* 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, 2012, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
CFS_INIT_LIST_HEAD(&peer->ibp_conns);
CFS_INIT_LIST_HEAD(&peer->ibp_tx_queue);
- cfs_write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
+ write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
/* always called with a ref on ni, which prevents ni being shutdown */
LASSERT (net->ibn_shutdown == 0);
/* npeers only grows with the global lock held */
cfs_atomic_inc(&net->ibn_npeers);
- cfs_write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
+ write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
*peerp = peer;
return 0;
int i;
unsigned long flags;
- cfs_read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
+ read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
for (i = 0; i < kiblnd_data.kib_peer_hash_size; i++) {
*nidp = peer->ibp_nid;
*count = cfs_atomic_read(&peer->ibp_refcount);
- cfs_read_unlock_irqrestore(&kiblnd_data.kib_global_lock,
- flags);
- return 0;
- }
- }
+ read_unlock_irqrestore(&kiblnd_data.kib_global_lock,
+ flags);
+ return 0;
+ }
+ }
- cfs_read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
+ read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
return -ENOENT;
}
unsigned long flags;
int rc = -ENOENT;
- cfs_write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
+ write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
if (nid != LNET_NID_ANY) {
lo = hi = kiblnd_nid2peerlist(nid) - kiblnd_data.kib_peers;
}
}
- cfs_write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
+ write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
kiblnd_txlist_done(ni, &zombies, -EIO);
int i;
unsigned long flags;
- cfs_read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
+ read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
for (i = 0; i < kiblnd_data.kib_peer_hash_size; i++) {
cfs_list_for_each (ptmp, &kiblnd_data.kib_peers[i]) {
conn = cfs_list_entry(ctmp, kib_conn_t,
ibc_list);
kiblnd_conn_addref(conn);
- cfs_read_unlock_irqrestore(&kiblnd_data.kib_global_lock,
- flags);
- return conn;
- }
- }
- }
+ read_unlock_irqrestore(&kiblnd_data.kib_global_lock,
+ flags);
+ return conn;
+ }
+ }
+ }
- cfs_read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
+ read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
return NULL;
}
void
kiblnd_debug_conn (kib_conn_t *conn)
{
- cfs_list_t *tmp;
- int i;
+ cfs_list_t *tmp;
+ int i;
- cfs_spin_lock(&conn->ibc_lock);
+ spin_lock(&conn->ibc_lock);
CDEBUG(D_CONSOLE, "conn[%d] %p [version %x] -> %s: \n",
cfs_atomic_read(&conn->ibc_refcount), conn,
for (i = 0; i < IBLND_RX_MSGS(conn->ibc_version); i++)
kiblnd_debug_rx(&conn->ibc_rxs[i]);
- cfs_spin_unlock(&conn->ibc_lock);
+ spin_unlock(&conn->ibc_lock);
}
int
* she must dispose of 'cmid'. (Actually I'd block forever if I tried
* to destroy 'cmid' here since I'm called from the CM which still has
* its ref on 'cmid'). */
- cfs_rwlock_t *glock = &kiblnd_data.kib_global_lock;
+ rwlock_t *glock = &kiblnd_data.kib_global_lock;
kib_net_t *net = peer->ibp_ni->ni_data;
- kib_dev_t *dev = net->ibn_dev;
+ kib_dev_t *dev;
struct ib_qp_init_attr *init_qp_attr;
struct kib_sched_info *sched;
kib_conn_t *conn;
