Home | History | Annotate | Line # | Download | only in raidframe
rf_acctrace.c revision 1.1
      1 /*	$NetBSD: rf_acctrace.c,v 1.1 1998/11/13 04:20:26 oster Exp $	*/
      2 /*
      3  * Copyright (c) 1995 Carnegie-Mellon University.
      4  * All rights reserved.
      5  *
      6  * Author: Mark Holland
      7  *
      8  * Permission to use, copy, modify and distribute this software and
      9  * its documentation is hereby granted, provided that both the copyright
     10  * notice and this permission notice appear in all copies of the
     11  * software, derivative works or modified versions, and any portions
     12  * thereof, and that both notices appear in supporting documentation.
     13  *
     14  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
     15  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
     16  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
     17  *
     18  * Carnegie Mellon requests users of this software to return to
     19  *
     20  *  Software Distribution Coordinator  or  Software.Distribution (at) CS.CMU.EDU
     21  *  School of Computer Science
     22  *  Carnegie Mellon University
     23  *  Pittsburgh PA 15213-3890
     24  *
     25  * any improvements or extensions that they make and grant Carnegie the
     26  * rights to redistribute these changes.
     27  */
     28 
     29 /*****************************************************************************
     30  *
     31  * acctrace.c -- code to support collecting information about each access
     32  *
     33  *****************************************************************************/
     34 
     35 /* :
     36  * Log: rf_acctrace.c,v
     37  * Revision 1.29  1996/07/27 23:36:08  jimz
     38  * Solaris port of simulator
     39  *
     40  * Revision 1.28  1996/07/17  21:00:58  jimz
     41  * clean up timer interface, tracing
     42  *
     43  * Revision 1.27  1996/06/14  14:35:24  jimz
     44  * clean up dfstrace protection
     45  *
     46  * Revision 1.26  1996/06/13  19:09:04  jimz
     47  * remove trace.dat file before beginning
     48  *
     49  * Revision 1.25  1996/06/12  04:41:26  jimz
     50  * tweaks to make genplot work with user-level driver
     51  * (mainly change stat collection)
     52  *
     53  * Revision 1.24  1996/06/10  11:55:47  jimz
     54  * Straightened out some per-array/not-per-array distinctions, fixed
     55  * a couple bugs related to confusion. Added shutdown lists. Removed
     56  * layout shutdown function (now subsumed by shutdown lists).
     57  *
     58  * Revision 1.23  1996/06/09  02:36:46  jimz
     59  * lots of little crufty cleanup- fixup whitespace
     60  * issues, comment #ifdefs, improve typing in some
     61  * places (esp size-related)
     62  *
     63  * Revision 1.22  1996/06/05  18:06:02  jimz
     64  * Major code cleanup. The Great Renaming is now done.
     65  * Better modularity. Better typing. Fixed a bunch of
     66  * synchronization bugs. Made a lot of global stuff
     67  * per-desc or per-array. Removed dead code.
     68  *
     69  * Revision 1.21  1996/05/31  22:26:54  jimz
     70  * fix a lot of mapping problems, memory allocation problems
     71  * found some weird lock issues, fixed 'em
     72  * more code cleanup
     73  *
     74  * Revision 1.20  1996/05/30  23:22:16  jimz
     75  * bugfixes of serialization, timing problems
     76  * more cleanup
     77  *
     78  * Revision 1.19  1996/05/30  12:59:18  jimz
     79  * make etimer happier, more portable
     80  *
     81  * Revision 1.18  1996/05/27  18:56:37  jimz
     82  * more code cleanup
     83  * better typing
     84  * compiles in all 3 environments
     85  *
     86  * Revision 1.17  1996/05/23  00:33:23  jimz
     87  * code cleanup: move all debug decls to rf_options.c, all extern
     88  * debug decls to rf_options.h, all debug vars preceded by rf_
     89  *
     90  * Revision 1.16  1996/05/20  16:15:49  jimz
     91  * switch to rf_{mutex,cond}_{init,destroy}
     92  *
     93  * Revision 1.15  1996/05/18  20:10:00  jimz
     94  * bit of cleanup to compile cleanly in kernel, once again
     95  *
     96  * Revision 1.14  1996/05/18  19:51:34  jimz
     97  * major code cleanup- fix syntax, make some types consistent,
     98  * add prototypes, clean out dead code, et cetera
     99  *
    100  * Revision 1.13  1995/11/30  16:26:43  wvcii
    101  * added copyright info
    102  *
    103  */
    104 
    105 #ifdef _KERNEL
    106 #define KERNEL
    107 #endif
    108 
    109 #include "rf_threadstuff.h"
    110 #include "rf_types.h"
    111 #include <sys/stat.h>
    112 #include <sys/types.h>
    113 
    114 #ifdef KERNEL
    115 #ifndef __NetBSD__
    116 #include <dfstrace.h>
    117 #endif /* !__NetBSD__ */
    118 #if DFSTRACE > 0
    119 #include <sys/dfs_log.h>
    120 #include <sys/dfstracebuf.h>
    121 #endif /* DFSTRACE > 0 */
    122 #endif /* KERNEL */
    123 
    124 #include "rf_debugMem.h"
    125 #include "rf_acctrace.h"
    126 #include "rf_general.h"
    127 #include "rf_raid.h"
    128 #include "rf_etimer.h"
    129 #include "rf_hist.h"
    130 #include "rf_shutdown.h"
    131 #include "rf_sys.h"
    132 
    133 static long numTracesSoFar;
    134 static int accessTraceBufCount = 0;
    135 static RF_AccTraceEntry_t *access_tracebuf;
    136 static long traceCount;
    137 
    138 int  rf_stopCollectingTraces;
    139 RF_DECLARE_MUTEX(rf_tracing_mutex)
    140 int rf_trace_fd;
    141 
    142 static void rf_ShutdownAccessTrace(void *);
    143 
    144 static void rf_ShutdownAccessTrace(ignored)
    145   void  *ignored;
    146 {
    147   if (rf_accessTraceBufSize) {
    148     if (accessTraceBufCount) rf_FlushAccessTraceBuf();
    149 #ifndef KERNEL
    150     close(rf_trace_fd);
    151 #endif /* !KERNEL */
    152     RF_Free(access_tracebuf, rf_accessTraceBufSize * sizeof(RF_AccTraceEntry_t));
    153   }
    154   rf_mutex_destroy(&rf_tracing_mutex);
    155 #if defined(KERNEL) && DFSTRACE > 0
    156   printf("RAIDFRAME: %d trace entries were sent to dfstrace\n",traceCount);
    157 #endif /* KERNEL && DFSTRACE > 0 */
    158 }
    159 
    160 int rf_ConfigureAccessTrace(listp)
    161   RF_ShutdownList_t  **listp;
    162 {
    163   int rc;
    164 
    165   numTracesSoFar = accessTraceBufCount = rf_stopCollectingTraces = 0;
    166   if (rf_accessTraceBufSize) {
    167     RF_Malloc(access_tracebuf, rf_accessTraceBufSize * sizeof(RF_AccTraceEntry_t), (RF_AccTraceEntry_t *));
    168     accessTraceBufCount = 0;
    169 #ifndef KERNEL
    170     rc = unlink("trace.dat");
    171     if (rc && (errno != ENOENT)) {
    172       perror("unlink");
    173       RF_ERRORMSG("Unable to remove existing trace.dat\n");
    174       return(errno);
    175     }
    176     if ((rf_trace_fd = open("trace.dat",O_WRONLY|O_CREAT|O_TRUNC, S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0 ) {
    177       perror("Unable to open trace.dat for output");
    178       return(errno);
    179     }
    180 #endif /* !KERNEL */
    181   }
    182   traceCount = 0;
    183   numTracesSoFar = 0;
    184   rc = rf_mutex_init(&rf_tracing_mutex);
    185   if (rc) {
    186     RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
    187       __LINE__, rc);
    188   }
    189   rc = rf_ShutdownCreate(listp, rf_ShutdownAccessTrace, NULL);
    190   if (rc) {
    191     RF_ERRORMSG3("Unable to add to shutdown list file %s line %d rc=%d\n", __FILE__,
    192       __LINE__, rc);
    193     if (rf_accessTraceBufSize) {
    194       RF_Free(access_tracebuf, rf_accessTraceBufSize * sizeof(RF_AccTraceEntry_t));
    195 #ifndef KERNEL
    196       close(rf_trace_fd);
    197 #endif /* !KERNEL */
    198       rf_mutex_destroy(&rf_tracing_mutex);
    199     }
    200   }
    201   return(rc);
    202 }
    203 
    204 /* install a trace record.  cause a flush to disk or to the trace collector daemon
    205  * if the trace buffer is at least 1/2 full.
    206  */
    207 void rf_LogTraceRec(raid, rec)
    208   RF_Raid_t           *raid;
    209   RF_AccTraceEntry_t  *rec;
    210 {
    211 	RF_AccTotals_t *acc = &raid->acc_totals;
    212 #if 0
    213 	RF_Etimer_t timer;
    214 	int i, n;
    215 #endif
    216 
    217   if (rf_stopCollectingTraces || ((rf_maxNumTraces >= 0) && (numTracesSoFar >= rf_maxNumTraces)))
    218     return;
    219 
    220 #ifndef KERNEL
    221   if (rf_accessTraceBufSize) {
    222     RF_LOCK_MUTEX(rf_tracing_mutex);
    223     numTracesSoFar++;
    224     bcopy((char *)rec, (char *)&access_tracebuf[ accessTraceBufCount++ ], sizeof(RF_AccTraceEntry_t));
    225     if (accessTraceBufCount == rf_accessTraceBufSize)
    226       rf_FlushAccessTraceBuf();
    227     RF_UNLOCK_MUTEX(rf_tracing_mutex);
    228   }
    229 #endif /* !KERNEL */
    230 #if defined(KERNEL) && DFSTRACE > 0
    231   rec->index = traceCount++;
    232   if (traceon & DFS_TRACE_RAIDFRAME) {
    233     dfs_log(DFS_NOTE, (char *) rec, (int) sizeof(*rec), 0);
    234   }
    235 #endif /* KERNEL && DFSTRACE > 0 */
    236 	/* update AccTotals for this device */
    237 	if (!raid->keep_acc_totals)
    238 		return;
    239 	acc->num_log_ents++;
    240 	if (rec->reconacc) {
    241 		acc->recon_start_to_fetch_us += rec->specific.recon.recon_start_to_fetch_us;
    242 		acc->recon_fetch_to_return_us += rec->specific.recon.recon_fetch_to_return_us;
    243 		acc->recon_return_to_submit_us += rec->specific.recon.recon_return_to_submit_us;
    244 		acc->recon_num_phys_ios += rec->num_phys_ios;
    245 		acc->recon_phys_io_us += rec->phys_io_us;
    246 		acc->recon_diskwait_us += rec->diskwait_us;
    247 		acc->recon_reccount++;
    248 	}
    249 	else {
    250 		RF_HIST_ADD(acc->tot_hist, rec->total_us);
    251 		RF_HIST_ADD(acc->dw_hist, rec->diskwait_us);
    252 		/* count of physical ios which are too big.  often due to thermal recalibration */
    253 		/* if bigvals > 0, you should probably ignore this data set */
    254 		if (rec->diskwait_us > 100000)
    255 			acc->bigvals++;
    256 		acc->total_us += rec->total_us;
    257 		acc->suspend_ovhd_us += rec->specific.user.suspend_ovhd_us;
    258 		acc->map_us += rec->specific.user.map_us;
    259 		acc->lock_us += rec->specific.user.lock_us;
    260 		acc->dag_create_us += rec->specific.user.dag_create_us;
    261 		acc->dag_retry_us += rec->specific.user.dag_retry_us;
    262 		acc->exec_us += rec->specific.user.exec_us;
    263 		acc->cleanup_us += rec->specific.user.cleanup_us;
    264 		acc->exec_engine_us += rec->specific.user.exec_engine_us;
    265 		acc->xor_us += rec->xor_us;
    266 		acc->q_us += rec->q_us;
    267 		acc->plog_us += rec->plog_us;
    268 		acc->diskqueue_us += rec->diskqueue_us;
    269 		acc->diskwait_us += rec->diskwait_us;
    270 		acc->num_phys_ios += rec->num_phys_ios;
    271 		acc->phys_io_us = rec->phys_io_us;
    272 		acc->user_reccount++;
    273 	}
    274 }
    275 
    276 
    277 /* assumes the tracing mutex is locked at entry.  In order to allow this to be called
    278  * from interrupt context, we don't do any copyouts here, but rather just wake trace
    279  * buffer collector thread.
    280  */
    281 void rf_FlushAccessTraceBuf()
    282 {
    283 #ifndef KERNEL
    284   int size = accessTraceBufCount * sizeof(RF_AccTraceEntry_t);
    285 
    286   if (write(rf_trace_fd, (char *) access_tracebuf, size) < size ) {
    287     fprintf(stderr, "Unable to write traces to file.  tracing disabled\n");
    288     RF_Free(access_tracebuf, rf_accessTraceBufSize * sizeof(RF_AccTraceEntry_t));
    289     rf_accessTraceBufSize = 0;
    290     close(rf_trace_fd);
    291   }
    292 #endif /* !KERNEL */
    293   accessTraceBufCount = 0;
    294 }
    295