b=13490,i=maxim:

-   fix credit flow deadlock in uptllnd.

diff --git a/lnet/ChangeLog b/lnet/ChangeLog
index b331348..b276ad9 100644 (file)
--- a/lnet/ChangeLog
+++ b/lnet/ChangeLog
@@ -18,6 +18,10 @@ Description:
 Details    : 
 
 Severity   : normal
 Details    : 
 
 Severity   : normal

+Bugzilla   : 13490
+Description: fix credit flow deadlock in uptllnd
+
+Severity   : normal

 Bugzilla   : 16308
 Description: finalize network operation in reasonable time
 Details    : conf-sanity test_32a couldn't stop ost and mds because it
 Bugzilla   : 16308
 Description: finalize network operation in reasonable time
 Details    : conf-sanity test_32a couldn't stop ost and mds because it

diff --git a/lnet/ulnds/ptllnd/ptllnd.c b/lnet/ulnds/ptllnd/ptllnd.c
index 4bddc38..3b504e7 100644 (file)
--- a/lnet/ulnds/ptllnd/ptllnd.c
+++ b/lnet/ulnds/ptllnd/ptllnd.c
@@ -44,13 +44,13 @@ lnd_t               the_ptllnd = {
         .lnd_type       = PTLLND,
         .lnd_startup    = ptllnd_startup,
         .lnd_shutdown   = ptllnd_shutdown,
         .lnd_type       = PTLLND,
         .lnd_startup    = ptllnd_startup,
         .lnd_shutdown   = ptllnd_shutdown,

-       .lnd_ctl        = ptllnd_ctl,
+        .lnd_ctl        = ptllnd_ctl,

         .lnd_send       = ptllnd_send,
         .lnd_recv       = ptllnd_recv,
         .lnd_eager_recv = ptllnd_eager_recv,
         .lnd_notify     = ptllnd_notify,
         .lnd_wait       = ptllnd_wait,
         .lnd_send       = ptllnd_send,
         .lnd_recv       = ptllnd_recv,
         .lnd_eager_recv = ptllnd_eager_recv,
         .lnd_notify     = ptllnd_notify,
         .lnd_wait       = ptllnd_wait,

-       .lnd_setasync   = ptllnd_setasync,
+        .lnd_setasync   = ptllnd_setasync,

 };
 
 static int ptllnd_ni_count = 0;
 };
 
 static int ptllnd_ni_count = 0;


@@ -61,112 +61,112 @@ static struct list_head ptllnd_history_list;
 void
 ptllnd_history_fini(void)
 {
 void
 ptllnd_history_fini(void)
 {

-       ptllnd_he_t *he;
-
-       while (!list_empty(&ptllnd_idle_history)) {
-               he = list_entry(ptllnd_idle_history.next,
-                               ptllnd_he_t, he_list);
-               
-               list_del(&he->he_list);
-               LIBCFS_FREE(he, sizeof(*he));
-       }
-       
-       while (!list_empty(&ptllnd_history_list)) {
-               he = list_entry(ptllnd_history_list.next,
-                               ptllnd_he_t, he_list);
-               
-               list_del(&he->he_list);
-               LIBCFS_FREE(he, sizeof(*he));
-       }
+        ptllnd_he_t *he;
+
+        while (!list_empty(&ptllnd_idle_history)) {
+                he = list_entry(ptllnd_idle_history.next,
+                                ptllnd_he_t, he_list);
+
+                list_del(&he->he_list);
+                LIBCFS_FREE(he, sizeof(*he));
+        }
+
+        while (!list_empty(&ptllnd_history_list)) {
+                he = list_entry(ptllnd_history_list.next,
+                                ptllnd_he_t, he_list);
+
+                list_del(&he->he_list);
+                LIBCFS_FREE(he, sizeof(*he));
+        }

 }
 
 int
 ptllnd_history_init(void)
 {
 }
 
 int
 ptllnd_history_init(void)
 {

-       int          i;
-       ptllnd_he_t *he;
-       int          n;
-       int          rc;
-       
-       CFS_INIT_LIST_HEAD(&ptllnd_idle_history);
-       CFS_INIT_LIST_HEAD(&ptllnd_history_list);
-       
-       rc = ptllnd_parse_int_tunable(&n, "PTLLND_HISTORY", 0);
-       if (rc != 0)
-               return rc;
-       
-       for (i = 0; i < n; i++) {
-               LIBCFS_ALLOC(he, sizeof(*he));
-               if (he == NULL) {
-                       ptllnd_history_fini();
-                       return -ENOMEM;
-               }
-               
-               list_add(&he->he_list, &ptllnd_idle_history);
-       }
-
-       PTLLND_HISTORY("Init");
-
-       return 0;
+        int          i;
+        ptllnd_he_t *he;
+        int          n;
+        int          rc;
+
+        CFS_INIT_LIST_HEAD(&ptllnd_idle_history);
+        CFS_INIT_LIST_HEAD(&ptllnd_history_list);
+
+        rc = ptllnd_parse_int_tunable(&n, "PTLLND_HISTORY", 0);
+        if (rc != 0)
+                return rc;
+
+        for (i = 0; i < n; i++) {
+                LIBCFS_ALLOC(he, sizeof(*he));
+                if (he == NULL) {
+                        ptllnd_history_fini();
+                        return -ENOMEM;
+                }
+
+                list_add(&he->he_list, &ptllnd_idle_history);
+        }
+
+        PTLLND_HISTORY("Init");
+
+        return 0;

 }
 
 void
 ptllnd_history(const char *fn, const char *file, const int line,
 }
 
 void
 ptllnd_history(const char *fn, const char *file, const int line,

-              const char *fmt, ...)
+               const char *fmt, ...)

 {
 {

-       static int     seq;
-       
+        static int     seq;
+

         va_list        ap;
         va_list        ap;

-       ptllnd_he_t   *he;
-       
-       if (!list_empty(&ptllnd_idle_history)) {
-               he = list_entry(ptllnd_idle_history.next,
-                               ptllnd_he_t, he_list);
-       } else if (!list_empty(&ptllnd_history_list)) {
-               he = list_entry(ptllnd_history_list.next,
-                               ptllnd_he_t, he_list);
-       } else {
-               return;
-       }
-
-       list_del(&he->he_list);
-       list_add_tail(&he->he_list, &ptllnd_history_list);
-
-       he->he_seq = seq++;
-       he->he_fn = fn;
-       he->he_file = file;
-       he->he_line = line;
-       gettimeofday(&he->he_time, NULL);
-       
-       va_start(ap, fmt);
-       vsnprintf(he->he_msg, sizeof(he->he_msg), fmt, ap);
-       va_end(ap);
+        ptllnd_he_t   *he;
+
+        if (!list_empty(&ptllnd_idle_history)) {
+                he = list_entry(ptllnd_idle_history.next,
+                                ptllnd_he_t, he_list);
+        } else if (!list_empty(&ptllnd_history_list)) {
+                he = list_entry(ptllnd_history_list.next,
+                                ptllnd_he_t, he_list);
+        } else {
+                return;
+        }
+
+        list_del(&he->he_list);
+        list_add_tail(&he->he_list, &ptllnd_history_list);
+
+        he->he_seq = seq++;
+        he->he_fn = fn;
+        he->he_file = file;
+        he->he_line = line;
+        gettimeofday(&he->he_time, NULL);
+
+        va_start(ap, fmt);
+        vsnprintf(he->he_msg, sizeof(he->he_msg), fmt, ap);
+        va_end(ap);

 }
 
 void
 ptllnd_dump_history(void)
 {
 }
 
 void
 ptllnd_dump_history(void)
 {

-       ptllnd_he_t    *he;
+        ptllnd_he_t    *he;

 
 

-       PTLLND_HISTORY("dumping...");
-       
-       while (!list_empty(&ptllnd_history_list)) {
-               he = list_entry(ptllnd_history_list.next,
-                               ptllnd_he_t, he_list);
+        PTLLND_HISTORY("dumping...");

 
 

-               list_del(&he->he_list);
-               
-               CDEBUG(D_WARNING, "%d %d.%06d (%s:%d:%s()) %s\n", he->he_seq,
-                      (int)he->he_time.tv_sec, (int)he->he_time.tv_usec,
-                      he->he_file, he->he_line, he->he_fn, he->he_msg);
+        while (!list_empty(&ptllnd_history_list)) {
+                he = list_entry(ptllnd_history_list.next,
+                                ptllnd_he_t, he_list);

 
 

-               list_add_tail(&he->he_list, &ptllnd_idle_history);
-       }
+                list_del(&he->he_list);

 
 

-       PTLLND_HISTORY("complete");
+                CDEBUG(D_WARNING, "%d %d.%06d (%s:%d:%s()) %s\n", he->he_seq,
+                       (int)he->he_time.tv_sec, (int)he->he_time.tv_usec,
+                       he->he_file, he->he_line, he->he_fn, he->he_msg);
+
+                list_add_tail(&he->he_list, &ptllnd_idle_history);
+        }
+
+        PTLLND_HISTORY("complete");

