diff options
Diffstat (limited to 'drivers/infiniband/ulp')
| -rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib.h | 34 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_ib.c | 27 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_main.c | 28 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_multicast.c | 23 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_verbs.c | 3 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/iser/Kconfig | 11 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/iser/Makefile | 4 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/iser/iscsi_iser.c | 790 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/iser/iscsi_iser.h | 354 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/iser/iser_initiator.c | 738 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/iser/iser_memory.c | 401 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/iser/iser_verbs.c | 827 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/srp/ib_srp.c | 482 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/srp/ib_srp.h | 33 |
14 files changed, 3541 insertions, 214 deletions
diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h index 12a1e0572ef2..491d2afaf5b4 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib.h +++ b/drivers/infiniband/ulp/ipoib/ipoib.h @@ -272,8 +272,7 @@ int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port); void ipoib_dev_cleanup(struct net_device *dev); void ipoib_mcast_join_task(void *dev_ptr); -void ipoib_mcast_send(struct net_device *dev, union ib_gid *mgid, - struct sk_buff *skb); +void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb); void ipoib_mcast_restart_task(void *dev_ptr); int ipoib_mcast_start_thread(struct net_device *dev); @@ -369,15 +368,26 @@ extern int ipoib_debug_level; #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */ -#define IPOIB_GID_FMT "%x:%x:%x:%x:%x:%x:%x:%x" - -#define IPOIB_GID_ARG(gid) be16_to_cpup((__be16 *) ((gid).raw + 0)), \ - be16_to_cpup((__be16 *) ((gid).raw + 2)), \ - be16_to_cpup((__be16 *) ((gid).raw + 4)), \ - be16_to_cpup((__be16 *) ((gid).raw + 6)), \ - be16_to_cpup((__be16 *) ((gid).raw + 8)), \ - be16_to_cpup((__be16 *) ((gid).raw + 10)), \ - be16_to_cpup((__be16 *) ((gid).raw + 12)), \ - be16_to_cpup((__be16 *) ((gid).raw + 14)) +#define IPOIB_GID_FMT "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:" \ + "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x" + +#define IPOIB_GID_RAW_ARG(gid) ((u8 *)(gid))[0], \ + ((u8 *)(gid))[1], \ + ((u8 *)(gid))[2], \ + ((u8 *)(gid))[3], \ + ((u8 *)(gid))[4], \ + ((u8 *)(gid))[5], \ + ((u8 *)(gid))[6], \ + ((u8 *)(gid))[7], \ + ((u8 *)(gid))[8], \ + ((u8 *)(gid))[9], \ + ((u8 *)(gid))[10],\ + ((u8 *)(gid))[11],\ + ((u8 *)(gid))[12],\ + ((u8 *)(gid))[13],\ + ((u8 *)(gid))[14],\ + ((u8 *)(gid))[15] + +#define IPOIB_GID_ARG(gid) IPOIB_GID_RAW_ARG((gid).raw) #endif /* _IPOIB_H */ diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c index 8406839b91cf..5033666b1481 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c @@ -84,15 +84,9 @@ void ipoib_free_ah(struct kref *kref) unsigned long flags; - if ((int) priv->tx_tail - (int) ah->last_send >= 0) { - ipoib_dbg(priv, "Freeing ah %p\n", ah->ah); - ib_destroy_ah(ah->ah); - kfree(ah); - } else { - spin_lock_irqsave(&priv->lock, flags); - list_add_tail(&ah->list, &priv->dead_ahs); - spin_unlock_irqrestore(&priv->lock, flags); - } + spin_lock_irqsave(&priv->lock, flags); + list_add_tail(&ah->list, &priv->dead_ahs); + spin_unlock_irqrestore(&priv->lock, flags); } static int ipoib_ib_post_receive(struct net_device *dev, int id) @@ -377,19 +371,16 @@ static void __ipoib_reap_ah(struct net_device *dev) struct ipoib_ah *ah, *tah; LIST_HEAD(remove_list); - spin_lock_irq(&priv->lock); + spin_lock_irq(&priv->tx_lock); + spin_lock(&priv->lock); list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list) if ((int) priv->tx_tail - (int) ah->last_send >= 0) { list_del(&ah->list); - list_add_tail(&ah->list, &remove_list); + ib_destroy_ah(ah->ah); + kfree(ah); } - spin_unlock_irq(&priv->lock); - - list_for_each_entry_safe(ah, tah, &remove_list, list) { - ipoib_dbg(priv, "Reaping ah %p\n", ah->ah); - ib_destroy_ah(ah->ah); - kfree(ah); - } + spin_unlock(&priv->lock); + spin_unlock_irq(&priv->tx_lock); } void ipoib_reap_ah(void *dev_ptr) diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c index cb078a7d0bf5..1c6ea1c682a5 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_main.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c @@ -185,8 +185,7 @@ static int ipoib_change_mtu(struct net_device *dev, int new_mtu) return 0; } -static struct ipoib_path *__path_find(struct net_device *dev, - union ib_gid *gid) +static struct ipoib_path *__path_find(struct net_device *dev, void *gid) { struct ipoib_dev_priv *priv = netdev_priv(dev); struct rb_node *n = priv->path_tree.rb_node; @@ -196,7 +195,7 @@ static struct ipoib_path *__path_find(struct net_device *dev, while (n) { path = rb_entry(n, struct ipoib_path, rb_node); - ret = memcmp(gid->raw, path->pathrec.dgid.raw, + ret = memcmp(gid, path->pathrec.dgid.raw, sizeof (union ib_gid)); if (ret < 0) @@ -424,8 +423,7 @@ static void path_rec_completion(int status, } } -static struct ipoib_path *path_rec_create(struct net_device *dev, - union ib_gid *gid) +static struct ipoib_path *path_rec_create(struct net_device *dev, void *gid) { struct ipoib_dev_priv *priv = netdev_priv(dev); struct ipoib_path *path; @@ -440,7 +438,7 @@ static struct ipoib_path *path_rec_create(struct net_device *dev, INIT_LIST_HEAD(&path->neigh_list); - memcpy(path->pathrec.dgid.raw, gid->raw, sizeof (union ib_gid)); + memcpy(path->pathrec.dgid.raw, gid, sizeof (union ib_gid)); path->pathrec.sgid = priv->local_gid; path->pathrec.pkey = cpu_to_be16(priv->pkey); path->pathrec.numb_path = 1; @@ -498,10 +496,9 @@ static void neigh_add_path(struct sk_buff *skb, struct net_device *dev) */ spin_lock(&priv->lock); - path = __path_find(dev, (union ib_gid *) (skb->dst->neighbour->ha + 4)); + path = __path_find(dev, skb->dst->neighbour->ha + 4); if (!path) { - path = path_rec_create(dev, - (union ib_gid *) (skb->dst->neighbour->ha + 4)); + path = path_rec_create(dev, skb->dst->neighbour->ha + 4); if (!path) goto err_path; @@ -551,7 +548,7 @@ static void ipoib_path_lookup(struct sk_buff *skb, struct net_device *dev) /* Add in the P_Key for multicasts */ skb->dst->neighbour->ha[8] = (priv->pkey >> 8) & 0xff; skb->dst->neighbour->ha[9] = priv->pkey & 0xff; - ipoib_mcast_send(dev, (union ib_gid *) (skb->dst->neighbour->ha + 4), skb); + ipoib_mcast_send(dev, skb->dst->neighbour->ha + 4, skb); } static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev, @@ -566,10 +563,9 @@ static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev, */ spin_lock(&priv->lock); - path = __path_find(dev, (union ib_gid *) (phdr->hwaddr + 4)); + path = __path_find(dev, phdr->hwaddr + 4); if (!path) { - path = path_rec_create(dev, - (union ib_gid *) (phdr->hwaddr + 4)); + path = path_rec_create(dev, phdr->hwaddr + 4); if (path) { /* put pseudoheader back on for next time */ skb_push(skb, sizeof *phdr); @@ -660,7 +656,7 @@ static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev) phdr->hwaddr[8] = (priv->pkey >> 8) & 0xff; phdr->hwaddr[9] = priv->pkey & 0xff; - ipoib_mcast_send(dev, (union ib_gid *) (phdr->hwaddr + 4), skb); + ipoib_mcast_send(dev, phdr->hwaddr + 4, skb); } else { /* unicast GID -- should be ARP or RARP reply */ @@ -671,7 +667,7 @@ static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev) skb->dst ? "neigh" : "dst", be16_to_cpup((__be16 *) skb->data), be32_to_cpup((__be32 *) phdr->hwaddr), - IPOIB_GID_ARG(*(union ib_gid *) (phdr->hwaddr + 4))); + IPOIB_GID_RAW_ARG(phdr->hwaddr + 4)); dev_kfree_skb_any(skb); ++priv->stats.tx_dropped; goto out; @@ -754,7 +750,7 @@ static void ipoib_neigh_destructor(struct neighbour *n) ipoib_dbg(priv, "neigh_destructor for %06x " IPOIB_GID_FMT "\n", be32_to_cpup((__be32 *) n->ha), - IPOIB_GID_ARG(*((union ib_gid *) (n->ha + 4)))); + IPOIB_GID_RAW_ARG(n->ha + 4)); spin_lock_irqsave(&priv->lock, flags); diff --git a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c index 1dae4b238252..ab40488182b3 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c @@ -154,7 +154,7 @@ static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev, return mcast; } -static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, union ib_gid *mgid) +static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid) { struct ipoib_dev_priv *priv = netdev_priv(dev); struct rb_node *n = priv->multicast_tree.rb_node; @@ -165,7 +165,7 @@ static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, union ib_g mcast = rb_entry(n, struct ipoib_mcast, rb_node); - ret = memcmp(mgid->raw, mcast->mcmember.mgid.raw, + ret = memcmp(mgid, mcast->mcmember.mgid.raw, sizeof (union ib_gid)); if (ret < 0) n = n->rb_left; @@ -694,8 +694,7 @@ static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast) return 0; } -void ipoib_mcast_send(struct net_device *dev, union ib_gid *mgid, - struct sk_buff *skb) +void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb) { struct ipoib_dev_priv *priv = netdev_priv(dev); struct ipoib_mcast *mcast; @@ -718,7 +717,7 @@ void ipoib_mcast_send(struct net_device *dev, union ib_gid *mgid, if (!mcast) { /* Let's create a new send only group now */ ipoib_dbg_mcast(priv, "setting up send only multicast group for " - IPOIB_GID_FMT "\n", IPOIB_GID_ARG(*mgid)); + IPOIB_GID_FMT "\n", IPOIB_GID_RAW_ARG(mgid)); mcast = ipoib_mcast_alloc(dev, 0); if (!mcast) { @@ -730,7 +729,7 @@ void ipoib_mcast_send(struct net_device *dev, union ib_gid *mgid, } set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags); - mcast->mcmember.mgid = *mgid; + memcpy(mcast->mcmember.mgid.raw, mgid, sizeof (union ib_gid)); __ipoib_mcast_add(dev, mcast); list_add_tail(&mcast->list, &priv->multicast_list); } @@ -821,7 +820,8 @@ void ipoib_mcast_restart_task(void *dev_ptr) ipoib_mcast_stop_thread(dev, 0); - spin_lock_irqsave(&dev->xmit_lock, flags); + local_irq_save(flags); + netif_tx_lock(dev); spin_lock(&priv->lock); /* @@ -864,8 +864,7 @@ void ipoib_mcast_restart_task(void *dev_ptr) if (mcast) { /* Destroy the send only entry */ - list_del(&mcast->list); - list_add_tail(&mcast->list, &remove_list); + list_move_tail(&mcast->list, &remove_list); rb_replace_node(&mcast->rb_node, &nmcast->rb_node, @@ -890,13 +889,13 @@ void ipoib_mcast_restart_task(void *dev_ptr) rb_erase(&mcast->rb_node, &priv->multicast_tree); /* Move to the remove list */ - list_del(&mcast->list); - list_add_tail(&mcast->list, &remove_list); + list_move_tail(&mcast->list, &remove_list); } } spin_unlock(&priv->lock); - spin_unlock_irqrestore(&dev->xmit_lock, flags); + netif_tx_unlock(dev); + local_irq_restore(flags); /* We have to cancel outside of the spinlock */ list_for_each_entry_safe(mcast, tmcast, &remove_list, list) { diff --git a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c index 1d49d1643c59..7b717c648f72 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c @@ -255,7 +255,8 @@ void ipoib_event(struct ib_event_handler *handler, record->event == IB_EVENT_PKEY_CHANGE || record->event == IB_EVENT_PORT_ACTIVE || record->event == IB_EVENT_LID_CHANGE || - record->event == IB_EVENT_SM_CHANGE) { + record->event == IB_EVENT_SM_CHANGE || + record->event == IB_EVENT_CLIENT_REREGISTER) { ipoib_dbg(priv, "Port state change event\n"); queue_work(ipoib_workqueue, &priv->flush_task); } diff --git a/drivers/infiniband/ulp/iser/Kconfig b/drivers/infiniband/ulp/iser/Kconfig new file mode 100644 index 000000000000..fead87d1eff9 --- /dev/null +++ b/drivers/infiniband/ulp/iser/Kconfig @@ -0,0 +1,11 @@ +config INFINIBAND_ISER + tristate "ISCSI RDMA Protocol" + depends on INFINIBAND && SCSI + select SCSI_ISCSI_ATTRS + ---help--- + Support for the ISCSI RDMA Protocol over InfiniBand. This + allows you to access storage devices that speak ISER/ISCSI + over InfiniBand. + + The ISER protocol is defined by IETF. + See <http://www.ietf.org/>. diff --git a/drivers/infiniband/ulp/iser/Makefile b/drivers/infiniband/ulp/iser/Makefile new file mode 100644 index 000000000000..fe6cd15f2317 --- /dev/null +++ b/drivers/infiniband/ulp/iser/Makefile @@ -0,0 +1,4 @@ +obj-$(CONFIG_INFINIBAND_ISER) += ib_iser.o + +ib_iser-y := iser_verbs.o iser_initiator.o iser_memory.o \ + iscsi_iser.o diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.c b/drivers/infiniband/ulp/iser/iscsi_iser.c new file mode 100644 index 000000000000..4c3f2de2a06e --- /dev/null +++ b/drivers/infiniband/ulp/iser/iscsi_iser.c @@ -0,0 +1,790 @@ +/* + * iSCSI Initiator over iSER Data-Path + * + * Copyright (C) 2004 Dmitry Yusupov + * Copyright (C) 2004 Alex Aizman + * Copyright (C) 2005 Mike Christie + * Copyright (c) 2005, 2006 Voltaire, Inc. All rights reserved. + * maintained by openib-general@openib.org + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Credits: + * Christoph Hellwig + * FUJITA Tomonori + * Arne Redlich + * Zhenyu Wang + * Modified by: + * Erez Zilber + * + * + * $Id: iscsi_iser.c 6965 2006-05-07 11:36:20Z ogerlitz $ + */ + +#include <linux/types.h> +#include <linux/list.h> +#include <linux/hardirq.h> +#include <linux/kfifo.h> +#include <linux/blkdev.h> +#include <linux/init.h> +#include <linux/ioctl.h> +#include <linux/devfs_fs_kernel.h> +#include <linux/cdev.h> +#include <linux/in.h> +#include <linux/net.h> +#include <linux/scatterlist.h> +#include <linux/delay.h> + +#include <net/sock.h> + +#include <asm/uaccess.h> + +#include <scsi/scsi_cmnd.h> +#include <scsi/scsi_device.h> +#include <scsi/scsi_eh.h> +#include <scsi/scsi_tcq.h> +#include <scsi/scsi_host.h> +#include <scsi/scsi.h> +#include <scsi/scsi_transport_iscsi.h> + +#include "iscsi_iser.h" + +static unsigned int iscsi_max_lun = 512; +module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO); + +int iser_debug_level = 0; + +MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover " + "v" DRV_VER " (" DRV_DATE ")"); +MODULE_LICENSE("Dual BSD/GPL"); +MODULE_AUTHOR("Alex Nezhinsky, Dan Bar Dov, Or Gerlitz"); + +module_param_named(debug_level, iser_debug_level, int, 0644); +MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:disabled)"); + +struct iser_global ig; + +void +iscsi_iser_recv(struct iscsi_conn *conn, + struct iscsi_hdr *hdr, char *rx_data, int rx_data_len) +{ + int rc = 0; + uint32_t ret_itt; + int datalen; + int ahslen; + + /* verify PDU length */ + datalen = ntoh24(hdr->dlength); + if (datalen != rx_data_len) { + printk(KERN_ERR "iscsi_iser: datalen %d (hdr) != %d (IB) \n", + datalen, rx_data_len); + rc = ISCSI_ERR_DATALEN; + goto error; + } + + /* read AHS */ + ahslen = hdr->hlength * 4; + + /* verify itt (itt encoding: age+cid+itt) */ + rc = iscsi_verify_itt(conn, hdr, &ret_itt); + + if (!rc) + rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len); + + if (rc && rc != ISCSI_ERR_NO_SCSI_CMD) + goto error; + + return; +error: + iscsi_conn_failure(conn, rc); +} + + +/** + * iscsi_iser_cmd_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands + * + **/ +static void +iscsi_iser_cmd_init(struct iscsi_cmd_task *ctask) +{ + struct iscsi_iser_conn *iser_conn = ctask->conn->dd_data; + struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + struct scsi_cmnd *sc = ctask->sc; + + iser_ctask->command_sent = 0; + iser_ctask->iser_conn = iser_conn; + + if (sc->sc_data_direction == DMA_TO_DEVICE) { + BUG_ON(ctask->total_length == 0); + /* bytes to be sent via RDMA operations */ + iser_ctask->rdma_data_count = ctask->total_length - + ctask->imm_count - + ctask->unsol_count; + + debug_scsi("cmd [itt %x total %d imm %d unsol_data %d " + "rdma_data %d]\n", + ctask->itt, ctask->total_length, ctask->imm_count, + ctask->unsol_count, iser_ctask->rdma_data_count); + } else + /* bytes to be sent via RDMA operations */ + iser_ctask->rdma_data_count = ctask->total_length; + + iser_ctask_rdma_init(iser_ctask); +} + +/** + * iscsi_mtask_xmit - xmit management(immediate) task + * @conn: iscsi connection + * @mtask: task management task + * + * Notes: + * The function can return -EAGAIN in which case caller must + * call it again later, or recover. '0' return code means successful + * xmit. + * + **/ +static int +iscsi_iser_mtask_xmit(struct iscsi_conn *conn, + struct iscsi_mgmt_task *mtask) +{ + int error = 0; + + debug_scsi("mtask deq [cid %d itt 0x%x]\n", conn->id, mtask->itt); + + error = iser_send_control(conn, mtask); + + /* since iser xmits control with zero copy, mtasks can not be recycled + * right after sending them. + * The recycling scheme is based on whether a response is expected + * - if yes, the mtask is recycled at iscsi_complete_pdu + * - if no, the mtask is recycled at iser_snd_completion + */ + if (error && error != -EAGAIN) + iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED); + + return error; +} + +static int +iscsi_iser_ctask_xmit_unsol_data(struct iscsi_conn *conn, + struct iscsi_cmd_task *ctask) +{ + struct iscsi_data hdr; + int error = 0; + struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + + /* Send data-out PDUs while there's still unsolicited data to send */ + while (ctask->unsol_count > 0) { + iscsi_prep_unsolicit_data_pdu(ctask, &hdr, + iser_ctask->rdma_data_count); + + debug_scsi("Sending data-out: itt 0x%x, data count %d\n", + hdr.itt, ctask->data_count); + + /* the buffer description has been passed with the command */ + /* Send the command */ + error = iser_send_data_out(conn, ctask, &hdr); + if (error) { + ctask->unsol_datasn--; + goto iscsi_iser_ctask_xmit_unsol_data_exit; + } + ctask->unsol_count -= ctask->data_count; + debug_scsi("Need to send %d more as data-out PDUs\n", + ctask->unsol_count); + } + +iscsi_iser_ctask_xmit_unsol_data_exit: + return error; +} + +static int +iscsi_iser_ctask_xmit(struct iscsi_conn *conn, + struct iscsi_cmd_task *ctask) +{ + struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + int error = 0; + + debug_scsi("ctask deq [cid %d itt 0x%x]\n", + conn->id, ctask->itt); + + /* + * serialize with TMF AbortTask + */ + if (ctask->mtask) + return error; + + /* Send the cmd PDU */ + if (!iser_ctask->command_sent) { + error = iser_send_command(conn, ctask); + if (error) + goto iscsi_iser_ctask_xmit_exit; + iser_ctask->command_sent = 1; + } + + /* Send unsolicited data-out PDU(s) if necessary */ + if (ctask->unsol_count) + error = iscsi_iser_ctask_xmit_unsol_data(conn, ctask); + + iscsi_iser_ctask_xmit_exit: + if (error && error != -EAGAIN) + iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED); + return error; +} + +static void +iscsi_iser_cleanup_ctask(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask) +{ + struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + + if (iser_ctask->status == ISER_TASK_STATUS_STARTED) { + iser_ctask->status = ISER_TASK_STATUS_COMPLETED; + iser_ctask_rdma_finalize(iser_ctask); + } +} + +static struct iser_conn * +iscsi_iser_ib_conn_lookup(__u64 ep_handle) +{ + struct iser_conn *ib_conn; + struct iser_conn *uib_conn = (struct iser_conn *)(unsigned long)ep_handle; + + mutex_lock(&ig.connlist_mutex); + list_for_each_entry(ib_conn, &ig.connlist, conn_list) { + if (ib_conn == uib_conn) { + mutex_unlock(&ig.connlist_mutex); + return ib_conn; + } + } + mutex_unlock(&ig.connlist_mutex); + iser_err("no conn exists for eph %llx\n",(unsigned long long)ep_handle); + return NULL; +} + +static struct iscsi_cls_conn * +iscsi_iser_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx) +{ + struct iscsi_conn *conn; + struct iscsi_cls_conn *cls_conn; + struct iscsi_iser_conn *iser_conn; + + cls_conn = iscsi_conn_setup(cls_session, conn_idx); + if (!cls_conn) + return NULL; + conn = cls_conn->dd_data; + + /* + * due to issues with the login code re iser sematics + * this not set in iscsi_conn_setup - FIXME + */ + conn->max_recv_dlength = 128; + + iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL); + if (!iser_conn) + goto conn_alloc_fail; + + /* currently this is the only field which need to be initiated */ + rwlock_init(&iser_conn->lock); + + conn->dd_data = iser_conn; + iser_conn->iscsi_conn = conn; + + return cls_conn; + +conn_alloc_fail: + iscsi_conn_teardown(cls_conn); + return NULL; +} + +static void +iscsi_iser_conn_destroy(struct iscsi_cls_conn *cls_conn) +{ + struct iscsi_conn *conn = cls_conn->dd_data; + struct iscsi_iser_conn *iser_conn = conn->dd_data; + + iscsi_conn_teardown(cls_conn); + kfree(iser_conn); +} + +static int +iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session, + struct iscsi_cls_conn *cls_conn, uint64_t transport_eph, + int is_leading) +{ + struct iscsi_conn *conn = cls_conn->dd_data; + struct iscsi_iser_conn *iser_conn; + struct iser_conn *ib_conn; + int error; + + error = iscsi_conn_bind(cls_session, cls_conn, is_leading); + if (error) + return error; + + /* the transport ep handle comes from user space so it must be + * verified against the global ib connections list */ + ib_conn = iscsi_iser_ib_conn_lookup(transport_eph); + if (!ib_conn) { + iser_err("can't bind eph %llx\n", + (unsigned long long)transport_eph); + return -EINVAL; + } + /* binds the iSER connection retrieved from the previously + * connected ep_handle to the iSCSI layer connection. exchanges + * connection pointers */ + iser_err("binding iscsi conn %p to iser_conn %p\n",conn,ib_conn); + iser_conn = conn->dd_data; + ib_conn->iser_conn = iser_conn; + iser_conn->ib_conn = ib_conn; + + conn->recv_lock = &iser_conn->lock; + + return 0; +} + +static int +iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn) +{ + struct iscsi_conn *conn = cls_conn->dd_data; + int err; + + err = iscsi_conn_start(cls_conn); + if (err) + return err; + + return iser_conn_set_full_featured_mode(conn); +} + +static void +iscsi_iser_conn_terminate(struct iscsi_conn *conn) +{ + struct iscsi_iser_conn *iser_conn = conn->dd_data; + struct iser_conn *ib_conn = iser_conn->ib_conn; + + BUG_ON(!ib_conn); + /* starts conn teardown process, waits until all previously * + * posted buffers get flushed, deallocates all conn resources */ + iser_conn_terminate(ib_conn); + iser_conn->ib_conn = NULL; + conn->recv_lock = NULL; +} + + +static struct iscsi_transport iscsi_iser_transport; + +static struct iscsi_cls_session * +iscsi_iser_session_create(struct iscsi_transport *iscsit, + struct scsi_transport_template *scsit, + uint32_t initial_cmdsn, uint32_t *hostno) +{ + struct iscsi_cls_session *cls_session; + struct iscsi_session *session; + int i; + uint32_t hn; + struct iscsi_cmd_task *ctask; + struct iscsi_mgmt_task *mtask; + struct iscsi_iser_cmd_task *iser_ctask; + struct iser_desc *desc; + + cls_session = iscsi_session_setup(iscsit, scsit, + sizeof(struct iscsi_iser_cmd_task), + sizeof(struct iser_desc), + initial_cmdsn, &hn); + if (!cls_session) + return NULL; + + *hostno = hn; + session = class_to_transport_session(cls_session); + + /* libiscsi setup itts, data and pool so just set desc fields */ + for (i = 0; i < session->cmds_max; i++) { + ctask = session->cmds[i]; + iser_ctask = ctask->dd_data; + ctask->hdr = (struct iscsi_cmd *)&iser_ctask->desc.iscsi_header; + } + + for (i = 0; i < session->mgmtpool_max; i++) { + mtask = session->mgmt_cmds[i]; + desc = mtask->dd_data; + mtask->hdr = &desc->iscsi_header; + desc->data = mtask->data; + } + + return cls_session; +} + +static int +iscsi_iser_conn_set_param(struct iscsi_cls_conn *cls_conn, + enum iscsi_param param, uint32_t value) +{ + struct iscsi_conn *conn = cls_conn->dd_data; + struct iscsi_session *session = conn->session; + + spin_lock_bh(&session->lock); + if (conn->c_stage != ISCSI_CONN_INITIAL_STAGE && + conn->stop_stage != STOP_CONN_RECOVER) { + printk(KERN_ERR "iscsi_iser: can not change parameter [%d]\n", + param); + spin_unlock_bh(&session->lock); + return 0; + } + spin_unlock_bh(&session->lock); + + switch (param) { + case ISCSI_PARAM_MAX_RECV_DLENGTH: + /* TBD */ + break; + case ISCSI_PARAM_MAX_XMIT_DLENGTH: + conn->max_xmit_dlength = value; + break; + case ISCSI_PARAM_HDRDGST_EN: + if (value) { + printk(KERN_ERR "DataDigest wasn't negotiated to None"); + return -EPROTO; + } + break; + case ISCSI_PARAM_DATADGST_EN: + if (value) { + printk(KERN_ERR "DataDigest wasn't negotiated to None"); + return -EPROTO; + } + break; + case ISCSI_PARAM_INITIAL_R2T_EN: + session->initial_r2t_en = value; + break; + case ISCSI_PARAM_IMM_DATA_EN: + session->imm_data_en = value; + break; + case ISCSI_PARAM_FIRST_BURST: + session->first_burst = value; + break; + case ISCSI_PARAM_MAX_BURST: + session->max_burst = value; + break; + case ISCSI_PARAM_PDU_INORDER_EN: + session->pdu_inorder_en = value; + break; + case ISCSI_PARAM_DATASEQ_INORDER_EN: + session->dataseq_inorder_en = value; + break; + case ISCSI_PARAM_ERL: + session->erl = value; + break; + case ISCSI_PARAM_IFMARKER_EN: + if (value) { + printk(KERN_ERR "IFMarker wasn't negotiated to No"); + return -EPROTO; + } + break; + case ISCSI_PARAM_OFMARKER_EN: + if (value) { + printk(KERN_ERR "OFMarker wasn't negotiated to No"); + return -EPROTO; + } + break; + default: + break; + } + + return 0; +} + +static int +iscsi_iser_session_get_param(struct iscsi_cls_session *cls_session, + enum iscsi_param param, uint32_t *value) +{ + struct Scsi_Host *shost = iscsi_session_to_shost(cls_session); + struct iscsi_session *session = iscsi_hostdata(shost->hostdata); + + switch (param) { + case ISCSI_PARAM_INITIAL_R2T_EN: + *value = session->initial_r2t_en; + break; + case ISCSI_PARAM_MAX_R2T: + *value = session->max_r2t; + break; + case ISCSI_PARAM_IMM_DATA_EN: + *value = session->imm_data_en; + break; + case ISCSI_PARAM_FIRST_BURST: + *value = session->first_burst; + break; + case ISCSI_PARAM_MAX_BURST: + *value = session->max_burst; + break; + case ISCSI_PARAM_PDU_INORDER_EN: + *value = session->pdu_inorder_en; + break; + case ISCSI_PARAM_DATASEQ_INORDER_EN: + *value = session->dataseq_inorder_en; + break; + case ISCSI_PARAM_ERL: + *value = session->erl; + break; + case ISCSI_PARAM_IFMARKER_EN: + *value = 0; + break; + case ISCSI_PARAM_OFMARKER_EN: + *value = 0; + break; + default: + return ISCSI_ERR_PARAM_NOT_FOUND; + } + + return 0; +} + +static int +iscsi_iser_conn_get_param(struct iscsi_cls_conn *cls_conn, + enum iscsi_param param, uint32_t *value) +{ + struct iscsi_conn *conn = cls_conn->dd_data; + + switch(param) { + case ISCSI_PARAM_MAX_RECV_DLENGTH: + *value = conn->max_recv_dlength; + break; + case ISCSI_PARAM_MAX_XMIT_DLENGTH: + *value = conn->max_xmit_dlength; + break; + case ISCSI_PARAM_HDRDGST_EN: + *value = 0; + break; + case ISCSI_PARAM_DATADGST_EN: + *value = 0; + break; + /*case ISCSI_PARAM_TARGET_RECV_DLENGTH: + *value = conn->target_recv_dlength; + break; + case ISCSI_PARAM_INITIATOR_RECV_DLENGTH: + *value = conn->initiator_recv_dlength; + break;*/ + default: + return ISCSI_ERR_PARAM_NOT_FOUND; + } + + return 0; +} + + +static void +iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats) +{ + struct iscsi_conn *conn = cls_conn->dd_data; + + stats->txdata_octets = conn->txdata_octets; + stats->rxdata_octets = conn->rxdata_octets; + stats->scsicmd_pdus = conn->scsicmd_pdus_cnt; + stats->dataout_pdus = conn->dataout_pdus_cnt; + stats->scsirsp_pdus = conn->scsirsp_pdus_cnt; + stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */ + stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */ + stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt; + stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt; + stats->custom_length = 3; + strcpy(stats->custom[0].desc, "qp_tx_queue_full"); + stats->custom[0].value = 0; /* TB iser_conn->qp_tx_queue_full; */ + strcpy(stats->custom[1].desc, "fmr_map_not_avail"); + stats->custom[1].value = 0; /* TB iser_conn->fmr_map_not_avail */; + strcpy(stats->custom[2].desc, "eh_abort_cnt"); + stats->custom[2].value = conn->eh_abort_cnt; +} + +static int +iscsi_iser_ep_connect(struct sockaddr *dst_addr, int non_blocking, + __u64 *ep_handle) +{ + int err; + struct iser_conn *ib_conn; + + err = iser_conn_init(&ib_conn); + if (err) + goto out; + + err = iser_connect(ib_conn, NULL, (struct sockaddr_in *)dst_addr, non_blocking); + if (!err) + *ep_handle = (__u64)(unsigned long)ib_conn; + +out: + return err; +} + +static int +iscsi_iser_ep_poll(__u64 ep_handle, int timeout_ms) +{ + struct iser_conn *ib_conn = iscsi_iser_ib_conn_lookup(ep_handle); + int rc; + + if (!ib_conn) + return -EINVAL; + + rc = wait_event_interruptible_timeout(ib_conn->wait, + ib_conn->state == ISER_CONN_UP, + msecs_to_jiffies(timeout_ms)); + + /* if conn establishment failed, return error code to iscsi */ + if (!rc && + (ib_conn->state == ISER_CONN_TERMINATING || + ib_conn->state == ISER_CONN_DOWN)) + rc = -1; + + iser_err("ib conn %p rc = %d\n", ib_conn, rc); + + if (rc > 0) + return 1; /* success, this is the equivalent of POLLOUT */ + else if (!rc) + return 0; /* timeout */ + else + return rc; /* signal */ +} + +static void +iscsi_iser_ep_disconnect(__u64 ep_handle) +{ + struct iser_conn *ib_conn = iscsi_iser_ib_conn_lookup(ep_handle); + + if (!ib_conn) + return; + + iser_err("ib conn %p state %d\n",ib_conn, ib_conn->state); + + iser_conn_terminate(ib_conn); +} + +static struct scsi_host_template iscsi_iser_sht = { + .name = "iSCSI Initiator over iSER, v." DRV_VER, + .queuecommand = iscsi_queuecommand, + .can_queue = ISCSI_XMIT_CMDS_MAX - 1, + .sg_tablesize = ISCSI_ISER_SG_TABLESIZE, + .cmd_per_lun = ISCSI_MAX_CMD_PER_LUN, + .eh_abort_handler = iscsi_eh_abort, + .eh_host_reset_handler = iscsi_eh_host_reset, + .use_clustering = DISABLE_CLUSTERING, + .proc_name = "iscsi_iser", + .this_id = -1, +}; + +static struct iscsi_transport iscsi_iser_transport = { + .owner = THIS_MODULE, + .name = "iser", + .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T, + .param_mask = ISCSI_MAX_RECV_DLENGTH | + ISCSI_MAX_XMIT_DLENGTH | + ISCSI_HDRDGST_EN | + ISCSI_DATADGST_EN | + ISCSI_INITIAL_R2T_EN | + ISCSI_MAX_R2T | + ISCSI_IMM_DATA_EN | + ISCSI_FIRST_BURST | + ISCSI_MAX_BURST | + ISCSI_PDU_INORDER_EN | + ISCSI_DATASEQ_INORDER_EN, + .host_template = &iscsi_iser_sht, + .conndata_size = sizeof(struct iscsi_conn), + .max_lun = ISCSI_ISER_MAX_LUN, + .max_cmd_len = ISCSI_ISER_MAX_CMD_LEN, + /* session management */ + .create_session = iscsi_iser_session_create, + .destroy_session = iscsi_session_teardown, + /* connection management */ + .create_conn = iscsi_iser_conn_create, + .bind_conn = iscsi_iser_conn_bind, + .destroy_conn = iscsi_iser_conn_destroy, + .set_param = iscsi_iser_conn_set_param, + .get_conn_param = iscsi_iser_conn_get_param, + .get_session_param = iscsi_iser_session_get_param, + .start_conn = iscsi_iser_conn_start, + .stop_conn = iscsi_conn_stop, + /* these are called as part of conn recovery */ + .suspend_conn_recv = NULL, /* FIXME is/how this relvant to iser? */ + .terminate_conn = iscsi_iser_conn_terminate, + /* IO */ + .send_pdu = iscsi_conn_send_pdu, + .get_stats = iscsi_iser_conn_get_stats, + .init_cmd_task = iscsi_iser_cmd_init, + .xmit_cmd_task = iscsi_iser_ctask_xmit, + .xmit_mgmt_task = iscsi_iser_mtask_xmit, + .cleanup_cmd_task = iscsi_iser_cleanup_ctask, + /* recovery */ + .session_recovery_timedout = iscsi_session_recovery_timedout, + + .ep_connect = iscsi_iser_ep_connect, + .ep_poll = iscsi_iser_ep_poll, + .ep_disconnect = iscsi_iser_ep_disconnect +}; + +static int __init iser_init(void) +{ + int err; + + iser_dbg("Starting iSER datamover...