static void cachedev_idlework_callback (void *idleworkparam, int idledevno)
{
struct cache_dev *cache = (struct cache_dev *)idleworkparam;
struct cache_dev_event *flushdesc;
ioreq_event *flushreq;
int blkno, bcount;
struct ioq *queue;
ASSERT (idledevno == cache->real_devno);
#ifdef DEBUG_CACHEDEV
fprintf(outputfile, "*** %f: Entered cachedev::cachedev_idlework_callback\n", simtime );
fflush(outputfile);
#endif
queue = (*cache->queuefind)(cache->queuefindparam, cache->real_devno);
if (ioqueue_get_number_in_queue (queue) != 0) {
return;
}
queue = (*cache->queuefind)(cache->queuefindparam, cache->cache_devno);
if (ioqueue_get_number_in_queue (queue) != 0) {
return;
}
if (cachedev_find_dirty_cache_blocks (cache, &blkno, &bcount) == 0) {
return;
}
/* Just assume that bufferspace is available */
cache->bufferspace += bcount;
if (cache->bufferspace > cache->stat.maxbufferspace) {
cache->stat.maxbufferspace = cache->bufferspace;
}
flushdesc = (struct cache_dev_event *) getfromextraq();
flushdesc->type = CACHE_EVENT_IDLEFLUSH_READ;
flushreq = (ioreq_event *) getfromextraq();
flushreq->buf = flushdesc;
flushreq->devno = cache->cache_devno;
flushreq->blkno = blkno;
flushreq->bcount = bcount;
flushreq->type = IO_ACCESS_ARRIVE;
flushreq->flags = READ;
(*cache->issuefunc)(cache->issueparam, flushreq);
cache->stat.destagereads++;
cache->stat.destagereadblocks += bcount;
}
static int logorg_shadowed_get_short_queue (logorg *currlogorg, ioreq_event *curr, int def)
{
int i, j;
int len;
int shortdev = -1;
int shortlen = 999999999;
int ties[MAXCOPIES];
j = -1;
for (i = 0; i < currlogorg->copies; i++) {
len = ioqueue_get_number_in_queue(currlogorg->devs[(curr->devno + (i * currlogorg->numdisks))].queue);
if (len >= 999999999) {
fprintf(stderr, "Haven't allowed for large enough 'dist's in logorg_shadowed_get_short_queue - %d\n", len);
exit(1);
}
if (len == shortlen) {
j++;
if (j >= MAXCOPIES) {
fprintf(stderr, "Haven't allowed for enough ties in logorg_shadowed_get_short_queue - %d\n", j);
exit(1);
}
ties[j] = i;
}
if (len < shortlen) {
shortlen = len;
shortdev = i;
j = -1;
}
}
if (shortdev == -1) {
fprintf(stderr, "Illegal condition in logorg_shadowed_get_short_queue\n");
exit(1);
} else if (j != -1) {
if (def == 1) {
i = (int) (DISKSIM_drand48() * (double) (j+2));
if (i != 0) {
shortdev = ties[(i-1)];
}
} else if (def == 2) {
j++;
ties[j] = shortdev;
shortdev = logorg_shadowed_get_short_dist(currlogorg, curr, (j+1), ties);
} else {
fprintf(stderr, "Unknown default tie breaker in logorg_shadowed_get_short_queue - %d\n", def);
exit(1);
}
}
return(shortdev);
}
event * iodriver_request (int iodriverno, ioreq_event *curr)
{
ioreq_event *temp = NULL;
ioreq_event *ret = NULL;
ioreq_event *retlist = NULL;
int numreqs;
/*
printf ("Entered iodriver_request - simtime %f, devno %d, blkno %lld, cause %d\n", simtime, curr->devno, curr->blkno, curr->cause);
fprintf (outputfile, "Entered iodriver_request - simtime %f, devno %d, blkno %lld, cause %d\n", simtime, curr->devno, curr->blkno, curr->cause);
