vv_return_t vvrc;
/* mapping must be active (it dropped fmrcount to 0) */
- LASSERT (tx->tx_md.md_active);
+ LASSERT (tx->tx_md.md_active);
vvrc = vv_unmap_fmr(kibnal_data.kib_hca,
1, &tx->tx_md.md_fmrhandle);
/* tx may have up to 2 lnet msgs to finalise */
lntmsg[0] = tx->tx_lntmsg[0]; tx->tx_lntmsg[0] = NULL;
lntmsg[1] = tx->tx_lntmsg[1]; tx->tx_lntmsg[1] = NULL;
-
+
if (tx->tx_conn != NULL) {
kibnal_conn_decref(tx->tx_conn);
tx->tx_conn = NULL;
}
kib_tx_t *
-kibnal_get_idle_tx (void)
+kibnal_get_idle_tx (void)
{
kib_tx_t *tx;
-
+
spin_lock(&kibnal_data.kib_tx_lock);
if (list_empty (&kibnal_data.kib_idle_txs)) {
LASSERT (tx->tx_conn == NULL);
LASSERT (tx->tx_lntmsg[0] == NULL);
LASSERT (tx->tx_lntmsg[1] == NULL);
-
+
return tx;
}
/* old peers don't reserve rxs for RDMA replies */
LASSERT (!rsrvd_credit ||
conn->ibc_version != IBNAL_MSG_VERSION_RDMAREPLYNOTRSRVD);
-
+
rx->rx_gl = (vv_scatgat_t) {
.v_address = KIBNAL_ADDR2SG(addr),
.l_key = rx->rx_lkey,
LASSERT (conn->ibc_state >= IBNAL_CONN_INIT);
LASSERT (rx->rx_nob >= 0); /* not posted */
- CDEBUG(D_NET, "posting rx [%d %x "LPX64"]\n",
+ CDEBUG(D_NET, "posting rx [%d %x "LPX64"]\n",
rx->rx_wrq.scatgat_list->length,
rx->rx_wrq.scatgat_list->l_key,
KIBNAL_SG2ADDR(rx->rx_wrq.scatgat_list->v_address));
kibnal_conn_decref(conn);
return 0;
}
-
+
rx->rx_nob = -1; /* flag posted */
-
+
spin_lock(&conn->ibc_lock);
/* Serialise vv_post_receive; it's not re-entrant on the same QP */
vvrc = vv_post_receive(kibnal_data.kib_hca,
return 0;
}
-
+
spin_unlock(&conn->ibc_lock);
- CERROR ("post rx -> %s failed %d\n",
+ CERROR ("post rx -> %s failed %d\n",
libcfs_nid2str(conn->ibc_peer->ibp_nid), vvrc);
rc = -EIO;
- kibnal_close_conn(rx->rx_conn, rc);
+ kibnal_close_conn(conn, rc);
/* No more posts for this rx; so lose its ref */
kibnal_conn_decref(conn);
return rc;
kibnal_find_waiting_tx_locked(kib_conn_t *conn, int txtype, __u64 cookie)
{
struct list_head *tmp;
-
+
list_for_each(tmp, &conn->ibc_active_txs) {
kib_tx_t *tx = list_entry(tmp, kib_tx_t, tx_list);
-
+
LASSERT (!tx->tx_queued);
LASSERT (tx->tx_sending != 0 || tx->tx_waiting);
if (tx->tx_status == 0) { /* success so far */
if (status < 0) { /* failed? */
tx->tx_status = status;
- } else if (txtype == IBNAL_MSG_GET_REQ) {
+ } else if (txtype == IBNAL_MSG_GET_REQ) {
lnet_set_reply_msg_len(kibnal_data.kib_ni,
tx->tx_lntmsg[1], status);
}
}
-
+
tx->tx_waiting = 0;
idle = !tx->tx_queued && (tx->tx_sending == 0);
list_del(&tx->tx_list);
spin_unlock(&conn->ibc_lock);
-
+
if (idle)
kibnal_tx_done(tx);
}
void
-kibnal_send_completion (kib_conn_t *conn, int type, int status, __u64 cookie)
+kibnal_send_completion (kib_conn_t *conn, int type, int status, __u64 cookie)
{
kib_tx_t *tx = kibnal_get_idle_tx();
-
+
if (tx == NULL) {
CERROR("Can't get tx for completion %x for %s\n",
type, libcfs_nid2str(conn->ibc_peer->ibp_nid));
return;
}
-
+
tx->tx_msg->ibm_u.completion.ibcm_status = status;
tx->tx_msg->ibm_u.completion.ibcm_cookie = cookie;
kibnal_init_tx_msg(tx, type, sizeof(kib_completion_msg_t));
-
+
kibnal_queue_tx(tx, conn);
}
CDEBUG (D_NET, "Received %x[%d] from %s\n",
msg->ibm_type, credits, libcfs_nid2str(conn->ibc_peer->ibp_nid));
-
+
if (credits != 0) {
/* Have I received credits that will let me send? */
spin_lock(&conn->ibc_lock);
msg->ibm_srcnid, rx, 0);
repost = rc < 0; /* repost on error */
break;
-
+
case IBNAL_MSG_PUT_REQ:
rc = lnet_parse(kibnal_data.kib_ni, &msg->ibm_u.putreq.ibprm_hdr,
msg->ibm_srcnid, rx, 1);
case IBNAL_MSG_PUT_NAK:
rsrvd_credit = 1; /* rdma reply (was pre-reserved) */
-
+
CWARN ("PUT_NACK from %s\n", libcfs_nid2str(conn->ibc_peer->ibp_nid));
- kibnal_handle_completion(conn, IBNAL_MSG_PUT_REQ,
+ kibnal_handle_completion(conn, IBNAL_MSG_PUT_REQ,
msg->ibm_u.completion.ibcm_status,
msg->ibm_u.completion.ibcm_cookie);
break;
tx->tx_nwrq = 0; /* overwrite PUT_REQ */
- rc2 = kibnal_init_rdma(tx, IBNAL_MSG_PUT_DONE,
+ rc2 = kibnal_init_rdma(tx, IBNAL_MSG_PUT_DONE,
kibnal_rd_size(&msg->ibm_u.putack.ibpam_rd),
&msg->ibm_u.putack.ibpam_rd,
msg->ibm_u.putack.ibpam_dst_cookie);
kibnal_queue_tx_locked(tx, conn);
spin_unlock(&conn->ibc_lock);
break;
-
+
case IBNAL_MSG_PUT_DONE:
/* This buffer was pre-reserved by not returning the credit
* when the PUT_REQ's buffer was reposted, so I just return it
case IBNAL_MSG_GET_DONE:
rsrvd_credit = 1; /* rdma reply (was pre-reserved) */
-
+
kibnal_handle_completion(conn, IBNAL_MSG_GET_REQ,
msg->ibm_u.completion.ibcm_status,
msg->ibm_u.completion.ibcm_cookie);
goto ignore;
if (vvrc != vv_comp_status_success) {
- CERROR("Rx from %s failed: %d\n",
+ CERROR("Rx from %s failed: %d\n",
libcfs_nid2str(conn->ibc_peer->ibp_nid), vvrc);
goto failed;
}
if (!lnet_ptlcompat_matchnid(conn->ibc_peer->ibp_nid,
msg->ibm_srcnid) ||