\n"); + + if (iscsi_max_lun < 1) { + printk(KERN_ERR "Invalid max_lun value of %u\n", iscsi_max_lun); + return -EINVAL; + } + + iscsi_iser_transport.max_lun = iscsi_max_lun; + + memset(&ig, 0, sizeof(struct iser_global)); + + ig.desc_cache = kmem_cache_create("iser_descriptors", + sizeof (struct iser_desc), + 0, SLAB_HWCACHE_ALIGN, + NULL, NULL); + if (ig.desc_cache == NULL) + return -ENOMEM; + + /* device init is called only after the first addr resolution */ + mutex_init(&ig.device_list_mutex); + INIT_LIST_HEAD(&ig.device_list); + mutex_init(&ig.connlist_mutex); + INIT_LIST_HEAD(&ig.connlist); + + if (!iscsi_register_transport(&iscsi_iser_transport)) { + iser_err("iscsi_register_transport failed\n"); + err = -EINVAL; + goto register_transport_failure; + } + + return 0; + +register_transport_failure: + kmem_cache_destroy(ig.desc_cache); + + return err; +} + +static void __exit iser_exit(void) +{ + iser_dbg("Removing iSER datamover...\n"); + iscsi_unregister_transport(&iscsi_iser_transport); + kmem_cache_destroy(ig.desc_cache); +} + +module_init(iser_init); +module_exit(iser_exit); diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.h b/drivers/infiniband/ulp/iser/iscsi_iser.h new file mode 100644 index 000000000000..3350ba690cfe --- /dev/null +++ b/drivers/infiniband/ulp/iser/iscsi_iser.h @@ -0,0 +1,354 @@ +/* + * iSER transport for the Open iSCSI Initiator & iSER transport internals + * + * Copyright (C) 2004 Dmitry Yusupov + * Copyright (C) 2004 Alex Aizman + * Copyright (C) 2005 Mike Christie + * based on code maintained by open-iscsi@googlegroups.com + * + * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved. + * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * $Id: iscsi_iser.h 7051 2006-05-10 12:29:11Z ogerlitz $ + */ +#ifndef __ISCSI_ISER_H__ +#define __ISCSI_ISER_H__ + +#include <linux/types.h> +#include <linux/net.h> +#include <scsi/libiscsi.h> +#include <scsi/scsi_transport_iscsi.h> + +#include <linux/wait.h> +#include <linux/sched.h> +#include <linux/list.h> +#include <linux/slab.h> +#include <linux/dma-mapping.h> +#include <linux/mutex.h> +#include <linux/mempool.h> +#include <linux/uio.h> + +#include <linux/socket.h> +#include <linux/in.h> +#include <linux/in6.h> + +#include <rdma/ib_verbs.h> +#include <rdma/ib_fmr_pool.h> +#include <rdma/rdma_cm.h> + +#define DRV_NAME "iser" +#define PFX DRV_NAME ": " +#define DRV_VER "0.1" +#define DRV_DATE "May 7th, 2006" + +#define iser_dbg(fmt, arg...) \ + do { \ + if (iser_debug_level > 0) \ + printk(KERN_DEBUG PFX "%s:" fmt,\ + __func__ , ## arg); \ + } while (0) + +#define iser_err(fmt, arg...) \ + do { \ + printk(KERN_ERR PFX "%s:" fmt, \ + __func__ , ## arg); \ + } while (0) + + /* support upto 512KB in one RDMA */ +#define ISCSI_ISER_SG_TABLESIZE (0x80000 >> PAGE_SHIFT) +#define ISCSI_ISER_MAX_LUN 256 +#define ISCSI_ISER_MAX_CMD_LEN 16 + +/* QP settings */ +/* Maximal bounds on received asynchronous PDUs */ +#define ISER_MAX_RX_MISC_PDUS 4 /* NOOP_IN(2) , ASYNC_EVENT(2) */ + +#define ISER_MAX_TX_MISC_PDUS 6 /* NOOP_OUT(2), TEXT(1), * + * SCSI_TMFUNC(2), LOGOUT(1) */ + +#define ISER_QP_MAX_RECV_DTOS (ISCSI_XMIT_CMDS_MAX + \ + ISER_MAX_RX_MISC_PDUS + \ + ISER_MAX_TX_MISC_PDUS) + +/* the max TX (send) WR supported by the iSER QP is defined by * + * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect * + * to have at max for SCSI command. The tx posting & completion handling code * + * supports -EAGAIN scheme where tx is suspended till the QP has room for more * + * send WR. D=8 comes from 64K/8K */ + +#define ISER_INFLIGHT_DATAOUTS 8 + +#define ISER_QP_MAX_REQ_DTOS (ISCSI_XMIT_CMDS_MAX * \ + (1 + ISER_INFLIGHT_DATAOUTS) + \ + ISER_MAX_TX_MISC_PDUS + \ + ISER_MAX_RX_MISC_PDUS) + +#define ISER_VER 0x10 +#define ISER_WSV 0x08 +#define ISER_RSV 0x04 + +struct iser_hdr { + u8 flags; + u8 rsvd[3]; + __be32 write_stag; /* write rkey */ + __be64 write_va; + __be32 read_stag; /* read rkey */ + __be64 read_va; +} __attribute__((packed)); + + +/* Length of an object name string */ +#define ISER_OBJECT_NAME_SIZE 64 + +enum iser_ib_conn_state { + ISER_CONN_INIT, /* descriptor allocd, no conn */ + ISER_CONN_PENDING, /* in the process of being established */ + ISER_CONN_UP, /* up and running */ + ISER_CONN_TERMINATING, /* in the process of being terminated */ + ISER_CONN_DOWN, /* shut down */ + ISER_CONN_STATES_NUM +}; + +enum iser_task_status { + ISER_TASK_STATUS_INIT = 0, + ISER_TASK_STATUS_STARTED, + ISER_TASK_STATUS_COMPLETED +}; + +enum iser_data_dir { + ISER_DIR_IN = 0, /* to initiator */ + ISER_DIR_OUT, /* from initiator */ + ISER_DIRS_NUM +}; + +struct iser_data_buf { + void *buf; /* pointer to the sg list */ + unsigned int size; /* num entries of this sg */ + unsigned long data_len; /* total data len */ + unsigned int dma_nents; /* returned by dma_map_sg */ + char *copy_buf; /* allocated copy buf for SGs unaligned * + * for rdma which are copied */ + struct scatterlist sg_single; /* SG-ified clone of a non SG SC or * + * unaligned SG */ + }; + +/* fwd declarations */ +struct iser_device; +struct iscsi_iser_conn; +struct iscsi_iser_cmd_task; + +struct iser_mem_reg { + u32 lkey; + u32 rkey; + u64 va; + u64 len; + void *mem_h; +}; + +struct iser_regd_buf { + struct iser_mem_reg reg; /* memory registration info */ + void *virt_addr; + struct iser_device *device; /* device->device for dma_unmap */ + dma_addr_t dma_addr; /* if non zero, addr for dma_unmap */ + enum dma_data_direction direction; /* direction for dma_unmap */ + unsigned int data_size; + atomic_t ref_count; /* refcount, freed when dec to 0 */ +}; + +#define MAX_REGD_BUF_VECTOR_LEN 2 + +struct iser_dto { + struct iscsi_iser_cmd_task *ctask; + struct iscsi_iser_conn *conn; + int notify_enable; + + /* vector of registered buffers */ + unsigned int regd_vector_len; + struct iser_regd_buf *regd[MAX_REGD_BUF_VECTOR_LEN]; + + /* offset into the registered buffer may be specified */ + unsigned int offset[MAX_REGD_BUF_VECTOR_LEN]; + + /* a smaller size may be specified, if 0, then full size is used */ + unsigned int used_sz[MAX_REGD_BUF_VECTOR_LEN]; +}; + +enum iser_desc_type { + ISCSI_RX, + ISCSI_TX_CONTROL , + ISCSI_TX_SCSI_COMMAND, + ISCSI_TX_DATAOUT +}; + +struct iser_desc { + struct iser_hdr iser_header; + struct iscsi_hdr iscsi_header; + struct iser_regd_buf hdr_regd_buf; + void *data; /* used by RX & TX_CONTROL */ + struct iser_regd_buf data_regd_buf; /* used by RX & TX_CONTROL */ + enum iser_desc_type type; + struct iser_dto dto; +}; + +struct iser_device { + struct ib_device *ib_device; + struct ib_pd *pd; + struct ib_cq *cq; + struct ib_mr *mr; + struct tasklet_struct cq_tasklet; + struct list_head ig_list; /* entry in ig devices list */ + int refcount; +}; + +struct iser_conn { + struct iscsi_iser_conn *iser_conn; /* iser conn for upcalls */ + enum iser_ib_conn_state state; /* rdma connection state */ + spinlock_t lock; /* used for state changes */ + struct iser_device *device; /* device context */ + struct rdma_cm_id *cma_id; /* CMA ID */ + struct ib_qp *qp; /* QP */ + struct ib_fmr_pool *fmr_pool; /* pool of IB FMRs */ + int disc_evt_flag; /* disconn event delivered */ + wait_queue_head_t wait; /* waitq for conn/disconn */ + atomic_t post_recv_buf_count; /* posted rx count */ + atomic_t post_send_buf_count; /* posted tx count */ + struct work_struct comperror_work; /* conn term sleepable ctx*/ + char name[ISER_OBJECT_NAME_SIZE]; + struct iser_page_vec *page_vec; /* represents SG to fmr maps* + * maps serialized as tx is*/ + struct list_head conn_list; /* entry in ig conn list */ +}; + +struct iscsi_iser_conn { + struct iscsi_conn *iscsi_conn;/* ptr to iscsi conn */ + struct iser_conn *ib_conn; /* iSER IB conn */ + + rwlock_t lock; +}; + +struct iscsi_iser_cmd_task { + struct iser_desc desc; + struct iscsi_iser_conn *iser_conn; + int rdma_data_count;/* RDMA bytes */ + enum iser_task_status status; + int command_sent; /* set if command sent */ + int dir[ISER_DIRS_NUM]; /* set if dir use*/ + struct iser_regd_buf rdma_regd[ISER_DIRS_NUM];/* regd rdma buf */ + struct iser_data_buf data[ISER_DIRS_NUM]; /* orig. data des*/ + struct iser_data_buf data_copy[ISER_DIRS_NUM];/* contig. copy */ +}; + +struct iser_page_vec { + u64 *pages; + int length; + int offset; + int data_size; +}; + +struct iser_global { + struct mutex device_list_mutex;/* */ + struct list_head device_list; /* all iSER devices */ + struct mutex connlist_mutex; + struct list_head connlist; /* all iSER IB connections */ + + kmem_cache_t *desc_cache; +}; + +extern struct iser_global ig; +extern int iser_debug_level; + +/* allocate connection resources needed for rdma functionality */ +int iser_conn_set_full_featured_mode(struct iscsi_conn *conn); + +int iser_send_control(struct iscsi_conn *conn, + struct iscsi_mgmt_task *mtask); + +int iser_send_command(struct iscsi_conn *conn, + struct iscsi_cmd_task *ctask); + +int iser_send_data_out(struct iscsi_conn *conn, + struct iscsi_cmd_task *ctask, + struct iscsi_data *hdr); + +void iscsi_iser_recv(struct iscsi_conn *conn, + struct iscsi_hdr *hdr, + char *rx_data, + int rx_data_len); + +int iser_conn_init(struct iser_conn **ib_conn); + +void iser_conn_terminate(struct iser_conn *ib_conn); + +void iser_conn_release(struct iser_conn *ib_conn); + +void iser_rcv_completion(struct iser_desc *desc, + unsigned long dto_xfer_len); + +void iser_snd_completion(struct iser_desc *desc); + +void iser_ctask_rdma_init(struct iscsi_iser_cmd_task *ctask); + +void iser_ctask_rdma_finalize(struct iscsi_iser_cmd_task *ctask); + +void iser_dto_buffs_release(struct iser_dto *dto); + +int iser_regd_buff_release(struct iser_regd_buf *regd_buf); + +void iser_reg_single(struct iser_device *device, + struct iser_regd_buf *regd_buf, + enum dma_data_direction direction); + +int iser_start_rdma_unaligned_sg(struct iscsi_iser_cmd_task *ctask, + enum iser_data_dir cmd_dir); + +void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_cmd_task *ctask, + enum iser_data_dir cmd_dir); + +int iser_reg_rdma_mem(struct iscsi_iser_cmd_task *ctask, + enum iser_data_dir cmd_dir); + +int iser_connect(struct iser_conn *ib_conn, + struct sockaddr_in *src_addr, + struct sockaddr_in *dst_addr, + int non_blocking); + +int iser_reg_page_vec(struct iser_conn *ib_conn, + struct iser_page_vec *page_vec, + struct iser_mem_reg *mem_reg); + +void iser_unreg_mem(struct iser_mem_reg *mem_reg); + +int iser_post_recv(struct iser_desc *rx_desc); +int iser_post_send(struct iser_desc *tx_desc); + +int iser_conn_state_comp(struct iser_conn *ib_conn, + enum iser_ib_conn_state comp); +#endif diff --git a/drivers/infiniband/ulp/iser/iser_initiator.c b/drivers/infiniband/ulp/iser/iser_initiator.c new file mode 100644 index 000000000000..ccf56f6f7236 --- /dev/null +++ b/drivers/infiniband/ulp/iser/iser_initiator.c @@ -0,0 +1,738 @@ +/* + * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * $Id: iser_initiator.c 6964 2006-05-07 11:11:43Z ogerlitz $ + */ +#include <linux/kernel.h> +#include <linux/slab.h> +#include <linux/mm.h> +#include <asm/io.h> +#include <asm/scatterlist.h> +#include <linux/scatterlist.h> +#include <linux/kfifo.h> +#include <scsi/scsi_cmnd.h> +#include <scsi/scsi_host.h> + +#include "iscsi_iser.h" + +/* Constant PDU lengths calculations */ +#define ISER_TOTAL_HEADERS_LEN (sizeof (struct iser_hdr) + \ + sizeof (struct iscsi_hdr)) + +/* iser_dto_add_regd_buff - increments the reference count for * + * the registered buffer & adds it to the DTO object */ +static void iser_dto_add_regd_buff(struct iser_dto *dto, + struct iser_regd_buf *regd_buf, + unsigned long use_offset, + unsigned long use_size) +{ + int add_idx; + + atomic_inc(®d_buf->ref_count); + + add_idx = dto->regd_vector_len; + dto->regd[add_idx] = regd_buf; + dto->used_sz[add_idx] = use_size; + dto->offset[add_idx] = use_offset; + + dto->regd_vector_len++; +} + +static int iser_dma_map_task_data(struct iscsi_iser_cmd_task *iser_ctask, + struct iser_data_buf *data, + enum iser_data_dir iser_dir, + enum dma_data_direction dma_dir) +{ + struct device *dma_device; + + iser_ctask->dir[iser_dir] = 1; + dma_device = iser_ctask->iser_conn->ib_conn->device->ib_device->dma_device; + + data->dma_nents = dma_map_sg(dma_device, data->buf, data->size, dma_dir); + if (data->dma_nents == 0) { + iser_err("dma_map_sg failed!!!\n"); + return -EINVAL; + } + return 0; +} + +static void iser_dma_unmap_task_data(struct iscsi_iser_cmd_task *iser_ctask) +{ + struct device *dma_device; + struct iser_data_buf *data; + + dma_device = iser_ctask->iser_conn->ib_conn->device->ib_device->dma_device; + + if (iser_ctask->dir[ISER_DIR_IN]) { + data = &iser_ctask->data[ISER_DIR_IN]; + dma_unmap_sg(dma_device, data->buf, data->size, DMA_FROM_DEVICE); + } + + if (iser_ctask->dir[ISER_DIR_OUT]) { + data = &iser_ctask->data[ISER_DIR_OUT]; + dma_unmap_sg(dma_device, data->buf, data->size, DMA_TO_DEVICE); + } +} + +/* Register user buffer memory and initialize passive rdma + * dto descriptor. Total data size is stored in + * iser_ctask->data[ISER_DIR_IN].data_len + */ +static int iser_prepare_read_cmd(struct iscsi_cmd_task *ctask, + unsigned int edtl) + +{ + struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + struct iser_regd_buf *regd_buf; + int err; + struct iser_hdr *hdr = &iser_ctask->desc.iser_header; + struct iser_data_buf *buf_in = &iser_ctask->data[ISER_DIR_IN]; + + err = iser_dma_map_task_data(iser_ctask, + buf_in, + ISER_DIR_IN, + DMA_FROM_DEVICE); + if (err) + return err; + + if (edtl > iser_ctask->data[ISER_DIR_IN].data_len) { + iser_err("Total data length: %ld, less than EDTL: " + "%d, in READ cmd BHS itt: %d, conn: 0x%p\n", + iser_ctask->data[ISER_DIR_IN].data_len, edtl, + ctask->itt, iser_ctask->iser_conn); + return -EINVAL; + } + + err = iser_reg_rdma_mem(iser_ctask,ISER_DIR_IN); + if (err) { + iser_err("Failed to set up Data-IN RDMA\n"); + return err; + } + regd_buf = &iser_ctask->rdma_regd[ISER_DIR_IN]; + + hdr->flags |= ISER_RSV; + hdr->read_stag = cpu_to_be32(regd_buf->reg.rkey); + hdr->read_va = cpu_to_be64(regd_buf->reg.va); + + iser_dbg("Cmd itt:%d READ tags RKEY:%#.4X VA:%#llX\n", + ctask->itt, regd_buf->reg.rkey, + (unsigned long long)regd_buf->reg.va); + + return 0; +} + +/* Register user buffer memory and initialize passive rdma + * dto descriptor. Total data size is stored in + * ctask->data[ISER_DIR_OUT].data_len + */ +static int +iser_prepare_write_cmd(struct iscsi_cmd_task *ctask, + unsigned int imm_sz, + unsigned int unsol_sz, + unsigned int edtl) +{ + struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + struct iser_regd_buf *regd_buf; + int err; + struct iser_dto *send_dto = &iser_ctask->desc.dto; + struct iser_hdr *hdr = &iser_ctask->desc.iser_header; + struct iser_data_buf *buf_out = &iser_ctask->data[ISER_DIR_OUT]; + + err = iser_dma_map_task_data(iser_ctask, + buf_out, + ISER_DIR_OUT, + DMA_TO_DEVICE); + if (err) + return err; + + if (edtl > iser_ctask->data[ISER_DIR_OUT].data_len) { + iser_err("Total data length: %ld, less than EDTL: %d, " + "in WRITE cmd BHS itt: %d, conn: 0x%p\n", + iser_ctask->data[ISER_DIR_OUT].data_len, + edtl, ctask->itt, ctask->conn); + return -EINVAL; + } + + err = iser_reg_rdma_mem(iser_ctask,ISER_DIR_OUT); + if (err != 0) { + iser_err("Failed to register write cmd RDMA mem\n"); + return err; + } + + regd_buf = &iser_ctask->rdma_regd[ISER_DIR_OUT]; + + if (unsol_sz < edtl) { + hdr->flags |= ISER_WSV; + hdr->write_stag = cpu_to_be32(regd_buf->reg.rkey); + hdr->write_va = cpu_to_be64(regd_buf->reg.va + unsol_sz); + + iser_dbg("Cmd itt:%d, WRITE tags, RKEY:%#.4X " + "VA:%#llX + unsol:%d\n", + ctask->itt, regd_buf->reg.rkey, + (unsigned long long)regd_buf->reg.va, unsol_sz); + } + + if (imm_sz > 0) { + iser_dbg("Cmd itt:%d, WRITE, adding imm.data sz: %d\n", + ctask->itt, imm_sz); + iser_dto_add_regd_buff(send_dto, + regd_buf, + 0, + imm_sz); + } + + return 0; +} + +/** + * iser_post_receive_control - allocates, initializes and posts receive DTO. + */ +static int iser_post_receive_control(struct iscsi_conn *conn) +{ + struct iscsi_iser_conn *iser_conn = conn->dd_data; + struct iser_desc *rx_desc; + struct iser_regd_buf *regd_hdr; + struct iser_regd_buf *regd_data; + struct iser_dto *recv_dto = NULL; + struct iser_device *device = iser_conn->ib_conn->device; + int rx_data_size, err = 0; + + rx_desc = kmem_cache_alloc(ig.desc_cache, GFP_NOIO); + if (rx_desc == NULL) { + iser_err("Failed to alloc desc for post recv\n"); + return -ENOMEM; + } + rx_desc->type = ISCSI_RX; + + /* for the login sequence we must support rx of upto 8K; login is done + * after conn create/bind (connect) and conn stop/bind (reconnect), + * what's common for both schemes is that the connection is not started + */ + if (conn->c_stage != ISCSI_CONN_STARTED) + rx_data_size = DEFAULT_MAX_RECV_DATA_SEGMENT_LENGTH; + else /* FIXME till user space sets conn->max_recv_dlength correctly */ + rx_data_size = 128; + + rx_desc->data = kmalloc(rx_data_size, GFP_NOIO); + if (rx_desc->data == NULL) { + iser_err("Failed to alloc data buf for post recv\n"); + err = -ENOMEM; + goto post_rx_kmalloc_failure; + } + + recv_dto = &rx_desc->dto; + recv_dto->conn = iser_conn; + recv_dto->regd_vector_len = 0; + + regd_hdr = &rx_desc->hdr_regd_buf; + memset(regd_hdr, 0, sizeof(struct iser_regd_buf)); + regd_hdr->device = device; + regd_hdr->virt_addr = rx_desc; /* == &rx_desc->iser_header */ + regd_hdr->data_size = ISER_TOTAL_HEADERS_LEN; + + iser_reg_single(device, regd_hdr, DMA_FROM_DEVICE); + + iser_dto_add_regd_buff(recv_dto, regd_hdr, 0, 0); + + regd_data = &rx_desc->data_regd_buf; + memset(regd_data, 0, sizeof(struct iser_regd_buf)); + regd_data->device = device; + regd_data->virt_addr = rx_desc->data; + regd_data->data_size = rx_data_size; + + iser_reg_single(device, regd_data, DMA_FROM_DEVICE); + + iser_dto_add_regd_buff(recv_dto, regd_data, 0, 0); + + err = iser_post_recv(rx_desc); + if (!err) + return 0; + + /* iser_post_recv failed */ + iser_dto_buffs_release(recv_dto); + kfree(rx_desc->data); +post_rx_kmalloc_failure: + kmem_cache_free(ig.desc_cache, rx_desc); + return err; +} + +/* creates a new tx descriptor and adds header regd buffer */ +static void iser_create_send_desc(struct iscsi_iser_conn *iser_conn, + struct iser_desc *tx_desc) +{ + struct iser_regd_buf *regd_hdr = &tx_desc->hdr_regd_buf; + struct iser_dto *send_dto = &tx_desc->dto; + + memset(regd_hdr, 0, sizeof(struct iser_regd_buf)); + regd_hdr->device = iser_conn->ib_conn->device; + regd_hdr->virt_addr = tx_desc; /* == &tx_desc->iser_header */ + regd_hdr->data_size = ISER_TOTAL_HEADERS_LEN; + + send_dto->conn = iser_conn; + send_dto->notify_enable = 1; + send_dto->regd_vector_len = 0; + + memset(&tx_desc->iser_header, 0, sizeof(struct iser_hdr)); + tx_desc->iser_header.flags = ISER_VER; + + iser_dto_add_regd_buff(send_dto, regd_hdr, 0, 0); +} + +/** + * iser_conn_set_full_featured_mode - (iSER API) + */ +int iser_conn_set_full_featured_mode(struct iscsi_conn *conn) +{ + struct iscsi_iser_conn *iser_conn = conn->dd_data; + + int i; + /* no need to keep it in a var, we are after login so if this should + * be negotiated, by now the result should be available here */ + int initial_post_recv_bufs_num = ISER_MAX_RX_MISC_PDUS; + + iser_dbg("Initially post: %d\n", initial_post_recv_bufs_num); + + /* Check that there is no posted recv or send buffers left - */ + /* they must be consumed during the login phase */ + BUG_ON(atomic_read(&iser_conn->ib_conn->post_recv_buf_count) != 0); + BUG_ON(atomic_read(&iser_conn->ib_conn->post_send_buf_count) != 0); + + /* Initial post receive buffers */ + for (i = 0; i < initial_post_recv_bufs_num; i++) { + if (iser_post_receive_control(conn) != 0) { + iser_err("Failed to post recv bufs at:%d conn:0x%p\n", + i, conn); + return -ENOMEM; + } + } + iser_dbg("Posted %d post recv bufs, conn:0x%p\n", i, conn); + return 0; +} + +static int +iser_check_xmit(struct iscsi_conn *conn, void *task) +{ + int rc = 0; + struct iscsi_iser_conn *iser_conn = conn->dd_data; + + write_lock_bh(conn->recv_lock); + if (atomic_read(&iser_conn->ib_conn->post_send_buf_count) == + ISER_QP_MAX_REQ_DTOS) { + iser_dbg("%ld can't xmit task %p, suspending tx\n",jiffies,task); + set_bit(ISCSI_SUSPEND_BIT, &conn->suspend_tx); + rc = -EAGAIN; + } + write_unlock_bh(conn->recv_lock); + return rc; +} + + +/** + * iser_send_command - send command PDU + */ +int iser_send_command(struct iscsi_conn *conn, + struct iscsi_cmd_task *ctask) +{ + struct iscsi_iser_conn *iser_conn = conn->dd_data; + struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + struct iser_dto *send_dto = NULL; + unsigned long edtl; + int err = 0; + struct iser_data_buf *data_buf; + + struct iscsi_cmd *hdr = ctask->hdr; + struct scsi_cmnd *sc = ctask->sc; + + if (!iser_conn_state_comp(iser_conn->ib_conn, ISER_CONN_UP)) { + iser_err("Failed to send, conn: 0x%p is not up\n", iser_conn->ib_conn); + return -EPERM; + } + if (iser_check_xmit(conn, ctask)) + return -EAGAIN; + + edtl = ntohl(hdr->data_length); + + /* build the tx desc regd header and add it to the tx desc dto */ + iser_ctask->desc.type = ISCSI_TX_SCSI_COMMAND; + send_dto = &iser_ctask->desc.dto; + send_dto->ctask = iser_ctask; + iser_create_send_desc(iser_conn, &iser_ctask->desc); + + if (hdr->flags & ISCSI_FLAG_CMD_READ) + data_buf = &iser_ctask->data[ISER_DIR_IN]; + else + data_buf = &iser_ctask->data[ISER_DIR_OUT]; + + if (sc->use_sg) { /* using a scatter list */ + data_buf->buf = sc->request_buffer; + data_buf->size = sc->use_sg; + } else if (sc->request_bufflen) { + /* using a single buffer - convert it into one entry SG */ + sg_init_one(&data_buf->sg_single, + sc->request_buffer, sc->request_bufflen); + data_buf->buf = &data_buf->sg_single; + data_buf->size = 1; + } + + data_buf->data_len = sc->request_bufflen; + + if (hdr->flags & ISCSI_FLAG_CMD_READ) { + err = iser_prepare_read_cmd(ctask, edtl); + if (err) + goto send_command_error; + } + if (hdr->flags & ISCSI_FLAG_CMD_WRITE) { + err = iser_prepare_write_cmd(ctask, + ctask->imm_count, + ctask->imm_count + + ctask->unsol_count, + edtl); + if (err) + goto send_command_error; + } + + iser_reg_single(iser_conn->ib_conn->device, + send_dto->regd[0], DMA_TO_DEVICE); + + if (iser_post_receive_control(conn) != 0) { + iser_err("post_recv failed!\n"); + err = -ENOMEM; + goto send_command_error; + } + + iser_ctask->status = ISER_TASK_STATUS_STARTED; + + err = iser_post_send(&iser_ctask->desc); + if (!err) + return 0; + +send_command_error: + iser_dto_buffs_release(send_dto); + iser_err("conn %p failed ctask->itt %d err %d\n",conn, ctask->itt, err); + return err; +} + +/** + * iser_send_data_out - send data out PDU + */ +int iser_send_data_out(struct iscsi_conn *conn, + struct iscsi_cmd_task *ctask, + struct iscsi_data *hdr) +{ + struct iscsi_iser_conn *iser_conn = conn->dd_data; + struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + struct iser_desc *tx_desc = NULL; + struct iser_dto *send_dto = NULL; + unsigned long buf_offset; + unsigned long data_seg_len; + unsigned int itt; + int err = 0; + + if (!iser_conn_state_comp(iser_conn->ib_conn, ISER_CONN_UP)) { + iser_err("Failed to send, conn: 0x%p is not up\n", iser_conn->ib_conn); + return -EPERM; + } + + if (iser_check_xmit(conn, ctask)) + return -EAGAIN; + + itt = ntohl(hdr->itt); + data_seg_len = ntoh24(hdr->dlength); + buf_offset = ntohl(hdr->offset); + + iser_dbg("%s itt %d dseg_len %d offset %d\n", + __func__,(int)itt,(int)data_seg_len,(int)buf_offset); + + tx_desc = kmem_cache_alloc(ig.desc_cache, GFP_NOIO); + if (tx_desc == NULL) { + iser_err("Failed to alloc desc for post dataout\n"); + return -ENOMEM; + } + + tx_desc->type = ISCSI_TX_DATAOUT; + memcpy(&tx_desc->iscsi_header, hdr, sizeof(struct iscsi_hdr)); + + /* build the tx desc regd header and add it to the tx desc dto */ + send_dto = &tx_desc->dto; + send_dto->ctask = iser_ctask; + iser_create_send_desc(iser_conn, tx_desc); + + iser_reg_single(iser_conn->ib_conn->device, + send_dto->regd[0], DMA_TO_DEVICE); + + /* all data was registered for RDMA, we can use the lkey */ + iser_dto_add_regd_buff(send_dto, + &iser_ctask->rdma_regd[ISER_DIR_OUT], + buf_offset, + data_seg_len); + + if (buf_offset + data_seg_len > iser_ctask->data[ISER_DIR_OUT].data_len) { + iser_err("Offset:%ld & DSL:%ld in Data-Out " + "inconsistent with total len:%ld, itt:%d\n", + buf_offset, data_seg_len, + iser_ctask->data[ISER_DIR_OUT].data_len, itt); + err = -EINVAL; + goto send_data_out_error; + } + iser_dbg("data-out itt: %d, offset: %ld, sz: %ld\n", + itt, buf_offset, data_seg_len); + + + err = iser_post_send(tx_desc); + if (!err) + return 0; + +send_data_out_error: + iser_dto_buffs_release(send_dto); + kmem_cache_free(ig.desc_cache, tx_desc); + iser_err("conn %p failed err %d\n",conn, err); + return err; +} + +int iser_send_control(struct iscsi_conn *conn, + struct iscsi_mgmt_task *mtask) +{ + struct iscsi_iser_conn *iser_conn = conn->dd_data; + struct iser_desc *mdesc = mtask->dd_data; + struct iser_dto *send_dto = NULL; + unsigned int itt; + unsigned long data_seg_len; + int err = 0; + unsigned char opcode; + struct iser_regd_buf *regd_buf; + struct iser_device *device; + + if (!iser_conn_state_comp(iser_conn->ib_conn, ISER_CONN_UP)) { + iser_err("Failed to send, conn: 0x%p is not up\n", iser_conn->ib_conn); + return -EPERM; + } + + if (iser_check_xmit(conn,mtask)) + return -EAGAIN; + + /* build the tx desc regd header and add it to the tx desc dto */ + mdesc->type = ISCSI_TX_CONTROL; + send_dto = &mdesc->dto; + send_dto->ctask = NULL; + iser_create_send_desc(iser_conn, mdesc); + + device = iser_conn->ib_conn->device; + + iser_reg_single(device, send_dto->regd[0], DMA_TO_DEVICE); + + itt = ntohl(mtask->hdr->itt); + opcode = mtask->hdr->opcode & ISCSI_OPCODE_MASK; + data_seg_len = ntoh24(mtask->hdr->dlength); + + if (data_seg_len > 0) { + regd_buf = &mdesc->data_regd_buf; + memset(regd_buf, 0, sizeof(struct iser_regd_buf)); + regd_buf->device = device; + regd_buf->virt_addr = mtask->data; + regd_buf->data_size = mtask->data_count; + iser_reg_single(device, regd_buf, + DMA_TO_DEVICE); + iser_dto_add_regd_buff(send_dto, regd_buf, + 0, + data_seg_len); + } + + if (iser_post_receive_control(conn) != 0) { + iser_err("post_rcv_buff failed!\n"); + err = -ENOMEM; + goto send_control_error; + } + + err = iser_post_send(mdesc); + if (!err) + return 0; + +send_control_error: + iser_dto_buffs_release(send_dto); + iser_err("conn %p failed err %d\n",conn, err); + return err; +} + +/** + * iser_rcv_dto_completion - recv DTO completion + */ +void iser_rcv_completion(struct iser_desc *rx_desc, + unsigned long dto_xfer_len) +{ + struct iser_dto *dto = &rx_desc->dto; + struct iscsi_iser_conn *conn = dto->conn; + struct iscsi_session *session = conn->iscsi_conn->session; + struct iscsi_cmd_task *ctask; + struct iscsi_iser_cmd_task *iser_ctask; + struct iscsi_hdr *hdr; + char *rx_data = NULL; + int rx_data_len = 0; + unsigned int itt; + unsigned char opcode; + + hdr = &rx_desc->iscsi_header; + + iser_dbg("op 0x%x itt 0x%x\n", hdr->opcode,hdr->itt); + + if (dto_xfer_len > ISER_TOTAL_HEADERS_LEN) { /* we have data */ + rx_data_len = dto_xfer_len - ISER_TOTAL_HEADERS_LEN; + rx_data = dto->regd[1]->virt_addr; + rx_data += dto->offset[1]; + } + + opcode = hdr->opcode & ISCSI_OPCODE_MASK; + + if (opcode == ISCSI_OP_SCSI_CMD_RSP) { + itt = hdr->itt & ISCSI_ITT_MASK; /* mask out cid and age bits */ + if (!(itt < session->cmds_max)) + iser_err("itt can't be matched to task!!!" + "conn %p opcode %d cmds_max %d itt %d\n", + conn->iscsi_conn,opcode,session->cmds_max,itt); + /* use the mapping given with the cmds array indexed by itt */ + ctask = (struct iscsi_cmd_task *)session->cmds[itt]; + iser_ctask = ctask->dd_data; + iser_dbg("itt %d ctask %p\n",itt,ctask); + iser_ctask->status = ISER_TASK_STATUS_COMPLETED; + iser_ctask_rdma_finalize(iser_ctask); + } + + iser_dto_buffs_release(dto); + + iscsi_iser_recv(conn->iscsi_conn, hdr, rx_data, rx_data_len); + + kfree(rx_desc->data); + kmem_cache_free(ig.desc_cache, rx_desc); + + /* decrementing conn->post_recv_buf_count only --after-- freeing the * + * task eliminates the need to worry on tasks which are completed in * + * parallel to the execution of iser_conn_term. So the code that waits * + * for the posted rx bufs refcount to become zero handles everything */ + atomic_dec(&conn->ib_conn->post_recv_buf_count); +} + +void iser_snd_completion(struct iser_desc *tx_desc) +{ + struct iser_dto *dto = &tx_desc->dto; + struct iscsi_iser_conn *iser_conn = dto->conn; + struct iscsi_conn *conn = iser_conn->iscsi_conn; + struct iscsi_mgmt_task *mtask; + + iser_dbg("Initiator, Data sent dto=0x%p\n", dto); + + iser_dto_buffs_release(dto); + + if (tx_desc->type == ISCSI_TX_DATAOUT) + kmem_cache_free(ig.desc_cache, tx_desc); + + atomic_dec(&iser_conn->ib_conn->post_send_buf_count); + + write_lock(conn->recv_lock); + if (conn->suspend_tx) { + iser_dbg("%ld resuming tx\n",jiffies); + clear_bit(ISCSI_SUSPEND_BIT, &conn->suspend_tx); + scsi_queue_work(conn->session->host, &conn->xmitwork); + } + write_unlock(conn->recv_lock); + + if (tx_desc->type == ISCSI_TX_CONTROL) { + /* this arithmetic is legal by libiscsi dd_data allocation */ + mtask = (void *) ((long)(void *)tx_desc - + sizeof(struct iscsi_mgmt_task)); + if (mtask->hdr->itt == cpu_to_be32(ISCSI_RESERVED_TAG)) { + struct iscsi_session *session = conn->session; + + spin_lock(&conn->session->lock); + list_del(&mtask->running); + __kfifo_put(session->mgmtpool.queue, (void*)&mtask, + sizeof(void*)); + spin_unlock(&session->lock); + } + } +} + +void iser_ctask_rdma_init(struct iscsi_iser_cmd_task *iser_ctask) + +{ + iser_ctask->status = ISER_TASK_STATUS_INIT; + + iser_ctask->dir[ISER_DIR_IN] = 0; + iser_ctask->dir[ISER_DIR_OUT] = 0; + + iser_ctask->data[ISER_DIR_IN].data_len = 0; + iser_ctask->data[ISER_DIR_OUT].data_len = 0; + + memset(&iser_ctask->rdma_regd[ISER_DIR_IN], 0, + sizeof(struct iser_regd_buf)); + memset(&iser_ctask->rdma_regd[ISER_DIR_OUT], 0, + sizeof(struct iser_regd_buf)); +} + +void iser_ctask_rdma_finalize(struct iscsi_iser_cmd_task *iser_ctask) +{ + int deferred; + + /* if we were reading, copy back to unaligned sglist, + * anyway dma_unmap and free the copy + */ + if (iser_ctask->data_copy[ISER_DIR_IN].copy_buf != NULL) + iser_finalize_rdma_unaligned_sg(iser_ctask, ISER_DIR_IN); + if (iser_ctask->data_copy[ISER_DIR_OUT].copy_buf != NULL) + iser_finalize_rdma_unaligned_sg(iser_ctask, ISER_DIR_OUT); + + if (iser_ctask->dir[ISER_DIR_IN]) { + deferred = iser_regd_buff_release + (&iser_ctask->rdma_regd[ISER_DIR_IN]); + if (deferred) { + iser_err("References remain for BUF-IN rdma reg\n"); + BUG(); + } + } + + if (iser_ctask->dir[ISER_DIR_OUT]) { + deferred = iser_regd_buff_release + (&iser_ctask->rdma_regd[ISER_DIR_OUT]); + if (deferred) { + iser_err("References remain for BUF-OUT rdma reg\n"); + BUG(); + } + } + + iser_dma_unmap_task_data(iser_ctask); +} + +void iser_dto_buffs_release(struct iser_dto *dto) +{ + int i; + + for (i = 0; i < dto->regd_vector_len; i++) + iser_regd_buff_release(dto->regd[i]); +} + diff --git a/drivers/infiniband/ulp/iser/iser_memory.c b/drivers/infiniband/ulp/iser/iser_memory.c new file mode 100644 index 000000000000..31950a522a1c --- /dev/null +++ b/drivers/infiniband/ulp/iser/iser_memory.c @@ -0,0 +1,401 @@ +/* + * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * $Id: iser_memory.c 6964 2006-05-07 11:11:43Z ogerlitz $ + */ +#include <linux/module.h> +#include <linux/kernel.h> +#include <linux/slab.h> +#include <linux/mm.h> +#include <asm/io.h> +#include <asm/scatterlist.h> +#include <linux/scatterlist.h> + +#include "iscsi_iser.h" + +#define ISER_KMALLOC_THRESHOLD 0x20000 /* 128K - kmalloc limit */ +/** + * Decrements the reference count for the + * registered buffer & releases it + * + * returns 0 if released, 1 if deferred + */ +int iser_regd_buff_release(struct iser_regd_buf *regd_buf) +{ + struct device *dma_device; + + if ((atomic_read(®d_buf->ref_count) == 0) || + atomic_dec_and_test(®d_buf->ref_count)) { + /* if we used the dma mr, unreg is just NOP */ + if (regd_buf->reg.rkey != 0) + iser_unreg_mem(®d_buf->reg); + + if (regd_buf->dma_addr) { + dma_device = regd_buf->device->ib_device->dma_device; + dma_unmap_single(dma_device, + regd_buf->dma_addr, + regd_buf->data_size, + regd_buf->direction); + } + /* else this regd buf is associated with task which we */ + /* dma_unmap_single/sg later */ + return 0; + } else { + iser_dbg("Release deferred, regd.buff: 0x%p\n", regd_buf); + return 1; + } +} + +/** + * iser_reg_single - fills registered buffer descriptor with + * registration information + */ +void iser_reg_single(struct iser_device *device, + struct iser_regd_buf *regd_buf, + enum dma_data_direction direction) +{ + dma_addr_t dma_addr; + + dma_addr = dma_map_single(device->ib_device->dma_device, + regd_buf->virt_addr, + regd_buf->data_size, direction); + BUG_ON(dma_mapping_error(dma_addr)); + + regd_buf->reg.lkey = device->mr->lkey; + regd_buf->reg.rkey = 0; /* indicate there's no need to unreg */ + regd_buf->reg.len = regd_buf->data_size; + regd_buf->reg.va = dma_addr; + + regd_buf->dma_addr = dma_addr; + regd_buf->direction = direction; +} + +/** + * iser_start_rdma_unaligned_sg + */ +int iser_start_rdma_unaligned_sg(struct iscsi_iser_cmd_task *iser_ctask, + enum iser_data_dir cmd_dir) +{ + int dma_nents; + struct device *dma_device; + char *mem = NULL; + struct iser_data_buf *data = &iser_ctask->data[cmd_dir]; + unsigned long cmd_data_len = data->data_len; + + if (cmd_data_len > ISER_KMALLOC_THRESHOLD) + mem = (void *)__get_free_pages(GFP_NOIO, + long_log2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT); + else + mem = kmalloc(cmd_data_len, GFP_NOIO); + + if (mem == NULL) { + iser_err("Failed to allocate mem size %d %d for copying sglist\n", + data->size,(int)cmd_data_len); + return -ENOMEM; + } + + if (cmd_dir == ISER_DIR_OUT) { + /* copy the unaligned sg the buffer which is used for RDMA */ + struct scatterlist *sg = (struct scatterlist *)data->buf; + int i; + char *p, *from; + + for (p = mem, i = 0; i < data->size; i++) { + from = kmap_atomic(sg[i].page, KM_USER0); + memcpy(p, + from + sg[i].offset, + sg[i].length); + kunmap_atomic(from, KM_USER0); + p += sg[i].length; + } + } + + sg_init_one(&iser_ctask->data_copy[cmd_dir].sg_single, mem, cmd_data_len); + iser_ctask->data_copy[cmd_dir].buf = + &iser_ctask->data_copy[cmd_dir].sg_single; + iser_ctask->data_copy[cmd_dir].size = 1; + + iser_ctask->data_copy[cmd_dir].copy_buf = mem; + + dma_device = iser_ctask->iser_conn->ib_conn->device->ib_device->dma_device; + + if (cmd_dir == ISER_DIR_OUT) + dma_nents = dma_map_sg(dma_device, + &iser_ctask->data_copy[cmd_dir].sg_single, + 1, DMA_TO_DEVICE); + else + dma_nents = dma_map_sg(dma_device, + &iser_ctask->data_copy[cmd_dir].sg_single, + 1, DMA_FROM_DEVICE); + + BUG_ON(dma_nents == 0); + + iser_ctask->data_copy[cmd_dir].dma_nents = dma_nents; + return 0; +} + +/** + * iser_finalize_rdma_unaligned_sg + */ +void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_cmd_task *iser_ctask, + enum iser_data_dir cmd_dir) +{ + struct device *dma_device; + struct iser_data_buf *mem_copy; + unsigned long cmd_data_len; + + dma_device = iser_ctask->iser_conn->ib_conn->device->ib_device->dma_device; + mem_copy = &iser_ctask->data_copy[cmd_dir]; + + if (cmd_dir == ISER_DIR_OUT) + dma_unmap_sg(dma_device, &mem_copy->sg_single, 1, + DMA_TO_DEVICE); + else + dma_unmap_sg(dma_device, &mem_copy->sg_single, 1, + DMA_FROM_DEVICE); + + if (cmd_dir == ISER_DIR_IN) { + char *mem; + struct scatterlist *sg; + unsigned char *p, *to; + unsigned int sg_size; + int i; + + /* copy back read RDMA to unaligned sg */ + mem = mem_copy->copy_buf; + + sg = (struct scatterlist *)iser_ctask->data[ISER_DIR_IN].buf; + sg_size = iser_ctask->data[ISER_DIR_IN].size; + + for (p = mem, i = 0; i < sg_size; i++){ + to = kmap_atomic(sg[i].page, KM_SOFTIRQ0); + memcpy(to + sg[i].offset, + p, + sg[i].length); + kunmap_atomic(to, KM_SOFTIRQ0); + p += sg[i].length; + } + } + + cmd_data_len = iser_ctask->data[cmd_dir].data_len; + + if (cmd_data_len > ISER_KMALLOC_THRESHOLD) + free_pages((unsigned long)mem_copy->copy_buf, + long_log2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT); + else + kfree(mem_copy->copy_buf); + + mem_copy->copy_buf = NULL; +} + +/** + * iser_sg_to_page_vec - Translates scatterlist entries to physical addresses + * and returns the length of resulting physical address array (may be less than + * the original due to possible compaction). + * + * we build a "page vec" under the assumption that the SG meets the RDMA + * alignment requirements. Other then the first and last SG elements, all + * the "internal" elements can be compacted into a list whose elements are + * dma addresses of physical pages. The code supports also the weird case + * where --few fragments of the same page-- are present in the SG as + * consecutive elements. Also, it handles one entry SG. + */ +static int iser_sg_to_page_vec(struct iser_data_buf *data, + struct iser_page_vec *page_vec) +{ + struct scatterlist *sg = (struct scatterlist *)data->buf; + dma_addr_t first_addr, last_addr, page; + int start_aligned, end_aligned; + unsigned int cur_page = 0; + unsigned long total_sz = 0; + int i; + + /* compute the offset of first element */ + page_vec->offset = (u64) sg[0].offset; + + for (i = 0; i < data->dma_nents; i++) { + total_sz += sg_dma_len(&sg[i]); + + first_addr = sg_dma_address(&sg[i]); + last_addr = first_addr + sg_dma_len(&sg[i]); + + start_aligned = !(first_addr & ~PAGE_MASK); + end_aligned = !(last_addr & ~PAGE_MASK); + + /* continue to collect page fragments till aligned or SG ends */ + while (!end_aligned && (i + 1 < data->dma_nents)) { + i++; + total_sz += sg_dma_len(&sg[i]); + last_addr = sg_dma_address(&sg[i]) + sg_dma_len(&sg[i]); + end_aligned = !(last_addr & ~PAGE_MASK); + } + + first_addr = first_addr & PAGE_MASK; + + for (page = first_addr; page < last_addr; page += PAGE_SIZE) + page_vec->pages[cur_page++] = page; + + } + page_vec->data_size = total_sz; + iser_dbg("page_vec->data_size:%d cur_page %d\n", page_vec->data_size,cur_page); + return cur_page; +} + +#define MASK_4K ((1UL << 12) - 1) /* 0xFFF */ +#define IS_4K_ALIGNED(addr) ((((unsigned long)addr) & MASK_4K) == 0) + +/** + * iser_data_buf_aligned_len - Tries to determine the maximal correctly aligned + * for RDMA sub-list of a scatter-gather list of memory buffers, and returns + * the number of entries which are aligned correctly. Supports the case where + * consecutive SG elements are actually fragments of the same physcial page. + */ +static unsigned int iser_data_buf_aligned_len(struct iser_data_buf *data) +{ + struct scatterlist *sg; + dma_addr_t end_addr, next_addr; + int i, cnt; + unsigned int ret_len = 0; + + sg = (struct scatterlist *)data->buf; + + for (cnt = 0, i = 0; i < data->dma_nents; i++, cnt++) { + /* iser_dbg("Checking sg iobuf [%d]: phys=0x%08lX " + "offset: %ld sz: %ld\n", i, + (unsigned long)page_to_phys(sg[i].page), + (unsigned long)sg[i].offset, + (unsigned long)sg[i].length); */ + end_addr = sg_dma_address(&sg[i]) + + sg_dma_len(&sg[i]); + /* iser_dbg("Checking sg iobuf end address " + "0x%08lX\n", end_addr); */ + if (i + 1 < data->dma_nents) { + next_addr = sg_dma_address(&sg[i+1]); + /* are i, i+1 fragments of the same page? */ + if (end_addr == next_addr) + continue; + else if (!IS_4K_ALIGNED(end_addr)) { + ret_len = cnt + 1; + break; + } + } + } + if (i == data->dma_nents) + ret_len = cnt; /* loop ended */ + iser_dbg("Found %d aligned entries out of %d in sg:0x%p\n", + ret_len, data->dma_nents, data); + return ret_len; +} + +static void iser_data_buf_dump(struct iser_data_buf *data) +{ + struct scatterlist *sg = (struct scatterlist *)data->buf; + int i; + + for (i = 0; i < data->size; i++) + iser_err("sg[%d] dma_addr:0x%lX page:0x%p " + "off:%d sz:%d dma_len:%d\n", + i, (unsigned long)sg_dma_address(&sg[i]), + sg[i].page, sg[i].offset, + sg[i].length,sg_dma_len(&sg[i])); +} + +static void iser_dump_page_vec(struct iser_page_vec *page_vec) +{ + int i; + + iser_err("page vec length %d data size %d\n", + page_vec->length, page_vec->data_size); + for (i = 0; i < page_vec->length; i++) + iser_err("%d %lx\n",i,(unsigned long)page_vec->pages[i]); +} + +static void iser_page_vec_build(struct iser_data_buf *data, + struct iser_page_vec *page_vec) +{ + int page_vec_len = 0; + + page_vec->length = 0; + page_vec->offset = 0; + + iser_dbg("Translating sg sz: %d\n", data->dma_nents); + page_vec_len = iser_sg_to_page_vec(data,page_vec); + iser_dbg("sg len %d page_vec_len %d\n", data->dma_nents,page_vec_len); + + page_vec->length = page_vec_len; + + if (page_vec_len * PAGE_SIZE < page_vec->data_size) { + iser_err("page_vec too short to hold this SG\n"); + iser_data_buf_dump(data); + iser_dump_page_vec(page_vec); + BUG(); + } +} + +/** + * iser_reg_rdma_mem - Registers memory intended for RDMA, + * obtaining rkey and va + * + * returns 0 on success, errno code on failure + */ +int iser_reg_rdma_mem(struct iscsi_iser_cmd_task *iser_ctask, + enum iser_data_dir cmd_dir) +{ + struct iser_conn *ib_conn = iser_ctask->iser_conn->ib_conn; + struct iser_data_buf *mem = &iser_ctask->data[cmd_dir]; + struct iser_regd_buf *regd_buf; + int aligned_len; + int err; + + regd_buf = &iser_ctask->rdma_regd[cmd_dir]; + + aligned_len = iser_data_buf_aligned_len(mem); + if (aligned_len != mem->size) { + iser_err("rdma alignment violation %d/%d aligned\n", + aligned_len, mem->size); + iser_data_buf_dump(mem); + /* allocate copy buf, if we are writing, copy the */ + /* unaligned scatterlist, dma map the copy */ + if (iser_start_rdma_unaligned_sg(iser_ctask, cmd_dir) != 0) + return -ENOMEM; + mem = &iser_ctask->data_copy[cmd_dir]; + } + + iser_page_vec_build(mem, ib_conn->page_vec); + err = iser_reg_page_vec(ib_conn, ib_conn->page_vec, ®d_buf->reg); + if (err) + return err; + + /* take a reference on this regd buf such that it will not be released * + * (eg in send dto completion) before we get the scsi response */ + atomic_inc(®d_buf->ref_count); + return 0; +} diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c new file mode 100644 index 000000000000..ff117bbf81b4 --- /dev/null +++ b/drivers/infiniband/ulp/iser/iser_verbs.c @@ -0,0 +1,827 @@ +/* + * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved. + * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * $Id: iser_verbs.c 7051 2006-05-10 12:29:11Z ogerlitz $ + */ +#include <asm/io.h> +#include <linux/kernel.h> +#include <linux/module.h> +#include <linux/smp_lock.h> +#include <linux/delay.h> +#include <linux/version.h> + +#include "iscsi_iser.h" + +#define ISCSI_ISER_MAX_CONN 8 +#define ISER_MAX_CQ_LEN ((ISER_QP_MAX_RECV_DTOS + \ + ISER_QP_MAX_REQ_DTOS) * \ + ISCSI_ISER_MAX_CONN) + +static void iser_cq_tasklet_fn(unsigned long data); +static void iser_cq_callback(struct ib_cq *cq, void *cq_context); +static void iser_comp_error_worker(void *data); + +static void iser_cq_event_callback(struct ib_event *cause, void *context) +{ + iser_err("got cq event %d \n", cause->event); +} + +static void iser_qp_event_callback(struct ib_event *cause, void *context) +{ + iser_err("got qp event %d\n",cause->event); +} + +/** + * iser_create_device_ib_res - creates Protection Domain (PD), Completion + * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with + * the adapator. + * + * returns 0 on success, -1 on failure + */ +static int iser_create_device_ib_res(struct iser_device *device) +{ + device->pd = ib_alloc_pd(device->ib_device); + if (IS_ERR(device->pd)) + goto pd_err; + + device->cq = ib_create_cq(device->ib_device, + iser_cq_callback, + iser_cq_event_callback, + (void *)device, + ISER_MAX_CQ_LEN); + if (IS_ERR(device->cq)) + goto cq_err; + + if (ib_req_notify_cq(device->cq, IB_CQ_NEXT_COMP)) + goto cq_arm_err; + + tasklet_init(&device->cq_tasklet, + iser_cq_tasklet_fn, + (unsigned long)device); + + device->mr = ib_get_dma_mr(device->pd, + IB_ACCESS_LOCAL_WRITE); + if (IS_ERR(device->mr)) + goto dma_mr_err; + + return 0; + +dma_mr_err: + tasklet_kill(&device->cq_tasklet); +cq_arm_err: + ib_destroy_cq(device->cq); +cq_err: + ib_dealloc_pd(device->pd); +pd_err: + iser_err("failed to allocate an IB resource\n"); + return -1; +} + +/** + * iser_free_device_ib_res - destory/dealloc/dereg the DMA MR, + * CQ and PD created with the device associated with the adapator. + */ +static void iser_free_device_ib_res(struct iser_device *device) +{ + BUG_ON(device->mr == NULL); + + tasklet_kill(&device->cq_tasklet); + + (void)ib_dereg_mr(device->mr); + (void)ib_destroy_cq(device->cq); + (void)ib_dealloc_pd(device->pd); + + device->mr = NULL; + device->cq = NULL; + device->pd = NULL; +} + +/** + * iser_create_ib_conn_res - Creates FMR pool and Queue-Pair (QP) + * + * returns 0 on success, -1 on failure + */ +static int iser_create_ib_conn_res(struct iser_conn *ib_conn) +{ + struct iser_device *device; + struct ib_qp_init_attr init_attr; + int ret; + struct ib_fmr_pool_param params; + + BUG_ON(ib_conn->device == NULL); + + device = ib_conn->device; + + ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) + + (sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)), + GFP_KERNEL); + if (!ib_conn->page_vec) { + ret = -ENOMEM; + goto alloc_err; + } + ib_conn->page_vec->pages = (u64 *) (ib_conn->page_vec + 1); + + params.page_shift = PAGE_SHIFT; + /* when the first/last SG element are not start/end * + * page aligned, the map whould be of N+1 pages */ + params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1; + /* make the pool size twice the max number of SCSI commands * + * the ML is expected to queue, watermark for unmap at 50% */ + params.pool_size = ISCSI_XMIT_CMDS_MAX * 2; + params.dirty_watermark = ISCSI_XMIT_CMDS_MAX; + params.cache = 0; + params.flush_function = NULL; + params.access = (IB_ACCESS_LOCAL_WRITE | + IB_ACCESS_REMOTE_WRITE | + IB_ACCESS_REMOTE_READ); + + ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, ¶ms); + if (IS_ERR(ib_conn->fmr_pool)) { + ret = PTR_ERR(ib_conn->fmr_pool); + goto fmr_pool_err; + } + + memset(&init_attr, 0, sizeof init_attr); + + init_attr.event_handler = iser_qp_event_callback; + init_attr.qp_context = (void *)ib_conn; + init_attr.send_cq = device->cq; + init_attr.recv_cq = device->cq; + init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS; + init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS; + init_attr.cap.max_send_sge = MAX_REGD_BUF_VECTOR_LEN; + init_attr.cap.max_recv_sge = 2; + init_attr.sq_sig_type = IB_SIGNAL_REQ_WR; + init_attr.qp_type = IB_QPT_RC; + + ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr); + if (ret) + goto qp_err; + + ib_conn->qp = ib_conn->cma_id->qp; + iser_err("setting conn %p cma_id %p: fmr_pool %p qp %p\n", + ib_conn, ib_conn->cma_id, + ib_conn->fmr_pool, ib_conn->cma_id->qp); + return ret; + +qp_err: + (void)ib_destroy_fmr_pool(ib_conn->fmr_pool); +fmr_pool_err: + kfree(ib_conn->page_vec); +alloc_err: + iser_err("unable to alloc mem or create resource, err %d\n", ret); + return ret; +} + +/** + * releases the FMR pool, QP and CMA ID objects, returns 0 on success, + * -1 on failure + */ +static int iser_free_ib_conn_res(struct iser_conn *ib_conn) +{ + BUG_ON(ib_conn == NULL); + + iser_err("freeing conn %p cma_id %p fmr pool %p qp %p\n", + ib_conn, ib_conn->cma_id, + ib_conn->fmr_pool, ib_conn->qp); + + /* qp is created only once both addr & route are resolved */ + if (ib_conn->fmr_pool != NULL) + ib_destroy_fmr_pool(ib_conn->fmr_pool); + + if (ib_conn->qp != NULL) + rdma_destroy_qp(ib_conn->cma_id); + + if (ib_conn->cma_id != NULL) + rdma_destroy_id(ib_conn->cma_id); + + ib_conn->fmr_pool = NULL; + ib_conn->qp = NULL; + ib_conn->cma_id = NULL; + kfree(ib_conn->page_vec); + + return 0; +} + +/** + * based on the resolved device node GUID see if there already allocated + * device for this device. If there's no such, create one. + */ +static +struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id) +{ + struct list_head *p_list; + struct iser_device *device = NULL; + + mutex_lock(&ig.device_list_mutex); + + p_list = ig.device_list.next; + while (p_list != &ig.device_list) { + device = list_entry(p_list, struct iser_device, ig_list); + /* find if there's a match using the node GUID */ + if (device->ib_device->node_guid == cma_id->device->node_guid) + break; + } + + if (device == NULL) { + device = kzalloc(sizeof *device, GFP_KERNEL); + if (device == NULL) + goto out; + /* assign this device to the device */ + device->ib_device = cma_id->device; + /* init the device and link it into ig device list */ + if (iser_create_device_ib_res(device)) { + kfree(device); + device = NULL; + goto out; + } + list_add(&device->ig_list, &ig.device_list); + } +out: + BUG_ON(device == NULL); + device->refcount++; + mutex_unlock(&ig.device_list_mutex); + return device; +} + +/* if there's no demand for this device, release it */ +static void iser_device_try_release(struct iser_device *device) +{ + mutex_lock(&ig.device_list_mutex); + device->refcount--; + iser_err("device %p refcount %d\n",device,device->refcount); + if (!device->refcount) { + iser_free_device_ib_res(device); + list_del(&device->ig_list); + kfree(device); + } + mutex_unlock(&ig.device_list_mutex); +} + +int iser_conn_state_comp(struct iser_conn *ib_conn, + enum iser_ib_conn_state comp) +{ + int ret; + + spin_lock_bh(&ib_conn->lock); + ret = (ib_conn->state == comp); + spin_unlock_bh(&ib_conn->lock); + return ret; +} + +static int iser_conn_state_comp_exch(struct iser_conn *ib_conn, + enum iser_ib_conn_state comp, + enum iser_ib_conn_state exch) +{ + int ret; + + spin_lock_bh(&ib_conn->lock); + if ((ret = (ib_conn->state == comp))) + ib_conn->state = exch; + spin_unlock_bh(&ib_conn->lock); + return ret; +} + +/** + * triggers start of the disconnect procedures and wait for them to be done + */ +void iser_conn_terminate(struct iser_conn *ib_conn) +{ + int err = 0; + + /* change the ib conn state only if the conn is UP, however always call + * rdma_disconnect since this is the only way to cause the CMA to change + * the QP state to ERROR + */ + + iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, ISER_CONN_TERMINATING); + err = rdma_disconnect(ib_conn->cma_id); + if (err) + iser_err("Failed to disconnect, conn: 0x%p err %d\n", + ib_conn,err); + + wait_event_interruptible(ib_conn->wait, + ib_conn->state == ISER_CONN_DOWN); + + iser_conn_release(ib_conn); +} + +static void iser_connect_error(struct rdma_cm_id *cma_id) +{ + struct iser_conn *ib_conn; + ib_conn = (struct iser_conn *)cma_id->context; + + ib_conn->state = ISER_CONN_DOWN; + wake_up_interruptible(&ib_conn->wait); +} + +static void iser_addr_handler(struct rdma_cm_id *cma_id) +{ + struct iser_device *device; + struct iser_conn *ib_conn; + int ret; + + device = iser_device_find_by_ib_device(cma_id); + ib_conn = (struct iser_conn *)cma_id->context; + ib_conn->device = device; + + ret = rdma_resolve_route(cma_id, 1000); + if (ret) { + iser_err("resolve route failed: %d\n", ret); + iser_connect_error(cma_id); + } + return; +} + +static void iser_route_handler(struct rdma_cm_id *cma_id) +{ + struct rdma_conn_param conn_param; + int ret; + + ret = iser_create_ib_conn_res((struct iser_conn *)cma_id->context); + if (ret) + goto failure; + + iser_dbg("path.mtu is %d setting it to %d\n", + cma_id->route.path_rec->mtu, IB_MTU_1024); + + /* we must set the MTU to 1024 as this is what the target is assuming */ + if (cma_id->route.path_rec->mtu > IB_MTU_1024) + cma_id->route.path_rec->mtu = IB_MTU_1024; + + memset(&conn_param, 0, sizeof conn_param); + conn_param.responder_resources = 4; + conn_param.initiator_depth = 1; + conn_param.retry_count = 7; + conn_param.rnr_retry_count = 6; + + ret = rdma_connect(cma_id, &conn_param); + if (ret) { + iser_err("failure connecting: %d\n", ret); + goto failure; + } + + return; +failure: + iser_connect_error(cma_id); +} + +static void iser_connected_handler(struct rdma_cm_id *cma_id) +{ + struct iser_conn *ib_conn; + + ib_conn = (struct iser_conn *)cma_id->context; + ib_conn->state = ISER_CONN_UP; + wake_up_interruptible(&ib_conn->wait); +} + +static void iser_disconnected_handler(struct rdma_cm_id *cma_id) +{ + struct iser_conn *ib_conn; + + ib_conn = (struct iser_conn *)cma_id->context; + ib_conn->disc_evt_flag = 1; + + /* getting here when the state is UP means that the conn is being * + * terminated asynchronously from the iSCSI layer's perspective. */ + if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, + ISER_CONN_TERMINATING)) + iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn, + ISCSI_ERR_CONN_FAILED); + + /* Complete the termination process if no posts are pending */ + if ((atomic_read(&ib_conn->post_recv_buf_count) == 0) && + (atomic_read(&ib_conn->post_send_buf_count) == 0)) { + ib_conn->state = ISER_CONN_DOWN; + wake_up_interruptible(&ib_conn->wait); + } +} + +static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) +{ + int ret = 0; + + iser_err("event %d conn %p id %p\n",event->event,cma_id->context,cma_id); + + switch (event->event) { + case RDMA_CM_EVENT_ADDR_RESOLVED: + iser_addr_handler(cma_id); + break; + case RDMA_CM_EVENT_ROUTE_RESOLVED: + iser_route_handler(cma_id); + break; + case RDMA_CM_EVENT_ESTABLISHED: + iser_connected_handler(cma_id); + break; + case RDMA_CM_EVENT_ADDR_ERROR: + case RDMA_CM_EVENT_ROUTE_ERROR: + case RDMA_CM_EVENT_CONNECT_ERROR: + case RDMA_CM_EVENT_UNREACHABLE: + case RDMA_CM_EVENT_REJECTED: + iser_err("event: %d, error: %d\n", event->event, event->status); + iser_connect_error(cma_id); + break; + case RDMA_CM_EVENT_DISCONNECTED: + iser_disconnected_handler(cma_id); + break; + case RDMA_CM_EVENT_DEVICE_REMOVAL: + BUG(); + break; + case RDMA_CM_EVENT_CONNECT_RESPONSE: + BUG(); + break; + case RDMA_CM_EVENT_CONNECT_REQUEST: + default: + break; + } + return ret; +} + +int iser_conn_init(struct iser_conn **ibconn) +{ + struct iser_conn *ib_conn; + + ib_conn = kzalloc(sizeof *ib_conn, GFP_KERNEL); + if (!ib_conn) { + iser_err("can't alloc memory for struct iser_conn\n"); + return -ENOMEM; + } + ib_conn->state = ISER_CONN_INIT; + init_waitqueue_head(&ib_conn->wait); + atomic_set(&ib_conn->post_recv_buf_count, 0); + atomic_set(&ib_conn->post_send_buf_count, 0); + INIT_WORK(&ib_conn->comperror_work, iser_comp_error_worker, + ib_conn); + INIT_LIST_HEAD(&ib_conn->conn_list); + spin_lock_init(&ib_conn->lock); + + *ibconn = ib_conn; + return 0; +} + + /** + * starts the process of connecting to the target + * sleeps untill the connection is established or rejected + */ +int iser_connect(struct iser_conn *ib_conn, + struct sockaddr_in *src_addr, + struct sockaddr_in *dst_addr, + int non_blocking) +{ + struct sockaddr *src, *dst; + int err = 0; + + sprintf(ib_conn->name,"%d.%d.%d.%d:%d", + NIPQUAD(dst_addr->sin_addr.s_addr), dst_addr->sin_port); + + /* the device is known only --after-- address resolution */ + ib_conn->device = NULL; + + iser_err("connecting to: %d.%d.%d.%d, port 0x%x\n", + NIPQUAD(dst_addr->sin_addr), dst_addr->sin_port); + + ib_conn->state = ISER_CONN_PENDING; + + ib_conn->cma_id = rdma_create_id(iser_cma_handler, + (void *)ib_conn, + RDMA_PS_TCP); + if (IS_ERR(ib_conn->cma_id)) { + err = PTR_ERR(ib_conn->cma_id); + iser_err("rdma_create_id failed: %d\n", err); + goto id_failure; + } + + src = (struct sockaddr *)src_addr; + dst = (struct sockaddr *)dst_addr; + err = rdma_resolve_addr(ib_conn->cma_id, src, dst, 1000); + if (err) { + iser_err("rdma_resolve_addr failed: %d\n", err); + goto addr_failure; + } + + if (!non_blocking) { + wait_event_interruptible(ib_conn->wait, + (ib_conn->state != ISER_CONN_PENDING)); + + if (ib_conn->state != ISER_CONN_UP) { + err = -EIO; + goto connect_failure; + } + } + + mutex_lock(&ig.connlist_mutex); + list_add(&ib_conn->conn_list, &ig.connlist); + mutex_unlock(&ig.connlist_mutex); + return 0; + +id_failure: + ib_conn->cma_id = NULL; +addr_failure: + ib_conn->state = ISER_CONN_DOWN; +connect_failure: + iser_conn_release(ib_conn); + return err; +} + +/** + * Frees all conn objects and deallocs conn descriptor + */ +void iser_conn_release(struct iser_conn *ib_conn) +{ + struct iser_device *device = ib_conn->device; + + BUG_ON(ib_conn->state != ISER_CONN_DOWN); + + mutex_lock(&ig.connlist_mutex); + list_del(&ib_conn->conn_list); + mutex_unlock(&ig.connlist_mutex); + + iser_free_ib_conn_res(ib_conn); + ib_conn->device = NULL; + /* on EVENT_ADDR_ERROR there's no device yet for this conn */ + if (device != NULL) + iser_device_try_release(device); + kfree(ib_conn); +} + + +/** + * iser_reg_page_vec - Register physical memory + * + * returns: 0 on success, errno code on failure + */ +int iser_reg_page_vec(struct iser_conn *ib_conn, + struct iser_page_vec *page_vec, + struct iser_mem_reg *mem_reg) +{ + struct ib_pool_fmr *mem; + u64 io_addr; + u64 *page_list; + int status; + + page_list = page_vec->pages; + io_addr = page_list[0]; + + mem = ib_fmr_pool_map_phys(ib_conn->fmr_pool, + page_list, + page_vec->length, + &io_addr); + + if (IS_ERR(mem)) { + status = (int)PTR_ERR(mem); + iser_err("ib_fmr_pool_map_phys failed: %d\n", status); + return status; + } + + mem_reg->lkey = mem->fmr->lkey; + mem_reg->rkey = mem->fmr->rkey; + mem_reg->len = page_vec->length * PAGE_SIZE; + mem_reg->va = io_addr; + mem_reg->mem_h = (void *)mem; + + mem_reg->va += page_vec->offset; + mem_reg->len = page_vec->data_size; + + iser_dbg("PHYSICAL Mem.register, [PHYS p_array: 0x%p, sz: %d, " + "entry[0]: (0x%08lx,%ld)] -> " + "[lkey: 0x%08X mem_h: 0x%p va: 0x%08lX sz: %ld]\n", + page_vec, page_vec->length, + (unsigned long)page_vec->pages[0], + (unsigned long)page_vec->data_size, + (unsigned int)mem_reg->lkey, mem_reg->mem_h, + (unsigned long)mem_reg->va, (unsigned long)mem_reg->len); + return 0; +} + +/** + * Unregister (previosuly registered) memory. + */ +void iser_unreg_mem(struct iser_mem_reg *reg) +{ + int ret; + + iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n",reg->mem_h); + + ret = ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h); + if (ret) + iser_err("ib_fmr_pool_unmap failed %d\n", ret); + + reg->mem_h = NULL; +} + +/** + * iser_dto_to_iov - builds IOV from a dto descriptor + */ +static void iser_dto_to_iov(struct iser_dto *dto, struct ib_sge *iov, int iov_len) +{ + int i; + struct ib_sge *sge; + struct iser_regd_buf *regd_buf; + + if (dto->regd_vector_len > iov_len) { + iser_err("iov size %d too small for posting dto of len %d\n", + iov_len, dto->regd_vector_len); + BUG(); + } + + for (i = 0; i < dto->regd_vector_len; i++) { + sge = &iov[i]; + regd_buf = dto->regd[i]; + + sge->addr = regd_buf->reg.va; + sge->length = regd_buf->reg.len; + sge->lkey = regd_buf->reg.lkey; + + if (dto->used_sz[i] > 0) /* Adjust size */ + sge->length = dto->used_sz[i]; + + /* offset and length should not exceed the regd buf length */ + if (sge->length + dto->offset[i] > regd_buf->reg.len) { + iser_err("Used len:%ld + offset:%d, exceed reg.buf.len:" + "%ld in dto:0x%p [%d], va:0x%08lX\n", + (unsigned long)sge->length, dto->offset[i], + (unsigned long)regd_buf->reg.len, dto, i, + (unsigned long)sge->addr); + BUG(); + } + + sge->addr += dto->offset[i]; /* Adjust offset */ + } +} + +/** + * iser_post_recv - Posts a receive buffer. + * + * returns 0 on success, -1 on failure + */ +int iser_post_recv(struct iser_desc *rx_desc) +{ + int ib_ret, ret_val = 0; + struct ib_recv_wr recv_wr, *recv_wr_failed; + struct ib_sge iov[2]; + struct iser_conn *ib_conn; + struct iser_dto *recv_dto = &rx_desc->dto; + + /* Retrieve conn */ + ib_conn = recv_dto->conn->ib_conn; + + iser_dto_to_iov(recv_dto, iov, 2); + + recv_wr.next = NULL; + recv_wr.sg_list = iov; + recv_wr.num_sge = recv_dto->regd_vector_len; + recv_wr.wr_id = (unsigned long)rx_desc; + + atomic_inc(&ib_conn->post_recv_buf_count); + ib_ret = ib_post_recv(ib_conn->qp, &recv_wr, &recv_wr_failed); + if (ib_ret) { + iser_err("ib_post_recv failed ret=%d\n", ib_ret); + atomic_dec(&ib_conn->post_recv_buf_count); + ret_val = -1; + } + + return ret_val; +} + +/** + * iser_start_send - Initiate a Send DTO operation + * + * returns 0 on success, -1 on failure + */ +int iser_post_send(struct iser_desc *tx_desc) +{ + int ib_ret, ret_val = 0; + struct ib_send_wr send_wr, *send_wr_failed; + struct ib_sge iov[MAX_REGD_BUF_VECTOR_LEN]; + struct iser_conn *ib_conn; + struct iser_dto *dto = &tx_desc->dto; + + ib_conn = dto->conn->ib_conn; + + iser_dto_to_iov(dto, iov, MAX_REGD_BUF_VECTOR_LEN); + + send_wr.next = NULL; + send_wr.wr_id = (unsigned long)tx_desc; + send_wr.sg_list = iov; + send_wr.num_sge = dto->regd_vector_len; + send_wr.opcode = IB_WR_SEND; + send_wr.send_flags = dto->notify_enable ? IB_SEND_SIGNALED : 0; + + atomic_inc(&ib_conn->post_send_buf_count); + + ib_ret = ib_post_send(ib_conn->qp, &send_wr, &send_wr_failed); + if (ib_ret) { + iser_err("Failed to start SEND DTO, dto: 0x%p, IOV len: %d\n", + dto, dto->regd_vector_len); + iser_err("ib_post_send failed, ret:%d\n", ib_ret); + atomic_dec(&ib_conn->post_send_buf_count); + ret_val = -1; + } + + return ret_val; +} + +static void iser_comp_error_worker(void *data) +{ + struct iser_conn *ib_conn = data; + + /* getting here when the state is UP means that the conn is being * + * terminated asynchronously from the iSCSI layer's perspective. */ + if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, + ISER_CONN_TERMINATING)) + iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn, + ISCSI_ERR_CONN_FAILED); + + /* complete the termination process if disconnect event was delivered * + * note there are no more non completed posts to the QP */ + if (ib_conn->disc_evt_flag) { + ib_conn->state = ISER_CONN_DOWN; + wake_up_interruptible(&ib_conn->wait); + } +} + +static void iser_handle_comp_error(struct iser_desc *desc) +{ + struct iser_dto *dto = &desc->dto; + struct iser_conn *ib_conn = dto->conn->ib_conn; + + iser_dto_buffs_release(dto); + + if (desc->type == ISCSI_RX) { + kfree(desc->data); + kmem_cache_free(ig.desc_cache, desc); + atomic_dec(&ib_conn->post_recv_buf_count); + } else { /* type is TX control/command/dataout */ + if (desc->type == ISCSI_TX_DATAOUT) + kmem_cache_free(ig.desc_cache, desc); + atomic_dec(&ib_conn->post_send_buf_count); + } + + if (atomic_read(&ib_conn->post_recv_buf_count) == 0 && + atomic_read(&ib_conn->post_send_buf_count) == 0) + schedule_work(&ib_conn->comperror_work); +} + +static void iser_cq_tasklet_fn(unsigned long data) +{ + struct iser_device *device = (struct iser_device *)data; + struct ib_cq *cq = device->cq; + struct ib_wc wc; + struct iser_desc *desc; + unsigned long xfer_len; + + while (ib_poll_cq(cq, 1, &wc) == 1) { + desc = (struct iser_desc *) (unsigned long) wc.wr_id; + BUG_ON(desc == NULL); + + if (wc.status == IB_WC_SUCCESS) { + if (desc->type == ISCSI_RX) { + xfer_len = (unsigned long)wc.byte_len; + iser_rcv_completion(desc, xfer_len); + } else /* type == ISCSI_TX_CONTROL/SCSI_CMD/DOUT */ + iser_snd_completion(desc); + } else { + iser_err("comp w. error op %d status %d\n",desc->type,wc.status); + iser_handle_comp_error(desc); + } + } + /* #warning "it is assumed here that arming CQ only once its empty" * + * " would not cause interrupts to be missed" */ + ib_req_notify_cq(cq, IB_CQ_NEXT_COMP); +} + +static void iser_cq_callback(struct ib_cq *cq, void *cq_context) +{ + struct iser_device *device = (struct iser_device *)cq_context; + + tasklet_schedule(&device->cq_tasklet); +} diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c index 9cbdffa08dc2..4e22afef7206 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.c +++ b/drivers/infiniband/ulp/srp/ib_srp.c @@ -62,6 +62,13 @@ MODULE_DESCRIPTION("InfiniBand SCSI RDMA Protocol initiator " "v" DRV_VERSION " (" DRV_RELDATE ")"); MODULE_LICENSE("Dual BSD/GPL"); +static int srp_sg_tablesize = SRP_DEF_SG_TABLESIZE; +static int srp_max_iu_len; + +module_param(srp_sg_tablesize, int, 0444); +MODULE_PARM_DESC(srp_sg_tablesize, + "Max number of gather/scatter entries per I/O (default is 12)"); + static int topspin_workarounds = 1; module_param(topspin_workarounds, int, 0444); @@ -105,7 +112,8 @@ static struct srp_iu *srp_alloc_iu(struct srp_host *host, size_t size, if (!iu->buf) goto out_free_iu; - iu->dma = dma_map_single(host->dev->dma_device, iu->buf, size, direction); + iu->dma = dma_map_single(host->dev->dev->dma_device, + iu->buf, size, direction); if (dma_mapping_error(iu->dma)) goto out_free_buf; @@ -127,7 +135,8 @@ static void srp_free_iu(struct srp_host *host, struct srp_iu *iu) if (!iu) return; - dma_unmap_single(host->dev->dma_device, iu->dma, iu->size, iu->direction); + dma_unmap_single(host->dev->dev->dma_device, + iu->dma, iu->size, iu->direction); kfree(iu->buf); kfree(iu); } @@ -147,7 +156,7 @@ static int srp_init_qp(struct srp_target_port *target, if (!attr) return -ENOMEM; - ret = ib_find_cached_pkey(target->srp_host->dev, + ret = ib_find_cached_pkey(target->srp_host->dev->dev, target->srp_host->port, be16_to_cpu(target->path.pkey), &attr->pkey_index); @@ -179,7 +188,7 @@ static int srp_create_target_ib(struct srp_target_port *target) if (!init_attr) return -ENOMEM; - target->cq = ib_create_cq(target->srp_host->dev, srp_completion, + target->cq = ib_create_cq(target->srp_host->dev->dev, srp_completion, NULL, target, SRP_CQ_SIZE); if (IS_ERR(target->cq)) { ret = PTR_ERR(target->cq); @@ -198,7 +207,7 @@ static int srp_create_target_ib(struct srp_target_port *target) init_attr->send_cq = target->cq; init_attr->recv_cq = target->cq; - target->qp = ib_create_qp(target->srp_host->pd, init_attr); + target->qp = ib_create_qp(target->srp_host->dev->pd, init_attr); if (IS_ERR(target->qp)) { ret = PTR_ERR(target->qp); ib_destroy_cq(target->cq); @@ -250,7 +259,7 @@ static int srp_lookup_path(struct srp_target_port *target) init_completion(&target->done); - target->path_query_id = ib_sa_path_rec_get(target->srp_host->dev, + target->path_query_id = ib_sa_path_rec_get(target->srp_host->dev->dev, target->srp_host->port, &target->path, IB_SA_PATH_REC_DGID | @@ -309,10 +318,32 @@ static int srp_send_req(struct srp_target_port *target) req->priv.opcode = SRP_LOGIN_REQ; req->priv.tag = 0; - req->priv.req_it_iu_len = cpu_to_be32(SRP_MAX_IU_LEN); + req->priv.req_it_iu_len = cpu_to_be32(srp_max_iu_len); req->priv.req_buf_fmt = cpu_to_be16(SRP_BUF_FORMAT_DIRECT | SRP_BUF_FORMAT_INDIRECT); - memcpy(req->priv.initiator_port_id, target->srp_host->initiator_port_id, 16); + /* + * In the published SRP specification (draft rev. 16a), the + * port identifier format is 8 bytes of ID extension followed + * by 8 bytes of GUID. Older drafts put the two halves in the + * opposite order, so that the GUID comes first. + * + * Targets conforming to these obsolete drafts can be + * recognized by the I/O Class they report. + */ + if (target->io_class == SRP_REV10_IB_IO_CLASS) { + memcpy(req->priv.initiator_port_id, + target->srp_host->initiator_port_id + 8, 8); + memcpy(req->priv.initiator_port_id + 8, + target->srp_host->initiator_port_id, 8); + memcpy(req->priv.target_port_id, &target->ioc_guid, 8); + memcpy(req->priv.target_port_id + 8, &target->id_ext, 8); + } else { + memcpy(req->priv.initiator_port_id, + target->srp_host->initiator_port_id, 16); + memcpy(req->priv.target_port_id, &target->id_ext, 8); + memcpy(req->priv.target_port_id + 8, &target->ioc_guid, 8); + } + /* * Topspin/Cisco SRP targets will reject our login unless we * zero out the first 8 bytes of our initiator port ID. The @@ -325,8 +356,6 @@ static int srp_send_req(struct srp_target_port *target) (unsigned long long) be64_to_cpu(target->ioc_guid)); memset(req->priv.initiator_port_id, 0, 8); } - memcpy(req->priv.target_port_id, &target->id_ext, 8); - memcpy(req->priv.target_port_id + 8, &target->ioc_guid, 8); status = ib_send_cm_req(target->cm_id, &req->param); @@ -359,9 +388,9 @@ static void srp_remove_work(void *target_ptr) target->state = SRP_TARGET_REMOVED; spin_unlock_irq(target->scsi_host->host_lock); - mutex_lock(&target->srp_host->target_mutex); + spin_lock(&target->srp_host->target_lock); list_del(&target->list); - mutex_unlock(&target->srp_host->target_mutex); + spin_unlock(&target->srp_host->target_lock); scsi_remove_host(target->scsi_host); ib_destroy_cm_id(target->cm_id); @@ -421,6 +450,11 @@ static void srp_unmap_data(struct scsi_cmnd *scmnd, scmnd->sc_data_direction != DMA_FROM_DEVICE)) return; + if (req->fmr) { + ib_fmr_pool_unmap(req->fmr); + req->fmr = NULL; + } + /* * This handling of non-SG commands can be killed when the * SCSI midlayer no longer generates non-SG commands. @@ -433,18 +467,30 @@ static void srp_unmap_data(struct scsi_cmnd *scmnd, scat = &req->fake_sg; } - dma_unmap_sg(target->srp_host->dev->dma_device, scat, nents, + dma_unmap_sg(target->srp_host->dev->dev->dma_device, scat, nents, scmnd->sc_data_direction); } +static void srp_remove_req(struct srp_target_port *target, struct srp_request *req) +{ + srp_unmap_data(req->scmnd, target, req); + list_move_tail(&req->list, &target->free_reqs); +} + +static void srp_reset_req(struct srp_target_port *target, struct srp_request *req) +{ + req->scmnd->result = DID_RESET << 16; + req->scmnd->scsi_done(req->scmnd); + srp_remove_req(target, req); +} + static int srp_reconnect_target(struct srp_target_port *target) { struct ib_cm_id *new_cm_id; struct ib_qp_attr qp_attr; - struct srp_request *req; + struct srp_request *req, *tmp; struct ib_wc wc; int ret; - int i; spin_lock_irq(target->scsi_host->host_lock); if (target->state != SRP_TARGET_LIVE) { @@ -459,7 +505,7 @@ static int srp_reconnect_target(struct srp_target_port *target) * Now get a new local CM ID so that we avoid confusing the * target in case things are really fouled up. */ - new_cm_id = ib_create_cm_id(target->srp_host->dev, + new_cm_id = ib_create_cm_id(target->srp_host->dev->dev, srp_cm_handler, target); if (IS_ERR(new_cm_id)) { ret = PTR_ERR(new_cm_id); @@ -480,19 +526,12 @@ static int srp_reconnect_target(struct srp_target_port *target) while (ib_poll_cq(target->cq, 1, &wc) > 0) ; /* nothing */ - list_for_each_entry(req, &target->req_queue, list) { - req->scmnd->result = DID_RESET << 16; - req->scmnd->scsi_done(req->scmnd); - srp_unmap_data(req->scmnd, target, req); - } + list_for_each_entry_safe(req, tmp, &target->req_queue, list) + srp_reset_req(target, req); target->rx_head = 0; target->tx_head = 0; target->tx_tail = 0; - INIT_LIST_HEAD(&target->free_reqs); - INIT_LIST_HEAD(&target->req_queue); - for (i = 0; i < SRP_SQ_SIZE; ++i) - list_add_tail(&target->req_ring[i].list, &target->free_reqs); ret = srp_connect_target(target); if (ret) @@ -528,14 +567,79 @@ err: return ret; } +static int srp_map_fmr(struct srp_device *dev, struct scatterlist *scat, + int sg_cnt, struct srp_request *req, + struct srp_direct_buf *buf) +{ + u64 io_addr = 0; + u64 *dma_pages; + u32 len; + int page_cnt; + int i, j; + int ret; + + if (!dev->fmr_pool) + return -ENODEV; + + len = page_cnt = 0; + for (i = 0; i < sg_cnt; ++i) { + if (sg_dma_address(&scat[i]) & ~dev->fmr_page_mask) { + if (i > 0) + return -EINVAL; + else + ++page_cnt; + } + if ((sg_dma_address(&scat[i]) + sg_dma_len(&scat[i])) & + ~dev->fmr_page_mask) { + if (i < sg_cnt - 1) + return -EINVAL; + else + ++page_cnt; + } + + len += sg_dma_len(&scat[i]); + } + + page_cnt += len >> dev->fmr_page_shift; + if (page_cnt > SRP_FMR_SIZE) + return -ENOMEM; + + dma_pages = kmalloc(sizeof (u64) * page_cnt, GFP_ATOMIC); + if (!dma_pages) + return -ENOMEM; + + page_cnt = 0; + for (i = 0; i < sg_cnt; ++i) + for (j = 0; j < sg_dma_len(&scat[i]); j += dev->fmr_page_size) + dma_pages[page_cnt++] = + (sg_dma_address(&scat[i]) & dev->fmr_page_mask) + j; + + req->fmr = ib_fmr_pool_map_phys(dev->fmr_pool, + dma_pages, page_cnt, &io_addr); + if (IS_ERR(req->fmr)) { + ret = PTR_ERR(req->fmr); + goto out; + } + + buf->va = cpu_to_be64(sg_dma_address(&scat[0]) & ~dev->fmr_page_mask); + buf->key = cpu_to_be32(req->fmr->fmr->rkey); + buf->len = cpu_to_be32(len); + + ret = 0; + +out: + kfree(dma_pages); + + return ret; +} + static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_target_port *target, struct srp_request *req) { struct scatterlist *scat; struct srp_cmd *cmd = req->cmd->buf; int len, nents, count; - int i; - u8 fmt; + u8 fmt = SRP_DATA_DESC_DIRECT; if (!scmnd->request_buffer || scmnd->sc_data_direction == DMA_NONE) return sizeof (struct srp_cmd); @@ -560,53 +664,63 @@ static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_target_port *target, sg_init_one(scat, scmnd->request_buffer, scmnd->request_bufflen); } - count = dma_map_sg(target->srp_host->dev->dma_device, scat, nents, - scmnd->sc_data_direction); + count = dma_map_sg(target->srp_host->dev->dev->dma_device, + scat, nents, scmnd->sc_data_direction); + + fmt = SRP_DATA_DESC_DIRECT; + len = sizeof (struct srp_cmd) + sizeof (struct srp_direct_buf); if (count == 1) { + /* + * The midlayer only generated a single gather/scatter + * entry, or DMA mapping coalesced everything to a + * single entry. So a direct descriptor along with + * the DMA MR suffices. + */ struct srp_direct_buf *buf = (void *) cmd->add_data; - fmt = SRP_DATA_DESC_DIRECT; - buf->va = cpu_to_be64(sg_dma_address(scat)); - buf->key = cpu_to_be32(target->srp_host->mr->rkey); + buf->key = cpu_to_be32(target->srp_host->dev->mr->rkey); buf->len = cpu_to_be32(sg_dma_len(scat)); - - len = sizeof (struct srp_cmd) + - sizeof (struct srp_direct_buf); - } else { + } else if (srp_map_fmr(target->srp_host->dev, scat, count, req, + (void *) cmd->add_data)) { + /* + * FMR mapping failed, and the scatterlist has more + * than one entry. Generate an indirect memory + * descriptor. + */ struct srp_indirect_buf *buf = (void *) cmd->add_data; u32 datalen = 0; + int i; fmt = SRP_DATA_DESC_INDIRECT; + len = sizeof (struct srp_cmd) + + sizeof (struct srp_indirect_buf) + + count * sizeof (struct srp_direct_buf); + + for (i = 0; i < count; ++i) { + buf->desc_list[i].va = + cpu_to_be64(sg_dma_address(&scat[i])); + buf->desc_list[i].key = + cpu_to_be32(target->srp_host->dev->mr->rkey); + buf->desc_list[i].len = + cpu_to_be32(sg_dma_len(&scat[i])); + datalen += sg_dma_len(&scat[i]); + } if (scmnd->sc_data_direction == DMA_TO_DEVICE) cmd->data_out_desc_cnt = count; else cmd->data_in_desc_cnt = count; - buf->table_desc.va = cpu_to_be64(req->cmd->dma + - sizeof *cmd + - sizeof *buf); + buf->table_desc.va = + cpu_to_be64(req->cmd->dma + sizeof *cmd + sizeof *buf); buf->table_desc.key = - cpu_to_be32(target->srp_host->mr->rkey); + cpu_to_be32(target->srp_host->dev->mr->rkey); buf->table_desc.len = cpu_to_be32(count * sizeof (struct srp_direct_buf)); - for (i = 0; i < count; ++i) { - buf->desc_list[i].va = cpu_to_be64(sg_dma_address(&scat[i])); - buf->desc_list[i].key = - cpu_to_be32(target->srp_host->mr->rkey); - buf->desc_list[i].len = cpu_to_be32(sg_dma_len(&scat[i])); - - datalen += sg_dma_len(&scat[i]); - } - buf->len = cpu_to_be32(datalen); - - len = sizeof (struct srp_cmd) + - sizeof (struct srp_indirect_buf) + - count * sizeof (struct srp_direct_buf); } if (scmnd->sc_data_direction == DMA_TO_DEVICE) @@ -617,12 +731,6 @@ static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_target_port *target, return len; } -static void srp_remove_req(struct srp_target_port *target, struct srp_request *req) -{ - srp_unmap_data(req->scmnd, target, req); - list_move_tail(&req->list, &target->free_reqs); -} - static void srp_process_rsp(struct srp_target_port *target, struct srp_rsp *rsp) { struct srp_request *req; @@ -689,7 +797,7 @@ static void srp_handle_recv(struct srp_target_port *target, struct ib_wc *wc) iu = target->rx_ring[wc->wr_id & ~SRP_OP_RECV]; - dma_sync_single_for_cpu(target->srp_host->dev->dma_device, iu->dma, + dma_sync_single_for_cpu(target->srp_host->dev->dev->dma_device, iu->dma, target->max_ti_iu_len, DMA_FROM_DEVICE); opcode = *(u8 *) iu->buf; @@ -726,7 +834,7 @@ static void srp_handle_recv(struct srp_target_port *target, struct ib_wc *wc) break; } - dma_sync_single_for_device(target->srp_host->dev->dma_device, iu->dma, + dma_sync_single_for_device(target->srp_host->dev->dev->dma_device, iu->dma, target->max_ti_iu_len, DMA_FROM_DEVICE); } @@ -770,7 +878,7 @@ static int __srp_post_recv(struct srp_target_port *target) list.addr = iu->dma; list.length = iu->size; - list.lkey = target->srp_host->mr->lkey; + list.lkey = target->srp_host->dev->mr->lkey; wr.next = NULL; wr.sg_list = &list; @@ -805,12 +913,8 @@ static struct srp_iu *__srp_get_tx_iu(struct srp_target_port *target) if (target->tx_head - target->tx_tail >= SRP_SQ_SIZE) return NULL; - if (unlikely(target->req_lim < 1)) { - if (printk_ratelimit()) - printk(KERN_DEBUG PFX "Target has req_lim %d\n", - target->req_lim); - return NULL; - } + if (unlikely(target->req_lim < 1)) + ++target->zero_req_lim; return target->tx_ring[target->tx_head & SRP_SQ_SIZE]; } @@ -828,7 +932,7 @@ static int __srp_post_send(struct srp_target_port *target, list.addr = iu->dma; list.length = len; - list.lkey = target->srp_host->mr->lkey; + list.lkey = target->srp_host->dev->mr->lkey; wr.next = NULL; wr.wr_id = target->tx_head & SRP_SQ_SIZE; @@ -870,8 +974,8 @@ static int srp_queuecommand(struct scsi_cmnd *scmnd, if (!iu) goto err; - dma_sync_single_for_cpu(target->srp_host->dev->dma_device, iu->dma, - SRP_MAX_IU_LEN, DMA_TO_DEVICE); + dma_sync_single_for_cpu(target->srp_host->dev->dev->dma_device, iu->dma, + srp_max_iu_len, DMA_TO_DEVICE); req = list_entry(target->free_reqs.next, struct srp_request, list); @@ -903,8 +1007,8 @@ static int srp_queuecommand(struct scsi_cmnd *scmnd, goto err_unmap; } - dma_sync_single_for_device(target->srp_host->dev->dma_device, iu->dma, - SRP_MAX_IU_LEN, DMA_TO_DEVICE); + dma_sync_single_for_device(target->srp_host->dev->dev->dma_device, iu->dma, + srp_max_iu_len, DMA_TO_DEVICE); if (__srp_post_send(target, iu, len)) { printk(KERN_ERR PFX "Send failed\n"); @@ -936,7 +1040,7 @@ static int srp_alloc_iu_bufs(struct srp_target_port *target) for (i = 0; i < SRP_SQ_SIZE + 1; ++i) { target->tx_ring[i] = srp_alloc_iu(target->srp_host, - SRP_MAX_IU_LEN, + srp_max_iu_len, GFP_KERNEL, DMA_TO_DEVICE); if (!target->tx_ring[i]) goto err; @@ -1107,11 +1211,10 @@ static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event) srp_cm_rej_handler(cm_id, event, target); break; - case IB_CM_MRA_RECEIVED: - printk(KERN_ERR PFX "MRA received\n"); - break; - - case IB_CM_DREP_RECEIVED: + case IB_CM_DREQ_RECEIVED: + printk(KERN_WARNING PFX "DREQ received - connection closed\n"); + if (ib_send_cm_drep(cm_id, NULL, 0)) + printk(KERN_ERR PFX "Sending CM DREP failed\n"); break; case IB_CM_TIMEWAIT_EXIT: @@ -1121,6 +1224,11 @@ static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event) target->status = 0; break; + case IB_CM_MRA_RECEIVED: + case IB_CM_DREQ_ERROR: + case IB_CM_DREP_RECEIVED: + break; + default: printk(KERN_WARNING PFX "Unhandled CM event %d\n", event->event); break; @@ -1239,11 +1347,8 @@ static int srp_reset_device(struct scsi_cmnd *scmnd) spin_lock_irq(target->scsi_host->host_lock); list_for_each_entry_safe(req, tmp, &target->req_queue, list) - if (req->scmnd->device == scmnd->device) { - req->scmnd->result = DID_RESET << 16; - req->scmnd->scsi_done(req->scmnd); - srp_remove_req(target, req); - } + if (req->scmnd->device == scmnd->device) + srp_reset_req(target, req); spin_unlock_irq(target->scsi_host->host_lock); @@ -1329,11 +1434,23 @@ static ssize_t show_dgid(struct class_device *cdev, char *buf) be16_to_cpu(((__be16 *) target->path.dgid.raw)[7])); } +static ssize_t show_zero_req_lim(struct class_device *cdev, char *buf) +{ + struct srp_target_port *target = host_to_target(class_to_shost(cdev)); + + if (target->state == SRP_TARGET_DEAD || + target->state == SRP_TARGET_REMOVED) + return -ENODEV; + + return sprintf(buf, "%d\n", target->zero_req_lim); +} + static CLASS_DEVICE_ATTR(id_ext, S_IRUGO, show_id_ext, NULL); static CLASS_DEVICE_ATTR(ioc_guid, S_IRUGO, show_ioc_guid, NULL); static CLASS_DEVICE_ATTR(service_id, S_IRUGO, show_service_id, NULL); static CLASS_DEVICE_ATTR(pkey, S_IRUGO, show_pkey, NULL); static CLASS_DEVICE_ATTR(dgid, S_IRUGO, show_dgid, NULL); +static CLASS_DEVICE_ATTR(zero_req_lim, S_IRUGO, show_zero_req_lim, NULL); static struct class_device_attribute *srp_host_attrs[] = { &class_device_attr_id_ext, @@ -1341,6 +1458,7 @@ static struct class_device_attribute *srp_host_attrs[] = { &class_device_attr_service_id, &class_device_attr_pkey, &class_device_attr_dgid, + &class_device_attr_zero_req_lim, NULL }; @@ -1354,7 +1472,6 @@ static struct scsi_host_template srp_template = { .eh_host_reset_handler = srp_reset_host, .can_queue = SRP_SQ_SIZE, .this_id = -1, - .sg_tablesize = SRP_MAX_INDIRECT, .cmd_per_lun = SRP_SQ_SIZE, .use_clustering = ENABLE_CLUSTERING, .shost_attrs = srp_host_attrs @@ -1365,18 +1482,17 @@ static int srp_add_target(struct srp_host *host, struct srp_target_port *target) sprintf(target->target_name, "SRP.T10:%016llX", (unsigned long long) be64_to_cpu(target->id_ext)); - if (scsi_add_host(target->scsi_host, host->dev->dma_device)) + if (scsi_add_host(target->scsi_host, host->dev->dev->dma_device)) return -ENODEV; - mutex_lock(&host->target_mutex); + spin_lock(&host->target_lock); list_add_tail(&target->list, &host->target_list); - mutex_unlock(&host->target_mutex); + spin_unlock(&host->target_lock); target->state = SRP_TARGET_LIVE; - /* XXX: are we supposed to have a definition of SCAN_WILD_CARD ?? */ scsi_scan_target(&target->scsi_host->shost_gendev, - 0, target->scsi_id, ~0, 0); + 0, target->scsi_id, SCAN_WILD_CARD, 0); return 0; } @@ -1410,6 +1526,8 @@ enum { SRP_OPT_PKEY = 1 << 3, SRP_OPT_SERVICE_ID = 1 << 4, SRP_OPT_MAX_SECT = 1 << 5, + SRP_OPT_MAX_CMD_PER_LUN = 1 << 6, + SRP_OPT_IO_CLASS = 1 << 7, SRP_OPT_ALL = (SRP_OPT_ID_EXT | SRP_OPT_IOC_GUID | SRP_OPT_DGID | @@ -1418,13 +1536,15 @@ enum { }; static match_table_t srp_opt_tokens = { - { SRP_OPT_ID_EXT, "id_ext=%s" }, - { SRP_OPT_IOC_GUID, "ioc_guid=%s" }, - { SRP_OPT_DGID, "dgid=%s" }, - { SRP_OPT_PKEY, "pkey=%x" }, - { SRP_OPT_SERVICE_ID, "service_id=%s" }, - { SRP_OPT_MAX_SECT, "max_sect=%d" }, - { SRP_OPT_ERR, NULL } + { SRP_OPT_ID_EXT, "id_ext=%s" }, + { SRP_OPT_IOC_GUID, "ioc_guid=%s" }, + { SRP_OPT_DGID, "dgid=%s" }, + { SRP_OPT_PKEY, "pkey=%x" }, + { SRP_OPT_SERVICE_ID, "service_id=%s" }, + { SRP_OPT_MAX_SECT, "max_sect=%d" }, + { SRP_OPT_MAX_CMD_PER_LUN, "max_cmd_per_lun=%d" }, + { SRP_OPT_IO_CLASS, "io_class=%x" }, + { SRP_OPT_ERR, NULL } }; static int srp_parse_options(const char *buf, struct srp_target_port *target) @@ -1500,6 +1620,29 @@ static int srp_parse_options(const char *buf, struct srp_target_port *target) target->scsi_host->max_sectors = token; break; + case SRP_OPT_MAX_CMD_PER_LUN: + if (match_int(args, &token)) { + printk(KERN_WARNING PFX "bad max cmd_per_lun parameter '%s'\n", p); + goto out; + } + target->scsi_host->cmd_per_lun = min(token, SRP_SQ_SIZE); + break; + + case SRP_OPT_IO_CLASS: + if (match_hex(args, &token)) { + printk(KERN_WARNING PFX "bad IO class parameter '%s' \n", p); + goto out; + } + if (token != SRP_REV10_IB_IO_CLASS && + token != SRP_REV16A_IB_IO_CLASS) { + printk(KERN_WARNING PFX "unknown IO class parameter value" + " %x specified (use %x or %x).\n", + token, SRP_REV10_IB_IO_CLASS, SRP_REV16A_IB_IO_CLASS); + goto out; + } + target->io_class = token; + break; + default: printk(KERN_WARNING PFX "unknown parameter or missing value " "'%s' in target creation request\n", p); @@ -1542,6 +1685,7 @@ static ssize_t srp_create_target(struct class_device *class_dev, target = host_to_target(target_host); memset(target, 0, sizeof *target); + target->io_class = SRP_REV16A_IB_IO_CLASS; target->scsi_host = target_host; target->srp_host = host; @@ -1558,7 +1702,7 @@ static ssize_t srp_create_target(struct class_device *class_dev, if (ret) goto err; - ib_get_cached_gid(host->dev, host->port, 0, &target->path.sgid); + ib_get_cached_gid(host->dev->dev, host->port, 0, &target->path.sgid); printk(KERN_DEBUG PFX "new target: id_ext %016llx ioc_guid %016llx pkey %04x " "service_id %016llx dgid %04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x\n", @@ -1579,7 +1723,7 @@ static ssize_t srp_create_target(struct class_device *class_dev, if (ret) goto err; - target->cm_id = ib_create_cm_id(host->dev, srp_cm_handler, target); + target->cm_id = ib_create_cm_id(host->dev->dev, srp_cm_handler, target); if (IS_ERR(target->cm_id)) { ret = PTR_ERR(target->cm_id); goto err_free; @@ -1619,7 +1763,7 @@ static ssize_t show_ibdev(struct class_device *class_dev, char *buf) struct srp_host *host = container_of(class_dev, struct srp_host, class_dev); - return sprintf(buf, "%s\n", host->dev->name); + return sprintf(buf, "%s\n", host->dev->dev->name); } static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL); @@ -1634,7 +1778,7 @@ static ssize_t show_port(struct class_device *class_dev, char *buf) static CLASS_DEVICE_ATTR(port, S_IRUGO, show_port, NULL); -static struct srp_host *srp_add_port(struct ib_device *device, u8 port) +static struct srp_host *srp_add_port(struct srp_device *device, u8 port) { struct srp_host *host; @@ -1643,32 +1787,21 @@ static struct srp_host *srp_add_port(struct ib_device *device, u8 port) return NULL; INIT_LIST_HEAD(&host->target_list); - mutex_init(&host->target_mutex); + spin_lock_init(&host->target_lock); init_completion(&host->released); host->dev = device; host->port = port; host->initiator_port_id[7] = port; - memcpy(host->initiator_port_id + 8, &device->node_guid, 8); - - host->pd = ib_alloc_pd(device); - if (IS_ERR(host->pd)) - goto err_free; - - host->mr = ib_get_dma_mr(host->pd, - IB_ACCESS_LOCAL_WRITE | - IB_ACCESS_REMOTE_READ | - IB_ACCESS_REMOTE_WRITE); - if (IS_ERR(host->mr)) - goto err_pd; + memcpy(host->initiator_port_id + 8, &device->dev->node_guid, 8); host->class_dev.class = &srp_class; - host->class_dev.dev = device->dma_device; + host->class_dev.dev = device->dev->dma_device; snprintf(host->class_dev.class_id, BUS_ID_SIZE, "srp-%s-%d", - device->name, port); + device->dev->name, port); if (class_device_register(&host->class_dev)) - goto err_mr; + goto free_host; if (class_device_create_file(&host->class_dev, &class_device_attr_add_target)) goto err_class; if (class_device_create_file(&host->class_dev, &class_device_attr_ibdev)) @@ -1681,13 +1814,7 @@ static struct srp_host *srp_add_port(struct ib_device *device, u8 port) err_class: class_device_unregister(&host->class_dev); -err_mr: - ib_dereg_mr(host->mr); - -err_pd: - ib_dealloc_pd(host->pd); - -err_free: +free_host: kfree(host); return NULL; @@ -1695,15 +1822,62 @@ err_free: static void srp_add_one(struct ib_device *device) { - struct list_head *dev_list; + struct srp_device *srp_dev; + struct ib_device_attr *dev_attr; + struct ib_fmr_pool_param fmr_param; struct srp_host *host; int s, e, p; - dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL); - if (!dev_list) + dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL); + if (!dev_attr) return; - INIT_LIST_HEAD(dev_list); + if (ib_query_device(device, dev_attr)) { + printk(KERN_WARNING PFX "Query device failed for %s\n", + device->name); + goto free_attr; + } + + srp_dev = kmalloc(sizeof *srp_dev, GFP_KERNEL); + if (!srp_dev) + goto free_attr; + + /* + * Use the smallest page size supported by the HCA, down to a + * minimum of 512 bytes (which is the smallest sector that a + * SCSI command will ever carry). + */ + srp_dev->fmr_page_shift = max(9, ffs(dev_attr->page_size_cap) - 1); + srp_dev->fmr_page_size = 1 << srp_dev->fmr_page_shift; + srp_dev->fmr_page_mask = ~((unsigned long) srp_dev->fmr_page_size - 1); + + INIT_LIST_HEAD(&srp_dev->dev_list); + + srp_dev->dev = device; + srp_dev->pd = ib_alloc_pd(device); + if (IS_ERR(srp_dev->pd)) + goto free_dev; + + srp_dev->mr = ib_get_dma_mr(srp_dev->pd, + IB_ACCESS_LOCAL_WRITE | + IB_ACCESS_REMOTE_READ | + IB_ACCESS_REMOTE_WRITE); + if (IS_ERR(srp_dev->mr)) + goto err_pd; + + memset(&fmr_param, 0, sizeof fmr_param); + fmr_param.pool_size = SRP_FMR_POOL_SIZE; + fmr_param.dirty_watermark = SRP_FMR_DIRTY_SIZE; + fmr_param.cache = 1; + fmr_param.max_pages_per_fmr = SRP_FMR_SIZE; + fmr_param.page_shift = srp_dev->fmr_page_shift; + fmr_param.access = (IB_ACCESS_LOCAL_WRITE | + IB_ACCESS_REMOTE_WRITE | + IB_ACCESS_REMOTE_READ); + + srp_dev->fmr_pool = ib_create_fmr_pool(srp_dev->pd, &fmr_param); + if (IS_ERR(srp_dev->fmr_pool)) + srp_dev->fmr_pool = NULL; if (device->node_type == IB_NODE_SWITCH) { s = 0; @@ -1714,25 +1888,35 @@ static void srp_add_one(struct ib_device *device) } for (p = s; p <= e; ++p) { - host = srp_add_port(device, p); + host = srp_add_port(srp_dev, p); if (host) - list_add_tail(&host->list, dev_list); + list_add_tail(&host->list, &srp_dev->dev_list); } - ib_set_client_data(device, &srp_client, dev_list); + ib_set_client_data(device, &srp_client, srp_dev); + + goto free_attr; + +err_pd: + ib_dealloc_pd(srp_dev->pd); + +free_dev: + kfree(srp_dev); + +free_attr: + kfree(dev_attr); } static void srp_remove_one(struct ib_device *device) { - struct list_head *dev_list; + struct srp_device *srp_dev; struct srp_host *host, *tmp_host; LIST_HEAD(target_list); struct srp_target_port *target, *tmp_target; - unsigned long flags; - dev_list = ib_get_client_data(device, &srp_client); + srp_dev = ib_get_client_data(device, &srp_client); - list_for_each_entry_safe(host, tmp_host, dev_list, list) { + list_for_each_entry_safe(host, tmp_host, &srp_dev->dev_list, list) { class_device_unregister(&host->class_dev); /* * Wait for the sysfs entry to go away, so that no new @@ -1744,15 +1928,13 @@ static void srp_remove_one(struct ib_device *device) * Mark all target ports as removed, so we stop queueing * commands and don't try to reconnect. */ - mutex_lock(&host->target_mutex); - list_for_each_entry_safe(target, tmp_target, - &host->target_list, list) { - spin_lock_irqsave(target->scsi_host->host_lock, flags); - if (target->state != SRP_TARGET_REMOVED) - target->state = SRP_TARGET_REMOVED; - spin_unlock_irqrestore(target->scsi_host->host_lock, flags); + spin_lock(&host->target_lock); + list_for_each_entry(target, &host->target_list, list) { + spin_lock_irq(target->scsi_host->host_lock); + target->state = SRP_TARGET_REMOVED; + spin_unlock_irq(target->scsi_host->host_lock); } - mutex_unlock(&host->target_mutex); + spin_unlock(&host->target_lock); /* * Wait for any reconnection tasks that may have @@ -1770,18 +1952,26 @@ static void srp_remove_one(struct ib_device *device) scsi_host_put(target->scsi_host); } - ib_dereg_mr(host->mr); - ib_dealloc_pd(host->pd); kfree(host); } - kfree(dev_list); + if (srp_dev->fmr_pool) + ib_destroy_fmr_pool(srp_dev->fmr_pool); + ib_dereg_mr(srp_dev->mr); + ib_dealloc_pd(srp_dev->pd); + + kfree(srp_dev); } static int __init srp_init_module(void) { int ret; + srp_template.sg_tablesize = srp_sg_tablesize; + srp_max_iu_len = (sizeof (struct srp_cmd) + + sizeof (struct srp_indirect_buf) + + srp_sg_tablesize * 16); + ret = class_register(&srp_class); if (ret) { printk(KERN_ERR PFX "couldn't register class infiniband_srp\n"); diff --git a/drivers/infiniband/ulp/srp/ib_srp.h b/drivers/infiniband/ulp/srp/ib_srp.h index c5cd43aae860..5b581fb8eb0d 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.h +++ b/drivers/infiniband/ulp/srp/ib_srp.h @@ -46,6 +46,7 @@ #include <rdma/ib_verbs.h> #include <rdma/ib_sa.h> #include <rdma/ib_cm.h> +#include <rdma/ib_fmr_pool.h> enum { SRP_PATH_REC_TIMEOUT_MS = 1000, @@ -55,20 +56,21 @@ enum { SRP_DLID_REDIRECT = 2, SRP_MAX_LUN = 512, - SRP_MAX_IU_LEN = 256, + SRP_DEF_SG_TABLESIZE = 12, SRP_RQ_SHIFT = 6, SRP_RQ_SIZE = 1 << SRP_RQ_SHIFT, SRP_SQ_SIZE = SRP_RQ_SIZE - 1, SRP_CQ_SIZE = SRP_SQ_SIZE + SRP_RQ_SIZE, - SRP_TAG_TSK_MGMT = 1 << (SRP_RQ_SHIFT + 1) + SRP_TAG_TSK_MGMT = 1 << (SRP_RQ_SHIFT + 1), + + SRP_FMR_SIZE = 256, + SRP_FMR_POOL_SIZE = 1024, + SRP_FMR_DIRTY_SIZE = SRP_FMR_POOL_SIZE / 4 }; #define SRP_OP_RECV (1 << 31) -#define SRP_MAX_INDIRECT ((SRP_MAX_IU_LEN - \ - sizeof (struct srp_cmd) - \ - sizeof (struct srp_indirect_buf)) / 16) enum srp_target_state { SRP_TARGET_LIVE, @@ -77,15 +79,24 @@ enum srp_target_state { SRP_TARGET_REMOVED }; -struct srp_host { - u8 initiator_port_id[16]; +struct srp_device { + struct list_head dev_list; struct ib_device *dev; - u8 port; struct ib_pd *pd; struct ib_mr *mr; + struct ib_fmr_pool *fmr_pool; + int fmr_page_shift; + int fmr_page_size; + unsigned long fmr_page_mask; +}; + +struct srp_host { + u8 initiator_port_id[16]; + struct srp_device *dev; + u8 port; struct class_device class_dev; struct list_head target_list; - struct mutex target_mutex; + spinlock_t target_lock; struct completion released; struct list_head list; }; @@ -95,6 +106,7 @@ struct srp_request { struct scsi_cmnd *scmnd; struct srp_iu *cmd; struct srp_iu *tsk_mgmt; + struct ib_pool_fmr *fmr; /* * Fake scatterlist used when scmnd->use_sg==0. Can be killed * when the SCSI midlayer no longer generates non-SG commands. @@ -110,6 +122,7 @@ struct srp_target_port { __be64 id_ext; __be64 ioc_guid; __be64 service_id; + u16 io_class; struct srp_host *srp_host; struct Scsi_Host *scsi_host; char target_name[32]; @@ -126,6 +139,8 @@ struct srp_target_port { int max_ti_iu_len; s32 req_lim; + int zero_req_lim; + unsigned rx_head; struct srp_iu *rx_ring[SRP_RQ_SIZE]; |