fprintf (stderr, "Entered iodriver_request - simtime %f, devno %d, blkno %lld, cause %d\n", simtime, curr->devno, curr->blkno, curr->cause);
*/
if (NULL != OUTIOS) {
fprintf(OUTIOS, "%.6f,%d,%lld,%d,%x,%d,%p\n", simtime, curr->devno, curr->blkno, curr->bcount, curr->flags, OVERALLQUEUE->base.listlen + 1, curr );
fflush( OUTIOS );
}
#if 0
fprintf (stderr, "Entered iodriver_request - simtime %f, devno %d, blkno %lld, cause %d\n", simtime, curr->devno, curr->blkno, curr->cause);
#endif
/* add to the overall queue to start tracking */
ret = ioreq_copy (curr);
ioqueue_add_new_request (OVERALLQUEUE, ret);
ret = NULL;
disksim->totalreqs++;
if ((disksim->checkpoint_iocnt > 0) && ((disksim->totalreqs % disksim->checkpoint_iocnt) == 0)) {
disksim_register_checkpoint (simtime);
}
if (disksim->totalreqs == disksim->warmup_iocnt) {
warmuptime = simtime;
resetstats();
}
numreqs = logorg_maprequest(sysorgs, numsysorgs, curr);
temp = curr->next;
for (; numreqs>0; numreqs--) {
/* Request list size must match numreqs */
ASSERT(curr != NULL);
curr->next = NULL;
if ((iodrivers[iodriverno]->consttime == IODRIVER_TRACED_QUEUE_TIMES) || (iodrivers[iodriverno]->consttime == IODRIVER_TRACED_BOTH_TIMES)) {
ret = ioreq_copy(curr);
ret->time = simtime + (double) ret->tempint1 / (double) 1000;
ret->type = IO_TRACE_REQUEST_START;
addtointq((event *) ret);
ret = NULL;
if ((curr->slotno == 1) && (ioqueue_get_number_in_queue(iodrivers[iodriverno]->devices[(curr->devno)].queue) == 0)) {
iodrivers[(iodriverno)]->devices[(curr->devno)].flag = 2;
iodrivers[(iodriverno)]->devices[(curr->devno)].lastevent = simtime;
}
}
ret = handle_new_request(iodrivers[iodriverno], curr);
if ((ret) && (iodrivers[iodriverno]->type == STANDALONE) && (ret->time == 0.0)) {
ret->type = IO_ACCESS_ARRIVE;
ret->time = simtime;
iodriver_schedule(iodriverno, ret);
} else if (ret) {
ret->type = IO_ACCESS_ARRIVE;
ret->next = retlist;
ret->prev = NULL;
retlist = ret;
}
curr = temp;
temp = (temp) ? temp->next : NULL;
}
if (iodrivers[iodriverno]->type == STANDALONE) {
while (retlist) {
ret = retlist;
retlist = ret->next;
ret->next = NULL;
ret->time += simtime;
addtointq((event *) ret);
}
}
/*
fprintf (outputfile, "leaving iodriver_request: retlist %p\n", retlist);
*/
return((event *) retlist);
}
void iodriver_access_complete (int iodriverno, intr_event *intrp)
{
int i;
int numreqs;
ioreq_event *tmp;
ioreq_event *del;
ioreq_event *req;
int devno;
int skip = 0;
ctlr *ctl = NULL;
time_t now;
if (iodrivers[iodriverno]->type == STANDALONE) {
req = ioreq_copy((ioreq_event *) intrp->infoptr);
}
else {
req = (ioreq_event *) intrp->infoptr;
}
#ifdef DEBUG_IODRIVER
fprintf (outputfile, "*** %f: iodriver_access_complete - devno %d, blkno %d, bcount %d, read %d\n", simtime, req->devno, req->blkno, req->bcount, (req->flags & READ));
fflush(outputfile);
#endif
time( & now );
disksim_exectrace( "Request completion: simtime %f, devno %d, blkno %lld, bcount %d, flags %X, time %s\n", simtime, req->devno, req->blkno, req->bcount, req->flags, asctime( localtime(& now)) );
if (iodrivers[iodriverno]->devices[(req->devno)].queuectlr != -1)
