LASSERT(rs != NULL);
LASSERT(rs->rs_nlocks < RS_MAX_LOCKS);
- if (req->rq_export->exp_disconnected) {
- ldlm_lock_decref(lock, mode);
- } else {
- idx = rs->rs_nlocks++;
- rs->rs_locks[idx] = *lock;
- rs->rs_modes[idx] = mode;
- rs->rs_difficult = 1;
- rs->rs_no_ack = no_ack;
- rs->rs_convert_lock = convert_lock;
- }
+ idx = rs->rs_nlocks++;
+ rs->rs_locks[idx] = *lock;
+ rs->rs_modes[idx] = mode;
+ rs->rs_difficult = 1;
+ rs->rs_no_ack = no_ack;
+ rs->rs_convert_lock = convert_lock;
}
EXPORT_SYMBOL(ptlrpc_save_lock);
}
weight = cfs_cpt_weight(svc->srv_cptable, 0);
- LASSERT(weight > 0);
for (; factor > 0 && weight > 0; factor--, weight -= fade)
nthrs += min(weight, fade) * factor;
*/
struct ptlrpc_service *
ptlrpc_register_service(struct ptlrpc_service_conf *conf,
+ struct kset *parent,
struct proc_dir_entry *proc_entry)
{
struct ptlrpc_service_cpt_conf *cconf = &conf->psc_cpt;
list_add(&service->srv_list, &ptlrpc_all_services);
mutex_unlock(&ptlrpc_all_services_mutex);
+ if (parent) {
+ rc = ptlrpc_sysfs_register_service(parent, service);
+ if (rc)
+ GOTO(failed, rc);
+ }
+
if (proc_entry != NULL)
ptlrpc_lprocfs_register_service(proc_entry, service);
spin_lock(&svcpt->scp_lock);
/*
* now all reqs including the embedded req has been
- * disposed, schedule request buffer for re-use.
+ * disposed, schedule request buffer for re-use
+ * or free it to drain some in excess.
*/
LASSERT(atomic_read(&rqbd->rqbd_req.rq_refcount) == 0);
- list_add_tail(&rqbd->rqbd_list, &svcpt->scp_rqbd_idle);
+ if (svcpt->scp_nrqbds_posted >=
+ svc->srv_nbuf_per_group &&
+ !test_req_buffer_pressure) {
+ /* like in ptlrpc_free_rqbd() */
+ svcpt->scp_nrqbds_total--;
+ OBD_FREE_LARGE(rqbd->rqbd_buffer,
+ svc->srv_buf_size);
+ OBD_FREE_PTR(rqbd);
+ } else {
+ list_add_tail(&rqbd->rqbd_list,
+ &svcpt->scp_rqbd_idle);
+ }
}
spin_unlock(&svcpt->scp_lock);
* we better wait for 3. */
exp->exp_obd->obd_eviction_timer =
cfs_time_current_sec() + 3 * PING_INTERVAL;
- CDEBUG(D_HA, "%s: Think about evicting %s from "CFS_TIME_T"\n",
+ CDEBUG(D_HA, "%s: Think about evicting %s from %ld\n",
exp->exp_obd->obd_name,
obd_export_nid2str(oldest_exp), oldest_time);
}
}
/* Set timer for closest deadline */
- next = (__s32)(array->paa_deadline - cfs_time_current_sec() -
+ next = (__s32)(array->paa_deadline - ktime_get_real_seconds() -
at_early_margin);
if (next <= 0) {
ptlrpc_at_timer((unsigned long)svcpt);
spin_lock(&svcpt->scp_at_lock);
LASSERT(list_empty(&req->rq_timed_list));
- index = (unsigned long)req->rq_deadline % array->paa_size;
+ div_u64_rem(req->rq_deadline, array->paa_size, &index);
if (array->paa_reqs_count[index] > 0) {
/* latest rpcs will have the latest deadlines in the list,
* so search backward. */
struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
struct ptlrpc_request *reqcopy;
struct lustre_msg *reqmsg;
- cfs_duration_t olddl = req->rq_deadline - cfs_time_current_sec();
- time_t newdl;
+ time64_t olddl = req->rq_deadline - ktime_get_real_seconds();
+ time64_t newdl;
int rc;
ENTRY;
/* deadline is when the client expects us to reply, margin is the
difference between clients' and servers' expectations */
DEBUG_REQ(D_ADAPTTO, req,