 }
 
 }
 

-void 
+void

 ptllnd_assert_wire_constants (void)
 {
         /* Wire protocol assertions generated by 'wirecheck'
 ptllnd_assert_wire_constants (void)
 {
         /* Wire protocol assertions generated by 'wirecheck'


@@ -272,10 +272,10 @@ ptllnd_get_tunables(lnet_ni_t *ni)
         int          rc;
         int          temp;
 
         int          rc;
         int          temp;
 

-       /*  Other tunable defaults depend on this */
-       rc = ptllnd_parse_int_tunable(&plni->plni_debug, "PTLLND_DEBUG", 0);
-       if (rc != 0)
-               return rc;
+        /*  Other tunable defaults depend on this */
+        rc = ptllnd_parse_int_tunable(&plni->plni_debug, "PTLLND_DEBUG", 0);
+        if (rc != 0)
+                return rc;

 
         rc = ptllnd_parse_int_tunable(&plni->plni_portal,
                                       "PTLLND_PORTAL", PTLLND_PORTAL);
 
         rc = ptllnd_parse_int_tunable(&plni->plni_portal,
                                       "PTLLND_PORTAL", PTLLND_PORTAL);


@@ -292,6 +292,11 @@ ptllnd_get_tunables(lnet_ni_t *ni)
                                       "PTLLND_PEERCREDITS", PTLLND_PEERCREDITS);
         if (rc != 0)
                 return rc;
                                       "PTLLND_PEERCREDITS", PTLLND_PEERCREDITS);
         if (rc != 0)
                 return rc;

+        /* kptl_msg_t::ptlm_credits is only a __u8 */
+        if (plni->plni_peer_credits > 255) {
+                CERROR("PTLLND_PEERCREDITS must be <= 255\n");
+                return -EINVAL;
+        }

 
         rc = ptllnd_parse_int_tunable(&max_msg_size,
                                       "PTLLND_MAX_MSG_SIZE",
 
         rc = ptllnd_parse_int_tunable(&max_msg_size,
                                       "PTLLND_MAX_MSG_SIZE",


@@ -320,56 +325,56 @@ ptllnd_get_tunables(lnet_ni_t *ni)
         if (rc != 0)
                 return rc;
 
         if (rc != 0)
                 return rc;
 

-       rc = ptllnd_parse_int_tunable(&plni->plni_checksum,
-                                     "PTLLND_CHECKSUM", 0);
-       if (rc != 0)
-               return rc;
-
-       rc = ptllnd_parse_int_tunable(&plni->plni_max_tx_history,
-                                     "PTLLND_TX_HISTORY",
-                                     plni->plni_debug ? 1024 : 0);
-       if (rc != 0)
-               return rc;
-
-       rc = ptllnd_parse_int_tunable(&plni->plni_abort_on_protocol_mismatch,
-                                     "PTLLND_ABORT_ON_PROTOCOL_MISMATCH", 1);
-       if (rc != 0)
-               return rc;
-
-       rc = ptllnd_parse_int_tunable(&plni->plni_abort_on_nak,
-                                     "PTLLND_ABORT_ON_NAK", 0);
-       if (rc != 0)
-               return rc;
-
-       rc = ptllnd_parse_int_tunable(&plni->plni_dump_on_nak,
-                                     "PTLLND_DUMP_ON_NAK", plni->plni_debug);
-       if (rc != 0)
-               return rc;
-
-       rc = ptllnd_parse_int_tunable(&plni->plni_watchdog_interval,
-                                     "PTLLND_WATCHDOG_INTERVAL", 1);
-       if (rc != 0)
-               return rc;
-       if (plni->plni_watchdog_interval <= 0)
-               plni->plni_watchdog_interval = 1;
-
-       rc = ptllnd_parse_int_tunable(&plni->plni_timeout,
-                                     "PTLLND_TIMEOUT", 50);
-       if (rc != 0)
-               return rc;
-
-       rc = ptllnd_parse_int_tunable(&plni->plni_long_wait,
-                                     "PTLLND_LONG_WAIT",
-                                     plni->plni_debug ? 5 : plni->plni_timeout);
-       if (rc != 0)
-               return rc;
-       plni->plni_long_wait *= 1000;           /* convert to mS */
+        rc = ptllnd_parse_int_tunable(&plni->plni_checksum,
+                                      "PTLLND_CHECKSUM", 0);
+        if (rc != 0)
+                return rc;
+
+        rc = ptllnd_parse_int_tunable(&plni->plni_max_tx_history,
+                                      "PTLLND_TX_HISTORY",
+                                      plni->plni_debug ? 1024 : 0);
+        if (rc != 0)
+                return rc;
+
+        rc = ptllnd_parse_int_tunable(&plni->plni_abort_on_protocol_mismatch,
+                                      "PTLLND_ABORT_ON_PROTOCOL_MISMATCH", 1);
+        if (rc != 0)
+                return rc;
+
+        rc = ptllnd_parse_int_tunable(&plni->plni_abort_on_nak,
+                                      "PTLLND_ABORT_ON_NAK", 0);
+        if (rc != 0)
+                return rc;
+
+        rc = ptllnd_parse_int_tunable(&plni->plni_dump_on_nak,
+                                      "PTLLND_DUMP_ON_NAK", plni->plni_debug);
+        if (rc != 0)
+                return rc;
+
+        rc = ptllnd_parse_int_tunable(&plni->plni_watchdog_interval,
+                                      "PTLLND_WATCHDOG_INTERVAL", 1);
+        if (rc != 0)
+                return rc;
+        if (plni->plni_watchdog_interval <= 0)
+                plni->plni_watchdog_interval = 1;
+
+        rc = ptllnd_parse_int_tunable(&plni->plni_timeout,
+                                      "PTLLND_TIMEOUT", 50);
+        if (rc != 0)
+                return rc;
+
+        rc = ptllnd_parse_int_tunable(&plni->plni_long_wait,
+                                      "PTLLND_LONG_WAIT",
+                                      plni->plni_debug ? 5 : plni->plni_timeout);
+        if (rc != 0)
+                return rc;
+        plni->plni_long_wait *= 1000;           /* convert to mS */

 
         plni->plni_max_msg_size = max_msg_size & ~7;
         if (plni->plni_max_msg_size < PTLLND_MIN_BUFFER_SIZE)
                 plni->plni_max_msg_size = PTLLND_MIN_BUFFER_SIZE;
 
         plni->plni_max_msg_size = max_msg_size & ~7;
         if (plni->plni_max_msg_size < PTLLND_MIN_BUFFER_SIZE)
                 plni->plni_max_msg_size = PTLLND_MIN_BUFFER_SIZE;

-       CLASSERT ((PTLLND_MIN_BUFFER_SIZE & 7) == 0);
-       CLASSERT (sizeof(kptl_msg_t) <= PTLLND_MIN_BUFFER_SIZE);
+        CLASSERT ((PTLLND_MIN_BUFFER_SIZE & 7) == 0);
+        CLASSERT (sizeof(kptl_msg_t) <= PTLLND_MIN_BUFFER_SIZE);

 
         plni->plni_buffer_size = plni->plni_max_msg_size * msgs_per_buffer;
 
 
         plni->plni_buffer_size = plni->plni_max_msg_size * msgs_per_buffer;
 


@@ -441,9 +446,9 @@ ptllnd_size_buffers (lnet_ni_t *ni, int delta)
         CDEBUG(D_NET, "nposted_buffers = %d (before)\n",plni->plni_nposted_buffers);
         CDEBUG(D_NET, "nbuffers = %d (before)\n",plni->plni_nbuffers);
 
         CDEBUG(D_NET, "nposted_buffers = %d (before)\n",plni->plni_nposted_buffers);
         CDEBUG(D_NET, "nbuffers = %d (before)\n",plni->plni_nbuffers);
 

-       plni->plni_nmsgs += delta;
-       LASSERT(plni->plni_nmsgs >= 0);
-       
+        plni->plni_nmsgs += delta;
+        LASSERT(plni->plni_nmsgs >= 0);
+

         nmsgs = plni->plni_nmsgs + plni->plni_msgs_spare;
 
         nbufs = (nmsgs * plni->plni_max_msg_size + plni->plni_buffer_size - 1) /
         nmsgs = plni->plni_nmsgs + plni->plni_msgs_spare;
 
         nbufs = (nmsgs * plni->plni_max_msg_size + plni->plni_buffer_size - 1) /


@@ -490,22 +495,22 @@ ptllnd_destroy_buffers (lnet_ni_t *ni)
 
                 LASSERT (plni->plni_nbuffers > 0);
                 if (buf->plb_posted) {
 
                 LASSERT (plni->plni_nbuffers > 0);
                 if (buf->plb_posted) {

-                       time_t   start = cfs_time_current_sec();
-                       int      w = plni->plni_long_wait;
+                        time_t   start = cfs_time_current_sec();
+                        int      w = plni->plni_long_wait;

 
                         LASSERT (plni->plni_nposted_buffers > 0);
 
 #ifdef LUSTRE_PORTALS_UNLINK_SEMANTICS
                         (void) PtlMDUnlink(buf->plb_md);
 
 
                         LASSERT (plni->plni_nposted_buffers > 0);
 
