diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2021-11-08 20:25:26 +0300 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2021-11-08 20:25:26 +0300 |
commit | bbdbeb0048b443082bcce5ed65a336bcc578a60e (patch) | |
tree | c5a4a8b719e07c8f747dba8b3c9cb22f1124671b /tools/perf/util/expr.y | |
parent | 1e9ed9360f80d13e41684ca458f01fdf922c7c57 (diff) | |
parent | 6b491a86b77c0dc323ca49f3a29a0f67178b75f8 (diff) | |
download | linux-bbdbeb0048b443082bcce5ed65a336bcc578a60e.tar.xz |
Merge tag 'perf-tools-for-v5.16-2021-11-07-without-bpftool-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux
Pull perf tools updates from Arnaldo Carvalho de Melo:
"perf annotate:
- Add riscv64 support.
- Add fusion logic for AMD microarchs.
perf record:
- Add an option to control the synthesizing behavior:
--synth <no|all|task|mmap|cgroup>
core:
- Allow controlling synthesizing PERF_RECORD_ metadata events during
record.
- perf.data reader prep work for multithreaded processing.
- Fix missing exclude_{host,guest} setting in PMUs that don't support
it and that were causing the feature detection code to disable it
for all events, even the ones in PMUs that support it.
- Fix the default use of precise events on AMD, that were always
falling back to non-precise because perf_event_attr.exclude_guest=1
was set and IBS does not have filtering capability, refusing
precise + exclude_guest.
- Add bitfield_swap() to handle branch_stack endian issue.
perf script:
- Show binary offsets for userspace addresses in callchains.
- Support instruction latency via new "ins_lat" selectable field.
- Add dlfilter-show-cycles
perf inject:
- Add vmlinux and ignore-vmlinux arguments, similar to other tools.
perf list:
- Display PMU prefix for partially supported hybrid cache events.
- Display hybrid PMU events with cpu type.
perf stat:
- Improve metrics documentation of data structures.
- Fix memory leaks in the metric code.
- Use NAN for missing event IDs.
- Don't compute unused events.
- Fix memory leak on error path.
- Encode and use metric-id as a metric qualifier.
- Allow metrics with no events.
- Avoid events for an 'if' constant result.
- Only add a referenced metric once.
- Simplify metric_refs calculation.
- Allow modifiers on metrics.
perf test:
- Add workload test of metric and metric groups.
- Workload test of all PMUs.
- vmlinux-kallsyms: Ignore hidden symbols.
- Add pmu-event test for event described as "config=".
- Verify more event members in pmu-events test.
- Add endian test for struct branch_flags on the sample-parsing test.
- Improve temp file cleanup in several tests.
perf daemon:
- Address MSAN warnings on send_cmd().
perf kmem:
- Improve man page for record options
perf srcline:
- Use long-running addr2line per DSO, greatly speeding up the
'srcline' sort order.
perf symbols:
- Ignore $a/$d symbols for ARM modules.
- Fix /proc/kcore access on 32 bit systems.
Kernel UAPI copies:
- Update copy of linux/socket.h with the kernel sources, no change in
tooling output.
libbpf:
- Pull in bpf_program__get_prog_info_linear() from libbpf, too much
specific to perf.
- Deprecate bpf_map__resize() in favor of bpf_map_set_max_entries()
- Install libbpf headers locally when building.
- Bump minimum LLVM C++ std to GNU++14.
libperf:
- Use binary search in perf_cpu_map__idx() as array are sorted.
libtracefs:
- Enable libtracefs dynamic linking.
libtraceevent:
- Increase logging when verbose.
Arch specific:
* PowerPC:
- Add support to expose instruction and data address registers as
part of extended regs.
Vendor events:
* JSON parser:
- Support ConfigCode to set the config= in PMUs
- Make the JSON parser more conformant when in strict mode.
* All JSON files:
- Fix all remaining invalid JSON files.
* ARM:
- Syntax corrections in Neoverse N1 json.
- Categorise the Neoverse V1 counters.
- Add new armv8 PMU events.
- Revise hip08 uncore events.
Hardware tracing:
* auxtrace:
- Add missing Z option to ITRACE_HELP.
- Add itrace A option to approximate IPC.
- Add itrace d+o option to direct debug log to stdout.