- !lnet_ptlcompat_matchnid(kibnal_data.kib_ni->ni_nid,
+ !lnet_ptlcompat_matchnid(kibnal_data.kib_ni->ni_nid,
msg->ibm_dstnid) ||
msg->ibm_srcstamp != conn->ibc_incarnation ||
msg->ibm_dststamp != kibnal_data.kib_incarnation) {
/* must check holding global lock to eliminate race */
if (conn->ibc_state < IBNAL_CONN_ESTABLISHED) {
list_add_tail(&rx->rx_list, &conn->ibc_early_rxs);
- write_unlock_irqrestore(&kibnal_data.kib_global_lock,
+ write_unlock_irqrestore(&kibnal_data.kib_global_lock,
flags);
return;
}
- write_unlock_irqrestore(&kibnal_data.kib_global_lock,
+ write_unlock_irqrestore(&kibnal_data.kib_global_lock,
flags);
}
kibnal_handle_rx(rx);
return;
-
+
failed:
CDEBUG(D_NET, "rx %p conn %p\n", rx, conn);
kibnal_close_conn(conn, -EIO);
LASSERT (page != NULL);
return page;
}
-#if CONFIG_HIGHMEM
+#ifdef CONFIG_HIGHMEM
if (vaddr >= PKMAP_BASE &&
vaddr < (PKMAP_BASE + LAST_PKMAP * PAGE_SIZE)) {
/* No highmem pages only used for bulk (kiov) I/O */
#if !IBNAL_USE_FMR
int
-kibnal_append_rdfrag(kib_rdma_desc_t *rd, int active, struct page *page,
+kibnal_append_rdfrag(kib_rdma_desc_t *rd, int active, struct page *page,
unsigned long page_offset, unsigned long len)
{
kib_rdma_frag_t *frag = &rd->rd_frags[rd->rd_nfrag];
/* NB this relies entirely on there being a single region for the whole
* of memory, since "high" memory will wrap in the (void *) cast! */
- vvrc = vv_get_gen_mr_attrib(kibnal_data.kib_hca,
+ vvrc = vv_get_gen_mr_attrib(kibnal_data.kib_hca,
(void *)((unsigned long)addr),
len, &mem_h, &l_key, &r_key);
LASSERT (vvrc == vv_return_ok);
kibnal_rf_set(frag, frag_addr, len);
- CDEBUG(D_NET,"map frag [%d][%d %x %08x%08x] "LPX64"\n",
- rd->rd_nfrag, frag->rf_nob, rd->rd_key,
+ CDEBUG(D_NET,"map frag [%d][%d %x %08x%08x] "LPX64"\n",
+ rd->rd_nfrag, frag->rf_nob, rd->rd_key,
frag->rf_addr_hi, frag->rf_addr_lo, frag_addr);
rd->rd_nfrag++;
}
int
-kibnal_setup_rd_iov(kib_tx_t *tx, kib_rdma_desc_t *rd,
+kibnal_setup_rd_iov(kib_tx_t *tx, kib_rdma_desc_t *rd,
vv_access_con_bit_mask_t access,
unsigned int niov, struct iovec *iov, int offset, int nob)
-
{
/* active if I'm sending */
int active = ((access & vv_acc_r_mem_write) == 0);
fragnob = min((int)(iov->iov_len - offset), nob);
fragnob = min(fragnob, (int)PAGE_SIZE - page_offset);
- rc = kibnal_append_rdfrag(rd, active, page,
+ rc = kibnal_append_rdfrag(rd, active, page,
page_offset, fragnob);
if (rc != 0)
return rc;
}
nob -= fragnob;
} while (nob > 0);
-
+
return 0;
}
int
-kibnal_setup_rd_kiov (kib_tx_t *tx, kib_rdma_desc_t *rd,
+kibnal_setup_rd_kiov (kib_tx_t *tx, kib_rdma_desc_t *rd,
vv_access_con_bit_mask_t access,
int nkiov, lnet_kiov_t *kiov, int offset, int nob)
{
do {
LASSERT (nkiov > 0);
fragnob = min((int)(kiov->kiov_len - offset), nob);
-
+
rc = kibnal_append_rdfrag(rd, active, kiov->kiov_page,
kiov->kiov_offset + offset,
fragnob);
vvrc = vv_map_fmr(kibnal_data.kib_hca, tx->tx_md.md_fmrhandle,
&map_props, &tx->tx_md.md_lkey, &tx->tx_md.md_rkey);
if (vvrc != vv_return_ok) {
- CERROR ("Can't map vaddr %p for %d in %d pages: %d\n",
+ CERROR ("Can't map vaddr %p for %d in %d pages: %d\n",
map_props.start, nob, npages, vvrc);
return -EFAULT;
}
kibnal_setup_rd_iov (kib_tx_t *tx, kib_rdma_desc_t *rd,
vv_access_con_bit_mask_t access,
unsigned int niov, struct iovec *iov, int offset, int nob)
-
{
/* active if I'm sending */
int active = ((access & vv_acc_r_mem_write) == 0);
}
vaddr = ((unsigned long)iov->iov_base) + offset;
-
+
page_offset = vaddr & (PAGE_SIZE - 1);
resid = nob;
npages = 0;
int resid;
int npages;
unsigned long page_offset;
-
+
CDEBUG(D_NET, "niov %d offset %d nob %d\n", nkiov, offset, nob);
LASSERT (nob > 0);
}
page_offset = kiov->kiov_offset + offset;
-
+
resid = offset + nob;
npages = 0;
LASSERT (nkiov > 0);
if ((npages > 0 && kiov->kiov_offset != 0) ||
- (resid > kiov->kiov_len &&
+ (resid > kiov->kiov_len &&
(kiov->kiov_offset + kiov->kiov_len) != PAGE_SIZE)) {
/* Can't have gaps */
CERROR ("Can't make payload contiguous in I/O VM:"
"page %d, offset %d, len %d \n",
npages, kiov->kiov_offset, kiov->kiov_len);
-
+
return -EINVAL;
}
libcfs_nid2str(conn->ibc_peer->ibp_nid));
return;
}
-
+
spin_lock(&conn->ibc_lock);
LASSERT (conn->ibc_nsends_posted <=
*kibnal_tunables.kib_concurrent_sends);
LASSERT (conn->ibc_reserved_credits >= 0);
-
+
while (conn->ibc_reserved_credits > 0 &&
!list_empty(&conn->ibc_tx_queue_rsrvd)) {
- LASSERT (conn->ibc_version !=
+ LASSERT (conn->ibc_version !=
IBNAL_MSG_VERSION_RDMAREPLYNOTRSRVD);
tx = list_entry(conn->ibc_tx_queue_rsrvd.next,
kib_tx_t, tx_list);
list_add_tail(&tx->tx_list, &conn->ibc_tx_queue);
conn->ibc_reserved_credits--;
}
-
+
if (list_empty(&conn->ibc_tx_queue) &&
list_empty(&conn->ibc_tx_queue_nocred) &&
(conn->ibc_outstanding_credits >= IBNAL_CREDIT_HIGHWATER ||
kibnal_send_keepalive(conn))) {
spin_unlock(&conn->ibc_lock);
-
+
tx = kibnal_get_idle_tx();
if (tx != NULL)
kibnal_init_tx_msg(tx, IBNAL_MSG_NOOP, 0);
spin_lock(&conn->ibc_lock);
-
+
if (tx != NULL)
kibnal_queue_tx_locked(tx, conn);
}
for (;;) {
if (!list_empty(&conn->ibc_tx_queue_nocred)) {
- LASSERT (conn->ibc_version !=
+ LASSERT (conn->ibc_version !=
IBNAL_MSG_VERSION_RDMAREPLYNOTRSRVD);
- tx = list_entry (conn->ibc_tx_queue_nocred.next,
+ tx = list_entry (conn->ibc_tx_queue_nocred.next,
kib_tx_t, tx_list);
consume_cred = 0;
} else if (!list_empty (&conn->ibc_tx_queue)) {
- tx = list_entry (conn->ibc_tx_queue.next,
+ tx = list_entry (conn->ibc_tx_queue.next,
kib_tx_t, tx_list);
consume_cred = 1;
} else {
/* nothing waiting */
break;
}
-
+
LASSERT (tx->tx_queued);
/* We rely on this for QP sizing */
LASSERT (tx->tx_nwrq > 0 && tx->tx_nwrq <= 1 + IBNAL_MAX_RDMA_FRAGS);
libcfs_nid2str(conn->ibc_peer->ibp_nid));
break;
}
-
+
if (consume_cred) {
if (conn->ibc_credits == 0) { /* no credits */
CDEBUG(D_NET, "%s: no credits\n",
libcfs_nid2str(conn->ibc_peer->ibp_nid));
break;
}
-
+
if (conn->ibc_credits == 1 && /* last credit reserved for */
conn->ibc_outstanding_credits == 0) { /* giving back credits */
CDEBUG(D_NET, "%s: not using last credit\n",
break;
}
}
-
+
list_del (&tx->tx_list);
tx->tx_queued = 0;
tx->tx_status = rc;
tx->tx_waiting = 0;
tx->tx_sending--;
-
+
done = (tx->tx_sending == 0);
if (done)
list_del (&tx->tx_list);
-
+
spin_unlock(&conn->ibc_lock);
-
+
if (conn->ibc_state == IBNAL_CONN_ESTABLISHED)
CERROR ("Error %d posting transmit to %s\n",
vvrc, libcfs_nid2str(conn->ibc_peer->ibp_nid));
tx->tx_waiting = 0;
tx->tx_status = -EIO;
}
-
+
idle = (tx->tx_sending == 0) && /* This is the final callback */
!tx->tx_waiting && /* Not waiting for peer */
!tx->tx_queued; /* Not re-queued (PUT_DONE) */
int nob = offsetof (kib_msg_t, ibm_u) + body_nob;
__u64 addr = (__u64)((unsigned long)((tx)->tx_msg));
- LASSERT (tx->tx_nwrq >= 0 &&
+ LASSERT (tx->tx_nwrq >= 0 &&
tx->tx_nwrq < (1 + IBNAL_MAX_RDMA_FRAGS));
LASSERT (nob <= IBNAL_MSG_SIZE);
wrq->type.send.solicited_event = 1;
wrq->type.send.immidiate_data_indicator = 0;
wrq->type.send.send_qp_type.rc_type.fance_indicator = 0;
-
+
tx->tx_nwrq++;
}
rc = -EPROTO;
break;
}
-
+
if (dstidx == dstrd->rd_nfrag) {
CERROR("Dst buffer exhausted: %d frags\n", dstidx);
rc = -EPROTO;
resid -= wrknob;
if (wrknob < srcfrag->rf_nob) {
- kibnal_rf_set(srcfrag,
- kibnal_rf_addr(srcfrag) + wrknob,
+ kibnal_rf_set(srcfrag,
+ kibnal_rf_addr(srcfrag) + wrknob,
srcfrag->rf_nob - wrknob);
} else {
srcfrag++;
srcidx++;
}
-
+
if (wrknob < dstfrag->rf_nob) {
kibnal_rf_set(dstfrag,
kibnal_rf_addr(dstfrag) + wrknob,
dstfrag++;
dstidx++;
}
-
+
tx->tx_nwrq++;
}
if (rc < 0) /* no RDMA if completing with failure */
tx->tx_nwrq = 0;
#endif
-
+
ibmsg->ibm_u.completion.ibcm_status = rc;
ibmsg->ibm_u.completion.ibcm_cookie = dstcookie;
kibnal_init_tx_msg(tx, type, sizeof (kib_completion_msg_t));
spin_lock(&conn->ibc_lock);
kibnal_queue_tx_locked (tx, conn);
spin_unlock(&conn->ibc_lock);
-
+
kibnal_check_sends(conn);
}
/* If I get here, I've committed to send, so I complete the tx with
* failure on any problems */
-
+
LASSERT (tx->tx_conn == NULL); /* only set when assigned a conn */
LASSERT (tx->tx_nwrq > 0); /* work items have been set up */
for (retry = 0; ; retry = 1) {
read_lock_irqsave(g_lock, flags);
-
+
peer = kibnal_find_peer_locked (nid);
if (peer != NULL) {
conn = kibnal_find_conn_locked (peer);
return;
}
}
-
+
/* Making one or more connections; I'll need a write lock... */
read_unlock(g_lock);
write_lock(g_lock);
if (rc != 0) {
CERROR("Can't add peer %s: %d\n",
libcfs_nid2str(nid), rc);
-
+
tx->tx_status = -EHOSTUNREACH;
tx->tx_waiting = 0;
kibnal_tx_done (tx);
/* Connection exists; queue message on it */
kibnal_conn_addref(conn); /* 1 ref for me... */
write_unlock_irqrestore(g_lock, flags);
-
+
kibnal_queue_tx (tx, conn);
kibnal_conn_decref(conn); /* ...until here */
return;
peer->ibp_arp_count = 1 + *kibnal_tunables.kib_arp_retries;
kibnal_schedule_peer_arp(peer);
}
-
+
/* A connection is being established; queue the message... */
list_add_tail (&tx->tx_list, &peer->ibp_tx_queue);
int
kibnal_send(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg)
{
- lnet_hdr_t *hdr = &lntmsg->msg_hdr;
- int type = lntmsg->msg_type;
+ lnet_hdr_t *hdr = &lntmsg->msg_hdr;
+ int type = lntmsg->msg_type;
lnet_process_id_t target = lntmsg->msg_target;
int target_is_router = lntmsg->msg_target_is_router;
int routing = lntmsg->msg_routing;
- unsigned int payload_niov = lntmsg->msg_niov;
- struct iovec *payload_iov = lntmsg->msg_iov;
+ unsigned int payload_niov = lntmsg->msg_niov;
+ struct iovec *payload_iov = lntmsg->msg_iov;
lnet_kiov_t *payload_kiov = lntmsg->msg_kiov;
unsigned int payload_offset = lntmsg->msg_offset;
unsigned int payload_nob = lntmsg->msg_len;
default:
LBUG();
return (-EIO);
-
+
case LNET_MSG_ACK:
LASSERT (payload_nob == 0);
break;
case LNET_MSG_GET:
if (routing || target_is_router)
break; /* send IMMEDIATE */
-
+
/* is the REPLY message too small for RDMA? */
nob = offsetof(kib_msg_t, ibm_u.immediate.ibim_payload[lntmsg->msg_md->md_length]);
if (nob <= IBNAL_MSG_SIZE)
libcfs_nid2str(target.nid));
return -ENOMEM;
}
-
+
ibmsg = tx->tx_msg;
ibmsg->ibm_u.get.ibgm_hdr = *hdr;
ibmsg->ibm_u.get.ibgm_cookie = tx->tx_cookie;
#else
{
int n = ibmsg->ibm_u.get.ibgm_rd.rd_nfrag;
-
+
nob = offsetof(kib_get_msg_t, ibgm_rd.rd_frags[n]);
}
#endif
kibnal_reply (lnet_ni_t *ni, kib_rx_t *rx, lnet_msg_t *lntmsg)
{
lnet_process_id_t target = lntmsg->msg_target;
- unsigned int niov = lntmsg->msg_niov;
- struct iovec *iov = lntmsg->msg_iov;
+ unsigned int niov = lntmsg->msg_niov;
+ struct iovec *iov = lntmsg->msg_iov;
lnet_kiov_t *kiov = lntmsg->msg_kiov;
unsigned int offset = lntmsg->msg_offset;
unsigned int nob = lntmsg->msg_len;
kib_tx_t *tx;
int rc;
-
+
tx = kibnal_get_idle_tx();
if (tx == NULL) {
CERROR("Can't get tx for REPLY to %s\n",
if (nob == 0)
rc = 0;
else if (kiov == NULL)
- rc = kibnal_setup_rd_iov(tx, tx->tx_rd, 0,
+ rc = kibnal_setup_rd_iov(tx, tx->tx_rd, 0,
niov, iov, offset, nob);
else
rc = kibnal_setup_rd_kiov(tx, tx->tx_rd, 0,
libcfs_nid2str(target.nid), rc);
goto failed_1;
}
-
+
rc = kibnal_init_rdma(tx, IBNAL_MSG_GET_DONE, nob,
&rx->rx_msg->ibm_u.get.ibgm_rd,
rx->rx_msg->ibm_u.get.ibgm_cookie);
if (rc < 0) {
- CERROR("Can't setup rdma for GET from %s: %d\n",
+ CERROR("Can't setup rdma for GET from %s: %d\n",
libcfs_nid2str(target.nid), rc);
goto failed_1;
}
-
+
if (rc == 0) {
/* No RDMA: local completion may happen now! */
lnet_finalize(ni, lntmsg, 0);
* completes */
tx->tx_lntmsg[0] = lntmsg;
}
-
+
kibnal_queue_tx(tx, rx->rx_conn);
return;
-
+
failed_1:
kibnal_tx_done(tx);
failed_0:
if (conn->ibc_version == IBNAL_MSG_VERSION_RDMAREPLYNOTRSRVD) {
/* Can't block if RDMA completions need normal credits */
- LCONSOLE_ERROR(0x129, "Dropping message from %s: no buffers "
- " free. %s is running an old version of LNET "
- "that may deadlock if messages wait for buffers)"
- "\n", libcfs_nid2str(conn->ibc_peer->ibp_nid),
- libcfs_nid2str(conn->ibc_peer->ibp_nid));
+ LCONSOLE_ERROR_MSG(0x129, "Dropping message from %s: no buffers"
+ " free. %s is running an old version of LNET "
+ "that may deadlock if messages wait for"
+ "buffers) \n",
+ libcfs_nid2str(conn->ibc_peer->ibp_nid),
+ libcfs_nid2str(conn->ibc_peer->ibp_nid));
return -EDEADLK;
}
-
+
*new_private = private;
return 0;
}
int nob;
int post_cred = 1;
int rc = 0;
-
+
LASSERT (mlen <= rlen);
LASSERT (!in_interrupt());
/* Either all pages or all vaddrs */
switch (rxmsg->ibm_type) {
default:
LBUG();
-
+
case IBNAL_MSG_IMMEDIATE:
nob = offsetof(kib_msg_t, ibm_u.immediate.ibim_payload[rlen]);
if (nob > rx->rx_nob) {
case IBNAL_MSG_PUT_REQ:
if (mlen == 0) {
lnet_finalize(ni, lntmsg, 0);
- kibnal_send_completion(rx->rx_conn, IBNAL_MSG_PUT_NAK, 0,
+ kibnal_send_completion(conn, IBNAL_MSG_PUT_NAK, 0,
rxmsg->ibm_u.putreq.ibprm_cookie);
break;
}
-
+
tx = kibnal_get_idle_tx();
if (tx == NULL) {
CERROR("Can't allocate tx for %s\n",
txmsg = tx->tx_msg;
if (kiov == NULL)
- rc = kibnal_setup_rd_iov(tx,
+ rc = kibnal_setup_rd_iov(tx,
&txmsg->ibm_u.putack.ibpam_rd,
vv_acc_r_mem_write,
niov, iov, offset, mlen);
libcfs_nid2str(conn->ibc_peer->ibp_nid), rc);
kibnal_tx_done(tx);
/* tell peer it's over */
- kibnal_send_completion(rx->rx_conn, IBNAL_MSG_PUT_NAK, rc,
+ kibnal_send_completion(conn, IBNAL_MSG_PUT_NAK, rc,
rxmsg->ibm_u.putreq.ibprm_cookie);
break;
}
kibnal_reply(ni, rx, lntmsg);
} else {
/* GET didn't match anything */
- kibnal_send_completion(rx->rx_conn, IBNAL_MSG_GET_DONE,
- -ENODATA,
+ kibnal_send_completion(conn, IBNAL_MSG_GET_DONE, -ENODATA,
rxmsg->ibm_u.get.ibgm_cookie);
}
break;
time_t last_alive = 0;
int error = 0;
unsigned long flags;
-
+
read_lock_irqsave(&kibnal_data.kib_global_lock, flags);
if (list_empty(&peer->ibp_conns) &&
peer->ibp_error != 0) {
error = peer->ibp_error;
peer->ibp_error = 0;
-
+
last_alive = cfs_time_current_sec() -
cfs_duration_sec(cfs_time_current() -
peer->ibp_last_alive);
}
-
+
read_unlock_irqrestore(&kibnal_data.kib_global_lock, flags);
-
+
if (error != 0)
lnet_notify(kibnal_data.kib_ni, peer->ibp_nid, 0, last_alive);
}
unsigned long flags;
kibnal_conn_addref(conn); /* ++ref for connd */
-
+
spin_lock_irqsave(&kibnal_data.kib_connd_lock, flags);
list_add_tail (&conn->ibc_list, &kibnal_data.kib_connd_conns);
wake_up (&kibnal_data.kib_connd_waitq);
-
+
spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags);
}
* already dealing with it (either to set it up or tear it down).
* Caller holds kib_global_lock exclusively in irq context */
kib_peer_t *peer = conn->ibc_peer;
-
+
LASSERT (error != 0 || conn->ibc_state >= IBNAL_CONN_ESTABLISHED);
if (error != 0 && conn->ibc_comms_error == 0)
if (conn->ibc_state != IBNAL_CONN_ESTABLISHED)
return; /* already being handled */
-
+
/* NB Can't take ibc_lock here (could be in IRQ context), without
* risking deadlock, so access to ibc_{tx_queue,active_txs} is racey */
list_empty(&conn->ibc_tx_queue_nocred) &&
list_empty(&conn->ibc_active_txs)) {
CDEBUG(D_NET, "closing conn to %s"
- " rx# "LPD64" tx# "LPD64"\n",
+ " rx# "LPD64" tx# "LPD64"\n",
libcfs_nid2str(peer->ibp_nid),
conn->ibc_txseq, conn->ibc_rxseq);
} else {
kibnal_close_conn (kib_conn_t *conn, int error)
{
unsigned long flags;
-
+
write_lock_irqsave(&kibnal_data.kib_global_lock, flags);
kibnal_close_conn_locked (conn, error);
-
+
write_unlock_irqrestore(&kibnal_data.kib_global_lock, flags);
}
LASSERT (!in_interrupt());
LASSERT (conn->ibc_state >= IBNAL_CONN_ESTABLISHED);
-
+
write_lock_irqsave(&kibnal_data.kib_global_lock, flags);
while (!list_empty(&conn->ibc_early_rxs)) {
rx = list_entry(conn->ibc_early_rxs.next,
kib_rx_t, rx_list);
list_del(&rx->rx_list);
write_unlock_irqrestore(&kibnal_data.kib_global_lock, flags);
-
+
kibnal_handle_rx(rx);
-
+
write_lock_irqsave(&kibnal_data.kib_global_lock, flags);
}
write_unlock_irqrestore(&kibnal_data.kib_global_lock, flags);
void
kibnal_abort_txs(kib_conn_t *conn, struct list_head *txs)
{
- LIST_HEAD (zombies);
+ LIST_HEAD (zombies);
struct list_head *tmp;
struct list_head *nxt;
kib_tx_t *tx;
} else {
LASSERT (tx->tx_queued);
}
-
+
tx->tx_status = -ECONNABORTED;
tx->tx_queued = 0;
tx->tx_waiting = 0;
-
+
if (tx->tx_sending == 0) {
list_del (&tx->tx_list);
list_add (&tx->tx_list, &zombies);
LASSERT (!in_interrupt());
LASSERT (current == kibnal_data.kib_connd);
LASSERT (conn->ibc_state >= IBNAL_CONN_INIT);
-
+
kibnal_set_conn_state(conn, IBNAL_CONN_DISCONNECTED);
/* move QP to error state to make posted work items complete */
LASSERT (peer->ibp_accepting != 0);
peer->ibp_accepting--;
}
-
+
if (peer->ibp_connecting != 0 ||
peer->ibp_accepting != 0) {
/* another connection attempt under way (loopback?)... */
peer->ibp_reconnect_interval =
MIN(peer->ibp_reconnect_interval,
*kibnal_tunables.kib_max_reconnect_interval);
-
- peer->ibp_reconnect_time = jiffies +
+
+ peer->ibp_reconnect_time = jiffies +
peer->ibp_reconnect_interval * HZ;
/* Take peer's blocked transmits to complete with error */
list_add(&zombies, &peer->ibp_tx_queue);
list_del_init(&peer->ibp_tx_queue);
-
+
if (kibnal_peer_active(peer) &&
(peer->ibp_persistence == 0)) {
/* failed connection attempt on non-persistent peer */
/* Can't have blocked transmits if there are connections */
LASSERT (list_empty(&peer->ibp_tx_queue));
}
-
+
write_unlock_irqrestore(&kibnal_data.kib_global_lock, flags);
kibnal_peer_notify(peer);
- if (list_empty (&zombies))
+ if (list_empty (&zombies))
return;
-
+
CDEBUG (D_NETERROR, "Deleting messages for %s: connection failed\n",
libcfs_nid2str(peer->ibp_nid));
} else {
LASSERT (peer->ibp_accepting > 0);
}
-
+
LIBCFS_FREE(conn->ibc_connvars, sizeof(*conn->ibc_connvars));
conn->ibc_connvars = NULL;
break;
}
- kibnal_peer_connect_failed(conn->ibc_peer, active, status);
+ kibnal_peer_connect_failed(peer, active, status);
kibnal_conn_disconnected(conn);
return;
}
} else {
LASSERT(conn->ibc_state == IBNAL_CONN_PASSIVE_WAIT);
}
-
+
conn->ibc_last_send = jiffies;
kibnal_set_conn_state(conn, IBNAL_CONN_ESTABLISHED);
kibnal_peer_alive(peer);
* peer instance... */
kibnal_conn_addref(conn); /* +1 ref for ibc_list */
list_add(&conn->ibc_list, &peer->ibp_conns);
- kibnal_close_stale_conns_locked (conn->ibc_peer,
- conn->ibc_incarnation);
+ kibnal_close_stale_conns_locked (peer, conn->ibc_incarnation);
if (!kibnal_peer_active(peer) || /* peer has been deleted */
conn->ibc_comms_error != 0 || /* comms error */
conn->ibc_disconnect) { /* need to disconnect */
-
+
/* start to shut down connection */
kibnal_close_conn_locked(conn, -ECONNABORTED);
/* grab pending txs while I have the lock */
list_add(&txs, &peer->ibp_tx_queue);
list_del_init(&peer->ibp_tx_queue);
-
+
peer->ibp_reconnect_interval = 0; /* OK to reconnect at any time */
write_unlock_irqrestore(&kibnal_data.kib_global_lock, flags);
kibnal_cm_callback(cm_cep_handle_t cep, cm_conn_data_t *cmdata, void *arg)
{
static cm_dreply_data_t drep; /* just zeroed space */
-
+
kib_conn_t *conn = (kib_conn_t *)arg;
unsigned long flags;
-
+
/* CAVEAT EMPTOR: tasklet context */
switch (cmdata->status) {
default:
LBUG();
-
+
case cm_event_disconn_request:
/* IBNAL_CONN_ACTIVE_RTU: gets closed in kibnal_connreq_done
* IBNAL_CONN_ESTABLISHED: I start it closing
}
write_unlock_irqrestore(&kibnal_data.kib_global_lock, flags);
break;
-
+
case cm_event_disconn_timeout:
case cm_event_disconn_reply:
write_lock_irqsave(&kibnal_data.kib_global_lock, flags);
write_unlock_irqrestore(&kibnal_data.kib_global_lock, flags);
break;
-
+
case cm_event_connected:
case cm_event_conn_timeout:
case cm_event_conn_reject:
switch (conn->ibc_connvars->cv_conndata.status) {
default:
LBUG();
-
+
case cm_event_connected:
kibnal_conn_addref(conn); /* ++ ref for CM callback */
rc = kibnal_set_qp_state(conn, vv_qp_state_rts);
* an error immediately... */
kibnal_connreq_done(conn, 0, 0);
break;
-
+
case cm_event_conn_timeout:
kibnal_connreq_done(conn, 0, -ETIMEDOUT);
break;
-
+
case cm_event_conn_reject:
kibnal_connreq_done(conn, 0, -ECONNRESET);
break;
kib_connvars_t *cv;
cm_return_t cmrc;
vv_return_t vvrc;
-
+
/* I'm the connd executing in thread context
* No concurrency problems with static data! */
LASSERT (!in_interrupt());
if (!lnet_ptlcompat_matchnid(kibnal_data.kib_ni->ni_nid,
rxmsg.ibm_dstnid)) {
CERROR("Can't accept %s: bad dst nid %s\n",
- libcfs_nid2str(rxmsg.ibm_srcnid),
+ libcfs_nid2str(rxmsg.ibm_srcnid),
libcfs_nid2str(rxmsg.ibm_dstnid));
reason = IBNAL_REJECT_FATAL;
goto reject;
if (rxmsg.ibm_u.connparams.ibcp_queue_depth != IBNAL_MSG_QUEUE_SIZE) {
CERROR("Can't accept %s: incompatible queue depth %d (%d wanted)\n",
- libcfs_nid2str(rxmsg.ibm_srcnid),
- rxmsg.ibm_u.connparams.ibcp_queue_depth,
+ libcfs_nid2str(rxmsg.ibm_srcnid),
+ rxmsg.ibm_u.connparams.ibcp_queue_depth,
IBNAL_MSG_QUEUE_SIZE);
reason = IBNAL_REJECT_FATAL;
goto reject;
if (rxmsg.ibm_u.connparams.ibcp_max_msg_size > IBNAL_MSG_SIZE) {
CERROR("Can't accept %s: message size %d too big (%d max)\n",
- libcfs_nid2str(rxmsg.ibm_srcnid),
- rxmsg.ibm_u.connparams.ibcp_max_msg_size,
+ libcfs_nid2str(rxmsg.ibm_srcnid),
+ rxmsg.ibm_u.connparams.ibcp_max_msg_size,
IBNAL_MSG_SIZE);
reason = IBNAL_REJECT_FATAL;
goto reject;
}
-
+
if (rxmsg.ibm_u.connparams.ibcp_max_frags > IBNAL_MAX_RDMA_FRAGS) {
CERROR("Can't accept %s: max frags %d too big (%d max)\n",
- libcfs_nid2str(rxmsg.ibm_srcnid),
- rxmsg.ibm_u.connparams.ibcp_max_frags,
+ libcfs_nid2str(rxmsg.ibm_srcnid),
+ rxmsg.ibm_u.connparams.ibcp_max_frags,
IBNAL_MAX_RDMA_FRAGS);
reason = IBNAL_REJECT_FATAL;
goto reject;
}
-
+
/* assume 'rxmsg.ibm_srcnid' is a new peer; create */
rc = kibnal_create_peer (&peer, rxmsg.ibm_srcnid);
if (rc != 0) {
write_lock_irqsave(g_lock, flags);
+ if (kibnal_data.kib_listen_handle == NULL) {
+ write_unlock_irqrestore(g_lock, flags);
+
+ CWARN ("Shutdown has started, rejecting connreq from %s\n",
+ libcfs_nid2str(rxmsg.ibm_srcnid));
+ kibnal_peer_decref(peer);
+ reason = IBNAL_REJECT_FATAL;
+ goto reject;
+ }
+
peer2 = kibnal_find_peer_locked(rxmsg.ibm_srcnid);
if (peer2 != NULL) {
- /* tie-break connection race in favour of the higher NID */
+ /* tie-break connection race in favour of the higher NID */
if (peer2->ibp_connecting != 0 &&
rxmsg.ibm_srcnid < kibnal_data.kib_ni->ni_nid) {
write_unlock_irqrestore(g_lock, flags);
CWARN("Conn race %s\n",
- libcfs_nid2str(peer2->ibp_nid));
+ libcfs_nid2str(rxmsg.ibm_srcnid));
kibnal_peer_decref(peer);
reason = IBNAL_REJECT_CONN_RACE;
write_unlock_irqrestore(g_lock, flags);
}
-
+
conn = kibnal_create_conn(cep);
if (conn == NULL) {
CERROR("Can't create conn for %s\n",
reason = IBNAL_REJECT_FATAL;
goto reject;
}
-
+
vvrc = pkey2pkey_index(kibnal_data.kib_hca, cv->cv_port,
cv->cv_path.pkey, &cv->cv_pkey_index);
if (vvrc != vv_return_ok) {
rc = kibnal_post_receives(conn);
if (rc != 0) {
- CERROR("Can't post receives for %s\n",
+ CERROR("Can't post receives for %s\n",
libcfs_nid2str(rxmsg.ibm_srcnid));
reason = IBNAL_REJECT_FATAL;
goto reject;
reason = IBNAL_REJECT_FATAL;
goto reject;
}
-
+
memset(&reply, 0, sizeof(reply));
reply.qpn = cv->cv_local_qpn;
reply.qkey = IBNAL_QKEY;
reply.failover_accepted = IBNAL_FAILOVER_ACCEPTED;
reply.rnr_retry_count = cv->cv_rnr_count;
reply.targ_ack_delay = kibnal_data.kib_hca_attrs.ack_delay;
-
+
/* setup txmsg... */
memset(&txmsg, 0, sizeof(txmsg));
- kibnal_init_msg(&txmsg, IBNAL_MSG_CONNACK,
+ kibnal_init_msg(&txmsg, IBNAL_MSG_CONNACK,
sizeof(txmsg.ibm_u.connparams));
LASSERT (txmsg.ibm_nob <= cm_REP_priv_data_len);
txmsg.ibm_u.connparams.ibcp_queue_depth = IBNAL_MSG_QUEUE_SIZE;
memcpy(&reply.priv_data, &txmsg, txmsg.ibm_nob);
kibnal_set_conn_state(conn, IBNAL_CONN_PASSIVE_WAIT);
-
+
cmrc = cm_accept(conn->ibc_cep, &reply, NULL,
kibnal_cm_callback, conn);
kibnal_set_conn_state(conn, IBNAL_CONN_INIT);
rc = -EIO;
reason = IBNAL_REJECT_FATAL;
-
+
reject:
CDEBUG(D_NET, "Rejecting connreq from %s\n",
libcfs_nid2str(rxmsg.ibm_srcnid));
cm_request_data_t *cmreq = &data->data.request;
kib_pcreq_t *pcr;
unsigned long flags;
-
+
LASSERT (arg == NULL);
if (data->status != cm_event_conn_request) {
pcr->pcr_cep = cep;
pcr->pcr_cmreq = *cmreq;
-
+
spin_lock_irqsave(&kibnal_data.kib_connd_lock, flags);
list_add_tail(&pcr->pcr_list, &kibnal_data.kib_connd_pcreqs);
wake_up(&kibnal_data.kib_connd_waitq);
-
- spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags);
+spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags);
}
void
-kibnal_active_connect_callback (cm_cep_handle_t cep, cm_conn_data_t *cd,
+kibnal_active_connect_callback (cm_cep_handle_t cep, cm_conn_data_t *cd,
void *arg)
{
/* CAVEAT EMPTOR: tasklet context */
{
static cm_request_data_t cmreq;
static kib_msg_t msg;
-
+
kib_connvars_t *cv = conn->ibc_connvars;
kib_peer_t *peer = conn->ibc_peer;
cm_return_t cmrc;
LASSERT (conn->ibc_state == IBNAL_CONN_ACTIVE_ARP);
memset(&cmreq, 0, sizeof(cmreq));
-
+
cmreq.sid = (__u64)(*kibnal_tunables.kib_service_number);
cmreq.cep_data.ca_guid = kibnal_data.kib_hca_attrs.guid;
cmreq.path_data.subn_local = IBNAL_LOCAL_SUB;
cmreq.path_data.path = cv->cv_path;
-
+
/* setup msg... */
memset(&msg, 0, sizeof(msg));
kibnal_init_msg(&msg, IBNAL_MSG_CONNREQ, sizeof(msg.ibm_u.connparams));
/* ...and copy into cmreq to avoid alignment issues */
memcpy(&cmreq.priv_data, &msg, msg.ibm_nob);
-
+
CDEBUG(D_NET, "Connecting %p to %s\n", conn,
libcfs_nid2str(peer->ibp_nid));
kibnal_conn_addref(conn); /* ++ref for CM callback */
kibnal_set_conn_state(conn, IBNAL_CONN_ACTIVE_CONNECT);
- cmrc = cm_connect(conn->ibc_cep, &cmreq,
+ cmrc = cm_connect(conn->ibc_cep, &cmreq,
kibnal_active_connect_callback, conn);
if (cmrc == cm_stat_success) {
CDEBUG(D_NET, "connection REQ sent to %s\n",
unsigned long flags;
cm_return_t cmrc;
cm_cep_handle_t cep;
-
+
LASSERT (conn->ibc_state == IBNAL_CONN_ACTIVE_CONNECT);
read_lock_irqsave(&kibnal_data.kib_global_lock, flags);
retry = (!list_empty(&peer->ibp_tx_queue) &&
peer->ibp_connecting == 1 &&
peer->ibp_accepting == 0);
-
+
read_unlock_irqrestore(&kibnal_data.kib_global_lock, flags);
if (!retry) {
if (msg.ibm_u.connparams.ibcp_queue_depth != IBNAL_MSG_QUEUE_SIZE) {
CERROR("%s has incompatible queue depth %d(%d wanted)\n",
- libcfs_nid2str(peer->ibp_nid),
+ libcfs_nid2str(peer->ibp_nid),
msg.ibm_u.connparams.ibcp_queue_depth,
IBNAL_MSG_QUEUE_SIZE);
kibnal_connreq_done(conn, 1, -EPROTO);
return;
}
-
+
if (msg.ibm_u.connparams.ibcp_max_msg_size > IBNAL_MSG_SIZE) {
CERROR("%s max message size %d too big (%d max)\n",
- libcfs_nid2str(peer->ibp_nid),
- msg.ibm_u.connparams.ibcp_max_msg_size,
+ libcfs_nid2str(peer->ibp_nid),
+ msg.ibm_u.connparams.ibcp_max_msg_size,
IBNAL_MSG_SIZE);
kibnal_connreq_done(conn, 1, -EPROTO);
return;
if (msg.ibm_u.connparams.ibcp_max_frags > IBNAL_MAX_RDMA_FRAGS) {
CERROR("%s max frags %d too big (%d max)\n",
libcfs_nid2str(peer->ibp_nid),
- msg.ibm_u.connparams.ibcp_max_frags,
+ msg.ibm_u.connparams.ibcp_max_frags,
IBNAL_MAX_RDMA_FRAGS);
kibnal_connreq_done(conn, 1, -EPROTO);
return;
}
-
+
read_lock_irqsave(&kibnal_data.kib_global_lock, flags);
if (lnet_ptlcompat_matchnid(kibnal_data.kib_ni->ni_nid,
msg.ibm_dstnid) &&
conn->ibc_reserved_credits = IBNAL_MSG_QUEUE_SIZE;
LASSERT (conn->ibc_credits + conn->ibc_reserved_credits
<= IBNAL_RX_MSGS);
-
+
rc = kibnal_post_receives(conn);
if (rc != 0) {
CERROR("Can't post receives for %s\n",
kibnal_connreq_done(conn, 1, rc);
return;
}
-
+
rc = kibnal_set_qp_state(conn, vv_qp_state_rtr);
if (rc != 0) {
kibnal_connreq_done(conn, 1, rc);
return;
}
-
+
rc = kibnal_set_qp_state(conn, vv_qp_state_rts);
if (rc != 0) {
kibnal_connreq_done(conn, 1, rc);
return;
}
-
+
kibnal_set_conn_state(conn, IBNAL_CONN_ACTIVE_RTU);
kibnal_conn_addref(conn); /* ++for CM callback */
-
+
memset(&rtu, 0, sizeof(rtu));
cmrc = cm_accept(conn->ibc_cep, NULL, &rtu,
kibnal_cm_callback, conn);
return;
}
- CERROR("cm_accept %s failed: %d\n",
+ CERROR("cm_accept %s failed: %d\n",
libcfs_nid2str(peer->ibp_nid), cmrc);
/* Back out of RTU: no callback coming */
kibnal_set_conn_state(conn, IBNAL_CONN_ACTIVE_CHECK_REPLY);
CERROR("conn -> %s rejected: fatal error %d\n",
libcfs_nid2str(peer->ibp_nid), why);
- if (why == IBNAL_REJECT_CONN_RACE)
+ if (why == IBNAL_REJECT_CONN_RACE)
kibnal_reconnect(conn, -EALREADY);
else
kibnal_connreq_done(conn, 1, -ECONNREFUSED);
return;
}
-
+
/* Fail unless it's worth retrying with an old proto
* version */
if (!(magic == IBNAL_MSG_MAGIC &&
conn->ibc_version = version;
CWARN ("Connection to %s refused: "
- "retrying with old protocol version 0x%x\n",
+ "retrying with old protocol version 0x%x\n",
libcfs_nid2str(peer->ibp_nid), version);
kibnal_reconnect(conn, -ECONNREFUSED);
return;
- } else if (cv->cv_conndata.data.reject.reason ==
+ } else if (cv->cv_conndata.data.reject.reason ==
cm_rej_code_stale_conn) {
-
- CWARN ("conn -> %s stale: retrying\n",
+
+ CWARN ("conn -> %s stale: retrying\n",
libcfs_nid2str(peer->ibp_nid));
kibnal_reconnect(conn, -ESTALE);
/* NOT REACHED */
}
- CDEBUG(D_NETERROR, "conn -> %s failed: %d\n",
+ CDEBUG(D_NETERROR, "conn -> %s failed: %d\n",
libcfs_nid2str(peer->ibp_nid), cv->cv_conndata.status);
kibnal_connreq_done(conn, 1, -ECONNABORTED);
}
LASSERT (current == kibnal_data.kib_connd);
LASSERT (conn->ibc_state == IBNAL_CONN_ACTIVE_ARP);
LASSERT (peer->ibp_arp_count > 0);
-
+
if (cv->cv_arprc != ibat_stat_ok) {
- CDEBUG(D_NETERROR, "Arp %s @ %u.%u.%u.%u failed: %d\n",
+ CDEBUG(D_NETERROR, "Arp %s @ %u.%u.%u.%u failed: %d\n",
libcfs_nid2str(peer->ibp_nid), HIPQUAD(peer->ibp_ip),
cv->cv_arprc);
goto failed;
vvrc = base_gid2port_num(kibnal_data.kib_hca, &path->sgid,
&cv->cv_port);
if (vvrc != vv_return_ok) {
- CWARN("base_gid2port_num failed for %s @ %u.%u.%u.%u: %d\n",
+ CWARN("base_gid2port_num failed for %s @ %u.%u.%u.%u: %d\n",
libcfs_nid2str(peer->ibp_nid),
HIPQUAD(peer->ibp_ip), vvrc);
goto failed;
vvrc = gid2gid_index(kibnal_data.kib_hca, cv->cv_port,
&path->sgid, &cv->cv_sgid_index);
if (vvrc != vv_return_ok) {
- CWARN("gid2gid_index failed for %s @ %u.%u.%u.%u: %d\n",
+ CWARN("gid2gid_index failed for %s @ %u.%u.%u.%u: %d\n",
libcfs_nid2str(peer->ibp_nid),
HIPQUAD(peer->ibp_ip), vvrc);
goto failed;
vvrc = pkey2pkey_index(kibnal_data.kib_hca, cv->cv_port,
path->pkey, &cv->cv_pkey_index);
if (vvrc != vv_return_ok) {
- CWARN("pkey2pkey_index failed for %s @ %u.%u.%u.%u: %d\n",
- libcfs_nid2str(peer->ibp_nid),
+ CWARN("pkey2pkey_index failed for %s @ %u.%u.%u.%u: %d\n",
+ libcfs_nid2str(peer->ibp_nid),
HIPQUAD(peer->ibp_ip), vvrc);
goto failed;
}
vvrc = port_num2base_gid(kibnal_data.kib_hca, cv->cv_port,
&path->sgid);
if (vvrc != vv_return_ok) {
- CWARN("port_num2base_gid failed for %s @ %u.%u.%u.%u: %d\n",
+ CWARN("port_num2base_gid failed for %s @ %u.%u.%u.%u: %d\n",
libcfs_nid2str(peer->ibp_ip),
HIPQUAD(peer->ibp_ip), vvrc);
goto failed;
vvrc = port_num2base_lid(kibnal_data.kib_hca, cv->cv_port,
&path->slid);
if (vvrc != vv_return_ok) {
- CWARN("port_num2base_lid failed for %s @ %u.%u.%u.%u: %d\n",
- libcfs_nid2str(peer->ibp_ip),
+ CWARN("port_num2base_lid failed for %s @ %u.%u.%u.%u: %d\n",
+ libcfs_nid2str(peer->ibp_ip),
HIPQUAD(peer->ibp_ip), vvrc);
goto failed;
}
peer->ibp_arp_count--;
if (peer->ibp_arp_count == 0) {
/* final ARP attempt failed */
- write_unlock_irqrestore(&kibnal_data.kib_global_lock,
+ write_unlock_irqrestore(&kibnal_data.kib_global_lock,
flags);
- CDEBUG(D_NETERROR, "Arp %s @ %u.%u.%u.%u failed (final attempt)\n",
+ CDEBUG(D_NETERROR, "Arp %s @ %u.%u.%u.%u failed (final attempt)\n",
libcfs_nid2str(peer->ibp_nid), HIPQUAD(peer->ibp_ip));
} else {
/* Retry ARP: ibp_connecting++ so terminating conn
* doesn't end peer's connection attempt */
peer->ibp_connecting++;
- write_unlock_irqrestore(&kibnal_data.kib_global_lock,
+ write_unlock_irqrestore(&kibnal_data.kib_global_lock,
flags);
CDEBUG(D_NETERROR, "Arp %s @ %u.%u.%u.%u failed (%d attempts left)\n",
- libcfs_nid2str(peer->ibp_nid), HIPQUAD(peer->ibp_ip),
+ libcfs_nid2str(peer->ibp_nid), HIPQUAD(peer->ibp_ip),
peer->ibp_arp_count);
-
+
kibnal_schedule_peer_arp(peer);
}
kibnal_connreq_done(conn, 1, -ENETUNREACH);
libcfs_nid2str(peer->ibp_nid), HIPQUAD(peer->ibp_ip), arprc);
else
CDEBUG(D_NET, "Arp %s at %u.%u.%u.%u OK: LID %s PATH %s\n",
- libcfs_nid2str(peer->ibp_nid), HIPQUAD(peer->ibp_ip),
+ libcfs_nid2str(peer->ibp_nid), HIPQUAD(peer->ibp_ip),
(arp_data->mask & IBAT_LID_VALID) == 0 ? "invalid" : "valid",
(arp_data->mask & IBAT_PRI_PATH_VALID) == 0 ? "invalid" : "valid");
conn->ibc_connvars->cv_arprc = arprc;
if (arprc == ibat_stat_ok)
conn->ibc_connvars->cv_arp = *arp_data;
-
+
kibnal_schedule_conn(conn);
kibnal_conn_decref(conn);
}
kibnal_set_conn_state(conn, IBNAL_CONN_ACTIVE_ARP);
- ibatrc = ibat_get_ib_data(htonl(peer->ibp_ip), INADDR_ANY,
+ ibatrc = ibat_get_ib_data(htonl(peer->ibp_ip), INADDR_ANY,
ibat_paths_primary,
- &conn->ibc_connvars->cv_arp,
+ &conn->ibc_connvars->cv_arp,
kibnal_arp_callback, conn, 0);
CDEBUG(D_NET,"ibatrc %d\n", ibatrc);
switch (ibatrc) {
default:
LBUG();
-
+
case ibat_stat_pending:
/* NB callback has my ref on conn */
break;
-
+
case ibat_stat_ok:
case ibat_stat_error:
case ibat_stat_timeout:
/* Handle timeout by closing the whole connection. We
* can only be sure RDMA activity has ceased once the
* QP has been modified. */
-
+
kibnal_conn_addref(conn); /* 1 ref for me... */
read_unlock_irqrestore(&kibnal_data.kib_global_lock,
kibnal_disconnect_conn (kib_conn_t *conn)
{
static cm_drequest_data_t dreq; /* just for the space */
-
+
cm_return_t cmrc;
unsigned long flags;
LASSERT (!in_interrupt());
LASSERT (current == kibnal_data.kib_connd);
-
+
write_lock_irqsave(&kibnal_data.kib_global_lock, flags);
if (conn->ibc_disconnect) {
kibnal_conn_disconnected(conn);
return;
}
-
+
LASSERT (conn->ibc_state == IBNAL_CONN_DISCONNECT1);
/* active disconnect */
if (!conn->ibc_disconnect) /* CM callback will never happen now */
kibnal_conn_decref(conn);
-
+
LASSERT (atomic_read(&conn->ibc_refcount) > 0);
LASSERT (conn->ibc_state == IBNAL_CONN_DISCONNECT1);
int dropped_lock;
int peer_index = 0;
unsigned long deadline = jiffies;
-
+
cfs_daemonize ("kibnal_connd");
cfs_block_allsigs ();
conn = list_entry (kibnal_data.kib_connd_zombies.next,
kib_conn_t, ibc_list);
list_del (&conn->ibc_list);
-
+
spin_unlock_irqrestore (&kibnal_data.kib_connd_lock, flags);
dropped_lock = 1;
pcr = list_entry(kibnal_data.kib_connd_pcreqs.next,
kib_pcreq_t, pcr_list);
list_del(&pcr->pcr_list);
-
+
spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags);
dropped_lock = 1;
spin_lock_irqsave(&kibnal_data.kib_connd_lock, flags);
}
-
+
if (!list_empty (&kibnal_data.kib_connd_peers)) {
peer = list_entry (kibnal_data.kib_connd_peers.next,
kib_peer_t, ibp_connd_list);
-
+
list_del_init (&peer->ibp_connd_list);
spin_unlock_irqrestore (&kibnal_data.kib_connd_lock, flags);
dropped_lock = 1;
conn = list_entry (kibnal_data.kib_connd_conns.next,
kib_conn_t, ibc_list);
list_del (&conn->ibc_list);
-
+
spin_unlock_irqrestore (&kibnal_data.kib_connd_lock, flags);
dropped_lock = 1;
switch (conn->ibc_state) {
default:
LBUG();
-
+
case IBNAL_CONN_ACTIVE_ARP:
kibnal_arp_done(conn);
break;
const int n = 4;
const int p = 1;
int chunk = kibnal_data.kib_peer_hash_size;
-
+
spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags);
dropped_lock = 1;
* interval. */
if (*kibnal_tunables.kib_timeout > n * p)
- chunk = (chunk * n * p) /
+ chunk = (chunk * n * p) /
*kibnal_tunables.kib_timeout;
if (chunk == 0)
chunk = 1;
for (i = 0; i < chunk; i++) {
kibnal_check_conns (peer_index);
- peer_index = (peer_index + 1) %
+ peer_index = (peer_index + 1) %
kibnal_data.kib_peer_hash_size;
}
if (dropped_lock)
continue;
-
+
/* Nothing to do for 'timeout' */
set_current_state (TASK_INTERRUPTIBLE);
add_wait_queue (&kibnal_data.kib_connd_waitq, &wait);
return (0);
}
-void
+void
kibnal_async_callback(vv_event_record_t ev)
{
- CERROR("type: %d, port: %d, data: "LPX64"\n",
+ CERROR("type: %d, port: %d, data: "LPX64"\n",
ev.event_type, ev.port_num, ev.type.data);
}
our_cond_resched();
busy_loops = 0;
-
+
spin_lock_irqsave(&kibnal_data.kib_sched_lock, flags);
}
kibnal_data.kib_checking_cq = 1;
/* Assume I'll exhaust the CQ */
kibnal_data.kib_ready = 0;
- spin_unlock_irqrestore(&kibnal_data.kib_sched_lock,
+ spin_unlock_irqrestore(&kibnal_data.kib_sched_lock,
flags);
-
- vvrc = vv_poll_for_completion(kibnal_data.kib_hca,
+
+ vvrc = vv_poll_for_completion(kibnal_data.kib_hca,
kibnal_data.kib_cq, &wc);
if (vvrc == vv_return_err_cq_empty) {
vvrc2 = vv_request_completion_notification(
- kibnal_data.kib_hca,
- kibnal_data.kib_cq,
+ kibnal_data.kib_hca,
+ kibnal_data.kib_cq,
vv_next_solicit_unsolicit_event);
LASSERT (vvrc2 == vv_return_ok);
}
* I give a scheduler on another CPU a chance
* to get the final SEND completion, so the tx
* descriptor can get freed as I inspect it. */
- CDEBUG(D_NETERROR, "RDMA failed: %d\n",
+ CDEBUG(D_NETERROR, "RDMA failed: %d\n",
wc.completion_status);
break;
default:
LBUG();
}
-
+
spin_lock_irqsave(&kibnal_data.kib_sched_lock, flags);
continue;
}