 #ifdef LUSTRE_PORTALS_UNLINK_SEMANTICS
                         (void) PtlMDUnlink(buf->plb_md);
 

-                       while (buf->plb_posted) {
-                               if (w > 0 && cfs_time_current_sec() > start + w/1000) {
-                                       CWARN("Waited %ds to unlink buffer\n",
-                                             (int)(cfs_time_current_sec() - start));
-                                       w *= 2;
-                               }
-                               ptllnd_wait(ni, w);
-                       }
+                        while (buf->plb_posted) {
+                                if (w > 0 && cfs_time_current_sec() > start + w/1000) {
+                                        CWARN("Waited %ds to unlink buffer\n",
+                                              (int)(cfs_time_current_sec() - start));
+                                        w *= 2;
+                                }
+                                ptllnd_wait(ni, w);
+                        }

 #else
                         while (buf->plb_posted) {
                                 rc = PtlMDUnlink(buf->plb_md);
 #else
                         while (buf->plb_posted) {
                                 rc = PtlMDUnlink(buf->plb_md);


@@ -515,12 +520,12 @@ ptllnd_destroy_buffers (lnet_ni_t *ni)
                                         break;
                                 }
                                 LASSERT (rc == PTL_MD_IN_USE);
                                         break;
                                 }
                                 LASSERT (rc == PTL_MD_IN_USE);

-                               if (w > 0 && cfs_time_current_sec() > start + w/1000) {
-                                       CWARN("Waited %ds to unlink buffer\n",
-                                             cfs_time_current_sec() - start);
-                                       w *= 2;
-                               }
-                               ptllnd_wait(ni, w);
+                                if (w > 0 && cfs_time_current_sec() > start + w/1000) {
+                                        CWARN("Waited %ds to unlink buffer\n",
+                                              cfs_time_current_sec() - start);
+                                        w *= 2;
+                                }
+                                ptllnd_wait(ni, w);

                         }
 #endif
                 }
                         }
 #endif
                 }


@@ -590,14 +595,14 @@ ptllnd_close_peers (lnet_ni_t *ni)
 int
 ptllnd_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
 {
 int
 ptllnd_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
 {

-       switch (cmd) {
-       case IOC_LIBCFS_DEBUG_PEER:
-               ptllnd_dump_debug(ni, *((lnet_process_id_t *)arg));
-               return 0;
-               
-       default:
-               return -EINVAL;
-       }
+        switch (cmd) {
+        case IOC_LIBCFS_DEBUG_PEER:
+                ptllnd_dump_debug(ni, *((lnet_process_id_t *)arg));
+                return 0;
+
+        default:
+                return -EINVAL;
+        }

 }
 
 __u64
 }
 
 __u64


@@ -615,25 +620,25 @@ ptllnd_shutdown (lnet_ni_t *ni)
 {
         ptllnd_ni_t *plni = ni->ni_data;
         int          rc;
 {
         ptllnd_ni_t *plni = ni->ni_data;
         int          rc;

-       time_t       start = cfs_time_current_sec();
-       int          w = plni->plni_long_wait;
+        time_t       start = cfs_time_current_sec();
+        int          w = plni->plni_long_wait;

 
         LASSERT (ptllnd_ni_count == 1);
 
         LASSERT (ptllnd_ni_count == 1);

-       plni->plni_max_tx_history = 0;
+        plni->plni_max_tx_history = 0;

 
 

-       ptllnd_cull_tx_history(plni);
+        ptllnd_cull_tx_history(plni);

 
         ptllnd_close_peers(ni);
         ptllnd_destroy_buffers(ni);
 
         while (plni->plni_npeers > 0) {
 
         ptllnd_close_peers(ni);
         ptllnd_destroy_buffers(ni);
 
         while (plni->plni_npeers > 0) {

-               if (w > 0 && cfs_time_current_sec() > start + w/1000) {
-                       CWARN("Waited %ds for peers to shutdown\n",
-                             (int)(cfs_time_current_sec() - start));
-                       w *= 2;
-               }
+                if (w > 0 && cfs_time_current_sec() > start + w/1000) {
+                        CWARN("Waited %ds for peers to shutdown\n",
+                              (int)(cfs_time_current_sec() - start));
+                        w *= 2;
+                }

                 ptllnd_wait(ni, w);
                 ptllnd_wait(ni, w);

-       }
+        }

 
         LASSERT (plni->plni_ntxs == 0);
         LASSERT (plni->plni_nrxs == 0);
 
         LASSERT (plni->plni_ntxs == 0);
         LASSERT (plni->plni_nrxs == 0);


@@ -655,9 +660,9 @@ ptllnd_startup (lnet_ni_t *ni)
         ptllnd_ni_t *plni;
         int          rc;
 
         ptllnd_ni_t *plni;
         int          rc;
 

-       /* could get limits from portals I guess... */
-       ni->ni_maxtxcredits =
-       ni->ni_peertxcredits = 1000;
+        /* could get limits from portals I guess... */
+        ni->ni_maxtxcredits =
+        ni->ni_peertxcredits = 1000;

 
         if (ptllnd_ni_count != 0) {
                 CERROR("Can't have > 1 instance of ptllnd\n");
 
         if (ptllnd_ni_count != 0) {
                 CERROR("Can't have > 1 instance of ptllnd\n");


@@ -666,12 +671,12 @@ ptllnd_startup (lnet_ni_t *ni)
 
         ptllnd_ni_count++;
 
 
         ptllnd_ni_count++;
 

-       rc = ptllnd_history_init();
-       if (rc != 0) {
-               CERROR("Can't init history\n");
-               goto failed0;
-       }
-       
+        rc = ptllnd_history_init();
+        if (rc != 0) {
+                CERROR("Can't init history\n");
+                goto failed0;
+        }
+

         LIBCFS_ALLOC(plni, sizeof(*plni));
         if (plni == NULL) {
                 CERROR("Can't allocate ptllnd state\n");
         LIBCFS_ALLOC(plni, sizeof(*plni));
         if (plni == NULL) {
                 CERROR("Can't allocate ptllnd state\n");


@@ -684,9 +689,9 @@ ptllnd_startup (lnet_ni_t *ni)
         plni->plni_stamp = ptllnd_get_timestamp();
         plni->plni_nrxs = 0;
         plni->plni_ntxs = 0;
         plni->plni_stamp = ptllnd_get_timestamp();
         plni->plni_nrxs = 0;
         plni->plni_ntxs = 0;

-       plni->plni_ntx_history = 0;
-       plni->plni_watchdog_peeridx = 0;
-       plni->plni_watchdog_nextt = cfs_time_current_sec();
+        plni->plni_ntx_history = 0;
+        plni->plni_watchdog_peeridx = 0;
+        plni->plni_watchdog_nextt = cfs_time_current_sec();

         CFS_INIT_LIST_HEAD(&plni->plni_zombie_txs);
         CFS_INIT_LIST_HEAD(&plni->plni_tx_history);
 
         CFS_INIT_LIST_HEAD(&plni->plni_zombie_txs);
         CFS_INIT_LIST_HEAD(&plni->plni_tx_history);
 


@@ -713,7 +718,7 @@ ptllnd_startup (lnet_ni_t *ni)
                        NULL, NULL, &plni->plni_nih);
         if (rc != PTL_OK && rc != PTL_IFACE_DUP) {
                 CERROR("PtlNIInit failed: %s(%d)\n",
                        NULL, NULL, &plni->plni_nih);
         if (rc != PTL_OK && rc != PTL_IFACE_DUP) {
                 CERROR("PtlNIInit failed: %s(%d)\n",

-                      ptllnd_errtype2str(rc), rc);
+                       ptllnd_errtype2str(rc), rc);

                 rc = -ENODEV;
                 goto failed2;
         }
                 rc = -ENODEV;
                 goto failed2;
         }


@@ -722,7 +727,7 @@ ptllnd_startup (lnet_ni_t *ni)
                         PTL_EQ_HANDLER_NONE, &plni->plni_eqh);
         if (rc != PTL_OK) {
                 CERROR("PtlEQAlloc failed: %s(%d)\n",
                         PTL_EQ_HANDLER_NONE, &plni->plni_eqh);
         if (rc != PTL_OK) {
                 CERROR("PtlEQAlloc failed: %s(%d)\n",

-                      ptllnd_errtype2str(rc), rc);
+                       ptllnd_errtype2str(rc), rc);

                 rc = -ENODEV;
                 goto failed3;
         }
                 rc = -ENODEV;
                 goto failed3;
         }


@@ -730,10 +735,10 @@ ptllnd_startup (lnet_ni_t *ni)
         /*
          * Fetch the Portals NID
          */
         /*
          * Fetch the Portals NID
          */

-       rc = PtlGetId(plni->plni_nih, &plni->plni_portals_id);
+        rc = PtlGetId(plni->plni_nih, &plni->plni_portals_id);

         if (rc != PTL_OK) {
                 CERROR ("PtlGetID failed : %s(%d)\n",
         if (rc != PTL_OK) {
                 CERROR ("PtlGetID failed : %s(%d)\n",

-                       ptllnd_errtype2str(rc), rc);
+                        ptllnd_errtype2str(rc), rc);

                 rc = -EINVAL;
                 goto failed4;
         }
                 rc = -EINVAL;
                 goto failed4;
         }


@@ -753,7 +758,7 @@ ptllnd_startup (lnet_ni_t *ni)
         if (rc != 0)
                 goto failed4;
 
         if (rc != 0)
                 goto failed4;
 

-       return 0;
+        return 0;

 
  failed4:
         ptllnd_destroy_buffers(ni);
 
  failed4:
         ptllnd_destroy_buffers(ni);


@@ -765,7 +770,7 @@ ptllnd_startup (lnet_ni_t *ni)
  failed1:
         LIBCFS_FREE(plni, sizeof(*plni));
  failed0:
  failed1:
         LIBCFS_FREE(plni, sizeof(*plni));
  failed0:

-       ptllnd_history_fini();
+        ptllnd_history_fini();

         ptllnd_ni_count--;
         CDEBUG(D_NET, "<<< rc=%d\n",rc);
         return rc;
         ptllnd_ni_count--;
         CDEBUG(D_NET, "<<< rc=%d\n",rc);
         return rc;

diff --git a/lnet/ulnds/ptllnd/ptllnd.h b/lnet/ulnds/ptllnd/ptllnd.h
index 87697e7..b8198b2 100644 (file)
--- a/lnet/ulnds/ptllnd/ptllnd.h
+++ b/lnet/ulnds/ptllnd/ptllnd.h
@@ -140,11 +140,13 @@ typedef struct
 
         int                        plp_max_msg_size;
         int                        plp_refcount;
 
         int                        plp_max_msg_size;
         int                        plp_refcount;

+        int                        plp_sent_hello:1;

         int                        plp_recvd_hello:1;
         int                        plp_closing:1;
         __u64                      plp_match;
         __u64                      plp_stamp;
         struct list_head           plp_txq;
         int                        plp_recvd_hello:1;
         int                        plp_closing:1;
         __u64                      plp_match;
         __u64                      plp_stamp;
         struct list_head           plp_txq;

+        struct list_head           plp_noopq;

         struct list_head           plp_activeq;
 } ptllnd_peer_t;
 
         struct list_head           plp_activeq;
 } ptllnd_peer_t;
 


@@ -271,13 +273,13 @@ ptllnd_peer_decref (ptllnd_peer_t *peer)
 static inline lnet_nid_t
 ptllnd_ptl2lnetnid(lnet_ni_t *ni, ptl_nid_t portals_nid)
 {
 static inline lnet_nid_t
 ptllnd_ptl2lnetnid(lnet_ni_t *ni, ptl_nid_t portals_nid)
 {

-       return LNET_MKNID(LNET_NIDNET(ni->ni_nid), portals_nid);
+        return LNET_MKNID(LNET_NIDNET(ni->ni_nid), portals_nid);

 }
 
 static inline ptl_nid_t
 ptllnd_lnet2ptlnid(lnet_nid_t lnet_nid)
 {
 }
 
 static inline ptl_nid_t
 ptllnd_lnet2ptlnid(lnet_nid_t lnet_nid)
 {

-       return LNET_NIDADDR(lnet_nid);
+        return LNET_NIDADDR(lnet_nid);

 }
 
 /*
 }
 
 /*

diff --git a/lnet/ulnds/ptllnd/ptllnd_cb.c b/lnet/ulnds/ptllnd/ptllnd_cb.c
index da6d277..c8431df 100644 (file)
--- a/lnet/ulnds/ptllnd/ptllnd_cb.c
+++ b/lnet/ulnds/ptllnd/ptllnd_cb.c
@@ -55,6 +55,8 @@ ptllnd_post_tx(ptllnd_tx_t *tx)
 {
         ptllnd_peer_t  *peer = tx->tx_peer;
 
 {
         ptllnd_peer_t  *peer = tx->tx_peer;
 

+        LASSERT (tx->tx_type != PTLLND_MSG_TYPE_NOOP);
+

         ptllnd_set_tx_deadline(tx);
         list_add_tail(&tx->tx_list, &peer->plp_txq);
         ptllnd_check_sends(peer);
         ptllnd_set_tx_deadline(tx);
         list_add_tail(&tx->tx_list, &peer->plp_txq);
         ptllnd_check_sends(peer);


@@ -67,7 +69,7 @@ ptllnd_ptlid2str(ptl_process_id_t id)
         static int  idx = 0;
 
         char   *str = strs[idx++];
         static int  idx = 0;
 
         char   *str = strs[idx++];

-        
+

         if (idx >= sizeof(strs)/sizeof(strs[0]))
                 idx = 0;
 
         if (idx >= sizeof(strs)/sizeof(strs[0]))
                 idx = 0;
 


@@ -88,6 +90,7 @@ ptllnd_destroy_peer(ptllnd_peer_t *peer)
         LASSERT (peer->plp_closing);
         LASSERT (plni->plni_npeers > 0);
         LASSERT (list_empty(&peer->plp_txq));
         LASSERT (peer->plp_closing);
         LASSERT (plni->plni_npeers > 0);
         LASSERT (list_empty(&peer->plp_txq));

+        LASSERT (list_empty(&peer->plp_noopq));

         LASSERT (list_empty(&peer->plp_activeq));
         plni->plni_npeers--;
         LIBCFS_FREE(peer, sizeof(*peer));
         LASSERT (list_empty(&peer->plp_activeq));
         plni->plni_npeers--;
         LIBCFS_FREE(peer, sizeof(*peer));


@@ -117,14 +120,16 @@ ptllnd_close_peer(ptllnd_peer_t *peer, int error)
         peer->plp_closing = 1;
 
         if (!list_empty(&peer->plp_txq) ||
         peer->plp_closing = 1;
 
         if (!list_empty(&peer->plp_txq) ||

+            !list_empty(&peer->plp_noopq) ||

             !list_empty(&peer->plp_activeq) ||
             error != 0) {
                 CWARN("Closing %s\n", libcfs_id2str(peer->plp_id));
                 if (plni->plni_debug)
                         ptllnd_dump_debug(ni, peer->plp_id);
         }
             !list_empty(&peer->plp_activeq) ||
             error != 0) {
                 CWARN("Closing %s\n", libcfs_id2str(peer->plp_id));
                 if (plni->plni_debug)
                         ptllnd_dump_debug(ni, peer->plp_id);
         }

-        
+

         ptllnd_abort_txs(plni, &peer->plp_txq);
         ptllnd_abort_txs(plni, &peer->plp_txq);

+        ptllnd_abort_txs(plni, &peer->plp_noopq);

         ptllnd_abort_txs(plni, &peer->plp_activeq);
 
         list_del(&peer->plp_list);
         ptllnd_abort_txs(plni, &peer->plp_activeq);
 
         list_del(&peer->plp_list);


@@ -136,16 +141,13 @@ ptllnd_find_peer(lnet_ni_t *ni, lnet_process_id_t id, int create)
 {
         ptllnd_ni_t       *plni = ni->ni_data;
         unsigned int       hash = LNET_NIDADDR(id.nid) % plni->plni_peer_hash_size;
 {
         ptllnd_ni_t       *plni = ni->ni_data;
         unsigned int       hash = LNET_NIDADDR(id.nid) % plni->plni_peer_hash_size;

-        struct list_head  *tmp;

         ptllnd_peer_t     *plp;
         ptllnd_tx_t       *tx;
         int                rc;
 
         LASSERT (LNET_NIDNET(id.nid) == LNET_NIDNET(ni->ni_nid));
 
         ptllnd_peer_t     *plp;
         ptllnd_tx_t       *tx;
         int                rc;
 
         LASSERT (LNET_NIDNET(id.nid) == LNET_NIDNET(ni->ni_nid));
 

-        list_for_each(tmp, &plni->plni_peer_hash[hash]) {
-                plp = list_entry(tmp, ptllnd_peer_t, plp_list);
-
+        list_for_each_entry (plp, &plni->plni_peer_hash[hash], plp_list) {

                 if (plp->plp_id.nid == id.nid &&
                     plp->plp_id.pid == id.pid) {
                         ptllnd_peer_addref(plp);
                 if (plp->plp_id.nid == id.nid &&
                     plp->plp_id.pid == id.pid) {
                         ptllnd_peer_addref(plp);


@@ -184,11 +186,13 @@ ptllnd_find_peer(lnet_ni_t *ni, lnet_process_id_t id, int create)
         plp->plp_extra_lazy_credits = 0;
         plp->plp_match = 0;
         plp->plp_stamp = 0;
         plp->plp_extra_lazy_credits = 0;
         plp->plp_match = 0;
         plp->plp_stamp = 0;

+        plp->plp_sent_hello = 0;

         plp->plp_recvd_hello = 0;
         plp->plp_closing = 0;
         plp->plp_refcount = 1;
         CFS_INIT_LIST_HEAD(&plp->plp_list);
         CFS_INIT_LIST_HEAD(&plp->plp_txq);
         plp->plp_recvd_hello = 0;
         plp->plp_closing = 0;
         plp->plp_refcount = 1;
         CFS_INIT_LIST_HEAD(&plp->plp_list);
         CFS_INIT_LIST_HEAD(&plp->plp_txq);

