Annotation of src/sys/dev/raidframe/rf_diskqueue.c, revision 1.60

1.60    ! oster       1: /*     $NetBSD: rf_diskqueue.c,v 1.59 2021/07/23 00:26:19 oster Exp $  */
1.1       oster       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@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: 
1.13      oster      29: /****************************************************************************
1.1       oster      30:  *
                     31:  * rf_diskqueue.c -- higher-level disk queue code
                     32:  *
                     33:  * the routines here are a generic wrapper around the actual queueing
1.6       oster      34:  * routines.  The code here implements thread scheduling, synchronization,
1.1       oster      35:  * and locking ops (see below) on top of the lower-level queueing code.
                     36:  *
1.13      oster      37:  * to support atomic RMW, we implement "locking operations".  When a
                     38:  * locking op is dispatched to the lower levels of the driver, the
                     39:  * queue is locked, and no further I/Os are dispatched until the queue
                     40:  * receives & completes a corresponding "unlocking operation".  This
                     41:  * code relies on the higher layers to guarantee that a locking op
                     42:  * will always be eventually followed by an unlocking op.  The model
                     43:  * is that the higher layers are structured so locking and unlocking
                     44:  * ops occur in pairs, i.e.  an unlocking op cannot be generated until
                     45:  * after a locking op reports completion.  There is no good way to
                     46:  * check to see that an unlocking op "corresponds" to the op that
                     47:  * currently has the queue locked, so we make no such attempt.  Since
                     48:  * by definition there can be only one locking op outstanding on a
                     49:  * disk, this should not be a problem.
                     50:  *
                     51:  * In the kernel, we allow multiple I/Os to be concurrently dispatched
                     52:  * to the disk driver.  In order to support locking ops in this
                     53:  * environment, when we decide to do a locking op, we stop dispatching
                     54:  * new I/Os and wait until all dispatched I/Os have completed before
                     55:  * dispatching the locking op.
                     56:  *
                     57:  * Unfortunately, the code is different in the 3 different operating
                     58:  * states (user level, kernel, simulator).  In the kernel, I/O is
                     59:  * non-blocking, and we have no disk threads to dispatch for us.
                     60:  * Therefore, we have to dispatch new I/Os to the scsi driver at the
                     61:  * time of enqueue, and also at the time of completion.  At user
                     62:  * level, I/O is blocking, and so only the disk threads may dispatch
                     63:  * I/Os.  Thus at user level, all we can do at enqueue time is enqueue
                     64:  * and wake up the disk thread to do the dispatch.
1.1       oster      65:  *
1.13      oster      66:  ****************************************************************************/
1.15      lukem      67: 
                     68: #include <sys/cdefs.h>
1.60    ! oster      69: __KERNEL_RCSID(0, "$NetBSD: rf_diskqueue.c,v 1.59 2021/07/23 00:26:19 oster Exp $");
1.1       oster      70: 
1.14      oster      71: #include <dev/raidframe/raidframevar.h>
                     72: 
1.1       oster      73: #include "rf_threadstuff.h"
                     74: #include "rf_raid.h"
                     75: #include "rf_diskqueue.h"
                     76: #include "rf_alloclist.h"
                     77: #include "rf_acctrace.h"
                     78: #include "rf_etimer.h"
                     79: #include "rf_general.h"
                     80: #include "rf_debugprint.h"
                     81: #include "rf_shutdown.h"
                     82: #include "rf_cvscan.h"
                     83: #include "rf_sstf.h"
                     84: #include "rf_fifo.h"
1.11      oster      85: #include "rf_kintf.h"
1.1       oster      86: 
1.59      oster      87: #include <sys/buf.h>
                     88: 
1.1       oster      89: static void rf_ShutdownDiskQueueSystem(void *);
                     90: 
1.21      oster      91: #ifndef RF_DEBUG_DISKQUEUE
                     92: #define RF_DEBUG_DISKQUEUE 0
                     93: #endif
                     94: 
                     95: #if RF_DEBUG_DISKQUEUE
1.1       oster      96: #define Dprintf1(s,a)         if (rf_queueDebug) rf_debug_printf(s,(void *)((unsigned long)a),NULL,NULL,NULL,NULL,NULL,NULL,NULL)
                     97: #define Dprintf2(s,a,b)       if (rf_queueDebug) rf_debug_printf(s,(void *)((unsigned long)a),(void *)((unsigned long)b),NULL,NULL,NULL,NULL,NULL,NULL)
                     98: #define Dprintf3(s,a,b,c)     if (rf_queueDebug) rf_debug_printf(s,(void *)((unsigned long)a),(void *)((unsigned long)b),(void *)((unsigned long)c),NULL,NULL,NULL,NULL,NULL)