- "%ssending early reply (deadline %+lds, margin %+lds) for "
+ "%ssending early reply (deadline %+llds, margin %+llds) for "
"%d+%d", AT_OFF ? "AT off - not " : "",
- olddl, olddl - at_get(&svcpt->scp_at_estimate),
+ (s64)olddl, (s64)(olddl - at_get(&svcpt->scp_at_estimate)),
at_get(&svcpt->scp_at_estimate), at_extra);
if (AT_OFF)
RETURN(0);
if (olddl < 0) {
- DEBUG_REQ(D_WARNING, req, "Already past deadline (%+lds), "
+ DEBUG_REQ(D_WARNING, req, "Already past deadline (%+llds), "
"not sending early reply. Consider increasing "
- "at_early_margin (%d)?", olddl, at_early_margin);
+ "at_early_margin (%d)?", (s64)olddl, at_early_margin);
/* Return an error so we're not re-added to the timed list. */
RETURN(-ETIMEDOUT);
* account for network latency). See ptlrpc_at_recv_early_reply
*/
at_measured(&svcpt->scp_at_estimate, at_extra +
- cfs_time_current_sec() -
+ ktime_get_real_seconds() -
req->rq_arrival_time.tv_sec);
newdl = req->rq_arrival_time.tv_sec +
at_get(&svcpt->scp_at_estimate);
/* Check to see if we've actually increased the deadline -
* we may be past adaptive_max */
if (req->rq_deadline >= newdl) {
- DEBUG_REQ(D_WARNING, req, "Couldn't add any time "
- "(%ld/%ld), not sending early reply\n",
- olddl, newdl - cfs_time_current_sec());
+ DEBUG_REQ(D_WARNING, req, "Couldn't add any time (%lld/%lld), not sending early reply\n",
+ (s64)olddl, (s64)(newdl - ktime_get_real_seconds()));
RETURN(-ETIMEDOUT);
}
struct ptlrpc_request *rq, *n;
struct list_head work_list;
__u32 index, count;
- time_t deadline;
- time_t now = cfs_time_current_sec();
+ time64_t deadline;
+ time64_t now = ktime_get_real_seconds();
cfs_duration_t delay;
int first, counter = 0;
ENTRY;
server will take. Send early replies to everyone expiring soon. */
INIT_LIST_HEAD(&work_list);
deadline = -1;
- index = (unsigned long)array->paa_deadline % array->paa_size;
+ div_u64_rem(array->paa_deadline, array->paa_size, &index);
count = array->paa_count;
while (count > 0) {
count -= array->paa_reqs_count[index];
LCONSOLE_WARN("%s: This server is not able to keep up with "
"request traffic (cpu-bound).\n",
svcpt->scp_service->srv_name);
- CWARN("earlyQ=%d reqQ=%d recA=%d, svcEst=%d, "
- "delay="CFS_DURATION_T"(jiff)\n",
+ CWARN("earlyQ=%d reqQ=%d recA=%d, svcEst=%d, delay=%ld(jiff)\n",
counter, svcpt->scp_nreqs_incoming,
svcpt->scp_nreqs_active,
at_get(&svcpt->scp_at_estimate), delay);
}
/* req_in handling should/must be fast */
- if (cfs_time_current_sec() - req->rq_arrival_time.tv_sec > 5)
- DEBUG_REQ(D_WARNING, req, "Slow req_in handling "CFS_DURATION_T"s",
- cfs_time_sub(cfs_time_current_sec(),
- req->rq_arrival_time.tv_sec));
+ if (ktime_get_real_seconds() - req->rq_arrival_time.tv_sec > 5)
+ DEBUG_REQ(D_WARNING, req, "Slow req_in handling %llds",
+ (s64)(ktime_get_real_seconds() -
+ req->rq_arrival_time.tv_sec));
/* Set rpc server deadline and add it to the timed list */
deadline = (lustre_msghdr_get_flags(req->rq_reqmsg) &
ptlrpc_server_handle_request(struct ptlrpc_service_part *svcpt,
struct ptlrpc_thread *thread)
{
- struct ptlrpc_service *svc = svcpt->scp_service;
- struct ptlrpc_request *request;
- struct timeval work_start;