{
int ctlrno = iodrivers[iodriverno]->devices[(req->devno)].queuectlr;
ctl = &iodrivers[iodriverno]->ctlrs[ctlrno];
tmp = ctl->oversized;
numreqs = 1;
while (((numreqs) || (tmp != ctl->oversized))
&& (tmp)
&& (tmp->next)
&& ((tmp->next->devno != req->devno)
|| (tmp->next->opid != req->opid)
|| (req->blkno < tmp->next->blkno)
|| (req->blkno >= (tmp->next->blkno + tmp->next->bcount))))
{
// fprintf (outputfile, "oversized request in list: opid %d, blkno %lld, bcount %d\n", tmp->opid, tmp->blkno, tmp->bcount);
numreqs = 0;
tmp = tmp->next;
}
if ((tmp)
&& (tmp->next->devno == req->devno)
&& (tmp->next->opid == req->opid)
&& (req->blkno >= tmp->next->blkno)
&& (req->blkno < (tmp->next->blkno + tmp->next->bcount)))
{
fprintf (outputfile, "%f, part of oversized request completed: opid %d, blkno %lld, bcount %d, maxreqsize %d\n", simtime, req->opid, req->blkno, req->bcount, ctl->maxreqsize);
if ((req->blkno + ctl->maxreqsize) < (tmp->next->blkno + tmp->next->bcount))
{
fprintf (outputfile, "more to go\n");
req->blkno += ctl->maxreqsize;
req->bcount = min(ctl->maxreqsize,
(tmp->next->blkno + tmp->next->bcount - req->blkno));
goto schedule_next;
}
else {
fprintf (outputfile, "done for real\n");
addtoextraq((event *) req);
req = tmp->next;
tmp->next = tmp->next->next;
if (ctl->oversized == req) {
ctl->oversized = (req != req->next) ? req->next : NULL;
}
req->next = NULL;
}
}
}
devno = req->devno;
req = ioqueue_physical_access_done(iodrivers[iodriverno]->devices[devno].queue, req);
if (ctl) {
ctl->numoutstanding--;
}
// special case for validate:
if (disksim->traceformat == VALIDATE) {
tmp = (ioreq_event *) getfromextraq();
io_validate_do_stats1();
tmp = iotrace_validate_get_ioreq_event(disksim->iotracefile, tmp);
if (tmp) {
io_validate_do_stats2(tmp);
tmp->type = IO_REQUEST_ARRIVE;
addtointq((event *) tmp);
disksim_exectrace("Request issue: simtime %f, devno %d, blkno %lld, time %f\n",
simtime, tmp->devno, tmp->blkno, tmp->time);
}
else {
disksim_simstop();
}
}
else if (disksim->closedios) {
tmp = (ioreq_event *) io_get_next_external_event(disksim->iotracefile);
if (tmp) {
io_using_external_event ((event *)tmp);
tmp->time = simtime + disksim->closedthinktime;
tmp->type = IO_REQUEST_ARRIVE;
addtointq((event *) tmp);
} else {
disksim_simstop();
}
}
while (req) {
tmp = req;
req = req->next;
tmp->next = NULL;
update_iodriver_statistics();
if ((numreqs = logorg_mapcomplete(sysorgs, numsysorgs, tmp)) == COMPLETE) {
/* update up overall I/O system stats for this completed request */
ioreq_event *temp = ioqueue_get_specific_request (OVERALLQUEUE, tmp);
ioreq_event *temp2 = ioqueue_physical_access_done (OVERALLQUEUE, temp);
ASSERT (temp2 != NULL);
addtoextraq((event *)temp);
temp = NULL;
if (iodrivers[iodriverno]->type != STANDALONE) {
iodriver_add_to_intrp_eventlist(intrp,
io_done_notify(tmp),
iodrivers[iodriverno]->scale);
} else {
io_done_notify (tmp);
}
}
else if (numreqs > 0) {
for (i = 0; i < numreqs; i++) {
del = tmp->next;
tmp->next = del->next;
del->next = NULL;
del->type = IO_REQUEST_ARRIVE;
del->flags |= MAPPED;
