiyun 				intel_pt_set_last_ip(decoder);
2110*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_TIP_PGD;
2111*4882a593Smuzhiyun 			return intel_pt_walk_tip(decoder);
2112*4882a593Smuzhiyun 
2113*4882a593Smuzhiyun 		case INTEL_PT_TIP_PGE: {
2114*4882a593Smuzhiyun 			decoder->pge = true;
2115*4882a593Smuzhiyun 			decoder->overflow = false;
2116*4882a593Smuzhiyun 			intel_pt_mtc_cyc_cnt_pge(decoder);
2117*4882a593Smuzhiyun 			if (decoder->packet.count == 0) {
2118*4882a593Smuzhiyun 				intel_pt_log_at("Skipping zero TIP.PGE",
2119*4882a593Smuzhiyun 						decoder->pos);
2120*4882a593Smuzhiyun 				break;
2121*4882a593Smuzhiyun 			}
2122*4882a593Smuzhiyun 			intel_pt_set_ip(decoder);
2123*4882a593Smuzhiyun 			decoder->state.from_ip = 0;
2124*4882a593Smuzhiyun 			decoder->state.to_ip = decoder->ip;
2125*4882a593Smuzhiyun 			decoder->state.type |= INTEL_PT_TRACE_BEGIN;
2126*4882a593Smuzhiyun 			/*
2127*4882a593Smuzhiyun 			 * In hop mode, resample to get the to_ip as an
2128*4882a593Smuzhiyun 			 * "instruction" sample.
2129*4882a593Smuzhiyun 			 */
2130*4882a593Smuzhiyun 			if (decoder->hop)
2131*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
2132*4882a593Smuzhiyun 			return 0;
2133*4882a593Smuzhiyun 		}
2134*4882a593Smuzhiyun 
2135*4882a593Smuzhiyun 		case INTEL_PT_OVF:
2136*4882a593Smuzhiyun 			return intel_pt_overflow(decoder);
2137*4882a593Smuzhiyun 
2138*4882a593Smuzhiyun 		case INTEL_PT_TIP:
2139*4882a593Smuzhiyun 			if (decoder->packet.count != 0)
2140*4882a593Smuzhiyun 				intel_pt_set_last_ip(decoder);
2141*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_TIP;
2142*4882a593Smuzhiyun 			return intel_pt_walk_tip(decoder);
2143*4882a593Smuzhiyun 
2144*4882a593Smuzhiyun 		case INTEL_PT_FUP:
2145*4882a593Smuzhiyun 			if (decoder->packet.count == 0) {
2146*4882a593Smuzhiyun 				intel_pt_log_at("Skipping zero FUP",
2147*4882a593Smuzhiyun 						decoder->pos);
2148*4882a593Smuzhiyun 				no_tip = false;
2149*4882a593Smuzhiyun 				break;
2150*4882a593Smuzhiyun 			}
2151*4882a593Smuzhiyun 			intel_pt_set_last_ip(decoder);
2152*4882a593Smuzhiyun 			if (!decoder->branch_enable || !decoder->pge) {
2153*4882a593Smuzhiyun 				decoder->ip = decoder->last_ip;
2154*4882a593Smuzhiyun 				if (intel_pt_fup_event(decoder))
2155*4882a593Smuzhiyun 					return 0;
2156*4882a593Smuzhiyun 				no_tip = false;
2157*4882a593Smuzhiyun 				break;
2158*4882a593Smuzhiyun 			}
2159*4882a593Smuzhiyun 			if (decoder->set_fup_mwait)
2160*4882a593Smuzhiyun 				no_tip = true;
2161*4882a593Smuzhiyun 			if (no_tip)
2162*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_FUP_NO_TIP;
2163*4882a593Smuzhiyun 			else
2164*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_FUP;
2165*4882a593Smuzhiyun 			err = intel_pt_walk_fup(decoder);
2166*4882a593Smuzhiyun 			if (err != -EAGAIN)
2167*4882a593Smuzhiyun 				return err;
2168*4882a593Smuzhiyun 			if (no_tip) {
2169*4882a593Smuzhiyun 				no_tip = false;
2170*4882a593Smuzhiyun 				break;
2171*4882a593Smuzhiyun 			}
2172*4882a593Smuzhiyun 			return intel_pt_walk_fup_tip(decoder);
2173*4882a593Smuzhiyun 
2174*4882a593Smuzhiyun 		case INTEL_PT_TRACESTOP:
2175*4882a593Smuzhiyun 			decoder->pge = false;
2176*4882a593Smuzhiyun 			decoder->continuous_period = false;
2177*4882a593Smuzhiyun 			intel_pt_clear_tx_flags(decoder);
2178*4882a593Smuzhiyun 			decoder->have_tma = false;
2179*4882a593Smuzhiyun 			break;
2180*4882a593Smuzhiyun 
2181*4882a593Smuzhiyun 		case INTEL_PT_PSB:
2182*4882a593Smuzhiyun 			decoder->last_ip = 0;
2183*4882a593Smuzhiyun 			decoder->have_last_ip = true;
2184*4882a593Smuzhiyun 			intel_pt_clear_stack(&decoder->stack);
2185*4882a593Smuzhiyun 			err = intel_pt_walk_psbend(decoder);
2186*4882a593Smuzhiyun 			if (err == -EAGAIN)
2187*4882a593Smuzhiyun 				goto next;
2188*4882a593Smuzhiyun 			if (err)
2189*4882a593Smuzhiyun 				return err;
2190*4882a593Smuzhiyun 			/*
2191*4882a593Smuzhiyun 			 * PSB+ CBR will not have changed but cater for the
2192*4882a593Smuzhiyun 			 * possibility of another CBR change that gets caught up
2193*4882a593Smuzhiyun 			 * in the PSB+.
2194*4882a593Smuzhiyun 			 */
2195*4882a593Smuzhiyun 			if (decoder->cbr != decoder->cbr_seen) {
2196*4882a593Smuzhiyun 				decoder->state.type = 0;
2197*4882a593Smuzhiyun 				return 0;
2198*4882a593Smuzhiyun 			}
2199*4882a593Smuzhiyun 			break;
2200*4882a593Smuzhiyun 
2201*4882a593Smuzhiyun 		case INTEL_PT_PIP:
2202*4882a593Smuzhiyun 			decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
2203*4882a593Smuzhiyun 			break;
2204*4882a593Smuzhiyun 
2205*4882a593Smuzhiyun 		case INTEL_PT_MTC:
2206*4882a593Smuzhiyun 			intel_pt_calc_mtc_timestamp(decoder);
2207*4882a593Smuzhiyun 			if (decoder->period_type != INTEL_PT_PERIOD_MTC)
2208*4882a593Smuzhiyun 				break;
2209*4882a593Smuzhiyun 			/*
2210*4882a593Smuzhiyun 			 * Ensure that there has been an instruction since the
2211*4882a593Smuzhiyun 			 * last MTC.
2212*4882a593Smuzhiyun 			 */
2213*4882a593Smuzhiyun 			if (!decoder->mtc_insn)
2214*4882a593Smuzhiyun 				break;
2215*4882a593Smuzhiyun 			decoder->mtc_insn = false;
2216*4882a593Smuzhiyun 			/* Ensure that there is a timestamp */
2217*4882a593Smuzhiyun 			if (!decoder->timestamp)
2218*4882a593Smuzhiyun 				break;
2219*4882a593Smuzhiyun 			decoder->state.type = INTEL_PT_INSTRUCTION;
2220*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
2221*4882a593Smuzhiyun 			decoder->state.to_ip = 0;
2222*4882a593Smuzhiyun 			decoder->mtc_insn = false;
2223*4882a593Smuzhiyun 			return 0;
2224*4882a593Smuzhiyun 
2225*4882a593Smuzhiyun 		case INTEL_PT_TSC:
2226*4882a593Smuzhiyun 			intel_pt_calc_tsc_timestamp(decoder);
2227*4882a593Smuzhiyun 			break;
2228*4882a593Smuzhiyun 
2229*4882a593Smuzhiyun 		case INTEL_PT_TMA:
2230*4882a593Smuzhiyun 			intel_pt_calc_tma(decoder);
2231*4882a593Smuzhiyun 			break;
2232*4882a593Smuzhiyun 
2233*4882a593Smuzhiyun 		case INTEL_PT_CYC:
2234*4882a593Smuzhiyun 			intel_pt_calc_cyc_timestamp(decoder);
2235*4882a593Smuzhiyun 			break;
2236*4882a593Smuzhiyun 
2237*4882a593Smuzhiyun 		case INTEL_PT_CBR:
2238*4882a593Smuzhiyun 			intel_pt_calc_cbr(decoder);
2239*4882a593Smuzhiyun 			if (decoder->cbr != decoder->cbr_seen) {
2240*4882a593Smuzhiyun 				decoder->state.type = 0;
2241*4882a593Smuzhiyun 				return 0;
2242*4882a593Smuzhiyun 			}
2243*4882a593Smuzhiyun 			break;
2244*4882a593Smuzhiyun 
2245*4882a593Smuzhiyun 		case INTEL_PT_MODE_EXEC:
2246*4882a593Smuzhiyun 			decoder->exec_mode = decoder->packet.payload;
2247*4882a593Smuzhiyun 			break;
2248*4882a593Smuzhiyun 
2249*4882a593Smuzhiyun 		case INTEL_PT_MODE_TSX:
2250*4882a593Smuzhiyun 			/* MODE_TSX need not be followed by FUP */
2251*4882a593Smuzhiyun 			if (!decoder->pge || decoder->in_psb) {
2252*4882a593Smuzhiyun 				intel_pt_update_in_tx(decoder);
2253*4882a593Smuzhiyun 				break;
2254*4882a593Smuzhiyun 			}
2255*4882a593Smuzhiyun 			err = intel_pt_mode_tsx(decoder, &no_tip);
2256*4882a593Smuzhiyun 			if (err)
2257*4882a593Smuzhiyun 				return err;
2258*4882a593Smuzhiyun 			goto next;
2259*4882a593Smuzhiyun 
2260*4882a593Smuzhiyun 		case INTEL_PT_BAD: /* Does not happen */
2261*4882a593Smuzhiyun 			return intel_pt_bug(decoder);
2262*4882a593Smuzhiyun 
2263*4882a593Smuzhiyun 		case INTEL_PT_PSBEND:
2264*4882a593Smuzhiyun 		case INTEL_PT_VMCS:
2265*4882a593Smuzhiyun 		case INTEL_PT_MNT:
2266*4882a593Smuzhiyun 		case INTEL_PT_PAD:
2267*4882a593Smuzhiyun 			break;
2268*4882a593Smuzhiyun 
2269*4882a593Smuzhiyun 		case INTEL_PT_PTWRITE_IP:
2270*4882a593Smuzhiyun 			decoder->fup_ptw_payload = decoder->packet.payload;
2271*4882a593Smuzhiyun 			err = intel_pt_get_next_packet(decoder);
2272*4882a593Smuzhiyun 			if (err)
2273*4882a593Smuzhiyun 				return err;
2274*4882a593Smuzhiyun 			if (decoder->packet.type == INTEL_PT_FUP) {
2275*4882a593Smuzhiyun 				decoder->set_fup_ptw = true;
2276*4882a593Smuzhiyun 				no_tip = true;
2277*4882a593Smuzhiyun 			} else {
2278*4882a593Smuzhiyun 				intel_pt_log_at("ERROR: Missing FUP after PTWRITE",
2279*4882a593Smuzhiyun 						decoder->pos);
2280*4882a593Smuzhiyun 			}
2281*4882a593Smuzhiyun 			goto next;
2282*4882a593Smuzhiyun 
2283*4882a593Smuzhiyun 		case INTEL_PT_PTWRITE:
2284*4882a593Smuzhiyun 			decoder->state.type = INTEL_PT_PTW;
2285*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
2286*4882a593Smuzhiyun 			decoder->state.to_ip = 0;
2287*4882a593Smuzhiyun 			decoder->state.ptw_payload = decoder->packet.payload;
2288*4882a593Smuzhiyun 			return 0;
2289*4882a593Smuzhiyun 
2290*4882a593Smuzhiyun 		case INTEL_PT_MWAIT:
2291*4882a593Smuzhiyun 			decoder->fup_mwait_payload = decoder->packet.payload;
2292*4882a593Smuzhiyun 			decoder->set_fup_mwait = true;
2293*4882a593Smuzhiyun 			break;
2294*4882a593Smuzhiyun 
2295*4882a593Smuzhiyun 		case INTEL_PT_PWRE:
2296*4882a593Smuzhiyun 			if (decoder->set_fup_mwait) {
2297*4882a593Smuzhiyun 				decoder->fup_pwre_payload =
2298*4882a593Smuzhiyun 							decoder->packet.payload;
2299*4882a593Smuzhiyun 				decoder->set_fup_pwre = true;
2300*4882a593Smuzhiyun 				break;
2301*4882a593Smuzhiyun 			}
2302*4882a593Smuzhiyun 			decoder->state.type = INTEL_PT_PWR_ENTRY;
2303*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
2304*4882a593Smuzhiyun 			decoder->state.to_ip = 0;
2305*4882a593Smuzhiyun 			decoder->state.pwrx_payload = decoder->packet.payload;
2306*4882a593Smuzhiyun 			return 0;
2307*4882a593Smuzhiyun 
2308*4882a593Smuzhiyun 		case INTEL_PT_EXSTOP_IP:
2309*4882a593Smuzhiyun 			err = intel_pt_get_next_packet(decoder);
2310*4882a593Smuzhiyun 			if (err)
2311*4882a593Smuzhiyun 				return err;
2312*4882a593Smuzhiyun 			if (decoder->packet.type == INTEL_PT_FUP) {
2313*4882a593Smuzhiyun 				decoder->set_fup_exstop = true;
2314*4882a593Smuzhiyun 				no_tip = true;
2315*4882a593Smuzhiyun 			} else {
2316*4882a593Smuzhiyun 				intel_pt_log_at("ERROR: Missing FUP after EXSTOP",
2317*4882a593Smuzhiyun 						decoder->pos);
2318*4882a593Smuzhiyun 			}
2319*4882a593Smuzhiyun 			goto next;
2320*4882a593Smuzhiyun 
2321*4882a593Smuzhiyun 		case INTEL_PT_EXSTOP:
2322*4882a593Smuzhiyun 			decoder->state.type = INTEL_PT_EX_STOP;
2323*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
2324*4882a593Smuzhiyun 			decoder->state.to_ip = 0;
2325*4882a593Smuzhiyun 			return 0;
2326*4882a593Smuzhiyun 
2327*4882a593Smuzhiyun 		case INTEL_PT_PWRX:
2328*4882a593Smuzhiyun 			decoder->state.type = INTEL_PT_PWR_EXIT;
2329*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
2330*4882a593Smuzhiyun 			decoder->state.to_ip = 0;
2331*4882a593Smuzhiyun 			decoder->state.pwrx_payload = decoder->packet.payload;
2332*4882a593Smuzhiyun 			return 0;
2333*4882a593Smuzhiyun 
2334*4882a593Smuzhiyun 		case INTEL_PT_BBP:
2335*4882a593Smuzhiyun 			intel_pt_bbp(decoder);
2336*4882a593Smuzhiyun 			break;
2337*4882a593Smuzhiyun 
2338*4882a593Smuzhiyun 		case INTEL_PT_BIP:
2339*4882a593Smuzhiyun 			intel_pt_bip(decoder);
2340*4882a593Smuzhiyun 			break;
2341*4882a593Smuzhiyun 
2342*4882a593Smuzhiyun 		case INTEL_PT_BEP:
2343*4882a593Smuzhiyun 			decoder->state.type = INTEL_PT_BLK_ITEMS;
2344*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
2345*4882a593Smuzhiyun 			decoder->state.to_ip = 0;
2346*4882a593Smuzhiyun 			return 0;
2347*4882a593Smuzhiyun 
2348*4882a593Smuzhiyun 		case INTEL_PT_BEP_IP:
2349*4882a593Smuzhiyun 			err = intel_pt_get_next_packet(decoder);
2350*4882a593Smuzhiyun 			if (err)
2351*4882a593Smuzhiyun 				return err;
2352*4882a593Smuzhiyun 			if (decoder->packet.type == INTEL_PT_FUP) {
2353*4882a593Smuzhiyun 				decoder->set_fup_bep = true;
2354*4882a593Smuzhiyun 				no_tip = true;
2355*4882a593Smuzhiyun 			} else {
2356*4882a593Smuzhiyun 				intel_pt_log_at("ERROR: Missing FUP after BEP",
2357*4882a593Smuzhiyun 						decoder->pos);
2358*4882a593Smuzhiyun 			}
2359*4882a593Smuzhiyun 			goto next;
2360*4882a593Smuzhiyun 
2361*4882a593Smuzhiyun 		default:
2362*4882a593Smuzhiyun 			return intel_pt_bug(decoder);
2363*4882a593Smuzhiyun 		}
2364*4882a593Smuzhiyun 	}
2365*4882a593Smuzhiyun }
2366*4882a593Smuzhiyun 
intel_pt_have_ip(struct intel_pt_decoder * decoder)2367*4882a593Smuzhiyun static inline bool intel_pt_have_ip(struct intel_pt_decoder *decoder)
2368*4882a593Smuzhiyun {
2369*4882a593Smuzhiyun 	return decoder->packet.count &&
2370*4882a593Smuzhiyun 	       (decoder->have_last_ip || decoder->packet.count == 3 ||
2371*4882a593Smuzhiyun 		decoder->packet.count == 6);
2372*4882a593Smuzhiyun }
2373*4882a593Smuzhiyun 
2374*4882a593Smuzhiyun /* Walk PSB+ packets to get in sync. */
intel_pt_walk_psb(struct intel_pt_decoder * decoder)2375*4882a593Smuzhiyun static int intel_pt_walk_psb(struct intel_pt_decoder *decoder)
2376*4882a593Smuzhiyun {
2377*4882a593Smuzhiyun 	int err;
2378*4882a593Smuzhiyun 
2379*4882a593Smuzhiyun 	decoder->in_psb = true;
2380*4882a593Smuzhiyun 
2381*4882a593Smuzhiyun 	while (1) {
2382*4882a593Smuzhiyun 		err = intel_pt_get_next_packet(decoder);
2383*4882a593Smuzhiyun 		if (err)
2384*4882a593Smuzhiyun 			goto out;
2385*4882a593Smuzhiyun 
2386*4882a593Smuzhiyun 		switch (decoder->packet.type) {
2387*4882a593Smuzhiyun 		case INTEL_PT_TIP_PGD:
2388*4882a593Smuzhiyun 			decoder->continuous_period = false;
2389*4882a593Smuzhiyun 			__fallthrough;
2390*4882a593Smuzhiyun 		case INTEL_PT_TIP_PGE:
2391*4882a593Smuzhiyun 		case INTEL_PT_TIP:
2392*4882a593Smuzhiyun 		case INTEL_PT_PTWRITE:
2393*4882a593Smuzhiyun 		case INTEL_PT_PTWRITE_IP:
2394*4882a593Smuzhiyun 		case INTEL_PT_EXSTOP:
2395*4882a593Smuzhiyun 		case INTEL_PT_EXSTOP_IP:
2396*4882a593Smuzhiyun 		case INTEL_PT_MWAIT:
2397*4882a593Smuzhiyun 		case INTEL_PT_PWRE:
2398*4882a593Smuzhiyun 		case INTEL_PT_PWRX:
2399*4882a593Smuzhiyun 		case INTEL_PT_BBP:
2400*4882a593Smuzhiyun 		case INTEL_PT_BIP:
2401*4882a593Smuzhiyun 		case INTEL_PT_BEP:
2402*4882a593Smuzhiyun 		case INTEL_PT_BEP_IP:
2403*4882a593Smuzhiyun 			intel_pt_log("ERROR: Unexpected packet\n");
2404*4882a593Smuzhiyun 			err = -ENOENT;
2405*4882a593Smuzhiyun 			goto out;
2406*4882a593Smuzhiyun 
2407*4882a593Smuzhiyun 		case INTEL_PT_FUP:
2408*4882a593Smuzhiyun 			decoder->pge = true;
2409*4882a593Smuzhiyun 			if (intel_pt_have_ip(decoder)) {
2410*4882a593Smuzhiyun 				uint64_t current_ip = decoder->ip;
2411*4882a593Smuzhiyun 
2412*4882a593Smuzhiyun 				intel_pt_set_ip(decoder);
2413*4882a593Smuzhiyun 				if (current_ip)
2414*4882a593Smuzhiyun 					intel_pt_log_to("Setting IP",
2415*4882a593Smuzhiyun 							decoder->ip);
2416*4882a593Smuzhiyun 			}
2417*4882a593Smuzhiyun 			break;
2418*4882a593Smuzhiyun 
2419*4882a593Smuzhiyun 		case INTEL_PT_MTC:
2420*4882a593Smuzhiyun 			intel_pt_calc_mtc_timestamp(decoder);
2421*4882a593Smuzhiyun 			break;
2422*4882a593Smuzhiyun 
2423*4882a593Smuzhiyun 		case INTEL_PT_TSC:
2424*4882a593Smuzhiyun 			intel_pt_calc_tsc_timestamp(decoder);
2425*4882a593Smuzhiyun 			break;
2426*4882a593Smuzhiyun 
2427*4882a593Smuzhiyun 		case INTEL_PT_TMA:
2428*4882a593Smuzhiyun 			intel_pt_calc_tma(decoder);
2429*4882a593Smuzhiyun 			break;
2430*4882a593Smuzhiyun 
2431*4882a593Smuzhiyun 		case INTEL_PT_CYC:
2432*4882a593Smuzhiyun 			intel_pt_calc_cyc_timestamp(decoder);
2433*4882a593Smuzhiyun 			break;
2434*4882a593Smuzhiyun 
2435*4882a593Smuzhiyun 		case INTEL_PT_CBR:
2436*4882a593Smuzhiyun 			intel_pt_calc_cbr(decoder);
2437*4882a593Smuzhiyun 			break;
2438*4882a593Smuzhiyun 
2439*4882a593Smuzhiyun 		case INTEL_PT_PIP:
2440*4882a593Smuzhiyun 			decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
2441*4882a593Smuzhiyun 			break;
2442*4882a593Smuzhiyun 
2443*4882a593Smuzhiyun 		case INTEL_PT_MODE_EXEC:
2444*4882a593Smuzhiyun 			decoder->exec_mode = decoder->packet.payload;
2445*4882a593Smuzhiyun 			break;
2446*4882a593Smuzhiyun 
2447*4882a593Smuzhiyun 		case INTEL_PT_MODE_TSX:
2448*4882a593Smuzhiyun 			intel_pt_update_in_tx(decoder);
2449*4882a593Smuzhiyun 			break;
2450*4882a593Smuzhiyun 
2451*4882a593Smuzhiyun 		case INTEL_PT_TRACESTOP:
2452*4882a593Smuzhiyun 			decoder->pge = false;
2453*4882a593Smuzhiyun 			decoder->continuous_period = false;
2454*4882a593Smuzhiyun 			intel_pt_clear_tx_flags(decoder);
2455*4882a593Smuzhiyun 			__fallthrough;
2456*4882a593Smuzhiyun 
2457*4882a593Smuzhiyun 		case INTEL_PT_TNT:
2458*4882a593Smuzhiyun 			decoder->have_tma = false;
2459*4882a593Smuzhiyun 			intel_pt_log("ERROR: Unexpected packet\n");
2460*4882a593Smuzhiyun 			if (decoder->ip)
2461*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_ERR4;
2462*4882a593Smuzhiyun 			else
2463*4882a593Smuzhiyun 				decoder->pkt_state = INTEL_PT_STATE_ERR3;
2464*4882a593Smuzhiyun 			err = -ENOENT;
2465*4882a593Smuzhiyun 			goto out;
2466*4882a593Smuzhiyun 
2467*4882a593Smuzhiyun 		case INTEL_PT_BAD: /* Does not happen */
2468*4882a593Smuzhiyun 			err = intel_pt_bug(decoder);
2469*4882a593Smuzhiyun 			goto out;
2470*4882a593Smuzhiyun 
2471*4882a593Smuzhiyun 		case INTEL_PT_OVF:
2472*4882a593Smuzhiyun 			err = intel_pt_overflow(decoder);
2473*4882a593Smuzhiyun 			goto out;
2474*4882a593Smuzhiyun 
2475*4882a593Smuzhiyun 		case INTEL_PT_PSBEND:
2476*4882a593Smuzhiyun 			err = 0;
2477*4882a593Smuzhiyun 			goto out;
2478*4882a593Smuzhiyun 
2479*4882a593Smuzhiyun 		case INTEL_PT_PSB:
2480*4882a593Smuzhiyun 		case INTEL_PT_VMCS:
2481*4882a593Smuzhiyun 		case INTEL_PT_MNT:
2482*4882a593Smuzhiyun 		case INTEL_PT_PAD:
2483*4882a593Smuzhiyun 		default:
2484*4882a593Smuzhiyun 			break;
2485*4882a593Smuzhiyun 		}
2486*4882a593Smuzhiyun 	}
2487*4882a593Smuzhiyun out:
2488*4882a593Smuzhiyun 	decoder->in_psb = false;
2489*4882a593Smuzhiyun 
2490*4882a593Smuzhiyun 	return err;
2491*4882a593Smuzhiyun }
2492*4882a593Smuzhiyun 
intel_pt_walk_to_ip(struct intel_pt_decoder * decoder)2493*4882a593Smuzhiyun static int intel_pt_walk_to_ip(struct intel_pt_decoder *decoder)
2494*4882a593Smuzhiyun {
2495*4882a593Smuzhiyun 	int err;
2496*4882a593Smuzhiyun 
2497*4882a593Smuzhiyun 	while (1) {
2498*4882a593Smuzhiyun 		err = intel_pt_get_next_packet(decoder);
2499*4882a593Smuzhiyun 		if (err)
2500*4882a593Smuzhiyun 			return err;
2501*4882a593Smuzhiyun 
2502*4882a593Smuzhiyun 		switch (decoder->packet.type) {
2503*4882a593Smuzhiyun 		case INTEL_PT_TIP_PGD:
2504*4882a593Smuzhiyun 			decoder->continuous_period = false;
2505*4882a593Smuzhiyun 			decoder->pge = false;
2506*4882a593Smuzhiyun 			if (intel_pt_have_ip(decoder))
2507*4882a593Smuzhiyun 				intel_pt_set_ip(decoder);
2508*4882a593Smuzhiyun 			if (!decoder->ip)
2509*4882a593Smuzhiyun 				break;
2510*4882a593Smuzhiyun 			decoder->state.type |= INTEL_PT_TRACE_END;
2511*4882a593Smuzhiyun 			return 0;
2512*4882a593Smuzhiyun 
2513*4882a593Smuzhiyun 		case INTEL_PT_TIP_PGE:
2514*4882a593Smuzhiyun 			decoder->pge = true;
2515*4882a593Smuzhiyun 			intel_pt_mtc_cyc_cnt_pge(decoder);
2516*4882a593Smuzhiyun 			if (intel_pt_have_ip(decoder))
2517*4882a593Smuzhiyun 				intel_pt_set_ip(decoder);
2518*4882a593Smuzhiyun 			if (!decoder->ip)
2519*4882a593Smuzhiyun 				break;
2520*4882a593Smuzhiyun 			decoder->state.type |= INTEL_PT_TRACE_BEGIN;
2521*4882a593Smuzhiyun 			return 0;
2522*4882a593Smuzhiyun 
2523*4882a593Smuzhiyun 		case INTEL_PT_TIP:
2524*4882a593Smuzhiyun 			decoder->pge = true;
2525*4882a593Smuzhiyun 			if (intel_pt_have_ip(decoder))
2526*4882a593Smuzhiyun 				intel_pt_set_ip(decoder);
2527*4882a593Smuzhiyun 			if (!decoder->ip)
2528*4882a593Smuzhiyun 				break;
2529*4882a593Smuzhiyun 			return 0;
2530*4882a593Smuzhiyun 
2531*4882a593Smuzhiyun 		case INTEL_PT_FUP:
2532*4882a593Smuzhiyun 			if (intel_pt_have_ip(decoder))
2533*4882a593Smuzhiyun 				intel_pt_set_ip(decoder);
2534*4882a593Smuzhiyun 			if (decoder->ip)
2535*4882a593Smuzhiyun 				return 0;
2536*4882a593Smuzhiyun 			break;
2537*4882a593Smuzhiyun 
2538*4882a593Smuzhiyun 		case INTEL_PT_MTC:
2539*4882a593Smuzhiyun 			intel_pt_calc_mtc_timestamp(decoder);
2540*4882a593Smuzhiyun 			break;
2541*4882a593Smuzhiyun 
2542*4882a593Smuzhiyun 		case INTEL_PT_TSC:
2543*4882a593Smuzhiyun 			intel_pt_calc_tsc_timestamp(decoder);
2544*4882a593Smuzhiyun 			break;
2545*4882a593Smuzhiyun 
2546*4882a593Smuzhiyun 		case INTEL_PT_TMA:
2547*4882a593Smuzhiyun 			intel_pt_calc_tma(decoder);
2548*4882a593Smuzhiyun 			break;
2549*4882a593Smuzhiyun 
2550*4882a593Smuzhiyun 		case INTEL_PT_CYC:
2551*4882a593Smuzhiyun 			intel_pt_calc_cyc_timestamp(decoder);
2552*4882a593Smuzhiyun 			break;
2553*4882a593Smuzhiyun 
2554*4882a593Smuzhiyun 		case INTEL_PT_CBR:
2555*4882a593Smuzhiyun 			intel_pt_calc_cbr(decoder);
2556*4882a593Smuzhiyun 			break;
2557*4882a593Smuzhiyun 
2558*4882a593Smuzhiyun 		case INTEL_PT_PIP:
2559*4882a593Smuzhiyun 			decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
2560*4882a593Smuzhiyun 			break;
2561*4882a593Smuzhiyun 
2562*4882a593Smuzhiyun 		case INTEL_PT_MODE_EXEC:
2563*4882a593Smuzhiyun 			decoder->exec_mode = decoder->packet.payload;
2564*4882a593Smuzhiyun 			break;
2565*4882a593Smuzhiyun 
2566*4882a593Smuzhiyun 		case INTEL_PT_MODE_TSX:
2567*4882a593Smuzhiyun 			intel_pt_update_in_tx(decoder);
2568*4882a593Smuzhiyun 			break;
2569*4882a593Smuzhiyun 
2570*4882a593Smuzhiyun 		case INTEL_PT_OVF:
2571*4882a593Smuzhiyun 			return intel_pt_overflow(decoder);
2572*4882a593Smuzhiyun 
2573*4882a593Smuzhiyun 		case INTEL_PT_BAD: /* Does not happen */
2574*4882a593Smuzhiyun 			return intel_pt_bug(decoder);
2575*4882a593Smuzhiyun 
2576*4882a593Smuzhiyun 		case INTEL_PT_TRACESTOP:
2577*4882a593Smuzhiyun 			decoder->pge = false;
2578*4882a593Smuzhiyun 			decoder->continuous_period = false;
2579*4882a593Smuzhiyun 			intel_pt_clear_tx_flags(decoder);
2580*4882a593Smuzhiyun 			decoder->have_tma = false;
2581*4882a593Smuzhiyun 			break;
2582*4882a593Smuzhiyun 
2583*4882a593Smuzhiyun 		case INTEL_PT_PSB:
2584*4882a593Smuzhiyun 			decoder->last_ip = 0;
2585*4882a593Smuzhiyun 			decoder->have_last_ip = true;
2586*4882a593Smuzhiyun 			intel_pt_clear_stack(&decoder->stack);
2587*4882a593Smuzhiyun 			err = intel_pt_walk_psb(decoder);
2588*4882a593Smuzhiyun 			if (err)
2589*4882a593Smuzhiyun 				return err;
2590*4882a593Smuzhiyun 			if (decoder->ip) {
2591*4882a593Smuzhiyun 				/* Do not have a sample */
2592*4882a593Smuzhiyun 				decoder->state.type = 0;
2593*4882a593Smuzhiyun 				return 0;
2594*4882a593Smuzhiyun 			}
2595*4882a593Smuzhiyun 			break;
2596*4882a593Smuzhiyun 
2597*4882a593Smuzhiyun 		case INTEL_PT_TNT:
2598*4882a593Smuzhiyun 		case INTEL_PT_PSBEND:
2599*4882a593Smuzhiyun 		case INTEL_PT_VMCS:
2600*4882a593Smuzhiyun 		case INTEL_PT_MNT:
2601*4882a593Smuzhiyun 		case INTEL_PT_PAD:
2602*4882a593Smuzhiyun 		case INTEL_PT_PTWRITE:
2603*4882a593Smuzhiyun 		case INTEL_PT_PTWRITE_IP:
2604*4882a593Smuzhiyun 		case INTEL_PT_EXSTOP:
2605*4882a593Smuzhiyun 		case INTEL_PT_EXSTOP_IP:
2606*4882a593Smuzhiyun 		case INTEL_PT_MWAIT:
2607*4882a593Smuzhiyun 		case INTEL_PT_PWRE:
2608*4882a593Smuzhiyun 		case INTEL_PT_PWRX:
2609*4882a593Smuzhiyun 		case INTEL_PT_BBP:
2610*4882a593Smuzhiyun 		case INTEL_PT_BIP:
2611*4882a593Smuzhiyun 		case INTEL_PT_BEP:
2612*4882a593Smuzhiyun 		case INTEL_PT_BEP_IP:
2613*4882a593Smuzhiyun 		default:
2614*4882a593Smuzhiyun 			break;
2615*4882a593Smuzhiyun 		}
2616*4882a593Smuzhiyun 	}
2617*4882a593Smuzhiyun }
2618*4882a593Smuzhiyun 
intel_pt_sync_ip(struct intel_pt_decoder * decoder)2619*4882a593Smuzhiyun static int intel_pt_sync_ip(struct intel_pt_decoder *decoder)
2620*4882a593Smuzhiyun {
2621*4882a593Smuzhiyun 	int err;
2622*4882a593Smuzhiyun 
2623*4882a593Smuzhiyun 	decoder->set_fup_tx_flags = false;
2624*4882a593Smuzhiyun 	decoder->set_fup_ptw = false;
2625*4882a593Smuzhiyun 	decoder->set_fup_mwait = false;
2626*4882a593Smuzhiyun 	decoder->set_fup_pwre = false;
2627*4882a593Smuzhiyun 	decoder->set_fup_exstop = false;
2628*4882a593Smuzhiyun 	decoder->set_fup_bep = false;
2629*4882a593Smuzhiyun 	decoder->overflow = false;
2630*4882a593Smuzhiyun 
2631*4882a593Smuzhiyun 	if (!decoder->branch_enable) {
2632*4882a593Smuzhiyun 		decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2633*4882a593Smuzhiyun 		decoder->state.type = 0; /* Do not have a sample */
2634*4882a593Smuzhiyun 		return 0;
2635*4882a593Smuzhiyun 	}
2636*4882a593Smuzhiyun 
2637*4882a593Smuzhiyun 	intel_pt_log("Scanning for full IP\n");
2638*4882a593Smuzhiyun 	err = intel_pt_walk_to_ip(decoder);
2639*4882a593Smuzhiyun 	if (err)
2640*4882a593Smuzhiyun 		return err;
2641*4882a593Smuzhiyun 
2642*4882a593Smuzhiyun 	/* In hop mode, resample to get the to_ip as an "instruction" sample */
2643*4882a593Smuzhiyun 	if (decoder->hop)
2644*4882a593Smuzhiyun 		decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
2645*4882a593Smuzhiyun 	else
2646*4882a593Smuzhiyun 		decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2647*4882a593Smuzhiyun 
2648*4882a593Smuzhiyun 	decoder->state.from_ip = 0;
2649*4882a593Smuzhiyun 	decoder->state.to_ip = decoder->ip;
2650*4882a593Smuzhiyun 	intel_pt_log_to("Setting IP", decoder->ip);
2651*4882a593Smuzhiyun 
2652*4882a593Smuzhiyun 	return 0;
2653*4882a593Smuzhiyun }
2654*4882a593Smuzhiyun 
intel_pt_part_psb(struct intel_pt_decoder * decoder)2655*4882a593Smuzhiyun static int intel_pt_part_psb(struct intel_pt_decoder *decoder)
2656*4882a593Smuzhiyun {
2657*4882a593Smuzhiyun 	const unsigned char *end = decoder->buf + decoder->len;
2658*4882a593Smuzhiyun 	size_t i;
2659*4882a593Smuzhiyun 
2660*4882a593Smuzhiyun 	for (i = INTEL_PT_PSB_LEN - 1; i; i--) {
2661*4882a593Smuzhiyun 		if (i > decoder->len)
2662*4882a593Smuzhiyun 			continue;
2663*4882a593Smuzhiyun 		if (!memcmp(end - i, INTEL_PT_PSB_STR, i))
2664*4882a593Smuzhiyun 			return i;
2665*4882a593Smuzhiyun 	}
2666*4882a593Smuzhiyun 	return 0;
2667*4882a593Smuzhiyun }
2668*4882a593Smuzhiyun 
intel_pt_rest_psb(struct intel_pt_decoder * decoder,int part_psb)2669*4882a593Smuzhiyun static int intel_pt_rest_psb(struct intel_pt_decoder *decoder, int part_psb)
2670*4882a593Smuzhiyun {
2671*4882a593Smuzhiyun 	size_t rest_psb = INTEL_PT_PSB_LEN - part_psb;
2672*4882a593Smuzhiyun 	const char *psb = INTEL_PT_PSB_STR;
2673*4882a593Smuzhiyun 
2674*4882a593Smuzhiyun 	if (rest_psb > decoder->len ||
2675*4882a593Smuzhiyun 	    memcmp(decoder->buf, psb + part_psb, rest_psb))
2676*4882a593Smuzhiyun 		return 0;
2677*4882a593Smuzhiyun 
2678*4882a593Smuzhiyun 	return rest_psb;
2679*4882a593Smuzhiyun }
2680*4882a593Smuzhiyun 
intel_pt_get_split_psb(struct intel_pt_decoder * decoder,int part_psb)2681*4882a593Smuzhiyun static int intel_pt_get_split_psb(struct intel_pt_decoder *decoder,
2682*4882a593Smuzhiyun 				  int part_psb)
2683*4882a593Smuzhiyun {
2684*4882a593Smuzhiyun 	int rest_psb, ret;
2685*4882a593Smuzhiyun 
2686*4882a593Smuzhiyun 	decoder->pos += decoder->len;
2687*4882a593Smuzhiyun 	decoder->len = 0;
2688*4882a593Smuzhiyun 
2689*4882a593Smuzhiyun 	ret = intel_pt_get_next_data(decoder, false);
2690*4882a593Smuzhiyun 	if (ret)
2691*4882a593Smuzhiyun 		return ret;
2692*4882a593Smuzhiyun 
2693*4882a593Smuzhiyun 	rest_psb = intel_pt_rest_psb(decoder, part_psb);
2694*4882a593Smuzhiyun 	if (!rest_psb)
2695*4882a593Smuzhiyun 		return 0;
2696*4882a593Smuzhiyun 
2697*4882a593Smuzhiyun 	decoder->pos -= part_psb;
2698*4882a593Smuzhiyun 	decoder->next_buf = decoder->buf + rest_psb;
2699*4882a593Smuzhiyun 	decoder->next_len = decoder->len - rest_psb;
2700*4882a593Smuzhiyun 	memcpy(decoder->temp_buf, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2701*4882a593Smuzhiyun 	decoder->buf = decoder->temp_buf;
2702*4882a593Smuzhiyun 	decoder->len = INTEL_PT_PSB_LEN;
2703*4882a593Smuzhiyun 
2704*4882a593Smuzhiyun 	return 0;
2705*4882a593Smuzhiyun }
2706*4882a593Smuzhiyun 
intel_pt_scan_for_psb(struct intel_pt_decoder * decoder)2707*4882a593Smuzhiyun static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder)
2708*4882a593Smuzhiyun {
2709*4882a593Smuzhiyun 	unsigned char *next;
2710*4882a593Smuzhiyun 	int ret;
2711*4882a593Smuzhiyun 
2712*4882a593Smuzhiyun 	intel_pt_log("Scanning for PSB\n");
2713*4882a593Smuzhiyun 	while (1) {
2714*4882a593Smuzhiyun 		if (!decoder->len) {
2715*4882a593Smuzhiyun 			ret = intel_pt_get_next_data(decoder, false);
2716*4882a593Smuzhiyun 			if (ret)
2717*4882a593Smuzhiyun 				return ret;
2718*4882a593Smuzhiyun 		}
2719*4882a593Smuzhiyun 
2720*4882a593Smuzhiyun 		next = memmem(decoder->buf, decoder->len, INTEL_PT_PSB_STR,
2721*4882a593Smuzhiyun 			      INTEL_PT_PSB_LEN);
2722*4882a593Smuzhiyun 		if (!next) {
2723*4882a593Smuzhiyun 			int part_psb;
2724*4882a593Smuzhiyun 
2725*4882a593Smuzhiyun 			part_psb = intel_pt_part_psb(decoder);
2726*4882a593Smuzhiyun 			if (part_psb) {
2727*4882a593Smuzhiyun 				ret = intel_pt_get_split_psb(decoder, part_psb);
2728*4882a593Smuzhiyun 				if (ret)
2729*4882a593Smuzhiyun 					return ret;
2730*4882a593Smuzhiyun 			} else {
2731*4882a593Smuzhiyun 				decoder->pos += decoder->len;
2732*4882a593Smuzhiyun 				decoder->len = 0;
2733*4882a593Smuzhiyun 			}
2734*4882a593Smuzhiyun 			continue;
2735*4882a593Smuzhiyun 		}
2736*4882a593Smuzhiyun 
2737*4882a593Smuzhiyun 		decoder->pkt_step = next - decoder->buf;
2738*4882a593Smuzhiyun 		return intel_pt_get_next_packet(decoder);
2739*4882a593Smuzhiyun 	}
2740*4882a593Smuzhiyun }
2741*4882a593Smuzhiyun 
intel_pt_sync(struct intel_pt_decoder * decoder)2742*4882a593Smuzhiyun static int intel_pt_sync(struct intel_pt_decoder *decoder)
2743*4882a593Smuzhiyun {
2744*4882a593Smuzhiyun 	int err;
2745*4882a593Smuzhiyun 
2746*4882a593Smuzhiyun 	decoder->pge = false;
2747*4882a593Smuzhiyun 	decoder->continuous_period = false;
2748*4882a593Smuzhiyun 	decoder->have_last_ip = false;
2749*4882a593Smuzhiyun 	decoder->last_ip = 0;
2750*4882a593Smuzhiyun 	decoder->ip = 0;
2751*4882a593Smuzhiyun 	intel_pt_clear_stack(&decoder->stack);
2752*4882a593Smuzhiyun 
2753*4882a593Smuzhiyun leap:
2754*4882a593Smuzhiyun 	err = intel_pt_scan_for_psb(decoder);
2755*4882a593Smuzhiyun 	if (err)
2756*4882a593Smuzhiyun 		return err;
2757*4882a593Smuzhiyun 
2758*4882a593Smuzhiyun 	decoder->have_last_ip = true;
2759*4882a593Smuzhiyun 	decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2760*4882a593Smuzhiyun 
2761*4882a593Smuzhiyun 	err = intel_pt_walk_psb(decoder);
2762*4882a593Smuzhiyun 	if (err)
2763*4882a593Smuzhiyun 		return err;
2764*4882a593Smuzhiyun 
2765*4882a593Smuzhiyun 	if (decoder->ip) {
2766*4882a593Smuzhiyun 		decoder->state.type = 0; /* Do not have a sample */
2767*4882a593Smuzhiyun 		/*
2768*4882a593Smuzhiyun 		 * In hop mode, resample to get the PSB FUP ip as an
2769*4882a593Smuzhiyun 		 * "instruction" sample.
2770*4882a593Smuzhiyun 		 */
2771*4882a593Smuzhiyun 		if (decoder->hop)
2772*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
2773*4882a593Smuzhiyun 		else
2774*4882a593Smuzhiyun 			decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2775*4882a593Smuzhiyun 	} else if (decoder->leap) {
2776*4882a593Smuzhiyun 		/*
2777*4882a593Smuzhiyun 		 * In leap mode, only PSB+ is decoded, so keeping leaping to the
2778*4882a593Smuzhiyun 		 * next PSB until there is an ip.
2779*4882a593Smuzhiyun 		 */
2780*4882a593Smuzhiyun 		goto leap;
2781*4882a593Smuzhiyun 	} else {
2782*4882a593Smuzhiyun 		return intel_pt_sync_ip(decoder);
2783*4882a593Smuzhiyun 	}
2784*4882a593Smuzhiyun 
2785*4882a593Smuzhiyun 	return 0;
2786*4882a593Smuzhiyun }
2787*4882a593Smuzhiyun 
intel_pt_est_timestamp(struct intel_pt_decoder * decoder)2788*4882a593Smuzhiyun static uint64_t intel_pt_est_timestamp(struct intel_pt_decoder *decoder)
2789*4882a593Smuzhiyun {
2790*4882a593Smuzhiyun 	uint64_t est = decoder->sample_insn_cnt << 1;
2791*4882a593Smuzhiyun 
2792*4882a593Smuzhiyun 	if (!decoder->cbr || !decoder->max_non_turbo_ratio)
2793*4882a593Smuzhiyun 		goto out;
2794*4882a593Smuzhiyun 
2795*4882a593Smuzhiyun 	est *= decoder->max_non_turbo_ratio;
2796*4882a593Smuzhiyun 	est /= decoder->cbr;
2797*4882a593Smuzhiyun out:
2798*4882a593Smuzhiyun 	return decoder->sample_timestamp + est;
2799*4882a593Smuzhiyun }
2800*4882a593Smuzhiyun 
intel_pt_decode(struct intel_pt_decoder * decoder)2801*4882a593Smuzhiyun const struct intel_pt_state *intel_pt_decode(struct intel_pt_decoder *decoder)
2802*4882a593Smuzhiyun {
2803*4882a593Smuzhiyun 	int err;
2804*4882a593Smuzhiyun 
2805*4882a593Smuzhiyun 	do {
2806*4882a593Smuzhiyun 		decoder->state.type = INTEL_PT_BRANCH;
2807*4882a593Smuzhiyun 		decoder->state.flags = 0;
2808*4882a593Smuzhiyun 
2809*4882a593Smuzhiyun 		switch (decoder->pkt_state) {
2810*4882a593Smuzhiyun 		case INTEL_PT_STATE_NO_PSB:
2811*4882a593Smuzhiyun 			err = intel_pt_sync(decoder);
2812*4882a593Smuzhiyun 			break;
2813*4882a593Smuzhiyun 		case INTEL_PT_STATE_NO_IP:
2814*4882a593Smuzhiyun 			decoder->have_last_ip = false;
2815*4882a593Smuzhiyun 			decoder->last_ip = 0;
2816*4882a593Smuzhiyun 			decoder->ip = 0;
2817*4882a593Smuzhiyun 			__fallthrough;
2818*4882a593Smuzhiyun 		case INTEL_PT_STATE_ERR_RESYNC:
2819*4882a593Smuzhiyun 			err = intel_pt_sync_ip(decoder);
2820*4882a593Smuzhiyun 			break;
2821*4882a593Smuzhiyun 		case INTEL_PT_STATE_IN_SYNC:
2822*4882a593Smuzhiyun 			err = intel_pt_walk_trace(decoder);
2823*4882a593Smuzhiyun 			break;
2824*4882a593Smuzhiyun 		case INTEL_PT_STATE_TNT:
2825*4882a593Smuzhiyun 		case INTEL_PT_STATE_TNT_CONT:
2826*4882a593Smuzhiyun 			err = intel_pt_walk_tnt(decoder);
2827*4882a593Smuzhiyun 			if (err == -EAGAIN)
2828*4882a593Smuzhiyun 				err = intel_pt_walk_trace(decoder);
2829*4882a593Smuzhiyun 			break;
2830*4882a593Smuzhiyun 		case INTEL_PT_STATE_TIP:
2831*4882a593Smuzhiyun 		case INTEL_PT_STATE_TIP_PGD:
2832*4882a593Smuzhiyun 			err = intel_pt_walk_tip(decoder);
2833*4882a593Smuzhiyun 			break;
2834*4882a593Smuzhiyun 		case INTEL_PT_STATE_FUP:
2835*4882a593Smuzhiyun 			err = intel_pt_walk_fup(decoder);
2836*4882a593Smuzhiyun 			if (err == -EAGAIN)
2837*4882a593Smuzhiyun 				err = intel_pt_walk_fup_tip(decoder);
2838*4882a593Smuzhiyun 			break;
2839*4882a593Smuzhiyun 		case INTEL_PT_STATE_FUP_NO_TIP:
2840*4882a593Smuzhiyun 			err = intel_pt_walk_fup(decoder);
2841*4882a593Smuzhiyun 			if (err == -EAGAIN)
2842*4882a593Smuzhiyun 				err = intel_pt_walk_trace(decoder);
2843*4882a593Smuzhiyun 			break;
2844*4882a593Smuzhiyun 		case INTEL_PT_STATE_RESAMPLE:
2845*4882a593Smuzhiyun 			err = intel_pt_resample(decoder);
2846*4882a593Smuzhiyun 			break;
2847*4882a593Smuzhiyun 		default:
2848*4882a593Smuzhiyun 			err = intel_pt_bug(decoder);
2849*4882a593Smuzhiyun 			break;
2850*4882a593Smuzhiyun 		}
2851*4882a593Smuzhiyun 	} while (err == -ENOLINK);
2852*4882a593Smuzhiyun 
2853*4882a593Smuzhiyun 	if (err) {
2854*4882a593Smuzhiyun 		decoder->state.err = intel_pt_ext_err(err);
2855*4882a593Smuzhiyun 		if (err != -EOVERFLOW)
2856*4882a593Smuzhiyun 			decoder->state.from_ip = decoder->ip;
2857*4882a593Smuzhiyun 		intel_pt_update_sample_time(decoder);
2858*4882a593Smuzhiyun 		decoder->sample_tot_cyc_cnt = decoder->tot_cyc_cnt;
2859*4882a593Smuzhiyun 	} else {
2860*4882a593Smuzhiyun 		decoder->state.err = 0;
2861*4882a593Smuzhiyun 		if (decoder->cbr != decoder->cbr_seen) {
2862*4882a593Smuzhiyun 			decoder->cbr_seen = decoder->cbr;
2863*4882a593Smuzhiyun 			if (!decoder->state.type) {
2864*4882a593Smuzhiyun 				decoder->state.from_ip = decoder->ip;
2865*4882a593Smuzhiyun 				decoder->state.to_ip = 0;
2866*4882a593Smuzhiyun 			}
2867*4882a593Smuzhiyun 			decoder->state.type |= INTEL_PT_CBR_CHG;
2868*4882a593Smuzhiyun 			decoder->state.cbr_payload = decoder->cbr_payload;
2869*4882a593Smuzhiyun 			decoder->state.cbr = decoder->cbr;
2870*4882a593Smuzhiyun 		}
2871*4882a593Smuzhiyun 		if (intel_pt_sample_time(decoder->pkt_state)) {
2872*4882a593Smuzhiyun 			intel_pt_update_sample_time(decoder);
2873*4882a593Smuzhiyun 			if (decoder->sample_cyc) {
2874*4882a593Smuzhiyun 				decoder->sample_tot_cyc_cnt = decoder->tot_cyc_cnt;
2875*4882a593Smuzhiyun 				decoder->state.flags |= INTEL_PT_SAMPLE_IPC;
2876*4882a593Smuzhiyun 				decoder->sample_cyc = false;
2877*4882a593Smuzhiyun 			}
2878*4882a593Smuzhiyun 		}
2879*4882a593Smuzhiyun 		/*
2880*4882a593Smuzhiyun 		 * When using only TSC/MTC to compute cycles, IPC can be
2881*4882a593Smuzhiyun 		 * sampled as soon as the cycle count changes.
2882*4882a593Smuzhiyun 		 */
2883*4882a593Smuzhiyun 		if (!decoder->have_cyc)
2884*4882a593Smuzhiyun 			decoder->state.flags |= INTEL_PT_SAMPLE_IPC;
2885*4882a593Smuzhiyun 	}
2886*4882a593Smuzhiyun 
2887*4882a593Smuzhiyun 	decoder->state.timestamp = decoder->sample_timestamp;
2888*4882a593Smuzhiyun 	decoder->state.est_timestamp = intel_pt_est_timestamp(decoder);
2889*4882a593Smuzhiyun 	decoder->state.cr3 = decoder->cr3;
2890*4882a593Smuzhiyun 	decoder->state.tot_insn_cnt = decoder->tot_insn_cnt;
2891*4882a593Smuzhiyun 	decoder->state.tot_cyc_cnt = decoder->sample_tot_cyc_cnt;
2892*4882a593Smuzhiyun 
2893*4882a593Smuzhiyun 	return &decoder->state;
2894*4882a593Smuzhiyun }
2895*4882a593Smuzhiyun 
2896*4882a593Smuzhiyun /**
2897*4882a593Smuzhiyun  * intel_pt_next_psb - move buffer pointer to the start of the next PSB packet.
2898*4882a593Smuzhiyun  * @buf: pointer to buffer pointer
2899*4882a593Smuzhiyun  * @len: size of buffer
2900*4882a593Smuzhiyun  *
2901*4882a593Smuzhiyun  * Updates the buffer pointer to point to the start of the next PSB packet if
2902*4882a593Smuzhiyun  * there is one, otherwise the buffer pointer is unchanged.  If @buf is updated,
2903*4882a593Smuzhiyun  * @len is adjusted accordingly.
2904*4882a593Smuzhiyun  *
2905*4882a593Smuzhiyun  * Return: %true if a PSB packet is found, %false otherwise.
2906*4882a593Smuzhiyun  */
intel_pt_next_psb(unsigned char ** buf,size_t * len)2907*4882a593Smuzhiyun static bool intel_pt_next_psb(unsigned char **buf, size_t *len)
2908*4882a593Smuzhiyun {
2909*4882a593Smuzhiyun 	unsigned char *next;
2910*4882a593Smuzhiyun 
2911*4882a593Smuzhiyun 	next = memmem(*buf, *len, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2912*4882a593Smuzhiyun 	if (next) {
2913*4882a593Smuzhiyun 		*len -= next - *buf;
2914*4882a593Smuzhiyun 		*buf = next;
2915*4882a593Smuzhiyun 		return true;
2916*4882a593Smuzhiyun 	}
2917*4882a593Smuzhiyun 	return false;
2918*4882a593Smuzhiyun }
2919*4882a593Smuzhiyun 
2920*4882a593Smuzhiyun /**
2921*4882a593Smuzhiyun  * intel_pt_step_psb - move buffer pointer to the start of the following PSB
2922*4882a593Smuzhiyun  *                     packet.
2923*4882a593Smuzhiyun  * @buf: pointer to buffer pointer
2924*4882a593Smuzhiyun  * @len: size of buffer
2925*4882a593Smuzhiyun  *
2926*4882a593Smuzhiyun  * Updates the buffer pointer to point to the start of the following PSB packet
2927*4882a593Smuzhiyun  * (skipping the PSB at @buf itself) if there is one, otherwise the buffer
2928*4882a593Smuzhiyun  * pointer is unchanged.  If @buf is updated, @len is adjusted accordingly.
2929*4882a593Smuzhiyun  *
2930*4882a593Smuzhiyun  * Return: %true if a PSB packet is found, %false otherwise.
2931*4882a593Smuzhiyun  */
intel_pt_step_psb(unsigned char ** buf,size_t * len)2932*4882a593Smuzhiyun static bool intel_pt_step_psb(unsigned char **buf, size_t *len)
2933*4882a593Smuzhiyun {
2934*4882a593Smuzhiyun 	unsigned char *next;
2935*4882a593Smuzhiyun 
2936*4882a593Smuzhiyun 	if (!*len)
2937*4882a593Smuzhiyun 		return false;
2938*4882a593Smuzhiyun 
2939*4882a593Smuzhiyun 	next = memmem(*buf + 1, *len - 1, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2940*4882a593Smuzhiyun 	if (next) {
2941*4882a593Smuzhiyun 		*len -= next - *buf;
2942*4882a593Smuzhiyun 		*buf = next;
2943*4882a593Smuzhiyun 		return true;
2944*4882a593Smuzhiyun 	}
2945*4882a593Smuzhiyun 	return false;
2946*4882a593Smuzhiyun }
2947*4882a593Smuzhiyun 
2948*4882a593Smuzhiyun /**
2949*4882a593Smuzhiyun  * intel_pt_last_psb - find the last PSB packet in a buffer.
2950*4882a593Smuzhiyun  * @buf: buffer
2951*4882a593Smuzhiyun  * @len: size of buffer
2952*4882a593Smuzhiyun  *
2953*4882a593Smuzhiyun  * This function finds the last PSB in a buffer.
2954*4882a593Smuzhiyun  *
2955*4882a593Smuzhiyun  * Return: A pointer to the last PSB in @buf if found, %NULL otherwise.
2956*4882a593Smuzhiyun  */
intel_pt_last_psb(unsigned char * buf,size_t len)2957*4882a593Smuzhiyun static unsigned char *intel_pt_last_psb(unsigned char *buf, size_t len)
2958*4882a593Smuzhiyun {
2959*4882a593Smuzhiyun 	const char *n = INTEL_PT_PSB_STR;
2960*4882a593Smuzhiyun 	unsigned char *p;
2961*4882a593Smuzhiyun 	size_t k;
2962*4882a593Smuzhiyun 
2963*4882a593Smuzhiyun 	if (len < INTEL_PT_PSB_LEN)
2964*4882a593Smuzhiyun 		return NULL;
2965*4882a593Smuzhiyun 
2966*4882a593Smuzhiyun 	k = len - INTEL_PT_PSB_LEN + 1;
2967*4882a593Smuzhiyun 	while (1) {
2968*4882a593Smuzhiyun 		p = memrchr(buf, n[0], k);
2969*4882a593Smuzhiyun 		if (!p)
2970*4882a593Smuzhiyun 			return NULL;
2971*4882a593Smuzhiyun 		if (!memcmp(p + 1, n + 1, INTEL_PT_PSB_LEN - 1))
2972*4882a593Smuzhiyun 			return p;
2973*4882a593Smuzhiyun 		k = p - buf;
2974*4882a593Smuzhiyun 		if (!k)
2975*4882a593Smuzhiyun 			return NULL;
2976*4882a593Smuzhiyun 	}
2977*4882a593Smuzhiyun }
2978*4882a593Smuzhiyun 
2979*4882a593Smuzhiyun /**
2980*4882a593Smuzhiyun  * intel_pt_next_tsc - find and return next TSC.
2981*4882a593Smuzhiyun  * @buf: buffer
2982*4882a593Smuzhiyun  * @len: size of buffer
2983*4882a593Smuzhiyun  * @tsc: TSC value returned
2984*4882a593Smuzhiyun  * @rem: returns remaining size when TSC is found
2985*4882a593Smuzhiyun  *
2986*4882a593Smuzhiyun  * Find a TSC packet in @buf and return the TSC value.  This function assumes
2987*4882a593Smuzhiyun  * that @buf starts at a PSB and that PSB+ will contain TSC and so stops if a
2988*4882a593Smuzhiyun  * PSBEND packet is found.
2989*4882a593Smuzhiyun  *
2990*4882a593Smuzhiyun  * Return: %true if TSC is found, false otherwise.
2991*4882a593Smuzhiyun  */
intel_pt_next_tsc(unsigned char * buf,size_t len,uint64_t * tsc,size_t * rem)2992*4882a593Smuzhiyun static bool intel_pt_next_tsc(unsigned char *buf, size_t len, uint64_t *tsc,
2993*4882a593Smuzhiyun 			      size_t *rem)
2994*4882a593Smuzhiyun {
2995*4882a593Smuzhiyun 	enum intel_pt_pkt_ctx ctx = INTEL_PT_NO_CTX;
2996*4882a593Smuzhiyun 	struct intel_pt_pkt packet;
2997*4882a593Smuzhiyun 	int ret;
2998*4882a593Smuzhiyun 
2999*4882a593Smuzhiyun 	while (len) {
3000*4882a593Smuzhiyun 		ret = intel_pt_get_packet(buf, len, &packet, &ctx);
3001*4882a593Smuzhiyun 		if (ret <= 0)
3002*4882a593Smuzhiyun 			return false;
3003*4882a593Smuzhiyun 		if (packet.type == INTEL_PT_TSC) {
3004*4882a593Smuzhiyun 			*tsc = packet.payload;
3005*4882a593Smuzhiyun 			*rem = len;
3006*4882a593Smuzhiyun 			return true;
3007*4882a593Smuzhiyun 		}
3008*4882a593Smuzhiyun 		if (packet.type == INTEL_PT_PSBEND)
3009*4882a593Smuzhiyun 			return false;
3010*4882a593Smuzhiyun 		buf += ret;
3011*4882a593Smuzhiyun 		len -= ret;
3012*4882a593Smuzhiyun 	}
3013*4882a593Smuzhiyun 	return false;
3014*4882a593Smuzhiyun }
3015*4882a593Smuzhiyun 
3016*4882a593Smuzhiyun /**
3017*4882a593Smuzhiyun  * intel_pt_tsc_cmp - compare 7-byte TSCs.
3018*4882a593Smuzhiyun  * @tsc1: first TSC to compare
3019*4882a593Smuzhiyun  * @tsc2: second TSC to compare
3020*4882a593Smuzhiyun  *
3021*4882a593Smuzhiyun  * This function compares 7-byte TSC values allowing for the possibility that
3022*4882a593Smuzhiyun  * TSC wrapped around.  Generally it is not possible to know if TSC has wrapped
3023*4882a593Smuzhiyun  * around so for that purpose this function assumes the absolute difference is
3024*4882a593Smuzhiyun  * less than half the maximum difference.
3025*4882a593Smuzhiyun  *
3026*4882a593Smuzhiyun  * Return: %-1 if @tsc1 is before @tsc2, %0 if @tsc1 == @tsc2, %1 if @tsc1 is
3027*4882a593Smuzhiyun  * after @tsc2.
