*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, Whamcloud, Inc.
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_LNET
+#include <linux/kthread.h>
#include <libcfs/libcfs.h>
#define CFS_WS_NAME_LEN 16
-typedef struct cfs_wi_sched {
- cfs_list_t ws_list; /* chain on global list */
-#ifdef __KERNEL__
+struct cfs_wi_sched {
+ struct list_head ws_list; /* chain on global list */
/** serialised workitems */
- cfs_spinlock_t ws_lock;
+ spinlock_t ws_lock;
/** where schedulers sleep */
- cfs_waitq_t ws_waitq;
-#endif
+ wait_queue_head_t ws_waitq;
/** concurrent workitems */
- cfs_list_t ws_runq;
+ struct list_head ws_runq;
/** rescheduled running-workitems, a workitem can be rescheduled
* while running in wi_action(), but we don't to execute it again
* unless it returns from wi_action(), so we put it on ws_rerunq
* while rescheduling, and move it to runq after it returns
* from wi_action() */
- cfs_list_t ws_rerunq;
+ struct list_head ws_rerunq;
/** CPT-table for this scheduler */
struct cfs_cpt_table *ws_cptab;
/** CPT id for affinity */
unsigned int ws_starting:1;
/** scheduler name */
char ws_name[CFS_WS_NAME_LEN];
-} cfs_wi_sched_t;
+};
-struct cfs_workitem_data {
+static struct cfs_workitem_data {
/** serialize */
- cfs_spinlock_t wi_glock;
+ spinlock_t wi_glock;
/** list of all schedulers */
- cfs_list_t wi_scheds;
+ struct list_head wi_scheds;
/** WI module is initialized */
int wi_init;
/** shutting down the whole WI module */
int wi_stopping;
} cfs_wi_data;
-#ifdef __KERNEL__
-static inline void
-cfs_wi_sched_lock(cfs_wi_sched_t *sched)
-{
- cfs_spin_lock(&sched->ws_lock);
-}
-
-static inline void
-cfs_wi_sched_unlock(cfs_wi_sched_t *sched)
-{
- cfs_spin_unlock(&sched->ws_lock);
-}
-
static inline int
-cfs_wi_sched_cansleep(cfs_wi_sched_t *sched)
+cfs_wi_sched_cansleep(struct cfs_wi_sched *sched)
{
- cfs_wi_sched_lock(sched);
+ spin_lock(&sched->ws_lock);
if (sched->ws_stopping) {
- cfs_wi_sched_unlock(sched);
- return 0;
- }
-
- if (!cfs_list_empty(&sched->ws_runq)) {
- cfs_wi_sched_unlock(sched);
- return 0;
- }
- cfs_wi_sched_unlock(sched);
- return 1;
-}
-
-#else /* !__KERNEL__ */
-
-static inline void
-cfs_wi_sched_lock(cfs_wi_sched_t *sched)
-{
- cfs_spin_lock(&cfs_wi_data.wi_glock);
-}
+ spin_unlock(&sched->ws_lock);
+ return 0;
+ }
-static inline void
-cfs_wi_sched_unlock(cfs_wi_sched_t *sched)
-{
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
+ if (!list_empty(&sched->ws_runq)) {
+ spin_unlock(&sched->ws_lock);
+ return 0;
+ }
+ spin_unlock(&sched->ws_lock);
+ return 1;
}
-#endif /* __KERNEL__ */
-
/* XXX:
* 0. it only works when called from wi->wi_action.
* 1. when it returns no one shall try to schedule the workitem.