- struct timeval work_end;
- long timediff;
- int fail_opc = 0;
+ struct ptlrpc_service *svc = svcpt->scp_service;
+ struct ptlrpc_request *request;
+ ktime_t work_start;
+ ktime_t work_end;
+ ktime_t arrived;
+ s64 timediff_usecs;
+ s64 arrived_usecs;
+ int fail_opc = 0;
ENTRY;
if(OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_DUMP_LOG))
libcfs_debug_dumplog();
- do_gettimeofday(&work_start);
- timediff = cfs_timeval_sub(&work_start, &request->rq_arrival_time,NULL);
+ work_start = ktime_get_real();
+ arrived = timespec64_to_ktime(request->rq_arrival_time);
+ timediff_usecs = ktime_us_delta(work_start, arrived);
if (likely(svc->srv_stats != NULL)) {
lprocfs_counter_add(svc->srv_stats, PTLRPC_REQWAIT_CNTR,
- timediff);
+ timediff_usecs);
lprocfs_counter_add(svc->srv_stats, PTLRPC_REQQDEPTH_CNTR,
svcpt->scp_nreqs_incoming);
lprocfs_counter_add(svc->srv_stats, PTLRPC_REQACTIVE_CNTR,
if (likely(request->rq_export)) {
if (unlikely(ptlrpc_check_req(request)))
goto put_conn;
- ptlrpc_update_export_timer(request->rq_export, timediff >> 19);
+ ptlrpc_update_export_timer(request->rq_export,
+ timediff_usecs >> 19);
}
/* Discard requests queued for longer than the deadline.
The deadline is increased if we send an early reply. */
- if (cfs_time_current_sec() > request->rq_deadline) {
- DEBUG_REQ(D_ERROR, request, "Dropping timed-out request from %s"
- ": deadline "CFS_DURATION_T":"CFS_DURATION_T"s ago\n",
+ if (ktime_get_real_seconds() > request->rq_deadline) {
+ DEBUG_REQ(D_ERROR, request, "Dropping timed-out request from %s: deadline %lld:%llds ago\n",
libcfs_id2str(request->rq_peer),
- cfs_time_sub(request->rq_deadline,
- request->rq_arrival_time.tv_sec),
- cfs_time_sub(cfs_time_current_sec(),
- request->rq_deadline));
+ request->rq_deadline -
+ request->rq_arrival_time.tv_sec,
+ ktime_get_real_seconds() - request->rq_deadline);
goto put_conn;
}
ptlrpc_rqphase_move(request, RQ_PHASE_COMPLETE);
put_conn:
- if (unlikely(cfs_time_current_sec() > request->rq_deadline)) {
- DEBUG_REQ(D_WARNING, request, "Request took longer "
- "than estimated ("CFS_DURATION_T":"CFS_DURATION_T"s);"
- " client may timeout.",
- cfs_time_sub(request->rq_deadline,
- request->rq_arrival_time.tv_sec),
- cfs_time_sub(cfs_time_current_sec(),
- request->rq_deadline));
- }
-
- do_gettimeofday(&work_end);
- timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
- CDEBUG(D_RPCTRACE, "Handled RPC pname:cluuid+ref:pid:xid:nid:opc "
- "%s:%s+%d:%d:x%llu:%s:%d Request procesed in "
- "%ldus (%ldus total) trans %llu rc %d/%d\n",
+ if (unlikely(ktime_get_real_seconds() > request->rq_deadline)) {
+ DEBUG_REQ(D_WARNING, request, "Request took longer than estimated (%lld:%llds); client may timeout.",
+ request->rq_deadline -
+ request->rq_arrival_time.tv_sec,
+ ktime_get_real_seconds() - request->rq_deadline);
+ }
+
+ work_end = ktime_get_real();
+ timediff_usecs = ktime_us_delta(work_end, work_start);
+ arrived_usecs = ktime_us_delta(work_end, arrived);
+ CDEBUG(D_RPCTRACE, "Handled RPC pname:cluuid+ref:pid:xid:nid:opc %s:%s+%d:%d:x%llu:%s:%d Request procesed in %lldus (%lldus total) trans %llu rc %d/%d\n",
current_comm(),
(request->rq_export ?