3028*4882a593Smuzhiyun  */
intel_pt_tsc_cmp(uint64_t tsc1,uint64_t tsc2)3029*4882a593Smuzhiyun static int intel_pt_tsc_cmp(uint64_t tsc1, uint64_t tsc2)
3030*4882a593Smuzhiyun {
3031*4882a593Smuzhiyun 	const uint64_t halfway = (1ULL << 55);
3032*4882a593Smuzhiyun 
3033*4882a593Smuzhiyun 	if (tsc1 == tsc2)
3034*4882a593Smuzhiyun 		return 0;
3035*4882a593Smuzhiyun 
3036*4882a593Smuzhiyun 	if (tsc1 < tsc2) {
3037*4882a593Smuzhiyun 		if (tsc2 - tsc1 < halfway)
3038*4882a593Smuzhiyun 			return -1;
3039*4882a593Smuzhiyun 		else
3040*4882a593Smuzhiyun 			return 1;
3041*4882a593Smuzhiyun 	} else {
3042*4882a593Smuzhiyun 		if (tsc1 - tsc2 < halfway)
3043*4882a593Smuzhiyun 			return 1;
3044*4882a593Smuzhiyun 		else
3045*4882a593Smuzhiyun 			return -1;
3046*4882a593Smuzhiyun 	}
3047*4882a593Smuzhiyun }
3048*4882a593Smuzhiyun 
3049*4882a593Smuzhiyun #define MAX_PADDING (PERF_AUXTRACE_RECORD_ALIGNMENT - 1)
3050*4882a593Smuzhiyun 
3051*4882a593Smuzhiyun /**
3052*4882a593Smuzhiyun  * adj_for_padding - adjust overlap to account for padding.
3053*4882a593Smuzhiyun  * @buf_b: second buffer
3054*4882a593Smuzhiyun  * @buf_a: first buffer
3055*4882a593Smuzhiyun  * @len_a: size of first buffer
3056*4882a593Smuzhiyun  *
3057*4882a593Smuzhiyun  * @buf_a might have up to 7 bytes of padding appended. Adjust the overlap
3058*4882a593Smuzhiyun  * accordingly.
3059*4882a593Smuzhiyun  *
3060*4882a593Smuzhiyun  * Return: A pointer into @buf_b from where non-overlapped data starts
3061*4882a593Smuzhiyun  */
adj_for_padding(unsigned char * buf_b,unsigned char * buf_a,size_t len_a)3062*4882a593Smuzhiyun static unsigned char *adj_for_padding(unsigned char *buf_b,
3063*4882a593Smuzhiyun 				      unsigned char *buf_a, size_t len_a)
3064*4882a593Smuzhiyun {
3065*4882a593Smuzhiyun 	unsigned char *p = buf_b - MAX_PADDING;
3066*4882a593Smuzhiyun 	unsigned char *q = buf_a + len_a - MAX_PADDING;
3067*4882a593Smuzhiyun 	int i;
3068*4882a593Smuzhiyun 
3069*4882a593Smuzhiyun 	for (i = MAX_PADDING; i; i--, p++, q++) {
3070*4882a593Smuzhiyun 		if (*p != *q)
3071*4882a593Smuzhiyun 			break;
3072*4882a593Smuzhiyun 	}
3073*4882a593Smuzhiyun 
3074*4882a593Smuzhiyun 	return p;
3075*4882a593Smuzhiyun }
3076*4882a593Smuzhiyun 
3077*4882a593Smuzhiyun /**
3078*4882a593Smuzhiyun  * intel_pt_find_overlap_tsc - determine start of non-overlapped trace data
3079*4882a593Smuzhiyun  *                             using TSC.
3080*4882a593Smuzhiyun  * @buf_a: first buffer
3081*4882a593Smuzhiyun  * @len_a: size of first buffer
3082*4882a593Smuzhiyun  * @buf_b: second buffer
3083*4882a593Smuzhiyun  * @len_b: size of second buffer
3084*4882a593Smuzhiyun  * @consecutive: returns true if there is data in buf_b that is consecutive
3085*4882a593Smuzhiyun  *               to buf_a
3086*4882a593Smuzhiyun  *
3087*4882a593Smuzhiyun  * If the trace contains TSC we can look at the last TSC of @buf_a and the
3088*4882a593Smuzhiyun  * first TSC of @buf_b in order to determine if the buffers overlap, and then
3089*4882a593Smuzhiyun  * walk forward in @buf_b until a later TSC is found.  A precondition is that
3090*4882a593Smuzhiyun  * @buf_a and @buf_b are positioned at a PSB.
3091*4882a593Smuzhiyun  *
3092*4882a593Smuzhiyun  * Return: A pointer into @buf_b from where non-overlapped data starts, or
3093*4882a593Smuzhiyun  * @buf_b + @len_b if there is no non-overlapped data.
3094*4882a593Smuzhiyun  */
intel_pt_find_overlap_tsc(unsigned char * buf_a,size_t len_a,unsigned char * buf_b,size_t len_b,bool * consecutive)3095*4882a593Smuzhiyun static unsigned char *intel_pt_find_overlap_tsc(unsigned char *buf_a,
3096*4882a593Smuzhiyun 						size_t len_a,
3097*4882a593Smuzhiyun 						unsigned char *buf_b,
3098*4882a593Smuzhiyun 						size_t len_b, bool *consecutive)
3099*4882a593Smuzhiyun {
3100*4882a593Smuzhiyun 	uint64_t tsc_a, tsc_b;
3101*4882a593Smuzhiyun 	unsigned char *p;
3102*4882a593Smuzhiyun 	size_t len, rem_a, rem_b;
3103*4882a593Smuzhiyun 
3104*4882a593Smuzhiyun 	p = intel_pt_last_psb(buf_a, len_a);
3105*4882a593Smuzhiyun 	if (!p)
3106*4882a593Smuzhiyun 		return buf_b; /* No PSB in buf_a => no overlap */
3107*4882a593Smuzhiyun 
3108*4882a593Smuzhiyun 	len = len_a - (p - buf_a);
3109*4882a593Smuzhiyun 	if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a)) {
3110*4882a593Smuzhiyun 		/* The last PSB+ in buf_a is incomplete, so go back one more */
3111*4882a593Smuzhiyun 		len_a -= len;
3112*4882a593Smuzhiyun 		p = intel_pt_last_psb(buf_a, len_a);
3113*4882a593Smuzhiyun 		if (!p)
3114*4882a593Smuzhiyun 			return buf_b; /* No full PSB+ => assume no overlap */
3115*4882a593Smuzhiyun 		len = len_a - (p - buf_a);
3116*4882a593Smuzhiyun 		if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a))
3117*4882a593Smuzhiyun 			return buf_b; /* No TSC in buf_a => assume no overlap */
3118*4882a593Smuzhiyun 	}
3119*4882a593Smuzhiyun 
3120*4882a593Smuzhiyun 	while (1) {
3121*4882a593Smuzhiyun 		/* Ignore PSB+ with no TSC */
3122*4882a593Smuzhiyun 		if (intel_pt_next_tsc(buf_b, len_b, &tsc_b, &rem_b)) {
3123*4882a593Smuzhiyun 			int cmp = intel_pt_tsc_cmp(tsc_a, tsc_b);
3124*4882a593Smuzhiyun 
3125*4882a593Smuzhiyun 			/* Same TSC, so buffers are consecutive */
3126*4882a593Smuzhiyun 			if (!cmp && rem_b >= rem_a) {
3127*4882a593Smuzhiyun 				unsigned char *start;
3128*4882a593Smuzhiyun 
3129*4882a593Smuzhiyun 				*consecutive = true;
3130*4882a593Smuzhiyun 				start = buf_b + len_b - (rem_b - rem_a);
3131*4882a593Smuzhiyun 				return adj_for_padding(start, buf_a, len_a);
3132*4882a593Smuzhiyun 			}
3133*4882a593Smuzhiyun 			if (cmp < 0)
3134*4882a593Smuzhiyun 				return buf_b; /* tsc_a < tsc_b => no overlap */
3135*4882a593Smuzhiyun 		}
3136*4882a593Smuzhiyun 
3137*4882a593Smuzhiyun 		if (!intel_pt_step_psb(&buf_b, &len_b))
3138*4882a593Smuzhiyun 			return buf_b + len_b; /* No PSB in buf_b => no data */
3139*4882a593Smuzhiyun 	}
3140*4882a593Smuzhiyun }
3141*4882a593Smuzhiyun 
3142*4882a593Smuzhiyun /**
3143*4882a593Smuzhiyun  * intel_pt_find_overlap - determine start of non-overlapped trace data.
3144*4882a593Smuzhiyun  * @buf_a: first buffer
3145*4882a593Smuzhiyun  * @len_a: size of first buffer
3146*4882a593Smuzhiyun  * @buf_b: second buffer
3147*4882a593Smuzhiyun  * @len_b: size of second buffer
3148*4882a593Smuzhiyun  * @have_tsc: can use TSC packets to detect overlap
3149*4882a593Smuzhiyun  * @consecutive: returns true if there is data in buf_b that is consecutive
3150*4882a593Smuzhiyun  *               to buf_a
3151*4882a593Smuzhiyun  *
3152*4882a593Smuzhiyun  * When trace samples or snapshots are recorded there is the possibility that
3153*4882a593Smuzhiyun  * the data overlaps.  Note that, for the purposes of decoding, data is only
3154*4882a593Smuzhiyun  * useful if it begins with a PSB packet.
3155*4882a593Smuzhiyun  *
3156*4882a593Smuzhiyun  * Return: A pointer into @buf_b from where non-overlapped data starts, or
3157*4882a593Smuzhiyun  * @buf_b + @len_b if there is no non-overlapped data.
3158*4882a593Smuzhiyun  */
intel_pt_find_overlap(unsigned char * buf_a,size_t len_a,unsigned char * buf_b,size_t len_b,bool have_tsc,bool * consecutive)3159*4882a593Smuzhiyun unsigned char *intel_pt_find_overlap(unsigned char *buf_a, size_t len_a,
3160*4882a593Smuzhiyun 				     unsigned char *buf_b, size_t len_b,
3161*4882a593Smuzhiyun 				     bool have_tsc, bool *consecutive)
3162*4882a593Smuzhiyun {
3163*4882a593Smuzhiyun 	unsigned char *found;
3164*4882a593Smuzhiyun 
3165*4882a593Smuzhiyun 	/* Buffer 'b' must start at PSB so throw away everything before that */
3166*4882a593Smuzhiyun 	if (!intel_pt_next_psb(&buf_b, &len_b))
3167*4882a593Smuzhiyun 		return buf_b + len_b; /* No PSB */
3168*4882a593Smuzhiyun 
3169*4882a593Smuzhiyun 	if (!intel_pt_next_psb(&buf_a, &len_a))
3170*4882a593Smuzhiyun 		return buf_b; /* No overlap */
3171*4882a593Smuzhiyun 
3172*4882a593Smuzhiyun 	if (have_tsc) {
3173*4882a593Smuzhiyun 		found = intel_pt_find_overlap_tsc(buf_a, len_a, buf_b, len_b,
3174*4882a593Smuzhiyun 						  consecutive);
3175*4882a593Smuzhiyun 		if (found)
3176*4882a593Smuzhiyun 			return found;
3177*4882a593Smuzhiyun 	}
3178*4882a593Smuzhiyun 
3179*4882a593Smuzhiyun 	/*
3180*4882a593Smuzhiyun 	 * Buffer 'b' cannot end within buffer 'a' so, for comparison purposes,
3181*4882a593Smuzhiyun 	 * we can ignore the first part of buffer 'a'.