+        CFS_INIT_LIST_HEAD(&plp->plp_noopq);

         CFS_INIT_LIST_HEAD(&plp->plp_activeq);
 
         ptllnd_peer_addref(plp);
         CFS_INIT_LIST_HEAD(&plp->plp_activeq);
 
         ptllnd_peer_addref(plp);


@@ -221,27 +225,27 @@ ptllnd_count_q(struct list_head *q)
 {
         struct list_head *e;
         int               n = 0;
 {
         struct list_head *e;
         int               n = 0;

-        
+

         list_for_each(e, q) {
                 n++;
         }
         list_for_each(e, q) {
                 n++;
         }

-        
+

         return n;
 }
 
 const char *
         return n;
 }
 
 const char *

-ptllnd_tx_typestr(int type) 
+ptllnd_tx_typestr(int type)

 {
         switch (type) {
         case PTLLND_RDMA_WRITE:
                 return "rdma_write";
 {
         switch (type) {
         case PTLLND_RDMA_WRITE:
                 return "rdma_write";

-                
+

         case PTLLND_RDMA_READ:
                 return "rdma_read";
 
         case PTLLND_MSG_TYPE_PUT:
                 return "put_req";
         case PTLLND_RDMA_READ:
                 return "rdma_read";
 
         case PTLLND_MSG_TYPE_PUT:
                 return "put_req";

-                
+

         case PTLLND_MSG_TYPE_GET:
                 return "get_req";
 
         case PTLLND_MSG_TYPE_GET:
                 return "get_req";
 


@@ -260,13 +264,13 @@ ptllnd_tx_typestr(int type)
 }
 
 void
 }
 
 void

-ptllnd_debug_tx(ptllnd_tx_t *tx) 
+ptllnd_debug_tx(ptllnd_tx_t *tx)

 {
         CDEBUG(D_WARNING, "%s %s b %ld.%06ld/%ld.%06ld"
                " r %ld.%06ld/%ld.%06ld status %d\n",
                ptllnd_tx_typestr(tx->tx_type),
                libcfs_id2str(tx->tx_peer->plp_id),
 {
         CDEBUG(D_WARNING, "%s %s b %ld.%06ld/%ld.%06ld"
                " r %ld.%06ld/%ld.%06ld status %d\n",
                ptllnd_tx_typestr(tx->tx_type),
                libcfs_id2str(tx->tx_peer->plp_id),

-               tx->tx_bulk_posted.tv_sec, tx->tx_bulk_posted.tv_usec, 
+               tx->tx_bulk_posted.tv_sec, tx->tx_bulk_posted.tv_usec,

                tx->tx_bulk_done.tv_sec, tx->tx_bulk_done.tv_usec,
                tx->tx_req_posted.tv_sec, tx->tx_req_posted.tv_usec,
                tx->tx_req_done.tv_sec, tx->tx_req_done.tv_usec,
                tx->tx_bulk_done.tv_sec, tx->tx_bulk_done.tv_usec,
                tx->tx_req_posted.tv_sec, tx->tx_req_posted.tv_usec,
                tx->tx_req_done.tv_sec, tx->tx_req_done.tv_usec,


@@ -277,59 +281,56 @@ void
 ptllnd_debug_peer(lnet_ni_t *ni, lnet_process_id_t id)
 {
         ptllnd_peer_t    *plp = ptllnd_find_peer(ni, id, 0);
 ptllnd_debug_peer(lnet_ni_t *ni, lnet_process_id_t id)
 {
         ptllnd_peer_t    *plp = ptllnd_find_peer(ni, id, 0);

-        struct list_head *tmp;

         ptllnd_ni_t      *plni = ni->ni_data;
         ptllnd_tx_t      *tx;
         ptllnd_ni_t      *plni = ni->ni_data;
         ptllnd_tx_t      *tx;

-        
+

         if (plp == NULL) {
                 CDEBUG(D_WARNING, "No peer %s\n", libcfs_id2str(id));
                 return;
         }
         if (plp == NULL) {
                 CDEBUG(D_WARNING, "No peer %s\n", libcfs_id2str(id));
                 return;
         }

-        
-        CDEBUG(D_WARNING, "%s %s%s [%d] "LPU64".%06d m "LPU64" q %d/%d c %d/%d+%d(%d)\n",
-               libcfs_id2str(id), 
-               plp->plp_recvd_hello ? "H" : "_",
-               plp->plp_closing     ? "C" : "_",
-               plp->plp_refcount,
-               plp->plp_stamp / 1000000, (int)(plp->plp_stamp % 1000000),
-               plp->plp_match,
-               ptllnd_count_q(&plp->plp_txq),
-               ptllnd_count_q(&plp->plp_activeq),
-               plp->plp_credits, plp->plp_outstanding_credits, plp->plp_sent_credits,
-               plni->plni_peer_credits + plp->plp_lazy_credits);
+
+        CWARN("%s %s%s [%d] "LPU64".%06d m "LPU64" q %d/%d/%d c %d/%d+%d(%d)\n",
+              libcfs_id2str(id),
+              plp->plp_recvd_hello ? "H" : "_",
+              plp->plp_closing     ? "C" : "_",
+              plp->plp_refcount,
+              plp->plp_stamp / 1000000, (int)(plp->plp_stamp % 1000000),
+              plp->plp_match,
+              ptllnd_count_q(&plp->plp_txq),
+              ptllnd_count_q(&plp->plp_noopq),
+              ptllnd_count_q(&plp->plp_activeq),
+              plp->plp_credits, plp->plp_outstanding_credits, plp->plp_sent_credits,
+              plni->plni_peer_credits + plp->plp_lazy_credits);

 
         CDEBUG(D_WARNING, "txq:\n");
 
         CDEBUG(D_WARNING, "txq:\n");

-        list_for_each (tmp, &plp->plp_txq) {
-                tx = list_entry(tmp, ptllnd_tx_t, tx_list);
-                
+        list_for_each_entry (tx, &plp->plp_txq, tx_list) {
+                ptllnd_debug_tx(tx);
+        }
+
+        CDEBUG(D_WARNING, "noopq:\n");
+        list_for_each_entry (tx, &plp->plp_noopq, tx_list) {

                 ptllnd_debug_tx(tx);
         }
 
         CDEBUG(D_WARNING, "activeq:\n");
                 ptllnd_debug_tx(tx);
         }
 
         CDEBUG(D_WARNING, "activeq:\n");

-        list_for_each (tmp, &plp->plp_activeq) {
-                tx = list_entry(tmp, ptllnd_tx_t, tx_list);
-                
+        list_for_each_entry (tx, &plp->plp_activeq, tx_list) {

                 ptllnd_debug_tx(tx);
         }
 
         CDEBUG(D_WARNING, "zombies:\n");
                 ptllnd_debug_tx(tx);
         }
 
         CDEBUG(D_WARNING, "zombies:\n");

-        list_for_each (tmp, &plni->plni_zombie_txs) {
-                tx = list_entry(tmp, ptllnd_tx_t, tx_list);
-                
+        list_for_each_entry (tx, &plni->plni_zombie_txs, tx_list) {

                 if (tx->tx_peer->plp_id.nid == id.nid &&
                     tx->tx_peer->plp_id.pid == id.pid)
                         ptllnd_debug_tx(tx);
         }
                 if (tx->tx_peer->plp_id.nid == id.nid &&
                     tx->tx_peer->plp_id.pid == id.pid)
                         ptllnd_debug_tx(tx);
         }

-        
+

         CDEBUG(D_WARNING, "history:\n");
         CDEBUG(D_WARNING, "history:\n");

-        list_for_each (tmp, &plni->plni_tx_history) {
-                tx = list_entry(tmp, ptllnd_tx_t, tx_list);
-                
+        list_for_each_entry (tx, &plni->plni_tx_history, tx_list) {

                 if (tx->tx_peer->plp_id.nid == id.nid &&
                     tx->tx_peer->plp_id.pid == id.pid)
                         ptllnd_debug_tx(tx);
         }
                 if (tx->tx_peer->plp_id.nid == id.nid &&
                     tx->tx_peer->plp_id.pid == id.pid)
                         ptllnd_debug_tx(tx);
         }

-        
+

         ptllnd_peer_decref(plp);
 }
 
         ptllnd_peer_decref(plp);
 }
 


@@ -354,7 +355,7 @@ ptllnd_notify(lnet_ni_t *ni, lnet_nid_t nid, int alive)
 
         id.nid = nid;
         id.pid = LUSTRE_SRV_LNET_PID;
 
         id.nid = nid;
         id.pid = LUSTRE_SRV_LNET_PID;

-        
+

         peer = ptllnd_find_peer(ni, id, 1);
         if (peer == NULL)
                 return;
         peer = ptllnd_find_peer(ni, id, 1);
         if (peer == NULL)
                 return;


@@ -367,10 +368,10 @@ ptllnd_notify(lnet_ni_t *ni, lnet_nid_t nid, int alive)
                               libcfs_id2str(id));
                         w *= 2;
                 }
                               libcfs_id2str(id));
                         w *= 2;
                 }

-                
+

                 ptllnd_wait(ni, w);
         }
                 ptllnd_wait(ni, w);
         }

-        
+

         ptllnd_peer_decref(peer);
 }
 
         ptllnd_peer_decref(peer);
 }
 


@@ -379,7 +380,7 @@ ptllnd_setasync(lnet_ni_t *ni, lnet_process_id_t id, int nasync)
 {
         ptllnd_peer_t *peer = ptllnd_find_peer(ni, id, nasync > 0);
         int            rc;
 {
         ptllnd_peer_t *peer = ptllnd_find_peer(ni, id, nasync > 0);
         int            rc;

-        
+

         if (peer == NULL)
                 return -ENOMEM;
 
         if (peer == NULL)
                 return -ENOMEM;
 


@@ -404,7 +405,7 @@ ptllnd_setasync(lnet_ni_t *ni, lnet_process_id_t id, int nasync)
 
         nasync -= peer->plp_extra_lazy_credits;
         peer->plp_extra_lazy_credits = 0;
 
         nasync -= peer->plp_extra_lazy_credits;
         peer->plp_extra_lazy_credits = 0;

-        
+

         rc = ptllnd_size_buffers(ni, nasync);
         if (rc == 0) {
                 peer->plp_lazy_credits += nasync;
         rc = ptllnd_size_buffers(ni, nasync);
         if (rc == 0) {
                 peer->plp_lazy_credits += nasync;


@@ -597,7 +598,7 @@ ptllnd_tx_done(ptllnd_tx_t *tx)
                 }
                 ptllnd_close_peer(peer, tx->tx_status);
         }
                 }
                 ptllnd_close_peer(peer, tx->tx_status);
         }

-        
+

         ptllnd_abort_tx(tx, &tx->tx_reqmdh);
         ptllnd_abort_tx(tx, &tx->tx_bulkmdh);
 
         ptllnd_abort_tx(tx, &tx->tx_reqmdh);
         ptllnd_abort_tx(tx, &tx->tx_bulkmdh);
 


@@ -619,7 +620,7 @@ ptllnd_tx_done(ptllnd_tx_t *tx)
 
         plni->plni_ntx_history++;
         list_add_tail(&tx->tx_list, &plni->plni_tx_history);
 
         plni->plni_ntx_history++;
         list_add_tail(&tx->tx_list, &plni->plni_tx_history);

-        
+

         ptllnd_cull_tx_history(plni);
 }
 
         ptllnd_cull_tx_history(plni);
 }
 


@@ -663,7 +664,7 @@ ptllnd_set_txiov(ptllnd_tx_t *tx,
 
                         piov[npiov].iov_base = iov[npiov].iov_base + temp_offset;
                         piov[npiov].iov_len = iov[npiov].iov_len - temp_offset;
 
                         piov[npiov].iov_base = iov[npiov].iov_base + temp_offset;
                         piov[npiov].iov_len = iov[npiov].iov_len - temp_offset;

-                        
+

                         if (piov[npiov].iov_len >= resid) {
                                 piov[npiov].iov_len = resid;
                                 npiov++;
                         if (piov[npiov].iov_len >= resid) {
                                 piov[npiov].iov_len = resid;
                                 npiov++;


@@ -759,11 +760,25 @@ ptllnd_post_buffer(ptllnd_buffer_t *buf)
         return -ENOMEM;
 }
 
         return -ENOMEM;
 }
 

+static inline int
+ptllnd_peer_send_noop (ptllnd_peer_t *peer)
+{
+        ptllnd_ni_t *plni = peer->plp_ni->ni_data;
+
+        if (!peer->plp_sent_hello ||
+            peer->plp_credits == 0 ||
+            !list_empty(&peer->plp_noopq) ||
+            peer->plp_outstanding_credits < PTLLND_CREDIT_HIGHWATER(plni))
+                return 0;
+
+        /* No tx to piggyback NOOP onto or no credit to send a tx */
+        return (list_empty(&peer->plp_txq) || peer->plp_credits == 1);
+}
+

 void
 ptllnd_check_sends(ptllnd_peer_t *peer)
 {
 void
 ptllnd_check_sends(ptllnd_peer_t *peer)
 {

-        lnet_ni_t      *ni = peer->plp_ni;
-        ptllnd_ni_t    *plni = ni->ni_data;
+        ptllnd_ni_t    *plni = peer->plp_ni->ni_data;

         ptllnd_tx_t    *tx;
         ptl_md_t        md;
         ptl_handle_md_t mdh;
         ptllnd_tx_t    *tx;
         ptl_md_t        md;
         ptl_handle_md_t mdh;


@@ -774,10 +789,7 @@ ptllnd_check_sends(ptllnd_peer_t *peer)
                peer->plp_outstanding_credits, peer->plp_sent_credits,
                plni->plni_peer_credits + peer->plp_lazy_credits);
 
                peer->plp_outstanding_credits, peer->plp_sent_credits,
                plni->plni_peer_credits + peer->plp_lazy_credits);
 

-        if (list_empty(&peer->plp_txq) &&
-            peer->plp_outstanding_credits >= PTLLND_CREDIT_HIGHWATER(plni) &&
-            peer->plp_credits != 0) {
-
+        if (ptllnd_peer_send_noop(peer)) {

                 tx = ptllnd_new_tx(peer, PTLLND_MSG_TYPE_NOOP, 0);
                 CDEBUG(D_NET, "NOOP tx=%p\n",tx);
                 if (tx == NULL) {
                 tx = ptllnd_new_tx(peer, PTLLND_MSG_TYPE_NOOP, 0);
                 CDEBUG(D_NET, "NOOP tx=%p\n",tx);
                 if (tx == NULL) {


@@ -785,12 +797,22 @@ ptllnd_check_sends(ptllnd_peer_t *peer)
                                libcfs_id2str(peer->plp_id));
                 } else {
                         ptllnd_set_tx_deadline(tx);
                                libcfs_id2str(peer->plp_id));
                 } else {
                         ptllnd_set_tx_deadline(tx);

-                        list_add_tail(&tx->tx_list, &peer->plp_txq);
+                        list_add_tail(&tx->tx_list, &peer->plp_noopq);

                 }
         }
 
                 }
         }
 

-        while (!list_empty(&peer->plp_txq)) {
-                tx = list_entry(peer->plp_txq.next, ptllnd_tx_t, tx_list);
+        for (;;) {
+                if (!list_empty(&peer->plp_noopq)) {
+                        LASSERT (peer->plp_sent_hello);
+                        tx = list_entry(peer->plp_noopq.next,
+                                        ptllnd_tx_t, tx_list);
+                } else if (!list_empty(&peer->plp_txq)) {
+                        tx = list_entry(peer->plp_txq.next,
+                                        ptllnd_tx_t, tx_list);
+                } else {
+                        /* nothing to send right now */
+                        break;
+                }

 
                 LASSERT (tx->tx_msgsize > 0);
 
 
                 LASSERT (tx->tx_msgsize > 0);
 


@@ -800,6 +822,14 @@ ptllnd_check_sends(ptllnd_peer_t *peer)
                          <= plni->plni_peer_credits + peer->plp_lazy_credits);
                 LASSERT (peer->plp_credits >= 0);
 
                          <= plni->plni_peer_credits + peer->plp_lazy_credits);
                 LASSERT (peer->plp_credits >= 0);
 

+                /* say HELLO first */
+                if (!peer->plp_sent_hello) {
+                        LASSERT (list_empty(&peer->plp_noopq));
+                        LASSERT (tx->tx_type == PTLLND_MSG_TYPE_HELLO);
+
+                        peer->plp_sent_hello = 1;
+                }
+

                 if (peer->plp_credits == 0) {   /* no credits */
                         PTLLND_HISTORY("%s[%d/%d+%d(%d)]: no creds for %p",
                                        libcfs_id2str(peer->plp_id),
                 if (peer->plp_credits == 0) {   /* no credits */
                         PTLLND_HISTORY("%s[%d/%d+%d(%d)]: no creds for %p",
                                        libcfs_id2str(peer->plp_id),


@@ -810,9 +840,11 @@ ptllnd_check_sends(ptllnd_peer_t *peer)
                                        peer->plp_lazy_credits, tx);
                         break;
                 }
                                        peer->plp_lazy_credits, tx);
                         break;
                 }

-                
-                if (peer->plp_credits == 1 &&   /* last credit reserved for */
-                    peer->plp_outstanding_credits == 0) { /* returning credits */
+
+                /* Last/Initial credit reserved for NOOP/HELLO */
+                if (peer->plp_credits == 1 &&
+                    tx->tx_type != PTLLND_MSG_TYPE_NOOP &&
+                    tx->tx_type != PTLLND_MSG_TYPE_HELLO) {

                         PTLLND_HISTORY("%s[%d/%d+%d(%d)]: too few creds for %p",
                                        libcfs_id2str(peer->plp_id),
                                        peer->plp_credits,
                         PTLLND_HISTORY("%s[%d/%d+%d(%d)]: too few creds for %p",
                                        libcfs_id2str(peer->plp_id),
                                        peer->plp_credits,


@@ -822,7 +854,7 @@ ptllnd_check_sends(ptllnd_peer_t *peer)
                                        peer->plp_lazy_credits, tx);
                         break;
                 }
                                        peer->plp_lazy_credits, tx);
                         break;
                 }

-                
+

                 list_del(&tx->tx_list);
                 list_add_tail(&tx->tx_list, &peer->plp_activeq);
 
                 list_del(&tx->tx_list);
                 list_add_tail(&tx->tx_list, &peer->plp_activeq);
 


@@ -830,9 +862,7 @@ ptllnd_check_sends(ptllnd_peer_t *peer)
                         ptllnd_msgtype2str(tx->tx_type),tx->tx_type);
 
                 if (tx->tx_type == PTLLND_MSG_TYPE_NOOP &&
                         ptllnd_msgtype2str(tx->tx_type),tx->tx_type);
 
                 if (tx->tx_type == PTLLND_MSG_TYPE_NOOP &&

-                    (!list_empty(&peer->plp_txq) ||
-                     peer->plp_outstanding_credits <
-                     PTLLND_CREDIT_HIGHWATER(plni))) {
+                    !ptllnd_peer_send_noop(peer)) {

                         /* redundant NOOP */
                         ptllnd_tx_done(tx);
                         continue;
                         /* redundant NOOP */
                         ptllnd_tx_done(tx);
                         continue;


@@ -878,7 +908,7 @@ ptllnd_check_sends(ptllnd_peer_t *peer)
 
                 LASSERT (tx->tx_type != PTLLND_RDMA_WRITE &&
                          tx->tx_type != PTLLND_RDMA_READ);
 
                 LASSERT (tx->tx_type != PTLLND_RDMA_WRITE &&
                          tx->tx_type != PTLLND_RDMA_READ);

-                
+

                 tx->tx_reqmdh = mdh;
                 gettimeofday(&tx->tx_req_posted, NULL);
 
                 tx->tx_reqmdh = mdh;
                 gettimeofday(&tx->tx_req_posted, NULL);
 


@@ -1130,7 +1160,7 @@ ptllnd_send(lnet_ni_t *ni, void *private, lnet_msg_t *msg)
 
         LASSERT (msg->msg_niov <= PTL_MD_MAX_IOV); /* !!! */
 
 
         LASSERT (msg->msg_niov <= PTL_MD_MAX_IOV); /* !!! */
 

-        CDEBUG(D_NET, "%s [%d]+%d,%d -> %s%s\n", 
+        CDEBUG(D_NET, "%s [%d]+%d,%d -> %s%s\n",

                lnet_msgtyp2str(msg->msg_type),
                msg->msg_niov, msg->msg_offset, msg->msg_len,
                libcfs_nid2str(msg->msg_target.nid),
                lnet_msgtyp2str(msg->msg_type),
                msg->msg_niov, msg->msg_offset, msg->msg_len,
                libcfs_nid2str(msg->msg_target.nid),


@@ -1141,7 +1171,7 @@ ptllnd_send(lnet_ni_t *ni, void *private, lnet_msg_t *msg)
                        libcfs_id2str(msg->msg_target));
                 return -EHOSTUNREACH;
         }
                        libcfs_id2str(msg->msg_target));
                 return -EHOSTUNREACH;
         }

-        
+

         plp = ptllnd_find_peer(ni, msg->msg_target, 1);
         if (plp == NULL)
                 return -ENOMEM;
         plp = ptllnd_find_peer(ni, msg->msg_target, 1);
         if (plp == NULL)
                 return -ENOMEM;


@@ -1223,8 +1253,7 @@ void
 ptllnd_rx_done(ptllnd_rx_t *rx)
 {
         ptllnd_peer_t *plp = rx->rx_peer;
 ptllnd_rx_done(ptllnd_rx_t *rx)
 {
         ptllnd_peer_t *plp = rx->rx_peer;

-        lnet_ni_t     *ni = plp->plp_ni;
-        ptllnd_ni_t   *plni = ni->ni_data;
+        ptllnd_ni_t   *plni = plp->plp_ni->ni_data;

 
         plp->plp_outstanding_credits++;
 
 
         plp->plp_outstanding_credits++;
 


@@ -1234,7 +1263,7 @@ ptllnd_rx_done(ptllnd_rx_t *rx)
                        plp->plp_sent_credits,
                        plni->plni_peer_credits + plp->plp_lazy_credits, rx);
 
                        plp->plp_sent_credits,
                        plni->plni_peer_credits + plp->plp_lazy_credits, rx);
 

-        ptllnd_check_sends(rx->rx_peer);
+        ptllnd_check_sends(plp);

 
         LASSERT (plni->plni_nrxs > 0);
         plni->plni_nrxs--;
 
         LASSERT (plni->plni_nrxs > 0);
         plni->plni_nrxs--;


@@ -1337,7 +1366,7 @@ ptllnd_parse_request(lnet_ni_t *ni, ptl_process_id_t initiator,
         msg_version = flip ? __swab16(msg->ptlm_version) : msg->ptlm_version;
 
         if (msg_version != PTLLND_MSG_VERSION) {
         msg_version = flip ? __swab16(msg->ptlm_version) : msg->ptlm_version;
 
         if (msg_version != PTLLND_MSG_VERSION) {

-                CERROR("Bad protocol version %04x from %s: %04x expected\n", 
+                CERROR("Bad protocol version %04x from %s: %04x expected\n",

                        (__u32)msg_version, ptllnd_ptlid2str(initiator), PTLLND_MSG_VERSION);
 
                 if (plni->plni_abort_on_protocol_mismatch)
                        (__u32)msg_version, ptllnd_ptlid2str(initiator), PTLLND_MSG_VERSION);
 
                 if (plni->plni_abort_on_protocol_mismatch)


@@ -1366,7 +1395,7 @@ ptllnd_parse_request(lnet_ni_t *ni, ptl_process_id_t initiator,
 
         msg->ptlm_version = msg_version;
         msg->ptlm_cksum = msg_cksum;
 
         msg->ptlm_version = msg_version;
         msg->ptlm_cksum = msg_cksum;

-        
+

         if (flip) {
                 /* NB stamps are opaque cookies */
                 __swab32s(&msg->ptlm_nob);
         if (flip) {
                 /* NB stamps are opaque cookies */
                 __swab32s(&msg->ptlm_nob);


@@ -1375,7 +1404,7 @@ ptllnd_parse_request(lnet_ni_t *ni, ptl_process_id_t initiator,
                 __swab32s(&msg->ptlm_srcpid);
                 __swab32s(&msg->ptlm_dstpid);
         }
                 __swab32s(&msg->ptlm_srcpid);
                 __swab32s(&msg->ptlm_dstpid);
         }

-        
+

         srcid.nid = msg->ptlm_srcnid;
         srcid.pid = msg->ptlm_srcpid;
 
         srcid.nid = msg->ptlm_srcnid;
         srcid.pid = msg->ptlm_srcpid;
 


@@ -1387,19 +1416,19 @@ ptllnd_parse_request(lnet_ni_t *ni, ptl_process_id_t initiator,
         }
 
         if (msg->ptlm_type == PTLLND_MSG_TYPE_NAK) {
         }
 
         if (msg->ptlm_type == PTLLND_MSG_TYPE_NAK) {

-                CERROR("NAK from %s (%s)\n", 
+                CERROR("NAK from %s (%s)\n",

                        libcfs_id2str(srcid),
                        ptllnd_ptlid2str(initiator));
 
                 if (plni->plni_dump_on_nak)
                         ptllnd_dump_debug(ni, srcid);
                        libcfs_id2str(srcid),
                        ptllnd_ptlid2str(initiator));
 
                 if (plni->plni_dump_on_nak)
                         ptllnd_dump_debug(ni, srcid);

-                
+

                 if (plni->plni_abort_on_nak)
                         abort();
                 if (plni->plni_abort_on_nak)
                         abort();

-                
+

                 return;
         }
                 return;
         }

-        
+

         if (msg->ptlm_dstnid != ni->ni_nid ||
             msg->ptlm_dstpid != the_lnet.ln_pid) {
                 CERROR("Bad dstid %s (%s expected) from %s\n",
         if (msg->ptlm_dstnid != ni->ni_nid ||
             msg->ptlm_dstpid != the_lnet.ln_pid) {
                 CERROR("Bad dstid %s (%s expected) from %s\n",


@@ -1459,7 +1488,7 @@ ptllnd_parse_request(lnet_ni_t *ni, ptl_process_id_t initiator,
                         __swab32s(&msg->ptlm_u.hello.kptlhm_max_msg_size);
                 }
                 break;
                         __swab32s(&msg->ptlm_u.hello.kptlhm_max_msg_size);
                 }
                 break;

-                
+

         case PTLLND_MSG_TYPE_NOOP:
                 break;
 
         case PTLLND_MSG_TYPE_NOOP:
                 break;
 


@@ -1509,19 +1538,16 @@ ptllnd_parse_request(lnet_ni_t *ni, ptl_process_id_t initiator,
         if (plp->plp_sent_credits == 0) {
                 CERROR("%s[%d/%d+%d(%d)]: unexpected message\n",
                        libcfs_id2str(plp->plp_id),
         if (plp->plp_sent_credits == 0) {
                 CERROR("%s[%d/%d+%d(%d)]: unexpected message\n",
                        libcfs_id2str(plp->plp_id),

-                       plp->plp_credits, plp->plp_outstanding_credits, 
+                       plp->plp_credits, plp->plp_outstanding_credits,

                        plp->plp_sent_credits,
                        plni->plni_peer_credits + plp->plp_lazy_credits);
                 return;
         }
         plp->plp_sent_credits--;
                        plp->plp_sent_credits,
                        plni->plni_peer_credits + plp->plp_lazy_credits);
                 return;
         }
         plp->plp_sent_credits--;

-        
+

         /* No check for credit overflow - the peer may post new buffers after
          * the startup handshake. */
         /* No check for credit overflow - the peer may post new buffers after
          * the startup handshake. */

-        if (msg->ptlm_credits > 0) {
-                plp->plp_credits += msg->ptlm_credits;
-                ptllnd_check_sends(plp);
-        }
+        plp->plp_credits += msg->ptlm_credits;

 
         /* All OK so far; assume the message is good... */
 
 
         /* All OK so far; assume the message is good... */
 


@@ -1551,6 +1577,9 @@ ptllnd_parse_request(lnet_ni_t *ni, ptl_process_id_t initiator,
                 break;
         }
 
                 break;
         }
 

+        if (msg->ptlm_credits > 0)
+                ptllnd_check_sends(plp);
+

         ptllnd_peer_decref(plp);
 }
 
         ptllnd_peer_decref(plp);
 }
 


@@ -1580,7 +1609,7 @@ ptllnd_buf_event (lnet_ni_t *ni, ptl_event_t *event)
                 /* Portals can't force message alignment - someone sending an
                  * odd-length message could misalign subsequent messages */
                 if ((event->mlength & 7) != 0) {
                 /* Portals can't force message alignment - someone sending an
                  * odd-length message could misalign subsequent messages */
                 if ((event->mlength & 7) != 0) {

-                        CERROR("Message from %s has odd length %llu: "
+                        CERROR("Message from %s has odd length %u: "

                                "probable version incompatibility\n",
                                ptllnd_ptlid2str(event->initiator),
                                event->mlength);
                                "probable version incompatibility\n",
                                ptllnd_ptlid2str(event->initiator),
                                event->mlength);


@@ -1655,7 +1684,7 @@ ptllnd_tx_event (lnet_ni_t *ni, ptl_event_t *event)
         LASSERT (!isreq != !isbulk);            /* always one and only 1 match */
 
         PTLLND_HISTORY("%s[%d/%d+%d(%d)]: TX done %p %s%s",
         LASSERT (!isreq != !isbulk);            /* always one and only 1 match */
 
         PTLLND_HISTORY("%s[%d/%d+%d(%d)]: TX done %p %s%s",

-                       libcfs_id2str(tx->tx_peer->plp_id), 
+                       libcfs_id2str(tx->tx_peer->plp_id),

                        tx->tx_peer->plp_credits,
                        tx->tx_peer->plp_outstanding_credits,
                        tx->tx_peer->plp_sent_credits,
                        tx->tx_peer->plp_credits,
                        tx->tx_peer->plp_outstanding_credits,
                        tx->tx_peer->plp_sent_credits,


@@ -1728,18 +1757,19 @@ ptllnd_tx_t *
 ptllnd_find_timed_out_tx(ptllnd_peer_t *peer)
 {
         time_t            now = cfs_time_current_sec();
 ptllnd_find_timed_out_tx(ptllnd_peer_t *peer)
 {
         time_t            now = cfs_time_current_sec();

-        struct list_head *tmp;
+        ptllnd_tx_t *tx;
+
+        list_for_each_entry (tx, &peer->plp_txq, tx_list) {
+                if (tx->tx_deadline < now)
+                        return tx;
+        }

 
 

-        list_for_each(tmp, &peer->plp_txq) {
-                ptllnd_tx_t *tx = list_entry(tmp, ptllnd_tx_t, tx_list);
-                
+        list_for_each_entry (tx, &peer->plp_noopq, tx_list) {

                 if (tx->tx_deadline < now)
                         return tx;
         }
                 if (tx->tx_deadline < now)
                         return tx;
         }

-        
-        list_for_each(tmp, &peer->plp_activeq) {
-                ptllnd_tx_t *tx = list_entry(tmp, ptllnd_tx_t, tx_list);
-                
+
+        list_for_each_entry (tx, &peer->plp_activeq, tx_list) {

                 if (tx->tx_deadline < now)
                         return tx;
         }
                 if (tx->tx_deadline < now)
                         return tx;
         }


@@ -1751,10 +1781,10 @@ void
 ptllnd_check_peer(ptllnd_peer_t *peer)
 {
         ptllnd_tx_t *tx = ptllnd_find_timed_out_tx(peer);
 ptllnd_check_peer(ptllnd_peer_t *peer)
 {
         ptllnd_tx_t *tx = ptllnd_find_timed_out_tx(peer);

-        
+

         if (tx == NULL)
                 return;
         if (tx == NULL)
                 return;

-        
+

         CERROR("%s: timed out\n", libcfs_id2str(peer->plp_id));
         ptllnd_close_peer(peer, -ETIMEDOUT);
 }
         CERROR("%s: timed out\n", libcfs_id2str(peer->plp_id));
         ptllnd_close_peer(peer, -ETIMEDOUT);
 }


@@ -1788,11 +1818,11 @@ ptllnd_watchdog (lnet_ni_t *ni, time_t now)
 
         for (i = 0; i < chunk; i++) {
                 hashlist = &plni->plni_peer_hash[plni->plni_watchdog_peeridx];
 
         for (i = 0; i < chunk; i++) {
                 hashlist = &plni->plni_peer_hash[plni->plni_watchdog_peeridx];

-                
+

                 list_for_each_safe(tmp, nxt, hashlist) {
                         ptllnd_check_peer(list_entry(tmp, ptllnd_peer_t, plp_list));
                 }
                 list_for_each_safe(tmp, nxt, hashlist) {
                         ptllnd_check_peer(list_entry(tmp, ptllnd_peer_t, plp_list));
                 }

-                
+

                 plni->plni_watchdog_peeridx = (plni->plni_watchdog_peeridx + 1) %
                                               plni->plni_peer_hash_size;
         }
                 plni->plni_watchdog_peeridx = (plni->plni_watchdog_peeridx + 1) %
                                               plni->plni_peer_hash_size;
         }


@@ -1811,7 +1841,7 @@ ptllnd_wait (lnet_ni_t *ni, int milliseconds)
         struct timeval         then;
         struct timeval         now;
         struct timeval         deadline;
         struct timeval         then;
         struct timeval         now;
         struct timeval         deadline;

-        
+

         ptllnd_ni_t   *plni = ni->ni_data;
         ptllnd_tx_t   *tx;
         ptl_event_t    event;
         ptllnd_ni_t   *plni = ni->ni_data;
         ptllnd_tx_t   *tx;
         ptl_event_t    event;


@@ -1841,7 +1871,7 @@ ptllnd_wait (lnet_ni_t *ni, int milliseconds)
 
         for (;;) {
                 gettimeofday(&then, NULL);
 
         for (;;) {
                 gettimeofday(&then, NULL);

-                
+

                 rc = PtlEQPoll(&plni->plni_eqh, 1, timeout, &event, &which);
 
                 gettimeofday(&now, NULL);
                 rc = PtlEQPoll(&plni->plni_eqh, 1, timeout, &event, &which);
 
                 gettimeofday(&now, NULL);


@@ -1862,7 +1892,7 @@ ptllnd_wait (lnet_ni_t *ni, int milliseconds)
                                 ptllnd_watchdog(ni, now.tv_sec);
                                 LASSERT (now.tv_sec < plni->plni_watchdog_nextt);
                         }
                                 ptllnd_watchdog(ni, now.tv_sec);
                                 LASSERT (now.tv_sec < plni->plni_watchdog_nextt);
                         }

-                        
+

                         if (now.tv_sec > deadline.tv_sec || /* timeout expired */
                             (now.tv_sec == deadline.tv_sec &&
                              now.tv_usec >= deadline.tv_usec))
                         if (now.tv_sec > deadline.tv_sec || /* timeout expired */
                             (now.tv_sec == deadline.tv_sec &&
                              now.tv_usec >= deadline.tv_usec))


@@ -1878,7 +1908,7 @@ ptllnd_wait (lnet_ni_t *ni, int milliseconds)
 
                         continue;
                 }
 
                         continue;
                 }

-                
+

                 LASSERT (rc == PTL_OK || rc == PTL_EQ_DROPPED);
 
                 if (rc == PTL_EQ_DROPPED)
                 LASSERT (rc == PTL_OK || rc == PTL_EQ_DROPPED);
 
                 if (rc == PTL_EQ_DROPPED)