diff options
author | Damjan Marion <damarion@cisco.com> | 2016-12-19 23:05:39 +0100 |
---|---|---|
committer | Damjan Marion <damarion@cisco.com> | 2016-12-28 12:25:14 +0100 |
commit | 7cd468a3d7dee7d6c92f69a0bb7061ae208ec727 (patch) | |
tree | 5de62f8dbd3a752f5a676ca600e43d2652d1ff1a /src/tools/perftool/cpelinreg.c | |
parent | 696f1adec0df3b8f161862566dd9c86174302658 (diff) |
Reorganize source tree to use single autotools instance
Change-Id: I7b51f88292e057c6443b12224486f2d0c9f8ae23
Signed-off-by: Damjan Marion <damarion@cisco.com>
Diffstat (limited to 'src/tools/perftool/cpelinreg.c')
-rw-r--r-- | src/tools/perftool/cpelinreg.c | 892 |
1 files changed, 892 insertions, 0 deletions
diff --git a/src/tools/perftool/cpelinreg.c b/src/tools/perftool/cpelinreg.c new file mode 100644 index 00000000000..115afad7fb2 --- /dev/null +++ b/src/tools/perftool/cpelinreg.c @@ -0,0 +1,892 @@ +/* + *------------------------------------------------------------------ + * Copyright (c) 2008-2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/* + * Search for O(N**2) functions bracketed by before/after + * events. The "before" event's datum is used as a tag, e.g. which function + * did we call that's strongly O(N). + */ + +#include <stdio.h> +#include <stdlib.h> +#include <netinet/in.h> +#include <string.h> +#include <sys/types.h> +#include <sys/stat.h> +#include <fcntl.h> +#include <sys/fcntl.h> +#include <sys/mman.h> +#include <unistd.h> +#include <ctype.h> +#include <vppinfra/clib.h> +#include <vppinfra/vec.h> +#include <vppinfra/hash.h> +#include <pwd.h> +#include <stdarg.h> +#include <time.h> +#include "cpel.h" + +FILE *g_ifp; +char *g_ifile; + +typedef unsigned long long ulonglong; + +void process_traces (void); +void record_instance (ulong tag, ulonglong time); +void report_actors (void); +void scatterplot_data(void); +int entry_event, exit_event; +int nokey; +char *version = "cpelinreg 2.0"; +int model_these[10]; +int model_index; +int summary_stats; +ulonglong first_start_time; +ulonglong last_end_time; +ulonglong total_time; +ulong scatterkey; +int inline_mokus; + +typedef struct bound_track_ { + u32 track_code; + u32 *start_datum; + u8 *dup_event; + int state; + u64 *start_time; + u64 thread_timestamp; + u64 time_thread_on_cpu; +} bound_track_t; + +bound_track_t *bound_tracks; +uword *the_trackdef_hash; + + +#define MAXSTACK 128 + +typedef struct instance_ { + struct instance_ *next; + ulonglong time; +}instance_t; + +typedef struct actor_ { + struct actor_ *next; + ulong key; + struct instance_ *first; + struct instance_ *last; + double a; + double b; + double min; + double max; + double mean; + double r; + ulong ninst; +} actor_t; + +#define NBUCKETS 1811 + +actor_t *hash[NBUCKETS]; + +actor_t *find_or_create_actor (ulong key) +{ + ulong bucket; + actor_t *ap; + u8 *mem; + + bucket = key % NBUCKETS; + + ap = hash[bucket]; + + if (ap == NULL) { + /* Ensure 8-byte alignment to avoid (double) alignment faults */ + mem = malloc(sizeof(*ap) + 4); + if (((uword)(mem)) & 0x7) + mem += 4; + ap = (actor_t *)mem; + + if (ap == NULL) { + fprintf (stderr, "out of memory...