LASSERT(net != NULL);
LASSERT(!cfs_in_interrupt());
+ dev = net->ibn_dev;
+
cpt = lnet_cpt_of_nid(peer->ibp_nid);
sched = kiblnd_data.kib_scheds[cpt];
CFS_INIT_LIST_HEAD(&conn->ibc_tx_queue_rsrvd);
CFS_INIT_LIST_HEAD(&conn->ibc_tx_queue_nocred);
CFS_INIT_LIST_HEAD(&conn->ibc_active_txs);
- cfs_spin_lock_init(&conn->ibc_lock);
+ spin_lock_init(&conn->ibc_lock);
LIBCFS_CPT_ALLOC(conn->ibc_connvars, lnet_cpt_table(), cpt,
sizeof(*conn->ibc_connvars));
- if (conn->ibc_connvars == NULL) {
- CERROR("Can't allocate in-progress connection state\n");
- goto failed_2;
- }
+ if (conn->ibc_connvars == NULL) {
+ CERROR("Can't allocate in-progress connection state\n");
+ goto failed_2;
+ }
- cfs_write_lock_irqsave(glock, flags);
- if (dev->ibd_failover) {
- cfs_write_unlock_irqrestore(glock, flags);
- CERROR("%s: failover in progress\n", dev->ibd_ifname);
- goto failed_2;
- }
+ write_lock_irqsave(glock, flags);
+ if (dev->ibd_failover) {
+ write_unlock_irqrestore(glock, flags);
+ CERROR("%s: failover in progress\n", dev->ibd_ifname);
+ goto failed_2;
+ }
- if (dev->ibd_hdev->ibh_ibdev != cmid->device) {
- /* wakeup failover thread and teardown connection */
- if (kiblnd_dev_can_failover(dev)) {
- cfs_list_add_tail(&dev->ibd_fail_list,
- &kiblnd_data.kib_failed_devs);
- cfs_waitq_signal(&kiblnd_data.kib_failover_waitq);
- }
+ if (dev->ibd_hdev->ibh_ibdev != cmid->device) {
+ /* wakeup failover thread and teardown connection */
+ if (kiblnd_dev_can_failover(dev)) {
+ cfs_list_add_tail(&dev->ibd_fail_list,
+ &kiblnd_data.kib_failed_devs);
+ wake_up(&kiblnd_data.kib_failover_waitq);
+ }
- cfs_write_unlock_irqrestore(glock, flags);
- CERROR("cmid HCA(%s), kib_dev(%s) need failover\n",
- cmid->device->name, dev->ibd_ifname);
- goto failed_2;
- }
+ write_unlock_irqrestore(glock, flags);
+ CERROR("cmid HCA(%s), kib_dev(%s) need failover\n",
+ cmid->device->name, dev->ibd_ifname);
+ goto failed_2;
+ }
kiblnd_hdev_addref_locked(dev->ibd_hdev);
conn->ibc_hdev = dev->ibd_hdev;
kiblnd_setup_mtu_locked(cmid);
- cfs_write_unlock_irqrestore(glock, flags);
+ write_unlock_irqrestore(glock, flags);
LIBCFS_CPT_ALLOC(conn->ibc_rxs, lnet_cpt_table(), cpt,
IBLND_RX_MSGS(version) * sizeof(kib_rx_t));
rc = kiblnd_alloc_pages(&conn->ibc_rx_pages, cpt,
IBLND_RX_MSG_PAGES(version));
- if (rc != 0)
- goto failed_2;
+ if (rc != 0)
+ goto failed_2;
- kiblnd_map_rx_descs(conn);
+ kiblnd_map_rx_descs(conn);
#ifdef HAVE_OFED_IB_COMP_VECTOR
cq = ib_create_cq(cmid->device,
/* correct # of posted buffers
* NB locking needed now I'm racing with completion */
- cfs_spin_lock_irqsave(&sched->ibs_lock, flags);
+ spin_lock_irqsave(&sched->ibs_lock, flags);
conn->ibc_nrx -= IBLND_RX_MSGS(version) - i;
- cfs_spin_unlock_irqrestore(&sched->ibs_lock, flags);
+ spin_unlock_irqrestore(&sched->ibs_lock, flags);
/* cmid will be destroyed by CM(ofed) after cm_callback
* returned, so we can't refer it anymore
unsigned long flags;
int count = 0;
- cfs_write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
+ write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
if (nid != LNET_NID_ANY)
lo = hi = kiblnd_nid2peerlist(nid) - kiblnd_data.kib_peers;
}
}
- cfs_write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
+ write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
/* wildcards always succeed */
if (nid == LNET_NID_ANY)
void
kiblnd_query (lnet_ni_t *ni, lnet_nid_t nid, cfs_time_t *when)
{
- cfs_time_t last_alive = 0;
- cfs_time_t now = cfs_time_current();
- cfs_rwlock_t *glock = &kiblnd_data.kib_global_lock;
- kib_peer_t *peer;
- unsigned long flags;
+ cfs_time_t last_alive = 0;
+ cfs_time_t now = cfs_time_current();
+ rwlock_t *glock = &kiblnd_data.kib_global_lock;
+ kib_peer_t *peer;
+ unsigned long flags;
- cfs_read_lock_irqsave(glock, flags);
+ read_lock_irqsave(glock, flags);
peer = kiblnd_find_peer_locked(nid);
if (peer != NULL) {
last_alive = peer->ibp_last_alive;
}
- cfs_read_unlock_irqrestore(glock, flags);
+ read_unlock_irqrestore(glock, flags);
if (last_alive != 0)
*when = last_alive;
for (i = 0; i < npages; i++) {
if (p->ibp_pages[i] != NULL)
- cfs_free_page(p->ibp_pages[i]);
+ __free_page(p->ibp_pages[i]);
}
LIBCFS_FREE(p, offsetof(kib_pages_t, ibp_pages[npages]));
for (i = 0; i < npages; i++) {
p->ibp_pages[i] = cfs_page_cpt_alloc(lnet_cpt_table(), cpt,
- CFS_ALLOC_IO);
+ __GFP_IO);
if (p->ibp_pages[i] == NULL) {
CERROR("Can't allocate page %d of %d\n", i, npages);
kiblnd_free_pages(p);
unsigned long flags;
int i = 0;
- cfs_read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
- while (dev->ibd_failover) {
- cfs_read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
- if (i++ % 50 == 0)
- CDEBUG(D_NET, "Wait for dev(%s) failover\n", dev->ibd_ifname);
- cfs_schedule_timeout(cfs_time_seconds(1) / 100);
+ read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
+ while (dev->ibd_failover) {
+ read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
+ if (i++ % 50 == 0)
+ CDEBUG(D_NET, "%s: Wait for failover\n",
+ dev->ibd_ifname);
+ schedule_timeout(cfs_time_seconds(1) / 100);
- cfs_read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
- }
+ read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
+ }
- kiblnd_hdev_addref_locked(dev->ibd_hdev);
- hdev = dev->ibd_hdev;
+ kiblnd_hdev_addref_locked(dev->ibd_hdev);
+ hdev = dev->ibd_hdev;
- cfs_read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
+ read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
- return hdev;
+ return hdev;
}
static void
kib_pages_t *txpgs = tpo->tpo_tx_pages;
kib_pool_t *pool = &tpo->tpo_pool;
kib_net_t *net = pool->po_owner->ps_net;
- kib_dev_t *dev = net->ibn_dev;
+ kib_dev_t *dev;
struct page *page;
kib_tx_t *tx;
int page_offset;
LASSERT (net != NULL);
+ dev = net->ibn_dev;
+
/* pre-mapped messages are not bigger than 1 page */
CLASSERT (IBLND_MSG_SIZE <= PAGE_SIZE);
if (fps->fps_net == NULL) /* intialized? */
return;
- cfs_spin_lock(&fps->fps_lock);
+ spin_lock(&fps->fps_lock);
while (!cfs_list_empty(&fps->fps_pool_list)) {
kib_fmr_pool_t *fpo = cfs_list_entry(fps->fps_pool_list.next,
cfs_list_add(&fpo->fpo_list, &fps->fps_failed_pool_list);
}
- cfs_spin_unlock(&fps->fps_lock);
+ spin_unlock(&fps->fps_lock);
}
static void
fps->fps_cpt = cpt;
fps->fps_pool_size = pool_size;
fps->fps_flush_trigger = flush_trigger;
- cfs_spin_lock_init(&fps->fps_lock);
+ spin_lock_init(&fps->fps_lock);
CFS_INIT_LIST_HEAD(&fps->fps_pool_list);
CFS_INIT_LIST_HEAD(&fps->fps_failed_pool_list);
fmr->fmr_pool = NULL;
fmr->fmr_pfmr = NULL;
- cfs_spin_lock(&fps->fps_lock);
+ spin_lock(&fps->fps_lock);
fpo->fpo_map_count --; /* decref the pool */
cfs_list_for_each_entry_safe(fpo, tmp, &fps->fps_pool_list, fpo_list) {
fps->fps_version ++;
}
}
- cfs_spin_unlock(&fps->fps_lock);
+ spin_unlock(&fps->fps_lock);
if (!cfs_list_empty(&zombies))
kiblnd_destroy_fmr_pool_list(&zombies);
int rc;
again:
- cfs_spin_lock(&fps->fps_lock);
- version = fps->fps_version;
- cfs_list_for_each_entry(fpo, &fps->fps_pool_list, fpo_list) {
- fpo->fpo_deadline = cfs_time_shift(IBLND_POOL_DEADLINE);
- fpo->fpo_map_count ++;
- cfs_spin_unlock(&fps->fps_lock);
+ spin_lock(&fps->fps_lock);
+ version = fps->fps_version;
+ cfs_list_for_each_entry(fpo, &fps->fps_pool_list, fpo_list) {
+ fpo->fpo_deadline = cfs_time_shift(IBLND_POOL_DEADLINE);
+ fpo->fpo_map_count++;
+ spin_unlock(&fps->fps_lock);
pfmr = ib_fmr_pool_map_phys(fpo->fpo_fmr_pool,
pages, npages, iov);
return 0;
}
- cfs_spin_lock(&fps->fps_lock);
- fpo->fpo_map_count --;
- if (PTR_ERR(pfmr) != -EAGAIN) {
- cfs_spin_unlock(&fps->fps_lock);
- return PTR_ERR(pfmr);
- }
+ spin_lock(&fps->fps_lock);
+ fpo->fpo_map_count--;
+ if (PTR_ERR(pfmr) != -EAGAIN) {
+ spin_unlock(&fps->fps_lock);
+ return PTR_ERR(pfmr);
+ }
- /* EAGAIN and ... */
- if (version != fps->fps_version) {
- cfs_spin_unlock(&fps->fps_lock);
- goto again;
- }
- }
+ /* EAGAIN and ... */
+ if (version != fps->fps_version) {
+ spin_unlock(&fps->fps_lock);
+ goto again;
+ }
+ }
- if (fps->fps_increasing) {
- cfs_spin_unlock(&fps->fps_lock);
- CDEBUG(D_NET, "Another thread is allocating new "
- "FMR pool, waiting for her to complete\n");
- cfs_schedule();
- goto again;
+ if (fps->fps_increasing) {
+ spin_unlock(&fps->fps_lock);
+ CDEBUG(D_NET, "Another thread is allocating new "
+ "FMR pool, waiting for her to complete\n");
+ schedule();
+ goto again;
- }
+ }
- if (cfs_time_before(cfs_time_current(), fps->fps_next_retry)) {
- /* someone failed recently */
- cfs_spin_unlock(&fps->fps_lock);
- return -EAGAIN;
- }
+ if (cfs_time_before(cfs_time_current(), fps->fps_next_retry)) {
+ /* someone failed recently */
+ spin_unlock(&fps->fps_lock);
+ return -EAGAIN;
+ }
- fps->fps_increasing = 1;
- cfs_spin_unlock(&fps->fps_lock);
+ fps->fps_increasing = 1;
+ spin_unlock(&fps->fps_lock);
- CDEBUG(D_NET, "Allocate new FMR pool\n");
- rc = kiblnd_create_fmr_pool(fps, &fpo);
- cfs_spin_lock(&fps->fps_lock);
- fps->fps_increasing = 0;
- if (rc == 0) {
- fps->fps_version ++;
- cfs_list_add_tail(&fpo->fpo_list, &fps->fps_pool_list);
- } else {
- fps->fps_next_retry = cfs_time_shift(IBLND_POOL_RETRY);
- }
- cfs_spin_unlock(&fps->fps_lock);
+ CDEBUG(D_NET, "Allocate new FMR pool\n");
+ rc = kiblnd_create_fmr_pool(fps, &fpo);
+ spin_lock(&fps->fps_lock);
+ fps->fps_increasing = 0;
+ if (rc == 0) {
+ fps->fps_version++;
+ cfs_list_add_tail(&fpo->fpo_list, &fps->fps_pool_list);
+ } else {
+ fps->fps_next_retry = cfs_time_shift(IBLND_POOL_RETRY);
+ }
+ spin_unlock(&fps->fps_lock);
- goto again;
+ goto again;
}
static void
if (ps->ps_net == NULL) /* intialized? */
return;
- cfs_spin_lock(&ps->ps_lock);
+ spin_lock(&ps->ps_lock);
while (!cfs_list_empty(&ps->ps_pool_list)) {