3182*4882a593Smuzhiyun 	 */
3183*4882a593Smuzhiyun 	while (len_b < len_a) {
3184*4882a593Smuzhiyun 		if (!intel_pt_step_psb(&buf_a, &len_a))
3185*4882a593Smuzhiyun 			return buf_b; /* No overlap */
3186*4882a593Smuzhiyun 	}
3187*4882a593Smuzhiyun 
3188*4882a593Smuzhiyun 	/* Now len_b >= len_a */
3189*4882a593Smuzhiyun 	while (1) {
3190*4882a593Smuzhiyun 		/* Potential overlap so check the bytes */
3191*4882a593Smuzhiyun 		found = memmem(buf_a, len_a, buf_b, len_a);
3192*4882a593Smuzhiyun 		if (found) {
3193*4882a593Smuzhiyun 			*consecutive = true;
3194*4882a593Smuzhiyun 			return adj_for_padding(buf_b + len_a, buf_a, len_a);
3195*4882a593Smuzhiyun 		}
3196*4882a593Smuzhiyun 
3197*4882a593Smuzhiyun 		/* Try again at next PSB in buffer 'a' */
3198*4882a593Smuzhiyun 		if (!intel_pt_step_psb(&buf_a, &len_a))
3199*4882a593Smuzhiyun 			return buf_b; /* No overlap */
3200*4882a593Smuzhiyun 	}
3201*4882a593Smuzhiyun }
3202*4882a593Smuzhiyun 
3203*4882a593Smuzhiyun /**
3204*4882a593Smuzhiyun  * struct fast_forward_data - data used by intel_pt_ff_cb().
3205*4882a593Smuzhiyun  * @timestamp: timestamp to fast forward towards
3206*4882a593Smuzhiyun  * @buf_timestamp: buffer timestamp of last buffer with trace data earlier than
3207*4882a593Smuzhiyun  *                 the fast forward timestamp.
3208*4882a593Smuzhiyun  */
3209*4882a593Smuzhiyun struct fast_forward_data {
3210*4882a593Smuzhiyun 	uint64_t timestamp;
3211*4882a593Smuzhiyun 	uint64_t buf_timestamp;
3212*4882a593Smuzhiyun };
3213*4882a593Smuzhiyun 
3214*4882a593Smuzhiyun /**
3215*4882a593Smuzhiyun  * intel_pt_ff_cb - fast forward lookahead callback.
3216*4882a593Smuzhiyun  * @buffer: Intel PT trace buffer
3217*4882a593Smuzhiyun  * @data: opaque pointer to fast forward data (struct fast_forward_data)
3218*4882a593Smuzhiyun  *
3219*4882a593Smuzhiyun  * Determine if @buffer trace is past the fast forward timestamp.
3220*4882a593Smuzhiyun  *
3221*4882a593Smuzhiyun  * Return: 1 (stop lookahead) if @buffer trace is past the fast forward
3222*4882a593Smuzhiyun  *         timestamp, and 0 otherwise.
3223*4882a593Smuzhiyun  */
intel_pt_ff_cb(struct intel_pt_buffer * buffer,void * data)3224*4882a593Smuzhiyun static int intel_pt_ff_cb(struct intel_pt_buffer *buffer, void *data)
3225*4882a593Smuzhiyun {
3226*4882a593Smuzhiyun 	struct fast_forward_data *d = data;
3227*4882a593Smuzhiyun 	unsigned char *buf;
3228*4882a593Smuzhiyun 	uint64_t tsc;
3229*4882a593Smuzhiyun 	size_t rem;
3230*4882a593Smuzhiyun 	size_t len;
3231*4882a593Smuzhiyun 
3232*4882a593Smuzhiyun 	buf = (unsigned char *)buffer->buf;
3233*4882a593Smuzhiyun 	len = buffer->len;
3234*4882a593Smuzhiyun 
3235*4882a593Smuzhiyun 	if (!intel_pt_next_psb(&buf, &len) ||
3236*4882a593Smuzhiyun 	    !intel_pt_next_tsc(buf, len, &tsc, &rem))
3237*4882a593Smuzhiyun 		return 0;
3238*4882a593Smuzhiyun 
3239*4882a593Smuzhiyun 	tsc = intel_pt_8b_tsc(tsc, buffer->ref_timestamp);
3240*4882a593Smuzhiyun 
3241*4882a593Smuzhiyun 	intel_pt_log("Buffer 1st timestamp " x64_fmt " ref timestamp " x64_fmt "\n",
3242*4882a593Smuzhiyun 		     tsc, buffer->ref_timestamp);
3243*4882a593Smuzhiyun 
3244*4882a593Smuzhiyun 	/*
3245*4882a593Smuzhiyun 	 * If the buffer contains a timestamp earlier that the fast forward
3246*4882a593Smuzhiyun 	 * timestamp, then record it, else stop.
3247*4882a593Smuzhiyun 	 */
3248*4882a593Smuzhiyun 	if (tsc < d->timestamp)
3249*4882a593Smuzhiyun 		d->buf_timestamp = buffer->ref_timestamp;
3250*4882a593Smuzhiyun 	else
3251*4882a593Smuzhiyun 		return 1;
3252*4882a593Smuzhiyun 
3253*4882a593Smuzhiyun 	return 0;
3254*4882a593Smuzhiyun }
3255*4882a593Smuzhiyun 
3256*4882a593Smuzhiyun /**
3257*4882a593Smuzhiyun  * intel_pt_fast_forward - reposition decoder forwards.
3258*4882a593Smuzhiyun  * @decoder: Intel PT decoder
3259*4882a593Smuzhiyun  * @timestamp: timestamp to fast forward towards
3260*4882a593Smuzhiyun  *
3261*4882a593Smuzhiyun  * Reposition decoder at the last PSB with a timestamp earlier than @timestamp.
3262*4882a593Smuzhiyun  *
3263*4882a593Smuzhiyun  * Return: 0 on success or negative error code on failure.
3264*4882a593Smuzhiyun  */
intel_pt_fast_forward(struct intel_pt_decoder * decoder,uint64_t timestamp)3265*4882a593Smuzhiyun int intel_pt_fast_forward(struct intel_pt_decoder *decoder, uint64_t timestamp)
3266*4882a593Smuzhiyun {
3267*4882a593Smuzhiyun 	struct fast_forward_data d = { .timestamp = timestamp };
3268*4882a593Smuzhiyun 	unsigned char *buf;
3269*4882a593Smuzhiyun 	size_t len;
3270*4882a593Smuzhiyun 	int err;
3271*4882a593Smuzhiyun 
3272*4882a593Smuzhiyun 	intel_pt_log("Fast forward towards timestamp " x64_fmt "\n", timestamp);
3273*4882a593Smuzhiyun 
3274*4882a593Smuzhiyun 	/* Find buffer timestamp of buffer to fast forward to */
3275*4882a593Smuzhiyun 	err = decoder->lookahead(decoder->data, intel_pt_ff_cb, &d);
3276*4882a593Smuzhiyun 	if (err < 0)
3277*4882a593Smuzhiyun 		return err;
3278*4882a593Smuzhiyun 
3279*4882a593Smuzhiyun 	/* Walk to buffer with same buffer timestamp */
3280*4882a593Smuzhiyun 	if (d.buf_timestamp) {
3281*4882a593Smuzhiyun 		do {
3282*4882a593Smuzhiyun 			decoder->pos += decoder->len;
3283*4882a593Smuzhiyun 			decoder->len = 0;
3284*4882a593Smuzhiyun 			err = intel_pt_get_next_data(decoder, true);
3285*4882a593Smuzhiyun 			/* -ENOLINK means non-consecutive trace */
3286*4882a593Smuzhiyun 			if (err && err != -ENOLINK)
3287*4882a593Smuzhiyun 				return err;
3288*4882a593Smuzhiyun 		} while (decoder->buf_timestamp != d.buf_timestamp);
3289*4882a593Smuzhiyun 	}
3290*4882a593Smuzhiyun 
3291*4882a593Smuzhiyun 	if (!decoder->buf)
3292*4882a593Smuzhiyun 		return 0;
3293*4882a593Smuzhiyun 
3294*4882a593Smuzhiyun 	buf = (unsigned char *)decoder->buf;
3295*4882a593Smuzhiyun 	len = decoder->len;
3296*4882a593Smuzhiyun 
3297*4882a593Smuzhiyun 	if (!intel_pt_next_psb(&buf, &len))
3298*4882a593Smuzhiyun 		return 0;
3299*4882a593Smuzhiyun 
3300*4882a593Smuzhiyun 	/*
3301*4882a593Smuzhiyun 	 * Walk PSBs while the PSB timestamp is less than the fast forward
3302*4882a593Smuzhiyun 	 * timestamp.
3303*4882a593Smuzhiyun 	 */
3304*4882a593Smuzhiyun 	do {
3305*4882a593Smuzhiyun 		uint64_t tsc;
3306*4882a593Smuzhiyun 		size_t rem;
3307*4882a593Smuzhiyun 
3308*4882a593Smuzhiyun 		if (!intel_pt_next_tsc(buf, len, &tsc, &rem))
3309*4882a593Smuzhiyun 			break;
3310*4882a593Smuzhiyun 		tsc = intel_pt_8b_tsc(tsc, decoder->buf_timestamp);
3311*4882a593Smuzhiyun 		/*
3312*4882a593Smuzhiyun 		 * A TSC packet can slip past MTC packets but, after fast
3313*4882a593Smuzhiyun 		 * forward, decoding starts at the TSC timestamp. That means
3314*4882a593Smuzhiyun 		 * the timestamps may not be exactly the same as the timestamps
3315*4882a593Smuzhiyun 		 * that would have been decoded without fast forward.
3316*4882a593Smuzhiyun 		 */
3317*4882a593Smuzhiyun 		if (tsc < timestamp) {
3318*4882a593Smuzhiyun 			intel_pt_log("Fast forward to next PSB timestamp " x64_fmt "\n", tsc);
3319*4882a593Smuzhiyun 			decoder->pos += decoder->len - len;
3320*4882a593Smuzhiyun 			decoder->buf = buf;
3321*4882a593Smuzhiyun 			decoder->len = len;
3322*4882a593Smuzhiyun 			intel_pt_reposition(decoder);
3323*4882a593Smuzhiyun 		} else {
3324*4882a593Smuzhiyun 			break;
3325*4882a593Smuzhiyun 		}
3326*4882a593Smuzhiyun 	} while (intel_pt_step_psb(&buf, &len));
3327*4882a593Smuzhiyun 
3328*4882a593Smuzhiyun 	return 0;
3329*4882a593Smuzhiyun }
3330