(char *)request->rq_export->exp_client_uuid.uuid : "0"),
request->rq_xid,
libcfs_id2str(request->rq_peer),
lustre_msg_get_opc(request->rq_reqmsg),
- timediff,
- cfs_timeval_sub(&work_end, &request->rq_arrival_time, NULL),
+ timediff_usecs,
+ arrived_usecs,
(request->rq_repmsg ?
lustre_msg_get_transno(request->rq_repmsg) :
request->rq_transno),
LASSERT(opc < LUSTRE_MAX_OPCODES);
lprocfs_counter_add(svc->srv_stats,
opc + EXTRA_MAX_OPCODES,
- timediff);
+ timediff_usecs);
}
}
if (unlikely(request->rq_early_count)) {
DEBUG_REQ(D_ADAPTTO, request,
- "sent %d early replies before finishing in "
- CFS_DURATION_T"s",
+ "sent %d early replies before finishing in %llds",
request->rq_early_count,
- cfs_time_sub(work_end.tv_sec,
- request->rq_arrival_time.tv_sec));
+ arrived_usecs / USEC_PER_SEC);
}
ptlrpc_server_finish_active_request(svcpt, request);
if (rs->rs_convert_lock &&
rs->rs_transno > exp->exp_last_committed) {
struct ldlm_lock *lock;
+ struct ldlm_lock *ack_locks[RS_MAX_LOCKS] = { NULL };
spin_lock(&rs->rs_lock);
if (rs->rs_convert_lock &&
rs->rs_transno > exp->exp_last_committed) {
nlocks = rs->rs_nlocks;
- while (nlocks-- > 0)
+ while (nlocks-- > 0) {
+ /*
+ * NB don't assume rs is always handled
+ * by the same service thread (see
+ * ptlrpc_hr_select, so REP-ACK hr may
+ * race with trans commit, while the
+ * latter will release locks, get locks
+ * here early to convert to COS mode
+ * safely.
