Merge branch 'add-P'
[blktrace.git] / btreplay / btreplay.c
1 /*
2  * Blktrace replay utility - Play traces back
3  *
4  * Copyright (C) 2007 Alan D. Brunelle <Alan.Brunelle@hp.com>
5  *
6  *  This program is free software; you can redistribute it and/or modify
7  *  it under the terms of the GNU General Public License as published by
8  *  the Free Software Foundation; either version 2 of the License, or
9  *  (at your option) any later version.
10  *
11  *  This program is distributed in the hope that it will be useful,
12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  *  GNU General Public License for more details.
15  *
16  *  You should have received a copy of the GNU General Public License
17  *  along with this program; if not, write to the Free Software
18  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
19  */
20
21 static char build_date[] = __DATE__ " at "__TIME__;
22
23 #include <assert.h>
24 #include <errno.h>
25 #include <fcntl.h>
26 #include <libaio.h>
27 #include <pthread.h>
28 #include <sched.h>
29 #include <signal.h>
30 #include <stdio.h>
31 #include <stdlib.h>
32 #include <string.h>
33 #include <time.h>
34 #include <unistd.h>
35 #include <sys/param.h>
36 #include <sys/stat.h>
37 #include <sys/time.h>
38 #include <sys/types.h>
39 #include <dirent.h>
40 #include <stdarg.h>
41
42 #if !defined(_GNU_SOURCE)
43 #       define _GNU_SOURCE
44 #endif
45 #include <getopt.h>
46
47 #include "list.h"
48 #include "btrecord.h"
49
50 /* 
51  * ========================================================================
52  * ==== STRUCTURE DEFINITIONS =============================================
53  * ========================================================================
54  */
55
56 /**
57  * Each device map has one of these:
58  * 
59  * @head:       Linked on to map_devs
60  * @from_dev:   Device name as seen on recorded system
61  * @to_dev:     Device name to be used on replay system
62  */
63 struct map_dev {
64         struct list_head head;
65         char *from_dev, *to_dev;
66 };
67
68 /**
69  * Each device name specified has one of these (until threads are created)
70  *
71  * @head:       Linked onto input_devs
72  * @devnm:      Device name -- 'sd*'
73  */
74 struct dev_info {
75         struct list_head head;
76         char *devnm;
77 };
78
79 /*
80  * Per input file information
81  *
82  * @head:       Used to link up on input_files
83  * @free_iocbs: List of free iocb's available for use
84  * @used_iocbs: List of iocb's currently outstanding
85  * @mutex:      Mutex used with condition variable to protect volatile values
86  * @cond:       Condition variable used when waiting on a volatile value change
87  * @naios_out:  Current number of AIOs outstanding on this context
88  * @naios_free: Number of AIOs on the free list (short cut for list_len)
89  * @send_wait:  Boolean: When true, the sub thread is waiting on free IOCBs
90  * @reap_wait:  Boolean: When true, the rec thread is waiting on used IOCBs
91  * @send_done:  Boolean: When true, the sub thread has completed work
92  * @reap_done:  Boolean: When true, the rec thread has completed work
93  * @sub_thread: Thread used to submit IOs.
94  * @rec_thread: Thread used to reclaim IOs.
95  * @ctx:        IO context
96  * @devnm:      Copy of the device name being managed by this thread
97  * @file_name:  Full name of the input file
98  * @cpu:        CPU this thread is pinned to
99  * @ifd:        Input file descriptor
100  * @ofd:        Output file descriptor
101  * @iterations: Remaining iterations to process
102  * @vfp:        For verbose dumping of actions performed
103  */
104 struct thr_info {
105         struct list_head head, free_iocbs, used_iocbs;
106         pthread_mutex_t mutex;
107         pthread_cond_t cond;
108         volatile long naios_out, naios_free;
109         volatile int send_wait, reap_wait, send_done, reap_done;
110         pthread_t sub_thread, rec_thread;
111         io_context_t ctx;
112         char *devnm, *file_name;
113         int cpu, ifd, ofd, iterations;
114         FILE *vfp;
115 };
116
117 /*
118  * Every Asynchronous IO used has one of these (naios per file/device).
119  *
120  * @iocb:       IOCB sent down via io_submit
121  * @head:       Linked onto file_list.free_iocbs or file_list.used_iocbs
122  * @tip:        Pointer to per-thread information this IO is associated with
123  * @nbytes:     Number of bytes in buffer associated with iocb
124  */
125 struct iocb_pkt {
126         struct iocb iocb;
127         struct list_head head;
128         struct thr_info *tip;
129         int nbytes;
130 };
131
132 /* 
133  * ========================================================================
134  * ==== GLOBAL VARIABLES ==================================================
135  * ========================================================================
136  */
137
138 static volatile int signal_done = 0;    // Boolean: Signal'ed, need to quit
139
140 static char *ibase = "replay";          // Input base name
141 static char *idir = ".";                // Input directory base
142 static int cpus_to_use = -1;            // Number of CPUs to use
143 static int def_iterations = 1;          // Default number of iterations
144 static int naios = 512;                 // Number of AIOs per thread
145 static int ncpus = 0;                   // Number of CPUs in the system
146 static int verbose = 0;                 // Boolean: Output some extra info
147 static int write_enabled = 0;           // Boolean: Enable writing
148 static __u64 genesis = ~0;              // Earliest time seen
149 static __u64 rgenesis;                  // Our start time
150 static size_t pgsize;                   // System Page size
151 static int nb_sec = 512;                // Number of bytes per sector
152 static LIST_HEAD(input_devs);           // List of devices to handle
153 static LIST_HEAD(input_files);          // List of input files to handle
154 static LIST_HEAD(map_devs);             // List of device maps
155 static int nfiles = 0;                  // Number of files to handle
156 static int no_stalls = 0;               // Boolean: Disable pre-stalls
157 static unsigned acc_factor = 1;         // Int: Acceleration factor
158 static int find_records = 0;            // Boolean: Find record files auto
159
160 /*
161  * Variables managed under control of condition variables.
162  *
163  * n_reclaims_done:     Counts number of reclaim threads that have completed.
164  * n_replays_done:      Counts number of replay threads that have completed.
165  * n_replays_ready:     Counts number of replay threads ready to start.
166  * n_iters_done:        Counts number of replay threads done one iteration.
167  * iter_start:          Starts an iteration for the replay threads.
168  */
169 static volatile int n_reclaims_done = 0;
170 static pthread_mutex_t reclaim_done_mutex = PTHREAD_MUTEX_INITIALIZER;
171 static pthread_cond_t reclaim_done_cond = PTHREAD_COND_INITIALIZER;
172
173 static volatile int n_replays_done = 0;
174 static pthread_mutex_t replay_done_mutex = PTHREAD_MUTEX_INITIALIZER;
175 static pthread_cond_t replay_done_cond = PTHREAD_COND_INITIALIZER;
176
177 static volatile int n_replays_ready = 0;
178 static pthread_mutex_t replay_ready_mutex = PTHREAD_MUTEX_INITIALIZER;
179 static pthread_cond_t replay_ready_cond = PTHREAD_COND_INITIALIZER;
180
181 static volatile int n_iters_done = 0;
182 static pthread_mutex_t iter_done_mutex = PTHREAD_MUTEX_INITIALIZER;
183 static pthread_cond_t iter_done_cond = PTHREAD_COND_INITIALIZER;
184
185 static volatile int iter_start = 0;
186 static pthread_mutex_t iter_start_mutex = PTHREAD_MUTEX_INITIALIZER;
187 static pthread_cond_t iter_start_cond = PTHREAD_COND_INITIALIZER;
188
189 /* 
190  * ========================================================================
191  * ==== FORWARD REFERENECES ===============================================
192  * ========================================================================
193  */
194
195 static void *replay_sub(void *arg);
196 static void *replay_rec(void *arg);
197 static char usage_str[];
198
199 /* 
200  * ========================================================================
201  * ==== INLINE ROUTINES ===================================================
202  * ========================================================================
203  */
204
205 /*
206  * The 'fatal' macro will output a perror message (if errstring is !NULL)
207  * and display a string (with variable arguments) and then exit with the 
208  * specified exit value.
209  */
210 #define ERR_ARGS                        1
211 #define ERR_SYSCALL                     2
212 static inline void fatal(const char *errstring, const int exitval,
213                          const char *fmt, ...)
214 {
215         va_list ap;
216
217         if (errstring)
218                 perror(errstring);
219
220         va_start(ap, fmt);
221         vfprintf(stderr, fmt, ap);
222         va_end(ap);
223
224         exit(exitval);
225         /*NOTREACHED*/
226 }
227
228 static inline long long unsigned du64_to_sec(__u64 du64)
229 {
230         return (long long unsigned)du64 / (1000 * 1000 * 1000);
231 }
232
233 static inline long long unsigned du64_to_nsec(__u64 du64)
234 {
235         return llabs((long long)du64) % (1000 * 1000 * 1000);
236 }
237
238 /**
239  * min - Return minimum of two integers
240  */
241 static inline int min(int a, int b)
242
243         return a < b ? a : b;
244 }
245
246 /**
247  * minl - Return minimum of two longs
248  */
249 static inline long minl(long a, long b)
250
251         return a < b ? a : b;
252 }
253
254 /**
255  * usage - Display usage string and version
256  */
257 static inline void usage(void)
258 {
259         fprintf(stderr, "Usage: btreplay -- version %s\n%s", 
260                 my_btversion, usage_str);
261 }
262
263 /**
264  * is_send_done - Returns true if sender should quit early
265  * @tip: Per-thread information
266  */
267 static inline int is_send_done(struct thr_info *tip)
268 {
269         return signal_done || tip->send_done;
270 }
271
272 /**
273  * is_reap_done - Returns true if reaper should quit early
274  * @tip: Per-thread information
275  */
276 static inline int is_reap_done(struct thr_info *tip)
277 {
278         return tip->send_done && tip->naios_out == 0;
279 }
280
281 /**
282  * ts2ns - Convert timespec values to a nanosecond value
283  */
284 #define NS_TICKS                ((__u64)1000 * (__u64)1000 * (__u64)1000)
285 static inline __u64 ts2ns(struct timespec *ts)
286 {
287         return ((__u64)(ts->tv_sec) * NS_TICKS) + (__u64)(ts->tv_nsec);
288 }
289
290 /**
291  * ts2ns - Convert timeval values to a nanosecond value
292  */
293 static inline __u64 tv2ns(struct timeval *tp)
294 {
295         return ((__u64)(tp->tv_sec)) + ((__u64)(tp->tv_usec) * (__u64)1000);
296 }
297
298 /**
299  * touch_memory - Force physical memory to be allocating it
300  * 
301  * For malloc()ed memory we need to /touch/ it to make it really
302  * exist. Otherwise, for write's (to storage) things may not work
303  * as planned - we see Linux just use a single area to /read/ from
304  * (as there isn't any memory that has been associated with the 
305  * allocated virtual addresses yet).
306  */
307 static inline void touch_memory(char *buf, size_t bsize)
308 {
309 #if defined(PREP_BUFS)
310         memset(buf, 0, bsize);
311 #else
312         size_t i;
313
314         for (i = 0; i < bsize; i += pgsize)
315                 buf[i] = 0;
316 #endif
317 }
318
319 /**
320  * buf_alloc - Returns a page-aligned buffer of the specified size
321  * @nbytes: Number of bytes to allocate
322  */
323 static inline void *buf_alloc(size_t nbytes)
324 {
325         void *buf;
326
327         if (posix_memalign(&buf, pgsize, nbytes)) {
328                 fatal("posix_memalign", ERR_SYSCALL, "Allocation failed\n");
329                 /*NOTREACHED*/
330         }
331
332         return buf;
333 }
334
335 /**
336  * gettime - Returns current time 
337  */
338 static inline __u64 gettime(void)
339 {
340         static int use_clock_gettime = -1;              // Which clock to use
341
342         if (use_clock_gettime < 0) {
343                 use_clock_gettime = clock_getres(CLOCK_MONOTONIC, NULL) == 0;
344                 if (use_clock_gettime) {
345                         struct timespec ts = {
346                                 .tv_sec = 0,
347                                 .tv_nsec = 0
348                         };
349                         clock_settime(CLOCK_MONOTONIC, &ts);
350                 }
351         }
352
353         if (use_clock_gettime) {
354                 struct timespec ts;
355                 clock_gettime(CLOCK_MONOTONIC, &ts);
356                 return ts2ns(&ts);
357         }
358         else {
359                 struct timeval tp;
360                 gettimeofday(&tp, NULL);
361                 return tv2ns(&tp);
362         }
363 }
364
365 /**
366  * setup_signal - Set up a signal handler for the specified signum
367  */
368 static inline void setup_signal(int signum, sighandler_t handler)
369 {
370         if (signal(signum, handler) == SIG_ERR) {
371                 fatal("signal", ERR_SYSCALL, "Failed to set signal %d\n",
372                         signum);
373                 /*NOTREACHED*/
374         }
375 }
376
377 /* 
378  * ========================================================================
379  * ==== CONDITION VARIABLE ROUTINES =======================================
380  * ========================================================================
381  */
382
383 /**
384  * __set_cv - Increments a variable under condition variable control.
385  * @pmp:        Pointer to the associated mutex
386  * @pcp:        Pointer to the associated condition variable
387  * @vp:         Pointer to the variable being incremented
388  * @mxv:        Max value for variable (Used only when ASSERTS are on)
389  */
390 static inline void __set_cv(pthread_mutex_t *pmp, pthread_cond_t *pcp,
391                             volatile int *vp, 
392                             __attribute__((__unused__))int mxv)
393 {
394         pthread_mutex_lock(pmp);
395         assert(*vp < mxv);
396         *vp += 1;
397         pthread_cond_signal(pcp);
398         pthread_mutex_unlock(pmp);
399 }
400
401 /**
402  * __wait_cv - Waits for a variable under cond var control to hit a value
403  * @pmp:        Pointer to the associated mutex
404  * @pcp:        Pointer to the associated condition variable
405  * @vp:         Pointer to the variable being incremented
406  * @mxv:        Value to wait for
407  */
408 static inline void __wait_cv(pthread_mutex_t *pmp, pthread_cond_t *pcp,
409                              volatile int *vp, int mxv)
410 {
411         pthread_mutex_lock(pmp);
412         while (*vp < mxv)
413                 pthread_cond_wait(pcp, pmp);
414         *vp = 0;
415         pthread_mutex_unlock(pmp);
416 }
417
418 static inline void set_reclaim_done(void)
419 {
420         __set_cv(&reclaim_done_mutex, &reclaim_done_cond, &n_reclaims_done,
421                  nfiles);
422 }
423
424 static inline void wait_reclaims_done(void)
425 {
426         __wait_cv(&reclaim_done_mutex, &reclaim_done_cond, &n_reclaims_done,
427                   nfiles);
428 }
429
430 static inline void set_replay_ready(void)
431 {
432         __set_cv(&replay_ready_mutex, &replay_ready_cond, &n_replays_ready,
433                  nfiles);
434 }
435
436 static inline void wait_replays_ready(void)
437 {
438         __wait_cv(&replay_ready_mutex, &replay_ready_cond, &n_replays_ready,
439                   nfiles);
440 }
441
442 static inline void set_replay_done(void)
443 {
444         __set_cv(&replay_done_mutex, &replay_done_cond, &n_replays_done,
445                 nfiles);
446 }
447
448 static inline void wait_replays_done(void)
449 {
450         __wait_cv(&replay_done_mutex, &replay_done_cond, &n_replays_done,
451                   nfiles);
452 }
453
454 static inline void set_iter_done(void)
455 {
456         __set_cv(&iter_done_mutex, &iter_done_cond, &n_iters_done,
457                 nfiles);
458 }
459
460 static inline void wait_iters_done(void)
461 {
462         __wait_cv(&iter_done_mutex, &iter_done_cond, &n_iters_done,
463                   nfiles);
464 }
465
466 /**
467  * wait_iter_start - Wait for an iteration to start 
468  * 
469  * This is /slightly/ different: we are waiting for a value to become
470  * non-zero, and then we decrement it and go on. 
471  */
472 static inline void wait_iter_start(void)
473 {
474         pthread_mutex_lock(&iter_start_mutex);
475         while (iter_start == 0)
476                 pthread_cond_wait(&iter_start_cond, &iter_start_mutex);
477         assert(1 <= iter_start && iter_start <= nfiles);
478         iter_start--;
479         pthread_mutex_unlock(&iter_start_mutex);
480 }
481
482 /**
483  * start_iter - Start an iteration at the replay thread level
484  */
485 static inline void start_iter(void)
486 {
487         pthread_mutex_lock(&iter_start_mutex);
488         assert(iter_start == 0);
489         iter_start = nfiles;
490         pthread_cond_broadcast(&iter_start_cond);
491         pthread_mutex_unlock(&iter_start_mutex);
492 }
493
494 /* 
495  * ========================================================================
496  * ==== CPU RELATED ROUTINES ==============================================
497  * ========================================================================
498  */
499
500 /**
501  * get_ncpus - Sets up the global 'ncpus' value
502  */
503 static void get_ncpus(void)
504 {
505         cpu_set_t cpus;
506
507         if (sched_getaffinity(getpid(), sizeof(cpus), &cpus)) {
508                 fatal("sched_getaffinity", ERR_SYSCALL, "Can't get CPU info\n");
509                 /*NOTREACHED*/
510         }
511
512         /*
513          * XXX This assumes (perhaps wrongly) that there are no /holes/ 
514          * XXX in the mask.
515          */
516         for (ncpus = 0; ncpus < CPU_SETSIZE && CPU_ISSET(ncpus, &cpus); ncpus++)
517                 ;
518         if (ncpus == 0) {
519                 fatal(NULL, ERR_SYSCALL, "Insufficient number of CPUs\n");
520                 /*NOTREACHED*/
521         }
522 }
523
524 /**
525  * pin_to_cpu - Pin this thread to a specific CPU
526  * @tip: Thread information
527  */
528 static void pin_to_cpu(struct thr_info *tip)
529 {
530         cpu_set_t cpus;
531
532         assert(0 <= tip->cpu && tip->cpu < ncpus);
533
534         CPU_ZERO(&cpus);
535         CPU_SET(tip->cpu, &cpus);
536         if (sched_setaffinity(getpid(), sizeof(cpus), &cpus)) {
537                 fatal("sched_setaffinity", ERR_SYSCALL, "Failed to pin CPU\n");
538                 /*NOTREACHED*/
539         }
540
541         if (verbose > 1) {
542                 int i;
543                 cpu_set_t now;
544
545                 (void)sched_getaffinity(getpid(), sizeof(now), &now);
546                 fprintf(tip->vfp, "Pinned to CPU %02d ", tip->cpu);
547                 for (i = 0; i < ncpus; i++)
548                         fprintf(tip->vfp, "%1d", CPU_ISSET(i, &now));
549                 fprintf(tip->vfp, "\n");
550         }
551 }
552
553 /* 
554  * ========================================================================
555  * ==== INPUT DEVICE HANDLERS =============================================
556  * ========================================================================
557  */
558
559 /**
560  * add_input_dev - Add a device ('sd*') to the list of devices to handle
561  */
562 static void add_input_dev(char *devnm)
563 {
564         struct list_head *p;
565         struct dev_info *dip;
566
567         __list_for_each(p, &input_devs) {
568                 dip = list_entry(p, struct dev_info, head);
569                 if (strcmp(dip->devnm, devnm) == 0)
570                         return;
571         }
572
573         dip = malloc(sizeof(*dip));
574         dip->devnm = strdup(devnm);
575         list_add_tail(&dip->head, &input_devs);
576 }
577
578 /**
579  * rem_input_dev - Remove resources associated with this device
580  */
581 static void rem_input_dev(struct dev_info *dip)
582 {
583         list_del(&dip->head);
584         free(dip->devnm);
585         free(dip);
586 }
587
588 static void find_input_devs(char *idir)
589 {
590         struct dirent *ent;
591         DIR *dir = opendir(idir);
592
593         if (dir == NULL) {
594                 fatal(idir, ERR_ARGS, "Unable to open %s\n", idir);
595                 /*NOTREACHED*/
596         }
597
598         while ((ent = readdir(dir)) != NULL) {
599                 char *p, *dsf = malloc(256);
600
601                 if (strstr(ent->d_name, ".replay.") == NULL)
602                         continue;
603
604                 dsf = strdup(ent->d_name);
605                 p = index(dsf, '.');
606                 assert(p != NULL);
607                 *p = '\0';
608                 add_input_dev(dsf);
609                 free(dsf);
610         }
611
612         closedir(dir);
613 }
614
615 /* 
616  * ========================================================================
617  * ==== MAP DEVICE INTERFACES =============================================
618  * ========================================================================
619  */
620
621 /**
622  * read_map_devs - Read in a set of device mapping from the provided file.
623  * @file_name:  File containing device maps
624  *
625  * We support the notion of multiple such files being specifed on the cmd line
626  */
627 static void read_map_devs(char *file_name)
628 {
629         FILE *fp;
630         char *from_dev, *to_dev;
631
632         fp = fopen(file_name, "r");
633         if (!fp) {
634                 fatal(file_name, ERR_SYSCALL, "Could not open map devs file\n");
635                 /*NOTREACHED*/
636         }
637
638         while (fscanf(fp, "%as %as", &from_dev, &to_dev) == 2) {
639                 struct map_dev *mdp = malloc(sizeof(*mdp));
640
641                 mdp->from_dev = from_dev;
642                 mdp->to_dev = to_dev;
643                 list_add_tail(&mdp->head, &map_devs);
644         }
645
646         fclose(fp);
647 }
648
649 /**
650  * release_map_devs - Release resources associated with device mappings.
651  */
652 static void release_map_devs(void)
653 {
654         struct list_head *p, *q;
655
656         list_for_each_safe(p, q, &map_devs) {
657                 struct map_dev *mdp = list_entry(p, struct map_dev, head);
658
659                 list_del(&mdp->head);
660
661                 free(mdp->from_dev);
662                 free(mdp->to_dev);
663                 free(mdp);
664         }
665 }
666
667 /**
668  * map_dev - Return the mapped device for that specified
669  * @from_dev:   Device name as seen on recorded system
670  *
671  * Note: If there is no such mapping, we return the same name.
672  */
673 static char *map_dev(char *from_dev)
674 {
675         struct list_head *p;
676
677         __list_for_each(p, &map_devs) {
678                 struct map_dev *mdp = list_entry(p, struct map_dev, head);
679
680                 if (strcmp(from_dev, mdp->from_dev) == 0)
681                         return mdp->to_dev;
682         }
683
684         return from_dev;
685 }
686
687 /* 
688  * ========================================================================
689  * ==== IOCB MANAGEMENT ROUTINES ==========================================
690  * ========================================================================
691  */
692
693 /**
694  * iocb_init - Initialize the fields of an IOCB
695  * @tip: Per-thread information
696  * iocbp: IOCB pointer to update
697  */
698 static void iocb_init(struct thr_info *tip, struct iocb_pkt *iocbp)
699 {
700         iocbp->tip = tip;
701         iocbp->nbytes = 0;
702         iocbp->iocb.u.c.buf = NULL;
703 }
704
705 /**
706  * iocb_setup - Set up an iocb with this AIOs information
707  * @iocbp: IOCB pointer to update
708  * @rw: Direction (0 == write, 1 == read)
709  * @n: Number of bytes to transfer
710  * @off: Offset (in bytes)
711  */
712 static void iocb_setup(struct iocb_pkt *iocbp, int rw, int n, long long off)
713 {
714         char *buf;
715         struct iocb *iop = &iocbp->iocb;
716
717         assert(rw == 0 || rw == 1);
718         assert(0 < n && (n % nb_sec) == 0);
719         assert(0 <= off);
720
721         if (iocbp->nbytes) {
722                 if (iocbp->nbytes >= n) {
723                         buf = iop->u.c.buf;
724                         goto prep;
725                 }
726
727                 assert(iop->u.c.buf);
728                 free(iop->u.c.buf);
729         }
730
731         buf = buf_alloc(n);
732         iocbp->nbytes = n;
733
734 prep:
735         if (rw)
736                 io_prep_pread(iop, iocbp->tip->ofd, buf, n, off);
737         else {
738                 assert(write_enabled);
739                 io_prep_pwrite(iop, iocbp->tip->ofd, buf, n, off);
740                 touch_memory(buf, n);
741         }
742
743         iop->data = iocbp;
744 }
745
746 /* 
747  * ========================================================================
748  * ==== PER-THREAD SET UP & TEAR DOWN =====================================
749  * ========================================================================
750  */
751
752 /**
753  * tip_init - Per thread initialization function
754  */
755 static void tip_init(struct thr_info *tip)
756 {
757         int i;
758
759         INIT_LIST_HEAD(&tip->free_iocbs);
760         INIT_LIST_HEAD(&tip->used_iocbs);
761
762         pthread_mutex_init(&tip->mutex, NULL);
763         pthread_cond_init(&tip->cond, NULL);
764
765         if (io_setup(naios, &tip->ctx)) {
766                 fatal("io_setup", ERR_SYSCALL, "io_setup failed\n");
767                 /*NOTREACHED*/
768         }
769
770         tip->ofd = -1;
771         tip->naios_out = 0;
772         tip->send_done = tip->reap_done = 0;
773         tip->send_wait = tip->reap_wait = 0;
774
775         memset(&tip->sub_thread, 0, sizeof(tip->sub_thread));
776         memset(&tip->rec_thread, 0, sizeof(tip->rec_thread));
777
778         for (i = 0; i < naios; i++) {
779                 struct iocb_pkt *iocbp = buf_alloc(sizeof(*iocbp));
780
781                 iocb_init(tip, iocbp);
782                 list_add_tail(&iocbp->head, &tip->free_iocbs);
783         }
784         tip->naios_free = naios;
785
786         if (verbose > 1) {
787                 char fn[MAXPATHLEN];
788
789                 sprintf(fn, "%s/%s.%s.%d.rep", idir, tip->devnm, ibase, 
790                         tip->cpu);
791                 tip->vfp = fopen(fn, "w");
792                 if (!tip->vfp) {
793                         fatal(fn, ERR_SYSCALL, "Failed to open report\n");
794                         /*NOTREACHED*/
795                 }
796
797                 setlinebuf(tip->vfp);
798         }
799
800         if (pthread_create(&tip->sub_thread, NULL, replay_sub, tip)) {
801                 fatal("pthread_create", ERR_SYSCALL, 
802                         "thread create failed\n");
803                 /*NOTREACHED*/
804         }
805
806         if (pthread_create(&tip->rec_thread, NULL, replay_rec, tip)) {
807                 fatal("pthread_create", ERR_SYSCALL, 
808                         "thread create failed\n");
809                 /*NOTREACHED*/
810         }
811 }
812
813 /**
814  * tip_release - Release resources associated with this thread
815  */
816 static void tip_release(struct thr_info *tip)
817 {
818         struct list_head *p, *q;
819
820         assert(tip->send_done);
821         assert(tip->reap_done);
822         assert(list_len(&tip->used_iocbs) == 0);
823         assert(tip->naios_free == naios);
824
825         if (pthread_join(tip->sub_thread, NULL)) {
826                 fatal("pthread_join", ERR_SYSCALL, "pthread sub join failed\n");
827                 /*NOTREACHED*/
828         }
829         if (pthread_join(tip->rec_thread, NULL)) {
830                 fatal("pthread_join", ERR_SYSCALL, "pthread rec join failed\n");
831                 /*NOTREACHED*/
832         }
833
834         io_destroy(tip->ctx);
835
836         list_splice(&tip->used_iocbs, &tip->free_iocbs);
837         list_for_each_safe(p, q, &tip->free_iocbs) {
838                 struct iocb_pkt *iocbp = list_entry(p, struct iocb_pkt, head);
839
840                 list_del(&iocbp->head);
841                 if (iocbp->nbytes) 
842                         free(iocbp->iocb.u.c.buf);
843                 free(iocbp);
844         }
845
846         pthread_cond_destroy(&tip->cond);
847         pthread_mutex_destroy(&tip->mutex);
848 }
849
850 /**
851  * add_input_file - Allocate and initialize per-input file structure
852  * @cpu: CPU for this file
853  * @devnm: Device name for this file
854  * @file_name: Fully qualifed input file name
855  */
856 static void add_input_file(int cpu, char *devnm, char *file_name)
857 {
858         struct stat buf;
859         struct io_file_hdr hdr;
860         struct thr_info *tip = buf_alloc(sizeof(*tip));
861         __u64 my_version = mk_btversion(btver_mjr, btver_mnr, btver_sub);
862
863         assert(0 <= cpu && cpu < ncpus);
864
865         memset(&hdr, 0, sizeof(hdr));
866         memset(tip, 0, sizeof(*tip));
867         tip->cpu = cpu % cpus_to_use;
868         tip->iterations = def_iterations;
869
870         tip->ifd = open(file_name, O_RDONLY);
871         if (tip->ifd < 0) {
872                 fatal(file_name, ERR_ARGS, "Unable to open\n");
873                 /*NOTREACHED*/
874         }
875         if (fstat(tip->ifd, &buf) < 0) {
876                 fatal(file_name, ERR_SYSCALL, "fstat failed\n");
877                 /*NOTREACHED*/
878         }
879         if (buf.st_size < (off_t)sizeof(hdr)) {
880                 if (verbose)
881                         fprintf(stderr, "\t%s empty\n", file_name);
882                 goto empty_file;
883         }
884
885         if (read(tip->ifd, &hdr, sizeof(hdr)) != sizeof(hdr)) {
886                 fatal(file_name, ERR_ARGS, "Header read failed\n");
887                 /*NOTREACHED*/
888         }
889
890         if (hdr.version != my_version) {
891                 fprintf(stderr, "%llx %llx %llx %llx\n", 
892                         (long long unsigned)hdr.version,
893                         (long long unsigned)hdr.genesis,
894                         (long long unsigned)hdr.nbunches,
895                         (long long unsigned)hdr.total_pkts);
896                 fatal(NULL, ERR_ARGS, 
897                         "BT version mismatch: %lx versus my %lx\n",
898                         (long)hdr.version, (long)my_version);
899                         
900         }
901
902         if (hdr.nbunches == 0) {
903 empty_file:
904                 close(tip->ifd);
905                 free(tip);
906                 return;
907         }
908
909         if (hdr.genesis < genesis) {
910                 if (verbose > 1)
911                         fprintf(stderr, "Setting genesis to %llu.%llu\n",
912                                 du64_to_sec(hdr.genesis),
913                                 du64_to_nsec(hdr.genesis));
914                 genesis = hdr.genesis;
915         }
916
917         tip->devnm = strdup(devnm);
918         tip->file_name = strdup(file_name);
919
920         list_add_tail(&tip->head, &input_files);
921
922         if (verbose)
923                 fprintf(stderr, "Added %s %llu\n", file_name, 
924                         (long long)hdr.genesis);
925 }
926
927 /**
928  * rem_input_file - Release resources associated with an input file
929  * @tip: Per-input file information
930  */
931 static void rem_input_file(struct thr_info *tip)
932 {
933         list_del(&tip->head);
934
935         tip_release(tip);
936
937         close(tip->ofd);
938         close(tip->ifd);
939         free(tip->file_name);
940         free(tip->devnm);
941         free(tip);
942 }
943
944 /**
945  * rem_input_files - Remove all input files
946  */
947 static void rem_input_files(void)
948 {
949         struct list_head *p, *q;
950
951         list_for_each_safe(p, q, &input_files) {
952                 rem_input_file(list_entry(p, struct thr_info, head));
953         }
954 }
955
956 /**
957  * __find_input_files - Find input files associated with this device (per cpu)
958  */
959 static void __find_input_files(struct dev_info *dip)
960 {
961         int cpu = 0;
962
963         for (;;) {
964                 char full_name[MAXPATHLEN];
965
966                 sprintf(full_name, "%s/%s.%s.%d", idir, dip->devnm, ibase, cpu);
967                 if (access(full_name, R_OK) != 0)
968                         break;
969
970                 add_input_file(cpu, dip->devnm, full_name);
971                 cpu++;
972         }
973
974         if (!cpu) {
975                 fatal(NULL, ERR_ARGS, "No traces found for %s\n", dip->devnm);
976                 /*NOTREACHED*/
977         }
978
979         rem_input_dev(dip);
980 }
981
982
983 /**
984  * find_input_files - Find input files for all devices
985  */
986 static void find_input_files(void)
987 {
988         struct list_head *p, *q;
989
990         list_for_each_safe(p, q, &input_devs) {
991                 __find_input_files(list_entry(p, struct dev_info, head));
992         }
993 }
994
995 /* 
996  * ========================================================================
997  * ==== RECLAIM ROUTINES ==================================================
998  * ========================================================================
999  */
1000
1001 /**
1002  * reap_wait_aios - Wait for and return number of outstanding AIOs
1003  *
1004  * Will return 0 if we are done
1005  */
1006 static int reap_wait_aios(struct thr_info *tip)
1007 {
1008         int naios = 0;
1009
1010         if (!is_reap_done(tip)) {
1011                 pthread_mutex_lock(&tip->mutex);
1012                 while (tip->naios_out == 0) {
1013                         tip->reap_wait = 1;
1014                         if (pthread_cond_wait(&tip->cond, &tip->mutex)) {
1015                                 fatal("pthread_cond_wait", ERR_SYSCALL, 
1016                                         "nfree_current cond wait failed\n");
1017                                 /*NOTREACHED*/
1018                         }
1019                 }
1020                 naios = tip->naios_out;
1021                 pthread_mutex_unlock(&tip->mutex);
1022         }
1023         assert(is_reap_done(tip) || naios > 0);
1024
1025         return is_reap_done(tip) ? 0 : naios;
1026 }
1027
1028 /**
1029  * reclaim_ios - Reclaim AIOs completed, recycle IOCBs
1030  * @tip: Per-thread information
1031  * @naios_out: Number of AIOs we have outstanding (min)
1032  */
1033 static void reclaim_ios(struct thr_info *tip, long naios_out)
1034 {
1035         long i, ndone;
1036         struct io_event *evp, events[naios_out];
1037
1038 again:
1039         assert(naios > 0);
1040         for (;;) {
1041                 ndone = io_getevents(tip->ctx, 1, naios_out, events, NULL);
1042                 if (ndone > 0)
1043                         break;
1044
1045                 if (errno && errno != EINTR) {
1046                         fatal("io_getevents", ERR_SYSCALL, 
1047                                 "io_getevents failed\n");
1048                         /*NOTREACHED*/
1049                 }
1050         }
1051         assert(0 < ndone && ndone <= naios_out);
1052
1053         pthread_mutex_lock(&tip->mutex);
1054         for (i = 0, evp = events; i < ndone; i++, evp++) {
1055                 struct iocb_pkt *iocbp = evp->data;
1056
1057                 if (evp->res != iocbp->iocb.u.c.nbytes) {
1058                         fatal(NULL, ERR_SYSCALL,
1059                               "Event failure %ld/%ld\t(%ld + %ld)\n",
1060                               (long)evp->res, (long)evp->res2,
1061                               (long)iocbp->iocb.u.c.offset / nb_sec, 
1062                               (long)iocbp->iocb.u.c.nbytes / nb_sec);
1063                         /*NOTREACHED*/
1064                 }
1065
1066                 list_move_tail(&iocbp->head, &tip->free_iocbs);
1067         }
1068
1069         tip->naios_free += ndone;
1070         tip->naios_out -= ndone;
1071         naios_out = minl(naios_out, tip->naios_out);
1072
1073         if (tip->send_wait) {
1074                 tip->send_wait = 0;
1075                 pthread_cond_signal(&tip->cond);
1076         }
1077         pthread_mutex_unlock(&tip->mutex);
1078
1079         /*
1080          * Short cut: If we /know/ there are some more AIOs, go handle them
1081          */
1082         if (naios_out)
1083                 goto again;
1084 }
1085
1086 /**
1087  * replay_rec - Worker thread to reclaim AIOs
1088  * @arg: Pointer to thread information
1089  */
1090 static void *replay_rec(void *arg)
1091 {
1092         long naios_out;
1093         struct thr_info *tip = arg;
1094
1095         while ((naios_out = reap_wait_aios(tip)) > 0) 
1096                 reclaim_ios(tip, naios_out);
1097
1098         assert(tip->send_done);
1099         tip->reap_done = 1;
1100         set_reclaim_done();
1101
1102         return NULL;
1103 }
1104
1105 /* 
1106  * ========================================================================
1107  * ==== REPLAY ROUTINES ===================================================
1108  * ========================================================================
1109  */
1110
1111 /**
1112  * next_bunch - Retrieve next bunch of AIOs to process
1113  * @tip: Per-thread information
1114  * @bunch: Bunch information
1115  *
1116  * Returns TRUE if we recovered a bunch of IOs, else hit EOF
1117  */
1118 static int next_bunch(struct thr_info *tip, struct io_bunch *bunch)
1119 {
1120         size_t count, result;
1121         
1122         result = read(tip->ifd, &bunch->hdr, sizeof(bunch->hdr));
1123         if (result != sizeof(bunch->hdr)) {
1124                 if (result == 0)
1125                         return 0;
1126
1127                 fatal(tip->file_name, ERR_SYSCALL, "Short hdr(%ld)\n", 
1128                         (long)result);
1129                 /*NOTREACHED*/
1130         }
1131         assert(bunch->hdr.npkts <= BT_MAX_PKTS);
1132
1133         count = bunch->hdr.npkts * sizeof(struct io_pkt);
1134         result = read(tip->ifd, &bunch->pkts, count);
1135         if (result != count) {
1136                 fatal(tip->file_name, ERR_SYSCALL, "Short pkts(%ld/%ld)\n", 
1137                         (long)result, (long)count);
1138                 /*NOTREACHED*/
1139         }
1140
1141         return 1;
1142 }
1143
1144 /**
1145  * nfree_current - Returns current number of AIOs that are free
1146  *
1147  * Will wait for available ones...
1148  *
1149  * Returns 0 if we have some condition that causes us to exit
1150  */
1151 static int nfree_current(struct thr_info *tip)
1152 {
1153         int nfree = 0;
1154
1155         pthread_mutex_lock(&tip->mutex);
1156         while (!is_send_done(tip) && ((nfree = tip->naios_free) == 0)) {
1157                 tip->send_wait = 1;
1158                 if (pthread_cond_wait(&tip->cond, &tip->mutex)) {
1159                         fatal("pthread_cond_wait", ERR_SYSCALL, 
1160                                 "nfree_current cond wait failed\n");
1161                         /*NOTREACHED*/
1162                 }
1163         }
1164         pthread_mutex_unlock(&tip->mutex);
1165
1166         return nfree;
1167 }
1168
1169 /**
1170  * stall - Stall for the number of nanoseconds requested
1171  *
1172  * We may be late, in which case we just return.
1173  */
1174 static void stall(struct thr_info *tip, long long oclock)
1175 {
1176         struct timespec req;
1177         long long dreal, tclock = gettime() - rgenesis;
1178
1179         oclock /= acc_factor;
1180         
1181         if (verbose > 1)
1182                 fprintf(tip->vfp, "   stall(%lld.%09lld, %lld.%09lld)\n",
1183                         du64_to_sec(oclock), du64_to_nsec(oclock),
1184                         du64_to_sec(tclock), du64_to_nsec(tclock));
1185
1186         while (!is_send_done(tip) && tclock < oclock) {
1187                 dreal = oclock - tclock;
1188                 req.tv_sec = dreal / (1000 * 1000 * 1000);
1189                 req.tv_nsec = dreal % (1000 * 1000 * 1000);
1190
1191                 if (verbose > 1) {
1192                         fprintf(tip->vfp, "++ stall(%lld.%09lld) ++\n",
1193                                 (long long)req.tv_sec,
1194                                 (long long)req.tv_nsec);
1195                 }
1196
1197                 if (nanosleep(&req, NULL) < 0 && signal_done)
1198                         break;
1199
1200                 tclock = gettime() - rgenesis;
1201         }
1202 }
1203
1204 /**
1205  * iocbs_map - Map a set of AIOs onto a set of IOCBs
1206  * @tip: Per-thread information
1207  * @list: List of AIOs created
1208  * @pkts: AIOs to map
1209  * @ntodo: Number of AIOs to map
1210  */
1211 static void iocbs_map(struct thr_info *tip, struct iocb **list, 
1212                                              struct io_pkt *pkts, int ntodo)
1213 {
1214         int i;
1215         struct io_pkt *pkt;
1216
1217         assert(0 < ntodo && ntodo <= naios);
1218
1219         pthread_mutex_lock(&tip->mutex);
1220         assert(ntodo <= list_len(&tip->free_iocbs));
1221         for (i = 0, pkt = pkts; i < ntodo; i++, pkt++) {
1222                 __u32 rw = pkt->rw;
1223                 struct iocb_pkt *iocbp;
1224
1225                 if (!pkt->rw && !write_enabled)
1226                         rw = 1;
1227
1228                 if (verbose > 1)
1229                         fprintf(tip->vfp, "\t%10llu + %10llu %c%c\n",
1230                                 (unsigned long long)pkt->sector, 
1231                                 (unsigned long long)pkt->nbytes / nb_sec,
1232                                 rw ? 'R' : 'W', 
1233                                 (rw == 1 && pkt->rw == 0) ? '!' : ' ');
1234                 
1235                 iocbp = list_entry(tip->free_iocbs.next, struct iocb_pkt, head);
1236                 iocb_setup(iocbp, rw, pkt->nbytes, pkt->sector * nb_sec);
1237
1238                 list_move_tail(&iocbp->head, &tip->used_iocbs);
1239                 list[i] = &iocbp->iocb;
1240         }
1241
1242         tip->naios_free -= ntodo;
1243         assert(tip->naios_free >= 0);
1244         pthread_mutex_unlock(&tip->mutex);
1245 }
1246
1247 /**
1248  * process_bunch - Process a bunch of requests
1249  * @tip: Per-thread information
1250  * @bunch: Bunch to process
1251  */
1252 static void process_bunch(struct thr_info *tip, struct io_bunch *bunch)
1253 {
1254         __u64 i = 0;
1255         struct iocb *list[bunch->hdr.npkts];
1256
1257         assert(0 < bunch->hdr.npkts && bunch->hdr.npkts <= BT_MAX_PKTS);
1258         while (!is_send_done(tip) && (i < bunch->hdr.npkts)) {
1259                 long ndone;
1260                 int ntodo = min(nfree_current(tip), bunch->hdr.npkts - i);
1261
1262                 assert(0 < ntodo && ntodo <= naios);
1263                 iocbs_map(tip, list, &bunch->pkts[i], ntodo);
1264                 if (!no_stalls)
1265                         stall(tip, bunch->hdr.time_stamp - genesis);
1266
1267                 if (ntodo) {
1268                         if (verbose > 1)
1269                                 fprintf(tip->vfp, "submit(%d)\n", ntodo);
1270                         ndone = io_submit(tip->ctx, ntodo, list);
1271                         if (ndone != (long)ntodo) {
1272                                 fatal("io_submit", ERR_SYSCALL,
1273                                         "%d: io_submit(%d:%ld) failed (%s)\n", 
1274                                         tip->cpu, ntodo, ndone, 
1275                                         strerror(labs(ndone)));
1276                                 /*NOTREACHED*/
1277                         }
1278
1279                         pthread_mutex_lock(&tip->mutex);
1280                         tip->naios_out += ndone;
1281                         assert(tip->naios_out <= naios);
1282                         if (tip->reap_wait) {
1283                                 tip->reap_wait = 0;
1284                                 pthread_cond_signal(&tip->cond);
1285                         }
1286                         pthread_mutex_unlock(&tip->mutex);
1287
1288                         i += ndone;
1289                         assert(i <= bunch->hdr.npkts);
1290                 }
1291         }
1292 }
1293
1294 /**
1295  * reset_input_file - Reset the input file for the next iteration
1296  * @tip: Thread information
1297  *
1298  * We also do a dummy read of the file header to get us to the first bunch.
1299  */
1300 static void reset_input_file(struct thr_info *tip)
1301 {
1302         struct io_file_hdr hdr;
1303
1304         lseek(tip->ifd, 0, 0);
1305
1306         if (read(tip->ifd, &hdr, sizeof(hdr)) != sizeof(hdr)) {
1307                 fatal(tip->file_name, ERR_ARGS, "Header reread failed\n");
1308                 /*NOTREACHED*/
1309         }
1310 }
1311
1312 /**
1313  * replay_sub - Worker thread to submit AIOs that are being replayed
1314  */
1315 static void *replay_sub(void *arg)
1316 {
1317         char path[MAXPATHLEN];
1318         struct io_bunch bunch;
1319         struct thr_info *tip = arg;
1320
1321         pin_to_cpu(tip);
1322
1323         sprintf(path, "/dev/%s", map_dev(tip->devnm));
1324         tip->ofd = open(path, O_RDWR | O_DIRECT | O_NOATIME);
1325         if (tip->ofd < 0) {
1326                 fatal(path, ERR_SYSCALL, "Failed device open\n");
1327                 /*NOTREACHED*/
1328         }
1329
1330         set_replay_ready();
1331         while (!is_send_done(tip) && tip->iterations--) {
1332                 wait_iter_start();
1333                 if (verbose > 1)
1334                         fprintf(tip->vfp, "\n=== %d ===\n", tip->iterations);
1335                 while (!is_send_done(tip) && next_bunch(tip, &bunch))
1336                         process_bunch(tip, &bunch);
1337                 set_iter_done();
1338                 reset_input_file(tip);
1339         }
1340         tip->send_done = 1;
1341         set_replay_done();
1342
1343         return NULL;
1344 }
1345
1346 /* 
1347  * ========================================================================
1348  * ==== COMMAND LINE ARGUMENT HANDLING ====================================
1349  * ========================================================================
1350  */
1351
1352 static char usage_str[] =                                               \
1353         "\n"                                                            \
1354         "\t[ -c <cpus> : --cpus=<cpus>           ] Default: 1\n"        \
1355         "\t[ -d <dir>  : --input-directory=<dir> ] Default: .\n"        \
1356         "\t[ -F        : --find-records          ] Default: Off\n"      \
1357         "\t[ -h        : --help                  ] Default: Off\n"      \
1358         "\t[ -i <base> : --input-base=<base>     ] Default: replay\n"   \
1359         "\t[ -I <iters>: --iterations=<iters>    ] Default: 1\n"        \
1360         "\t[ -M <file> : --map-devs=<file>       ] Default: None\n"     \
1361         "\t[ -N        : --no-stalls             ] Default: Off\n"      \
1362         "\t[ -x        : --acc-factor            ] Default: 1\n"        \
1363         "\t[ -v        : --verbose               ] Default: Off\n"      \
1364         "\t[ -V        : --version               ] Default: Off\n"      \
1365         "\t[ -W        : --write-enable          ] Default: Off\n"      \
1366         "\t<dev...>                                Default: None\n"     \
1367         "\n";
1368
1369 #define S_OPTS  "c:d:Fhi:I:M:Nx:t:vVW"
1370 static struct option l_opts[] = {
1371         {
1372                 .name = "cpus",
1373                 .has_arg = required_argument,
1374                 .flag = NULL,
1375                 .val = 'c'
1376         },
1377         {
1378                 .name = "input-directory",
1379                 .has_arg = required_argument,
1380                 .flag = NULL,
1381                 .val = 'd'
1382         },
1383         {
1384                 .name = "find-records",
1385                 .has_arg = no_argument,
1386                 .flag = NULL,
1387                 .val = 'F'
1388         },
1389         {
1390                 .name = "help",
1391                 .has_arg = no_argument,
1392                 .flag = NULL,
1393                 .val = 'h'
1394         },
1395         {
1396                 .name = "input-base",
1397                 .has_arg = required_argument,
1398                 .flag = NULL,
1399                 .val = 'i'
1400         },
1401         {
1402                 .name = "iterations",
1403                 .has_arg = required_argument,
1404                 .flag = NULL,
1405                 .val = 'I'
1406         },
1407         {
1408                 .name = "map-devs",
1409                 .has_arg = required_argument,
1410                 .flag = NULL,
1411                 .val = 'M'
1412         },
1413         {
1414                 .name = "no-stalls",
1415                 .has_arg = no_argument,
1416                 .flag = NULL,
1417                 .val = 'N'
1418         },
1419         {
1420                 .name = "acc-factor",
1421                 .has_arg = required_argument,
1422                 .flag = NULL,
1423                 .val = 'x'
1424         },
1425         {
1426                 .name = "verbose",
1427                 .has_arg = no_argument,
1428                 .flag = NULL,
1429                 .val = 'v'
1430         },
1431         {
1432                 .name = "version",
1433                 .has_arg = no_argument,
1434                 .flag = NULL,
1435                 .val = 'V'
1436         },
1437         {
1438                 .name = "write-enable",
1439                 .has_arg = no_argument,
1440                 .flag = NULL,
1441                 .val = 'W'
1442         },
1443         {
1444                 .name = NULL
1445         }
1446 };
1447
1448 /**
1449  * handle_args: Parse passed in argument list
1450  * @argc: Number of arguments in argv
1451  * @argv: Arguments passed in
1452  *
1453  * Does rudimentary parameter verification as well.
1454  */
1455 static void handle_args(int argc, char *argv[])
1456 {
1457         int c;
1458         int r;
1459
1460         while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) {
1461                 switch (c) {
1462                 case 'c': 
1463                         cpus_to_use = atoi(optarg);
1464                         if (cpus_to_use <= 0 || cpus_to_use > ncpus) {
1465                                 fatal(NULL, ERR_ARGS, 
1466                                       "Invalid number of cpus %d (0<x<%d)\n",
1467                                       cpus_to_use, ncpus);
1468                                 /*NOTREACHED*/
1469                         }
1470                         break;
1471
1472                 case 'd':
1473                         idir = optarg;
1474                         if (access(idir, R_OK | X_OK) != 0) {
1475                                 fatal(idir, ERR_ARGS, 
1476                                       "Invalid input directory specified\n");
1477                                 /*NOTREACHED*/
1478                         }
1479                         break;
1480
1481                 case 'F': 
1482                         find_records = 1;
1483                         break;
1484
1485                 case 'h': 
1486                         usage(); 
1487                         exit(0);
1488                         /*NOTREACHED*/
1489
1490                 case 'i': 
1491                         ibase = optarg;
1492                         break;
1493
1494                 case 'I':
1495                         def_iterations = atoi(optarg);
1496                         if (def_iterations <= 0) {
1497                                 fprintf(stderr, 
1498                                         "Invalid number of iterations %d\n",
1499                                         def_iterations);
1500                                 exit(ERR_ARGS);
1501                                 /*NOTREACHED*/
1502                         }
1503                         break;
1504
1505                 case 'M':
1506                         read_map_devs(optarg);
1507                         break;
1508
1509                 case 'N':
1510                         no_stalls = 1;
1511                         break;
1512
1513                 case 'x':
1514                         r = sscanf(optarg,"%u",&acc_factor);
1515                         if (r!=1) {
1516                                 fprintf(stderr,
1517                                         "Invalid acceleration factor\n");
1518                                 exit(ERR_ARGS);
1519                                 /*NOTREACHED*/
1520                         }
1521                         break;
1522
1523                 case 'V':
1524                         fprintf(stderr, "btreplay -- version %s\n", 
1525                                 my_btversion);
1526                         fprintf(stderr, "            Built on %s\n", 
1527                                 build_date);
1528                         exit(0);
1529                         /*NOTREACHED*/
1530
1531                 case 'v':
1532                         verbose++;
1533                         break;
1534
1535                 case 'W':
1536                         write_enabled = 1;
1537                         break;
1538
1539                 default:
1540                         usage();
1541                         fatal(NULL, ERR_ARGS, 
1542                               "Invalid command line argument %c\n", c);
1543                         /*NOTREACHED*/
1544                 }
1545         }
1546
1547         while (optind < argc)
1548                 add_input_dev(argv[optind++]);
1549
1550         if (find_records)
1551                 find_input_devs(idir);
1552
1553         if (list_len(&input_devs) == 0) {
1554                 fatal(NULL, ERR_ARGS, "Missing required input dev name(s)\n");
1555                 /*NOTREACHED*/
1556         }
1557
1558         if (cpus_to_use < 0)
1559                 cpus_to_use = ncpus;
1560 }
1561
1562 /* 
1563  * ========================================================================
1564  * ==== MAIN ROUTINE ======================================================
1565  * ========================================================================
1566  */
1567
1568 /**
1569  * set_signal_done - Signal handler, catches signals & sets signal_done
1570  */
1571 static void set_signal_done(__attribute__((__unused__))int signum)
1572 {
1573         signal_done = 1;
1574 }
1575
1576 /**
1577  * main - 
1578  * @argc: Number of arguments
1579  * @argv: Array of arguments
1580  */
1581 int main(int argc, char *argv[])
1582 {
1583         int i;
1584         struct list_head *p;
1585
1586         pgsize = getpagesize();
1587         assert(pgsize > 0);
1588
1589         setup_signal(SIGINT, set_signal_done);
1590         setup_signal(SIGTERM, set_signal_done);
1591
1592         get_ncpus();
1593         handle_args(argc, argv);
1594         find_input_files();
1595
1596         nfiles = list_len(&input_files);
1597         __list_for_each(p, &input_files) {
1598                 tip_init(list_entry(p, struct thr_info, head));
1599         }
1600
1601         wait_replays_ready();
1602         for (i = 0; i < def_iterations; i++) {
1603                 rgenesis = gettime();
1604                 start_iter();
1605                 if (verbose)
1606                         fprintf(stderr, "I");
1607                 wait_iters_done();
1608         }
1609
1610         wait_replays_done();
1611         wait_reclaims_done();
1612
1613         if (verbose)
1614                 fprintf(stderr, "\n");
1615
1616         rem_input_files();
1617         release_map_devs();
1618
1619         return 0;
1620 }