*/
void
-cfs_wi_exit(struct cfs_wi_sched *sched, cfs_workitem_t *wi)
+cfs_wi_exit(struct cfs_wi_sched *sched, struct cfs_workitem *wi)
{
- LASSERT(!cfs_in_interrupt()); /* because we use plain spinlock */
+ LASSERT(!in_interrupt()); /* because we use plain spinlock */
LASSERT(!sched->ws_stopping);
- cfs_wi_sched_lock(sched);
+ spin_lock(&sched->ws_lock);
-#ifdef __KERNEL__
LASSERT(wi->wi_running);
-#endif
+
if (wi->wi_scheduled) { /* cancel pending schedules */
- LASSERT(!cfs_list_empty(&wi->wi_list));
- cfs_list_del_init(&wi->wi_list);
+ LASSERT(!list_empty(&wi->wi_list));
+ list_del_init(&wi->wi_list);
LASSERT(sched->ws_nscheduled > 0);
sched->ws_nscheduled--;
}
- LASSERT(cfs_list_empty(&wi->wi_list));
+ LASSERT(list_empty(&wi->wi_list));
wi->wi_scheduled = 1; /* LBUG future schedule attempts */
- cfs_wi_sched_unlock(sched);
-
- return;
+ spin_unlock(&sched->ws_lock);
}
EXPORT_SYMBOL(cfs_wi_exit);
* cancel schedule request of workitem \a wi
*/
int
-cfs_wi_deschedule(struct cfs_wi_sched *sched, cfs_workitem_t *wi)
+cfs_wi_deschedule(struct cfs_wi_sched *sched, struct cfs_workitem *wi)
{
int rc;
- LASSERT(!cfs_in_interrupt()); /* because we use plain spinlock */
+ LASSERT(!in_interrupt()); /* because we use plain spinlock */
LASSERT(!sched->ws_stopping);
/*
* means the workitem will not be scheduled and will not have
* any race with wi_action.
*/
- cfs_wi_sched_lock(sched);
+ spin_lock(&sched->ws_lock);
rc = !(wi->wi_running);
if (wi->wi_scheduled) { /* cancel pending schedules */
- LASSERT(!cfs_list_empty(&wi->wi_list));
- cfs_list_del_init(&wi->wi_list);
+ LASSERT(!list_empty(&wi->wi_list));
+ list_del_init(&wi->wi_list);
LASSERT(sched->ws_nscheduled > 0);
sched->ws_nscheduled--;
- wi->wi_scheduled = 0;
- }
+ wi->wi_scheduled = 0;
+ }
- LASSERT (cfs_list_empty(&wi->wi_list));
+ LASSERT (list_empty(&wi->wi_list));
- cfs_wi_sched_unlock(sched);
- return rc;
+ spin_unlock(&sched->ws_lock);
+ return rc;
}
EXPORT_SYMBOL(cfs_wi_deschedule);
* be added, and even dynamic creation of serialised queues might be supported.
*/
void
-cfs_wi_schedule(struct cfs_wi_sched *sched, cfs_workitem_t *wi)
+cfs_wi_schedule(struct cfs_wi_sched *sched, struct cfs_workitem *wi)
{
- LASSERT(!cfs_in_interrupt()); /* because we use plain spinlock */
+ LASSERT(!in_interrupt()); /* because we use plain spinlock */
LASSERT(!sched->ws_stopping);
- cfs_wi_sched_lock(sched);
+ spin_lock(&sched->ws_lock);
- if (!wi->wi_scheduled) {
- LASSERT (cfs_list_empty(&wi->wi_list));
+ if (!wi->wi_scheduled) {
+ LASSERT (list_empty(&wi->wi_list));
- wi->wi_scheduled = 1;
+ wi->wi_scheduled = 1;
sched->ws_nscheduled++;
- if (!wi->wi_running) {
- cfs_list_add_tail(&wi->wi_list, &sched->ws_runq);
-#ifdef __KERNEL__
- cfs_waitq_signal(&sched->ws_waitq);
-#endif
- } else {