+ */
+ lock = ldlm_handle2lock(
+ &rs->rs_locks[nlocks]);
+ LASSERT(lock);
+ ack_locks[nlocks] = lock;
rs->rs_modes[nlocks] = LCK_COS;
+ }
nlocks = rs->rs_nlocks;
rs->rs_convert_lock = 0;
/* clear rs_scheduled so that commit callback
spin_unlock(&rs->rs_lock);
while (nlocks-- > 0) {
- lock = ldlm_handle2lock(
- &rs->rs_locks[nlocks]);
- LASSERT(lock != NULL);
+ lock = ack_locks[nlocks];
ldlm_lock_downgrade(lock, LCK_COS);
LDLM_LOCK_PUT(lock);
}
struct ptlrpc_hr_partition *hrp;
struct ptlrpc_hr_thread *hrt;
int rc;
+ int cpt;
int i;
- int j;
int weight;
ENTRY;
weight = cpumask_weight(topology_sibling_cpumask(smp_processor_id()));
- cfs_percpt_for_each(hrp, i, ptlrpc_hr.hr_partitions) {
- hrp->hrp_cpt = i;
+ cfs_percpt_for_each(hrp, cpt, ptlrpc_hr.hr_partitions) {
+ hrp->hrp_cpt = cpt;
atomic_set(&hrp->hrp_nstarted, 0);
atomic_set(&hrp->hrp_nstopped, 0);
- hrp->hrp_nthrs = cfs_cpt_weight(ptlrpc_hr.hr_cpt_table, i);
-
+ hrp->hrp_nthrs = cfs_cpt_weight(ptlrpc_hr.hr_cpt_table, cpt);
hrp->hrp_nthrs /= weight;
if (hrp->hrp_nthrs == 0)
hrp->hrp_nthrs = 1;
- OBD_CPT_ALLOC(hrp->hrp_thrs, ptlrpc_hr.hr_cpt_table, i,
+ OBD_CPT_ALLOC(hrp->hrp_thrs, ptlrpc_hr.hr_cpt_table, cpt,
hrp->hrp_nthrs * sizeof(*hrt));
if (hrp->hrp_thrs == NULL)
GOTO(out, rc = -ENOMEM);
- for (j = 0; j < hrp->hrp_nthrs; j++) {
- hrt = &hrp->hrp_thrs[j];
+ for (i = 0; i < hrp->hrp_nthrs; i++) {
+ hrt = &hrp->hrp_thrs[i];
- hrt->hrt_id = j;
+ hrt->hrt_id = i;
hrt->hrt_partition = hrp;
init_waitqueue_head(&hrt->hrt_waitq);
spin_lock_init(&hrt->hrt_lock);
void ptlrpc_hr_fini(void)
{
struct ptlrpc_hr_partition *hrp;
- int i;
+ int cpt;
if (ptlrpc_hr.hr_partitions == NULL)
return;
ptlrpc_stop_hr_threads();
- cfs_percpt_for_each(hrp, i, ptlrpc_hr.hr_partitions) {
+ cfs_percpt_for_each(hrp, cpt, ptlrpc_hr.hr_partitions) {
if (hrp->hrp_thrs != NULL) {
OBD_FREE(hrp->hrp_thrs,
hrp->hrp_nthrs * sizeof(hrp->hrp_thrs[0]));
ptlrpc_service_nrs_cleanup(service);
ptlrpc_lprocfs_unregister_service(service);
+ ptlrpc_sysfs_unregister_service(service);
ptlrpc_service_free(service);
* to be shot, so it's intentionally non-aggressive. */
static int ptlrpc_svcpt_health_check(struct ptlrpc_service_part *svcpt)
{
- struct ptlrpc_request *request = NULL;
- struct timeval right_now;
- long timediff;
+ struct ptlrpc_request *request = NULL;
+ struct timespec64 right_now;
+ struct timespec64 timediff;
- do_gettimeofday(&right_now);
+ ktime_get_real_ts64(&right_now);
spin_lock(&svcpt->scp_req_lock);
/* How long has the next entry been waiting? */
return 0;
}
- timediff = cfs_timeval_sub(&right_now, &request->rq_arrival_time, NULL);
+ timediff = timespec64_sub(right_now, request->rq_arrival_time);
spin_unlock(&svcpt->scp_req_lock);
- if ((timediff / ONE_MILLION) >
+ if ((timediff.tv_sec) >
(AT_OFF ? obd_timeout * 3 / 2 : at_max)) {
- CERROR("%s: unhealthy - request has been waiting %lds\n",
- svcpt->scp_service->srv_name, timediff / ONE_MILLION);
+ CERROR("%s: unhealthy - request has been waiting %llds\n",
+ svcpt->scp_service->srv_name, (s64)timediff.tv_sec);
return -1;
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff