Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / staging / lustre / lnet / klnds / o2iblnd / o2iblnd_modparams.c
diff --git a/kernel/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd_modparams.c b/kernel/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd_modparams.c
new file mode 100644 (file)
index 0000000..eedf01a
--- /dev/null
@@ -0,0 +1,230 @@
+/*
+ * GPL HEADER START
+ *
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
+ *
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
+ *
+ * Copyright (c) 2012, Intel Corporation.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lnet/klnds/o2iblnd/o2iblnd_modparams.c
+ *
+ * Author: Eric Barton <eric@bartonsoftware.com>
+ */
+
+#include "o2iblnd.h"
+
+static int service = 987;
+module_param(service, int, 0444);
+MODULE_PARM_DESC(service, "service number (within RDMA_PS_TCP)");
+
+static int cksum;
+module_param(cksum, int, 0644);
+MODULE_PARM_DESC(cksum, "set non-zero to enable message (not RDMA) checksums");
+
+static int timeout = 50;
+module_param(timeout, int, 0644);
+MODULE_PARM_DESC(timeout, "timeout (seconds)");
+
+/* Number of threads in each scheduler pool which is percpt,
+ * we will estimate reasonable value based on CPUs if it's set to zero. */
+static int nscheds;
+module_param(nscheds, int, 0444);
+MODULE_PARM_DESC(nscheds, "number of threads in each scheduler pool");
+
+/* NB: this value is shared by all CPTs, it can grow at runtime */
+static int ntx = 512;
+module_param(ntx, int, 0444);
+MODULE_PARM_DESC(ntx, "# of message descriptors allocated for each pool");
+
+/* NB: this value is shared by all CPTs */
+static int credits = 256;
+module_param(credits, int, 0444);
+MODULE_PARM_DESC(credits, "# concurrent sends");
+
+static int peer_credits = 8;
+module_param(peer_credits, int, 0444);
+MODULE_PARM_DESC(peer_credits, "# concurrent sends to 1 peer");
+
+static int peer_credits_hiw;
+module_param(peer_credits_hiw, int, 0444);
+MODULE_PARM_DESC(peer_credits_hiw, "when eagerly to return credits");
+
+static int peer_buffer_credits;
+module_param(peer_buffer_credits, int, 0444);
+MODULE_PARM_DESC(peer_buffer_credits, "# per-peer router buffer credits");
+
+static int peer_timeout = 180;
+module_param(peer_timeout, int, 0444);
+MODULE_PARM_DESC(peer_timeout, "Seconds without aliveness news to declare peer dead (<=0 to disable)");
+
+static char *ipif_name = "ib0";
+module_param(ipif_name, charp, 0444);
+MODULE_PARM_DESC(ipif_name, "IPoIB interface name");
+
+static int retry_count = 5;
+module_param(retry_count, int, 0644);
+MODULE_PARM_DESC(retry_count, "Retransmissions when no ACK received");
+
+static int rnr_retry_count = 6;
+module_param(rnr_retry_count, int, 0644);
+MODULE_PARM_DESC(rnr_retry_count, "RNR retransmissions");
+
+static int keepalive = 100;
+module_param(keepalive, int, 0644);
+MODULE_PARM_DESC(keepalive, "Idle time in seconds before sending a keepalive");
+
+static int ib_mtu;
+module_param(ib_mtu, int, 0444);
+MODULE_PARM_DESC(ib_mtu, "IB MTU 256/512/1024/2048/4096");
+
+static int concurrent_sends;
+module_param(concurrent_sends, int, 0444);
+MODULE_PARM_DESC(concurrent_sends, "send work-queue sizing");
+
+static int map_on_demand;
+module_param(map_on_demand, int, 0444);
+MODULE_PARM_DESC(map_on_demand, "map on demand");
+
+/* NB: this value is shared by all CPTs, it can grow at runtime */
+static int fmr_pool_size = 512;
+module_param(fmr_pool_size, int, 0444);
+MODULE_PARM_DESC(fmr_pool_size, "size of fmr pool on each CPT (>= ntx / 4)");
+
+/* NB: this value is shared by all CPTs, it can grow at runtime */
+static int fmr_flush_trigger = 384;
+module_param(fmr_flush_trigger, int, 0444);
+MODULE_PARM_DESC(fmr_flush_trigger, "# dirty FMRs that triggers pool flush");
+
+static int fmr_cache = 1;
+module_param(fmr_cache, int, 0444);
+MODULE_PARM_DESC(fmr_cache, "non-zero to enable FMR caching");
+
+/* NB: this value is shared by all CPTs, it can grow at runtime */
+static int pmr_pool_size = 512;
+module_param(pmr_pool_size, int, 0444);
+MODULE_PARM_DESC(pmr_pool_size, "size of MR cache pmr pool on each CPT");
+
+/*
+ * 0: disable failover
+ * 1: enable failover if necessary
+ * 2: force to failover (for debug)
+ */
+static int dev_failover;
+module_param(dev_failover, int, 0444);
+MODULE_PARM_DESC(dev_failover, "HCA failover for bonding (0 off, 1 on, other values reserved)");