- cfs_list_add(&wi->wi_list, &sched->ws_rerunq);
- }
- }
+ if (!wi->wi_running) {
+ list_add_tail(&wi->wi_list, &sched->ws_runq);
+ wake_up(&sched->ws_waitq);
+ } else {
+ list_add(&wi->wi_list, &sched->ws_rerunq);
+ }
+ }
- LASSERT (!cfs_list_empty(&wi->wi_list));
- cfs_wi_sched_unlock(sched);
- return;
+ LASSERT (!list_empty(&wi->wi_list));
+ spin_unlock(&sched->ws_lock);
}
EXPORT_SYMBOL(cfs_wi_schedule);
-#ifdef __KERNEL__
-
static int
-cfs_wi_scheduler (void *arg)
+cfs_wi_scheduler(void *arg)
{
- struct cfs_wi_sched *sched = (cfs_wi_sched_t *)arg;
- char name[16];
-
- if (sched->ws_cptab != NULL && sched->ws_cpt >= 0) {
- snprintf(name, sizeof(name), "%s_%02d_%02d",
- sched->ws_name, sched->ws_cpt, sched->ws_nthreads);
- } else {
- snprintf(name, sizeof(name), "%s_%02d",
- sched->ws_name, sched->ws_nthreads);
- }
+ struct cfs_wi_sched *sched = (struct cfs_wi_sched *)arg;
- cfs_daemonize(name);
cfs_block_allsigs();
/* CPT affinity scheduler? */
if (sched->ws_cptab != NULL)
- cfs_cpt_bind(sched->ws_cptab, sched->ws_cpt);
+ if (cfs_cpt_bind(sched->ws_cptab, sched->ws_cpt) != 0)
+ CWARN("Unable to bind %s on CPU partition %d\n",
+ sched->ws_name, sched->ws_cpt);
- cfs_spin_lock(&cfs_wi_data.wi_glock);
+ spin_lock(&cfs_wi_data.wi_glock);
LASSERT(sched->ws_starting == 1);
sched->ws_starting--;
sched->ws_nthreads++;
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
+ spin_unlock(&cfs_wi_data.wi_glock);
- cfs_wi_sched_lock(sched);
+ spin_lock(&sched->ws_lock);
while (!sched->ws_stopping) {
- int nloops = 0;
- int rc;
- cfs_workitem_t *wi;
-
- while (!cfs_list_empty(&sched->ws_runq) &&
- nloops < CFS_WI_RESCHED) {
- wi = cfs_list_entry(sched->ws_runq.next,
- cfs_workitem_t, wi_list);
+ int nloops = 0;
+ int rc;
+ struct cfs_workitem *wi;
+
+ while (!list_empty(&sched->ws_runq) &&
+ nloops < CFS_WI_RESCHED) {
+ wi = list_entry(sched->ws_runq.next,
+ struct cfs_workitem, wi_list);
LASSERT(wi->wi_scheduled && !wi->wi_running);
- cfs_list_del_init(&wi->wi_list);
+ list_del_init(&wi->wi_list);
LASSERT(sched->ws_nscheduled > 0);
sched->ws_nscheduled--;
wi->wi_running = 1;
wi->wi_scheduled = 0;
-
- cfs_wi_sched_unlock(sched);
+ spin_unlock(&sched->ws_lock);
nloops++;
rc = (*wi->wi_action) (wi);
- cfs_wi_sched_lock(sched);
+ spin_lock(&sched->ws_lock);
if (rc != 0) /* WI should be dead, even be freed! */
continue;
- wi->wi_running = 0;
- if (cfs_list_empty(&wi->wi_list))
+ wi->wi_running = 0;
+ if (list_empty(&wi->wi_list))
continue;
LASSERT(wi->wi_scheduled);
- /* wi is rescheduled, should be on rerunq now, we
- * move it to runq so it can run action now */
- cfs_list_move_tail(&wi->wi_list, &sched->ws_runq);
+ /* wi is rescheduled, should be on rerunq now, we
+ * move it to runq so it can run action now */
+ list_move_tail(&wi->wi_list, &sched->ws_runq);