kib_pool_t *po = cfs_list_entry(ps->ps_pool_list.next,
kib_pool_t, po_list);
else
cfs_list_add(&po->po_list, &ps->ps_failed_pool_list);
}
- cfs_spin_unlock(&ps->ps_lock);
+ spin_unlock(&ps->ps_lock);
}
static void
ps->ps_node_init = nd_init;
ps->ps_node_fini = nd_fini;
ps->ps_pool_size = size;
- strncpy(ps->ps_name, name, IBLND_POOL_NAME_LEN);
- cfs_spin_lock_init(&ps->ps_lock);
+ if (strlcpy(ps->ps_name, name, sizeof(ps->ps_name))
+ >= sizeof(ps->ps_name))
+ return -E2BIG;
+ spin_lock_init(&ps->ps_lock);
CFS_INIT_LIST_HEAD(&ps->ps_pool_list);
CFS_INIT_LIST_HEAD(&ps->ps_failed_pool_list);
kib_pool_t *tmp;
cfs_time_t now = cfs_time_current();
- cfs_spin_lock(&ps->ps_lock);
+ spin_lock(&ps->ps_lock);
if (ps->ps_node_fini != NULL)
ps->ps_node_fini(pool, node);
if (kiblnd_pool_is_idle(pool, now))
cfs_list_move(&pool->po_list, &zombies);
}
- cfs_spin_unlock(&ps->ps_lock);
+ spin_unlock(&ps->ps_lock);
- if (!cfs_list_empty(&zombies))
- kiblnd_destroy_pool_list(&zombies);
+ if (!cfs_list_empty(&zombies))
+ kiblnd_destroy_pool_list(&zombies);
}
cfs_list_t *
int rc;
again:
- cfs_spin_lock(&ps->ps_lock);
+ spin_lock(&ps->ps_lock);
cfs_list_for_each_entry(pool, &ps->ps_pool_list, po_list) {
if (cfs_list_empty(&pool->po_free_list))
continue;
/* still hold the lock */
ps->ps_node_init(pool, node);
}
- cfs_spin_unlock(&ps->ps_lock);
- return node;
- }
+ spin_unlock(&ps->ps_lock);
+ return node;
+ }
- /* no available tx pool and ... */
- if (ps->ps_increasing) {
- /* another thread is allocating a new pool */
- cfs_spin_unlock(&ps->ps_lock);
+ /* no available tx pool and ... */
+ if (ps->ps_increasing) {
+ /* another thread is allocating a new pool */
+ spin_unlock(&ps->ps_lock);
CDEBUG(D_NET, "Another thread is allocating new "
"%s pool, waiting for her to complete\n",
ps->ps_name);
- cfs_schedule();
+ schedule();
goto again;
}
- if (cfs_time_before(cfs_time_current(), ps->ps_next_retry)) {
- /* someone failed recently */
- cfs_spin_unlock(&ps->ps_lock);
- return NULL;
- }
+ if (cfs_time_before(cfs_time_current(), ps->ps_next_retry)) {
+ /* someone failed recently */
+ spin_unlock(&ps->ps_lock);
+ return NULL;
+ }
- ps->ps_increasing = 1;
- cfs_spin_unlock(&ps->ps_lock);
+ ps->ps_increasing = 1;
+ spin_unlock(&ps->ps_lock);
- CDEBUG(D_NET, "%s pool exhausted, allocate new pool\n", ps->ps_name);
+ CDEBUG(D_NET, "%s pool exhausted, allocate new pool\n", ps->ps_name);
- rc = ps->ps_pool_create(ps, ps->ps_pool_size, &pool);
+ rc = ps->ps_pool_create(ps, ps->ps_pool_size, &pool);
- cfs_spin_lock(&ps->ps_lock);
+ spin_lock(&ps->ps_lock);
ps->ps_increasing = 0;
if (rc == 0) {
cfs_list_add_tail(&pool->po_list, &ps->ps_pool_list);
CERROR("Can't allocate new %s pool because out of memory\n",
ps->ps_name);
}
- cfs_spin_unlock(&ps->ps_lock);
+ spin_unlock(&ps->ps_lock);
- goto again;
+ goto again;
}
void
if (tx->tx_frags == NULL)
break;
+ sg_init_table(tx->tx_frags, IBLND_MAX_RDMA_FRAGS);
+
LIBCFS_CPT_ALLOC(tx->tx_wrq, lnet_cpt_table(), ps->ps_cpt,
(1 + IBLND_MAX_RDMA_FRAGS) *
sizeof(*tx->tx_wrq));
int rc;
int i;
- net->ibn_tx_ps = cfs_percpt_alloc(lnet_cpt_table(),
- sizeof(kib_tx_poolset_t));
- if (net->ibn_tx_ps == NULL) {
- CERROR("Failed to allocate tx pool array\n");
- return -ENOMEM;
- }
-
- for (i = 0; i < ncpts; i++) {
- cpt = (cpts == NULL) ? i : cpts[i];
- rc = kiblnd_init_poolset(&net->ibn_tx_ps[cpt]->tps_poolset,
- cpt, net, "TX",
- kiblnd_tx_pool_size(ncpts),
- kiblnd_create_tx_pool,
- kiblnd_destroy_tx_pool,
- kiblnd_tx_init, NULL);
- if (rc != 0) {
- CERROR("Failed to initialize TX pool\n");
- goto failed;
- }
- }
-
- cfs_read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
+ read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
if (*kiblnd_tunables.kib_map_on_demand == 0 &&
net->ibn_dev->ibd_hdev->ibh_nmrs == 1) {
- cfs_read_unlock_irqrestore(&kiblnd_data.kib_global_lock,
+ read_unlock_irqrestore(&kiblnd_data.kib_global_lock,
flags);
- return 0;
+ goto create_tx_pool;
}
- cfs_read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
+ read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
if (*kiblnd_tunables.kib_fmr_pool_size <
*kiblnd_tunables.kib_ntx / 4) {
CERROR("Can't set fmr pool size (%d) < ntx / 4(%d)\n",
*kiblnd_tunables.kib_fmr_pool_size,
*kiblnd_tunables.kib_ntx / 4);
+ rc = -EINVAL;
goto failed;
}
+ /* TX pool must be created later than FMR/PMR, see LU-2268
+ * for details */
+ LASSERT(net->ibn_tx_ps == NULL);
+
/* premapping can fail if ibd_nmr > 1, so we always create
* FMR/PMR pool and map-on-demand if premapping failed */
sizeof(kib_fmr_poolset_t));
if (net->ibn_fmr_ps == NULL) {
CERROR("Failed to allocate FMR pool array\n");
+ rc = -ENOMEM;
goto failed;
}
kiblnd_fmr_flush_trigger(ncpts));
if (rc == -ENOSYS && i == 0) /* no FMR */
break; /* create PMR pool */
- if (rc != 0)
- goto failed; /* a real error */
+
+ if (rc != 0) { /* a real error */
+ CERROR("Can't initialize FMR pool for CPT %d: %d\n",
+ cpt, rc);
+ goto failed;
+ }
+ }
+
+ if (i > 0) {
+ LASSERT(i == ncpts);
+ goto create_tx_pool;
}
cfs_percpt_free(net->ibn_fmr_ps);
net->ibn_fmr_ps = NULL;
+ CWARN("Device does not support FMR, failing back to PMR\n");
+
if (*kiblnd_tunables.kib_pmr_pool_size <
*kiblnd_tunables.kib_ntx / 4) {
CERROR("Can't set pmr pool size (%d) < ntx / 4(%d)\n",
*kiblnd_tunables.kib_pmr_pool_size,
*kiblnd_tunables.kib_ntx / 4);
+ rc = -EINVAL;
goto failed;
}
sizeof(kib_pmr_poolset_t));
if (net->ibn_pmr_ps == NULL) {
CERROR("Failed to allocate PMR pool array\n");
+ rc = -ENOMEM;
goto failed;
}
kiblnd_pmr_pool_size(ncpts),
kiblnd_create_pmr_pool,
kiblnd_destroy_pmr_pool, NULL, NULL);
- if (rc != 0)
+ if (rc != 0) {
+ CERROR("Can't initialize PMR pool for CPT %d: %d\n",
+ cpt, rc);
goto failed;
+ }
}
- return 0;
+ create_tx_pool:
+ net->ibn_tx_ps = cfs_percpt_alloc(lnet_cpt_table(),
+ sizeof(kib_tx_poolset_t));
+ if (net->ibn_tx_ps == NULL) {
+ CERROR("Failed to allocate tx pool array\n");
+ rc = -ENOMEM;
+ goto failed;
+ }
+ for (i = 0; i < ncpts; i++) {
+ cpt = (cpts == NULL) ? i : cpts[i];
+ rc = kiblnd_init_poolset(&net->ibn_tx_ps[cpt]->tps_poolset,
+ cpt, net, "TX",
+ kiblnd_tx_pool_size(ncpts),
+ kiblnd_create_tx_pool,
+ kiblnd_destroy_tx_pool,
+ kiblnd_tx_init, NULL);
+ if (rc != 0) {
+ CERROR("Can't initialize TX pool for CPT %d: %d\n",
+ cpt, rc);
+ goto failed;
+ }
+ }
+
+ return 0;
failed:
kiblnd_net_fini_pools(net);
+ LASSERT(rc != 0);
return rc;
}
* because we can fail to create new listener.
* But we have to close it now, otherwise rdma_bind_addr
* will return EADDRINUSE... How crap! */
- cfs_write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
+ write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
- cmid = dev->ibd_hdev->ibh_cmid;
- /* make next schedule of kiblnd_dev_need_failover
- * will return 1 for me */
- dev->ibd_hdev->ibh_cmid = NULL;
- cfs_write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
+ cmid = dev->ibd_hdev->ibh_cmid;
+ /* make next schedule of kiblnd_dev_need_failover()
+ * return 1 for me */
+ dev->ibd_hdev->ibh_cmid = NULL;
+ write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
rdma_destroy_id(cmid);
}
goto out;
}
- cfs_write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
+ write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
old = dev->ibd_hdev;
dev->ibd_hdev = hdev; /* take over the refcount */
}
}
- cfs_write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
+ write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
out:
if (!cfs_list_empty(&zombie_tpo))
kiblnd_destroy_pool_list(&zombie_tpo);
LASSERT (cfs_list_empty(&kiblnd_data.kib_connd_zombies));
LASSERT (cfs_list_empty(&kiblnd_data.kib_connd_conns));
- /* flag threads to terminate; wake and wait for them to die */
- kiblnd_data.kib_shutdown = 1;
+ /* flag threads to terminate; wake and wait for them to die */
+ kiblnd_data.kib_shutdown = 1;
/* NB: we really want to stop scheduler threads net by net
* instead of the whole module, this should be improved
* with dynamic configuration LNet */
cfs_percpt_for_each(sched, i, kiblnd_data.kib_scheds)
- cfs_waitq_broadcast(&sched->ibs_waitq);
+ wake_up_all(&sched->ibs_waitq);
- cfs_waitq_broadcast(&kiblnd_data.kib_connd_waitq);
- cfs_waitq_broadcast(&kiblnd_data.kib_failover_waitq);
+ wake_up_all(&kiblnd_data.kib_connd_waitq);
+ wake_up_all(&kiblnd_data.kib_failover_waitq);
- i = 2;
- while (cfs_atomic_read(&kiblnd_data.kib_nthreads) != 0) {
+ i = 2;
+ while (cfs_atomic_read(&kiblnd_data.kib_nthreads) != 0) {
i++;
CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, /* power of 2? */
"Waiting for %d threads to terminate\n",
CDEBUG(D_MALLOC, "after LND base cleanup: kmem %d\n",
cfs_atomic_read(&libcfs_kmemory));
- kiblnd_data.kib_init = IBLND_INIT_NOTHING;
- PORTAL_MODULE_UNUSE;
+ kiblnd_data.kib_init = IBLND_INIT_NOTHING;
+ module_put(THIS_MODULE);
}
void
kiblnd_shutdown (lnet_ni_t *ni)
{
kib_net_t *net = ni->ni_data;
- cfs_rwlock_t *g_lock = &kiblnd_data.kib_global_lock;
+ rwlock_t *g_lock = &kiblnd_data.kib_global_lock;
int i;
unsigned long flags;
CDEBUG(D_MALLOC, "before LND net cleanup: kmem %d\n",
cfs_atomic_read(&libcfs_kmemory));
- cfs_write_lock_irqsave(g_lock, flags);
- net->ibn_shutdown = 1;
- cfs_write_unlock_irqrestore(g_lock, flags);
+ write_lock_irqsave(g_lock, flags);
+ net->ibn_shutdown = 1;
+ write_unlock_irqrestore(g_lock, flags);
switch (net->ibn_init) {
default:
kiblnd_net_fini_pools(net);
- cfs_write_lock_irqsave(g_lock, flags);
- LASSERT (net->ibn_dev->ibd_nnets > 0);
- net->ibn_dev->ibd_nnets--;
- cfs_list_del(&net->ibn_list);
- cfs_write_unlock_irqrestore(g_lock, flags);
+ write_lock_irqsave(g_lock, flags);
+ LASSERT(net->ibn_dev->ibd_nnets > 0);
+ net->ibn_dev->ibd_nnets--;
+ cfs_list_del(&net->ibn_list);
+ write_unlock_irqrestore(g_lock, flags);
/* fall through */
LASSERT (kiblnd_data.kib_init == IBLND_INIT_NOTHING);
- PORTAL_MODULE_USE;