\n"); + exit (1); + } + ap->next = 0; + ap->key = key; + ap->first = 0; + ap->last = 0; + ap->a = 0.00; + ap->b = 0.00; + hash [bucket] = ap; + return (ap); + } + + while (ap) { + if (ap->key == key) + return (ap); + ap = ap->next; + } + + mem = malloc(sizeof(*ap)+4); + if (((uword)(mem) & 0x7)) + mem += 4; + ap = (actor_t *)mem; + + if (ap == NULL) { + fprintf (stderr, "out of memory...\n"); + exit (1); + } + ap->key = key; + ap->first = 0; + ap->last = 0; + ap->a = 0.00; + ap->b = 0.00; + + ap->next = hash[bucket]; + hash[bucket] = ap; + + return (ap); +} + +void record_instance (ulong key, ulonglong time) +{ + actor_t *ap; + instance_t *ip; + + if (nokey) + key = 0; + + ap = find_or_create_actor (key); + + ip = (instance_t *)malloc(sizeof(*ip)); + if (ip == NULL) { + fprintf (stderr, "out of memory...\n"); + exit (1); + } + ip->time = time; + ip->next = 0; + + if (ap->first == 0) { + ap->first = ip; + ap->last = ip; + ap->ninst = 1; + } else { + ap->last->next = ip; + ap->last = ip; + ap->ninst++; + } +} + +#define NINSTANCE 200000 + +double x[NINSTANCE]; +double y[NINSTANCE]; + +int actor_compare (const void *arg1, const void *arg2) +{ + double e10k1, e10k2; + actor_t **a1 = (actor_t **)arg1; + actor_t **a2 = (actor_t **)arg2; + double ninst1, ninst2; + + ninst1 = ((double)((*a1)->ninst)); + ninst2 = ((double)((*a2)->ninst)); + + e10k1 = ninst1 * ((*a1)->mean); + e10k2 = ninst2 * ((*a2)->mean); + + if (e10k1 < e10k2) + return (1); + else if (e10k1 == e10k2) + return (0); + else + return (-1); +} + +void report_actors (void) +{ + int i; + actor_t *ap; + instance_t *ip; + int nactors = 0; + int ninstance; + actor_t **actor_vector; + double e10k; + extern void linreg (double *x, double *y, int nitems, double *a, double *b, + double *minp, double *maxp, double *meanp, double *r); + + for (i = 0; i < NBUCKETS; i++) { + ap = hash[i]; + if (ap == NULL) + continue; + while (ap) { + nactors++; + ninstance = 0; + + ip = ap->first; + + while (ip) { + if (ninstance < NINSTANCE) { + x[ninstance] = ninstance; + y[ninstance] = ((double)ip->time); + ninstance++; + } + ip = ip->next; + } + if (ninstance > 1) { +#if DEBUG > 0 + int j; + + for (j = 0; j < ninstance; j++) { + printf("x[%d] = %10.2f, y[%d] = %10.2f\n", + j, x[j], j, y[j]); + } +#endif + + linreg (x, y, ninstance, &ap->a, &ap->b, &ap->min, + &ap->max, &ap->mean, &ap->r); + } else { + ap->a = 0.00; + ap->b = 0.00; + } + + ap = ap->next; + } + } + + actor_vector = (actor_t **)malloc (nactors*sizeof(*actor_vector)); + nactors = 0; + + for (i = 0; i < NBUCKETS; i++) { + ap = hash[i]; + if (ap == NULL) + continue; + while (ap) { + if ((ap->a != 0.00) || (ap->b != 0.00)) { + actor_vector[nactors++] = ap; + } + ap = ap->next; + } + } + + qsort (actor_vector, nactors, sizeof (actor_t *), actor_compare); + + if (summary_stats) + printf("NInst Offset Slope T(Ninst) Min Max Avg %%InstTime R Key"); + else + printf("NInst Offset Slope T(Ninst) Key"); + + for (i = 0; i < model_index; i++) { + printf ("T @ %-8d ", model_these[i]); + } + + printf ("\n"); + + for (i = 0; i < nactors; i++) { + int j; + double ninst; + double pcttot; + ap = actor_vector[i]; + ninst = ap->ninst; + + e10k = ninst * (ap->a + ap->b*((ninst-1.0)/2.0)); + + if (ap->ninst) { + if (summary_stats) { + pcttot = (e10k / ((double)total_time)) * 100.0; + printf ("%6ld %11.2f %11.2f %11.2f %11.2f %11.2f %11.2f %11.2f %11.2f 0x%08lx ", + ap->ninst, ap->a, ap->b, e10k, ap->min, + ap->max, ap->mean, pcttot, ap->r, ap->key); + } + else + printf ("%6ld %11.2f %11.2f %11.2f 0x%08lx ", + ap->ninst, ap->a, ap->b, e10k, ap->key); + + for (j = 0; j < model_index; j++) { + ninst = model_these[j]; + e10k = ninst * (ap->a + ap->b*((ninst-1.0)/2.0)); + printf ("%10.2f ", e10k); + } + printf ("\n"); + } + } +} + +void scatterplot_data(void) +{ + actor_t *ap; + int i; + instance_t *ip; + double time; + int count=0; + + for (i = 0; i < NBUCKETS; i++) { + ap = hash[i]; + if (ap == NULL) + continue; + while (ap) { + if (ap->key == scatterkey){ + ip = ap->first; + while (ip) { + time = ((double)ip->time); + printf ("%d\t%.0f\n", count++, time); + ip = ip->next; + } + return; + } + ap = ap->next; + } + } +} + + +void fatal(char *s) +{ + fprintf(stderr, "%s", s); + fprintf(stderr, "\n"); + exit(1); +} + +typedef enum { + PASS1=1, +} pass_t; + +typedef struct { + int (*pass1)(cpel_section_header_t *, int, FILE *); +} section_processor_t; + +int bad_section(cpel_section_header_t *sh, int verbose, FILE *ofp) +{ + fprintf(ofp, "Bad (type 0) section, skipped...\n"); + return(0); +} + +int noop_pass(cpel_section_header_t *sh, int verbose, FILE *ofp) +{ + return(0); +} + +int unsupported_pass (cpel_section_header_t *sh, int verbose, FILE *ofp) +{ + if (verbose) { + fprintf(ofp, "Unsupported type %d section\n", + ntohl(sh->section_type)); + } + return(0); +} + +int trackdef_pass(cpel_section_header_t *sh, int verbose, FILE *ofp) +{ + int i, nevents; + track_definition_section_header_t *tdh; + track_definition_t *tp; + u32 track_code; + uword *p; + bound_track_t *btp; + + tdh = (track_definition_section_header_t *)(sh+1); + nevents = ntohl(tdh->number_of_track_definitions); + + if (verbose) { + fprintf(stderr, "Track Definition Section: %d definitions\n", + nevents); + } + + tp = (track_definition_t *)(tdh+1); + + for (i = 0; i < nevents; i++) { + track_code = ntohl(tp->track); + p = hash_get(the_trackdef_hash, track_code); + if (p) { + fprintf(ofp, "track %d redefined, retain first definition\n", + track_code); + continue; + } + vec_add2(bound_tracks, btp, 1); + btp->track_code = track_code; + hash_set(the_trackdef_hash, track_code, btp - bound_tracks); + tp++; + } + return (0); +} + + +int event_pass (cpel_section_header_t *sh, int verbose, FILE *ofp) +{ + event_section_header_t *eh; + event_entry_t *ep; + f64 ticks_per_us; + long output_count; + long dup_events = 0; + ulonglong end_time = 0; + double t; + int sp, ancestor; + int nevents, i; + u64 now; + u64 time0, time1; + double d; + u32 last_track_code = 0xdeafb00b; + u32 track_code; + u32 event_code, event_datum; + bound_track_t *tp = 0; + uword *p; + + output_count = 0; + total_time = 0; + + eh = (event_section_header_t *)(sh+1); + nevents = ntohl(eh->number_of_events); + ticks_per_us = ((double)ntohl(eh->clock_ticks_per_second))/1e6; + + if (verbose) { + fprintf(ofp, "%.3f ticks_per_us\n", ticks_per_us); + } + + ep = (event_entry_t *)(eh+1); + + time0 = ntohl (ep->time[0]); + time1 = ntohl (ep->time[1]); + + now = (((u64) time0)<<32) | time1; + d = now; + d /= ticks_per_us; + first_start_time = d; + + for (i = 0; i < nevents; i++) { + time0 = ntohl (ep->time[0]); + time1 = ntohl (ep->time[1]); + + now = (((u64) time0)<<32) | time1; + + /* Convert from bus ticks to