perf intel-pt-decoder: Report instruction bytes and length in sample
PATCH 1/2 perf intel-pt-decoder: Report instruction bytes and length in sample
From: Andi Kleen <[email protected]>
Change the Intel PT decoder to pass up the length and the instruction bytes of the decoded or sampled instruction in the perf sample.
The decoder already knows this information, we just need to pass it up. Since it is only a couple of movs it is not very expensive.
Used in the next patch.
--- tools/perf/util/event.h | 3 +++ tools/perf/util/intel-pt-decoder/intel-pt-decoder.c | 2 ++ tools/perf/util/intel-pt-decoder/intel-pt-decoder.h | 3 +++ tools/perf/util/intel-pt-decoder/intel-pt-insn-decoder.h | 2 +- tools/perf/util/intel-pt.c | 5 +++++ 5 files changed, 14 insertions(+), 1 deletion(-)
tools/perf/util/event.h +3 -0
@@ -177,6 +177,8 @@ enum { PERF_IP_FLAG_TRACE_BEGIN |\ PERF_IP_FLAG_TRACE_END) +#define MAX_INSN 16 +struct perf_sample {u64 ip;u32 pid, tid;
@@ -193,6 +195,7 @@ struct perf_sample {u32 flags;u16 insn_len;u8 cpumode; +char insn[MAX_INSN ];void *raw_data;struct ip_callchain *callchain;struct branch_stack *branch_stack;
tools/perf/util/intel-pt-decoder/intel-pt-decoder.c +2 -0
@@ -949,6 +949,8 @@ out: out_no_progress: decoder->state.insn_op = intel_pt_insn->op; decoder->state.insn_len = intel_pt_insn->length; +memcpy (decoder->state.insn, intel_pt_insn->buf, +sizeof (decoder->state.insn));if (decoder->tx_flags &INTEL_PT_IN_TX ) decoder->state.flags |=INTEL_PT_IN_TX ;
tools/perf/util/intel-pt-decoder/intel-pt-decoder.h +3 -0
@@ -20,6 +20,8 @@#include <stddef.h> #include <stdbool.h> +#define MAX_INSN 16 +#include "intel-pt-insn-decoder.h" #define INTEL_PT_IN_TX (1 << 0)
@@ -66,6 +68,7 @@ struct intel_pt_state {uint32_t flags;enum intel_pt_insn_op insn_op;int insn_len; +char insn[MAX_INSN ]; };struct intel_pt_insn ;
tools/perf/util/intel-pt-decoder/intel-pt-insn-decoder.h +1 -1
@@ -20,7 +20,7 @@#include <stdint.h> #define INTEL_PT_INSN_DESC_MAX 32 -#define INTEL_PT_INSN_DBG_BUF_SZ 16 +#define INTEL_PT_INSN_DBG_BUF_SZ 16/* Must be >= MAX_INSN */ enum intel_pt_insn_op {INTEL_PT_OP_OTHER ,
tools/perf/util/intel-pt.c +5 -0
@@ -140,6 +140,7 @@ struct intel_pt_queue {u32 flags;u16 insn_len;u64 last_insn_cnt; +char insn[MAX_INSN ]; }; static void intel_pt_dump(struct intel_pt *pt __maybe_unused,
@@ -817,6 +818,7 @@ static void intel_pt_sample_flags(struct intel_pt_queue *ptq)if (ptq->state->flags &INTEL_PT_IN_TX ) ptq->flags |=PERF_IP_FLAG_IN_TX ; ptq->insn_len = ptq->state->insn_len; +memcpy (ptq->insn, ptq->state->insn,MAX_INSN ); } }
@@ -997,6 +999,7 @@ static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq) sample.cpu = ptq->cpu; sample.flags = ptq->flags; sample.insn_len = ptq->insn_len; +memcpy (sample.insn, ptq->insn,MAX_INSN );/* * perf report cannot handle events without a branch stack when using
@@ -1058,6 +1061,7 @@ static int intel_pt_synth_instruction_sample(struct intel_pt_queue *ptq) sample.cpu = ptq->cpu; sample.flags = ptq->flags; sample.insn_len = ptq->insn_len; +memcpy (sample.insn, ptq->insn,MAX_INSN ); ptq->last_insn_cnt = ptq->state->tot_insn_cnt;
@@ -1120,6 +1124,7 @@ static int intel_pt_synth_transaction_sample(struct intel_pt_queue *ptq) sample.cpu = ptq->cpu; sample.flags = ptq->flags; sample.insn_len = ptq->insn_len; +memcpy (sample.insn, ptq->insn,MAX_INSN );if (pt->synth_opts.callchain) { thread_stack__sample(ptq->thread, ptq->chain,
Signature · 1 line
2.5.5
PATCH 2/2 perf tools: Support insn and insnlen in perf script
From: Andi Kleen <[email protected]>
When looking at Intel PT traces with perf script it is useful to have some indication of the instruction. Dump the instruction bytes and instruction length, which can be used for simple pattern analysis in scripts.
% perf record -e intel_pt// foo
% perf script --itrace=i0ns -F ip,insn,insnlen
ffffffff8101232f ilen: 5 insn: 0f 1f 44 00 00 ffffffff81012334 ilen: 1 insn: 5b ffffffff81012335 ilen: 1 insn: 5d ffffffff81012336 ilen: 1 insn: c3 ffffffff810123e3 ilen: 1 insn: 5b ffffffff810123e4 ilen: 2 insn: 41 5c ffffffff810123e6 ilen: 1 insn: 5d ffffffff810123e7 ilen: 1 insn: c3 ffffffff810124a6 ilen: 2 insn: 31 c0 ffffffff810124a8 ilen: 9 insn: 41 83 bc 24 a8 01 00 00 01 ffffffff810124b1 ilen: 2 insn: 75 87
...
--- tools/perf/Documentation/perf-script.txt | 6 +++++- tools/perf/builtin-script.c | 25 +++++++++++++++++++++++-- 2 files changed, 28 insertions(+), 3 deletions(-)
tools/perf/Documentation/perf-script.txt +5 -1
@@ -117,7 +117,7 @@ OPTIONS Comma separated list of fields to print. Options are: comm, tid, pid, time, cpu, event, trace, ip, sym, dso, addr, symoff, srcline, period, iregs, brstack, brstacksym, flags, bpf-output, - callindent. Field list can be prepended with the type, trace, sw or hw, + callindent, insn, insnlen. Field list can be prepended with the type, trace, sw or hw, to indicate to which event type the field list applies. e.g., -F sw:comm,tid,time,ip,sym and -F trace:time,cpu,trace
@@ -181,6 +181,10 @@ OPTIONS Instruction Trace decoding. For calls and returns, it will display the name of the symbol indented with spaces to reflect the stack depth. + When doing instruction trace decoding insn and insnlen give the + instruction bytes and the instruction length of the current + instruction. + Finally, a user may not set fields to none for all event types. i.e., -F "" is not allowed.
tools/perf/builtin-script.c +23 -2
@@ -66,6 +66,8 @@ enum perf_output_field {PERF_OUTPUT_WEIGHT =1U <<18 ,PERF_OUTPUT_BPF_OUTPUT =1U <<19 ,PERF_OUTPUT_CALLINDENT =1U <<20 , +PERF_OUTPUT_INSN =1U <<21 , + PERF_OUTPUT_INSNLEN = 1U << 22, }; struct output_option {
@@ -93,6 +95,8 @@ struct output_option { {.str = "weight", .field = PERF_OUTPUT_WEIGHT}, {.str = "bpf-output", .field = PERF_OUTPUT_BPF_OUTPUT}, {.str = "callindent", .field = PERF_OUTPUT_CALLINDENT}, + {.str = "insn", .field = PERF_OUTPUT_INSN}, + {.str = "insnlen", .field = PERF_OUTPUT_INSNLEN}, };/* default set to maintain compatibility with current format */
@@ -624,6 +628,21 @@ static void print_sample_callindent(struct perf_sample *sample,printf ("%*s" , spacing - len,"" ); } + +static void print_insn (struct perf_sample *sample, +struct perf_event_attr *attr) +{ +if (PRINT_FIELD (INSNLEN )) +printf (" ilen: %d" , sample->insn_len); +if (PRINT_FIELD (INSN )) { +int i; + +printf (" insn:" ); +for (i =0 ; i < sample->insn_len; i++) +printf (" %02x" , (unsigned char )sample->insn[i]); + } +} + static void print_sample_bts(struct perf_sample *sample, struct perf_evsel *evsel, struct thread *thread,
@@ -668,6 +687,8 @@ static void print_sample_bts(struct perf_sample *sample,if (print_srcline_last)map__fprintf_srcline (al->map, al->addr,"\n " , stdout); +print_insn (sample, attr); +printf ("\n" ); }
@@ -911,7 +932,7 @@ static void process_event(struct perf_script *script,if (perf_evsel__is_bpf_output (evsel) &&PRINT_FIELD (BPF_OUTPUT ))print_sample_bpf_output (sample); - +print_insn (sample, attr);printf ("\n" ); }
@@ -2124,7 +2145,7 @@ int cmd_script(int argc, const char **argv, const char *prefix __maybe_unused) "Valid types: hw,sw,trace,raw. " "Fields: comm,tid,pid,time,cpu,event,trace,ip,sym,dso," "addr,symoff,period,iregs,brstack,brstacksym,flags," - "bpf-output,callindent", parse_output_fields), + "bpf-output,callindent,insn,insnlen", parse_output_fields),OPT_BOOLEAN ('a' ,"all-cpus" , &system_wide,"system-wide collection from all CPUs" ), OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
Signature · 1 line
2.5.5
Adrian Hunter
reviewerQuoted 7 lines · Andi Kleen
From: Andi Kleen <[email protected]>
Change the Intel PT decoder to pass up the length and the instruction bytes of the decoded or sampled instruction in the perf sample.
The decoder already knows this information, we just need to pass it up. Since it is only a couple of movs it is not very expensive.
The decoder doesn't always fill the instruction buffer because it caches the results from walking basic blocks. That means the bytes from the last instruction in the basic block would need to be added to the cache.
I wonder if we shouldn't look at something more sophisticated e.g. a python script that can do full disassembly.
Quoted 6 lines · Adrian Hunter
Quoted 5 lines · Andi Kleen
From: Andi Kleen <[email protected]>
Change the Intel PT decoder to pass up the length and the instruction bytes of the decoded or sampled instruction in the perf sample.
Quoted 6 lines
Quoted 2 lines
The decoder already knows this information, we just need to pass it up. Since it is only a couple of movs it is not very expensive.
The decoder doesn't always fill the instruction buffer because it caches the results from walking basic blocks. That means the bytes from the last instruction in the basic block would need to be added to the cache.
I fixed this and posted a new version.
Quoted 3 lines
I wonder if we shouldn't look at something more sophisticated e.g. a python script that can do full disassembly.
I had an earlier version that used a disassembler library, but that wasn't appreciated, so now switched to this simpler method. Assembler output can be done with a sed + objdump hack now.
-Andi