1.21      oster      99: #else
                    100: #define Dprintf1(s,a)
                    101: #define Dprintf2(s,a,b)
                    102: #define Dprintf3(s,a,b,c)
                    103: #endif
1.1       oster     104: 
1.13      oster     105: /*****************************************************************************
1.1       oster     106:  *
1.13      oster     107:  * the disk queue switch defines all the functions used in the
                    108:  * different queueing disciplines queue ID, init routine, enqueue
                    109:  * routine, dequeue routine
1.1       oster     110:  *
1.13      oster     111:  ****************************************************************************/
1.1       oster     112: 
1.22      jdolecek  113: static const RF_DiskQueueSW_t diskqueuesw[] = {
1.6       oster     114:        {"fifo",                /* FIFO */
                    115:                rf_FifoCreate,
                    116:                rf_FifoEnqueue,
                    117:                rf_FifoDequeue,
                    118:                rf_FifoPeek,
1.1       oster     119:        rf_FifoPromote},
                    120: 
1.6       oster     121:        {"cvscan",              /* cvscan */
                    122:                rf_CvscanCreate,
                    123:                rf_CvscanEnqueue,
                    124:                rf_CvscanDequeue,
                    125:                rf_CvscanPeek,
                    126:        rf_CvscanPromote},
                    127: 
                    128:        {"sstf",                /* shortest seek time first */
                    129:                rf_SstfCreate,
                    130:                rf_SstfEnqueue,
                    131:                rf_SstfDequeue,
                    132:                rf_SstfPeek,
1.1       oster     133:        rf_SstfPromote},
                    134: 
1.6       oster     135:        {"scan",                /* SCAN (two-way elevator) */
                    136:                rf_ScanCreate,
                    137:                rf_SstfEnqueue,
                    138:                rf_ScanDequeue,
                    139:                rf_ScanPeek,
1.1       oster     140:        rf_SstfPromote},
                    141: 
1.6       oster     142:        {"cscan",               /* CSCAN (one-way elevator) */
                    143:                rf_CscanCreate,
                    144:                rf_SstfEnqueue,
                    145:                rf_CscanDequeue,
                    146:                rf_CscanPeek,
1.1       oster     147:        rf_SstfPromote},
                    148: 
                    149: };
                    150: #define NUM_DISK_QUEUE_TYPES (sizeof(diskqueuesw)/sizeof(RF_DiskQueueSW_t))
                    151: 
1.59      oster     152: 
1.1       oster     153: #define RF_MAX_FREE_DQD 256
1.31      oster     154: #define RF_MIN_FREE_DQD  64
1.1       oster     155: 
1.59      oster     156: /* XXX: scale these... */
                    157: #define RF_MAX_FREE_BUFIO 256
                    158: #define RF_MIN_FREE_BUFIO  64
                    159: 
                    160: 
1.1       oster     161: 
1.6       oster     162: /* configures a single disk queue */
1.9       oster     163: 
1.53      mrg       164: static void
                    165: rf_ShutdownDiskQueue(void *arg)
                    166: {
                    167:        RF_DiskQueue_t *diskqueue = arg;
                    168: 
                    169:        rf_destroy_mutex2(diskqueue->mutex);
                    170: }
                    171: 
1.40      perry     172: int
1.27      oster     173: rf_ConfigureDiskQueue(RF_Raid_t *raidPtr, RF_DiskQueue_t *diskqueue,
                    174:                      RF_RowCol_t c, const RF_DiskQueueSW_t *p,
                    175:                      RF_SectorCount_t sectPerDisk, dev_t dev,
                    176:                      int maxOutstanding, RF_ShutdownList_t **listp,
                    177:                      RF_AllocListElem_t *clList)
1.6       oster     178: {
                    179:        diskqueue->col = c;
                    180:        diskqueue->qPtr = p;
                    181:        diskqueue->qHdr = (p->Create) (sectPerDisk, clList, listp);
                    182:        diskqueue->dev = dev;
                    183:        diskqueue->numOutstanding = 0;
                    184:        diskqueue->queueLength = 0;
                    185:        diskqueue->maxOutstanding = maxOutstanding;
                    186:        diskqueue->curPriority = RF_IO_NORMAL_PRIORITY;
                    187:        diskqueue->flags = 0;
                    188:        diskqueue->raidPtr = raidPtr;
1.23      oster     189:        diskqueue->rf_cinfo = &raidPtr->raid_cinfo[c];
1.53      mrg       190:        rf_init_mutex2(diskqueue->mutex, IPL_VM);
                    191:        rf_ShutdownCreate(listp, rf_ShutdownDiskQueue, diskqueue);
1.6       oster     192:        return (0);
1.1       oster     193: }
                    194: 
1.40      perry     195: static void
1.60    ! oster     196: rf_ShutdownDiskQueueSystem(void *arg)
1.6       oster     197: {
1.60    ! oster     198:        RF_Raid_t *raidPtr;
        !           199: 
        !           200:        raidPtr = (RF_Raid_t *) arg;
        !           201:        
        !           202:        pool_destroy(&raidPtr->pools.dqd);
        !           203:        pool_destroy(&raidPtr->pools.bufio);
1.1       oster     204: }
                    205: 
1.30      oster     206: int
1.60    ! oster     207: rf_ConfigureDiskQueueSystem(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
        !           208:                            RF_Config_t *cfgPtr)
        !           209: 
1.6       oster     210: {
                    211: 
1.60    ! oster     212:        rf_pool_init(raidPtr, raidPtr->poolNames.dqd, &raidPtr->pools.dqd, sizeof(RF_DiskQueueData_t),
        !           213:                     "dqd", RF_MIN_FREE_DQD, RF_MAX_FREE_DQD);
        !           214:        rf_pool_init(raidPtr, raidPtr->poolNames.bufio, &raidPtr->pools.bufio, sizeof(buf_t),
        !           215:                     "bufio", RF_MIN_FREE_BUFIO, RF_MAX_FREE_BUFIO);
        !           216:        rf_ShutdownCreate(listp, rf_ShutdownDiskQueueSystem, raidPtr);
1.24      oster     217: 
1.6       oster     218:        return (0);
                    219: }
                    220: 
1.40      perry     221: int
1.27      oster     222: rf_ConfigureDiskQueues(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
                    223:                       RF_Config_t *cfgPtr)
1.6       oster     224: {
1.23      oster     225:        RF_DiskQueue_t *diskQueues, *spareQueues;
1.22      jdolecek  226:        const RF_DiskQueueSW_t *p;
1.23      oster     227:        RF_RowCol_t r,c;
1.6       oster     228:        int     rc, i;
                    229: 
                    230:        raidPtr->maxQueueDepth = cfgPtr->maxOutstandingDiskReqs;
                    231: 
                    232:        for (p = NULL, i = 0; i < NUM_DISK_QUEUE_TYPES; i++) {
                    233:                if (!strcmp(diskqueuesw[i].queueType, cfgPtr->diskQueueType)) {
                    234:                        p = &diskqueuesw[i];
                    235:                        break;
                    236:                }
                    237:        }
                    238:        if (p == NULL) {
                    239:                RF_ERRORMSG2("Unknown queue type \"%s\".  Using %s\n", cfgPtr->diskQueueType, diskqueuesw[0].queueType);
                    240:                p = &diskqueuesw[0];
                    241:        }
1.10      oster     242:        raidPtr->qType = p;
1.23      oster     243: 
1.54      christos  244:        diskQueues = RF_MallocAndAdd(
                    245:            (raidPtr->numCol + RF_MAXSPARE) * sizeof(*diskQueues),
                    246:            raidPtr->cleanupList);
1.23      oster     247:        if (diskQueues == NULL)
1.6       oster     248:                return (ENOMEM);
                    249:        raidPtr->Queues = diskQueues;
1.23      oster     250: 
                    251:        for (c = 0; c < raidPtr->numCol; c++) {
                    252:                rc = rf_ConfigureDiskQueue(raidPtr, &diskQueues[c],
                    253:                                           c, p,
1.40      perry     254:                                           raidPtr->sectorsPerDisk,
1.23      oster     255:                                           raidPtr->Disks[c].dev,
1.40      perry     256:                                           cfgPtr->maxOutstandingDiskReqs,
1.23      oster     257:                                           listp, raidPtr->cleanupList);
                    258:                if (rc)
                    259:                        return (rc);
1.6       oster     260:        }
                    261: 
1.23      oster     262:        spareQueues = &raidPtr->Queues[raidPtr->numCol];
1.6       oster     263:        for (r = 0; r < raidPtr->numSpare; r++) {
1.9       oster     264:                rc = rf_ConfigureDiskQueue(raidPtr, &spareQueues[r],
1.23      oster     265:                                           raidPtr->numCol + r, p,
                    266:                                           raidPtr->sectorsPerDisk,
                    267:                                           raidPtr->Disks[raidPtr->numCol + r].dev,
                    268:                                           cfgPtr->maxOutstandingDiskReqs, listp,
                    269:                                           raidPtr->cleanupList);
1.6       oster     270:                if (rc)
                    271:                        return (rc);
                    272:        }
                    273:        return (0);
                    274: }
1.1       oster     275: /* Enqueue a disk I/O
                    276:  *
                    277:  * In the kernel, I/O is non-blocking and so we'd like to have multiple
                    278:  * I/Os outstanding on the physical disks when possible.