- memset(&kiblnd_data, 0, sizeof(kiblnd_data)); /* zero pointers, flags etc */
+ try_module_get(THIS_MODULE);
+ memset(&kiblnd_data, 0, sizeof(kiblnd_data)); /* zero pointers, flags etc */
- cfs_rwlock_init(&kiblnd_data.kib_global_lock);
+ rwlock_init(&kiblnd_data.kib_global_lock);
CFS_INIT_LIST_HEAD(&kiblnd_data.kib_devs);
CFS_INIT_LIST_HEAD(&kiblnd_data.kib_failed_devs);
for (i = 0; i < kiblnd_data.kib_peer_hash_size; i++)
CFS_INIT_LIST_HEAD(&kiblnd_data.kib_peers[i]);
- cfs_spin_lock_init(&kiblnd_data.kib_connd_lock);
- CFS_INIT_LIST_HEAD(&kiblnd_data.kib_connd_conns);
- CFS_INIT_LIST_HEAD(&kiblnd_data.kib_connd_zombies);
- cfs_waitq_init(&kiblnd_data.kib_connd_waitq);
- cfs_waitq_init(&kiblnd_data.kib_failover_waitq);
+ spin_lock_init(&kiblnd_data.kib_connd_lock);
+ CFS_INIT_LIST_HEAD(&kiblnd_data.kib_connd_conns);
+ CFS_INIT_LIST_HEAD(&kiblnd_data.kib_connd_zombies);
+ init_waitqueue_head(&kiblnd_data.kib_connd_waitq);
+ init_waitqueue_head(&kiblnd_data.kib_failover_waitq);
kiblnd_data.kib_scheds = cfs_percpt_alloc(lnet_cpt_table(),
sizeof(*sched));
cfs_percpt_for_each(sched, i, kiblnd_data.kib_scheds) {
int nthrs;
- cfs_spin_lock_init(&sched->ibs_lock);
+ spin_lock_init(&sched->ibs_lock);
CFS_INIT_LIST_HEAD(&sched->ibs_conns);
- cfs_waitq_init(&sched->ibs_waitq);
+ init_waitqueue_head(&sched->ibs_waitq);
nthrs = cfs_cpt_weight(lnet_cpt_table(), i);
if (*kiblnd_tunables.kib_nscheds > 0) {
kiblnd_data.kib_init = IBLND_INIT_DATA;
/*****************************************************/
- rc = kiblnd_thread_start(kiblnd_connd, NULL);
+ rc = kiblnd_thread_start(kiblnd_connd, NULL, "kiblnd_connd");
if (rc != 0) {
CERROR("Can't spawn o2iblnd connd: %d\n", rc);
goto failed;
}
- if (*kiblnd_tunables.kib_dev_failover != 0)
- rc = kiblnd_thread_start(kiblnd_failover_thread, NULL);
+ if (*kiblnd_tunables.kib_dev_failover != 0)
+ rc = kiblnd_thread_start(kiblnd_failover_thread, NULL,
+ "kiblnd_failover");
if (rc != 0) {
CERROR("Can't spawn o2iblnd failover thread: %d\n", rc);
for (i = 0; i < nthrs; i++) {
long id;
-
+ char name[20];
id = KIB_THREAD_ID(sched->ibs_cpt, sched->ibs_nthreads + i);
- rc = kiblnd_thread_start(kiblnd_scheduler, (void *)id);
+ snprintf(name, sizeof(name), "kiblnd_sd_%02ld_%02ld",
+ KIB_THREAD_CPT(id), KIB_THREAD_TID(id));
+ rc = kiblnd_thread_start(kiblnd_scheduler, (void *)id, name);
if (rc == 0)
continue;
if (net == NULL)
goto failed;
- memset(net, 0, sizeof(*net));
+ memset(net, 0, sizeof(*net));
- cfs_gettimeofday(&tv);
- net->ibn_incarnation = (((__u64)tv.tv_sec) * 1000000) + tv.tv_usec;
+ do_gettimeofday(&tv);
+ net->ibn_incarnation = (((__u64)tv.tv_sec) * 1000000) + tv.tv_usec;
ni->ni_peertimeout = *kiblnd_tunables.kib_peertimeout;
ni->ni_maxtxcredits = *kiblnd_tunables.kib_credits;
goto failed;
}
- cfs_write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
- ibdev->ibd_nnets++;
- cfs_list_add_tail(&net->ibn_list, &ibdev->ibd_nets);
- cfs_write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
+ write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
+ ibdev->ibd_nnets++;
+ cfs_list_add_tail(&net->ibn_list, &ibdev->ibd_nets);
+ write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
net->ibn_init = IBLND_INIT_ALL;
return 0;
failed:
- if (net->ibn_dev == NULL && ibdev != NULL)
+ if (net != NULL && net->ibn_dev == NULL && ibdev != NULL)
kiblnd_destroy_dev(ibdev);
kiblnd_shutdown(ni);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff