diff options
Diffstat (limited to 'drivers/infiniband/ulp/iser')
| -rw-r--r-- | drivers/infiniband/ulp/iser/Kconfig | 4 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/iser/iscsi_iser.c | 644 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/iser/iscsi_iser.h | 305 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/iser/iser_initiator.c | 819 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/iser/iser_memory.c | 750 | ||||
| -rw-r--r-- | drivers/infiniband/ulp/iser/iser_verbs.c | 878 |
6 files changed, 2207 insertions, 1193 deletions
diff --git a/drivers/infiniband/ulp/iser/Kconfig b/drivers/infiniband/ulp/iser/Kconfig index 77dedba829e..d00af71a2cf 100644 --- a/drivers/infiniband/ulp/iser/Kconfig +++ b/drivers/infiniband/ulp/iser/Kconfig @@ -1,6 +1,6 @@ config INFINIBAND_ISER tristate "iSCSI Extensions for RDMA (iSER)" - depends on SCSI && INET + depends on SCSI && INET && INFINIBAND_ADDR_TRANS select SCSI_ISCSI_ATTRS ---help--- Support for the iSCSI Extensions for RDMA (iSER) Protocol @@ -9,4 +9,4 @@ config INFINIBAND_ISER The iSER protocol is defined by IETF. See <http://www.ietf.org/rfc/rfc5046.txt> - and <http://www.infinibandta.org/members/spec/Annex_iSER.PDF> + and <http://members.infinibandta.org/kwspub/spec/Annex_iSER.PDF> diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.c b/drivers/infiniband/ulp/iser/iscsi_iser.c index be1b9fbd416..eb7973957a6 100644 --- a/drivers/infiniband/ulp/iser/iscsi_iser.c +++ b/drivers/infiniband/ulp/iser/iscsi_iser.c @@ -5,6 +5,7 @@ * Copyright (C) 2004 Alex Aizman * Copyright (C) 2005 Mike Christie * Copyright (c) 2005, 2006 Voltaire, Inc. All rights reserved. + * Copyright (c) 2013-2014 Mellanox Technologies. All rights reserved. * maintained by openib-general@openib.org * * This software is available to you under a choice of one of two @@ -42,9 +43,6 @@ * Zhenyu Wang * Modified by: * Erez Zilber - * - * - * $Id: iscsi_iser.c 6965 2006-05-07 11:36:20Z ogerlitz $ */ #include <linux/types.h> @@ -59,6 +57,8 @@ #include <linux/net.h> #include <linux/scatterlist.h> #include <linux/delay.h> +#include <linux/slab.h> +#include <linux/module.h> #include <net/sock.h> @@ -74,19 +74,32 @@ #include "iscsi_iser.h" +static struct scsi_host_template iscsi_iser_sht; +static struct iscsi_transport iscsi_iser_transport; +static struct scsi_transport_template *iscsi_iser_scsi_transport; + static unsigned int iscsi_max_lun = 512; module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO); int iser_debug_level = 0; +bool iser_pi_enable = false; +int iser_pi_guard = 0; -MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover " - "v" DRV_VER " (" DRV_DATE ")"); +MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover"); MODULE_LICENSE("Dual BSD/GPL"); MODULE_AUTHOR("Alex Nezhinsky, Dan Bar Dov, Or Gerlitz"); +MODULE_VERSION(DRV_VER); module_param_named(debug_level, iser_debug_level, int, 0644); MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:disabled)"); +module_param_named(pi_enable, iser_pi_enable, bool, 0644); +MODULE_PARM_DESC(pi_enable, "Enable T10-PI offload support (default:disabled)"); + +module_param_named(pi_guard, iser_pi_guard, int, 0644); +MODULE_PARM_DESC(pi_guard, "T10-PI guard_type, 0:CRC|1:IP_CSUM (default:CRC)"); + +static struct workqueue_struct *release_wq; struct iser_global ig; void @@ -94,28 +107,26 @@ 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); + if (datalen > rx_data_len || (datalen + 4) < rx_data_len) { + iser_err("wrong datalen %d (hdr), %d (IB)\n", + datalen, rx_data_len); rc = ISCSI_ERR_DATALEN; goto error; } + if (datalen != rx_data_len) + iser_dbg("aligned datalen (%d) hdr, %d (IB)\n", + datalen, rx_data_len); + /* 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); - + rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len); if (rc && rc != ISCSI_ERR_NO_SCSI_CMD) goto error; @@ -124,27 +135,65 @@ error: iscsi_conn_failure(conn, rc); } +static int iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode) +{ + struct iscsi_iser_task *iser_task = task->dd_data; + + task->hdr = (struct iscsi_hdr *)&iser_task->desc.iscsi_header; + task->hdr_max = sizeof(iser_task->desc.iscsi_header); + return 0; +} + +int iser_initialize_task_headers(struct iscsi_task *task, + struct iser_tx_desc *tx_desc) +{ + struct iser_conn *ib_conn = task->conn->dd_data; + struct iser_device *device = ib_conn->device; + struct iscsi_iser_task *iser_task = task->dd_data; + u64 dma_addr; + + dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc, + ISER_HEADERS_LEN, DMA_TO_DEVICE); + if (ib_dma_mapping_error(device->ib_device, dma_addr)) + return -ENOMEM; + + tx_desc->dma_addr = dma_addr; + tx_desc->tx_sg[0].addr = tx_desc->dma_addr; + tx_desc->tx_sg[0].length = ISER_HEADERS_LEN; + tx_desc->tx_sg[0].lkey = device->mr->lkey; + iser_task->ib_conn = ib_conn; + return 0; +} /** - * iscsi_iser_cmd_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands + * iscsi_iser_task_init - Initialize task + * @task: iscsi task * - **/ + * Initialize the task for the scsi command or mgmt command. + */ static int -iscsi_iser_cmd_init(struct iscsi_cmd_task *ctask) +iscsi_iser_task_init(struct iscsi_task *task) { - struct iscsi_iser_conn *iser_conn = ctask->conn->dd_data; - struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + struct iscsi_iser_task *iser_task = task->dd_data; + + if (iser_initialize_task_headers(task, &iser_task->desc)) + return -ENOMEM; + + /* mgmt task */ + if (!task->sc) + return 0; + + iser_task->command_sent = 0; + iser_task_rdma_init(iser_task); + iser_task->sc = task->sc; - iser_ctask->command_sent = 0; - iser_ctask->iser_conn = iser_conn; - iser_ctask_rdma_init(iser_ctask); return 0; } /** - * iscsi_mtask_xmit - xmit management(immediate) task + * iscsi_iser_mtask_xmit - xmit management(immediate) task * @conn: iscsi connection - * @mtask: task management task + * @task: task management task * * Notes: * The function can return -EAGAIN in which case caller must @@ -153,119 +202,120 @@ iscsi_iser_cmd_init(struct iscsi_cmd_task *ctask) * **/ static int -iscsi_iser_mtask_xmit(struct iscsi_conn *conn, - struct iscsi_mgmt_task *mtask) +iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task) { int error = 0; - debug_scsi("mtask deq [cid %d itt 0x%x]\n", conn->id, mtask->itt); + iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt); - error = iser_send_control(conn, mtask); + error = iser_send_control(conn, task); - /* since iser xmits control with zero copy, mtasks can not be recycled + /* since iser xmits control with zero copy, tasks 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 yes, the task is recycled at iscsi_complete_pdu + * - if no, the task is recycled at iser_snd_completion */ - if (error && error != -ENOBUFS) - 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) +iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn, + struct iscsi_task *task) { - struct iscsi_data hdr; + struct iscsi_r2t_info *r2t = &task->unsol_r2t; + struct iscsi_data hdr; int error = 0; /* Send data-out PDUs while there's still unsolicited data to send */ - while (ctask->unsol_count > 0) { - iscsi_prep_unsolicit_data_pdu(ctask, &hdr); - debug_scsi("Sending data-out: itt 0x%x, data count %d\n", - hdr.itt, ctask->data_count); + while (iscsi_task_has_unsol_data(task)) { + iscsi_prep_data_out_pdu(task, r2t, &hdr); + iser_dbg("Sending data-out: itt 0x%x, data count %d\n", + hdr.itt, r2t->data_count); /* the buffer description has been passed with the command */ /* Send the command */ - error = iser_send_data_out(conn, ctask, &hdr); + error = iser_send_data_out(conn, task, &hdr); if (error) { - ctask->unsol_datasn--; - goto iscsi_iser_ctask_xmit_unsol_data_exit; + r2t->datasn--; + goto iscsi_iser_task_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); + r2t->sent += r2t->data_count; + iser_dbg("Need to send %d more as data-out PDUs\n", + r2t->data_length - r2t->sent); } -iscsi_iser_ctask_xmit_unsol_data_exit: +iscsi_iser_task_xmit_unsol_data_exit: return error; } static int -iscsi_iser_ctask_xmit(struct iscsi_conn *conn, - struct iscsi_cmd_task *ctask) +iscsi_iser_task_xmit(struct iscsi_task *task) { - struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + struct iscsi_conn *conn = task->conn; + struct iscsi_iser_task *iser_task = task->dd_data; int error = 0; - if (ctask->sc->sc_data_direction == DMA_TO_DEVICE) { - BUG_ON(scsi_bufflen(ctask->sc) == 0); + if (!task->sc) + return iscsi_iser_mtask_xmit(conn, task); + + if (task->sc->sc_data_direction == DMA_TO_DEVICE) { + BUG_ON(scsi_bufflen(task->sc) == 0); - debug_scsi("cmd [itt %x total %d imm %d unsol_data %d\n", - ctask->itt, scsi_bufflen(ctask->sc), - ctask->imm_count, ctask->unsol_count); + iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n", + task->itt, scsi_bufflen(task->sc), + task->imm_count, task->unsol_r2t.data_length); } - debug_scsi("ctask deq [cid %d itt 0x%x]\n", - conn->id, ctask->itt); + iser_dbg("ctask xmit [cid %d itt 0x%x]\n", + conn->id, task->itt); /* Send the cmd PDU */ - if (!iser_ctask->command_sent) { - error = iser_send_command(conn, ctask); + if (!iser_task->command_sent) { + error = iser_send_command(conn, task); if (error) - goto iscsi_iser_ctask_xmit_exit; - iser_ctask->command_sent = 1; + goto iscsi_iser_task_xmit_exit; + iser_task->command_sent = 1; } /* Send unsolicited data-out PDU(s) if necessary */ - if (ctask->unsol_count) - error = iscsi_iser_ctask_xmit_unsol_data(conn, ctask); + if (iscsi_task_has_unsol_data(task)) + error = iscsi_iser_task_xmit_unsol_data(conn, task); - iscsi_iser_ctask_xmit_exit: - if (error && error != -ENOBUFS) - iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED); + iscsi_iser_task_xmit_exit: return error; } -static void -iscsi_iser_cleanup_ctask(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask) +static void iscsi_iser_cleanup_task(struct iscsi_task *task) { - struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + struct iscsi_iser_task *iser_task = task->dd_data; + struct iser_tx_desc *tx_desc = &iser_task->desc; + struct iser_conn *ib_conn = task->conn->dd_data; + struct iser_device *device = ib_conn->device; + + ib_dma_unmap_single(device->ib_device, + tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE); - if (iser_ctask->status == ISER_TASK_STATUS_STARTED) { - iser_ctask->status = ISER_TASK_STATUS_COMPLETED; - iser_ctask_rdma_finalize(iser_ctask); + /* mgmt tasks do not need special cleanup */ + if (!task->sc) + return; + + if (iser_task->status == ISER_TASK_STATUS_STARTED) { + iser_task->status = ISER_TASK_STATUS_COMPLETED; + iser_task_rdma_finalize(iser_task); } } -static struct iser_conn * -iscsi_iser_ib_conn_lookup(__u64 ep_handle) +static u8 iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector) { - struct iser_conn *ib_conn; - struct iser_conn *uib_conn = (struct iser_conn *)(unsigned long)ep_handle; + struct iscsi_iser_task *iser_task = task->dd_data; - 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; + if (iser_task->dir[ISER_DIR_IN]) + return iser_check_task_pi_status(iser_task, ISER_DIR_IN, + sector); + else + return iser_check_task_pi_status(iser_task, ISER_DIR_OUT, + sector); } static struct iscsi_cls_conn * @@ -273,9 +323,8 @@ 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); + cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx); if (!cls_conn) return NULL; conn = cls_conn->dd_data; @@ -284,35 +333,9 @@ iscsi_iser_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx) * 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; + conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN; 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); - if (iser_conn->ib_conn) - iser_conn->ib_conn->iser_conn = NULL; - kfree(iser_conn); } static int @@ -321,8 +344,9 @@ iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session, int is_leading) { struct iscsi_conn *conn = cls_conn->dd_data; - struct iscsi_iser_conn *iser_conn; + struct iscsi_session *session; struct iser_conn *ib_conn; + struct iscsi_endpoint *ep; int error; error = iscsi_conn_bind(cls_session, cls_conn, is_leading); @@ -331,21 +355,25 @@ iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session, /* 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) { + ep = iscsi_lookup_endpoint(transport_eph); + if (!ep) { iser_err("can't bind eph %llx\n", (unsigned long long)transport_eph); return -EINVAL; } + ib_conn = ep->dd_data; + + session = conn->session; + if (iser_alloc_rx_descriptors(ib_conn, session)) + return -ENOMEM; + /* 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; + iser_info("binding iscsi conn %p to ib_conn %p\n", conn, ib_conn); - conn->recv_lock = &iser_conn->lock; + conn->dd_data = ib_conn; + ib_conn->iscsi_conn = conn; return 0; } @@ -353,65 +381,118 @@ iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session, static int iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn) { - struct iscsi_conn *conn = cls_conn->dd_data; - int err; + struct iscsi_conn *iscsi_conn; + struct iser_conn *ib_conn; - err = iser_conn_set_full_featured_mode(conn); - if (err) - return err; + iscsi_conn = cls_conn->dd_data; + ib_conn = iscsi_conn->dd_data; + reinit_completion(&ib_conn->stop_completion); return iscsi_conn_start(cls_conn); } -static struct iscsi_transport iscsi_iser_transport; +static void +iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag) +{ + struct iscsi_conn *conn = cls_conn->dd_data; + struct iser_conn *ib_conn = conn->dd_data; + + iser_dbg("stopping iscsi_conn: %p, ib_conn: %p\n", conn, ib_conn); + iscsi_conn_stop(cls_conn, flag); + + /* + * Userspace may have goofed up and not bound the connection or + * might have only partially setup the connection. + */ + if (ib_conn) { + conn->dd_data = NULL; + complete(&ib_conn->stop_completion); + } +} + +static void iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session) +{ + struct Scsi_Host *shost = iscsi_session_to_shost(cls_session); + + iscsi_session_teardown(cls_session); + iscsi_host_remove(shost); + iscsi_host_free(shost); +} + +static inline unsigned int +iser_dif_prot_caps(int prot_caps) +{ + return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ? SHOST_DIF_TYPE1_PROTECTION | + SHOST_DIX_TYPE1_PROTECTION : 0) | + ((prot_caps & IB_PROT_T10DIF_TYPE_2) ? SHOST_DIF_TYPE2_PROTECTION | + SHOST_DIX_TYPE2_PROTECTION : 0) | + ((prot_caps & IB_PROT_T10DIF_TYPE_3) ? SHOST_DIF_TYPE3_PROTECTION | + SHOST_DIX_TYPE3_PROTECTION : 0); +} static struct iscsi_cls_session * -iscsi_iser_session_create(struct iscsi_transport *iscsit, - struct scsi_transport_template *scsit, - uint16_t cmds_max, uint16_t qdepth, - uint32_t initial_cmdsn, uint32_t *hostno) +iscsi_iser_session_create(struct iscsi_endpoint *ep, + uint16_t cmds_max, uint16_t qdepth, + uint32_t initial_cmdsn) { 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; + struct Scsi_Host *shost; + struct iser_conn *ib_conn = NULL; + + shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0); + if (!shost) + return NULL; + shost->transportt = iscsi_iser_scsi_transport; + shost->cmd_per_lun = qdepth; + shost->max_lun = iscsi_max_lun; + shost->max_id = 0; + shost->max_channel = 0; + shost->max_cmd_len = 16; /* - * we do not support setting can_queue cmd_per_lun from userspace yet - * because we preallocate so many resources + * older userspace tools (before 2.0-870) did not pass us + * the leading conn's ep so this will be NULL; */ - cls_session = iscsi_session_setup(iscsit, scsit, - ISCSI_DEF_XMIT_CMDS_MAX, - ISCSI_MAX_CMD_PER_LUN, - sizeof(struct iscsi_iser_cmd_task), - sizeof(struct iser_desc), - initial_cmdsn, &hn); - if (!cls_session) - return NULL; + if (ep) { + ib_conn = ep->dd_data; + if (ib_conn->pi_support) { + u32 sig_caps = ib_conn->device->dev_attr.sig_prot_cap; + + scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps)); + if (iser_pi_guard) + scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP); + else + scsi_host_set_guard(shost, SHOST_DIX_GUARD_CRC); + } + } - *hostno = hn; - session = class_to_transport_session(cls_session); + if (iscsi_host_add(shost, + ep ? ib_conn->device->ib_device->dma_device : NULL)) + goto free_host; - /* 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; - ctask->hdr_max = sizeof(iser_ctask->desc.iscsi_header); + if (cmds_max > ISER_DEF_XMIT_CMDS_MAX) { + iser_info("cmds_max changed from %u to %u\n", + cmds_max, ISER_DEF_XMIT_CMDS_MAX); + cmds_max = ISER_DEF_XMIT_CMDS_MAX; } - 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; - } + cls_session = iscsi_session_setup(&iscsi_iser_transport, shost, + cmds_max, 0, + sizeof(struct iscsi_iser_task), + initial_cmdsn, 0); + if (!cls_session) + goto remove_host; + session = cls_session->dd_data; + shost->can_queue = session->scsi_cmds_max; return cls_session; + +remove_host: + iscsi_host_remove(shost); +free_host: + iscsi_host_free(shost); + return NULL; } static int @@ -427,28 +508,28 @@ iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn, case ISCSI_PARAM_HDRDGST_EN: sscanf(buf, "%d", &value); if (value) { - printk(KERN_ERR "DataDigest wasn't negotiated to None"); + iser_err("DataDigest wasn't negotiated to None\n"); return -EPROTO; } break; case ISCSI_PARAM_DATADGST_EN: sscanf(buf, "%d", &value); if (value) { - printk(KERN_ERR "DataDigest wasn't negotiated to None"); + iser_err("DataDigest wasn't negotiated to None\n"); return -EPROTO; } break; case ISCSI_PARAM_IFMARKER_EN: sscanf(buf, "%d", &value); if (value) { - printk(KERN_ERR "IFMarker wasn't negotiated to No"); + iser_err("IFMarker wasn't negotiated to No\n"); return -EPROTO; } break; case ISCSI_PARAM_OFMARKER_EN: sscanf(buf, "%d", &value); if (value) { - printk(KERN_ERR "OFMarker wasn't negotiated to No"); + iser_err("OFMarker wasn't negotiated to No\n"); return -EPROTO; } break; @@ -473,43 +554,71 @@ iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *s 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; + stats->custom_length = 4; 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; + strcpy(stats->custom[3].desc, "fmr_unalign_cnt"); + stats->custom[3].value = conn->fmr_unalign_cnt; } -static int -iscsi_iser_ep_connect(struct sockaddr *dst_addr, int non_blocking, - __u64 *ep_handle) +static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep, + enum iscsi_param param, char *buf) +{ + struct iser_conn *ib_conn = ep->dd_data; + int len; + + switch (param) { + case ISCSI_PARAM_CONN_PORT: + case ISCSI_PARAM_CONN_ADDRESS: + if (!ib_conn || !ib_conn->cma_id) + return -ENOTCONN; + + return iscsi_conn_get_addr_param((struct sockaddr_storage *) + &ib_conn->cma_id->route.addr.dst_addr, + param, buf); + break; + default: + return -ENOSYS; + } + + return len; +} + +static struct iscsi_endpoint * +iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr, + int non_blocking) { int err; struct iser_conn *ib_conn; + struct iscsi_endpoint *ep; - err = iser_conn_init(&ib_conn); - if (err) - goto out; + ep = iscsi_create_endpoint(sizeof(*ib_conn)); + if (!ep) + return ERR_PTR(-ENOMEM); - err = iser_connect(ib_conn, NULL, (struct sockaddr_in *)dst_addr, non_blocking); - if (!err) - *ep_handle = (__u64)(unsigned long)ib_conn; + ib_conn = ep->dd_data; + ib_conn->ep = ep; + iser_conn_init(ib_conn); -out: - return err; + err = iser_connect(ib_conn, NULL, (struct sockaddr_in *)dst_addr, + non_blocking); + if (err) + return ERR_PTR(err); + + return ep; } static int -iscsi_iser_ep_poll(__u64 ep_handle, int timeout_ms) +iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms) { - struct iser_conn *ib_conn = iscsi_iser_ib_conn_lookup(ep_handle); + struct iser_conn *ib_conn; int rc; - if (!ib_conn) - return -EINVAL; - + ib_conn = ep->dd_data; rc = wait_event_interruptible_timeout(ib_conn->wait, ib_conn->state == ISER_CONN_UP, msecs_to_jiffies(timeout_ms)); @@ -520,7 +629,7 @@ iscsi_iser_ep_poll(__u64 ep_handle, int timeout_ms) ib_conn->state == ISER_CONN_DOWN)) rc = -1; - iser_err("ib conn %p rc = %d\n", ib_conn, rc); + iser_info("ib conn %p rc = %d\n", ib_conn, rc); if (rc > 0) return 1; /* success, this is the equivalent of POLLOUT */ @@ -531,30 +640,93 @@ iscsi_iser_ep_poll(__u64 ep_handle, int timeout_ms) } static void -iscsi_iser_ep_disconnect(__u64 ep_handle) +iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep) { struct iser_conn *ib_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); + ib_conn = ep->dd_data; + iser_info("ep %p ib conn %p state %d\n", ep, ib_conn, ib_conn->state); iser_conn_terminate(ib_conn); + + /* + * if iser_conn and iscsi_conn are bound, we must wait iscsi_conn_stop + * call and ISER_CONN_DOWN state before freeing the iser resources. + * otherwise we are safe to free resources immediately. + */ + if (ib_conn->iscsi_conn) { + INIT_WORK(&ib_conn->release_work, iser_release_work); + queue_work(release_wq, &ib_conn->release_work); + } else { + iser_conn_release(ib_conn); + } +} + +static umode_t iser_attr_is_visible(int param_type, int param) +{ + switch (param_type) { + case ISCSI_HOST_PARAM: + switch (param) { + case ISCSI_HOST_PARAM_NETDEV_NAME: + case ISCSI_HOST_PARAM_HWADDRESS: + case ISCSI_HOST_PARAM_INITIATOR_NAME: + return S_IRUGO; + default: + return 0; + } + case ISCSI_PARAM: + switch (param) { + case ISCSI_PARAM_MAX_RECV_DLENGTH: + case ISCSI_PARAM_MAX_XMIT_DLENGTH: + case ISCSI_PARAM_HDRDGST_EN: + case ISCSI_PARAM_DATADGST_EN: + case ISCSI_PARAM_CONN_ADDRESS: + case ISCSI_PARAM_CONN_PORT: + case ISCSI_PARAM_EXP_STATSN: + case ISCSI_PARAM_PERSISTENT_ADDRESS: + case ISCSI_PARAM_PERSISTENT_PORT: + case ISCSI_PARAM_PING_TMO: + case ISCSI_PARAM_RECV_TMO: + case ISCSI_PARAM_INITIAL_R2T_EN: + case ISCSI_PARAM_MAX_R2T: + case ISCSI_PARAM_IMM_DATA_EN: + case ISCSI_PARAM_FIRST_BURST: + case ISCSI_PARAM_MAX_BURST: + case ISCSI_PARAM_PDU_INORDER_EN: + case ISCSI_PARAM_DATASEQ_INORDER_EN: + case ISCSI_PARAM_TARGET_NAME: + case ISCSI_PARAM_TPGT: + case ISCSI_PARAM_USERNAME: + case ISCSI_PARAM_PASSWORD: + case ISCSI_PARAM_USERNAME_IN: + case ISCSI_PARAM_PASSWORD_IN: + case ISCSI_PARAM_FAST_ABORT: + case ISCSI_PARAM_ABORT_TMO: + case ISCSI_PARAM_LU_RESET_TMO: + case ISCSI_PARAM_TGT_RESET_TMO: + case ISCSI_PARAM_IFACE_NAME: + case ISCSI_PARAM_INITIATOR_NAME: + case ISCSI_PARAM_DISCOVERY_SESS: + return S_IRUGO; + default: + return 0; + } + } + + return 0; } static struct scsi_host_template iscsi_iser_sht = { .module = THIS_MODULE, - .name = "iSCSI Initiator over iSER, v." DRV_VER, + .name = "iSCSI Initiator over iSER", .queuecommand = iscsi_queuecommand, .change_queue_depth = iscsi_change_queue_depth, - .can_queue = ISCSI_DEF_XMIT_CMDS_MAX - 1, .sg_tablesize = ISCSI_ISER_SG_TABLESIZE, .max_sectors = 1024, - .cmd_per_lun = ISCSI_MAX_CMD_PER_LUN, + .cmd_per_lun = ISER_DEF_CMD_PER_LUN, .eh_abort_handler = iscsi_eh_abort, .eh_device_reset_handler= iscsi_eh_device_reset, - .eh_host_reset_handler = iscsi_eh_host_reset, + .eh_target_reset_handler = iscsi_eh_recover_target, + .target_alloc = iscsi_target_alloc, .use_clustering = DISABLE_CLUSTERING, .proc_name = "iscsi_iser", .this_id = -1, @@ -563,55 +735,32 @@ static struct scsi_host_template iscsi_iser_sht = { 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 | - ISCSI_EXP_STATSN | - ISCSI_PERSISTENT_PORT | - ISCSI_PERSISTENT_ADDRESS | - ISCSI_TARGET_NAME | ISCSI_TPGT | - ISCSI_USERNAME | ISCSI_PASSWORD | - ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN | - ISCSI_FAST_ABORT | ISCSI_ABORT_TMO | - ISCSI_PING_TMO | ISCSI_RECV_TMO, - .host_param_mask = ISCSI_HOST_HWADDRESS | - ISCSI_HOST_NETDEV_NAME | - ISCSI_HOST_INITIATOR_NAME, - .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, + .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO, /* session management */ .create_session = iscsi_iser_session_create, - .destroy_session = iscsi_session_teardown, + .destroy_session = iscsi_iser_session_destroy, /* connection management */ .create_conn = iscsi_iser_conn_create, .bind_conn = iscsi_iser_conn_bind, - .destroy_conn = iscsi_iser_conn_destroy, + .destroy_conn = iscsi_conn_teardown, + .attr_is_visible = iser_attr_is_visible, .set_param = iscsi_iser_set_param, .get_conn_param = iscsi_conn_get_param, + .get_ep_param = iscsi_iser_get_ep_param, .get_session_param = iscsi_session_get_param, .start_conn = iscsi_iser_conn_start, - .stop_conn = iscsi_conn_stop, + .stop_conn = iscsi_iser_conn_stop, /* iscsi host params */ .get_host_param = iscsi_host_get_param, .set_host_param = iscsi_host_set_param, /* 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, + .init_task = iscsi_iser_task_init, + .xmit_task = iscsi_iser_task_xmit, + .cleanup_task = iscsi_iser_cleanup_task, + .alloc_pdu = iscsi_iser_pdu_alloc, + .check_protection = iscsi_iser_check_protection, /* recovery */ .session_recovery_timedout = iscsi_session_recovery_timedout, @@ -627,16 +776,14 @@ static int __init iser_init(void) iser_dbg("Starting iSER datamover...\n"); if (iscsi_max_lun < 1) { - printk(KERN_ERR "Invalid max_lun value of %u\n", iscsi_max_lun); + iser_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), + sizeof(struct iser_tx_desc), 0, SLAB_HWCACHE_ALIGN, NULL); if (ig.desc_cache == NULL) @@ -648,7 +795,15 @@ static int __init iser_init(void) mutex_init(&ig.connlist_mutex); INIT_LIST_HEAD(&ig.connlist); - if (!iscsi_register_transport(&iscsi_iser_transport)) { + release_wq = alloc_workqueue("release workqueue", 0, 0); + if (!release_wq) { + iser_err("failed to allocate release workqueue\n"); + return -ENOMEM; + } + + iscsi_iser_scsi_transport = iscsi_register_transport( + &iscsi_iser_transport); + if (!iscsi_iser_scsi_transport) { iser_err("iscsi_register_transport failed\n"); err = -EINVAL; goto register_transport_failure; @@ -664,7 +819,24 @@ register_transport_failure: static void __exit iser_exit(void) { + struct iser_conn *ib_conn, *n; + int connlist_empty; + iser_dbg("Removing iSER datamover...\n"); + destroy_workqueue(release_wq); + + mutex_lock(&ig.connlist_mutex); + connlist_empty = list_empty(&ig.connlist); + mutex_unlock(&ig.connlist_mutex); + + if (!connlist_empty) { + iser_err("Error cleanup stage completed but we still have iser " + "connections, destroying them anyway.\n"); + list_for_each_entry_safe(ib_conn, n, &ig.connlist, conn_list) { + iser_conn_release(ib_conn); + } + } + iscsi_unregister_transport(&iscsi_iser_transport); kmem_cache_destroy(ig.desc_cache); } diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.h b/drivers/infiniband/ulp/iser/iscsi_iser.h index 1ee867b1b34..97cd385bf7f 100644 --- a/drivers/infiniband/ulp/iser/iscsi_iser.h +++ b/drivers/infiniband/ulp/iser/iscsi_iser.h @@ -8,6 +8,7 @@ * * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved. * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved. + * Copyright (c) 2013-2014 Mellanox Technologies. 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 @@ -36,17 +37,19 @@ * 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 <linux/printk.h> #include <scsi/libiscsi.h> #include <scsi/scsi_transport_iscsi.h> +#include <scsi/scsi_cmnd.h> +#include <scsi/scsi_device.h> +#include <linux/interrupt.h> #include <linux/wait.h> #include <linux/sched.h> #include <linux/list.h> @@ -66,16 +69,29 @@ #define DRV_NAME "iser" #define PFX DRV_NAME ": " -#define DRV_VER "0.1" -#define DRV_DATE "May 7th, 2006" +#define DRV_VER "1.4" #define iser_dbg(fmt, arg...) \ do { \ - if (iser_debug_level > 0) \ + if (iser_debug_level > 2) \ printk(KERN_DEBUG PFX "%s:" fmt,\ __func__ , ## arg); \ } while (0) +#define iser_warn(fmt, arg...) \ + do { \ + if (iser_debug_level > 0) \ + pr_warn(PFX "%s:" fmt, \ + __func__ , ## arg); \ + } while (0) + +#define iser_info(fmt, arg...) \ + do { \ + if (iser_debug_level > 1) \ + pr_info(PFX "%s:" fmt, \ + __func__ , ## arg); \ + } while (0) + #define iser_err(fmt, arg...) \ do { \ printk(KERN_ERR PFX "%s:" fmt, \ @@ -83,13 +99,18 @@ } while (0) #define SHIFT_4K 12 -#define SIZE_4K (1UL << SHIFT_4K) +#define SIZE_4K (1ULL << SHIFT_4K) #define MASK_4K (~(SIZE_4K-1)) - /* support upto 512KB in one RDMA */ + /* support up to 512KB in one RDMA */ #define ISCSI_ISER_SG_TABLESIZE (0x80000 >> SHIFT_4K) -#define ISCSI_ISER_MAX_LUN 256 -#define ISCSI_ISER_MAX_CMD_LEN 16 +#define ISER_DEF_XMIT_CMDS_DEFAULT 512 +#if ISCSI_DEF_XMIT_CMDS_MAX > ISER_DEF_XMIT_CMDS_DEFAULT + #define ISER_DEF_XMIT_CMDS_MAX ISCSI_DEF_XMIT_CMDS_MAX +#else + #define ISER_DEF_XMIT_CMDS_MAX ISER_DEF_XMIT_CMDS_DEFAULT +#endif +#define ISER_DEF_CMD_PER_LUN ISER_DEF_XMIT_CMDS_MAX /* QP settings */ /* Maximal bounds on received asynchronous PDUs */ @@ -98,9 +119,9 @@ #define ISER_MAX_TX_MISC_PDUS 6 /* NOOP_OUT(2), TEXT(1), * * SCSI_TMFUNC(2), LOGOUT(1) */ -#define ISER_QP_MAX_RECV_DTOS (ISCSI_DEF_XMIT_CMDS_MAX + \ - ISER_MAX_RX_MISC_PDUS + \ - ISER_MAX_TX_MISC_PDUS) +#define ISER_QP_MAX_RECV_DTOS (ISER_DEF_XMIT_CMDS_MAX) + +#define ISER_MIN_POSTED_RX (ISER_DEF_XMIT_CMDS_MAX >> 2) /* 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 * @@ -110,15 +131,26 @@ #define ISER_INFLIGHT_DATAOUTS 8 -#define ISER_QP_MAX_REQ_DTOS (ISCSI_DEF_XMIT_CMDS_MAX * \ +#define ISER_QP_MAX_REQ_DTOS (ISER_DEF_XMIT_CMDS_MAX * \ (1 + ISER_INFLIGHT_DATAOUTS) + \ ISER_MAX_TX_MISC_PDUS + \ ISER_MAX_RX_MISC_PDUS) +/* Max registration work requests per command */ +#define ISER_MAX_REG_WR_PER_CMD 5 + +/* For Signature we don't support DATAOUTs so no need to make room for them */ +#define ISER_QP_SIG_MAX_REQ_DTOS (ISER_DEF_XMIT_CMDS_MAX * \ + (1 + ISER_MAX_REG_WR_PER_CMD) + \ + ISER_MAX_TX_MISC_PDUS + \ + ISER_MAX_RX_MISC_PDUS) + #define ISER_VER 0x10 #define ISER_WSV 0x08 #define ISER_RSV 0x04 +#define ISER_FASTREG_LI_WRID 0xffffffffffffffffULL + struct iser_hdr { u8 flags; u8 rsvd[3]; @@ -129,6 +161,21 @@ struct iser_hdr { } __attribute__((packed)); +#define ISER_ZBVA_NOT_SUPPORTED 0x80 +#define ISER_SEND_W_INV_NOT_SUPPORTED 0x40 + +struct iser_cm_hdr { + u8 flags; + u8 rsvd[3]; +} __packed; + +/* Constant PDU lengths calculations */ +#define ISER_HEADERS_LEN (sizeof(struct iser_hdr) + sizeof(struct iscsi_hdr)) + +#define ISER_RECV_DATA_SEG_LEN 128 +#define ISER_RX_PAYLOAD_SIZE (ISER_HEADERS_LEN + ISER_RECV_DATA_SEG_LEN) +#define ISER_RX_LOGIN_SIZE (ISER_HEADERS_LEN + ISCSI_DEF_MAX_RECV_SEG_LEN) + /* Length of an object name string */ #define ISER_OBJECT_NAME_SIZE 64 @@ -166,8 +213,9 @@ struct iser_data_buf { /* fwd declarations */ struct iser_device; -struct iscsi_iser_conn; -struct iscsi_iser_cmd_task; +struct iser_cq_desc; +struct iscsi_iser_task; +struct iscsi_endpoint; struct iser_mem_reg { u32 lkey; @@ -175,98 +223,154 @@ struct iser_mem_reg { u64 va; u64 len; void *mem_h; - int is_fmr; + int is_mr; }; struct iser_regd_buf { struct iser_mem_reg reg; /* memory registration info */ void *virt_addr; struct iser_device *device; /* device->device for dma_unmap */ - u64 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 iser_conn *ib_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_tx_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; + u64 dma_addr; + /* sg[0] points to iser/iscsi headers, sg[1] optionally points to either + of immediate data, unsolicited data-out or control (login,text) */ + struct ib_sge tx_sg[2]; + int num_sge; }; +#define ISER_RX_PAD_SIZE (256 - (ISER_RX_PAYLOAD_SIZE + \ + sizeof(u64) + sizeof(struct ib_sge))) +struct iser_rx_desc { + struct iser_hdr iser_header; + struct iscsi_hdr iscsi_header; + char data[ISER_RECV_DATA_SEG_LEN]; + u64 dma_addr; + struct ib_sge rx_sg; + char pad[ISER_RX_PAD_SIZE]; +} __attribute__((packed)); + +#define ISER_MAX_CQ 4 + +struct iser_conn; +struct iscsi_iser_task; + struct iser_device { struct ib_device *ib_device; struct ib_pd *pd; - struct ib_cq *cq; + struct ib_device_attr dev_attr; + struct ib_cq *rx_cq[ISER_MAX_CQ]; + struct ib_cq *tx_cq[ISER_MAX_CQ]; struct ib_mr *mr; - struct tasklet_struct cq_tasklet; + struct tasklet_struct cq_tasklet[ISER_MAX_CQ]; + struct ib_event_handler event_handler; struct list_head ig_list; /* entry in ig devices list */ int refcount; + int cq_active_qps[ISER_MAX_CQ]; + int cqs_used; + struct iser_cq_desc *cq_desc; + int (*iser_alloc_rdma_reg_res)(struct iser_conn *ib_conn, + unsigned cmds_max); + void (*iser_free_rdma_reg_res)(struct iser_conn *ib_conn); + int (*iser_reg_rdma_mem)(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir); + void (*iser_unreg_rdma_mem)(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir); +}; + +#define ISER_CHECK_GUARD 0xc0 +#define ISER_CHECK_REFTAG 0x0f +#define ISER_CHECK_APPTAG 0x30 + +enum iser_reg_indicator { + ISER_DATA_KEY_VALID = 1 << 0, + ISER_PROT_KEY_VALID = 1 << 1, + ISER_SIG_KEY_VALID = 1 << 2, + ISER_FASTREG_PROTECTED = 1 << 3, +}; + +struct iser_pi_context { + struct ib_mr *prot_mr; + struct ib_fast_reg_page_list *prot_frpl; + struct ib_mr *sig_mr; +}; + +struct fast_reg_descriptor { + struct list_head list; + /* For fast registration - FRWR */ + struct ib_mr *data_mr; + struct ib_fast_reg_page_list *data_frpl; + struct iser_pi_context *pi_ctx; + /* registration indicators container */ + u8 reg_indicators; }; struct iser_conn { - struct iscsi_iser_conn *iser_conn; /* iser conn for upcalls */ + struct iscsi_conn *iscsi_conn; + struct iscsi_endpoint *ep; enum iser_ib_conn_state state; /* rdma connection state */ + atomic_t refcount; 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 */ + unsigned qp_max_recv_dtos; /* num of rx buffers */ + unsigned qp_max_recv_dtos_mask; /* above minus 1 */ + unsigned min_posted_rx; /* qp_max_recv_dtos >> 2 */ + int post_recv_buf_count; /* posted rx count */ atomic_t post_send_buf_count; /* posted tx count */ char name[ISER_OBJECT_NAME_SIZE]; - struct iser_page_vec *page_vec; /* represents SG to fmr maps* - * maps serialized as tx is*/ + struct work_struct release_work; + struct completion stop_completion; 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; + char *login_buf; + char *login_req_buf, *login_resp_buf; + u64 login_req_dma, login_resp_dma; + unsigned int rx_desc_head; + struct iser_rx_desc *rx_descs; + struct ib_recv_wr rx_wr[ISER_MIN_POSTED_RX]; + bool pi_support; + + /* Connection memory registration pool */ + union { + struct { + struct ib_fmr_pool *pool; /* pool of IB FMRs */ + struct iser_page_vec *page_vec; /* represents SG to fmr maps* + * maps serialized as tx is*/ + } fmr; + struct { + struct list_head pool; + int pool_size; + } fastreg; + }; }; -struct iscsi_iser_cmd_task { - struct iser_desc desc; - struct iscsi_iser_conn *iser_conn; +struct iscsi_iser_task { + struct iser_tx_desc desc; + struct iser_conn *ib_conn; enum iser_task_status status; + struct scsi_cmnd *sc; 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_data_buf prot[ISER_DIRS_NUM]; /* prot desc */ + struct iser_data_buf prot_copy[ISER_DIRS_NUM];/* prot copy */ }; struct iser_page_vec { @@ -276,6 +380,11 @@ struct iser_page_vec { int data_size; }; +struct iser_cq_desc { + struct iser_device *device; + int cq_index; +}; + struct iser_global { struct mutex device_list_mutex;/* */ struct list_head device_list; /* all iSER devices */ @@ -287,51 +396,56 @@ struct iser_global { extern struct iser_global ig; extern int iser_debug_level; +extern bool iser_pi_enable; +extern int iser_pi_guard; /* 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_control(struct iscsi_conn *conn, + struct iscsi_task *task); -int iser_send_command(struct iscsi_conn *conn, - struct iscsi_cmd_task *ctask); +int iser_send_command(struct iscsi_conn *conn, + struct iscsi_task *task); -int iser_send_data_out(struct iscsi_conn *conn, - struct iscsi_cmd_task *ctask, - struct iscsi_data *hdr); +int iser_send_data_out(struct iscsi_conn *conn, + struct iscsi_task *task, + 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_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_conn_terminate(struct iser_conn *ib_conn); -void iser_snd_completion(struct iser_desc *desc); +void iser_release_work(struct work_struct *work); -void iser_ctask_rdma_init(struct iscsi_iser_cmd_task *ctask); +void iser_rcv_completion(struct iser_rx_desc *desc, + unsigned long dto_xfer_len, + struct iser_conn *ib_conn); -void iser_ctask_rdma_finalize(struct iscsi_iser_cmd_task *ctask); +void iser_snd_completion(struct iser_tx_desc *desc, struct iser_conn *ib_conn); -void iser_dto_buffs_release(struct iser_dto *dto); +void iser_task_rdma_init(struct iscsi_iser_task *task); -int iser_regd_buff_release(struct iser_regd_buf *regd_buf); +void iser_task_rdma_finalize(struct iscsi_iser_task *task); -void iser_reg_single(struct iser_device *device, - struct iser_regd_buf *regd_buf, - enum dma_data_direction direction); +void iser_free_rx_descriptors(struct iser_conn *ib_conn); -void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_cmd_task *ctask, - enum iser_data_dir cmd_dir); +void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *iser_task, + struct iser_data_buf *mem, + struct iser_data_buf *mem_copy, + 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_reg_rdma_mem_fmr(struct iscsi_iser_task *task, + enum iser_data_dir cmd_dir); +int iser_reg_rdma_mem_fastreg(struct iscsi_iser_task *task, + enum iser_data_dir cmd_dir); int iser_connect(struct iser_conn *ib_conn, struct sockaddr_in *src_addr, @@ -342,18 +456,29 @@ 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); +void iser_unreg_mem_fmr(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir); +void iser_unreg_mem_fastreg(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir); -int iser_conn_state_comp(struct iser_conn *ib_conn, - enum iser_ib_conn_state comp); +int iser_post_recvl(struct iser_conn *ib_conn); +int iser_post_recvm(struct iser_conn *ib_conn, int count); +int iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc); -int iser_dma_map_task_data(struct iscsi_iser_cmd_task *iser_ctask, +int iser_dma_map_task_data(struct iscsi_iser_task *iser_task, struct iser_data_buf *data, enum iser_data_dir iser_dir, enum dma_data_direction dma_dir); -void iser_dma_unmap_task_data(struct iscsi_iser_cmd_task *iser_ctask); +void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task, + struct iser_data_buf *data); +int iser_initialize_task_headers(struct iscsi_task *task, + struct iser_tx_desc *tx_desc); +int iser_alloc_rx_descriptors(struct iser_conn *ib_conn, struct iscsi_session *session); +int iser_create_fmr_pool(struct iser_conn *ib_conn, unsigned cmds_max); +void iser_free_fmr_pool(struct iser_conn *ib_conn); +int iser_create_fastreg_pool(struct iser_conn *ib_conn, unsigned cmds_max); +void iser_free_fastreg_pool(struct iser_conn *ib_conn); +u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir, sector_t *sector); #endif diff --git a/drivers/infiniband/ulp/iser/iser_initiator.c b/drivers/infiniband/ulp/iser/iser_initiator.c index 83247f1fdf7..8d44a406063 100644 --- a/drivers/infiniband/ulp/iser/iser_initiator.c +++ b/drivers/infiniband/ulp/iser/iser_initiator.c @@ -1,5 +1,6 @@ /* * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved. + * Copyright (c) 2013-2014 Mellanox Technologies. 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 @@ -28,8 +29,6 @@ * 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> @@ -41,115 +40,101 @@ #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++; -} - /* Register user buffer memory and initialize passive rdma - * dto descriptor. Total data size is stored in - * iser_ctask->data[ISER_DIR_IN].data_len + * dto descriptor. Data size is stored in + * task->data[ISER_DIR_IN].data_len, Protection size + * os stored in task->prot[ISER_DIR_IN].data_len */ -static int iser_prepare_read_cmd(struct iscsi_cmd_task *ctask, - unsigned int edtl) +static int iser_prepare_read_cmd(struct iscsi_task *task) { - struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + struct iscsi_iser_task *iser_task = task->dd_data; + struct iser_device *device = iser_task->ib_conn->device; 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]; + struct iser_hdr *hdr = &iser_task->desc.iser_header; + struct iser_data_buf *buf_in = &iser_task->data[ISER_DIR_IN]; - err = iser_dma_map_task_data(iser_ctask, + err = iser_dma_map_task_data(iser_task, 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; + if (scsi_prot_sg_count(iser_task->sc)) { + struct iser_data_buf *pbuf_in = &iser_task->prot[ISER_DIR_IN]; + + err = iser_dma_map_task_data(iser_task, + pbuf_in, + ISER_DIR_IN, + DMA_FROM_DEVICE); + if (err) + return err; } - err = iser_reg_rdma_mem(iser_ctask,ISER_DIR_IN); + err = device->iser_reg_rdma_mem(iser_task, 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]; + regd_buf = &iser_task->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, + task->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 + * dto descriptor. Data size is stored in + * task->data[ISER_DIR_OUT].data_len, Protection size + * is stored at task->prot[ISER_DIR_OUT].data_len */ static int -iser_prepare_write_cmd(struct iscsi_cmd_task *ctask, +iser_prepare_write_cmd(struct iscsi_task *task, unsigned int imm_sz, unsigned int unsol_sz, unsigned int edtl) { - struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data; + struct iscsi_iser_task *iser_task = task->dd_data; + struct iser_device *device = iser_task->ib_conn->device; 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]; + struct iser_hdr *hdr = &iser_task->desc.iser_header; + struct iser_data_buf *buf_out = &iser_task->data[ISER_DIR_OUT]; + struct ib_sge *tx_dsg = &iser_task->desc.tx_sg[1]; - err = iser_dma_map_task_data(iser_ctask, + err = iser_dma_map_task_data(iser_task, 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; + if (scsi_prot_sg_count(iser_task->sc)) { + struct iser_data_buf *pbuf_out = &iser_task->prot[ISER_DIR_OUT]; + + err = iser_dma_map_task_data(iser_task, + pbuf_out, + ISER_DIR_OUT, + DMA_TO_DEVICE); + if (err) + return err; } - err = iser_reg_rdma_mem(iser_ctask,ISER_DIR_OUT); + err = device->iser_reg_rdma_mem(iser_task, 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]; + regd_buf = &iser_task->rdma_regd[ISER_DIR_OUT]; if (unsol_sz < edtl) { hdr->flags |= ISER_WSV; @@ -158,299 +143,344 @@ iser_prepare_write_cmd(struct iscsi_cmd_task *ctask, iser_dbg("Cmd itt:%d, WRITE tags, RKEY:%#.4X " "VA:%#llX + unsol:%d\n", - ctask->itt, regd_buf->reg.rkey, + task->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); + task->itt, imm_sz); + tx_dsg->addr = regd_buf->reg.va; + tx_dsg->length = imm_sz; + tx_dsg->lkey = regd_buf->reg.lkey; + iser_task->desc.num_sge = 2; } return 0; } -/** - * iser_post_receive_control - allocates, initializes and posts receive DTO. - */ -static int iser_post_receive_control(struct iscsi_conn *conn) +/* creates a new tx descriptor and adds header regd buffer */ +static void iser_create_send_desc(struct iser_conn *ib_conn, + struct iser_tx_desc *tx_desc) { - 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; + struct iser_device *device = ib_conn->device; - /* 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 = ISCSI_DEF_MAX_RECV_SEG_LEN; - 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; - } + ib_dma_sync_single_for_cpu(device->ib_device, + tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE); - recv_dto = &rx_desc->dto; - recv_dto->ib_conn = iser_conn->ib_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; + memset(&tx_desc->iser_header, 0, sizeof(struct iser_hdr)); + tx_desc->iser_header.flags = ISER_VER; - iser_reg_single(device, regd_hdr, DMA_FROM_DEVICE); + tx_desc->num_sge = 1; - iser_dto_add_regd_buff(recv_dto, regd_hdr, 0, 0); + if (tx_desc->tx_sg[0].lkey != device->mr->lkey) { + tx_desc->tx_sg[0].lkey = device->mr->lkey; + iser_dbg("sdesc %p lkey mismatch, fixing\n", tx_desc); + } +} - 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; +static void iser_free_login_buf(struct iser_conn *ib_conn) +{ + if (!ib_conn->login_buf) + return; - iser_reg_single(device, regd_data, DMA_FROM_DEVICE); + if (ib_conn->login_req_dma) + ib_dma_unmap_single(ib_conn->device->ib_device, + ib_conn->login_req_dma, + ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE); - iser_dto_add_regd_buff(recv_dto, regd_data, 0, 0); + if (ib_conn->login_resp_dma) + ib_dma_unmap_single(ib_conn->device->ib_device, + ib_conn->login_resp_dma, + ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE); - err = iser_post_recv(rx_desc); - if (!err) - return 0; + kfree(ib_conn->login_buf); - /* 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; + /* make sure we never redo any unmapping */ + ib_conn->login_req_dma = 0; + ib_conn->login_resp_dma = 0; + ib_conn->login_buf = NULL; } -/* 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) +static int iser_alloc_login_buf(struct iser_conn *ib_conn) { - struct iser_regd_buf *regd_hdr = &tx_desc->hdr_regd_buf; - struct iser_dto *send_dto = &tx_desc->dto; + struct iser_device *device; + int req_err, resp_err; - 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; + BUG_ON(ib_conn->device == NULL); - send_dto->ib_conn = iser_conn->ib_conn; - send_dto->notify_enable = 1; - send_dto->regd_vector_len = 0; + device = ib_conn->device; - memset(&tx_desc->iser_header, 0, sizeof(struct iser_hdr)); - tx_desc->iser_header.flags = ISER_VER; + ib_conn->login_buf = kmalloc(ISCSI_DEF_MAX_RECV_SEG_LEN + + ISER_RX_LOGIN_SIZE, GFP_KERNEL); + if (!ib_conn->login_buf) + goto out_err; + + ib_conn->login_req_buf = ib_conn->login_buf; + ib_conn->login_resp_buf = ib_conn->login_buf + + ISCSI_DEF_MAX_RECV_SEG_LEN; + + ib_conn->login_req_dma = ib_dma_map_single(ib_conn->device->ib_device, + (void *)ib_conn->login_req_buf, + ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE); - iser_dto_add_regd_buff(send_dto, regd_hdr, 0, 0); + ib_conn->login_resp_dma = ib_dma_map_single(ib_conn->device->ib_device, + (void *)ib_conn->login_resp_buf, + ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE); + + req_err = ib_dma_mapping_error(device->ib_device, + ib_conn->login_req_dma); + resp_err = ib_dma_mapping_error(device->ib_device, + ib_conn->login_resp_dma); + + if (req_err || resp_err) { + if (req_err) + ib_conn->login_req_dma = 0; + if (resp_err) + ib_conn->login_resp_dma = 0; + goto free_login_buf; + } + return 0; + +free_login_buf: + iser_free_login_buf(ib_conn); + +out_err: + iser_err("unable to alloc or map login buf\n"); + return -ENOMEM; } -/** - * iser_conn_set_full_featured_mode - (iSER API) - */ -int iser_conn_set_full_featured_mode(struct iscsi_conn *conn) +int iser_alloc_rx_descriptors(struct iser_conn *ib_conn, struct iscsi_session *session) { - struct iscsi_iser_conn *iser_conn = conn->dd_data; + int i, j; + u64 dma_addr; + struct iser_rx_desc *rx_desc; + struct ib_sge *rx_sg; + struct iser_device *device = ib_conn->device; - 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; + ib_conn->qp_max_recv_dtos = session->cmds_max; + ib_conn->qp_max_recv_dtos_mask = session->cmds_max - 1; /* cmds_max is 2^N */ + ib_conn->min_posted_rx = ib_conn->qp_max_recv_dtos >> 2; - iser_dbg("Initially post: %d\n", initial_post_recv_bufs_num); + if (device->iser_alloc_rdma_reg_res(ib_conn, session->scsi_cmds_max)) + goto create_rdma_reg_res_failed; - /* 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); + if (iser_alloc_login_buf(ib_conn)) + goto alloc_login_buf_fail; - /* 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; - } + ib_conn->rx_descs = kmalloc(session->cmds_max * + sizeof(struct iser_rx_desc), GFP_KERNEL); + if (!ib_conn->rx_descs) + goto rx_desc_alloc_fail; + + rx_desc = ib_conn->rx_descs; + + for (i = 0; i < ib_conn->qp_max_recv_dtos; i++, rx_desc++) { + dma_addr = ib_dma_map_single(device->ib_device, (void *)rx_desc, + ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE); + if (ib_dma_mapping_error(device->ib_device, dma_addr)) + goto rx_desc_dma_map_failed; + + rx_desc->dma_addr = dma_addr; + + rx_sg = &rx_desc->rx_sg; + rx_sg->addr = rx_desc->dma_addr; + rx_sg->length = ISER_RX_PAYLOAD_SIZE; + rx_sg->lkey = device->mr->lkey; } - iser_dbg("Posted %d post recv bufs, conn:0x%p\n", i, conn); + + ib_conn->rx_desc_head = 0; return 0; + +rx_desc_dma_map_failed: + rx_desc = ib_conn->rx_descs; + for (j = 0; j < i; j++, rx_desc++) + ib_dma_unmap_single(device->ib_device, rx_desc->dma_addr, + ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE); + kfree(ib_conn->rx_descs); + ib_conn->rx_descs = NULL; +rx_desc_alloc_fail: + iser_free_login_buf(ib_conn); +alloc_login_buf_fail: + device->iser_free_rdma_reg_res(ib_conn); +create_rdma_reg_res_failed: + iser_err("failed allocating rx descriptors / data buffers\n"); + return -ENOMEM; } -static int -iser_check_xmit(struct iscsi_conn *conn, void *task) +void iser_free_rx_descriptors(struct iser_conn *ib_conn) { - struct iscsi_iser_conn *iser_conn = conn->dd_data; + int i; + struct iser_rx_desc *rx_desc; + struct iser_device *device = ib_conn->device; - if (atomic_read(&iser_conn->ib_conn->post_send_buf_count) == - ISER_QP_MAX_REQ_DTOS) { - iser_dbg("%ld can't xmit task %p\n",jiffies,task); - return -ENOBUFS; - } - return 0; + if (!ib_conn->rx_descs) + goto free_login_buf; + + if (device->iser_free_rdma_reg_res) + device->iser_free_rdma_reg_res(ib_conn); + + rx_desc = ib_conn->rx_descs; + for (i = 0; i < ib_conn->qp_max_recv_dtos; i++, rx_desc++) + ib_dma_unmap_single(device->ib_device, rx_desc->dma_addr, + ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE); + kfree(ib_conn->rx_descs); + /* make sure we never redo any unmapping */ + ib_conn->rx_descs = NULL; + +free_login_buf: + iser_free_login_buf(ib_conn); } +static int iser_post_rx_bufs(struct iscsi_conn *conn, struct iscsi_hdr *req) +{ + struct iser_conn *ib_conn = conn->dd_data; + struct iscsi_session *session = conn->session; + + iser_dbg("req op %x flags %x\n", req->opcode, req->flags); + /* check if this is the last login - going to full feature phase */ + if ((req->flags & ISCSI_FULL_FEATURE_PHASE) != ISCSI_FULL_FEATURE_PHASE) + return 0; + + /* + * Check that there is one posted recv buffer (for the last login + * response) and no posted send buffers left - they must have been + * consumed during previous login phases. + */ + WARN_ON(ib_conn->post_recv_buf_count != 1); + WARN_ON(atomic_read(&ib_conn->post_send_buf_count) != 0); + + if (session->discovery_sess) { + iser_info("Discovery session, re-using login RX buffer\n"); + return 0; + } else + iser_info("Normal session, posting batch of RX %d buffers\n", + ib_conn->min_posted_rx); + + /* Initial post receive buffers */ + if (iser_post_recvm(ib_conn, ib_conn->min_posted_rx)) + return -ENOMEM; + + return 0; +} /** * iser_send_command - send command PDU */ -int iser_send_command(struct iscsi_conn *conn, - struct iscsi_cmd_task *ctask) +int iser_send_command(struct iscsi_conn *conn, + struct iscsi_task *task) { - 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; + struct iser_conn *ib_conn = conn->dd_data; + struct iscsi_iser_task *iser_task = task->dd_data; 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 -ENOBUFS; + int err; + struct iser_data_buf *data_buf, *prot_buf; + struct iscsi_scsi_req *hdr = (struct iscsi_scsi_req *)task->hdr; + struct scsi_cmnd *sc = task->sc; + struct iser_tx_desc *tx_desc = &iser_task->desc; 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); + tx_desc->type = ISCSI_TX_SCSI_COMMAND; + iser_create_send_desc(ib_conn, tx_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 (hdr->flags & ISCSI_FLAG_CMD_READ) { + data_buf = &iser_task->data[ISER_DIR_IN]; + prot_buf = &iser_task->prot[ISER_DIR_IN]; + } else { + data_buf = &iser_task->data[ISER_DIR_OUT]; + prot_buf = &iser_task->prot[ISER_DIR_OUT]; + } if (scsi_sg_count(sc)) { /* using a scatter list */ data_buf->buf = scsi_sglist(sc); data_buf->size = scsi_sg_count(sc); } - data_buf->data_len = scsi_bufflen(sc); + if (scsi_prot_sg_count(sc)) { + prot_buf->buf = scsi_prot_sglist(sc); + prot_buf->size = scsi_prot_sg_count(sc); + prot_buf->data_len = data_buf->data_len >> + ilog2(sc->device->sector_size) * 8; + } + if (hdr->flags & ISCSI_FLAG_CMD_READ) { - err = iser_prepare_read_cmd(ctask, edtl); + err = iser_prepare_read_cmd(task); 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, + err = iser_prepare_write_cmd(task, + task->imm_count, + task->imm_count + + task->unsol_r2t.data_length, 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; + iser_task->status = ISER_TASK_STATUS_STARTED; - err = iser_post_send(&iser_ctask->desc); + err = iser_post_send(ib_conn, tx_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); + iser_err("conn %p failed task->itt %d err %d\n",conn, task->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, +int iser_send_data_out(struct iscsi_conn *conn, + struct iscsi_task *task, 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; + struct iser_conn *ib_conn = conn->dd_data; + struct iscsi_iser_task *iser_task = task->dd_data; + struct iser_tx_desc *tx_desc = NULL; + struct iser_regd_buf *regd_buf; unsigned long buf_offset; unsigned long data_seg_len; - unsigned int itt; + uint32_t itt; int err = 0; + struct ib_sge *tx_dsg; - 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 -ENOBUFS; - - itt = ntohl(hdr->itt); + itt = (__force uint32_t)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); + tx_desc = kmem_cache_zalloc(ig.desc_cache, GFP_ATOMIC); if (tx_desc == NULL) { iser_err("Failed to alloc desc for post dataout\n"); return -ENOMEM; } tx_desc->type = ISCSI_TX_DATAOUT; + tx_desc->iser_header.flags = ISER_VER; 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); + /* build the tx desc */ + iser_initialize_task_headers(task, tx_desc); - /* 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); + regd_buf = &iser_task->rdma_regd[ISER_DIR_OUT]; + tx_dsg = &tx_desc->tx_sg[1]; + tx_dsg->addr = regd_buf->reg.va + buf_offset; + tx_dsg->length = data_seg_len; + tx_dsg->lkey = regd_buf->reg.lkey; + tx_desc->num_sge = 2; - if (buf_offset + data_seg_len > iser_ctask->data[ISER_DIR_OUT].data_len) { + if (buf_offset + data_seg_len > iser_task->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); + iser_task->data[ISER_DIR_OUT].data_len, itt); err = -EINVAL; goto send_data_out_error; } @@ -458,73 +488,73 @@ int iser_send_data_out(struct iscsi_conn *conn, itt, buf_offset, data_seg_len); - err = iser_post_send(tx_desc); + err = iser_post_send(ib_conn, 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_task *task) { - struct iscsi_iser_conn *iser_conn = conn->dd_data; - struct iser_desc *mdesc = mtask->dd_data; - struct iser_dto *send_dto = NULL; + struct iser_conn *ib_conn = conn->dd_data; + struct iscsi_iser_task *iser_task = task->dd_data; + struct iser_tx_desc *mdesc = &iser_task->desc; unsigned long data_seg_len; int err = 0; - 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 -ENOBUFS; - /* 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_create_send_desc(ib_conn, mdesc); - iser_reg_single(device, send_dto->regd[0], DMA_TO_DEVICE); + device = ib_conn->device; - data_seg_len = ntoh24(mtask->hdr->dlength); + data_seg_len = ntoh24(task->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); + struct ib_sge *tx_dsg = &mdesc->tx_sg[1]; + if (task != conn->login_task) { + iser_err("data present on non login task!!!\n"); + goto send_control_error; + } + + ib_dma_sync_single_for_cpu(device->ib_device, + ib_conn->login_req_dma, task->data_count, + DMA_TO_DEVICE); + + memcpy(ib_conn->login_req_buf, task->data, task->data_count); + + ib_dma_sync_single_for_device(device->ib_device, + ib_conn->login_req_dma, task->data_count, + DMA_TO_DEVICE); + + tx_dsg->addr = ib_conn->login_req_dma; + tx_dsg->length = task->data_count; + tx_dsg->lkey = device->mr->lkey; + mdesc->num_sge = 2; } - if (iser_post_receive_control(conn) != 0) { - iser_err("post_rcv_buff failed!\n"); - err = -ENOMEM; - goto send_control_error; + if (task == conn->login_task) { + iser_dbg("op %x dsl %lx, posting login rx buffer\n", + task->hdr->opcode, data_seg_len); + err = iser_post_recvl(ib_conn); + if (err) + goto send_control_error; + err = iser_post_rx_bufs(conn, task->hdr); + if (err) + goto send_control_error; } - err = iser_post_send(mdesc); + err = iser_post_send(ib_conn, 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; } @@ -532,164 +562,159 @@ send_control_error: /** * iser_rcv_dto_completion - recv DTO completion */ -void iser_rcv_completion(struct iser_desc *rx_desc, - unsigned long dto_xfer_len) +void iser_rcv_completion(struct iser_rx_desc *rx_desc, + unsigned long rx_xfer_len, + struct iser_conn *ib_conn) { - struct iser_dto *dto = &rx_desc->dto; - struct iscsi_iser_conn *conn = dto->ib_conn->iser_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]; + u64 rx_dma; + int rx_buflen, outstanding, count, err; + + /* differentiate between login to all other PDUs */ + if ((char *)rx_desc == ib_conn->login_resp_buf) { + rx_dma = ib_conn->login_resp_dma; + rx_buflen = ISER_RX_LOGIN_SIZE; + } else { + rx_dma = rx_desc->dma_addr; + rx_buflen = ISER_RX_PAYLOAD_SIZE; } - opcode = hdr->opcode & ISCSI_OPCODE_MASK; - - if (opcode == ISCSI_OP_SCSI_CMD_RSP) { - itt = get_itt(hdr->itt); /* 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); - } + ib_dma_sync_single_for_cpu(ib_conn->device->ib_device, rx_dma, + rx_buflen, DMA_FROM_DEVICE); - iser_dto_buffs_release(dto); + hdr = &rx_desc->iscsi_header; + + iser_dbg("op 0x%x itt 0x%x dlen %d\n", hdr->opcode, + hdr->itt, (int)(rx_xfer_len - ISER_HEADERS_LEN)); - iscsi_iser_recv(conn->iscsi_conn, hdr, rx_data, rx_data_len); + iscsi_iser_recv(ib_conn->iscsi_conn, hdr, rx_desc->data, + rx_xfer_len - ISER_HEADERS_LEN); - kfree(rx_desc->data); - kmem_cache_free(ig.desc_cache, rx_desc); + ib_dma_sync_single_for_device(ib_conn->device->ib_device, rx_dma, + rx_buflen, DMA_FROM_DEVICE); /* 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); -} + ib_conn->post_recv_buf_count--; -void iser_snd_completion(struct iser_desc *tx_desc) -{ - struct iser_dto *dto = &tx_desc->dto; - struct iser_conn *ib_conn = dto->ib_conn; - struct iscsi_iser_conn *iser_conn = ib_conn->iser_conn; - struct iscsi_conn *conn = iser_conn->iscsi_conn; - struct iscsi_mgmt_task *mtask; - int resume_tx = 0; + if (rx_dma == ib_conn->login_resp_dma) + return; - iser_dbg("Initiator, Data sent dto=0x%p\n", dto); + outstanding = ib_conn->post_recv_buf_count; + if (outstanding + ib_conn->min_posted_rx <= ib_conn->qp_max_recv_dtos) { + count = min(ib_conn->qp_max_recv_dtos - outstanding, + ib_conn->min_posted_rx); + err = iser_post_recvm(ib_conn, count); + if (err) + iser_err("posting %d rx bufs err %d\n", count, err); + } +} - iser_dto_buffs_release(dto); +void iser_snd_completion(struct iser_tx_desc *tx_desc, + struct iser_conn *ib_conn) +{ + struct iscsi_task *task; + struct iser_device *device = ib_conn->device; - if (tx_desc->type == ISCSI_TX_DATAOUT) + if (tx_desc->type == ISCSI_TX_DATAOUT) { + ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr, + ISER_HEADERS_LEN, DMA_TO_DEVICE); kmem_cache_free(ig.desc_cache, tx_desc); - - if (atomic_read(&iser_conn->ib_conn->post_send_buf_count) == - ISER_QP_MAX_REQ_DTOS) - resume_tx = 1; + tx_desc = NULL; + } atomic_dec(&ib_conn->post_send_buf_count); - if (resume_tx) { - iser_dbg("%ld resuming tx\n",jiffies); - scsi_queue_work(conn->session->host, &conn->xmitwork); - } - - if (tx_desc->type == ISCSI_TX_CONTROL) { + if (tx_desc && 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 == RESERVED_ITT) { - struct iscsi_session *session = conn->session; - - spin_lock(&conn->session->lock); - iscsi_free_mgmt_task(conn, mtask); - spin_unlock(&session->lock); - } + task = (void *) ((long)(void *)tx_desc - + sizeof(struct iscsi_task)); + if (task->hdr->itt == RESERVED_ITT) + iscsi_put_task(task); } } -void iser_ctask_rdma_init(struct iscsi_iser_cmd_task *iser_ctask) +void iser_task_rdma_init(struct iscsi_iser_task *iser_task) { - iser_ctask->status = ISER_TASK_STATUS_INIT; + iser_task->status = ISER_TASK_STATUS_INIT; + + iser_task->dir[ISER_DIR_IN] = 0; + iser_task->dir[ISER_DIR_OUT] = 0; - iser_ctask->dir[ISER_DIR_IN] = 0; - iser_ctask->dir[ISER_DIR_OUT] = 0; + iser_task->data[ISER_DIR_IN].data_len = 0; + iser_task->data[ISER_DIR_OUT].data_len = 0; - iser_ctask->data[ISER_DIR_IN].data_len = 0; - iser_ctask->data[ISER_DIR_OUT].data_len = 0; + iser_task->prot[ISER_DIR_IN].data_len = 0; + iser_task->prot[ISER_DIR_OUT].data_len = 0; - memset(&iser_ctask->rdma_regd[ISER_DIR_IN], 0, + memset(&iser_task->rdma_regd[ISER_DIR_IN], 0, sizeof(struct iser_regd_buf)); - memset(&iser_ctask->rdma_regd[ISER_DIR_OUT], 0, + memset(&iser_task->rdma_regd[ISER_DIR_OUT], 0, sizeof(struct iser_regd_buf)); } -void iser_ctask_rdma_finalize(struct iscsi_iser_cmd_task *iser_ctask) +void iser_task_rdma_finalize(struct iscsi_iser_task *iser_task) { - int deferred; - int is_rdma_aligned = 1; - struct iser_regd_buf *regd; + struct iser_device *device = iser_task->ib_conn->device; + int is_rdma_data_aligned = 1; + int is_rdma_prot_aligned = 1; + int prot_count = scsi_prot_sg_count(iser_task->sc); /* 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) { - is_rdma_aligned = 0; - iser_finalize_rdma_unaligned_sg(iser_ctask, ISER_DIR_IN); - } - if (iser_ctask->data_copy[ISER_DIR_OUT].copy_buf != NULL) { - is_rdma_aligned = 0; - iser_finalize_rdma_unaligned_sg(iser_ctask, ISER_DIR_OUT); + if (iser_task->data_copy[ISER_DIR_IN].copy_buf != NULL) { + is_rdma_data_aligned = 0; + iser_finalize_rdma_unaligned_sg(iser_task, + &iser_task->data[ISER_DIR_IN], + &iser_task->data_copy[ISER_DIR_IN], + ISER_DIR_IN); } - if (iser_ctask->dir[ISER_DIR_IN]) { - regd = &iser_ctask->rdma_regd[ISER_DIR_IN]; - deferred = iser_regd_buff_release(regd); - if (deferred) { - iser_err("%d references remain for BUF-IN rdma reg\n", - atomic_read(®d->ref_count)); - } + if (iser_task->data_copy[ISER_DIR_OUT].copy_buf != NULL) { + is_rdma_data_aligned = 0; + iser_finalize_rdma_unaligned_sg(iser_task, + &iser_task->data[ISER_DIR_OUT], + &iser_task->data_copy[ISER_DIR_OUT], + ISER_DIR_OUT); } - if (iser_ctask->dir[ISER_DIR_OUT]) { - regd = &iser_ctask->rdma_regd[ISER_DIR_OUT]; - deferred = iser_regd_buff_release(regd); - if (deferred) { - iser_err("%d references remain for BUF-OUT rdma reg\n", - atomic_read(®d->ref_count)); - } + if (iser_task->prot_copy[ISER_DIR_IN].copy_buf != NULL) { + is_rdma_prot_aligned = 0; + iser_finalize_rdma_unaligned_sg(iser_task, + &iser_task->prot[ISER_DIR_IN], + &iser_task->prot_copy[ISER_DIR_IN], + ISER_DIR_IN); } - /* if the data was unaligned, it was already unmapped and then copied */ - if (is_rdma_aligned) - iser_dma_unmap_task_data(iser_ctask); -} + if (iser_task->prot_copy[ISER_DIR_OUT].copy_buf != NULL) { + is_rdma_prot_aligned = 0; + iser_finalize_rdma_unaligned_sg(iser_task, + &iser_task->prot[ISER_DIR_OUT], + &iser_task->prot_copy[ISER_DIR_OUT], + ISER_DIR_OUT); + } -void iser_dto_buffs_release(struct iser_dto *dto) -{ - int i; + if (iser_task->dir[ISER_DIR_IN]) { + device->iser_unreg_rdma_mem(iser_task, ISER_DIR_IN); + if (is_rdma_data_aligned) + iser_dma_unmap_task_data(iser_task, + &iser_task->data[ISER_DIR_IN]); + if (prot_count && is_rdma_prot_aligned) + iser_dma_unmap_task_data(iser_task, + &iser_task->prot[ISER_DIR_IN]); + } - for (i = 0; i < dto->regd_vector_len; i++) - iser_regd_buff_release(dto->regd[i]); + if (iser_task->dir[ISER_DIR_OUT]) { + device->iser_unreg_rdma_mem(iser_task, ISER_DIR_OUT); + if (is_rdma_data_aligned) + iser_dma_unmap_task_data(iser_task, + &iser_task->data[ISER_DIR_OUT]); + if (prot_count && is_rdma_prot_aligned) + iser_dma_unmap_task_data(iser_task, + &iser_task->prot[ISER_DIR_OUT]); + } } - diff --git a/drivers/infiniband/ulp/iser/iser_memory.c b/drivers/infiniband/ulp/iser/iser_memory.c index 4a17743a639..47acd3ad3a1 100644 --- a/drivers/infiniband/ulp/iser/iser_memory.c +++ b/drivers/infiniband/ulp/iser/iser_memory.c @@ -1,5 +1,6 @@ /* * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved. + * Copyright (c) 2013-2014 Mellanox Technologies. 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 @@ -28,8 +29,6 @@ * 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> @@ -43,136 +42,83 @@ #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 ib_device *dev; - - 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.is_fmr) - iser_unreg_mem(®d_buf->reg); - - if (regd_buf->dma_addr) { - dev = regd_buf->device->ib_device; - ib_dma_unmap_single(dev, - 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) -{ - u64 dma_addr; - - dma_addr = ib_dma_map_single(device->ib_device, - regd_buf->virt_addr, - regd_buf->data_size, direction); - BUG_ON(ib_dma_mapping_error(device->ib_device, dma_addr)); - - regd_buf->reg.lkey = device->mr->lkey; - regd_buf->reg.len = regd_buf->data_size; - regd_buf->reg.va = dma_addr; - regd_buf->reg.is_fmr = 0; - - regd_buf->dma_addr = dma_addr; - regd_buf->direction = direction; -} - -/** * iser_start_rdma_unaligned_sg */ -static int iser_start_rdma_unaligned_sg(struct iscsi_iser_cmd_task *iser_ctask, +static int iser_start_rdma_unaligned_sg(struct iscsi_iser_task *iser_task, + struct iser_data_buf *data, + struct iser_data_buf *data_copy, enum iser_data_dir cmd_dir) { - int dma_nents; - struct ib_device *dev; + struct ib_device *dev = iser_task->ib_conn->device->ib_device; + struct scatterlist *sgl = (struct scatterlist *)data->buf; + struct scatterlist *sg; char *mem = NULL; - struct iser_data_buf *data = &iser_ctask->data[cmd_dir]; - unsigned long cmd_data_len = data->data_len; + unsigned long cmd_data_len = 0; + int dma_nents, i; + + for_each_sg(sgl, sg, data->size, i) + cmd_data_len += ib_sg_dma_len(dev, sg); if (cmd_data_len > ISER_KMALLOC_THRESHOLD) - mem = (void *)__get_free_pages(GFP_NOIO, + mem = (void *)__get_free_pages(GFP_ATOMIC, ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT); else - mem = kmalloc(cmd_data_len, GFP_NOIO); + mem = kmalloc(cmd_data_len, GFP_ATOMIC); if (mem == NULL) { iser_err("Failed to allocate mem size %d %d for copying sglist\n", - data->size,(int)cmd_data_len); + 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 *sgl = (struct scatterlist *)data->buf; - struct scatterlist *sg; int i; char *p, *from; + sgl = (struct scatterlist *)data->buf; p = mem; for_each_sg(sgl, sg, data->size, i) { - from = kmap_atomic(sg_page(sg), KM_USER0); + from = kmap_atomic(sg_page(sg)); memcpy(p, from + sg->offset, sg->length); - kunmap_atomic(from, KM_USER0); + kunmap_atomic(from); p += sg->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; + sg_init_one(&data_copy->sg_single, mem, cmd_data_len); + data_copy->buf = &data_copy->sg_single; + data_copy->size = 1; + data_copy->copy_buf = mem; - dev = iser_ctask->iser_conn->ib_conn->device->ib_device; - dma_nents = ib_dma_map_sg(dev, - &iser_ctask->data_copy[cmd_dir].sg_single, - 1, + dma_nents = ib_dma_map_sg(dev, &data_copy->sg_single, 1, (cmd_dir == ISER_DIR_OUT) ? DMA_TO_DEVICE : DMA_FROM_DEVICE); BUG_ON(dma_nents == 0); - iser_ctask->data_copy[cmd_dir].dma_nents = dma_nents; + data_copy->dma_nents = dma_nents; + data_copy->data_len = cmd_data_len; + 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) + +void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *iser_task, + struct iser_data_buf *data, + struct iser_data_buf *data_copy, + enum iser_data_dir cmd_dir) { struct ib_device *dev; - struct iser_data_buf *mem_copy; unsigned long cmd_data_len; - dev = iser_ctask->iser_conn->ib_conn->device->ib_device; - mem_copy = &iser_ctask->data_copy[cmd_dir]; + dev = iser_task->ib_conn->device->ib_device; - ib_dma_unmap_sg(dev, &mem_copy->sg_single, 1, + ib_dma_unmap_sg(dev, &data_copy->sg_single, 1, (cmd_dir == ISER_DIR_OUT) ? DMA_TO_DEVICE : DMA_FROM_DEVICE); @@ -184,33 +130,35 @@ void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_cmd_task *iser_ctask, int i; /* copy back read RDMA to unaligned sg */ - mem = mem_copy->copy_buf; + mem = data_copy->copy_buf; - sgl = (struct scatterlist *)iser_ctask->data[ISER_DIR_IN].buf; - sg_size = iser_ctask->data[ISER_DIR_IN].size; + sgl = (struct scatterlist *)data->buf; + sg_size = data->size; p = mem; for_each_sg(sgl, sg, sg_size, i) { - to = kmap_atomic(sg_page(sg), KM_SOFTIRQ0); + to = kmap_atomic(sg_page(sg)); memcpy(to + sg->offset, p, sg->length); - kunmap_atomic(to, KM_SOFTIRQ0); + kunmap_atomic(to); p += sg->length; } } - cmd_data_len = iser_ctask->data[cmd_dir].data_len; + cmd_data_len = data->data_len; if (cmd_data_len > ISER_KMALLOC_THRESHOLD) - free_pages((unsigned long)mem_copy->copy_buf, + free_pages((unsigned long)data_copy->copy_buf, ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT); else - kfree(mem_copy->copy_buf); + kfree(data_copy->copy_buf); - mem_copy->copy_buf = NULL; + data_copy->copy_buf = NULL; } +#define IS_4K_ALIGNED(addr) ((((unsigned long)addr) & ~MASK_4K) == 0) + /** * iser_sg_to_page_vec - Translates scatterlist entries to physical addresses * and returns the length of resulting physical address array (may be less than @@ -223,62 +171,53 @@ void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_cmd_task *iser_ctask, * 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 ib_device *ibdev) + struct ib_device *ibdev, u64 *pages, + int *offset, int *data_size) { - struct scatterlist *sgl = (struct scatterlist *)data->buf; - struct scatterlist *sg; - u64 first_addr, last_addr, page; - int end_aligned; - unsigned int cur_page = 0; + struct scatterlist *sg, *sgl = (struct scatterlist *)data->buf; + u64 start_addr, end_addr, page, chunk_start = 0; unsigned long total_sz = 0; - int i; + unsigned int dma_len; + int i, new_chunk, cur_page, last_ent = data->dma_nents - 1; /* compute the offset of first element */ - page_vec->offset = (u64) sgl[0].offset & ~MASK_4K; + *offset = (u64) sgl[0].offset & ~MASK_4K; + new_chunk = 1; + cur_page = 0; for_each_sg(sgl, sg, data->dma_nents, i) { - unsigned int dma_len = ib_sg_dma_len(ibdev, sg); - + start_addr = ib_sg_dma_address(ibdev, sg); + if (new_chunk) + chunk_start = start_addr; + dma_len = ib_sg_dma_len(ibdev, sg); + end_addr = start_addr + dma_len; total_sz += dma_len; - first_addr = ib_sg_dma_address(ibdev, sg); - last_addr = first_addr + dma_len; - - end_aligned = !(last_addr & ~MASK_4K); - - /* continue to collect page fragments till aligned or SG ends */ - while (!end_aligned && (i + 1 < data->dma_nents)) { - sg = sg_next(sg); - i++; - dma_len = ib_sg_dma_len(ibdev, sg); - total_sz += dma_len; - last_addr = ib_sg_dma_address(ibdev, sg) + dma_len; - end_aligned = !(last_addr & ~MASK_4K); + /* collect page fragments until aligned or end of SG list */ + if (!IS_4K_ALIGNED(end_addr) && i < last_ent) { + new_chunk = 0; + continue; } - - /* handle the 1st page in the 1st DMA element */ - if (cur_page == 0) { - page = first_addr & MASK_4K; - page_vec->pages[cur_page] = page; - cur_page++; + new_chunk = 1; + + /* address of the first page in the contiguous chunk; + masking relevant for the very first SG entry, + which might be unaligned */ + page = chunk_start & MASK_4K; + do { + pages[cur_page++] = page; page += SIZE_4K; - } else - page = first_addr; - - for (; page < last_addr; page += SIZE_4K) { - page_vec->pages[cur_page] = page; - cur_page++; - } - + } while (page < end_addr); } - page_vec->data_size = total_sz; - iser_dbg("page_vec->data_size:%d cur_page %d\n", page_vec->data_size,cur_page); + + *data_size = total_sz; + iser_dbg("page_vec->data_size:%d cur_page %d\n", + *data_size, cur_page); return cur_page; } -#define IS_4K_ALIGNED(addr) ((((unsigned long)addr) & ~MASK_4K) == 0) /** * iser_data_buf_aligned_len - Tries to determine the maximal correctly aligned @@ -286,42 +225,40 @@ static int iser_sg_to_page_vec(struct iser_data_buf *data, * 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 ib_device *ibdev) +static int iser_data_buf_aligned_len(struct iser_data_buf *data, + struct ib_device *ibdev) { - struct scatterlist *sgl, *sg; - u64 end_addr, next_addr; - int i, cnt; - unsigned int ret_len = 0; + struct scatterlist *sgl, *sg, *next_sg = NULL; + u64 start_addr, end_addr; + int i, ret_len, start_check = 0; + + if (data->dma_nents == 1) + return 1; sgl = (struct scatterlist *)data->buf; + start_addr = ib_sg_dma_address(ibdev, sgl); - cnt = 0; for_each_sg(sgl, sg, data->dma_nents, i) { - /* iser_dbg("Checking sg iobuf [%d]: phys=0x%08lX " - "offset: %ld sz: %ld\n", i, - (unsigned long)sg_phys(sg), - (unsigned long)sg->offset, - (unsigned long)sg->length); */ - end_addr = ib_sg_dma_address(ibdev, sg) + - ib_sg_dma_len(ibdev, sg); - /* iser_dbg("Checking sg iobuf end address " - "0x%08lX\n", end_addr); */ - if (i + 1 < data->dma_nents) { - next_addr = ib_sg_dma_address(ibdev, sg_next(sg)); - /* are i, i+1 fragments of the same page? */ - if (end_addr == next_addr) { - cnt++; - continue; - } else if (!IS_4K_ALIGNED(end_addr)) { - ret_len = cnt + 1; - break; - } - } - cnt++; + if (start_check && !IS_4K_ALIGNED(start_addr)) + break; + + next_sg = sg_next(sg); + if (!next_sg) + break; + + end_addr = start_addr + ib_sg_dma_len(ibdev, sg); + start_addr = ib_sg_dma_address(ibdev, next_sg); + + if (end_addr == start_addr) { + start_check = 0; + continue; + } else + start_check = 1; + + if (!IS_4K_ALIGNED(end_addr)) + break; } - if (i == data->dma_nents) - ret_len = cnt; /* loop ended */ + ret_len = (next_sg) ? i : i+1; iser_dbg("Found %d aligned entries out of %d in sg:0x%p\n", ret_len, data->dma_nents, data); return ret_len; @@ -335,7 +272,7 @@ static void iser_data_buf_dump(struct iser_data_buf *data, int i; for_each_sg(sgl, sg, data->dma_nents, i) - iser_err("sg[%d] dma_addr:0x%lX page:0x%p " + iser_dbg("sg[%d] dma_addr:0x%lX page:0x%p " "off:0x%x sz:0x%x dma_len:0x%x\n", i, (unsigned long)ib_sg_dma_address(ibdev, sg), sg_page(sg), sg->offset, @@ -362,8 +299,10 @@ static void iser_page_vec_build(struct iser_data_buf *data, 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, ibdev); - iser_dbg("sg len %d page_vec_len %d\n", data->dma_nents,page_vec_len); + page_vec_len = iser_sg_to_page_vec(data, ibdev, page_vec->pages, + &page_vec->offset, + &page_vec->data_size); + iser_dbg("sg len %d page_vec_len %d\n", data->dma_nents, page_vec_len); page_vec->length = page_vec_len; @@ -375,15 +314,15 @@ static void iser_page_vec_build(struct iser_data_buf *data, } } -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) +int iser_dma_map_task_data(struct iscsi_iser_task *iser_task, + struct iser_data_buf *data, + enum iser_data_dir iser_dir, + enum dma_data_direction dma_dir) { struct ib_device *dev; - iser_ctask->dir[iser_dir] = 1; - dev = iser_ctask->iser_conn->ib_conn->device->ib_device; + iser_task->dir[iser_dir] = 1; + dev = iser_task->ib_conn->device->ib_device; data->dma_nents = ib_dma_map_sg(dev, data->buf, data->size, dma_dir); if (data->dma_nents == 0) { @@ -393,59 +332,73 @@ int iser_dma_map_task_data(struct iscsi_iser_cmd_task *iser_ctask, return 0; } -void iser_dma_unmap_task_data(struct iscsi_iser_cmd_task *iser_ctask) +void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task, + struct iser_data_buf *data) { struct ib_device *dev; - struct iser_data_buf *data; - dev = iser_ctask->iser_conn->ib_conn->device->ib_device; + dev = iser_task->ib_conn->device->ib_device; + ib_dma_unmap_sg(dev, data->buf, data->size, DMA_FROM_DEVICE); +} - if (iser_ctask->dir[ISER_DIR_IN]) { - data = &iser_ctask->data[ISER_DIR_IN]; - ib_dma_unmap_sg(dev, data->buf, data->size, DMA_FROM_DEVICE); - } +static int fall_to_bounce_buf(struct iscsi_iser_task *iser_task, + struct ib_device *ibdev, + struct iser_data_buf *mem, + struct iser_data_buf *mem_copy, + enum iser_data_dir cmd_dir, + int aligned_len) +{ + struct iscsi_conn *iscsi_conn = iser_task->ib_conn->iscsi_conn; - if (iser_ctask->dir[ISER_DIR_OUT]) { - data = &iser_ctask->data[ISER_DIR_OUT]; - ib_dma_unmap_sg(dev, data->buf, data->size, DMA_TO_DEVICE); - } + iscsi_conn->fmr_unalign_cnt++; + iser_warn("rdma alignment violation (%d/%d aligned) or FMR not supported\n", + aligned_len, mem->size); + + if (iser_debug_level > 0) + iser_data_buf_dump(mem, ibdev); + + /* unmap the command data before accessing it */ + iser_dma_unmap_task_data(iser_task, mem); + + /* allocate copy buf, if we are writing, copy the */ + /* unaligned scatterlist, dma map the copy */ + if (iser_start_rdma_unaligned_sg(iser_task, mem, mem_copy, cmd_dir) != 0) + return -ENOMEM; + + return 0; } /** - * iser_reg_rdma_mem - Registers memory intended for RDMA, - * obtaining rkey and va + * iser_reg_rdma_mem_fmr - Registers memory intended for RDMA, + * using FMR (if possible) 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) +int iser_reg_rdma_mem_fmr(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir) { - struct iser_conn *ib_conn = iser_ctask->iser_conn->ib_conn; + struct iser_conn *ib_conn = iser_task->ib_conn; struct iser_device *device = ib_conn->device; struct ib_device *ibdev = device->ib_device; - struct iser_data_buf *mem = &iser_ctask->data[cmd_dir]; + struct iser_data_buf *mem = &iser_task->data[cmd_dir]; struct iser_regd_buf *regd_buf; int aligned_len; int err; int i; struct scatterlist *sg; - regd_buf = &iser_ctask->rdma_regd[cmd_dir]; + regd_buf = &iser_task->rdma_regd[cmd_dir]; aligned_len = iser_data_buf_aligned_len(mem, ibdev); if (aligned_len != mem->dma_nents) { - iser_err("rdma alignment violation %d/%d aligned\n", - aligned_len, mem->size); - iser_data_buf_dump(mem, ibdev); - - /* unmap the command data before accessing it */ - iser_dma_unmap_task_data(iser_ctask); - - /* 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]; + err = fall_to_bounce_buf(iser_task, ibdev, mem, + &iser_task->data_copy[cmd_dir], + cmd_dir, aligned_len); + if (err) { + iser_err("failed to allocate bounce buffer\n"); + return err; + } + mem = &iser_task->data_copy[cmd_dir]; } /* if there a single dma entry, FMR is not needed */ @@ -456,7 +409,7 @@ int iser_reg_rdma_mem(struct iscsi_iser_cmd_task *iser_ctask, regd_buf->reg.rkey = device->mr->rkey; regd_buf->reg.len = ib_sg_dma_len(ibdev, &sg[0]); regd_buf->reg.va = ib_sg_dma_address(ibdev, &sg[0]); - regd_buf->reg.is_fmr = 0; + regd_buf->reg.is_mr = 0; iser_dbg("PHYSICAL Mem.register: lkey: 0x%08X rkey: 0x%08X " "va: 0x%08lX sz: %ld]\n", @@ -465,24 +418,383 @@ int iser_reg_rdma_mem(struct iscsi_iser_cmd_task *iser_ctask, (unsigned long)regd_buf->reg.va, (unsigned long)regd_buf->reg.len); } else { /* use FMR for multiple dma entries */ - iser_page_vec_build(mem, ib_conn->page_vec, ibdev); - err = iser_reg_page_vec(ib_conn, ib_conn->page_vec, ®d_buf->reg); - if (err) { + iser_page_vec_build(mem, ib_conn->fmr.page_vec, ibdev); + err = iser_reg_page_vec(ib_conn, ib_conn->fmr.page_vec, + ®d_buf->reg); + if (err && err != -EAGAIN) { iser_data_buf_dump(mem, ibdev); - iser_err("mem->dma_nents = %d (dlength = 0x%x)\n", mem->dma_nents, - ntoh24(iser_ctask->desc.iscsi_header.dlength)); + iser_err("mem->dma_nents = %d (dlength = 0x%x)\n", + mem->dma_nents, + ntoh24(iser_task->desc.iscsi_header.dlength)); iser_err("page_vec: data_size = 0x%x, length = %d, offset = 0x%x\n", - ib_conn->page_vec->data_size, ib_conn->page_vec->length, - ib_conn->page_vec->offset); - for (i=0 ; i<ib_conn->page_vec->length ; i++) + ib_conn->fmr.page_vec->data_size, + ib_conn->fmr.page_vec->length, + ib_conn->fmr.page_vec->offset); + for (i = 0; i < ib_conn->fmr.page_vec->length; i++) iser_err("page_vec[%d] = 0x%llx\n", i, - (unsigned long long) ib_conn->page_vec->pages[i]); + (unsigned long long) ib_conn->fmr.page_vec->pages[i]); + } + if (err) + return err; + } + return 0; +} + +static inline enum ib_t10_dif_type +scsi2ib_prot_type(unsigned char prot_type) +{ + switch (prot_type) { + case SCSI_PROT_DIF_TYPE0: + return IB_T10DIF_NONE; + case SCSI_PROT_DIF_TYPE1: + return IB_T10DIF_TYPE1; + case SCSI_PROT_DIF_TYPE2: + return IB_T10DIF_TYPE2; + case SCSI_PROT_DIF_TYPE3: + return IB_T10DIF_TYPE3; + default: + return IB_T10DIF_NONE; + } +} + + +static int +iser_set_sig_attrs(struct scsi_cmnd *sc, struct ib_sig_attrs *sig_attrs) +{ + unsigned char scsi_ptype = scsi_get_prot_type(sc); + + sig_attrs->mem.sig_type = IB_SIG_TYPE_T10_DIF; + sig_attrs->wire.sig_type = IB_SIG_TYPE_T10_DIF; + sig_attrs->mem.sig.dif.pi_interval = sc->device->sector_size; + sig_attrs->wire.sig.dif.pi_interval = sc->device->sector_size; + + switch (scsi_get_prot_op(sc)) { + case SCSI_PROT_WRITE_INSERT: + case SCSI_PROT_READ_STRIP: + sig_attrs->mem.sig.dif.type = IB_T10DIF_NONE; + sig_attrs->wire.sig.dif.type = scsi2ib_prot_type(scsi_ptype); + sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC; + sig_attrs->wire.sig.dif.ref_tag = scsi_get_lba(sc) & + 0xffffffff; + break; + case SCSI_PROT_READ_INSERT: + case SCSI_PROT_WRITE_STRIP: + sig_attrs->mem.sig.dif.type = scsi2ib_prot_type(scsi_ptype); + sig_attrs->mem.sig.dif.bg_type = IB_T10DIF_CRC; + sig_attrs->mem.sig.dif.ref_tag = scsi_get_lba(sc) & + 0xffffffff; + sig_attrs->wire.sig.dif.type = IB_T10DIF_NONE; + break; + case SCSI_PROT_READ_PASS: + case SCSI_PROT_WRITE_PASS: + sig_attrs->mem.sig.dif.type = scsi2ib_prot_type(scsi_ptype); + sig_attrs->mem.sig.dif.bg_type = IB_T10DIF_CRC; + sig_attrs->mem.sig.dif.ref_tag = scsi_get_lba(sc) & + 0xffffffff; + sig_attrs->wire.sig.dif.type = scsi2ib_prot_type(scsi_ptype); + sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC; + sig_attrs->wire.sig.dif.ref_tag = scsi_get_lba(sc) & + 0xffffffff; + break; + default: + iser_err("Unsupported PI operation %d\n", + scsi_get_prot_op(sc)); + return -EINVAL; + } + return 0; +} + + +static int +iser_set_prot_checks(struct scsi_cmnd *sc, u8 *mask) +{ + switch (scsi_get_prot_type(sc)) { + case SCSI_PROT_DIF_TYPE0: + *mask = 0x0; + break; + case SCSI_PROT_DIF_TYPE1: + case SCSI_PROT_DIF_TYPE2: + *mask = ISER_CHECK_GUARD | ISER_CHECK_REFTAG; + break; + case SCSI_PROT_DIF_TYPE3: + *mask = ISER_CHECK_GUARD; + break; + default: + iser_err("Unsupported protection type %d\n", + scsi_get_prot_type(sc)); + return -EINVAL; + } + + return 0; +} + +static int +iser_reg_sig_mr(struct iscsi_iser_task *iser_task, + struct fast_reg_descriptor *desc, struct ib_sge *data_sge, + struct ib_sge *prot_sge, struct ib_sge *sig_sge) +{ + struct iser_conn *ib_conn = iser_task->ib_conn; + struct iser_pi_context *pi_ctx = desc->pi_ctx; + struct ib_send_wr sig_wr, inv_wr; + struct ib_send_wr *bad_wr, *wr = NULL; + struct ib_sig_attrs sig_attrs; + int ret; + u32 key; + + memset(&sig_attrs, 0, sizeof(sig_attrs)); + ret = iser_set_sig_attrs(iser_task->sc, &sig_attrs); + if (ret) + goto err; + + ret = iser_set_prot_checks(iser_task->sc, &sig_attrs.check_mask); + if (ret) + goto err; + + if (!(desc->reg_indicators & ISER_SIG_KEY_VALID)) { + memset(&inv_wr, 0, sizeof(inv_wr)); + inv_wr.opcode = IB_WR_LOCAL_INV; + inv_wr.wr_id = ISER_FASTREG_LI_WRID; + inv_wr.ex.invalidate_rkey = pi_ctx->sig_mr->rkey; + wr = &inv_wr; + /* Bump the key */ + key = (u8)(pi_ctx->sig_mr->rkey & 0x000000FF); + ib_update_fast_reg_key(pi_ctx->sig_mr, ++key); + } + + memset(&sig_wr, 0, sizeof(sig_wr)); + sig_wr.opcode = IB_WR_REG_SIG_MR; + sig_wr.wr_id = ISER_FASTREG_LI_WRID; + sig_wr.sg_list = data_sge; + sig_wr.num_sge = 1; + sig_wr.wr.sig_handover.sig_attrs = &sig_attrs; + sig_wr.wr.sig_handover.sig_mr = pi_ctx->sig_mr; + if (scsi_prot_sg_count(iser_task->sc)) + sig_wr.wr.sig_handover.prot = prot_sge; + sig_wr.wr.sig_handover.access_flags = IB_ACCESS_LOCAL_WRITE | + IB_ACCESS_REMOTE_READ | + IB_ACCESS_REMOTE_WRITE; + + if (!wr) + wr = &sig_wr; + else + wr->next = &sig_wr; + + ret = ib_post_send(ib_conn->qp, wr, &bad_wr); + if (ret) { + iser_err("reg_sig_mr failed, ret:%d\n", ret); + goto err; + } + desc->reg_indicators &= ~ISER_SIG_KEY_VALID; + + sig_sge->lkey = pi_ctx->sig_mr->lkey; + sig_sge->addr = 0; + sig_sge->length = data_sge->length + prot_sge->length; + if (scsi_get_prot_op(iser_task->sc) == SCSI_PROT_WRITE_INSERT || + scsi_get_prot_op(iser_task->sc) == SCSI_PROT_READ_STRIP) { + sig_sge->length += (data_sge->length / + iser_task->sc->device->sector_size) * 8; + } + + iser_dbg("sig_sge: addr: 0x%llx length: %u lkey: 0x%x\n", + sig_sge->addr, sig_sge->length, + sig_sge->lkey); +err: + return ret; +} + +static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task, + struct iser_regd_buf *regd_buf, + struct iser_data_buf *mem, + enum iser_reg_indicator ind, + struct ib_sge *sge) +{ + struct fast_reg_descriptor *desc = regd_buf->reg.mem_h; + struct iser_conn *ib_conn = iser_task->ib_conn; + struct iser_device *device = ib_conn->device; + struct ib_device *ibdev = device->ib_device; + struct ib_mr *mr; + struct ib_fast_reg_page_list *frpl; + struct ib_send_wr fastreg_wr, inv_wr; + struct ib_send_wr *bad_wr, *wr = NULL; + u8 key; + int ret, offset, size, plen; + + /* if there a single dma entry, dma mr suffices */ + if (mem->dma_nents == 1) { + struct scatterlist *sg = (struct scatterlist *)mem->buf; + + sge->lkey = device->mr->lkey; + sge->addr = ib_sg_dma_address(ibdev, &sg[0]); + sge->length = ib_sg_dma_len(ibdev, &sg[0]); + + iser_dbg("Single DMA entry: lkey=0x%x, addr=0x%llx, length=0x%x\n", + sge->lkey, sge->addr, sge->length); + return 0; + } + + if (ind == ISER_DATA_KEY_VALID) { + mr = desc->data_mr; + frpl = desc->data_frpl; + } else { + mr = desc->pi_ctx->prot_mr; + frpl = desc->pi_ctx->prot_frpl; + } + + plen = iser_sg_to_page_vec(mem, device->ib_device, frpl->page_list, + &offset, &size); + if (plen * SIZE_4K < size) { + iser_err("fast reg page_list too short to hold this SG\n"); + return -EINVAL; + } + + if (!(desc->reg_indicators & ind)) { + memset(&inv_wr, 0, sizeof(inv_wr)); + inv_wr.wr_id = ISER_FASTREG_LI_WRID; + inv_wr.opcode = IB_WR_LOCAL_INV; + inv_wr.ex.invalidate_rkey = mr->rkey; + wr = &inv_wr; + /* Bump the key */ + key = (u8)(mr->rkey & 0x000000FF); + ib_update_fast_reg_key(mr, ++key); + } + + /* Prepare FASTREG WR */ + memset(&fastreg_wr, 0, sizeof(fastreg_wr)); + fastreg_wr.wr_id = ISER_FASTREG_LI_WRID; + fastreg_wr.opcode = IB_WR_FAST_REG_MR; + fastreg_wr.wr.fast_reg.iova_start = frpl->page_list[0] + offset; + fastreg_wr.wr.fast_reg.page_list = frpl; + fastreg_wr.wr.fast_reg.page_list_len = plen; + fastreg_wr.wr.fast_reg.page_shift = SHIFT_4K; + fastreg_wr.wr.fast_reg.length = size; + fastreg_wr.wr.fast_reg.rkey = mr->rkey; + fastreg_wr.wr.fast_reg.access_flags = (IB_ACCESS_LOCAL_WRITE | + IB_ACCESS_REMOTE_WRITE | + IB_ACCESS_REMOTE_READ); + + if (!wr) + wr = &fastreg_wr; + else + wr->next = &fastreg_wr; + + ret = ib_post_send(ib_conn->qp, wr, &bad_wr); + if (ret) { + iser_err("fast registration failed, ret:%d\n", ret); + return ret; + } + desc->reg_indicators &= ~ind; + + sge->lkey = mr->lkey; + sge->addr = frpl->page_list[0] + offset; + sge->length = size; + + return ret; +} + +/** + * iser_reg_rdma_mem_fastreg - Registers memory intended for RDMA, + * using Fast Registration WR (if possible) obtaining rkey and va + * + * returns 0 on success, errno code on failure + */ +int iser_reg_rdma_mem_fastreg(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir) +{ + struct iser_conn *ib_conn = iser_task->ib_conn; + struct iser_device *device = ib_conn->device; + struct ib_device *ibdev = device->ib_device; + struct iser_data_buf *mem = &iser_task->data[cmd_dir]; + struct iser_regd_buf *regd_buf = &iser_task->rdma_regd[cmd_dir]; + struct fast_reg_descriptor *desc = NULL; + struct ib_sge data_sge; + int err, aligned_len; + unsigned long flags; + + aligned_len = iser_data_buf_aligned_len(mem, ibdev); + if (aligned_len != mem->dma_nents) { + err = fall_to_bounce_buf(iser_task, ibdev, mem, + &iser_task->data_copy[cmd_dir], + cmd_dir, aligned_len); + if (err) { + iser_err("failed to allocate bounce buffer\n"); return err; } + mem = &iser_task->data_copy[cmd_dir]; + } + + if (mem->dma_nents != 1 || + scsi_get_prot_op(iser_task->sc) != SCSI_PROT_NORMAL) { + spin_lock_irqsave(&ib_conn->lock, flags); + desc = list_first_entry(&ib_conn->fastreg.pool, + struct fast_reg_descriptor, list); + list_del(&desc->list); + spin_unlock_irqrestore(&ib_conn->lock, flags); + regd_buf->reg.mem_h = desc; + } + + err = iser_fast_reg_mr(iser_task, regd_buf, mem, + ISER_DATA_KEY_VALID, &data_sge); + if (err) + goto err_reg; + + if (scsi_get_prot_op(iser_task->sc) != SCSI_PROT_NORMAL) { + struct ib_sge prot_sge, sig_sge; + + memset(&prot_sge, 0, sizeof(prot_sge)); + if (scsi_prot_sg_count(iser_task->sc)) { + mem = &iser_task->prot[cmd_dir]; + aligned_len = iser_data_buf_aligned_len(mem, ibdev); + if (aligned_len != mem->dma_nents) { + err = fall_to_bounce_buf(iser_task, ibdev, mem, + &iser_task->prot_copy[cmd_dir], + cmd_dir, aligned_len); + if (err) { + iser_err("failed to allocate bounce buffer\n"); + return err; + } + mem = &iser_task->prot_copy[cmd_dir]; + } + + err = iser_fast_reg_mr(iser_task, regd_buf, mem, + ISER_PROT_KEY_VALID, &prot_sge); + if (err) + goto err_reg; + } + + err = iser_reg_sig_mr(iser_task, desc, &data_sge, + &prot_sge, &sig_sge); + if (err) { + iser_err("Failed to register signature mr\n"); + return err; + } + desc->reg_indicators |= ISER_FASTREG_PROTECTED; + + regd_buf->reg.lkey = sig_sge.lkey; + regd_buf->reg.rkey = desc->pi_ctx->sig_mr->rkey; + regd_buf->reg.va = sig_sge.addr; + regd_buf->reg.len = sig_sge.length; + regd_buf->reg.is_mr = 1; + } else { + if (desc) { + regd_buf->reg.rkey = desc->data_mr->rkey; + regd_buf->reg.is_mr = 1; + } else { + regd_buf->reg.rkey = device->mr->rkey; + regd_buf->reg.is_mr = 0; + } + + regd_buf->reg.lkey = data_sge.lkey; + regd_buf->reg.va = data_sge.addr; + regd_buf->reg.len = data_sge.length; } - /* 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; +err_reg: + if (desc) { + spin_lock_irqsave(&ib_conn->lock, flags); + list_add_tail(&desc->list, &ib_conn->fastreg.pool); + spin_unlock_irqrestore(&ib_conn->lock, flags); + } + + return err; } diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c index 993f0a8ff28..ea01075f9f9 100644 --- a/drivers/infiniband/ulp/iser/iser_verbs.c +++ b/drivers/infiniband/ulp/iser/iser_verbs.c @@ -1,6 +1,7 @@ /* * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved. * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved. + * Copyright (c) 2013-2014 Mellanox Technologies. 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 @@ -29,20 +30,17 @@ * 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 <linux/kernel.h> #include <linux/module.h> +#include <linux/slab.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) +#define ISER_MAX_RX_CQ_LEN (ISER_QP_MAX_RECV_DTOS * ISCSI_ISER_MAX_CONN) +#define ISER_MAX_TX_CQ_LEN (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); @@ -57,6 +55,13 @@ static void iser_qp_event_callback(struct ib_event *cause, void *context) iser_err("got qp event %d\n",cause->event); } +static void iser_event_handler(struct ib_event_handler *handler, + struct ib_event *event) +{ + iser_err("async event %d on device %s port %d\n", event->event, + event->device->name, event->element.port_num); +} + /** * iser_create_device_ib_res - creates Protection Domain (PD), Completion * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with @@ -66,24 +71,78 @@ static void iser_qp_event_callback(struct ib_event *cause, void *context) */ static int iser_create_device_ib_res(struct iser_device *device) { + struct iser_cq_desc *cq_desc; + struct ib_device_attr *dev_attr = &device->dev_attr; + int ret, i, j; + + ret = ib_query_device(device->ib_device, dev_attr); + if (ret) { + pr_warn("Query device failed for %s\n", device->ib_device->name); + return ret; + } + + /* Assign function handles - based on FMR support */ + if (device->ib_device->alloc_fmr && device->ib_device->dealloc_fmr && + device->ib_device->map_phys_fmr && device->ib_device->unmap_fmr) { + iser_info("FMR supported, using FMR for registration\n"); + device->iser_alloc_rdma_reg_res = iser_create_fmr_pool; + device->iser_free_rdma_reg_res = iser_free_fmr_pool; + device->iser_reg_rdma_mem = iser_reg_rdma_mem_fmr; + device->iser_unreg_rdma_mem = iser_unreg_mem_fmr; + } else + if (dev_attr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) { + iser_info("FastReg supported, using FastReg for registration\n"); + device->iser_alloc_rdma_reg_res = iser_create_fastreg_pool; + device->iser_free_rdma_reg_res = iser_free_fastreg_pool; + device->iser_reg_rdma_mem = iser_reg_rdma_mem_fastreg; + device->iser_unreg_rdma_mem = iser_unreg_mem_fastreg; + } else { + iser_err("IB device does not support FMRs nor FastRegs, can't register memory\n"); + return -1; + } + + device->cqs_used = min(ISER_MAX_CQ, device->ib_device->num_comp_vectors); + iser_info("using %d CQs, device %s supports %d vectors\n", + device->cqs_used, device->ib_device->name, + device->ib_device->num_comp_vectors); + + device->cq_desc = kmalloc(sizeof(struct iser_cq_desc) * device->cqs_used, + GFP_KERNEL); + if (device->cq_desc == NULL) + goto cq_desc_err; + cq_desc = device->cq_desc; + 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, 0); - if (IS_ERR(device->cq)) - goto cq_err; + for (i = 0; i < device->cqs_used; i++) { + cq_desc[i].device = device; + cq_desc[i].cq_index = i; + + device->rx_cq[i] = ib_create_cq(device->ib_device, + iser_cq_callback, + iser_cq_event_callback, + (void *)&cq_desc[i], + ISER_MAX_RX_CQ_LEN, i); + if (IS_ERR(device->rx_cq[i])) + goto cq_err; + + device->tx_cq[i] = ib_create_cq(device->ib_device, + NULL, iser_cq_event_callback, + (void *)&cq_desc[i], + ISER_MAX_TX_CQ_LEN, i); - if (ib_req_notify_cq(device->cq, IB_CQ_NEXT_COMP)) - goto cq_arm_err; + if (IS_ERR(device->tx_cq[i])) + goto cq_err; - tasklet_init(&device->cq_tasklet, - iser_cq_tasklet_fn, - (unsigned long)device); + if (ib_req_notify_cq(device->rx_cq[i], IB_CQ_NEXT_COMP)) + goto cq_err; + + tasklet_init(&device->cq_tasklet[i], + iser_cq_tasklet_fn, + (unsigned long)&cq_desc[i]); + } device->mr = ib_get_dma_mr(device->pd, IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | @@ -91,15 +150,29 @@ static int iser_create_device_ib_res(struct iser_device *device) if (IS_ERR(device->mr)) goto dma_mr_err; + INIT_IB_EVENT_HANDLER(&device->event_handler, device->ib_device, + iser_event_handler); + if (ib_register_event_handler(&device->event_handler)) + goto handler_err; + return 0; +handler_err: + ib_dereg_mr(device->mr); dma_mr_err: - tasklet_kill(&device->cq_tasklet); -cq_arm_err: - ib_destroy_cq(device->cq); + for (j = 0; j < device->cqs_used; j++) + tasklet_kill(&device->cq_tasklet[j]); cq_err: + for (j = 0; j < i; j++) { + if (device->tx_cq[j]) + ib_destroy_cq(device->tx_cq[j]); + if (device->rx_cq[j]) + ib_destroy_cq(device->rx_cq[j]); + } ib_dealloc_pd(device->pd); pd_err: + kfree(device->cq_desc); +cq_desc_err: iser_err("failed to allocate an IB resource\n"); return -1; } @@ -110,43 +183,45 @@ pd_err: */ static void iser_free_device_ib_res(struct iser_device *device) { + int i; BUG_ON(device->mr == NULL); - tasklet_kill(&device->cq_tasklet); + for (i = 0; i < device->cqs_used; i++) { + tasklet_kill(&device->cq_tasklet[i]); + (void)ib_destroy_cq(device->tx_cq[i]); + (void)ib_destroy_cq(device->rx_cq[i]); + device->tx_cq[i] = NULL; + device->rx_cq[i] = NULL; + } + (void)ib_unregister_event_handler(&device->event_handler); (void)ib_dereg_mr(device->mr); - (void)ib_destroy_cq(device->cq); (void)ib_dealloc_pd(device->pd); + kfree(device->cq_desc); + device->mr = NULL; - device->cq = NULL; device->pd = NULL; } /** - * iser_create_ib_conn_res - Creates FMR pool and Queue-Pair (QP) + * iser_create_fmr_pool - Creates FMR pool and page_vector * - * returns 0 on success, -1 on failure + * returns 0 on success, or errno code on failure */ -static int iser_create_ib_conn_res(struct iser_conn *ib_conn) +int iser_create_fmr_pool(struct iser_conn *ib_conn, unsigned cmds_max) { - struct iser_device *device; - struct ib_qp_init_attr init_attr; - int ret; + struct iser_device *device = ib_conn->device; struct ib_fmr_pool_param params; + int ret = -ENOMEM; - BUG_ON(ib_conn->device == NULL); - - device = ib_conn->device; + ib_conn->fmr.page_vec = kmalloc(sizeof(*ib_conn->fmr.page_vec) + + (sizeof(u64)*(ISCSI_ISER_SG_TABLESIZE + 1)), + GFP_KERNEL); + if (!ib_conn->fmr.page_vec) + return ret; - 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); + ib_conn->fmr.page_vec->pages = (u64 *)(ib_conn->fmr.page_vec + 1); params.page_shift = SHIFT_4K; /* when the first/last SG element are not start/end * @@ -154,78 +229,290 @@ static int iser_create_ib_conn_res(struct iser_conn *ib_conn) 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_DEF_XMIT_CMDS_MAX * 2; - params.dirty_watermark = ISCSI_DEF_XMIT_CMDS_MAX; + params.pool_size = cmds_max * 2; + params.dirty_watermark = 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; + ib_conn->fmr.pool = ib_create_fmr_pool(device->pd, ¶ms); + if (!IS_ERR(ib_conn->fmr.pool)) + return 0; + + /* no FMR => no need for page_vec */ + kfree(ib_conn->fmr.page_vec); + ib_conn->fmr.page_vec = NULL; + + ret = PTR_ERR(ib_conn->fmr.pool); + ib_conn->fmr.pool = NULL; + if (ret != -ENOSYS) { + iser_err("FMR allocation failed, err %d\n", ret); + return ret; + } else { + iser_warn("FMRs are not supported, using unaligned mode\n"); + return 0; } +} + +/** + * iser_free_fmr_pool - releases the FMR pool and page vec + */ +void iser_free_fmr_pool(struct iser_conn *ib_conn) +{ + iser_info("freeing conn %p fmr pool %p\n", + ib_conn, ib_conn->fmr.pool); + + if (ib_conn->fmr.pool != NULL) + ib_destroy_fmr_pool(ib_conn->fmr.pool); + + ib_conn->fmr.pool = NULL; + + kfree(ib_conn->fmr.page_vec); + ib_conn->fmr.page_vec = NULL; +} + +static int +iser_create_fastreg_desc(struct ib_device *ib_device, struct ib_pd *pd, + bool pi_enable, struct fast_reg_descriptor *desc) +{ + int ret; + + desc->data_frpl = ib_alloc_fast_reg_page_list(ib_device, + ISCSI_ISER_SG_TABLESIZE + 1); + if (IS_ERR(desc->data_frpl)) { + ret = PTR_ERR(desc->data_frpl); + iser_err("Failed to allocate ib_fast_reg_page_list err=%d\n", + ret); + return PTR_ERR(desc->data_frpl); + } + + desc->data_mr = ib_alloc_fast_reg_mr(pd, ISCSI_ISER_SG_TABLESIZE + 1); + if (IS_ERR(desc->data_mr)) { + ret = PTR_ERR(desc->data_mr); + iser_err("Failed to allocate ib_fast_reg_mr err=%d\n", ret); + goto fast_reg_mr_failure; + } + desc->reg_indicators |= ISER_DATA_KEY_VALID; + + if (pi_enable) { + struct ib_mr_init_attr mr_init_attr = {0}; + struct iser_pi_context *pi_ctx = NULL; + + desc->pi_ctx = kzalloc(sizeof(*desc->pi_ctx), GFP_KERNEL); + if (!desc->pi_ctx) { + iser_err("Failed to allocate pi context\n"); + ret = -ENOMEM; + goto pi_ctx_alloc_failure; + } + pi_ctx = desc->pi_ctx; + + pi_ctx->prot_frpl = ib_alloc_fast_reg_page_list(ib_device, + ISCSI_ISER_SG_TABLESIZE); + if (IS_ERR(pi_ctx->prot_frpl)) { + ret = PTR_ERR(pi_ctx->prot_frpl); + iser_err("Failed to allocate prot frpl ret=%d\n", + ret); + goto prot_frpl_failure; + } + + pi_ctx->prot_mr = ib_alloc_fast_reg_mr(pd, + ISCSI_ISER_SG_TABLESIZE + 1); + if (IS_ERR(pi_ctx->prot_mr)) { + ret = PTR_ERR(pi_ctx->prot_mr); + iser_err("Failed to allocate prot frmr ret=%d\n", + ret); + goto prot_mr_failure; + } + desc->reg_indicators |= ISER_PROT_KEY_VALID; + + mr_init_attr.max_reg_descriptors = 2; + mr_init_attr.flags |= IB_MR_SIGNATURE_EN; + pi_ctx->sig_mr = ib_create_mr(pd, &mr_init_attr); + if (IS_ERR(pi_ctx->sig_mr)) { + ret = PTR_ERR(pi_ctx->sig_mr); + iser_err("Failed to allocate signature enabled mr err=%d\n", + ret); + goto sig_mr_failure; + } + desc->reg_indicators |= ISER_SIG_KEY_VALID; + } + desc->reg_indicators &= ~ISER_FASTREG_PROTECTED; + + iser_dbg("Create fr_desc %p page_list %p\n", + desc, desc->data_frpl->page_list); + + return 0; +sig_mr_failure: + ib_dereg_mr(desc->pi_ctx->prot_mr); +prot_mr_failure: + ib_free_fast_reg_page_list(desc->pi_ctx->prot_frpl); +prot_frpl_failure: + kfree(desc->pi_ctx); +pi_ctx_alloc_failure: + ib_dereg_mr(desc->data_mr); +fast_reg_mr_failure: + ib_free_fast_reg_page_list(desc->data_frpl); + + return ret; +} + +/** + * iser_create_fastreg_pool - Creates pool of fast_reg descriptors + * for fast registration work requests. + * returns 0 on success, or errno code on failure + */ +int iser_create_fastreg_pool(struct iser_conn *ib_conn, unsigned cmds_max) +{ + struct iser_device *device = ib_conn->device; + struct fast_reg_descriptor *desc; + int i, ret; + + INIT_LIST_HEAD(&ib_conn->fastreg.pool); + ib_conn->fastreg.pool_size = 0; + for (i = 0; i < cmds_max; i++) { + desc = kzalloc(sizeof(*desc), GFP_KERNEL); + if (!desc) { + iser_err("Failed to allocate a new fast_reg descriptor\n"); + ret = -ENOMEM; + goto err; + } + + ret = iser_create_fastreg_desc(device->ib_device, device->pd, + ib_conn->pi_support, desc); + if (ret) { + iser_err("Failed to create fastreg descriptor err=%d\n", + ret); + kfree(desc); + goto err; + } + + list_add_tail(&desc->list, &ib_conn->fastreg.pool); + ib_conn->fastreg.pool_size++; + } + + return 0; + +err: + iser_free_fastreg_pool(ib_conn); + return ret; +} + +/** + * iser_free_fastreg_pool - releases the pool of fast_reg descriptors + */ +void iser_free_fastreg_pool(struct iser_conn *ib_conn) +{ + struct fast_reg_descriptor *desc, *tmp; + int i = 0; + + if (list_empty(&ib_conn->fastreg.pool)) + return; + + iser_info("freeing conn %p fr pool\n", ib_conn); + + list_for_each_entry_safe(desc, tmp, &ib_conn->fastreg.pool, list) { + list_del(&desc->list); + ib_free_fast_reg_page_list(desc->data_frpl); + ib_dereg_mr(desc->data_mr); + if (desc->pi_ctx) { + ib_free_fast_reg_page_list(desc->pi_ctx->prot_frpl); + ib_dereg_mr(desc->pi_ctx->prot_mr); + ib_destroy_mr(desc->pi_ctx->sig_mr); + kfree(desc->pi_ctx); + } + kfree(desc); + ++i; + } + + if (i < ib_conn->fastreg.pool_size) + iser_warn("pool still has %d regions registered\n", + ib_conn->fastreg.pool_size - i); +} + +/** + * iser_create_ib_conn_res - 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 = -ENOMEM; + int index, min_index = 0; + + BUG_ON(ib_conn->device == NULL); + + device = ib_conn->device; memset(&init_attr, 0, sizeof init_attr); + mutex_lock(&ig.connlist_mutex); + /* select the CQ with the minimal number of usages */ + for (index = 0; index < device->cqs_used; index++) + if (device->cq_active_qps[index] < + device->cq_active_qps[min_index]) + min_index = index; + device->cq_active_qps[min_index]++; + mutex_unlock(&ig.connlist_mutex); + iser_info("cq index %d used for ib_conn %p\n", min_index, ib_conn); + 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.send_cq = device->tx_cq[min_index]; + init_attr.recv_cq = device->rx_cq[min_index]; 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.cap.max_send_sge = 2; + init_attr.cap.max_recv_sge = 1; init_attr.sq_sig_type = IB_SIGNAL_REQ_WR; init_attr.qp_type = IB_QPT_RC; + if (ib_conn->pi_support) { + init_attr.cap.max_send_wr = ISER_QP_SIG_MAX_REQ_DTOS; + init_attr.create_flags |= IB_QP_CREATE_SIGNATURE_EN; + } else { + init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS; + } ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr); if (ret) - goto qp_err; + goto out_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); + iser_info("setting conn %p cma_id %p qp %p\n", + ib_conn, ib_conn->cma_id, + 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: +out_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, + * releases the QP objects, returns 0 on success, * -1 on failure */ static int iser_free_ib_conn_res(struct iser_conn *ib_conn) { + int cq_index; 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); + iser_info("freeing conn %p cma_id %p qp %p\n", + ib_conn, ib_conn->cma_id, + 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->qp != NULL) { + cq_index = ((struct iser_cq_desc *)ib_conn->qp->recv_cq->cq_context)->cq_index; + ib_conn->device->cq_active_qps[cq_index]--; - if (ib_conn->cma_id != NULL) - rdma_destroy_id(ib_conn->cma_id); + rdma_destroy_qp(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; } @@ -272,7 +559,7 @@ 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); + iser_info("device %p refcount %d\n", device, device->refcount); if (!device->refcount) { iser_free_device_ib_res(device); list_del(&device->ig_list); @@ -281,17 +568,6 @@ static void iser_device_try_release(struct iser_device *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) @@ -305,27 +581,46 @@ static int iser_conn_state_comp_exch(struct iser_conn *ib_conn, return ret; } +void iser_release_work(struct work_struct *work) +{ + struct iser_conn *ib_conn; + + ib_conn = container_of(work, struct iser_conn, release_work); + + /* wait for .conn_stop callback */ + wait_for_completion(&ib_conn->stop_completion); + + /* wait for the qp`s post send and post receive buffers to empty */ + wait_event_interruptible(ib_conn->wait, + ib_conn->state == ISER_CONN_DOWN); + + iser_conn_release(ib_conn); +} + /** * Frees all conn objects and deallocs conn descriptor */ -static void iser_conn_release(struct iser_conn *ib_conn) +void iser_conn_release(struct iser_conn *ib_conn) { struct iser_device *device = ib_conn->device; - BUG_ON(ib_conn->state != ISER_CONN_DOWN); + BUG_ON(ib_conn->state == ISER_CONN_UP); mutex_lock(&ig.connlist_mutex); list_del(&ib_conn->conn_list); mutex_unlock(&ig.connlist_mutex); - + iser_free_rx_descriptors(ib_conn); 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); - if (ib_conn->iser_conn) - ib_conn->iser_conn->ib_conn = NULL; - kfree(ib_conn); + /* if cma handler context, the caller actually destroy the id */ + if (ib_conn->cma_id != NULL) { + rdma_destroy_id(ib_conn->cma_id); + ib_conn->cma_id = NULL; + } + iscsi_destroy_endpoint(ib_conn->ep); } /** @@ -345,16 +640,12 @@ void iser_conn_terminate(struct iser_conn *ib_conn) 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; @@ -377,10 +668,24 @@ static void iser_addr_handler(struct rdma_cm_id *cma_id) ib_conn = (struct iser_conn *)cma_id->context; ib_conn->device = device; + /* connection T10-PI support */ + if (iser_pi_enable) { + if (!(device->dev_attr.device_cap_flags & + IB_DEVICE_SIGNATURE_HANDOVER)) { + iser_warn("T10-PI requested but not supported on %s, " + "continue without T10-PI\n", + ib_conn->device->ib_device->name); + ib_conn->pi_support = false; + } else { + ib_conn->pi_support = true; + } + } + ret = rdma_resolve_route(cma_id, 1000); if (ret) { iser_err("resolve route failed: %d\n", ret); iser_connect_error(cma_id); + return; } } @@ -388,24 +693,24 @@ static void iser_route_handler(struct rdma_cm_id *cma_id) { struct rdma_conn_param conn_param; int ret; + struct iser_cm_hdr req_hdr; 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; + memset(&req_hdr, 0, sizeof(req_hdr)); + req_hdr.flags = (ISER_ZBVA_NOT_SUPPORTED | + ISER_SEND_W_INV_NOT_SUPPORTED); + conn_param.private_data = (void *)&req_hdr; + conn_param.private_data_len = sizeof(struct iser_cm_hdr); + ret = rdma_connect(cma_id, &conn_param); if (ret) { iser_err("failure connecting: %d\n", ret); @@ -420,10 +725,15 @@ failure: static void iser_connected_handler(struct rdma_cm_id *cma_id) { struct iser_conn *ib_conn; + struct ib_qp_attr attr; + struct ib_qp_init_attr init_attr; + + (void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr); + iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num); ib_conn = (struct iser_conn *)cma_id->context; - ib_conn->state = ISER_CONN_UP; - wake_up_interruptible(&ib_conn->wait); + if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_PENDING, ISER_CONN_UP)) + wake_up_interruptible(&ib_conn->wait); } static void iser_disconnected_handler(struct rdma_cm_id *cma_id) @@ -431,17 +741,19 @@ 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); + ISER_CONN_TERMINATING)){ + if (ib_conn->iscsi_conn) + iscsi_conn_failure(ib_conn->iscsi_conn, ISCSI_ERR_CONN_FAILED); + else + iser_err("iscsi_iser connection isn't bound\n"); + } /* Complete the termination process if no posts are pending */ - if ((atomic_read(&ib_conn->post_recv_buf_count) == 0) && + if (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); @@ -450,9 +762,8 @@ static void iser_disconnected_handler(struct rdma_cm_id *cma_id) 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); + iser_info("event %d status %d conn %p id %p\n", + event->event, event->status, cma_id->context, cma_id); switch (event->event) { case RDMA_CM_EVENT_ADDR_RESOLVED: @@ -469,46 +780,34 @@ static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *eve 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: - iser_err("Device removal is currently unsupported\n"); - BUG(); + case RDMA_CM_EVENT_ADDR_CHANGE: + iser_disconnected_handler(cma_id); break; default: iser_err("Unexpected RDMA CM event (%d)\n", event->event); break; } - return ret; + return 0; } -int iser_conn_init(struct iser_conn **ibconn) +void iser_conn_init(struct iser_conn *ib_conn) { - 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); + ib_conn->post_recv_buf_count = 0; atomic_set(&ib_conn->post_send_buf_count, 0); + init_completion(&ib_conn->stop_completion); 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 + * sleeps until the connection is established or rejected */ int iser_connect(struct iser_conn *ib_conn, struct sockaddr_in *src_addr, @@ -518,20 +817,20 @@ int iser_connect(struct iser_conn *ib_conn, 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); + sprintf(ib_conn->name, "%pI4:%d", + &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); + iser_info("connecting to: %pI4, port 0x%x\n", + &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); + RDMA_PS_TCP, IB_QPT_RC); if (IS_ERR(ib_conn->cma_id)) { err = PTR_ERR(ib_conn->cma_id); iser_err("rdma_create_id failed: %d\n", err); @@ -587,7 +886,7 @@ int iser_reg_page_vec(struct iser_conn *ib_conn, page_list = page_vec->pages; io_addr = page_list[0]; - mem = ib_fmr_pool_map_phys(ib_conn->fmr_pool, + mem = ib_fmr_pool_map_phys(ib_conn->fmr.pool, page_list, page_vec->length, io_addr); @@ -602,7 +901,7 @@ int iser_reg_page_vec(struct iser_conn *ib_conn, mem_reg->rkey = mem->fmr->rkey; mem_reg->len = page_vec->length * SIZE_4K; mem_reg->va = io_addr; - mem_reg->is_fmr = 1; + mem_reg->is_mr = 1; mem_reg->mem_h = (void *)mem; mem_reg->va += page_vec->offset; @@ -620,12 +919,18 @@ int iser_reg_page_vec(struct iser_conn *ib_conn, } /** - * Unregister (previosuly registered) memory. + * Unregister (previosuly registered using FMR) memory. + * If memory is non-FMR does nothing. */ -void iser_unreg_mem(struct iser_mem_reg *reg) +void iser_unreg_mem_fmr(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir) { + struct iser_mem_reg *reg = &iser_task->rdma_regd[cmd_dir].reg; int ret; + if (!reg->is_mr) + return; + iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n",reg->mem_h); ret = ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h); @@ -635,185 +940,260 @@ void iser_unreg_mem(struct iser_mem_reg *reg) 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) +void iser_unreg_mem_fastreg(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir) { - int i; - struct ib_sge *sge; - struct iser_regd_buf *regd_buf; + struct iser_mem_reg *reg = &iser_task->rdma_regd[cmd_dir].reg; + struct iser_conn *ib_conn = iser_task->ib_conn; + struct fast_reg_descriptor *desc = reg->mem_h; - 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(); - } + if (!reg->is_mr) + return; - for (i = 0; i < dto->regd_vector_len; i++) { - sge = &iov[i]; - regd_buf = dto->regd[i]; + reg->mem_h = NULL; + reg->is_mr = 0; + spin_lock_bh(&ib_conn->lock); + list_add_tail(&desc->list, &ib_conn->fastreg.pool); + spin_unlock_bh(&ib_conn->lock); +} - sge->addr = regd_buf->reg.va; - sge->length = regd_buf->reg.len; - sge->lkey = regd_buf->reg.lkey; +int iser_post_recvl(struct iser_conn *ib_conn) +{ + struct ib_recv_wr rx_wr, *rx_wr_failed; + struct ib_sge sge; + int ib_ret; - if (dto->used_sz[i] > 0) /* Adjust size */ - sge->length = dto->used_sz[i]; + sge.addr = ib_conn->login_resp_dma; + sge.length = ISER_RX_LOGIN_SIZE; + sge.lkey = ib_conn->device->mr->lkey; - /* 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(); - } + rx_wr.wr_id = (unsigned long)ib_conn->login_resp_buf; + rx_wr.sg_list = &sge; + rx_wr.num_sge = 1; + rx_wr.next = NULL; - sge->addr += dto->offset[i]; /* Adjust offset */ + ib_conn->post_recv_buf_count++; + ib_ret = ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed); + if (ib_ret) { + iser_err("ib_post_recv failed ret=%d\n", ib_ret); + ib_conn->post_recv_buf_count--; } + return ib_ret; } -/** - * iser_post_recv - Posts a receive buffer. - * - * returns 0 on success, -1 on failure - */ -int iser_post_recv(struct iser_desc *rx_desc) +int iser_post_recvm(struct iser_conn *ib_conn, int count) { - 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->ib_conn; - - iser_dto_to_iov(recv_dto, iov, 2); + struct ib_recv_wr *rx_wr, *rx_wr_failed; + int i, ib_ret; + unsigned int my_rx_head = ib_conn->rx_desc_head; + struct iser_rx_desc *rx_desc; + + for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) { + rx_desc = &ib_conn->rx_descs[my_rx_head]; + rx_wr->wr_id = (unsigned long)rx_desc; + rx_wr->sg_list = &rx_desc->rx_sg; + rx_wr->num_sge = 1; + rx_wr->next = rx_wr + 1; + my_rx_head = (my_rx_head + 1) & ib_conn->qp_max_recv_dtos_mask; + } - 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; + rx_wr--; + rx_wr->next = NULL; /* mark end of work requests list */ - atomic_inc(&ib_conn->post_recv_buf_count); - ib_ret = ib_post_recv(ib_conn->qp, &recv_wr, &recv_wr_failed); + ib_conn->post_recv_buf_count += count; + ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &rx_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; + ib_conn->post_recv_buf_count -= count; + } else + ib_conn->rx_desc_head = my_rx_head; + return ib_ret; } + /** * 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 iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc) { - int ib_ret, ret_val = 0; + int ib_ret; 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->ib_conn; - iser_dto_to_iov(dto, iov, MAX_REGD_BUF_VECTOR_LEN); + ib_dma_sync_single_for_device(ib_conn->device->ib_device, + tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE); 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.sg_list = tx_desc->tx_sg; + send_wr.num_sge = tx_desc->num_sge; send_wr.opcode = IB_WR_SEND; - send_wr.send_flags = dto->notify_enable ? IB_SEND_SIGNALED : 0; + send_wr.send_flags = IB_SEND_SIGNALED; 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; + return ib_ret; } -static void iser_handle_comp_error(struct iser_desc *desc) +static void iser_handle_comp_error(struct iser_tx_desc *desc, + struct iser_conn *ib_conn) { - struct iser_dto *dto = &desc->dto; - struct iser_conn *ib_conn = dto->ib_conn; - - iser_dto_buffs_release(dto); - - if (desc->type == ISCSI_RX) { - kfree(desc->data); + if (desc && desc->type == ISCSI_TX_DATAOUT) 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 && + if (ib_conn->post_recv_buf_count == 0 && atomic_read(&ib_conn->post_send_buf_count) == 0) { /* 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_conn_failure(ib_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); + /* no more non completed posts to the QP, complete the + * termination process w.o worrying on disconnect event */ + ib_conn->state = ISER_CONN_DOWN; + wake_up_interruptible(&ib_conn->wait); + } +} + +static int iser_drain_tx_cq(struct iser_device *device, int cq_index) +{ + struct ib_cq *cq = device->tx_cq[cq_index]; + struct ib_wc wc; + struct iser_tx_desc *tx_desc; + struct iser_conn *ib_conn; + int completed_tx = 0; + + while (ib_poll_cq(cq, 1, &wc) == 1) { + tx_desc = (struct iser_tx_desc *) (unsigned long) wc.wr_id; + ib_conn = wc.qp->qp_context; + if (wc.status == IB_WC_SUCCESS) { + if (wc.opcode == IB_WC_SEND) + iser_snd_completion(tx_desc, ib_conn); + else + iser_err("expected opcode %d got %d\n", + IB_WC_SEND, wc.opcode); + } else { + iser_err("tx id %llx status %d vend_err %x\n", + wc.wr_id, wc.status, wc.vendor_err); + if (wc.wr_id != ISER_FASTREG_LI_WRID) { + atomic_dec(&ib_conn->post_send_buf_count); + iser_handle_comp_error(tx_desc, ib_conn); + } } + completed_tx++; } + return completed_tx; } + static void iser_cq_tasklet_fn(unsigned long data) { - struct iser_device *device = (struct iser_device *)data; - struct ib_cq *cq = device->cq; + struct iser_cq_desc *cq_desc = (struct iser_cq_desc *)data; + struct iser_device *device = cq_desc->device; + int cq_index = cq_desc->cq_index; + struct ib_cq *cq = device->rx_cq[cq_index]; struct ib_wc wc; - struct iser_desc *desc; + struct iser_rx_desc *desc; unsigned long xfer_len; + struct iser_conn *ib_conn; + int completed_tx, completed_rx = 0; + + /* First do tx drain, so in a case where we have rx flushes and a successful + * tx completion we will still go through completion error handling. + */ + completed_tx = iser_drain_tx_cq(device, cq_index); while (ib_poll_cq(cq, 1, &wc) == 1) { - desc = (struct iser_desc *) (unsigned long) wc.wr_id; + desc = (struct iser_rx_desc *) (unsigned long) wc.wr_id; BUG_ON(desc == NULL); - + ib_conn = wc.qp->qp_context; if (wc.status == IB_WC_SUCCESS) { - if (desc->type == ISCSI_RX) { + if (wc.opcode == IB_WC_RECV) { 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); + iser_rcv_completion(desc, xfer_len, ib_conn); + } else + iser_err("expected opcode %d got %d\n", + IB_WC_RECV, wc.opcode); } else { - iser_err("comp w. error op %d status %d\n",desc->type,wc.status); - iser_handle_comp_error(desc); + if (wc.status != IB_WC_WR_FLUSH_ERR) + iser_err("rx id %llx status %d vend_err %x\n", + wc.wr_id, wc.status, wc.vendor_err); + ib_conn->post_recv_buf_count--; + iser_handle_comp_error(NULL, ib_conn); } + completed_rx++; + if (!(completed_rx & 63)) + completed_tx += iser_drain_tx_cq(device, cq_index); } /* #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); + + iser_dbg("got %d rx %d tx completions\n", completed_rx, completed_tx); } static void iser_cq_callback(struct ib_cq *cq, void *cq_context) { - struct iser_device *device = (struct iser_device *)cq_context; + struct iser_cq_desc *cq_desc = (struct iser_cq_desc *)cq_context; + struct iser_device *device = cq_desc->device; + int cq_index = cq_desc->cq_index; + + tasklet_schedule(&device->cq_tasklet[cq_index]); +} + +u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir, sector_t *sector) +{ + struct iser_mem_reg *reg = &iser_task->rdma_regd[cmd_dir].reg; + struct fast_reg_descriptor *desc = reg->mem_h; + unsigned long sector_size = iser_task->sc->device->sector_size; + struct ib_mr_status mr_status; + int ret; - tasklet_schedule(&device->cq_tasklet); + if (desc && desc->reg_indicators & ISER_FASTREG_PROTECTED) { + desc->reg_indicators &= ~ISER_FASTREG_PROTECTED; + ret = ib_check_mr_status(desc->pi_ctx->sig_mr, + IB_MR_CHECK_SIG_STATUS, &mr_status); + if (ret) { + pr_err("ib_check_mr_status failed, ret %d\n", ret); + goto err; + } + + if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) { + sector_t sector_off = mr_status.sig_err.sig_err_offset; + + do_div(sector_off, sector_size + 8); + *sector = scsi_get_lba(iser_task->sc) + sector_off; + + pr_err("PI error found type %d at sector %llx " + "expected %x vs actual %x\n", + mr_status.sig_err.err_type, + (unsigned long long)*sector, + mr_status.sig_err.expected, + mr_status.sig_err.actual); + + switch (mr_status.sig_err.err_type) { + case IB_SIG_BAD_GUARD: + return 0x1; + case IB_SIG_BAD_REFTAG: + return 0x3; + case IB_SIG_BAD_APPTAG: + return 0x2; + } + } + } + + return 0; +err: + /* Not alot we can do here, return ambiguous guard error */ + return 0x1; } |