* Intel PT:
- Add support for PERF_RECORD_AUX_OUTPUT_HW_ID
- Support itrace A option to approximate IPC
- Support itrace d+o option to direct debug log to stdout"
* tag 'perf-tools-for-v5.16-2021-11-07-without-bpftool-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux: (120 commits)
perf build: Install libbpf headers locally when building
perf MANIFEST: Add bpftool files to allow building with BUILD_BPF_SKEL=1
perf metric: Fix memory leaks
perf parse-event: Add init and exit to parse_event_error
perf parse-events: Rename parse_events_error functions
perf stat: Fix memory leak on error path
perf tools: Use __BYTE_ORDER__
perf inject: Add vmlinux and ignore-vmlinux arguments
perf tools: Check vmlinux/kallsyms arguments in all tools
perf tools: Refactor out kernel symbol argument sanity checking
perf symbols: Ignore $a/$d symbols for ARM modules
perf evsel: Don't set exclude_guest by default
perf evsel: Fix missing exclude_{host,guest} setting
perf bpf: Add missing free to bpf_event__print_bpf_prog_info()
perf beauty: Update copy of linux/socket.h with the kernel sources
perf clang: Fixes for more recent LLVM/clang
tools: Bump minimum LLVM C++ std to GNU++14
perf bpf: Pull in bpf_program__get_prog_info_linear()
Revert "perf bench futex: Add support for 32-bit systems with 64-bit time_t"
perf test sample-parsing: Add endian test for struct branch_flags
...
Diffstat (limited to 'tools/perf/util/expr.y')
-rw-r--r-- | tools/perf/util/expr.y | 325 |
1 files changed, 244 insertions, 81 deletions
diff --git a/tools/perf/util/expr.y b/tools/perf/util/expr.y index b2ada8f8309a..f969dfa525bd 100644 --- a/tools/perf/util/expr.y +++ b/tools/perf/util/expr.y @@ -1,42 +1,43 @@ /* Simple expression parser */ %{ #define YYDEBUG 1 -#include <stdio.h> -#include "util.h" +#include <assert.h> +#include <math.h> #include "util/debug.h" -#include <stdlib.h> // strtod() +#include "smt.h" #define IN_EXPR_Y 1 #include "expr.h" -#include "smt.h" -#include <string.h> - -static double d_ratio(double val0, double val1) -{ - if (val1 == 0) { - return 0; - } - return val0 / val1; -} - %} %define api.pure full %parse-param { double *final_val } %parse-param { struct expr_parse_ctx *ctx } +%parse-param { bool compute_ids } %parse-param {void *scanner} %lex-param {void* scanner} %union { double num; char *str; + struct ids { + /* + * When creating ids, holds the working set of event ids. NULL + * implies the set is empty. + */ + struct hashmap *ids; + /* + * The metric value. When not creating ids this is the value + * read from a counter, a constant or some computed value. When + * creating ids the value is either a constant or BOTTOM. NAN is + * used as the special BOTTOM value, representing a "set of all + * values" case. + */ + double val; + } ids; } -%token EXPR_PARSE EXPR_OTHER EXPR_ERROR -%token <num> NUMBER -%token <str> ID -%destructor { free ($$); } <str> -%token MIN MAX IF ELSE SMT_ON D_RATIO +%token ID NUMBER MIN MAX IF ELSE SMT_ON D_RATIO EXPR_ERROR %left MIN MAX IF %left '|' %left '^' @@ -45,83 +46,245 @@ static double d_ratio(double val0, double val1) %left '-' '+' %left '*' '/' '%' %left NEG NOT -%type <num> expr if_expr +%type <num> NUMBER +%type <str> ID +%destructor { free ($$); } <str> +%type <ids> expr if_expr +%destructor { ids__free($$.ids); } <ids> %{ static void expr_error(double *final_val __maybe_unused, struct expr_parse_ctx *ctx __maybe_unused, + bool compute_ids __maybe_unused, void *scanner, const char *s) { pr_debug("%s\n", s); } +/* + * During compute ids, the special "bottom" value uses NAN to represent the set + * of all values. NAN is selected as it isn't a useful constant value. + */ +#define BOTTOM NAN + +/* During computing ids, does val represent a constant (non-BOTTOM) value? */ +static bool is_const(double val) +{ + return isfinite(val); +} + +static struct ids union_expr(struct ids ids1, struct ids ids2) +{ + struct ids result = { + .val = BOTTOM, + .ids = ids__union(ids1.ids, ids2.ids), + }; + return result; +} + +/* + * If we're not computing ids or $1 and $3 are constants, compute the new + * constant value using OP. Its invariant that there are no ids. If computing + * ids for non-constants union the set of IDs that must be computed. + */ +#define BINARY_LONG_OP(RESULT, OP, LHS, RHS) \ + if (!compute_ids || (is_const(LHS.val) && is_const(RHS.val))) { \ + assert(LHS.ids == NULL); \ + assert(RHS.ids == NULL); \ + RESULT.val = (long)LHS.val OP (long)RHS.val; \ + RESULT.ids = NULL; \ + } else { \ + RESULT = union_expr(LHS, RHS); \ + } + +#define BINARY_OP(RESULT, OP, LHS, RHS) \ + if (!compute_ids || (is_const(LHS.val) && is_const(RHS.val))) { \ + assert(LHS.ids == NULL); \ + assert(RHS.ids == NULL); \ + RESULT.val = LHS.val OP RHS.val; \ + RESULT.ids = NULL; \ + } else { \ + RESULT = union_expr(LHS, RHS); \ + } + %} %% -start: -EXPR_PARSE all_expr -| -EXPR_OTHER all_other +start: if_expr +{ + if (compute_ids) + ctx->ids = ids__union($1.ids, ctx->ids); -all_other: all_other other -| + if (final_val) + *final_val = $1.val; +} +; + +if_expr: expr IF expr ELSE expr +{ + if (fpclassify($3.val) == FP_ZERO) { + /* + * The IF expression evaluated to 0 so treat as false, take the + * ELSE and discard everything else. + */ + $$.val = $5.val; + $$.ids = $5.ids; + ids__free($1.ids); + ids__free($3.ids); + } else if (!compute_ids || is_const($3.val)) { + /* + * If ids aren't computed then treat the expression as true. If + * ids are being computed and the IF expr is a non-zero + * constant, then also evaluate the true case. + */ + $$.val = $1.val; + $$.ids = $1.ids; + ids__free($3.ids); + ids__free($5.ids); + } else if ($1.val == $5.val) { + /* + * LHS == RHS, so both are an identical constant. No need to + * evaluate any events. + */ + $$.val = $1.val; + $$.ids = NULL; + ids__free($1.ids); + ids__free($3.ids); + ids__free($5.ids); + } else { + /* + * Value is either the LHS or RHS and we need the IF expression + * to compute it. + */ + $$ = union_expr($1, union_expr($3, $5)); + } +} +| expr +; -other: ID +expr: NUMBER { - expr__add_id(ctx, $1); -} -| -MIN | MAX | IF | ELSE | SMT_ON | NUMBER | '|' | '^' | '&' | '-' | '+' | '*' | '/' | '%' | '(' | ')' | ',' -| -'<' | '>' | D_RATIO - -all_expr: if_expr { *final_val = $1; } - ; - -if_expr: - expr IF expr ELSE expr { $$ = $3 ? $1 : $5; } - | expr - ; - -expr: NUMBER - | ID { - struct expr_id_data *data; - - if (expr__resolve_id(ctx, $1, &data)) { - free($1); - YYABORT; - } - - $$ = expr_id_data__value(data); - free($1); - } - | expr '|' expr { $$ = (long)$1 | (long)$3; } - | expr '&' expr { $$ = (long)$1 & (long)$3; } - | expr '^' expr { $$ = (long)$1 ^ (long)$3; } - | expr '<' expr { $$ = $1 < $3; } - | expr '>' expr { $$ = $1 > $3; } - | expr '+' expr { $$ = $1 + $3; } - | expr '-' expr { $$ = $1 - $3; } - | expr '*' expr { $$ = $1 * $3; } - | expr '/' expr { if ($3 == 0) { - pr_debug("division by zero\n"); - YYABORT; - } - $$ = $1 / $3; - } - | expr '%' expr { if ((long)$3 == 0) { - pr_debug("division by zero\n"); - YYABORT; - } - $$ = (long)$1 % (long)$3; - } - | '-' expr %prec NEG { $$ = -$2; } - | '(' if_expr ')' { $$ = $2; } - | MIN '(' expr ',' expr ')' { $$ = $3 < $5 ? $3 : $5; } - | MAX '(' expr ',' expr ')' { $$ = $3 > $5 ? $3 : $5; } - | SMT_ON { $$ = smt_on() > 0; } - | D_RATIO '(' expr ',' expr ')' { $$ = d_ratio($3,$5); } - ; + $$.val = $1; + $$.ids = NULL; +} +| ID +{ + if (!compute_ids) { + /* + * Compute the event's value from ID. If the ID isn't known then + * it isn't used to compute the formula so set to NAN. + */ + struct expr_id_data *data; + + $$.val = NAN; + if (expr__resolve_id(ctx, $1, &data) == 0) + $$.val = expr_id_data__value(data); + + $$.ids = NULL; + free($1); + } else { + /* + * Set the value to BOTTOM to show that any value is possible + * when the event is computed. Create a set of just the ID. + */ + $$.val = BOTTOM; + $$.ids = ids__new(); + if (!$$.ids || ids__insert($$.ids, $1)) + YYABORT; + } +} +| expr '|' expr { BINARY_LONG_OP($$, |, $1, $3); } +| expr '&' expr { BINARY_LONG_OP($$, &, $1, $3); } +| expr '^' expr { BINARY_LONG_OP($$, ^, $1, $3); } +| expr '<' expr { BINARY_OP($$, <, $1, $3); } +| expr '>' expr { BINARY_OP($$, >, $1, $3); } +| expr '+' expr { BINARY_OP($$, +, $1, $3); } +| expr '-' expr { BINARY_OP($$, -, $1, $3); } +| expr '*' expr { BINARY_OP($$, *, $1, $3); } +| expr '/' expr +{ + if (fpclassify($3.val) == FP_ZERO) { + pr_debug("division by zero\n"); + YYABORT; + } else if (!compute_ids || (is_const($1.val) && is_const($3.val))) { + assert($1.ids == NULL); + assert($3.ids == NULL); + $$.val = $1.val / $3.val; + $$.ids = NULL; + } else { + /* LHS and/or RHS need computing from event IDs so union. */ + $$ = union_expr($1, $3); + } +} +| expr '%' expr +{ + if (fpclassify($3.val) == FP_ZERO) { + pr_debug("division by zero\n"); + YYABORT; + } else if (!compute_ids || (is_const($1.val) && is_const($3.val))) { + assert($1.ids == NULL); + assert($3.ids == NULL); + $$.val = (long)$1.val % (long)$3.val; + $$.ids = NULL; + } else { + /* LHS and/or RHS need computing from event IDs so union. */ + $$ = union_expr($1, $3); + } +} +| D_RATIO '(' expr ',' expr ')' +{ + if (fpclassify($5.val) == FP_ZERO) { + /* + * Division by constant zero always yields zero and no events + * are necessary. + */ + assert($5.ids == NULL); + $$.val = 0.0; + $$.ids = NULL; + ids__free($3.ids); + } else if (!compute_ids || (is_const($3.val) && is_const($5.val))) { + assert($3.ids == NULL); + assert($5.ids == NULL); + $$.val = $3.val / $5.val; + $$.ids = NULL; + } else { + /* LHS and/or RHS need computing from event IDs so union. */ + $$ = union_expr($3, $5); + } +} +| '-' expr %prec NEG +{ + $$.val = -$2.val; + $$.ids = $2.ids; +} +| '(' if_expr ')' +{ + $$ = $2; +} +| MIN '(' expr ',' expr ')' +{ + if (!compute_ids) { + $$.val = $3.val < $5.val ? $3.val : $5.val; + $$.ids = NULL; + } else { + $$ = union_expr($3, $5); + } +} +| MAX '(' expr ',' expr ')' +{ + if (!compute_ids) { + $$.val = $3.val > $5.val ? $3.val : $5.val; + $$.ids = NULL; + } else { + $$ = union_expr($3, $5); + } +} +| SMT_ON +{ + $$.val = smt_on() > 0 ? 1.0 : 0.0; + $$.ids = NULL; +} +; %% |