+
+
+static int require_privileged_port;
+module_param(require_privileged_port, int, 0644);
+MODULE_PARM_DESC(require_privileged_port, "require privileged port when accepting connection");
+
+static int use_privileged_port = 1;
+module_param(use_privileged_port, int, 0644);
+MODULE_PARM_DESC(use_privileged_port, "use privileged port when initiating connection");
+
+kib_tunables_t kiblnd_tunables = {
+       .kib_dev_failover          = &dev_failover,
+       .kib_service            = &service,
+       .kib_cksum                = &cksum,
+       .kib_timeout            = &timeout,
+       .kib_keepalive        = &keepalive,
+       .kib_ntx                    = &ntx,
+       .kib_credits            = &credits,
+       .kib_peertxcredits        = &peer_credits,
+       .kib_peercredits_hiw    = &peer_credits_hiw,
+       .kib_peerrtrcredits      = &peer_buffer_credits,
+       .kib_peertimeout            = &peer_timeout,
+       .kib_default_ipif          = &ipif_name,
+       .kib_retry_count            = &retry_count,
+       .kib_rnr_retry_count    = &rnr_retry_count,
+       .kib_concurrent_sends       = &concurrent_sends,
+       .kib_ib_mtu              = &ib_mtu,
+       .kib_map_on_demand        = &map_on_demand,
+       .kib_fmr_pool_size        = &fmr_pool_size,
+       .kib_fmr_flush_trigger      = &fmr_flush_trigger,
+       .kib_fmr_cache        = &fmr_cache,
+       .kib_pmr_pool_size        = &pmr_pool_size,
+       .kib_require_priv_port      = &require_privileged_port,
+       .kib_use_priv_port          = &use_privileged_port,
+       .kib_nscheds                = &nscheds
+};
+
+int
+kiblnd_tunables_init(void)
+{
+       if (kiblnd_translate_mtu(*kiblnd_tunables.kib_ib_mtu) < 0) {
+               CERROR("Invalid ib_mtu %d, expected 256/512/1024/2048/4096\n",
+                      *kiblnd_tunables.kib_ib_mtu);
+               return -EINVAL;
+       }
+
+       if (*kiblnd_tunables.kib_peertxcredits < IBLND_CREDITS_DEFAULT)
+               *kiblnd_tunables.kib_peertxcredits = IBLND_CREDITS_DEFAULT;
+
+       if (*kiblnd_tunables.kib_peertxcredits > IBLND_CREDITS_MAX)
+               *kiblnd_tunables.kib_peertxcredits = IBLND_CREDITS_MAX;
+
+       if (*kiblnd_tunables.kib_peertxcredits > *kiblnd_tunables.kib_credits)
+               *kiblnd_tunables.kib_peertxcredits = *kiblnd_tunables.kib_credits;
+
+       if (*kiblnd_tunables.kib_peercredits_hiw < *kiblnd_tunables.kib_peertxcredits / 2)
+               *kiblnd_tunables.kib_peercredits_hiw = *kiblnd_tunables.kib_peertxcredits / 2;
+
+       if (*kiblnd_tunables.kib_peercredits_hiw >= *kiblnd_tunables.kib_peertxcredits)
+               *kiblnd_tunables.kib_peercredits_hiw = *kiblnd_tunables.kib_peertxcredits - 1;
+
+       if (*kiblnd_tunables.kib_map_on_demand < 0 ||
+           *kiblnd_tunables.kib_map_on_demand > IBLND_MAX_RDMA_FRAGS)
+               *kiblnd_tunables.kib_map_on_demand = 0; /* disable map-on-demand */
+
+       if (*kiblnd_tunables.kib_map_on_demand == 1)
+               *kiblnd_tunables.kib_map_on_demand = 2; /* don't make sense to create map if only one fragment */
+
+       if (*kiblnd_tunables.kib_concurrent_sends == 0) {
+               if (*kiblnd_tunables.kib_map_on_demand > 0 &&
+                   *kiblnd_tunables.kib_map_on_demand <= IBLND_MAX_RDMA_FRAGS / 8)
+                       *kiblnd_tunables.kib_concurrent_sends = (*kiblnd_tunables.kib_peertxcredits) * 2;
+               else
+                       *kiblnd_tunables.kib_concurrent_sends = (*kiblnd_tunables.kib_peertxcredits);
+       }
+
+       if (*kiblnd_tunables.kib_concurrent_sends > *kiblnd_tunables.kib_peertxcredits * 2)
+               *kiblnd_tunables.kib_concurrent_sends = *kiblnd_tunables.kib_peertxcredits * 2;
+
+       if (*kiblnd_tunables.kib_concurrent_sends < *kiblnd_tunables.kib_peertxcredits / 2)
+               *kiblnd_tunables.kib_concurrent_sends = *kiblnd_tunables.kib_peertxcredits / 2;
+
+       if (*kiblnd_tunables.kib_concurrent_sends < *kiblnd_tunables.kib_peertxcredits) {
+               CWARN("Concurrent sends %d is lower than message queue size: %d, performance may drop slightly.\n",
+                     *kiblnd_tunables.kib_concurrent_sends, *kiblnd_tunables.kib_peertxcredits);
+       }
+
+       return 0;
+}