skip |= (del->devno == devno);
if (iodrivers[iodriverno]->type == STANDALONE)
{
del->time += simtime + 0.0000000001; /* to affect an ordering */
addtointq((event *) del);
}
else {
iodriver_add_to_intrp_eventlist(intrp,
(event *) del,
iodrivers[iodriverno]->scale);
}
}
}
addtoextraq((event *) tmp);
}
if ((iodrivers[iodriverno]->consttime == IODRIVER_TRACED_QUEUE_TIMES)
|| (iodrivers[iodriverno]->consttime == IODRIVER_TRACED_BOTH_TIMES))
{
if (ioqueue_get_number_in_queue(iodrivers[iodriverno]->devices[devno].queue) > 0)
{
iodrivers[iodriverno]->devices[devno].flag = 1;
iodrivers[iodriverno]->devices[devno].lastevent = simtime;
}
return;
}
if (skip) {
return;
}
// fprintf(outputfile, "iodriver_access_complete:: calling ioqueue_get_next_request\n");
req = ioqueue_get_next_request(iodrivers[iodriverno]->devices[devno].queue);
// fprintf (outputfile, "next scheduled: req %p, req->blkno %d, req->flags %x\n", req, ((req) ? req->blkno : 0), ((req) ? req->flags : 0));
schedule_next:
if (req) {
req->type = IO_ACCESS_ARRIVE;
req->next = NULL;
if (ctl) {
ctl->numoutstanding++;
}
if (iodrivers[iodriverno]->type == STANDALONE) {
req->time = simtime;
addtointq((event *) req);
}
else {
iodriver_add_to_intrp_eventlist(intrp,
(event *) req,
iodrivers[iodriverno]->scale);
}
}
}
static void ssd_activate_elem(ssd_t *currdisk, int elem_num)
{
ioreq_event *req;
ssd_req **read_reqs;
ssd_req **write_reqs;
int i;
int read_total = 0;
int write_total = 0;
double schtime = 0;
int max_reqs;
int tot_reqs_issued;
double max_time_taken = 0;
ssd_element *elem = &currdisk->elements[elem_num];
// if the media is busy, we can't do anything, so return
if (elem->media_busy == TRUE) {
return;
}
ASSERT(ioqueue_get_reqoutstanding(elem->queue) == 0);
// we can invoke cleaning in the background whether there
// is request waiting or not
if (currdisk->params.cleaning_in_background) {
// if cleaning was invoked, wait until
// it is over ...
if (ssd_invoke_element_cleaning(elem_num, currdisk)) {
return;
}
}
ASSERT(elem->metadata.reqs_waiting == ioqueue_get_number_in_queue(elem->queue));
if (elem->metadata.reqs_waiting > 0) {
// invoke cleaning in foreground when there are requests waiting
if (!currdisk->params.cleaning_in_background) {
// if cleaning was invoked, wait until
// it is over ...
if (ssd_invoke_element_cleaning(elem_num, currdisk)) {
return;
}
}
// how many reqs can we issue at once
if (currdisk->params.copy_back == SSD_COPY_BACK_DISABLE) {
max_reqs = 1;
} else {
if (currdisk->params.num_parunits == 1) {
max_reqs = 1;
} else {
max_reqs = MAX_REQS_ELEM_QUEUE;
}
}
// ideally, we should issue one req per plane, overlapping them all.
// in order to simplify the overlapping strategy, let's issue
// requests of the same type together.
read_reqs = (ssd_req **) malloc(max_reqs * sizeof(ssd_req *));
write_reqs = (ssd_req **) malloc(max_reqs * sizeof(ssd_req *));
// collect the requests
while ((req = ioqueue_get_next_request(elem->queue)) != NULL) {
int found = 0;
elem->metadata.reqs_waiting --;
// see if we already have the same request in the list.
// this usually doesn't happen -- but on synthetic traces
// this weird case can occur.
if (req->flags & READ) {
found = ssd_already_present(read_reqs, read_total, req);
} else {
found = ssd_already_present(write_reqs, write_total, req);
}
if (!found) {
// this is a valid request
ssd_req *r = malloc(sizeof(ssd_req));
r->blk = req->blkno;
r->count = req->bcount;
r->is_read = req->flags & READ;
r->org_req = req;
r->plane_num = -1; // we don't know to which plane this req will be directed at
if (req->flags & READ) {
read_reqs[read_total] = r;
read_total ++;
} else {
write_reqs[write_total] = r;
write_total ++;
}
// if we have more reqs than we can handle, quit
if ((read_total >= max_reqs) ||
(write_total >= max_reqs)) {
break;
}
} else {
// throw this request -- it doesn't make sense
stat_update (&currdisk->stat.acctimestats, 0);
req->time = simtime;
req->ssd_elem_num = elem_num;
req->type = DEVICE_ACCESS_COMPLETE;
addtointq ((event *)req);
}
}
if (read_total > 0) {
// first issue all the read requests (it doesn't matter what we
// issue first). i chose read because reads are mostly synchronous.
// find the time taken to serve these requests.
ssd_compute_access_time(currdisk, elem_num, read_reqs, read_total);
// add an event for each request completion
for (i = 0; i < read_total; i ++) {
elem->media_busy = TRUE;
// find the maximum time taken by a request
if (schtime < read_reqs[i]->schtime) {
schtime = read_reqs[i]->schtime;
}
stat_update (&currdisk->stat.acctimestats, read_reqs[i]->acctime);
read_reqs[i]->org_req->time = simtime + read_reqs[i]->schtime;
read_reqs[i]->org_req->ssd_elem_num = elem_num;
read_reqs[i]->org_req->type = DEVICE_ACCESS_COMPLETE;
//printf("R: blk %d elem %d acctime %f simtime %f\n", read_reqs[i]->blk,
// elem_num, read_reqs[i]->acctime, read_reqs[i]->org_req->time);
addtointq ((event *)read_reqs[i]->org_req);
free(read_reqs[i]);
}
}
free(read_reqs);
max_time_taken = schtime;
if (write_total > 0) {
// next issue the write requests
ssd_compute_access_time(currdisk, elem_num, write_reqs, write_total);
// add an event for each request completion.
// note that we can issue the writes only after all the reads above are
// over. so, include the maximum read time when creating the event.
for (i = 0; i < write_total; i ++) {
elem->media_busy = TRUE;
stat_update (&currdisk->stat.acctimestats, write_reqs[i]->acctime);
write_reqs[i]->org_req->time = simtime + schtime + write_reqs[i]->schtime;
//printf("blk %d elem %d acc time %f\n", write_reqs[i]->blk, elem_num, write_reqs[i]->acctime);
if (max_time_taken < (schtime+write_reqs[i]->schtime)) {
max_time_taken = (schtime+write_reqs[i]->schtime);
}
write_reqs[i]->org_req->ssd_elem_num = elem_num;
write_reqs[i]->org_req->type = DEVICE_ACCESS_COMPLETE;
//printf("W: blk %d elem %d acctime %f simtime %f\n", write_reqs[i]->blk,
// elem_num, write_reqs[i]->acctime, write_reqs[i]->org_req->time);
addtointq ((event *)write_reqs[i]->org_req);
free(write_reqs[i]);
}
}
free(write_reqs);
// statistics
tot_reqs_issued = read_total + write_total;
ASSERT(tot_reqs_issued > 0);
currdisk->elements[elem_num].stat.tot_reqs_issued += tot_reqs_issued;
currdisk->elements[elem_num].stat.tot_time_taken += max_time_taken;
}
}