usec */ + d = now; + d /= ticks_per_us; + + now = d; + + track_code = ntohl(ep->track); + event_code = ntohl(ep->event_code); + event_datum = ntohl(ep->event_datum); + + if (track_code != last_track_code) { + if (tp) { + tp->thread_timestamp += now - tp->time_thread_on_cpu; + tp->time_thread_on_cpu = 0; + } + p = hash_get(the_trackdef_hash, track_code); + if (!p) { + /* synthesize a new track */ + vec_add2(bound_tracks, tp, 1); + tp->track_code = track_code; + hash_set(the_trackdef_hash, track_code, tp - bound_tracks); + } else { + tp = bound_tracks + p[0]; + } + last_track_code = track_code; + tp->time_thread_on_cpu = now; + } + + if (event_code != entry_event && + event_code != exit_event) { + ep++; + continue; + } + + again: + switch (tp->state) { + case 0: /* not in state */ + /* Another exit event? Stack pop */ + if (event_code == exit_event) { + /* Only if we have something on the stack */ + if (vec_len(tp->start_datum) > 0) { + tp->state = 1; + goto again; + } else { + fprintf (stderr, + "End event before start event, key 0x%x.", + ntohl(ep->event_datum)); + fprintf (stderr, " Interpret results carefully...\n"); + } + } + + tp->state = 1; + if (vec_len(tp->start_datum) >= MAXSTACK) { + int j; + + fprintf (stderr, "stack overflow..\n"); + for (j = vec_len(tp->start_datum)-1; j >= 0; j--) { + fprintf(stderr, "stack[%d]: datum 0x%x\n", + j, tp->start_datum[j]); + } + fprintf (stderr, + "Stack overflow... This occurs when " + "(start, datum)...(end, datum) events\n" + "are not properly paired.\n\n" + "A typical scenario looks like this:\n\n" + " ...\n" + " ELOG(..., START_EVENT, datum);\n" + " if (condition)\n" + " return; /*oops, forgot the end event*/\n" + " ELOG(..., END_EVENT, datum);\n" + " ...\n\n" + "The datum stack dump (above) should make it clear\n" + "where to start looking for a sneak path...\n"); + + exit (1); + } + vec_add1(tp->start_datum, event_datum); + vec_add1(tp->start_time, (tp->thread_timestamp + (now - tp->time_thread_on_cpu))); +#ifdef HAVING_TROUBLE + printf ("sp %lld key 0x%x start time %llu\n", + (long long) vec_len(tp->start_time)-1, event_datum, + (unsigned long long) + tp->start_time [vec_len(tp->start_time)-1]); + printf ("timestamp %llu, now %llu, thread on cpu %llu\n", + (unsigned long long) tp->thread_timestamp, + (unsigned long long) now, + (unsigned long long) tp->time_thread_on_cpu); +#endif + + + + /* + * Multiple identical enter events? If the user knows that + * gcc is producing bogus events due to inline functions, + * trash the duplicate. + */ + if (inline_mokus + && vec_len (tp->start_datum) > 1 + && tp->start_datum [vec_len(tp->start_datum)-1] == + tp->start_datum [vec_len(tp->start_datum)-2]) { + vec_add1 (tp->dup_event, 1); + } else { + vec_add1 (tp->dup_event, 0); + } + + + ep++; + continue; + + case 1: /* in state */ + /* Another entry event? Stack push*/ + if (event_code == entry_event) { + tp->state = 0; + goto again; + } + + if (vec_len(tp->start_datum) == 0) { + fprintf (stderr, "Stack underflow...\n"); + exit (1); + } + + sp = vec_len(tp->start_time)-1; + + end_time = tp->thread_timestamp + (now - tp->time_thread_on_cpu); + + if (!tp->dup_event[sp]) { +#ifdef HAVING_TROUBLE + printf ("sp %d key 0x%x charged %llu\n", sp, + tp->start_datum[sp], end_time - tp->start_time[sp]); + printf (" start %llu, end %llu\n", (unsigned long long) tp->start_time[sp], + (unsigned long long) end_time); +#endif + + record_instance (tp->start_datum[sp], (end_time - + tp->start_time[sp])); + + /* Factor out our time from surrounding services, if any */ + for (ancestor = sp-1; ancestor >= 0; ancestor--) { +#ifdef HAVING_TROUBLE + printf ("Factor out %lld from key 0x%08x\n", + (end_time - tp->start_time[sp]), tp->start_datum[ancestor]); +#endif + tp->start_time[ancestor] += (end_time - tp->start_time[sp]); + } + output_count++; + total_time += (end_time - tp->start_time[sp]); + tp->state = 0; + } else { + dup_events++; + } + _vec_len(tp->start_datum) = sp; + _vec_len(tp->start_time) = sp; + _vec_len(tp->dup_event) = sp; + } + + ep++; + } + last_end_time = now; + + if (scatterkey) { + scatterplot_data(); + exit (0); + } + + if (output_count) { + t = (double)total_time; + printf ("%ld instances of state, %.2f microseconds average\n", + output_count, t / output_count); + + printf ("Total instrumented runtime: %.2f microseconds\n", + ((double)total_time)); + printf ("Total runtime: %lld microseconds\n", + last_end_time - first_start_time); + + t /= (double)(last_end_time - first_start_time); + t *= 100.0; + + if (dup_events) { + printf ("Suppressed %ld duplicate state entry events\n", + dup_events); + } + printf ("Instrumented code accounts for %.2f%% of total time.\n\n", + t); + report_actors(); + } else { + printf ("No instances of state...\n"); + } + + return(0); +} + +/* + * Note: If necessary, add passes / columns to this table to + * handle section order dependencies. + */ + +section_processor_t processors[CPEL_NUM_SECTION_TYPES+1] = +{ + {unsupported_pass}, /* type 0 -- f**ked */ + {noop_pass}, /* type 1 -- STRTAB */ + {noop_pass}, /* type 2 -- SYMTAB */ + {noop_pass}, /* type 3 -- EVTDEF */ + {trackdef_pass}, /* type 4 -- TRACKDEF */ + {event_pass}, /* type 5 -- EVENTS */ +}; + +int process_section(cpel_section_header_t *sh, int verbose, FILE *ofp, + pass_t pass) +{ + u32 type; + type = ntohl(sh->section_type); + int rv; + int (*fp)(cpel_section_header_t *, int, FILE *); + + if (type > CPEL_NUM_SECTION_TYPES) { + fprintf(stderr, "Unknown section type %d\n", type); + return(1); + } + switch(pass) { + case PASS1: + fp = processors[type].pass1; + break; + + default: + fprintf(stderr, "Unknown pass %d\n", pass); + return(1); + } + + rv = (*fp)(sh, verbose, ofp); + + return(rv); +} + +char *mapfile (char *file) +{ + struct stat statb; + char *rv; + int maphfile; + size_t mapfsize; + + maphfile = open (file, O_RDONLY); + + if (maphfile < 0) + { + fprintf (stderr, "Couldn't read %s, skipping it...\n", file); + return (NULL); + } + + if (fstat (maphfile, &statb) < 0) + { + fprintf (stderr, "Couldn't get size of %s, skipping it...\n", file); + return (NULL); + } + + /* Don't try to mmap directories, FIFOs, semaphores, etc. */ + if (! (statb.st_mode & S_IFREG)) { + fprintf (stderr, "%s is not a regular file, skipping it...\n", file); + return (NULL); + } + + mapfsize = statb.st_size; + + if (mapfsize < 3) + { + fprintf (stderr, "%s zero-length, skipping it...\n", file); + close (maphfile); + return (NULL); + } + + rv = mmap (0, mapfsize, PROT_READ, MAP_SHARED, maphfile, 0); + + if (rv == 0) + { + fprintf (stderr, "%s problem mapping, I quit...\n", file); + exit (-1); + } + close (maphfile); + return (rv); +} + +int process_file (u8 *cpel, int verbose) +{ + cpel_file_header_t *fh; + cpel_section_header_t *sh; + u16 nsections; + u32 section_size; + int i; + FILE *ofp = stderr; + + /* First, the file header */ + fh = (cpel_file_header_t *)cpel; + if (fh->endian_version != CPEL_FILE_VERSION) { + if (fh->endian_version & CPEL_FILE_LITTLE_ENDIAN) { + fprintf(stderr, "Little endian data format not supported\n"); + return(1); + } + fprintf(stderr, "Unsupported file version 0x%x\n", + fh->endian_version); + return(1); + } + nsections = ntohs(fh->nsections); + + /* + * Take a passe through the file. + */ + sh = (cpel_section_header_t *)(fh+1); + for (i = 0; i < nsections; i++) { + section_size = ntohl(sh->data_length); + + if(verbose) { + fprintf(ofp, "Section type %d, size %d\n", + ntohl(sh->section_type), + section_size); + } + + if(process_section(sh, verbose, ofp, PASS1)) + return(1); + + sh++; + sh = (cpel_section_header_t *)(((u8 *)sh)+section_size); + } + + return(0); +} + +/**************************************************************************** +* main - +****************************************************************************/ + +int main (int argc, char **argv) +{ + int curarg = 1; + u8 *cpel = 0; + int verbose = 0; + + if (argc < 6) + { + fprintf (stderr, "usage: cpelinreg -i <file>\n"); + fprintf (stderr, " -s start-event --e end-event [-nokey]\n"); + fprintf (stderr, " [-m <ninst-to-model>][-xtra-stats]\n"); + fprintf (stderr, " [-keyscatterplot <hex-key>]\n\n"); + fprintf (stderr, "%s\n", version); + exit (1); + } + + while (curarg < argc) { + if (!strncmp (argv[curarg], "-ifile", 2)) { + curarg++; + g_ifile = argv[curarg++]; + continue; + } + if (!strncmp (argv[curarg], "-start", 2)) { + curarg++; + entry_event = atol (argv [curarg++]); + continue; + } + if (!strncmp (argv[curarg], "-end", 2)) { + curarg++; + exit_event = atol (argv [curarg++]); + continue; + } + + if (!strncmp(argv[curarg], "-badinlines", 2)) { + curarg++; + inline_mokus = 1; + continue; + } + + if (!strncmp (argv[curarg], "-x", 2)) { + curarg++; + summary_stats=1; + continue; + } + if (!strncmp (argv[curarg], "-nokey", 2)) { + curarg++; + nokey = 1; + continue; + } + if (!strncmp (argv[curarg], "-keyscatterplot", 2)) { + curarg++; + sscanf (argv[curarg], "%lx", &scatterkey); + curarg++; + continue; + } + + if (!strncmp (argv[curarg], "-model", 2)) { + if (model_index >= sizeof(model_these) / sizeof(int)) { + fprintf (stderr, "Too many model requests\n"); + exit (1); + } + curarg++; + model_these[model_index++] = atol (argv [curarg++]); + continue; + } + if (!strncmp (argv[curarg], "-verbose", 2)) { + verbose++; + curarg++; + continue; + } + + fprintf (stderr, "unknown switch '%s'\n", argv[curarg]); + exit (1); + } + + cpel = (u8 *)mapfile(g_ifile); + + if (cpel == NULL) + { + fprintf (stderr, "Couldn't open %s\n", g_ifile); + exit (1); + } + + printf ("Extracting state info from %s\nentry_event %d, exit_event %d\n", + g_ifile, entry_event, exit_event); + if (nokey) { + printf ("All state instances mapped to a single actor chain\n"); + } + + the_trackdef_hash = hash_create (0, sizeof (uword)); + + process_file(cpel, verbose); + exit (0); +} |