}
- if (!cfs_list_empty(&sched->ws_runq)) {
- cfs_wi_sched_unlock(sched);
- /* don't sleep because some workitems still
- * expect me to come back soon */
- cfs_cond_resched();
- cfs_wi_sched_lock(sched);
- continue;
- }
+ if (!list_empty(&sched->ws_runq)) {
+ spin_unlock(&sched->ws_lock);
+ /* don't sleep because some workitems still
+ * expect me to come back soon */
+ cond_resched();
+ spin_lock(&sched->ws_lock);
+ continue;
+ }
- cfs_wi_sched_unlock(sched);
- cfs_wait_event_interruptible_exclusive(sched->ws_waitq,
- !cfs_wi_sched_cansleep(sched), rc);
- cfs_wi_sched_lock(sched);
+ spin_unlock(&sched->ws_lock);
+ rc = wait_event_interruptible_exclusive(sched->ws_waitq,
+ !cfs_wi_sched_cansleep(sched));
+ spin_lock(&sched->ws_lock);
}
- cfs_wi_sched_unlock(sched);
+ spin_unlock(&sched->ws_lock);
- cfs_spin_lock(&cfs_wi_data.wi_glock);
+ spin_lock(&cfs_wi_data.wi_glock);
sched->ws_nthreads--;
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
+ spin_unlock(&cfs_wi_data.wi_glock);
- return 0;
-}
-
-#else /* __KERNEL__ */
-
-int
-cfs_wi_check_events (void)
-{
- int n = 0;
- cfs_workitem_t *wi;
-
- cfs_spin_lock(&cfs_wi_data.wi_glock);
-
- for (;;) {
- struct cfs_wi_sched *sched = NULL;
- struct cfs_wi_sched *tmp;
-
- /** rerunq is always empty for userspace */
- cfs_list_for_each_entry(tmp,
- &cfs_wi_data.wi_scheds, ws_list) {
- if (!cfs_list_empty(&tmp->ws_runq)) {
- sched = tmp;
- break;
- }
- }
-
- if (sched == NULL)
- break;
-
- wi = cfs_list_entry(sched->ws_runq.next,
- cfs_workitem_t, wi_list);
- cfs_list_del_init(&wi->wi_list);
-
- LASSERT(sched->ws_nscheduled > 0);
- sched->ws_nscheduled--;
-
- LASSERT (wi->wi_scheduled);
- wi->wi_scheduled = 0;
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
-
- n++;
- (*wi->wi_action) (wi);
-
- cfs_spin_lock(&cfs_wi_data.wi_glock);
- }
-
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
- return n;
+ return 0;
}
-#endif
-
void
cfs_wi_sched_destroy(struct cfs_wi_sched *sched)
{
- int i;
-
LASSERT(cfs_wi_data.wi_init);
LASSERT(!cfs_wi_data.wi_stopping);
- cfs_spin_lock(&cfs_wi_data.wi_glock);
+ spin_lock(&cfs_wi_data.wi_glock);
if (sched->ws_stopping) {
CDEBUG(D_INFO, "%s is in progress of stopping\n",
sched->ws_name);
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
+ spin_unlock(&cfs_wi_data.wi_glock);
return;
}
- LASSERT(!cfs_list_empty(&sched->ws_list));
+ LASSERT(!list_empty(&sched->ws_list));
sched->ws_stopping = 1;
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
+ spin_unlock(&cfs_wi_data.wi_glock);
+
+ wake_up_all(&sched->ws_waitq);
- i = 2;
-#ifdef __KERNEL__
- cfs_waitq_broadcast(&sched->ws_waitq);
+ spin_lock(&cfs_wi_data.wi_glock);
+ {
+ int i = 2;
- cfs_spin_lock(&cfs_wi_data.wi_glock);
- while (sched->ws_nthreads > 0) {
- CDEBUG(IS_PO2(++i) ? D_WARNING : D_NET,