                    279:  *
                    280:  * when any request arrives at a queue, we have two choices:
                    281:  *    dispatch it to the lower levels
                    282:  *    queue it up
                    283:  *
                    284:  * kernel rules for when to do what:
                    285:  *    unlocking req  :  always dispatch it
                    286:  *    normal req     :  queue empty => dispatch it & set priority
                    287:  *                      queue not full & priority is ok => dispatch it
                    288:  *                      else queue it
                    289:  */
1.40      perry     290: void
1.27      oster     291: rf_DiskIOEnqueue(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req, int pri)
1.6       oster     292: {
                    293:        RF_ETIMER_START(req->qtime);
                    294:        RF_ASSERT(req->type == RF_IO_TYPE_NOP || req->numSector);
                    295:        req->priority = pri;
                    296: 
1.21      oster     297: #if RF_DEBUG_DISKQUEUE
1.6       oster     298:        if (rf_queueDebug && (req->numSector == 0)) {
                    299:                printf("Warning: Enqueueing zero-sector access\n");
                    300:        }
1.21      oster     301: #endif
1.6       oster     302:        RF_LOCK_QUEUE_MUTEX(queue, "DiskIOEnqueue");
1.52      oster     303:        if (RF_OK_TO_DISPATCH(queue, req)) {
                    304:                Dprintf2("Dispatching pri %d regular op to c %d (ok to dispatch)\n", pri, queue->col);
                    305:                rf_DispatchKernelIO(queue, req);
                    306:        } else {
                    307:                queue->queueLength++;   /* increment count of number of requests waiting in this queue */
                    308:                Dprintf2("Enqueueing pri %d regular op to c %d (not ok to dispatch)\n", pri, queue->col);
                    309:                req->queue = (void *) queue;
                    310:                (queue->qPtr->Enqueue) (queue->qHdr, req, pri);
1.6       oster     311:        }
                    312:        RF_UNLOCK_QUEUE_MUTEX(queue, "DiskIOEnqueue");
1.1       oster     313: }
1.6       oster     314: 
1.1       oster     315: 
1.52      oster     316: /* get the next set of I/Os started */
1.40      perry     317: void
1.27      oster     318: rf_DiskIOComplete(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req, int status)
1.6       oster     319: {
                    320:        int     done = 0;
                    321: 
                    322:        RF_LOCK_QUEUE_MUTEX(queue, "DiskIOComplete");
                    323:        queue->numOutstanding--;
                    324:        RF_ASSERT(queue->numOutstanding >= 0);
                    325: 
                    326:        /* dispatch requests to the disk until we find one that we can't. */
                    327:        /* no reason to continue once we've filled up the queue */
                    328:        /* no reason to even start if the queue is locked */
                    329: 
1.52      oster     330:        while (!done && !RF_QUEUE_FULL(queue)) {
                    331:                req = (queue->qPtr->Dequeue) (queue->qHdr);
1.6       oster     332:                if (req) {
1.52      oster     333:                        Dprintf2("DiskIOComplete: extracting pri %d req from queue at c %d\n", req->priority, queue->col);
                    334:                        queue->queueLength--;   /* decrement count of number of requests waiting in this queue */
1.6       oster     335:                        RF_ASSERT(queue->queueLength >= 0);
1.52      oster     336:                        if (RF_OK_TO_DISPATCH(queue, req)) {
                    337:                                Dprintf2("DiskIOComplete: dispatching pri %d regular req to c %d (ok to dispatch)\n", req->priority, queue->col);
                    338:                                rf_DispatchKernelIO(queue, req);
                    339:                        } else {        
                    340:                                /* we can't dispatch it, so just re-enqueue it.  
                    341:                                   potential trouble here if disk queues batch reqs */
                    342:                                Dprintf2("DiskIOComplete: re-enqueueing pri %d regular req to c %d\n", req->priority, queue->col);
                    343:                                queue->queueLength++;
                    344:                                (queue->qPtr->Enqueue) (queue->qHdr, req, req->priority);
                    345:                                done = 1;
                    346:                        }
                    347:                } else {        
                    348:                        Dprintf1("DiskIOComplete: no more requests to extract.\n", "");
                    349:                        done = 1;
1.6       oster     350:                }
                    351:        }
                    352: 
                    353:        RF_UNLOCK_QUEUE_MUTEX(queue, "DiskIOComplete");
1.1       oster     354: }
                    355: /* promotes accesses tagged with the given parityStripeID from low priority
                    356:  * to normal priority.  This promotion is optional, meaning that a queue
                    357:  * need not implement it.  If there is no promotion routine associated with
                    358:  * a queue, this routine does nothing and returns -1.
                    359:  */
1.40      perry     360: int
1.27      oster     361: rf_DiskIOPromote(RF_DiskQueue_t *queue, RF_StripeNum_t parityStripeID,
                    362:                 RF_ReconUnitNum_t which_ru)
1.6       oster     363: {
                    364:        int     retval;
                    365: 
                    366:        if (!queue->qPtr->Promote)
                    367:                return (-1);
                    368:        RF_LOCK_QUEUE_MUTEX(queue, "DiskIOPromote");
                    369:        retval = (queue->qPtr->Promote) (queue->qHdr, parityStripeID, which_ru);
                    370:        RF_UNLOCK_QUEUE_MUTEX(queue, "DiskIOPromote");
                    371:        return (retval);
                    372: }
                    373: 
                    374: RF_DiskQueueData_t *
1.27      oster     375: rf_CreateDiskQueueData(RF_IoType_t typ, RF_SectorNum_t ssect,
1.49      christos  376:                       RF_SectorCount_t nsect, void *bf,
1.27      oster     377:                       RF_StripeNum_t parityStripeID,
                    378:                       RF_ReconUnitNum_t which_ru,
1.56      christos  379:                       void (*wakeF) (void *, int), void *arg,
1.37      oster     380:                       RF_AccTraceEntry_t *tracerec, RF_Raid_t *raidPtr,
1.57      jdolecek  381:                       RF_DiskQueueDataFlags_t flags, const struct buf *mbp,
1.38      oster     382:                       int waitflag)
1.6       oster     383: {
                    384:        RF_DiskQueueData_t *p;
                    385: 
1.60    ! oster     386:        p = pool_get(&raidPtr->pools.dqd, PR_WAITOK | PR_ZERO);
1.59      oster     387:        KASSERT(p != NULL);
1.38      oster     388: 
1.59      oster     389:        /* Obtain a buffer from our own pool.  It is possible for the
                    390:           regular getiobuf() to run out of memory and return NULL.
                    391:           We need to guarantee that never happens, as RAIDframe
                    392:           doesn't have a good way to recover if memory allocation
                    393:           fails here.
                    394:        */
1.60    ! oster     395:        p->bp = pool_get(&raidPtr->pools.bufio, PR_WAITOK | PR_ZERO);
1.59      oster     396:        KASSERT(p->bp != NULL);
                    397:        
                    398:        buf_init(p->bp);
                    399:                
1.51      reinoud   400:        SET(p->bp->b_cflags, BC_BUSY);  /* mark buffer busy */
1.57      jdolecek  401:        if (mbp) {
                    402:                SET(p->bp->b_flags, mbp->b_flags & rf_b_pass);
                    403:                p->bp->b_proc = mbp->b_proc;
                    404:        }
1.6       oster     405: 
                    406:        p->sectorOffset = ssect + rf_protectedSectors;
                    407:        p->numSector = nsect;
                    408:        p->type = typ;
1.41      christos  409:        p->buf = bf;
1.6       oster     410:        p->parityStripeID = parityStripeID;
                    411:        p->which_ru = which_ru;
                    412:        p->CompleteFunc = wakeF;
                    413:        p->argument = arg;
1.39      oster     414:        p->next = NULL;
1.6       oster     415:        p->tracerec = tracerec;
                    416:        p->priority = RF_IO_NORMAL_PRIORITY;
                    417:        p->raidPtr = raidPtr;
                    418:        p->flags = flags;
                    419:        return (p);
                    420: }
                    421: 
1.40      perry     422: void
1.27      oster     423: rf_FreeDiskQueueData(RF_DiskQueueData_t *p)
1.1       oster     424: {
1.60    ! oster     425:        pool_put(&p->raidPtr->pools.bufio, p->bp);
        !           426:        pool_put(&p->raidPtr->pools.dqd, p);
1.1       oster     427: }

CVSweb <webmaster@jp.NetBSD.org>