- "waiting for %d threads of WI sched[%s] to terminate\n",
- sched->ws_nthreads, sched->ws_name);
+ while (sched->ws_nthreads > 0) {
+ CDEBUG(is_power_of_2(++i / 20) ? D_WARNING : D_NET,
+ "waiting %us for %d %s worker threads to exit\n",
+ i / 20, sched->ws_nthreads, sched->ws_name);
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
- cfs_pause(cfs_time_seconds(1) / 20);
- cfs_spin_lock(&cfs_wi_data.wi_glock);
+ spin_unlock(&cfs_wi_data.wi_glock);
+ schedule_timeout_uninterruptible(cfs_time_seconds(1)
+ / 20);
+ spin_lock(&cfs_wi_data.wi_glock);
+ }
}
- cfs_list_del(&sched->ws_list);
+ list_del(&sched->ws_list);
+
+ spin_unlock(&cfs_wi_data.wi_glock);
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
-#else
- SET_BUT_UNUSED(i);
-#endif
LASSERT(sched->ws_nscheduled == 0);
LIBCFS_FREE(sched, sizeof(*sched));
int cpt, int nthrs, struct cfs_wi_sched **sched_pp)
{
struct cfs_wi_sched *sched;
- int rc;
LASSERT(cfs_wi_data.wi_init);
LASSERT(!cfs_wi_data.wi_stopping);
if (sched == NULL)
return -ENOMEM;
- strncpy(sched->ws_name, name, CFS_WS_NAME_LEN);
+ if (strlen(name) > sizeof(sched->ws_name)-1) {
+ LIBCFS_FREE(sched, sizeof(*sched));
+ return -E2BIG;
+ }
+ strlcpy(sched->ws_name, name, sizeof(sched->ws_name));
+
sched->ws_cptab = cptab;
sched->ws_cpt = cpt;
-#ifdef __KERNEL__
- cfs_spin_lock_init(&sched->ws_lock);
- cfs_waitq_init(&sched->ws_waitq);
-#endif
- CFS_INIT_LIST_HEAD(&sched->ws_runq);
- CFS_INIT_LIST_HEAD(&sched->ws_rerunq);
- CFS_INIT_LIST_HEAD(&sched->ws_list);
-
- rc = 0;
-#ifdef __KERNEL__
- while (nthrs > 0) {
- cfs_spin_lock(&cfs_wi_data.wi_glock);
+ spin_lock_init(&sched->ws_lock);
+ init_waitqueue_head(&sched->ws_waitq);
+
+ INIT_LIST_HEAD(&sched->ws_runq);
+ INIT_LIST_HEAD(&sched->ws_rerunq);
+ INIT_LIST_HEAD(&sched->ws_list);
+
+ for (; nthrs > 0; nthrs--) {
+ char name[16];
+ struct task_struct *task;
+
+ spin_lock(&cfs_wi_data.wi_glock);
while (sched->ws_starting > 0) {
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
- cfs_schedule();
- cfs_spin_lock(&cfs_wi_data.wi_glock);
+ spin_unlock(&cfs_wi_data.wi_glock);
+ schedule();
+ spin_lock(&cfs_wi_data.wi_glock);
}
sched->ws_starting++;
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
-
- rc = cfs_create_thread(cfs_wi_scheduler, sched, 0);
- if (rc >= 0) {
- nthrs--;
- continue;
+ spin_unlock(&cfs_wi_data.wi_glock);
+
+ if (sched->ws_cptab != NULL && sched->ws_cpt >= 0) {
+ snprintf(name, sizeof(name), "%s_%02d_%02d",
+ sched->ws_name, sched->ws_cpt,
+ sched->ws_nthreads);
+ } else {
+ snprintf(name, sizeof(name), "%s_%02d",
+ sched->ws_name, sched->ws_nthreads);
}
- CERROR("Failed to create thread for WI scheduler %s: %d\n",
- name, rc);
+ task = kthread_run(cfs_wi_scheduler, sched, name);
+ if (IS_ERR(task)) {
+ int rc = PTR_ERR(task);
- cfs_spin_lock(&cfs_wi_data.wi_glock);
+ CERROR("Failed to create thread for "
+ "WI scheduler %s: %d\n", name, rc);
- /* make up for cfs_wi_sched_destroy */
- cfs_list_add(&sched->ws_list, &cfs_wi_data.wi_scheds);
- sched->ws_starting--;
+ spin_lock(&cfs_wi_data.wi_glock);
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
+ /* make up for cfs_wi_sched_destroy */
+ list_add(&sched->ws_list, &cfs_wi_data.wi_scheds);
+ sched->ws_starting--;
- cfs_wi_sched_destroy(sched);
- return rc;
+ spin_unlock(&cfs_wi_data.wi_glock);
+
+ cfs_wi_sched_destroy(sched);
+ return rc;
+ }
}
-#else
- SET_BUT_UNUSED(rc);
-#endif
- cfs_spin_lock(&cfs_wi_data.wi_glock);
- cfs_list_add(&sched->ws_list, &cfs_wi_data.wi_scheds);
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
+
+ spin_lock(&cfs_wi_data.wi_glock);
+ list_add(&sched->ws_list, &cfs_wi_data.wi_scheds);
+ spin_unlock(&cfs_wi_data.wi_glock);
*sched_pp = sched;
return 0;
int
cfs_wi_startup(void)
{
- memset(&cfs_wi_data, 0, sizeof(cfs_wi_data));
+ memset(&cfs_wi_data, 0, sizeof(struct cfs_workitem_data));
- cfs_spin_lock_init(&cfs_wi_data.wi_glock);
- CFS_INIT_LIST_HEAD(&cfs_wi_data.wi_scheds);
+ spin_lock_init(&cfs_wi_data.wi_glock);
+ INIT_LIST_HEAD(&cfs_wi_data.wi_scheds);
cfs_wi_data.wi_init = 1;
return 0;
{
struct cfs_wi_sched *sched;
- cfs_spin_lock(&cfs_wi_data.wi_glock);
+ spin_lock(&cfs_wi_data.wi_glock);
cfs_wi_data.wi_stopping = 1;
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
+ spin_unlock(&cfs_wi_data.wi_glock);
-#ifdef __KERNEL__
/* nobody should contend on this list */
- cfs_list_for_each_entry(sched, &cfs_wi_data.wi_scheds, ws_list) {
+ list_for_each_entry(sched, &cfs_wi_data.wi_scheds, ws_list) {
sched->ws_stopping = 1;
- cfs_waitq_broadcast(&sched->ws_waitq);
+ wake_up_all(&sched->ws_waitq);
}
- cfs_list_for_each_entry(sched, &cfs_wi_data.wi_scheds, ws_list) {
- cfs_spin_lock(&cfs_wi_data.wi_glock);
+ list_for_each_entry(sched, &cfs_wi_data.wi_scheds, ws_list) {
+ spin_lock(&cfs_wi_data.wi_glock);
while (sched->ws_nthreads != 0) {
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
- cfs_pause(cfs_time_seconds(1) / 20);
- cfs_spin_lock(&cfs_wi_data.wi_glock);
+ spin_unlock(&cfs_wi_data.wi_glock);
+ schedule_timeout_uninterruptible(cfs_time_seconds(1)
+ / 20);
+ spin_lock(&cfs_wi_data.wi_glock);
}
- cfs_spin_unlock(&cfs_wi_data.wi_glock);
+ spin_unlock(&cfs_wi_data.wi_glock);
}
-#endif
- while (!cfs_list_empty(&cfs_wi_data.wi_scheds)) {
- sched = cfs_list_entry(cfs_wi_data.wi_scheds.next,
+
+ while (!list_empty(&cfs_wi_data.wi_scheds)) {
+ sched = list_entry(cfs_wi_data.wi_scheds.next,
struct cfs_wi_sched, ws_list);
- cfs_list_del(&sched->ws_list);
+ list_del(&sched->ws_list);
LIBCFS_FREE(sched, sizeof(*sched));
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff