]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/commitdiff
Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target...
authorLinus Torvalds <torvalds@linux-foundation.org>
Fri, 5 Aug 2016 00:04:37 +0000 (20:04 -0400)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 5 Aug 2016 00:04:37 +0000 (20:04 -0400)
Pull SCSI target updates from Nicholas Bellinger:
 "The most notable item is IBM virtual SCSI target driver, that was
  originally ported to target-core back in 2010 by Tomo-san, and has
  been brought forward to v4.x code by Bryant Ly, Michael Cyr and co
  over the last months.

  Also included are two ORDERED task related bug-fixes Bryant + Michael
  found along the way using ibmvscsis with AIX guests, plus a few
  miscellaneous target-core + iscsi-target bug-fixes with associated
  stable tags"

* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending:
  target: fix spelling mistake: "limitiation" -> "limitation"
  target: Fix residual overflow handling in target_complete_cmd_with_length
  tcm_fc: set and unset FCP_SPPF_TARG_FCN
  iscsi-target: Fix panic when adding second TCP connection to iSCSI session
  ibmvscsis: Initial commit of IBM VSCSI Tgt Driver
  target: Fix ordered task CHECK_CONDITION early exception handling
  target: Fix ordered task target_setup_cmd_from_cdb exception hang
  target: Fix max_unmap_lba_count calc overflow
  target: Fix race between iscsi-target connection shutdown + ABORT_TASK
  target: Fix missing complete during ABORT_TASK + CMD_T_FABRIC_STOP

24 files changed:
MAINTAINERS
drivers/scsi/Kconfig
drivers/scsi/Makefile
drivers/scsi/ibmvscsi/ibmvfc.h
drivers/scsi/ibmvscsi/ibmvscsi.h
drivers/scsi/ibmvscsi/viosrp.h [deleted file]
drivers/scsi/ibmvscsi_tgt/Makefile [new file with mode: 0644]
drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c [new file with mode: 0644]
drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.h [new file with mode: 0644]
drivers/scsi/ibmvscsi_tgt/libsrp.c [new file with mode: 0644]
drivers/scsi/ibmvscsi_tgt/libsrp.h [new file with mode: 0644]
drivers/target/iscsi/iscsi_target.c
drivers/target/iscsi/iscsi_target_login.c
drivers/target/target_core_device.c
drivers/target/target_core_file.c
drivers/target/target_core_iblock.c
drivers/target/target_core_internal.h
drivers/target/target_core_sbc.c
drivers/target/target_core_transport.c
drivers/target/tcm_fc/tfc_sess.c
include/scsi/viosrp.h [new file with mode: 0644]
include/target/target_core_backend.h
include/target/target_core_base.h
include/target/target_core_fabric.h

index bafc8043d4f0df1858f9220ebdab17e932bf4bbc..f518e69e5f295238aae4f4f59f1acfc41863a39e 100644 (file)
@@ -5831,7 +5831,15 @@ M:       Tyrel Datwyler <tyreld@linux.vnet.ibm.com>
 L:     linux-scsi@vger.kernel.org
 S:     Supported
 F:     drivers/scsi/ibmvscsi/ibmvscsi*
-F:     drivers/scsi/ibmvscsi/viosrp.h
+F:     include/scsi/viosrp.h
+
+IBM Power Virtual SCSI Device Target Driver
+M:     Bryant G. Ly <bryantly@linux.vnet.ibm.com>
+M:     Michael Cyr <mikecyr@linux.vnet.ibm.com>
+L:     linux-scsi@vger.kernel.org
+L:     target-devel@vger.kernel.org
+S:     Supported
+F:     drivers/scsi/ibmvscsi_tgt/
 
 IBM Power Virtual FC Device Drivers
 M:     Tyrel Datwyler <tyreld@linux.vnet.ibm.com>
index 1918f5483b23debd923aa5412e9797a2e6bbb46d..7d1b4317eccc1ddea1b380db3900052b731377f7 100644 (file)
@@ -838,6 +838,23 @@ config SCSI_IBMVSCSI
          To compile this driver as a module, choose M here: the
          module will be called ibmvscsi.
 
+config SCSI_IBMVSCSIS
+       tristate "IBM Virtual SCSI Server support"
+       depends on PPC_PSERIES && TARGET_CORE && SCSI && PCI
+       help
+         This is the IBM POWER Virtual SCSI Target Server
+         This driver uses the SRP protocol for communication betwen servers
+         guest and/or the host that run on the same server.
+         More information on VSCSI protocol can be found at www.power.org
+
+         The userspace configuration needed to initialize the driver can be
+         be found here:
+
+         https://github.com/powervm/ibmvscsis/wiki/Configuration
+
+         To compile this driver as a module, choose M here: the
+         module will be called ibmvscsis.
+
 config SCSI_IBMVFC
        tristate "IBM Virtual FC support"
        depends on PPC_PSERIES && SCSI
index 862ab4efad61e90de24c506675801ec8d16ecff5..d5397987e731b7affcc507ece87f0360bb70327b 100644 (file)
@@ -128,6 +128,7 @@ obj-$(CONFIG_SCSI_SNI_53C710)       += 53c700.o sni_53c710.o
 obj-$(CONFIG_SCSI_NSP32)       += nsp32.o
 obj-$(CONFIG_SCSI_IPR)         += ipr.o
 obj-$(CONFIG_SCSI_IBMVSCSI)    += ibmvscsi/
+obj-$(CONFIG_SCSI_IBMVSCSIS)   += ibmvscsi_tgt/
 obj-$(CONFIG_SCSI_IBMVFC)      += ibmvscsi/
 obj-$(CONFIG_SCSI_HPTIOP)      += hptiop.o
 obj-$(CONFIG_SCSI_STEX)                += stex.o
index 8fae03215a85d2d26920146617cbfc68c8a845ed..5c70a52ad3466cdb31a8207ac9b664285ad797c6 100644 (file)
@@ -26,7 +26,7 @@
 
 #include <linux/list.h>
 #include <linux/types.h>
-#include "viosrp.h"
+#include <scsi/viosrp.h>
 
 #define IBMVFC_NAME    "ibmvfc"
 #define IBMVFC_DRIVER_VERSION          "1.0.11"
index 1067367395cd4fa216f160cf6578195c42e7cbd0..e0f6c3aeb4eef35aa390afaa5d7cd6174d8f7a48 100644 (file)
@@ -33,7 +33,7 @@
 #include <linux/list.h>
 #include <linux/completion.h>
 #include <linux/interrupt.h>
-#include "viosrp.h"
+#include <scsi/viosrp.h>
 
 struct scsi_cmnd;
 struct Scsi_Host;
diff --git a/drivers/scsi/ibmvscsi/viosrp.h b/drivers/scsi/ibmvscsi/viosrp.h
deleted file mode 100644 (file)
index c1ab8a4..0000000
+++ /dev/null
@@ -1,225 +0,0 @@
-/*****************************************************************************/
-/* srp.h -- SCSI RDMA Protocol definitions                                   */
-/*                                                                           */
-/* Written By: Colin Devilbis, IBM Corporation                               */
-/*                                                                           */
-/* Copyright (C) 2003 IBM Corporation                                        */
-/*                                                                           */
-/* This program is free software; you can redistribute it and/or modify      */
-/* it under the terms of the GNU General Public License as published by      */
-/* the Free Software Foundation; either version 2 of the License, or         */
-/* (at your option) any later version.                                       */
-/*                                                                           */
-/* This program is distributed in the hope that it will be useful,           */
-/* but WITHOUT ANY WARRANTY; without even the implied warranty of            */
-/* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the             */
-/* GNU General Public License for more details.                              */
-/*                                                                           */
-/* You should have received a copy of the GNU General Public License         */
-/* along with this program; if not, write to the Free Software               */
-/* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */
-/*                                                                           */
-/*                                                                           */
-/* This file contains structures and definitions for IBM RPA (RS/6000        */
-/* platform architecture) implementation of the SRP (SCSI RDMA Protocol)     */
-/* standard.  SRP is used on IBM iSeries and pSeries platforms to send SCSI  */
-/* commands between logical partitions.                                      */
-/*                                                                           */
-/* SRP Information Units (IUs) are sent on a "Command/Response Queue" (CRQ)  */
-/* between partitions.  The definitions in this file are architected,        */
-/* and cannot be changed without breaking compatibility with other versions  */
-/* of Linux and other operating systems (AIX, OS/400) that talk this protocol*/
-/* between logical partitions                                                */
-/*****************************************************************************/
-#ifndef VIOSRP_H
-#define VIOSRP_H
-#include <scsi/srp.h>
-
-#define SRP_VERSION "16.a"
-#define SRP_MAX_IU_LEN 256
-#define SRP_MAX_LOC_LEN 32
-
-union srp_iu {
-       struct srp_login_req login_req;
-       struct srp_login_rsp login_rsp;
-       struct srp_login_rej login_rej;
-       struct srp_i_logout i_logout;
-       struct srp_t_logout t_logout;
-       struct srp_tsk_mgmt tsk_mgmt;
-       struct srp_cmd cmd;
-       struct srp_rsp rsp;
-       u8 reserved[SRP_MAX_IU_LEN];
-};
-
-enum viosrp_crq_headers {
-       VIOSRP_CRQ_FREE = 0x00,
-       VIOSRP_CRQ_CMD_RSP = 0x80,
-       VIOSRP_CRQ_INIT_RSP = 0xC0,
-       VIOSRP_CRQ_XPORT_EVENT = 0xFF
-};
-
-enum viosrp_crq_init_formats {
-       VIOSRP_CRQ_INIT = 0x01,
-       VIOSRP_CRQ_INIT_COMPLETE = 0x02
-};
-
-enum viosrp_crq_formats {
-       VIOSRP_SRP_FORMAT = 0x01,
-       VIOSRP_MAD_FORMAT = 0x02,
-       VIOSRP_OS400_FORMAT = 0x03,
-       VIOSRP_AIX_FORMAT = 0x04,
-       VIOSRP_LINUX_FORMAT = 0x05,
-       VIOSRP_INLINE_FORMAT = 0x06
-};
-
-enum viosrp_crq_status {
-       VIOSRP_OK = 0x0,
-       VIOSRP_NONRECOVERABLE_ERR = 0x1,
-       VIOSRP_VIOLATES_MAX_XFER = 0x2,
-       VIOSRP_PARTNER_PANIC = 0x3,
-       VIOSRP_DEVICE_BUSY = 0x8,
-       VIOSRP_ADAPTER_FAIL = 0x10,
-       VIOSRP_OK2 = 0x99,
-};
-
-struct viosrp_crq {
-       u8 valid;               /* used by RPA */
-       u8 format;              /* SCSI vs out-of-band */
-       u8 reserved;
-       u8 status;              /* non-scsi failure? (e.g. DMA failure) */
-       __be16 timeout;         /* in seconds */
-       __be16 IU_length;               /* in bytes */
-       __be64 IU_data_ptr;     /* the TCE for transferring data */
-};
-
-/* MADs are Management requests above and beyond the IUs defined in the SRP
- * standard.  
- */
-enum viosrp_mad_types {
-       VIOSRP_EMPTY_IU_TYPE = 0x01,
-       VIOSRP_ERROR_LOG_TYPE = 0x02,
-       VIOSRP_ADAPTER_INFO_TYPE = 0x03,
-       VIOSRP_CAPABILITIES_TYPE = 0x05,
-       VIOSRP_ENABLE_FAST_FAIL = 0x08,
-};
-
-enum viosrp_mad_status {
-       VIOSRP_MAD_SUCCESS = 0x00,
-       VIOSRP_MAD_NOT_SUPPORTED = 0xF1,
-       VIOSRP_MAD_FAILED = 0xF7,
-};
-
-enum viosrp_capability_type {
-       MIGRATION_CAPABILITIES = 0x01,
-       RESERVATION_CAPABILITIES = 0x02,
-};
-
-enum viosrp_capability_support {
-       SERVER_DOES_NOT_SUPPORTS_CAP = 0x0,
-       SERVER_SUPPORTS_CAP = 0x01,
-       SERVER_CAP_DATA = 0x02,
-};
-
-enum viosrp_reserve_type {
-       CLIENT_RESERVE_SCSI_2 = 0x01,
-};
-
-enum viosrp_capability_flag {
-       CLIENT_MIGRATED = 0x01,
-       CLIENT_RECONNECT = 0x02,
-       CAP_LIST_SUPPORTED = 0x04,
-       CAP_LIST_DATA = 0x08,
-};
-
-/* 
- * Common MAD header
- */
-struct mad_common {
-       __be32 type;
-       __be16 status;
-       __be16 length;
-       __be64 tag;
-};
-
-/*
- * All SRP (and MAD) requests normally flow from the
- * client to the server.  There is no way for the server to send
- * an asynchronous message back to the client.  The Empty IU is used
- * to hang out a meaningless request to the server so that it can respond
- * asynchrouously with something like a SCSI AER 
- */
-struct viosrp_empty_iu {
-       struct mad_common common;
-       __be64 buffer;
-       __be32 port;
-};
-
-struct viosrp_error_log {
-       struct mad_common common;
-       __be64 buffer;
-};
-
-struct viosrp_adapter_info {
-       struct mad_common common;
-       __be64 buffer;
-};
-
-struct viosrp_fast_fail {
-       struct mad_common common;
-};
-
-struct viosrp_capabilities {
-       struct mad_common common;
-       __be64 buffer;
-};
-
-struct mad_capability_common {
-       __be32 cap_type;
-       __be16 length;
-       __be16 server_support;
-};
-
-struct mad_reserve_cap {
-       struct mad_capability_common common;
-       __be32 type;
-};
-
-struct mad_migration_cap {
-       struct mad_capability_common common;
-       __be32 ecl;
-};
-
-struct capabilities{
-       __be32 flags;
-       char name[SRP_MAX_LOC_LEN];
-       char loc[SRP_MAX_LOC_LEN];
-       struct mad_migration_cap migration;
-       struct mad_reserve_cap reserve;
-};
-
-union mad_iu {
-       struct viosrp_empty_iu empty_iu;
-       struct viosrp_error_log error_log;
-       struct viosrp_adapter_info adapter_info;
-       struct viosrp_fast_fail fast_fail;
-       struct viosrp_capabilities capabilities;
-};
-
-union viosrp_iu {
-       union srp_iu srp;
-       union mad_iu mad;
-};
-
-struct mad_adapter_info_data {
-       char srp_version[8];
-       char partition_name[96];
-       __be32 partition_number;
-#define SRP_MAD_VERSION_1 1
-       __be32 mad_version;
-#define SRP_MAD_OS_LINUX 2
-#define SRP_MAD_OS_AIX 3
-       __be32 os_type;
-       __be32 port_max_txu[8]; /* per-port maximum transfer */
-};
-
-#endif
diff --git a/drivers/scsi/ibmvscsi_tgt/Makefile b/drivers/scsi/ibmvscsi_tgt/Makefile
new file mode 100644 (file)
index 0000000..0c060ce
--- /dev/null
@@ -0,0 +1,3 @@
+obj-$(CONFIG_SCSI_IBMVSCSIS)   += ibmvscsis.o
+
+ibmvscsis-y := libsrp.o ibmvscsi_tgt.o
diff --git a/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c b/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c
new file mode 100644 (file)
index 0000000..b29fef9
--- /dev/null
@@ -0,0 +1,4087 @@
+/*******************************************************************************
+ * IBM Virtual SCSI Target Driver
+ * Copyright (C) 2003-2005 Dave Boutcher (boutcher@us.ibm.com) IBM Corp.
+ *                        Santiago Leon (santil@us.ibm.com) IBM Corp.
+ *                        Linda Xie (lxie@us.ibm.com) IBM Corp.
+ *
+ * Copyright (C) 2005-2011 FUJITA Tomonori <tomof@acm.org>
+ * Copyright (C) 2010 Nicholas A. Bellinger <nab@kernel.org>
+ *
+ * Authors: Bryant G. Ly <bryantly@linux.vnet.ibm.com>
+ * Authors: Michael Cyr <mikecyr@linux.vnet.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ ****************************************************************************/
+
+#define pr_fmt(fmt)     KBUILD_MODNAME ": " fmt
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/types.h>
+#include <linux/list.h>
+#include <linux/string.h>
+
+#include <target/target_core_base.h>
+#include <target/target_core_fabric.h>
+
+#include <asm/hvcall.h>
+#include <asm/vio.h>
+
+#include <scsi/viosrp.h>
+
+#include "ibmvscsi_tgt.h"
+
+#define IBMVSCSIS_VERSION      "v0.2"
+
+#define        INITIAL_SRP_LIMIT       800
+#define        DEFAULT_MAX_SECTORS     256
+
+static uint max_vdma_size = MAX_H_COPY_RDMA;
+
+static char system_id[SYS_ID_NAME_LEN] = "";
+static char partition_name[PARTITION_NAMELEN] = "UNKNOWN";
+static uint partition_number = -1;
+
+/* Adapter list and lock to control it */
+static DEFINE_SPINLOCK(ibmvscsis_dev_lock);
+static LIST_HEAD(ibmvscsis_dev_list);
+
+static long ibmvscsis_parse_command(struct scsi_info *vscsi,
+                                   struct viosrp_crq *crq);
+
+static void ibmvscsis_adapter_idle(struct scsi_info *vscsi);
+
+static void ibmvscsis_determine_resid(struct se_cmd *se_cmd,
+                                     struct srp_rsp *rsp)
+{
+       u32 residual_count = se_cmd->residual_count;
+
+       if (!residual_count)
+               return;
+
+       if (se_cmd->se_cmd_flags & SCF_UNDERFLOW_BIT) {
+               if (se_cmd->data_direction == DMA_TO_DEVICE) {
+                       /* residual data from an underflow write */
+                       rsp->flags = SRP_RSP_FLAG_DOUNDER;
+                       rsp->data_out_res_cnt = cpu_to_be32(residual_count);
+               } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
+                       /* residual data from an underflow read */
+                       rsp->flags = SRP_RSP_FLAG_DIUNDER;
+                       rsp->data_in_res_cnt = cpu_to_be32(residual_count);
+               }
+       } else if (se_cmd->se_cmd_flags & SCF_OVERFLOW_BIT) {
+               if (se_cmd->data_direction == DMA_TO_DEVICE) {
+                       /*  residual data from an overflow write */
+                       rsp->flags = SRP_RSP_FLAG_DOOVER;
+                       rsp->data_out_res_cnt = cpu_to_be32(residual_count);
+               } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
+                       /* residual data from an overflow read */
+                       rsp->flags = SRP_RSP_FLAG_DIOVER;
+                       rsp->data_in_res_cnt = cpu_to_be32(residual_count);
+               }
+       }
+}
+
+/**
+ * connection_broken() - Determine if the connection to the client is good
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * This function attempts to send a ping MAD to the client. If the call to
+ * queue the request returns H_CLOSED then the connection has been broken
+ * and the function returns TRUE.
+ *
+ * EXECUTION ENVIRONMENT:
+ *      Interrupt or Process environment
+ */
+static bool connection_broken(struct scsi_info *vscsi)
+{
+       struct viosrp_crq *crq;
+       u64 buffer[2] = { 0, 0 };
+       long h_return_code;
+       bool rc = false;
+
+       /* create a PING crq */
+       crq = (struct viosrp_crq *)&buffer;
+       crq->valid = VALID_CMD_RESP_EL;
+       crq->format = MESSAGE_IN_CRQ;
+       crq->status = PING;
+
+       h_return_code = h_send_crq(vscsi->dds.unit_id,
+                                  cpu_to_be64(buffer[MSG_HI]),
+                                  cpu_to_be64(buffer[MSG_LOW]));
+
+       pr_debug("connection_broken: rc %ld\n", h_return_code);
+
+       if (h_return_code == H_CLOSED)
+               rc = true;
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_unregister_command_q() - Helper Function-Unregister Command Queue
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * This function calls h_free_q then frees the interrupt bit etc.
+ * It must release the lock before doing so because of the time it can take
+ * for h_free_crq in PHYP
+ * NOTE: the caller must make sure that state and or flags will prevent
+ *      interrupt handler from scheduling work.
+ * NOTE: anyone calling this function may need to set the CRQ_CLOSED flag
+ *      we can't do it here, because we don't have the lock
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Process level
+ */
+static long ibmvscsis_unregister_command_q(struct scsi_info *vscsi)
+{
+       long qrc;
+       long rc = ADAPT_SUCCESS;
+       int ticks = 0;
+
+       do {
+               qrc = h_free_crq(vscsi->dds.unit_id);
+               switch (qrc) {
+               case H_SUCCESS:
+                       break;
+
+               case H_HARDWARE:
+               case H_PARAMETER:
+                       dev_err(&vscsi->dev, "unregister_command_q: error from h_free_crq %ld\n",
+                               qrc);
+                       rc = ERROR;
+                       break;
+
+               case H_BUSY:
+               case H_LONG_BUSY_ORDER_1_MSEC:
+                       /* msleep not good for small values */
+                       usleep_range(1000, 2000);
+                       ticks += 1;
+                       break;
+               case H_LONG_BUSY_ORDER_10_MSEC:
+                       usleep_range(10000, 20000);
+                       ticks += 10;
+                       break;
+               case H_LONG_BUSY_ORDER_100_MSEC:
+                       msleep(100);
+                       ticks += 100;
+                       break;
+               case H_LONG_BUSY_ORDER_1_SEC:
+                       ssleep(1);
+                       ticks += 1000;
+                       break;
+               case H_LONG_BUSY_ORDER_10_SEC:
+                       ssleep(10);
+                       ticks += 10000;
+                       break;
+               case H_LONG_BUSY_ORDER_100_SEC:
+                       ssleep(100);
+                       ticks += 100000;
+                       break;
+               default:
+                       dev_err(&vscsi->dev, "unregister_command_q: unknown error %ld from h_free_crq\n",
+                               qrc);
+                       rc = ERROR;
+                       break;
+               }
+
+               /*
+                * dont wait more then 300 seconds
+                * ticks are in milliseconds more or less
+                */
+               if (ticks > 300000 && qrc != H_SUCCESS) {
+                       rc = ERROR;
+                       dev_err(&vscsi->dev, "Excessive wait for h_free_crq\n");
+               }
+       } while (qrc != H_SUCCESS && rc == ADAPT_SUCCESS);
+
+       pr_debug("Freeing CRQ: phyp rc %ld, rc %ld\n", qrc, rc);
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_delete_client_info() - Helper function to Delete Client Info
+ * @vscsi:     Pointer to our adapter structure
+ * @client_closed:     True if client closed its queue
+ *
+ * Deletes information specific to the client when the client goes away
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt or Process
+ */
+static void ibmvscsis_delete_client_info(struct scsi_info *vscsi,
+                                        bool client_closed)
+{
+       vscsi->client_cap = 0;
+
+       /*
+        * Some things we don't want to clear if we're closing the queue,
+        * because some clients don't resend the host handshake when they
+        * get a transport event.
+        */
+       if (client_closed)
+               vscsi->client_data.os_type = 0;
+}
+
+/**
+ * ibmvscsis_free_command_q() - Free Command Queue
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * This function calls unregister_command_q, then clears interrupts and
+ * any pending interrupt acknowledgments associated with the command q.
+ * It also clears memory if there is no error.
+ *
+ * PHYP did not meet the PAPR architecture so that we must give up the
+ * lock. This causes a timing hole regarding state change.  To close the
+ * hole this routine does accounting on any change that occurred during
+ * the time the lock is not held.
+ * NOTE: must give up and then acquire the interrupt lock, the caller must
+ *      make sure that state and or flags will prevent interrupt handler from
+ *      scheduling work.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Process level, interrupt lock is held
+ */
+static long ibmvscsis_free_command_q(struct scsi_info *vscsi)
+{
+       int bytes;
+       u32 flags_under_lock;
+       u16 state_under_lock;
+       long rc = ADAPT_SUCCESS;
+
+       if (!(vscsi->flags & CRQ_CLOSED)) {
+               vio_disable_interrupts(vscsi->dma_dev);
+
+               state_under_lock = vscsi->new_state;
+               flags_under_lock = vscsi->flags;
+               vscsi->phyp_acr_state = 0;
+               vscsi->phyp_acr_flags = 0;
+
+               spin_unlock_bh(&vscsi->intr_lock);
+               rc = ibmvscsis_unregister_command_q(vscsi);
+               spin_lock_bh(&vscsi->intr_lock);
+
+               if (state_under_lock != vscsi->new_state)
+                       vscsi->phyp_acr_state = vscsi->new_state;
+
+               vscsi->phyp_acr_flags = ((~flags_under_lock) & vscsi->flags);
+
+               if (rc == ADAPT_SUCCESS) {
+                       bytes = vscsi->cmd_q.size * PAGE_SIZE;
+                       memset(vscsi->cmd_q.base_addr, 0, bytes);
+                       vscsi->cmd_q.index = 0;
+                       vscsi->flags |= CRQ_CLOSED;
+
+                       ibmvscsis_delete_client_info(vscsi, false);
+               }
+
+               pr_debug("free_command_q: flags 0x%x, state 0x%hx, acr_flags 0x%x, acr_state 0x%hx\n",
+                        vscsi->flags, vscsi->state, vscsi->phyp_acr_flags,
+                        vscsi->phyp_acr_state);
+       }
+       return rc;
+}
+
+/**
+ * ibmvscsis_cmd_q_dequeue() - Get valid Command element
+ * @mask:      Mask to use in case index wraps
+ * @current_index:     Current index into command queue
+ * @base_addr: Pointer to start of command queue
+ *
+ * Returns a pointer to a valid command element or NULL, if the command
+ * queue is empty
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt environment, interrupt lock held
+ */
+static struct viosrp_crq *ibmvscsis_cmd_q_dequeue(uint mask,
+                                                 uint *current_index,
+                                                 struct viosrp_crq *base_addr)
+{
+       struct viosrp_crq *ptr;
+
+       ptr = base_addr + *current_index;
+
+       if (ptr->valid) {
+               *current_index = (*current_index + 1) & mask;
+               dma_rmb();
+       } else {
+               ptr = NULL;
+       }
+
+       return ptr;
+}
+
+/**
+ * ibmvscsis_send_init_message() -  send initialize message to the client
+ * @vscsi:     Pointer to our adapter structure
+ * @format:    Which Init Message format to send
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt environment interrupt lock held
+ */
+static long ibmvscsis_send_init_message(struct scsi_info *vscsi, u8 format)
+{
+       struct viosrp_crq *crq;
+       u64 buffer[2] = { 0, 0 };
+       long rc;
+
+       crq = (struct viosrp_crq *)&buffer;
+       crq->valid = VALID_INIT_MSG;
+       crq->format = format;
+       rc = h_send_crq(vscsi->dds.unit_id, cpu_to_be64(buffer[MSG_HI]),
+                       cpu_to_be64(buffer[MSG_LOW]));
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_check_init_msg() - Check init message valid
+ * @vscsi:     Pointer to our adapter structure
+ * @format:    Pointer to return format of Init Message, if any.
+ *             Set to UNUSED_FORMAT if no Init Message in queue.
+ *
+ * Checks if an initialize message was queued by the initiatior
+ * after the queue was created and before the interrupt was enabled.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Process level only, interrupt lock held
+ */
+static long ibmvscsis_check_init_msg(struct scsi_info *vscsi, uint *format)
+{
+       struct viosrp_crq *crq;
+       long rc = ADAPT_SUCCESS;
+
+       crq = ibmvscsis_cmd_q_dequeue(vscsi->cmd_q.mask, &vscsi->cmd_q.index,
+                                     vscsi->cmd_q.base_addr);
+       if (!crq) {
+               *format = (uint)UNUSED_FORMAT;
+       } else if (crq->valid == VALID_INIT_MSG && crq->format == INIT_MSG) {
+               *format = (uint)INIT_MSG;
+               crq->valid = INVALIDATE_CMD_RESP_EL;
+               dma_rmb();
+
+               /*
+                * the caller has ensured no initialize message was
+                * sent after the queue was
+                * created so there should be no other message on the queue.
+                */
+               crq = ibmvscsis_cmd_q_dequeue(vscsi->cmd_q.mask,
+                                             &vscsi->cmd_q.index,
+                                             vscsi->cmd_q.base_addr);
+               if (crq) {
+                       *format = (uint)(crq->format);
+                       rc =  ERROR;
+                       crq->valid = INVALIDATE_CMD_RESP_EL;
+                       dma_rmb();
+               }
+       } else {
+               *format = (uint)(crq->format);
+               rc =  ERROR;
+               crq->valid = INVALIDATE_CMD_RESP_EL;
+               dma_rmb();
+       }
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_establish_new_q() - Establish new CRQ queue
+ * @vscsi:     Pointer to our adapter structure
+ * @new_state: New state being established after resetting the queue
+ *
+ * Must be called with interrupt lock held.
+ */
+static long ibmvscsis_establish_new_q(struct scsi_info *vscsi,  uint new_state)
+{
+       long rc = ADAPT_SUCCESS;
+       uint format;
+
+       vscsi->flags &= PRESERVE_FLAG_FIELDS;
+       vscsi->rsp_q_timer.timer_pops = 0;
+       vscsi->debit = 0;
+       vscsi->credit = 0;
+
+       rc = vio_enable_interrupts(vscsi->dma_dev);
+       if (rc) {
+               pr_warn("reset_queue: failed to enable interrupts, rc %ld\n",
+                       rc);
+               return rc;
+       }
+
+       rc = ibmvscsis_check_init_msg(vscsi, &format);
+       if (rc) {
+               dev_err(&vscsi->dev, "reset_queue: check_init_msg failed, rc %ld\n",
+                       rc);
+               return rc;
+       }
+
+       if (format == UNUSED_FORMAT && new_state == WAIT_CONNECTION) {
+               rc = ibmvscsis_send_init_message(vscsi, INIT_MSG);
+               switch (rc) {
+               case H_SUCCESS:
+               case H_DROPPED:
+               case H_CLOSED:
+                       rc = ADAPT_SUCCESS;
+                       break;
+
+               case H_PARAMETER:
+               case H_HARDWARE:
+                       break;
+
+               default:
+                       vscsi->state = UNDEFINED;
+                       rc = H_HARDWARE;
+                       break;
+               }
+       }
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_reset_queue() - Reset CRQ Queue
+ * @vscsi:     Pointer to our adapter structure
+ * @new_state: New state to establish after resetting the queue
+ *
+ * This function calls h_free_q and then calls h_reg_q and does all
+ * of the bookkeeping to get us back to where we can communicate.
+ *
+ * Actually, we don't always call h_free_crq.  A problem was discovered
+ * where one partition would close and reopen his queue, which would
+ * cause his partner to get a transport event, which would cause him to
+ * close and reopen his queue, which would cause the original partition
+ * to get a transport event, etc., etc.  To prevent this, we don't
+ * actually close our queue if the client initiated the reset, (i.e.
+ * either we got a transport event or we have detected that the client's
+ * queue is gone)
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Process environment, called with interrupt lock held
+ */
+static void ibmvscsis_reset_queue(struct scsi_info *vscsi, uint new_state)
+{
+       int bytes;
+       long rc = ADAPT_SUCCESS;
+
+       pr_debug("reset_queue: flags 0x%x\n", vscsi->flags);
+
+       /* don't reset, the client did it for us */
+       if (vscsi->flags & (CLIENT_FAILED | TRANS_EVENT)) {
+               vscsi->flags &=  PRESERVE_FLAG_FIELDS;
+               vscsi->rsp_q_timer.timer_pops = 0;
+               vscsi->debit = 0;
+               vscsi->credit = 0;
+               vscsi->state = new_state;
+               vio_enable_interrupts(vscsi->dma_dev);
+       } else {
+               rc = ibmvscsis_free_command_q(vscsi);
+               if (rc == ADAPT_SUCCESS) {
+                       vscsi->state = new_state;
+
+                       bytes = vscsi->cmd_q.size * PAGE_SIZE;
+                       rc = h_reg_crq(vscsi->dds.unit_id,
+                                      vscsi->cmd_q.crq_token, bytes);
+                       if (rc == H_CLOSED || rc == H_SUCCESS) {
+                               rc = ibmvscsis_establish_new_q(vscsi,
+                                                              new_state);
+                       }
+
+                       if (rc != ADAPT_SUCCESS) {
+                               pr_debug("reset_queue: reg_crq rc %ld\n", rc);
+
+                               vscsi->state = ERR_DISCONNECTED;
+                               vscsi->flags |=  RESPONSE_Q_DOWN;
+                               ibmvscsis_free_command_q(vscsi);
+                       }
+               } else {
+                       vscsi->state = ERR_DISCONNECTED;
+                       vscsi->flags |= RESPONSE_Q_DOWN;
+               }
+       }
+}
+
+/**
+ * ibmvscsis_free_cmd_resources() - Free command resources
+ * @vscsi:     Pointer to our adapter structure
+ * @cmd:       Command which is not longer in use
+ *
+ * Must be called with interrupt lock held.
+ */
+static void ibmvscsis_free_cmd_resources(struct scsi_info *vscsi,
+                                        struct ibmvscsis_cmd *cmd)
+{
+       struct iu_entry *iue = cmd->iue;
+
+       switch (cmd->type) {
+       case TASK_MANAGEMENT:
+       case SCSI_CDB:
+               /*
+                * When the queue goes down this value is cleared, so it
+                * cannot be cleared in this general purpose function.
+                */
+               if (vscsi->debit)
+                       vscsi->debit -= 1;
+               break;
+       case ADAPTER_MAD:
+               vscsi->flags &= ~PROCESSING_MAD;
+               break;
+       case UNSET_TYPE:
+               break;
+       default:
+               dev_err(&vscsi->dev, "free_cmd_resources unknown type %d\n",
+                       cmd->type);
+               break;
+       }
+
+       cmd->iue = NULL;
+       list_add_tail(&cmd->list, &vscsi->free_cmd);
+       srp_iu_put(iue);
+
+       if (list_empty(&vscsi->active_q) && list_empty(&vscsi->schedule_q) &&
+           list_empty(&vscsi->waiting_rsp) && (vscsi->flags & WAIT_FOR_IDLE)) {
+               vscsi->flags &= ~WAIT_FOR_IDLE;
+               complete(&vscsi->wait_idle);
+       }
+}
+
+/**
+ * ibmvscsis_disconnect() - Helper function to disconnect
+ * @work:      Pointer to work_struct, gives access to our adapter structure
+ *
+ * An error has occurred or the driver received a Transport event,
+ * and the driver is requesting that the command queue be de-registered
+ * in a safe manner. If there is no outstanding I/O then we can stop the
+ * queue. If we are restarting the queue it will be reflected in the
+ * the state of the adapter.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Process environment
+ */
+static void ibmvscsis_disconnect(struct work_struct *work)
+{
+       struct scsi_info *vscsi = container_of(work, struct scsi_info,
+                                              proc_work);
+       u16 new_state;
+       bool wait_idle = false;
+       long rc = ADAPT_SUCCESS;
+
+       spin_lock_bh(&vscsi->intr_lock);
+       new_state = vscsi->new_state;
+       vscsi->new_state = 0;
+
+       pr_debug("disconnect: flags 0x%x, state 0x%hx\n", vscsi->flags,
+                vscsi->state);
+
+       /*
+        * check which state we are in and see if we
+        * should transitition to the new state
+        */
+       switch (vscsi->state) {
+       /*  Should never be called while in this state. */
+       case NO_QUEUE:
+       /*
+        * Can never transition from this state;
+        * igonore errors and logout.
+        */
+       case UNCONFIGURING:
+               break;
+
+       /* can transition from this state to UNCONFIGURING */
+       case ERR_DISCONNECT:
+               if (new_state == UNCONFIGURING)
+                       vscsi->state = new_state;
+               break;
+
+       /*
+        * Can transition from this state to to unconfiguring
+        * or err disconnect.
+        */
+       case ERR_DISCONNECT_RECONNECT:
+               switch (new_state) {
+               case UNCONFIGURING:
+               case ERR_DISCONNECT:
+                       vscsi->state = new_state;
+                       break;
+
+               case WAIT_IDLE:
+                       break;
+               default:
+                       break;
+               }
+               break;
+
+       /* can transition from this state to UNCONFIGURING */
+       case ERR_DISCONNECTED:
+               if (new_state == UNCONFIGURING)
+                       vscsi->state = new_state;
+               break;
+
+       /*
+        * If this is a transition into an error state.
+        * a client is attempting to establish a connection
+        * and has violated the RPA protocol.
+        * There can be nothing pending on the adapter although
+        * there can be requests in the command queue.
+        */
+       case WAIT_ENABLED:
+       case PART_UP_WAIT_ENAB:
+               switch (new_state) {
+               case ERR_DISCONNECT:
+                       vscsi->flags |= RESPONSE_Q_DOWN;
+                       vscsi->state = new_state;
+                       vscsi->flags &= ~(SCHEDULE_DISCONNECT |
+                                         DISCONNECT_SCHEDULED);
+                       ibmvscsis_free_command_q(vscsi);
+                       break;
+               case ERR_DISCONNECT_RECONNECT:
+                       ibmvscsis_reset_queue(vscsi, WAIT_ENABLED);
+                       break;
+
+               /* should never happen */
+               case WAIT_IDLE:
+                       rc = ERROR;
+                       dev_err(&vscsi->dev, "disconnect: invalid state %d for WAIT_IDLE\n",
+                               vscsi->state);
+                       break;
+               }
+               break;
+
+       case WAIT_IDLE:
+               switch (new_state) {
+               case ERR_DISCONNECT:
+               case ERR_DISCONNECT_RECONNECT:
+                       vscsi->state = new_state;
+                       break;
+               }
+               break;
+
+       /*
+        * Initiator has not done a successful srp login
+        * or has done a successful srp logout ( adapter was not
+        * busy). In the first case there can be responses queued
+        * waiting for space on the initiators response queue (MAD)
+        * The second case the adapter is idle. Assume the worse case,
+        * i.e. the second case.
+        */
+       case WAIT_CONNECTION:
+       case CONNECTED:
+       case SRP_PROCESSING:
+               wait_idle = true;
+               vscsi->state = new_state;
+               break;
+
+       /* can transition from this state to UNCONFIGURING */
+       case UNDEFINED:
+               if (new_state == UNCONFIGURING)
+                       vscsi->state = new_state;
+               break;
+       default:
+               break;
+       }
+
+       if (wait_idle) {
+               pr_debug("disconnect start wait, active %d, sched %d\n",
+                        (int)list_empty(&vscsi->active_q),
+                        (int)list_empty(&vscsi->schedule_q));
+               if (!list_empty(&vscsi->active_q) ||
+                   !list_empty(&vscsi->schedule_q)) {
+                       vscsi->flags |= WAIT_FOR_IDLE;
+                       pr_debug("disconnect flags 0x%x\n", vscsi->flags);
+                       /*
+                        * This routine is can not be called with the interrupt
+                        * lock held.
+                        */
+                       spin_unlock_bh(&vscsi->intr_lock);
+                       wait_for_completion(&vscsi->wait_idle);
+                       spin_lock_bh(&vscsi->intr_lock);
+               }
+               pr_debug("disconnect stop wait\n");
+
+               ibmvscsis_adapter_idle(vscsi);
+       }
+
+       spin_unlock_bh(&vscsi->intr_lock);
+}
+
+/**
+ * ibmvscsis_post_disconnect() - Schedule the disconnect
+ * @vscsi:     Pointer to our adapter structure
+ * @new_state: State to move to after disconnecting
+ * @flag_bits: Flags to turn on in adapter structure
+ *
+ * If it's already been scheduled, then see if we need to "upgrade"
+ * the new state (if the one passed in is more "severe" than the
+ * previous one).
+ *
+ * PRECONDITION:
+ *     interrupt lock is held
+ */
+static void ibmvscsis_post_disconnect(struct scsi_info *vscsi, uint new_state,
+                                     uint flag_bits)
+{
+       uint state;
+
+       /* check the validity of the new state */
+       switch (new_state) {
+       case UNCONFIGURING:
+       case ERR_DISCONNECT:
+       case ERR_DISCONNECT_RECONNECT:
+       case WAIT_IDLE:
+               break;
+
+       default:
+               dev_err(&vscsi->dev, "post_disconnect: Invalid new state %d\n",
+                       new_state);
+               return;
+       }
+
+       vscsi->flags |= flag_bits;
+
+       pr_debug("post_disconnect: new_state 0x%x, flag_bits 0x%x, vscsi->flags 0x%x, state %hx\n",
+                new_state, flag_bits, vscsi->flags, vscsi->state);
+
+       if (!(vscsi->flags & (DISCONNECT_SCHEDULED | SCHEDULE_DISCONNECT))) {
+               vscsi->flags |= SCHEDULE_DISCONNECT;
+               vscsi->new_state = new_state;
+
+               INIT_WORK(&vscsi->proc_work, ibmvscsis_disconnect);
+               (void)queue_work(vscsi->work_q, &vscsi->proc_work);
+       } else {
+               if (vscsi->new_state)
+                       state = vscsi->new_state;
+               else
+                       state = vscsi->state;
+
+               switch (state) {
+               case NO_QUEUE:
+               case UNCONFIGURING:
+                       break;
+
+               case ERR_DISCONNECTED:
+               case ERR_DISCONNECT:
+               case UNDEFINED:
+                       if (new_state == UNCONFIGURING)
+                               vscsi->new_state = new_state;
+                       break;
+
+               case ERR_DISCONNECT_RECONNECT:
+                       switch (new_state) {
+                       case UNCONFIGURING:
+                       case ERR_DISCONNECT:
+                               vscsi->new_state = new_state;
+                               break;
+                       default:
+                               break;
+                       }
+                       break;
+
+               case WAIT_ENABLED:
+               case PART_UP_WAIT_ENAB:
+               case WAIT_IDLE:
+               case WAIT_CONNECTION:
+               case CONNECTED:
+               case SRP_PROCESSING:
+                       vscsi->new_state = new_state;
+                       break;
+
+               default:
+                       break;
+               }
+       }
+
+       pr_debug("Leaving post_disconnect: flags 0x%x, new_state 0x%x\n",
+                vscsi->flags, vscsi->new_state);
+}
+
+/**
+ * ibmvscsis_trans_event() - Handle a Transport Event
+ * @vscsi:     Pointer to our adapter structure
+ * @crq:       Pointer to CRQ entry containing the Transport Event
+ *
+ * Do the logic to close the I_T nexus.  This function may not
+ * behave to specification.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt, interrupt lock held
+ */
+static long ibmvscsis_trans_event(struct scsi_info *vscsi,
+                                 struct viosrp_crq *crq)
+{
+       long rc = ADAPT_SUCCESS;
+
+       pr_debug("trans_event: format %d, flags 0x%x, state 0x%hx\n",
+                (int)crq->format, vscsi->flags, vscsi->state);
+
+       switch (crq->format) {
+       case MIGRATED:
+       case PARTNER_FAILED:
+       case PARTNER_DEREGISTER:
+               ibmvscsis_delete_client_info(vscsi, true);
+               break;
+
+       default:
+               rc = ERROR;
+               dev_err(&vscsi->dev, "trans_event: invalid format %d\n",
+                       (uint)crq->format);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT,
+                                         RESPONSE_Q_DOWN);
+               break;
+       }
+
+       if (rc == ADAPT_SUCCESS) {
+               switch (vscsi->state) {
+               case NO_QUEUE:
+               case ERR_DISCONNECTED:
+               case UNDEFINED:
+                       break;
+
+               case UNCONFIGURING:
+                       vscsi->flags |= (RESPONSE_Q_DOWN | TRANS_EVENT);
+                       break;
+
+               case WAIT_ENABLED:
+                       break;
+
+               case WAIT_CONNECTION:
+                       break;
+
+               case CONNECTED:
+                       ibmvscsis_post_disconnect(vscsi, WAIT_IDLE,
+                                                 (RESPONSE_Q_DOWN |
+                                                  TRANS_EVENT));
+                       break;
+
+               case PART_UP_WAIT_ENAB:
+                       vscsi->state = WAIT_ENABLED;
+                       break;
+
+               case SRP_PROCESSING:
+                       if ((vscsi->debit > 0) ||
+                           !list_empty(&vscsi->schedule_q) ||
+                           !list_empty(&vscsi->waiting_rsp) ||
+                           !list_empty(&vscsi->active_q)) {
+                               pr_debug("debit %d, sched %d, wait %d, active %d\n",
+                                        vscsi->debit,
+                                        (int)list_empty(&vscsi->schedule_q),
+                                        (int)list_empty(&vscsi->waiting_rsp),
+                                        (int)list_empty(&vscsi->active_q));
+                               pr_warn("connection lost with outstanding work\n");
+                       } else {
+                               pr_debug("trans_event: SRP Processing, but no outstanding work\n");
+                       }
+
+                       ibmvscsis_post_disconnect(vscsi, WAIT_IDLE,
+                                                 (RESPONSE_Q_DOWN |
+                                                  TRANS_EVENT));
+                       break;
+
+               case ERR_DISCONNECT:
+               case ERR_DISCONNECT_RECONNECT:
+               case WAIT_IDLE:
+                       vscsi->flags |= (RESPONSE_Q_DOWN | TRANS_EVENT);
+                       break;
+               }
+       }
+
+       rc =  vscsi->flags & SCHEDULE_DISCONNECT;
+
+       pr_debug("Leaving trans_event: flags 0x%x, state 0x%hx, rc %ld\n",
+                vscsi->flags, vscsi->state, rc);
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_poll_cmd_q() - Poll Command Queue
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * Called to handle command elements that may have arrived while
+ * interrupts were disabled.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     intr_lock must be held
+ */
+static void ibmvscsis_poll_cmd_q(struct scsi_info *vscsi)
+{
+       struct viosrp_crq *crq;
+       long rc;
+       bool ack = true;
+       volatile u8 valid;
+
+       pr_debug("poll_cmd_q: flags 0x%x, state 0x%hx, q index %ud\n",
+                vscsi->flags, vscsi->state, vscsi->cmd_q.index);
+
+       rc = vscsi->flags & SCHEDULE_DISCONNECT;
+       crq = vscsi->cmd_q.base_addr + vscsi->cmd_q.index;
+       valid = crq->valid;
+       dma_rmb();
+
+       while (valid) {
+poll_work:
+               vscsi->cmd_q.index =
+                       (vscsi->cmd_q.index + 1) & vscsi->cmd_q.mask;
+
+               if (!rc) {
+                       rc = ibmvscsis_parse_command(vscsi, crq);
+               } else {
+                       if ((uint)crq->valid == VALID_TRANS_EVENT) {
+                               /*
+                                * must service the transport layer events even
+                                * in an error state, dont break out until all
+                                * the consecutive transport events have been
+                                * processed
+                                */
+                               rc = ibmvscsis_trans_event(vscsi, crq);
+                       } else if (vscsi->flags & TRANS_EVENT) {
+                               /*
+                                * if a tranport event has occurred leave
+                                * everything but transport events on the queue
+                                */
+                               pr_debug("poll_cmd_q, ignoring\n");
+
+                               /*
+                                * need to decrement the queue index so we can
+                                * look at the elment again
+                                */
+                               if (vscsi->cmd_q.index)
+                                       vscsi->cmd_q.index -= 1;
+                               else
+                                       /*
+                                        * index is at 0 it just wrapped.
+                                        * have it index last element in q
+                                        */
+                                       vscsi->cmd_q.index = vscsi->cmd_q.mask;
+                               break;
+                       }
+               }
+
+               crq->valid = INVALIDATE_CMD_RESP_EL;
+
+               crq = vscsi->cmd_q.base_addr + vscsi->cmd_q.index;
+               valid = crq->valid;
+               dma_rmb();
+       }
+
+       if (!rc) {
+               if (ack) {
+                       vio_enable_interrupts(vscsi->dma_dev);
+                       ack = false;
+                       pr_debug("poll_cmd_q, reenabling interrupts\n");
+               }
+               valid = crq->valid;
+               dma_rmb();
+               if (valid)
+                       goto poll_work;
+       }
+
+       pr_debug("Leaving poll_cmd_q: rc %ld\n", rc);
+}
+
+/**
+ * ibmvscsis_free_cmd_qs() - Free elements in queue
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * Free all of the elements on all queues that are waiting for
+ * whatever reason.
+ *
+ * PRECONDITION:
+ *     Called with interrupt lock held
+ */
+static void ibmvscsis_free_cmd_qs(struct scsi_info *vscsi)
+{
+       struct ibmvscsis_cmd *cmd, *nxt;
+
+       pr_debug("free_cmd_qs: waiting_rsp empty %d, timer starter %d\n",
+                (int)list_empty(&vscsi->waiting_rsp),
+                vscsi->rsp_q_timer.started);
+
+       list_for_each_entry_safe(cmd, nxt, &vscsi->waiting_rsp, list) {
+               list_del(&cmd->list);
+               ibmvscsis_free_cmd_resources(vscsi, cmd);
+       }
+}
+
+/**
+ * ibmvscsis_get_free_cmd() - Get free command from list
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * Must be called with interrupt lock held.
+ */
+static struct ibmvscsis_cmd *ibmvscsis_get_free_cmd(struct scsi_info *vscsi)
+{
+       struct ibmvscsis_cmd *cmd = NULL;
+       struct iu_entry *iue;
+
+       iue = srp_iu_get(&vscsi->target);
+       if (iue) {
+               cmd = list_first_entry_or_null(&vscsi->free_cmd,
+                                              struct ibmvscsis_cmd, list);
+               if (cmd) {
+                       list_del(&cmd->list);
+                       cmd->iue = iue;
+                       cmd->type = UNSET_TYPE;
+                       memset(&cmd->se_cmd, 0, sizeof(cmd->se_cmd));
+               } else {
+                       srp_iu_put(iue);
+               }
+       }
+
+       return cmd;
+}
+
+/**
+ * ibmvscsis_adapter_idle() - Helper function to handle idle adapter
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * This function is called when the adapter is idle when the driver
+ * is attempting to clear an error condition.
+ * The adapter is considered busy if any of its cmd queues
+ * are non-empty. This function can be invoked
+ * from the off level disconnect function.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Process environment called with interrupt lock held
+ */
+static void ibmvscsis_adapter_idle(struct scsi_info *vscsi)
+{
+       int free_qs = false;
+
+       pr_debug("adapter_idle: flags 0x%x, state 0x%hx\n", vscsi->flags,
+                vscsi->state);
+
+       /* Only need to free qs if we're disconnecting from client */
+       if (vscsi->state != WAIT_CONNECTION || vscsi->flags & TRANS_EVENT)
+               free_qs = true;
+
+       switch (vscsi->state) {
+       case ERR_DISCONNECT_RECONNECT:
+               ibmvscsis_reset_queue(vscsi, WAIT_CONNECTION);
+               pr_debug("adapter_idle, disc_rec: flags 0x%x\n", vscsi->flags);
+               break;
+
+       case ERR_DISCONNECT:
+               ibmvscsis_free_command_q(vscsi);
+               vscsi->flags &= ~DISCONNECT_SCHEDULED;
+               vscsi->flags |= RESPONSE_Q_DOWN;
+               vscsi->state = ERR_DISCONNECTED;
+               pr_debug("adapter_idle, disc: flags 0x%x, state 0x%hx\n",
+                        vscsi->flags, vscsi->state);
+               break;
+
+       case WAIT_IDLE:
+               vscsi->rsp_q_timer.timer_pops = 0;
+               vscsi->debit = 0;
+               vscsi->credit = 0;
+               if (vscsi->flags & TRANS_EVENT) {
+                       vscsi->state = WAIT_CONNECTION;
+                       vscsi->flags &= PRESERVE_FLAG_FIELDS;
+               } else {
+                       vscsi->state = CONNECTED;
+                       vscsi->flags &= ~DISCONNECT_SCHEDULED;
+               }
+
+               pr_debug("adapter_idle, wait: flags 0x%x, state 0x%hx\n",
+                        vscsi->flags, vscsi->state);
+               ibmvscsis_poll_cmd_q(vscsi);
+               break;
+
+       case ERR_DISCONNECTED:
+               vscsi->flags &= ~DISCONNECT_SCHEDULED;
+               pr_debug("adapter_idle, disconnected: flags 0x%x, state 0x%hx\n",
+                        vscsi->flags, vscsi->state);
+               break;
+
+       default:
+               dev_err(&vscsi->dev, "adapter_idle: in invalid state %d\n",
+                       vscsi->state);
+               break;
+       }
+
+       if (free_qs)
+               ibmvscsis_free_cmd_qs(vscsi);
+
+       /*
+        * There is a timing window where we could lose a disconnect request.
+        * The known path to this window occurs during the DISCONNECT_RECONNECT
+        * case above: reset_queue calls free_command_q, which will release the
+        * interrupt lock.  During that time, a new post_disconnect call can be
+        * made with a "more severe" state (DISCONNECT or UNCONFIGURING).
+        * Because the DISCONNECT_SCHEDULED flag is already set, post_disconnect
+        * will only set the new_state.  Now free_command_q reacquires the intr
+        * lock and clears the DISCONNECT_SCHEDULED flag (using PRESERVE_FLAG_
+        * FIELDS), and the disconnect is lost.  This is particularly bad when
+        * the new disconnect was for UNCONFIGURING, since the unconfigure hangs
+        * forever.
+        * Fix is that free command queue sets acr state and acr flags if there
+        * is a change under the lock
+        * note free command queue writes to this state it clears it
+        * before releasing the lock, different drivers call the free command
+        * queue different times so dont initialize above
+        */
+       if (vscsi->phyp_acr_state != 0) {
+               /*
+                * set any bits in flags that may have been cleared by
+                * a call to free command queue in switch statement
+                * or reset queue
+                */
+               vscsi->flags |= vscsi->phyp_acr_flags;
+               ibmvscsis_post_disconnect(vscsi, vscsi->phyp_acr_state, 0);
+               vscsi->phyp_acr_state = 0;
+               vscsi->phyp_acr_flags = 0;
+
+               pr_debug("adapter_idle: flags 0x%x, state 0x%hx, acr_flags 0x%x, acr_state 0x%hx\n",
+                        vscsi->flags, vscsi->state, vscsi->phyp_acr_flags,
+                        vscsi->phyp_acr_state);
+       }
+
+       pr_debug("Leaving adapter_idle: flags 0x%x, state 0x%hx, new_state 0x%x\n",
+                vscsi->flags, vscsi->state, vscsi->new_state);
+}
+
+/**
+ * ibmvscsis_copy_crq_packet() - Copy CRQ Packet
+ * @vscsi:     Pointer to our adapter structure
+ * @cmd:       Pointer to command element to use to process the request
+ * @crq:       Pointer to CRQ entry containing the request
+ *
+ * Copy the srp information unit from the hosted
+ * partition using remote dma
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt, interrupt lock held
+ */
+static long ibmvscsis_copy_crq_packet(struct scsi_info *vscsi,
+                                     struct ibmvscsis_cmd *cmd,
+                                     struct viosrp_crq *crq)
+{
+       struct iu_entry *iue = cmd->iue;
+       long rc = 0;
+       u16 len;
+
+       len = be16_to_cpu(crq->IU_length);
+       if ((len > SRP_MAX_IU_LEN) || (len == 0)) {
+               dev_err(&vscsi->dev, "copy_crq: Invalid len %d passed", len);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+               return SRP_VIOLATION;
+       }
+
+       rc = h_copy_rdma(len, vscsi->dds.window[REMOTE].liobn,
+                        be64_to_cpu(crq->IU_data_ptr),
+                        vscsi->dds.window[LOCAL].liobn, iue->sbuf->dma);
+
+       switch (rc) {
+       case H_SUCCESS:
+               cmd->init_time = mftb();
+               iue->remote_token = crq->IU_data_ptr;
+               iue->iu_len = len;
+               pr_debug("copy_crq: ioba 0x%llx, init_time 0x%llx\n",
+                        be64_to_cpu(crq->IU_data_ptr), cmd->init_time);
+               break;
+       case H_PERMISSION:
+               if (connection_broken(vscsi))
+                       ibmvscsis_post_disconnect(vscsi,
+                                                 ERR_DISCONNECT_RECONNECT,
+                                                 (RESPONSE_Q_DOWN |
+                                                  CLIENT_FAILED));
+               else
+                       ibmvscsis_post_disconnect(vscsi,
+                                                 ERR_DISCONNECT_RECONNECT, 0);
+
+               dev_err(&vscsi->dev, "copy_crq: h_copy_rdma failed, rc %ld\n",
+                       rc);
+               break;
+       case H_DEST_PARM:
+       case H_SOURCE_PARM:
+       default:
+               dev_err(&vscsi->dev, "copy_crq: h_copy_rdma failed, rc %ld\n",
+                       rc);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+               break;
+       }
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_adapter_info - Service an Adapter Info MAnagement Data gram
+ * @vscsi:     Pointer to our adapter structure
+ * @iue:       Information Unit containing the Adapter Info MAD request
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt adpater lock is held
+ */
+static long ibmvscsis_adapter_info(struct scsi_info *vscsi,
+                                  struct iu_entry *iue)
+{
+       struct viosrp_adapter_info *mad = &vio_iu(iue)->mad.adapter_info;
+       struct mad_adapter_info_data *info;
+       uint flag_bits = 0;
+       dma_addr_t token;
+       long rc;
+
+       mad->common.status = cpu_to_be16(VIOSRP_MAD_SUCCESS);
+
+       if (be16_to_cpu(mad->common.length) > sizeof(*info)) {
+               mad->common.status = cpu_to_be16(VIOSRP_MAD_FAILED);
+               return 0;
+       }
+
+       info = dma_alloc_coherent(&vscsi->dma_dev->dev, sizeof(*info), &token,
+                                 GFP_KERNEL);
+       if (!info) {
+               dev_err(&vscsi->dev, "bad dma_alloc_coherent %p\n",
+                       iue->target);
+               mad->common.status = cpu_to_be16(VIOSRP_MAD_FAILED);
+               return 0;
+       }
+
+       /* Get remote info */
+       rc = h_copy_rdma(be16_to_cpu(mad->common.length),
+                        vscsi->dds.window[REMOTE].liobn,
+                        be64_to_cpu(mad->buffer),
+                        vscsi->dds.window[LOCAL].liobn, token);
+
+       if (rc != H_SUCCESS) {
+               if (rc == H_PERMISSION) {
+                       if (connection_broken(vscsi))
+                               flag_bits = (RESPONSE_Q_DOWN | CLIENT_FAILED);
+               }
+               pr_warn("adapter_info: h_copy_rdma from client failed, rc %ld\n",
+                       rc);
+               pr_debug("adapter_info: ioba 0x%llx, flags 0x%x, flag_bits 0x%x\n",
+                        be64_to_cpu(mad->buffer), vscsi->flags, flag_bits);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT,
+                                         flag_bits);
+               goto free_dma;
+       }
+
+       /*
+        * Copy client info, but ignore partition number, which we
+        * already got from phyp - unless we failed to get it from
+        * phyp (e.g. if we're running on a p5 system).
+        */
+       if (vscsi->client_data.partition_number == 0)
+               vscsi->client_data.partition_number =
+                       be32_to_cpu(info->partition_number);
+       strncpy(vscsi->client_data.srp_version, info->srp_version,
+               sizeof(vscsi->client_data.srp_version));
+       strncpy(vscsi->client_data.partition_name, info->partition_name,
+               sizeof(vscsi->client_data.partition_name));
+       vscsi->client_data.mad_version = be32_to_cpu(info->mad_version);
+       vscsi->client_data.os_type = be32_to_cpu(info->os_type);
+
+       /* Copy our info */
+       strncpy(info->srp_version, SRP_VERSION,
+               sizeof(info->srp_version));
+       strncpy(info->partition_name, vscsi->dds.partition_name,
+               sizeof(info->partition_name));
+       info->partition_number = cpu_to_be32(vscsi->dds.partition_num);
+       info->mad_version = cpu_to_be32(MAD_VERSION_1);
+       info->os_type = cpu_to_be32(LINUX);
+       memset(&info->port_max_txu[0], 0, sizeof(info->port_max_txu));
+       info->port_max_txu[0] = cpu_to_be32(128 * PAGE_SIZE);
+
+       dma_wmb();
+       rc = h_copy_rdma(sizeof(*info), vscsi->dds.window[LOCAL].liobn,
+                        token, vscsi->dds.window[REMOTE].liobn,
+                        be64_to_cpu(mad->buffer));
+       switch (rc) {
+       case H_SUCCESS:
+               break;
+
+       case H_SOURCE_PARM:
+       case H_DEST_PARM:
+       case H_PERMISSION:
+               if (connection_broken(vscsi))
+                       flag_bits = (RESPONSE_Q_DOWN | CLIENT_FAILED);
+       default:
+               dev_err(&vscsi->dev, "adapter_info: h_copy_rdma to client failed, rc %ld\n",
+                       rc);
+               ibmvscsis_post_disconnect(vscsi,
+                                         ERR_DISCONNECT_RECONNECT,
+                                         flag_bits);
+               break;
+       }
+
+free_dma:
+       dma_free_coherent(&vscsi->dma_dev->dev, sizeof(*info), info, token);
+       pr_debug("Leaving adapter_info, rc %ld\n", rc);
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_cap_mad() - Service a Capabilities MAnagement Data gram
+ * @vscsi:     Pointer to our adapter structure
+ * @iue:       Information Unit containing the Capabilities MAD request
+ *
+ * NOTE: if you return an error from this routine you must be
+ * disconnecting or you will cause a hang
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt called with adapter lock held
+ */
+static int ibmvscsis_cap_mad(struct scsi_info *vscsi, struct iu_entry *iue)
+{
+       struct viosrp_capabilities *mad = &vio_iu(iue)->mad.capabilities;
+       struct capabilities *cap;
+       struct mad_capability_common *common;
+       dma_addr_t token;
+       u16 olen, len, status, min_len, cap_len;
+       u32 flag;
+       uint flag_bits = 0;
+       long rc = 0;
+
+       olen = be16_to_cpu(mad->common.length);
+       /*
+        * struct capabilities hardcodes a couple capabilities after the
+        * header, but the capabilities can actually be in any order.
+        */
+       min_len = offsetof(struct capabilities, migration);
+       if ((olen < min_len) || (olen > PAGE_SIZE)) {
+               pr_warn("cap_mad: invalid len %d\n", olen);
+               mad->common.status = cpu_to_be16(VIOSRP_MAD_FAILED);
+               return 0;
+       }
+
+       cap = dma_alloc_coherent(&vscsi->dma_dev->dev, olen, &token,
+                                GFP_KERNEL);
+       if (!cap) {
+               dev_err(&vscsi->dev, "bad dma_alloc_coherent %p\n",
+                       iue->target);
+               mad->common.status = cpu_to_be16(VIOSRP_MAD_FAILED);
+               return 0;
+       }
+       rc = h_copy_rdma(olen, vscsi->dds.window[REMOTE].liobn,
+                        be64_to_cpu(mad->buffer),
+                        vscsi->dds.window[LOCAL].liobn, token);
+       if (rc == H_SUCCESS) {
+               strncpy(cap->name, dev_name(&vscsi->dma_dev->dev),
+                       SRP_MAX_LOC_LEN);
+
+               len = olen - min_len;
+               status = VIOSRP_MAD_SUCCESS;
+               common = (struct mad_capability_common *)&cap->migration;
+
+               while ((len > 0) && (status == VIOSRP_MAD_SUCCESS) && !rc) {
+                       pr_debug("cap_mad: len left %hd, cap type %d, cap len %hd\n",
+                                len, be32_to_cpu(common->cap_type),
+                                be16_to_cpu(common->length));
+
+                       cap_len = be16_to_cpu(common->length);
+                       if (cap_len > len) {
+                               dev_err(&vscsi->dev, "cap_mad: cap len mismatch with total len\n");
+                               status = VIOSRP_MAD_FAILED;
+                               break;
+                       }
+
+                       if (cap_len == 0) {
+                               dev_err(&vscsi->dev, "cap_mad: cap len is 0\n");
+                               status = VIOSRP_MAD_FAILED;
+                               break;
+                       }
+
+                       switch (common->cap_type) {
+                       default:
+                               pr_debug("cap_mad: unsupported capability\n");
+                               common->server_support = 0;
+                               flag = cpu_to_be32((u32)CAP_LIST_SUPPORTED);
+                               cap->flags &= ~flag;
+                               break;
+                       }
+
+                       len = len - cap_len;
+                       common = (struct mad_capability_common *)
+                               ((char *)common + cap_len);
+               }
+
+               mad->common.status = cpu_to_be16(status);
+
+               dma_wmb();
+               rc = h_copy_rdma(olen, vscsi->dds.window[LOCAL].liobn, token,
+                                vscsi->dds.window[REMOTE].liobn,
+                                be64_to_cpu(mad->buffer));
+
+               if (rc != H_SUCCESS) {
+                       pr_debug("cap_mad: failed to copy to client, rc %ld\n",
+                                rc);
+
+                       if (rc == H_PERMISSION) {
+                               if (connection_broken(vscsi))
+                                       flag_bits = (RESPONSE_Q_DOWN |
+                                                    CLIENT_FAILED);
+                       }
+
+                       pr_warn("cap_mad: error copying data to client, rc %ld\n",
+                               rc);
+                       ibmvscsis_post_disconnect(vscsi,
+                                                 ERR_DISCONNECT_RECONNECT,
+                                                 flag_bits);
+               }
+       }
+
+       dma_free_coherent(&vscsi->dma_dev->dev, olen, cap, token);
+
+       pr_debug("Leaving cap_mad, rc %ld, client_cap 0x%x\n",
+                rc, vscsi->client_cap);
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_process_mad() - Service a MAnagement Data gram
+ * @vscsi:     Pointer to our adapter structure
+ * @iue:       Information Unit containing the MAD request
+ *
+ * Must be called with interrupt lock held.
+ */
+static long ibmvscsis_process_mad(struct scsi_info *vscsi, struct iu_entry *iue)
+{
+       struct mad_common *mad = (struct mad_common *)&vio_iu(iue)->mad;
+       struct viosrp_empty_iu *empty;
+       long rc = ADAPT_SUCCESS;
+
+       switch (be32_to_cpu(mad->type)) {
+       case VIOSRP_EMPTY_IU_TYPE:
+               empty = &vio_iu(iue)->mad.empty_iu;
+               vscsi->empty_iu_id = be64_to_cpu(empty->buffer);
+               vscsi->empty_iu_tag = be64_to_cpu(empty->common.tag);
+               mad->status = cpu_to_be16(VIOSRP_MAD_SUCCESS);
+               break;
+       case VIOSRP_ADAPTER_INFO_TYPE:
+               rc = ibmvscsis_adapter_info(vscsi, iue);
+               break;
+       case VIOSRP_CAPABILITIES_TYPE:
+               rc = ibmvscsis_cap_mad(vscsi, iue);
+               break;
+       case VIOSRP_ENABLE_FAST_FAIL:
+               if (vscsi->state == CONNECTED) {
+                       vscsi->fast_fail = true;
+                       mad->status = cpu_to_be16(VIOSRP_MAD_SUCCESS);
+               } else {
+                       pr_warn("fast fail mad sent after login\n");
+                       mad->status = cpu_to_be16(VIOSRP_MAD_FAILED);
+               }
+               break;
+       default:
+               mad->status = cpu_to_be16(VIOSRP_MAD_NOT_SUPPORTED);
+               break;
+       }
+
+       return rc;
+}
+
+/**
+ * srp_snd_msg_failed() - Handle an error when sending a response
+ * @vscsi:     Pointer to our adapter structure
+ * @rc:                The return code from the h_send_crq command
+ *
+ * Must be called with interrupt lock held.
+ */
+static void srp_snd_msg_failed(struct scsi_info *vscsi, long rc)
+{
+       ktime_t kt;
+
+       if (rc != H_DROPPED) {
+               ibmvscsis_free_cmd_qs(vscsi);
+
+               if (rc == H_CLOSED)
+                       vscsi->flags |= CLIENT_FAILED;
+
+               /* don't flag the same problem multiple times */
+               if (!(vscsi->flags & RESPONSE_Q_DOWN)) {
+                       vscsi->flags |= RESPONSE_Q_DOWN;
+                       if (!(vscsi->state & (ERR_DISCONNECT |
+                                             ERR_DISCONNECT_RECONNECT |
+                                             ERR_DISCONNECTED | UNDEFINED))) {
+                               dev_err(&vscsi->dev, "snd_msg_failed: setting RESPONSE_Q_DOWN, state 0x%hx, flags 0x%x, rc %ld\n",
+                                       vscsi->state, vscsi->flags, rc);
+                       }
+                       ibmvscsis_post_disconnect(vscsi,
+                                                 ERR_DISCONNECT_RECONNECT, 0);
+               }
+               return;
+       }
+
+       /*
+        * The response queue is full.
+        * If the server is processing SRP requests, i.e.
+        * the client has successfully done an
+        * SRP_LOGIN, then it will wait forever for room in
+        * the queue.  However if the system admin
+        * is attempting to unconfigure the server then one
+        * or more children will be in a state where
+        * they are being removed. So if there is even one
+        * child being removed then the driver assumes
+        * the system admin is attempting to break the
+        * connection with the client and MAX_TIMER_POPS
+        * is honored.
+        */
+       if ((vscsi->rsp_q_timer.timer_pops < MAX_TIMER_POPS) ||
+           (vscsi->state == SRP_PROCESSING)) {
+               pr_debug("snd_msg_failed: response queue full, flags 0x%x, timer started %d, pops %d\n",
+                        vscsi->flags, (int)vscsi->rsp_q_timer.started,
+                        vscsi->rsp_q_timer.timer_pops);
+
+               /*
+                * Check if the timer is running; if it
+                * is not then start it up.
+                */
+               if (!vscsi->rsp_q_timer.started) {
+                       if (vscsi->rsp_q_timer.timer_pops <
+                           MAX_TIMER_POPS) {
+                               kt = ktime_set(0, WAIT_NANO_SECONDS);
+                       } else {
+                               /*
+                                * slide the timeslice if the maximum
+                                * timer pops have already happened
+                                */
+                               kt = ktime_set(WAIT_SECONDS, 0);
+                       }
+
+                       vscsi->rsp_q_timer.started = true;
+                       hrtimer_start(&vscsi->rsp_q_timer.timer, kt,
+                                     HRTIMER_MODE_REL);
+               }
+       } else {
+               /*
+                * TBD: Do we need to worry about this? Need to get
+                *      remove working.
+                */
+               /*
+                * waited a long time and it appears the system admin
+                * is bring this driver down
+                */
+               vscsi->flags |= RESPONSE_Q_DOWN;
+               ibmvscsis_free_cmd_qs(vscsi);
+               /*
+                * if the driver is already attempting to disconnect
+                * from the client and has already logged an error
+                * trace this event but don't put it in the error log
+                */
+               if (!(vscsi->state & (ERR_DISCONNECT |
+                                     ERR_DISCONNECT_RECONNECT |
+                                     ERR_DISCONNECTED | UNDEFINED))) {
+                       dev_err(&vscsi->dev, "client crq full too long\n");
+                       ibmvscsis_post_disconnect(vscsi,
+                                                 ERR_DISCONNECT_RECONNECT,
+                                                 0);
+               }
+       }
+}
+
+/**
+ * ibmvscsis_send_messages() - Send a Response
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * Send a response, first checking the waiting queue. Responses are
+ * sent in order they are received. If the response cannot be sent,
+ * because the client queue is full, it stays on the waiting queue.
+ *
+ * PRECONDITION:
+ *     Called with interrupt lock held
+ */
+static void ibmvscsis_send_messages(struct scsi_info *vscsi)
+{
+       u64 msg_hi = 0;
+       /* note do not attmempt to access the IU_data_ptr with this pointer
+        * it is not valid
+        */
+       struct viosrp_crq *crq = (struct viosrp_crq *)&msg_hi;
+       struct ibmvscsis_cmd *cmd, *nxt;
+       struct iu_entry *iue;
+       long rc = ADAPT_SUCCESS;
+
+       if (!(vscsi->flags & RESPONSE_Q_DOWN)) {
+               list_for_each_entry_safe(cmd, nxt, &vscsi->waiting_rsp, list) {
+                       pr_debug("send_messages cmd %p\n", cmd);
+
+                       iue = cmd->iue;
+
+                       crq->valid = VALID_CMD_RESP_EL;
+                       crq->format = cmd->rsp.format;
+
+                       if (cmd->flags & CMD_FAST_FAIL)
+                               crq->status = VIOSRP_ADAPTER_FAIL;
+
+                       crq->IU_length = cpu_to_be16(cmd->rsp.len);
+
+                       rc = h_send_crq(vscsi->dma_dev->unit_address,
+                                       be64_to_cpu(msg_hi),
+                                       be64_to_cpu(cmd->rsp.tag));
+
+                       pr_debug("send_messages: tag 0x%llx, rc %ld\n",
+                                be64_to_cpu(cmd->rsp.tag), rc);
+
+                       /* if all ok free up the command element resources */
+                       if (rc == H_SUCCESS) {
+                               /* some movement has occurred */
+                               vscsi->rsp_q_timer.timer_pops = 0;
+                               list_del(&cmd->list);
+
+                               ibmvscsis_free_cmd_resources(vscsi, cmd);
+                       } else {
+                               srp_snd_msg_failed(vscsi, rc);
+                               break;
+                       }
+               }
+
+               if (!rc) {
+                       /*
+                        * The timer could pop with the queue empty.  If
+                        * this happens, rc will always indicate a
+                        * success; clear the pop count.
+                        */
+                       vscsi->rsp_q_timer.timer_pops = 0;
+               }
+       } else {
+               ibmvscsis_free_cmd_qs(vscsi);
+       }
+}
+
+/* Called with intr lock held */
+static void ibmvscsis_send_mad_resp(struct scsi_info *vscsi,
+                                   struct ibmvscsis_cmd *cmd,
+                                   struct viosrp_crq *crq)
+{
+       struct iu_entry *iue = cmd->iue;
+       struct mad_common *mad = (struct mad_common *)&vio_iu(iue)->mad;
+       uint flag_bits = 0;
+       long rc;
+
+       dma_wmb();
+       rc = h_copy_rdma(sizeof(struct mad_common),
+                        vscsi->dds.window[LOCAL].liobn, iue->sbuf->dma,
+                        vscsi->dds.window[REMOTE].liobn,
+                        be64_to_cpu(crq->IU_data_ptr));
+       if (!rc) {
+               cmd->rsp.format = VIOSRP_MAD_FORMAT;
+               cmd->rsp.len = sizeof(struct mad_common);
+               cmd->rsp.tag = mad->tag;
+               list_add_tail(&cmd->list, &vscsi->waiting_rsp);
+               ibmvscsis_send_messages(vscsi);
+       } else {
+               pr_debug("Error sending mad response, rc %ld\n", rc);
+               if (rc == H_PERMISSION) {
+                       if (connection_broken(vscsi))
+                               flag_bits = (RESPONSE_Q_DOWN | CLIENT_FAILED);
+               }
+               dev_err(&vscsi->dev, "mad: failed to copy to client, rc %ld\n",
+                       rc);
+
+               ibmvscsis_free_cmd_resources(vscsi, cmd);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT,
+                                         flag_bits);
+       }
+}
+
+/**
+ * ibmvscsis_mad() - Service a MAnagement Data gram.
+ * @vscsi:     Pointer to our adapter structure
+ * @crq:       Pointer to the CRQ entry containing the MAD request
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt  called with adapter lock held
+ */
+static long ibmvscsis_mad(struct scsi_info *vscsi, struct viosrp_crq *crq)
+{
+       struct iu_entry *iue;
+       struct ibmvscsis_cmd *cmd;
+       struct mad_common *mad;
+       long rc = ADAPT_SUCCESS;
+
+       switch (vscsi->state) {
+               /*
+                * We have not exchanged Init Msgs yet, so this MAD was sent
+                * before the last Transport Event; client will not be
+                * expecting a response.
+                */
+       case WAIT_CONNECTION:
+               pr_debug("mad: in Wait Connection state, ignoring MAD, flags %d\n",
+                        vscsi->flags);
+               return ADAPT_SUCCESS;
+
+       case SRP_PROCESSING:
+       case CONNECTED:
+               break;
+
+               /*
+                * We should never get here while we're in these states.
+                * Just log an error and get out.
+                */
+       case UNCONFIGURING:
+       case WAIT_IDLE:
+       case ERR_DISCONNECT:
+       case ERR_DISCONNECT_RECONNECT:
+       default:
+               dev_err(&vscsi->dev, "mad: invalid adapter state %d for mad\n",
+                       vscsi->state);
+               return ADAPT_SUCCESS;
+       }
+
+       cmd = ibmvscsis_get_free_cmd(vscsi);
+       if (!cmd) {
+               dev_err(&vscsi->dev, "mad: failed to get cmd, debit %d\n",
+                       vscsi->debit);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+               return ERROR;
+       }
+       iue = cmd->iue;
+       cmd->type = ADAPTER_MAD;
+
+       rc = ibmvscsis_copy_crq_packet(vscsi, cmd, crq);
+       if (!rc) {
+               mad = (struct mad_common *)&vio_iu(iue)->mad;
+
+               pr_debug("mad: type %d\n", be32_to_cpu(mad->type));
+
+               if (be16_to_cpu(mad->length) < 0) {
+                       dev_err(&vscsi->dev, "mad: length is < 0\n");
+                       ibmvscsis_post_disconnect(vscsi,
+                                                 ERR_DISCONNECT_RECONNECT, 0);
+                       rc = SRP_VIOLATION;
+               } else {
+                       rc = ibmvscsis_process_mad(vscsi, iue);
+               }
+
+               pr_debug("mad: status %hd, rc %ld\n", be16_to_cpu(mad->status),
+                        rc);
+
+               if (!rc)
+                       ibmvscsis_send_mad_resp(vscsi, cmd, crq);
+       } else {
+               ibmvscsis_free_cmd_resources(vscsi, cmd);
+       }
+
+       pr_debug("Leaving mad, rc %ld\n", rc);
+       return rc;
+}
+
+/**
+ * ibmvscsis_login_rsp() - Create/copy a login response notice to the client
+ * @vscsi:     Pointer to our adapter structure
+ * @cmd:       Pointer to the command for the SRP Login request
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt, interrupt lock held
+ */
+static long ibmvscsis_login_rsp(struct scsi_info *vscsi,
+                               struct ibmvscsis_cmd *cmd)
+{
+       struct iu_entry *iue = cmd->iue;
+       struct srp_login_rsp *rsp = &vio_iu(iue)->srp.login_rsp;
+       struct format_code *fmt;
+       uint flag_bits = 0;
+       long rc = ADAPT_SUCCESS;
+
+       memset(rsp, 0, sizeof(struct srp_login_rsp));
+
+       rsp->opcode = SRP_LOGIN_RSP;
+       rsp->req_lim_delta = cpu_to_be32(vscsi->request_limit);
+       rsp->tag = cmd->rsp.tag;
+       rsp->max_it_iu_len = cpu_to_be32(SRP_MAX_IU_LEN);
+       rsp->max_ti_iu_len = cpu_to_be32(SRP_MAX_IU_LEN);
+       fmt = (struct format_code *)&rsp->buf_fmt;
+       fmt->buffers = SUPPORTED_FORMATS;
+       vscsi->credit = 0;
+
+       cmd->rsp.len = sizeof(struct srp_login_rsp);
+
+       dma_wmb();
+       rc = h_copy_rdma(cmd->rsp.len, vscsi->dds.window[LOCAL].liobn,
+                        iue->sbuf->dma, vscsi->dds.window[REMOTE].liobn,
+                        be64_to_cpu(iue->remote_token));
+
+       switch (rc) {
+       case H_SUCCESS:
+               break;
+
+       case H_PERMISSION:
+               if (connection_broken(vscsi))
+                       flag_bits = RESPONSE_Q_DOWN | CLIENT_FAILED;
+               dev_err(&vscsi->dev, "login_rsp: error copying to client, rc %ld\n",
+                       rc);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT,
+                                         flag_bits);
+               break;
+       case H_SOURCE_PARM:
+       case H_DEST_PARM:
+       default:
+               dev_err(&vscsi->dev, "login_rsp: error copying to client, rc %ld\n",
+                       rc);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+               break;
+       }
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_srp_login_rej() - Create/copy a login rejection notice to client
+ * @vscsi:     Pointer to our adapter structure
+ * @cmd:       Pointer to the command for the SRP Login request
+ * @reason:    The reason the SRP Login is being rejected, per SRP protocol
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt, interrupt lock held
+ */
+static long ibmvscsis_srp_login_rej(struct scsi_info *vscsi,
+                                   struct ibmvscsis_cmd *cmd, u32 reason)
+{
+       struct iu_entry *iue = cmd->iue;
+       struct srp_login_rej *rej = &vio_iu(iue)->srp.login_rej;
+       struct format_code *fmt;
+       uint flag_bits = 0;
+       long rc = ADAPT_SUCCESS;
+
+       memset(rej, 0, sizeof(*rej));
+
+       rej->opcode = SRP_LOGIN_REJ;
+       rej->reason = cpu_to_be32(reason);
+       rej->tag = cmd->rsp.tag;
+       fmt = (struct format_code *)&rej->buf_fmt;
+       fmt->buffers = SUPPORTED_FORMATS;
+
+       cmd->rsp.len = sizeof(*rej);
+
+       dma_wmb();
+       rc = h_copy_rdma(cmd->rsp.len, vscsi->dds.window[LOCAL].liobn,
+                        iue->sbuf->dma, vscsi->dds.window[REMOTE].liobn,
+                        be64_to_cpu(iue->remote_token));
+
+       switch (rc) {
+       case H_SUCCESS:
+               break;
+       case H_PERMISSION:
+               if (connection_broken(vscsi))
+                       flag_bits =  RESPONSE_Q_DOWN | CLIENT_FAILED;
+               dev_err(&vscsi->dev, "login_rej: error copying to client, rc %ld\n",
+                       rc);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT,
+                                         flag_bits);
+               break;
+       case H_SOURCE_PARM:
+       case H_DEST_PARM:
+       default:
+               dev_err(&vscsi->dev, "login_rej: error copying to client, rc %ld\n",
+                       rc);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+               break;
+       }
+
+       return rc;
+}
+
+static int ibmvscsis_make_nexus(struct ibmvscsis_tport *tport)
+{
+       char *name = tport->tport_name;
+       struct ibmvscsis_nexus *nexus;
+       int rc;
+
+       if (tport->ibmv_nexus) {
+               pr_debug("tport->ibmv_nexus already exists\n");
+               return 0;
+       }
+
+       nexus = kzalloc(sizeof(*nexus), GFP_KERNEL);
+       if (!nexus) {
+               pr_err("Unable to allocate struct ibmvscsis_nexus\n");
+               return -ENOMEM;
+       }
+
+       nexus->se_sess = target_alloc_session(&tport->se_tpg, 0, 0,
+                                             TARGET_PROT_NORMAL, name, nexus,
+                                             NULL);
+       if (IS_ERR(nexus->se_sess)) {
+               rc = PTR_ERR(nexus->se_sess);
+               goto transport_init_fail;
+       }
+
+       tport->ibmv_nexus = nexus;
+
+       return 0;
+
+transport_init_fail:
+       kfree(nexus);
+       return rc;
+}
+
+static int ibmvscsis_drop_nexus(struct ibmvscsis_tport *tport)
+{
+       struct se_session *se_sess;
+       struct ibmvscsis_nexus *nexus;
+
+       nexus = tport->ibmv_nexus;
+       if (!nexus)
+               return -ENODEV;
+
+       se_sess = nexus->se_sess;
+       if (!se_sess)
+               return -ENODEV;
+
+       /*
+        * Release the SCSI I_T Nexus to the emulated ibmvscsis Target Port
+        */
+       transport_deregister_session(se_sess);
+       tport->ibmv_nexus = NULL;
+       kfree(nexus);
+
+       return 0;
+}
+
+/**
+ * ibmvscsis_srp_login() - Process an SRP Login Request
+ * @vscsi:     Pointer to our adapter structure
+ * @cmd:       Command element to use to process the SRP Login request
+ * @crq:       Pointer to CRQ entry containing the SRP Login request
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt, called with interrupt lock held
+ */
+static long ibmvscsis_srp_login(struct scsi_info *vscsi,
+                               struct ibmvscsis_cmd *cmd,
+                               struct viosrp_crq *crq)
+{
+       struct iu_entry *iue = cmd->iue;
+       struct srp_login_req *req = &vio_iu(iue)->srp.login_req;
+       struct port_id {
+               __be64 id_extension;
+               __be64 io_guid;
+       } *iport, *tport;
+       struct format_code *fmt;
+       u32 reason = 0x0;
+       long rc = ADAPT_SUCCESS;
+
+       iport = (struct port_id *)req->initiator_port_id;
+       tport = (struct port_id *)req->target_port_id;
+       fmt = (struct format_code *)&req->req_buf_fmt;
+       if (be32_to_cpu(req->req_it_iu_len) > SRP_MAX_IU_LEN)
+               reason = SRP_LOGIN_REJ_REQ_IT_IU_LENGTH_TOO_LARGE;
+       else if (be32_to_cpu(req->req_it_iu_len) < 64)
+               reason = SRP_LOGIN_REJ_UNABLE_ESTABLISH_CHANNEL;
+       else if ((be64_to_cpu(iport->id_extension) > (MAX_NUM_PORTS - 1)) ||
+                (be64_to_cpu(tport->id_extension) > (MAX_NUM_PORTS - 1)))
+               reason = SRP_LOGIN_REJ_UNABLE_ASSOCIATE_CHANNEL;
+       else if (req->req_flags & SRP_MULTICHAN_MULTI)
+               reason = SRP_LOGIN_REJ_MULTI_CHANNEL_UNSUPPORTED;
+       else if (fmt->buffers & (~SUPPORTED_FORMATS))
+               reason = SRP_LOGIN_REJ_UNSUPPORTED_DESCRIPTOR_FMT;
+       else if ((fmt->buffers | SUPPORTED_FORMATS) == 0)
+               reason = SRP_LOGIN_REJ_UNSUPPORTED_DESCRIPTOR_FMT;
+
+       if (vscsi->state == SRP_PROCESSING)
+               reason = SRP_LOGIN_REJ_CHANNEL_LIMIT_REACHED;
+
+       rc = ibmvscsis_make_nexus(&vscsi->tport);
+       if (rc)
+               reason = SRP_LOGIN_REJ_UNABLE_ESTABLISH_CHANNEL;
+
+       cmd->rsp.format = VIOSRP_SRP_FORMAT;
+       cmd->rsp.tag = req->tag;
+
+       pr_debug("srp_login: reason 0x%x\n", reason);
+
+       if (reason)
+               rc = ibmvscsis_srp_login_rej(vscsi, cmd, reason);
+       else
+               rc = ibmvscsis_login_rsp(vscsi, cmd);
+
+       if (!rc) {
+               if (!reason)
+                       vscsi->state = SRP_PROCESSING;
+
+               list_add_tail(&cmd->list, &vscsi->waiting_rsp);
+               ibmvscsis_send_messages(vscsi);
+       } else {
+               ibmvscsis_free_cmd_resources(vscsi, cmd);
+       }
+
+       pr_debug("Leaving srp_login, rc %ld\n", rc);
+       return rc;
+}
+
+/**
+ * ibmvscsis_srp_i_logout() - Helper Function to close I_T Nexus
+ * @vscsi:     Pointer to our adapter structure
+ * @cmd:       Command element to use to process the Implicit Logout request
+ * @crq:       Pointer to CRQ entry containing the Implicit Logout request
+ *
+ * Do the logic to close the I_T nexus.  This function may not
+ * behave to specification.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt, interrupt lock held
+ */
+static long ibmvscsis_srp_i_logout(struct scsi_info *vscsi,
+                                  struct ibmvscsis_cmd *cmd,
+                                  struct viosrp_crq *crq)
+{
+       struct iu_entry *iue = cmd->iue;
+       struct srp_i_logout *log_out = &vio_iu(iue)->srp.i_logout;
+       long rc = ADAPT_SUCCESS;
+
+       if ((vscsi->debit > 0) || !list_empty(&vscsi->schedule_q) ||
+           !list_empty(&vscsi->waiting_rsp)) {
+               dev_err(&vscsi->dev, "i_logout: outstanding work\n");
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT, 0);
+       } else {
+               cmd->rsp.format = SRP_FORMAT;
+               cmd->rsp.tag = log_out->tag;
+               cmd->rsp.len = sizeof(struct mad_common);
+               list_add_tail(&cmd->list, &vscsi->waiting_rsp);
+               ibmvscsis_send_messages(vscsi);
+
+               ibmvscsis_post_disconnect(vscsi, WAIT_IDLE, 0);
+       }
+
+       return rc;
+}
+
+/* Called with intr lock held */
+static void ibmvscsis_srp_cmd(struct scsi_info *vscsi, struct viosrp_crq *crq)
+{
+       struct ibmvscsis_cmd *cmd;
+       struct iu_entry *iue;
+       struct srp_cmd *srp;
+       struct srp_tsk_mgmt *tsk;
+       long rc;
+
+       if (vscsi->request_limit - vscsi->debit <= 0) {
+               /* Client has exceeded request limit */
+               dev_err(&vscsi->dev, "Client exceeded the request limit (%d), debit %d\n",
+                       vscsi->request_limit, vscsi->debit);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+               return;
+       }
+
+       cmd = ibmvscsis_get_free_cmd(vscsi);
+       if (!cmd) {
+               dev_err(&vscsi->dev, "srp_cmd failed to get cmd, debit %d\n",
+                       vscsi->debit);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+               return;
+       }
+       iue = cmd->iue;
+       srp = &vio_iu(iue)->srp.cmd;
+
+       rc = ibmvscsis_copy_crq_packet(vscsi, cmd, crq);
+       if (rc) {
+               ibmvscsis_free_cmd_resources(vscsi, cmd);
+               return;
+       }
+
+       if (vscsi->state == SRP_PROCESSING) {
+               switch (srp->opcode) {
+               case SRP_LOGIN_REQ:
+                       rc = ibmvscsis_srp_login(vscsi, cmd, crq);
+                       break;
+
+               case SRP_TSK_MGMT:
+                       tsk = &vio_iu(iue)->srp.tsk_mgmt;
+                       pr_debug("tsk_mgmt tag: %llu (0x%llx)\n", tsk->tag,
+                                tsk->tag);
+                       cmd->rsp.tag = tsk->tag;
+                       vscsi->debit += 1;
+                       cmd->type = TASK_MANAGEMENT;
+                       list_add_tail(&cmd->list, &vscsi->schedule_q);
+                       queue_work(vscsi->work_q, &cmd->work);
+                       break;
+
+               case SRP_CMD:
+                       pr_debug("srp_cmd tag: %llu (0x%llx)\n", srp->tag,
+                                srp->tag);
+                       cmd->rsp.tag = srp->tag;
+                       vscsi->debit += 1;
+                       cmd->type = SCSI_CDB;
+                       /*
+                        * We want to keep track of work waiting for
+                        * the workqueue.
+                        */
+                       list_add_tail(&cmd->list, &vscsi->schedule_q);
+                       queue_work(vscsi->work_q, &cmd->work);
+                       break;
+
+               case SRP_I_LOGOUT:
+                       rc = ibmvscsis_srp_i_logout(vscsi, cmd, crq);
+                       break;
+
+               case SRP_CRED_RSP:
+               case SRP_AER_RSP:
+               default:
+                       ibmvscsis_free_cmd_resources(vscsi, cmd);
+                       dev_err(&vscsi->dev, "invalid srp cmd, opcode %d\n",
+                               (uint)srp->opcode);
+                       ibmvscsis_post_disconnect(vscsi,
+                                                 ERR_DISCONNECT_RECONNECT, 0);
+                       break;
+               }
+       } else if (srp->opcode == SRP_LOGIN_REQ && vscsi->state == CONNECTED) {
+               rc = ibmvscsis_srp_login(vscsi, cmd, crq);
+       } else {
+               ibmvscsis_free_cmd_resources(vscsi, cmd);
+               dev_err(&vscsi->dev, "Invalid state %d to handle srp cmd\n",
+                       vscsi->state);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+       }
+}
+
+/**
+ * ibmvscsis_ping_response() - Respond to a ping request
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * Let the client know that the server is alive and waiting on
+ * its native I/O stack.
+ * If any type of error occurs from the call to queue a ping
+ * response then the client is either not accepting or receiving
+ * interrupts.  Disconnect with an error.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt, interrupt lock held
+ */
+static long ibmvscsis_ping_response(struct scsi_info *vscsi)
+{
+       struct viosrp_crq *crq;
+       u64 buffer[2] = { 0, 0 };
+       long rc;
+
+       crq = (struct viosrp_crq *)&buffer;
+       crq->valid = VALID_CMD_RESP_EL;
+       crq->format = (u8)MESSAGE_IN_CRQ;
+       crq->status = PING_RESPONSE;
+
+       rc = h_send_crq(vscsi->dds.unit_id, cpu_to_be64(buffer[MSG_HI]),
+                       cpu_to_be64(buffer[MSG_LOW]));
+
+       switch (rc) {
+       case H_SUCCESS:
+               break;
+       case H_CLOSED:
+               vscsi->flags |= CLIENT_FAILED;
+       case H_DROPPED:
+               vscsi->flags |= RESPONSE_Q_DOWN;
+       case H_REMOTE_PARM:
+               dev_err(&vscsi->dev, "ping_response: h_send_crq failed, rc %ld\n",
+                       rc);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+               break;
+       default:
+               dev_err(&vscsi->dev, "ping_response: h_send_crq returned unknown rc %ld\n",
+                       rc);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT, 0);
+               break;
+       }
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_handle_init_compl_msg() - Respond to an Init Complete Message
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * Must be called with interrupt lock held.
+ */
+static long ibmvscsis_handle_init_compl_msg(struct scsi_info *vscsi)
+{
+       long rc = ADAPT_SUCCESS;
+
+       switch (vscsi->state) {
+       case NO_QUEUE:
+       case ERR_DISCONNECT:
+       case ERR_DISCONNECT_RECONNECT:
+       case ERR_DISCONNECTED:
+       case UNCONFIGURING:
+       case UNDEFINED:
+               rc = ERROR;
+               break;
+
+       case WAIT_CONNECTION:
+               vscsi->state = CONNECTED;
+               break;
+
+       case WAIT_IDLE:
+       case SRP_PROCESSING:
+       case CONNECTED:
+       case WAIT_ENABLED:
+       case PART_UP_WAIT_ENAB:
+       default:
+               rc = ERROR;
+               dev_err(&vscsi->dev, "init_msg: invalid state %d to get init compl msg\n",
+                       vscsi->state);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+               break;
+       }
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_handle_init_msg() - Respond to an Init Message
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * Must be called with interrupt lock held.
+ */
+static long ibmvscsis_handle_init_msg(struct scsi_info *vscsi)
+{
+       long rc = ADAPT_SUCCESS;
+
+       switch (vscsi->state) {
+       case WAIT_ENABLED:
+               vscsi->state = PART_UP_WAIT_ENAB;
+               break;
+
+       case WAIT_CONNECTION:
+               rc = ibmvscsis_send_init_message(vscsi, INIT_COMPLETE_MSG);
+               switch (rc) {
+               case H_SUCCESS:
+                       vscsi->state = CONNECTED;
+                       break;
+
+               case H_PARAMETER:
+                       dev_err(&vscsi->dev, "init_msg: failed to send, rc %ld\n",
+                               rc);
+                       ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT, 0);
+                       break;
+
+               case H_DROPPED:
+                       dev_err(&vscsi->dev, "init_msg: failed to send, rc %ld\n",
+                               rc);
+                       rc = ERROR;
+                       ibmvscsis_post_disconnect(vscsi,
+                                                 ERR_DISCONNECT_RECONNECT, 0);
+                       break;
+
+               case H_CLOSED:
+                       pr_warn("init_msg: failed to send, rc %ld\n", rc);
+                       rc = 0;
+                       break;
+               }
+               break;
+
+       case UNDEFINED:
+               rc = ERROR;
+               break;
+
+       case UNCONFIGURING:
+               break;
+
+       case PART_UP_WAIT_ENAB:
+       case CONNECTED:
+       case SRP_PROCESSING:
+       case WAIT_IDLE:
+       case NO_QUEUE:
+       case ERR_DISCONNECT:
+       case ERR_DISCONNECT_RECONNECT:
+       case ERR_DISCONNECTED:
+       default:
+               rc = ERROR;
+               dev_err(&vscsi->dev, "init_msg: invalid state %d to get init msg\n",
+                       vscsi->state);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+               break;
+       }
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_init_msg() - Respond to an init message
+ * @vscsi:     Pointer to our adapter structure
+ * @crq:       Pointer to CRQ element containing the Init Message
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt, interrupt lock held
+ */
+static long ibmvscsis_init_msg(struct scsi_info *vscsi, struct viosrp_crq *crq)
+{
+       long rc = ADAPT_SUCCESS;
+
+       pr_debug("init_msg: state 0x%hx\n", vscsi->state);
+
+       rc = h_vioctl(vscsi->dds.unit_id, H_GET_PARTNER_INFO,
+                     (u64)vscsi->map_ioba | ((u64)PAGE_SIZE << 32), 0, 0, 0,
+                     0);
+       if (rc == H_SUCCESS) {
+               vscsi->client_data.partition_number =
+                       be64_to_cpu(*(u64 *)vscsi->map_buf);
+               pr_debug("init_msg, part num %d\n",
+                        vscsi->client_data.partition_number);
+       } else {
+               pr_debug("init_msg h_vioctl rc %ld\n", rc);
+               rc = ADAPT_SUCCESS;
+       }
+
+       if (crq->format == INIT_MSG) {
+               rc = ibmvscsis_handle_init_msg(vscsi);
+       } else if (crq->format == INIT_COMPLETE_MSG) {
+               rc = ibmvscsis_handle_init_compl_msg(vscsi);
+       } else {
+               rc = ERROR;
+               dev_err(&vscsi->dev, "init_msg: invalid format %d\n",
+                       (uint)crq->format);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+       }
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_parse_command() - Parse an element taken from the cmd rsp queue.
+ * @vscsi:     Pointer to our adapter structure
+ * @crq:       Pointer to CRQ element containing the SRP request
+ *
+ * This function will return success if the command queue element is valid
+ * and the srp iu or MAD request it pointed to was also valid.  That does
+ * not mean that an error was not returned to the client.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Interrupt, intr lock held
+ */
+static long ibmvscsis_parse_command(struct scsi_info *vscsi,
+                                   struct viosrp_crq *crq)
+{
+       long rc = ADAPT_SUCCESS;
+
+       switch (crq->valid) {
+       case VALID_CMD_RESP_EL:
+               switch (crq->format) {
+               case OS400_FORMAT:
+               case AIX_FORMAT:
+               case LINUX_FORMAT:
+               case MAD_FORMAT:
+                       if (vscsi->flags & PROCESSING_MAD) {
+                               rc = ERROR;
+                               dev_err(&vscsi->dev, "parse_command: already processing mad\n");
+                               ibmvscsis_post_disconnect(vscsi,
+                                                      ERR_DISCONNECT_RECONNECT,
+                                                      0);
+                       } else {
+                               vscsi->flags |= PROCESSING_MAD;
+                               rc = ibmvscsis_mad(vscsi, crq);
+                       }
+                       break;
+
+               case SRP_FORMAT:
+                       ibmvscsis_srp_cmd(vscsi, crq);
+                       break;
+
+               case MESSAGE_IN_CRQ:
+                       if (crq->status == PING)
+                               ibmvscsis_ping_response(vscsi);
+                       break;
+
+               default:
+                       dev_err(&vscsi->dev, "parse_command: invalid format %d\n",
+                               (uint)crq->format);
+                       ibmvscsis_post_disconnect(vscsi,
+                                                 ERR_DISCONNECT_RECONNECT, 0);
+                       break;
+               }
+               break;
+
+       case VALID_TRANS_EVENT:
+               rc =  ibmvscsis_trans_event(vscsi, crq);
+               break;
+
+       case VALID_INIT_MSG:
+               rc = ibmvscsis_init_msg(vscsi, crq);
+               break;
+
+       default:
+               dev_err(&vscsi->dev, "parse_command: invalid valid field %d\n",
+                       (uint)crq->valid);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+               break;
+       }
+
+       /*
+        * Return only what the interrupt handler cares
+        * about. Most errors we keep right on trucking.
+        */
+       rc = vscsi->flags & SCHEDULE_DISCONNECT;
+
+       return rc;
+}
+
+static int read_dma_window(struct scsi_info *vscsi)
+{
+       struct vio_dev *vdev = vscsi->dma_dev;
+       const __be32 *dma_window;
+       const __be32 *prop;
+
+       /* TODO Using of_parse_dma_window would be better, but it doesn't give
+        * a way to read multiple windows without already knowing the size of
+        * a window or the number of windows.
+        */
+       dma_window = (const __be32 *)vio_get_attribute(vdev,
+                                                      "ibm,my-dma-window",
+                                                      NULL);
+       if (!dma_window) {
+               pr_err("Couldn't find ibm,my-dma-window property\n");
+               return -1;
+       }
+
+       vscsi->dds.window[LOCAL].liobn = be32_to_cpu(*dma_window);
+       dma_window++;
+
+       prop = (const __be32 *)vio_get_attribute(vdev, "ibm,#dma-address-cells",
+                                                NULL);
+       if (!prop) {
+               pr_warn("Couldn't find ibm,#dma-address-cells property\n");
+               dma_window++;
+       } else {
+               dma_window += be32_to_cpu(*prop);
+       }
+
+       prop = (const __be32 *)vio_get_attribute(vdev, "ibm,#dma-size-cells",
+                                                NULL);
+       if (!prop) {
+               pr_warn("Couldn't find ibm,#dma-size-cells property\n");
+               dma_window++;
+       } else {
+               dma_window += be32_to_cpu(*prop);
+       }
+
+       /* dma_window should point to the second window now */
+       vscsi->dds.window[REMOTE].liobn = be32_to_cpu(*dma_window);
+
+       return 0;
+}
+
+static struct ibmvscsis_tport *ibmvscsis_lookup_port(const char *name)
+{
+       struct ibmvscsis_tport *tport = NULL;
+       struct vio_dev *vdev;
+       struct scsi_info *vscsi;
+
+       spin_lock_bh(&ibmvscsis_dev_lock);
+       list_for_each_entry(vscsi, &ibmvscsis_dev_list, list) {
+               vdev = vscsi->dma_dev;
+               if (!strcmp(dev_name(&vdev->dev), name)) {
+                       tport = &vscsi->tport;
+                       break;
+               }
+       }
+       spin_unlock_bh(&ibmvscsis_dev_lock);
+
+       return tport;
+}
+
+/**
+ * ibmvscsis_parse_cmd() - Parse SRP Command
+ * @vscsi:     Pointer to our adapter structure
+ * @cmd:       Pointer to command element with SRP command
+ *
+ * Parse the srp command; if it is valid then submit it to tcm.
+ * Note: The return code does not reflect the status of the SCSI CDB.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Process level
+ */
+static void ibmvscsis_parse_cmd(struct scsi_info *vscsi,
+                               struct ibmvscsis_cmd *cmd)
+{
+       struct iu_entry *iue = cmd->iue;
+       struct srp_cmd *srp = (struct srp_cmd *)iue->sbuf->buf;
+       struct ibmvscsis_nexus *nexus;
+       u64 data_len = 0;
+       enum dma_data_direction dir;
+       int attr = 0;
+       int rc = 0;
+
+       nexus = vscsi->tport.ibmv_nexus;
+       /*
+        * additional length in bytes.  Note that the SRP spec says that
+        * additional length is in 4-byte words, but technically the
+        * additional length field is only the upper 6 bits of the byte.
+        * The lower 2 bits are reserved.  If the lower 2 bits are 0 (as
+        * all reserved fields should be), then interpreting the byte as
+        * an int will yield the length in bytes.
+        */
+       if (srp->add_cdb_len & 0x03) {
+               dev_err(&vscsi->dev, "parse_cmd: reserved bits set in IU\n");
+               spin_lock_bh(&vscsi->intr_lock);
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+               ibmvscsis_free_cmd_resources(vscsi, cmd);
+               spin_unlock_bh(&vscsi->intr_lock);
+               return;
+       }
+
+       if (srp_get_desc_table(srp, &dir, &data_len)) {
+               dev_err(&vscsi->dev, "0x%llx: parsing SRP descriptor table failed.\n",
+                       srp->tag);
+               goto fail;
+               return;
+       }
+
+       cmd->rsp.sol_not = srp->sol_not;
+
+       switch (srp->task_attr) {
+       case SRP_SIMPLE_TASK:
+               attr = TCM_SIMPLE_TAG;
+               break;
+       case SRP_ORDERED_TASK:
+               attr = TCM_ORDERED_TAG;
+               break;
+       case SRP_HEAD_TASK:
+               attr = TCM_HEAD_TAG;
+               break;
+       case SRP_ACA_TASK:
+               attr = TCM_ACA_TAG;
+               break;
+       default:
+               dev_err(&vscsi->dev, "Invalid task attribute %d\n",
+                       srp->task_attr);
+               goto fail;
+       }
+
+       cmd->se_cmd.tag = be64_to_cpu(srp->tag);
+
+       spin_lock_bh(&vscsi->intr_lock);
+       list_add_tail(&cmd->list, &vscsi->active_q);
+       spin_unlock_bh(&vscsi->intr_lock);
+
+       srp->lun.scsi_lun[0] &= 0x3f;
+
+       pr_debug("calling submit_cmd, se_cmd %p, lun 0x%llx, cdb 0x%x, attr:%d\n",
+                &cmd->se_cmd, scsilun_to_int(&srp->lun), (int)srp->cdb[0],
+                attr);
+
+       rc = target_submit_cmd(&cmd->se_cmd, nexus->se_sess, srp->cdb,
+                              cmd->sense_buf, scsilun_to_int(&srp->lun),
+                              data_len, attr, dir, 0);
+       if (rc) {
+               dev_err(&vscsi->dev, "target_submit_cmd failed, rc %d\n", rc);
+               goto fail;
+       }
+       return;
+
+fail:
+       spin_lock_bh(&vscsi->intr_lock);
+       ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+       spin_unlock_bh(&vscsi->intr_lock);
+}
+
+/**
+ * ibmvscsis_parse_task() - Parse SRP Task Management Request
+ * @vscsi:     Pointer to our adapter structure
+ * @cmd:       Pointer to command element with SRP task management request
+ *
+ * Parse the srp task management request; if it is valid then submit it to tcm.
+ * Note: The return code does not reflect the status of the task management
+ * request.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Processor level
+ */
+static void ibmvscsis_parse_task(struct scsi_info *vscsi,
+                                struct ibmvscsis_cmd *cmd)
+{
+       struct iu_entry *iue = cmd->iue;
+       struct srp_tsk_mgmt *srp_tsk = &vio_iu(iue)->srp.tsk_mgmt;
+       int tcm_type;
+       u64 tag_to_abort = 0;
+       int rc = 0;
+       struct ibmvscsis_nexus *nexus;
+
+       nexus = vscsi->tport.ibmv_nexus;
+
+       cmd->rsp.sol_not = srp_tsk->sol_not;
+
+       switch (srp_tsk->tsk_mgmt_func) {
+       case SRP_TSK_ABORT_TASK:
+               tcm_type = TMR_ABORT_TASK;
+               tag_to_abort = be64_to_cpu(srp_tsk->task_tag);
+               break;
+       case SRP_TSK_ABORT_TASK_SET:
+               tcm_type = TMR_ABORT_TASK_SET;
+               break;
+       case SRP_TSK_CLEAR_TASK_SET:
+               tcm_type = TMR_CLEAR_TASK_SET;
+               break;
+       case SRP_TSK_LUN_RESET:
+               tcm_type = TMR_LUN_RESET;
+               break;
+       case SRP_TSK_CLEAR_ACA:
+               tcm_type = TMR_CLEAR_ACA;
+               break;
+       default:
+               dev_err(&vscsi->dev, "unknown task mgmt func %d\n",
+                       srp_tsk->tsk_mgmt_func);
+               cmd->se_cmd.se_tmr_req->response =
+                       TMR_TASK_MGMT_FUNCTION_NOT_SUPPORTED;
+               rc = -1;
+               break;
+       }
+
+       if (!rc) {
+               cmd->se_cmd.tag = be64_to_cpu(srp_tsk->tag);
+
+               spin_lock_bh(&vscsi->intr_lock);
+               list_add_tail(&cmd->list, &vscsi->active_q);
+               spin_unlock_bh(&vscsi->intr_lock);
+
+               srp_tsk->lun.scsi_lun[0] &= 0x3f;
+
+               pr_debug("calling submit_tmr, func %d\n",
+                        srp_tsk->tsk_mgmt_func);
+               rc = target_submit_tmr(&cmd->se_cmd, nexus->se_sess, NULL,
+                                      scsilun_to_int(&srp_tsk->lun), srp_tsk,
+                                      tcm_type, GFP_KERNEL, tag_to_abort, 0);
+               if (rc) {
+                       dev_err(&vscsi->dev, "target_submit_tmr failed, rc %d\n",
+                               rc);
+                       cmd->se_cmd.se_tmr_req->response =
+                               TMR_FUNCTION_REJECTED;
+               }
+       }
+
+       if (rc)
+               transport_send_check_condition_and_sense(&cmd->se_cmd, 0, 0);
+}
+
+static void ibmvscsis_scheduler(struct work_struct *work)
+{
+       struct ibmvscsis_cmd *cmd = container_of(work, struct ibmvscsis_cmd,
+                                                work);
+       struct scsi_info *vscsi = cmd->adapter;
+
+       spin_lock_bh(&vscsi->intr_lock);
+
+       /* Remove from schedule_q */
+       list_del(&cmd->list);
+
+       /* Don't submit cmd if we're disconnecting */
+       if (vscsi->flags & (SCHEDULE_DISCONNECT | DISCONNECT_SCHEDULED)) {
+               ibmvscsis_free_cmd_resources(vscsi, cmd);
+
+               /* ibmvscsis_disconnect might be waiting for us */
+               if (list_empty(&vscsi->active_q) &&
+                   list_empty(&vscsi->schedule_q) &&
+                   (vscsi->flags & WAIT_FOR_IDLE)) {
+                       vscsi->flags &= ~WAIT_FOR_IDLE;
+                       complete(&vscsi->wait_idle);
+               }
+
+               spin_unlock_bh(&vscsi->intr_lock);
+               return;
+       }
+
+       spin_unlock_bh(&vscsi->intr_lock);
+
+       switch (cmd->type) {
+       case SCSI_CDB:
+               ibmvscsis_parse_cmd(vscsi, cmd);
+               break;
+       case TASK_MANAGEMENT:
+               ibmvscsis_parse_task(vscsi, cmd);
+               break;
+       default:
+               dev_err(&vscsi->dev, "scheduler, invalid cmd type %d\n",
+                       cmd->type);
+               spin_lock_bh(&vscsi->intr_lock);
+               ibmvscsis_free_cmd_resources(vscsi, cmd);
+               spin_unlock_bh(&vscsi->intr_lock);
+               break;
+       }
+}
+
+static int ibmvscsis_alloc_cmds(struct scsi_info *vscsi, int num)
+{
+       struct ibmvscsis_cmd *cmd;
+       int i;
+
+       INIT_LIST_HEAD(&vscsi->free_cmd);
+       vscsi->cmd_pool = kcalloc(num, sizeof(struct ibmvscsis_cmd),
+                                 GFP_KERNEL);
+       if (!vscsi->cmd_pool)
+               return -ENOMEM;
+
+       for (i = 0, cmd = (struct ibmvscsis_cmd *)vscsi->cmd_pool; i < num;
+            i++, cmd++) {
+               cmd->adapter = vscsi;
+               INIT_WORK(&cmd->work, ibmvscsis_scheduler);
+               list_add_tail(&cmd->list, &vscsi->free_cmd);
+       }
+
+       return 0;
+}
+
+static void ibmvscsis_free_cmds(struct scsi_info *vscsi)
+{
+       kfree(vscsi->cmd_pool);
+       vscsi->cmd_pool = NULL;
+       INIT_LIST_HEAD(&vscsi->free_cmd);
+}
+
+/**
+ * ibmvscsis_service_wait_q() - Service Waiting Queue
+ * @timer:     Pointer to timer which has expired
+ *
+ * This routine is called when the timer pops to service the waiting
+ * queue. Elements on the queue have completed, their responses have been
+ * copied to the client, but the client's response queue was full so
+ * the queue message could not be sent. The routine grabs the proper locks
+ * and calls send messages.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     called at interrupt level
+ */
+static enum hrtimer_restart ibmvscsis_service_wait_q(struct hrtimer *timer)
+{
+       struct timer_cb *p_timer = container_of(timer, struct timer_cb, timer);
+       struct scsi_info *vscsi = container_of(p_timer, struct scsi_info,
+                                              rsp_q_timer);
+
+       spin_lock_bh(&vscsi->intr_lock);
+       p_timer->timer_pops += 1;
+       p_timer->started = false;
+       ibmvscsis_send_messages(vscsi);
+       spin_unlock_bh(&vscsi->intr_lock);
+
+       return HRTIMER_NORESTART;
+}
+
+static long ibmvscsis_alloctimer(struct scsi_info *vscsi)
+{
+       struct timer_cb *p_timer;
+
+       p_timer = &vscsi->rsp_q_timer;
+       hrtimer_init(&p_timer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+
+       p_timer->timer.function = ibmvscsis_service_wait_q;
+       p_timer->started = false;
+       p_timer->timer_pops = 0;
+
+       return ADAPT_SUCCESS;
+}
+
+static void ibmvscsis_freetimer(struct scsi_info *vscsi)
+{
+       struct timer_cb *p_timer;
+
+       p_timer = &vscsi->rsp_q_timer;
+
+       (void)hrtimer_cancel(&p_timer->timer);
+
+       p_timer->started = false;
+       p_timer->timer_pops = 0;
+}
+
+static irqreturn_t ibmvscsis_interrupt(int dummy, void *data)
+{
+       struct scsi_info *vscsi = data;
+
+       vio_disable_interrupts(vscsi->dma_dev);
+       tasklet_schedule(&vscsi->work_task);
+
+       return IRQ_HANDLED;
+}
+
+/**
+ * ibmvscsis_check_q() - Helper function to Check Init Message Valid
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * Checks if a initialize message was queued by the initiatior
+ * while the timing window was open.  This function is called from
+ * probe after the CRQ is created and interrupts are enabled.
+ * It would only be used by adapters who wait for some event before
+ * completing the init handshake with the client.  For ibmvscsi, this
+ * event is waiting for the port to be enabled.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Process level only, interrupt lock held
+ */
+static long ibmvscsis_check_q(struct scsi_info *vscsi)
+{
+       uint format;
+       long rc;
+
+       rc = ibmvscsis_check_init_msg(vscsi, &format);
+       if (rc)
+               ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT_RECONNECT, 0);
+       else if (format == UNUSED_FORMAT)
+               vscsi->state = WAIT_ENABLED;
+       else
+               vscsi->state = PART_UP_WAIT_ENAB;
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_enable_change_state() - Set new state based on enabled status
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * This function determines our new state now that we are enabled.  This
+ * may involve sending an Init Complete message to the client.
+ *
+ * Must be called with interrupt lock held.
+ */
+static long ibmvscsis_enable_change_state(struct scsi_info *vscsi)
+{
+       long rc = ADAPT_SUCCESS;
+
+handle_state_change:
+       switch (vscsi->state) {
+       case WAIT_ENABLED:
+               rc = ibmvscsis_send_init_message(vscsi, INIT_MSG);
+               switch (rc) {
+               case H_SUCCESS:
+               case H_DROPPED:
+               case H_CLOSED:
+                       vscsi->state =  WAIT_CONNECTION;
+                       rc = ADAPT_SUCCESS;
+                       break;
+
+               case H_PARAMETER:
+                       break;
+
+               case H_HARDWARE:
+                       break;
+
+               default:
+                       vscsi->state = UNDEFINED;
+                       rc = H_HARDWARE;
+                       break;
+               }
+               break;
+       case PART_UP_WAIT_ENAB:
+               rc = ibmvscsis_send_init_message(vscsi, INIT_COMPLETE_MSG);
+               switch (rc) {
+               case H_SUCCESS:
+                       vscsi->state = CONNECTED;
+                       rc = ADAPT_SUCCESS;
+                       break;
+
+               case H_DROPPED:
+               case H_CLOSED:
+                       vscsi->state = WAIT_ENABLED;
+                       goto handle_state_change;
+
+               case H_PARAMETER:
+                       break;
+
+               case H_HARDWARE:
+                       break;
+
+               default:
+                       rc = H_HARDWARE;
+                       break;
+               }
+               break;
+
+       case WAIT_CONNECTION:
+       case WAIT_IDLE:
+       case SRP_PROCESSING:
+       case CONNECTED:
+               rc = ADAPT_SUCCESS;
+               break;
+               /* should not be able to get here */
+       case UNCONFIGURING:
+               rc = ERROR;
+               vscsi->state = UNDEFINED;
+               break;
+
+               /* driver should never allow this to happen */
+       case ERR_DISCONNECT:
+       case ERR_DISCONNECT_RECONNECT:
+       default:
+               dev_err(&vscsi->dev, "in invalid state %d during enable_change_state\n",
+                       vscsi->state);
+               rc = ADAPT_SUCCESS;
+               break;
+       }
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_create_command_q() - Create Command Queue
+ * @vscsi:     Pointer to our adapter structure
+ * @num_cmds:  Currently unused.  In the future, may be used to determine
+ *             the size of the CRQ.
+ *
+ * Allocates memory for command queue maps remote memory into an ioba
+ * initializes the command response queue
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Process level only
+ */
+static long ibmvscsis_create_command_q(struct scsi_info *vscsi, int num_cmds)
+{
+       long rc = 0;
+       int pages;
+       struct vio_dev *vdev = vscsi->dma_dev;
+
+       /* We might support multiple pages in the future, but just 1 for now */
+       pages = 1;
+
+       vscsi->cmd_q.size = pages;
+
+       vscsi->cmd_q.base_addr =
+               (struct viosrp_crq *)get_zeroed_page(GFP_KERNEL);
+       if (!vscsi->cmd_q.base_addr)
+               return -ENOMEM;
+
+       vscsi->cmd_q.mask = ((uint)pages * CRQ_PER_PAGE) - 1;
+
+       vscsi->cmd_q.crq_token = dma_map_single(&vdev->dev,
+                                               vscsi->cmd_q.base_addr,
+                                               PAGE_SIZE, DMA_BIDIRECTIONAL);
+       if (dma_mapping_error(&vdev->dev, vscsi->cmd_q.crq_token)) {
+               free_page((unsigned long)vscsi->cmd_q.base_addr);
+               return -ENOMEM;
+       }
+
+       rc =  h_reg_crq(vscsi->dds.unit_id, vscsi->cmd_q.crq_token, PAGE_SIZE);
+       if (rc) {
+               if (rc == H_CLOSED) {
+                       vscsi->state = WAIT_ENABLED;
+                       rc = 0;
+               } else {
+                       dma_unmap_single(&vdev->dev, vscsi->cmd_q.crq_token,
+                                        PAGE_SIZE, DMA_BIDIRECTIONAL);
+                       free_page((unsigned long)vscsi->cmd_q.base_addr);
+                       rc = -ENODEV;
+               }
+       } else {
+               vscsi->state = WAIT_ENABLED;
+       }
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_destroy_command_q - Destroy Command Queue
+ * @vscsi:     Pointer to our adapter structure
+ *
+ * Releases memory for command queue and unmaps mapped remote memory.
+ *
+ * EXECUTION ENVIRONMENT:
+ *     Process level only
+ */
+static void ibmvscsis_destroy_command_q(struct scsi_info *vscsi)
+{
+       dma_unmap_single(&vscsi->dma_dev->dev, vscsi->cmd_q.crq_token,
+                        PAGE_SIZE, DMA_BIDIRECTIONAL);
+       free_page((unsigned long)vscsi->cmd_q.base_addr);
+       vscsi->cmd_q.base_addr = NULL;
+       vscsi->state = NO_QUEUE;
+}
+
+static u8 ibmvscsis_fast_fail(struct scsi_info *vscsi,
+                             struct ibmvscsis_cmd *cmd)
+{
+       struct iu_entry *iue = cmd->iue;
+       struct se_cmd *se_cmd = &cmd->se_cmd;
+       struct srp_cmd *srp = (struct srp_cmd *)iue->sbuf->buf;
+       struct scsi_sense_hdr sshdr;
+       u8 rc = se_cmd->scsi_status;
+
+       if (vscsi->fast_fail && (READ_CMD(srp->cdb) || WRITE_CMD(srp->cdb)))
+               if (scsi_normalize_sense(se_cmd->sense_buffer,
+                                        se_cmd->scsi_sense_length, &sshdr))
+                       if (sshdr.sense_key == HARDWARE_ERROR &&
+                           (se_cmd->residual_count == 0 ||
+                            se_cmd->residual_count == se_cmd->data_length)) {
+                               rc = NO_SENSE;
+                               cmd->flags |= CMD_FAST_FAIL;
+                       }
+
+       return rc;
+}
+
+/**
+ * srp_build_response() - Build an SRP response buffer
+ * @vscsi:     Pointer to our adapter structure
+ * @cmd:       Pointer to command for which to send the response
+ * @len_p:     Where to return the length of the IU response sent.  This
+ *             is needed to construct the CRQ response.
+ *
+ * Build the SRP response buffer and copy it to the client's memory space.
+ */
+static long srp_build_response(struct scsi_info *vscsi,
+                              struct ibmvscsis_cmd *cmd, uint *len_p)
+{
+       struct iu_entry *iue = cmd->iue;
+       struct se_cmd *se_cmd = &cmd->se_cmd;
+       struct srp_rsp *rsp;
+       uint len;
+       u32 rsp_code;
+       char *data;
+       u32 *tsk_status;
+       long rc = ADAPT_SUCCESS;
+
+       spin_lock_bh(&vscsi->intr_lock);
+
+       rsp = &vio_iu(iue)->srp.rsp;
+       len = sizeof(*rsp);
+       memset(rsp, 0, len);
+       data = rsp->data;
+
+       rsp->opcode = SRP_RSP;
+
+       if (vscsi->credit > 0 && vscsi->state == SRP_PROCESSING)
+               rsp->req_lim_delta = cpu_to_be32(vscsi->credit);
+       else
+               rsp->req_lim_delta = cpu_to_be32(1 + vscsi->credit);
+       rsp->tag = cmd->rsp.tag;
+       rsp->flags = 0;
+
+       if (cmd->type == SCSI_CDB) {
+               rsp->status = ibmvscsis_fast_fail(vscsi, cmd);
+               if (rsp->status) {
+                       pr_debug("build_resp: cmd %p, scsi status %d\n", cmd,
+                                (int)rsp->status);
+                       ibmvscsis_determine_resid(se_cmd, rsp);
+                       if (se_cmd->scsi_sense_length && se_cmd->sense_buffer) {
+                               rsp->sense_data_len =
+                                       cpu_to_be32(se_cmd->scsi_sense_length);
+                               rsp->flags |= SRP_RSP_FLAG_SNSVALID;
+                               len += se_cmd->scsi_sense_length;
+                               memcpy(data, se_cmd->sense_buffer,
+                                      se_cmd->scsi_sense_length);
+                       }
+                       rsp->sol_not = (cmd->rsp.sol_not & UCSOLNT) >>
+                               UCSOLNT_RESP_SHIFT;
+               } else if (cmd->flags & CMD_FAST_FAIL) {
+                       pr_debug("build_resp: cmd %p, fast fail\n", cmd);
+                       rsp->sol_not = (cmd->rsp.sol_not & UCSOLNT) >>
+                               UCSOLNT_RESP_SHIFT;
+               } else {
+                       rsp->sol_not = (cmd->rsp.sol_not & SCSOLNT) >>
+                               SCSOLNT_RESP_SHIFT;
+               }
+       } else {
+               /* this is task management */
+               rsp->status = 0;
+               rsp->resp_data_len = cpu_to_be32(4);
+               rsp->flags |= SRP_RSP_FLAG_RSPVALID;
+
+               switch (se_cmd->se_tmr_req->response) {
+               case TMR_FUNCTION_COMPLETE:
+               case TMR_TASK_DOES_NOT_EXIST:
+                       rsp_code = SRP_TASK_MANAGEMENT_FUNCTION_COMPLETE;
+                       rsp->sol_not = (cmd->rsp.sol_not & SCSOLNT) >>
+                               SCSOLNT_RESP_SHIFT;
+                       break;
+               case TMR_TASK_MGMT_FUNCTION_NOT_SUPPORTED:
+               case TMR_LUN_DOES_NOT_EXIST:
+                       rsp_code = SRP_TASK_MANAGEMENT_FUNCTION_NOT_SUPPORTED;
+                       rsp->sol_not = (cmd->rsp.sol_not & UCSOLNT) >>
+                               UCSOLNT_RESP_SHIFT;
+                       break;
+               case TMR_FUNCTION_FAILED:
+               case TMR_FUNCTION_REJECTED:
+               default:
+                       rsp_code = SRP_TASK_MANAGEMENT_FUNCTION_FAILED;
+                       rsp->sol_not = (cmd->rsp.sol_not & UCSOLNT) >>
+                               UCSOLNT_RESP_SHIFT;
+                       break;
+               }
+
+               tsk_status = (u32 *)data;
+               *tsk_status = cpu_to_be32(rsp_code);
+               data = (char *)(tsk_status + 1);
+               len += 4;
+       }
+
+       dma_wmb();
+       rc = h_copy_rdma(len, vscsi->dds.window[LOCAL].liobn, iue->sbuf->dma,
+                        vscsi->dds.window[REMOTE].liobn,
+                        be64_to_cpu(iue->remote_token));
+
+       switch (rc) {
+       case H_SUCCESS:
+               vscsi->credit = 0;
+               *len_p = len;
+               break;
+       case H_PERMISSION:
+               if (connection_broken(vscsi))
+                       vscsi->flags |= RESPONSE_Q_DOWN | CLIENT_FAILED;
+
+               dev_err(&vscsi->dev, "build_response: error copying to client, rc %ld, flags 0x%x, state 0x%hx\n",
+                       rc, vscsi->flags, vscsi->state);
+               break;
+       case H_SOURCE_PARM:
+       case H_DEST_PARM:
+       default:
+               dev_err(&vscsi->dev, "build_response: error copying to client, rc %ld\n",
+                       rc);
+               break;
+       }
+
+       spin_unlock_bh(&vscsi->intr_lock);
+
+       return rc;
+}
+
+static int ibmvscsis_rdma(struct ibmvscsis_cmd *cmd, struct scatterlist *sg,
+                         int nsg, struct srp_direct_buf *md, int nmd,
+                         enum dma_data_direction dir, unsigned int bytes)
+{
+       struct iu_entry *iue = cmd->iue;
+       struct srp_target *target = iue->target;
+       struct scsi_info *vscsi = target->ldata;
+       struct scatterlist *sgp;
+       dma_addr_t client_ioba, server_ioba;
+       ulong buf_len;
+       ulong client_len, server_len;
+       int md_idx;
+       long tx_len;
+       long rc = 0;
+
+       pr_debug("rdma: dir %d, bytes 0x%x\n", dir, bytes);
+
+       if (bytes == 0)
+               return 0;
+
+       sgp = sg;
+       client_len = 0;
+       server_len = 0;
+       md_idx = 0;
+       tx_len = bytes;
+
+       do {
+               if (client_len == 0) {
+                       if (md_idx >= nmd) {
+                               dev_err(&vscsi->dev, "rdma: ran out of client memory descriptors\n");
+                               rc = -EIO;
+                               break;
+                       }
+                       client_ioba = be64_to_cpu(md[md_idx].va);
+                       client_len = be32_to_cpu(md[md_idx].len);
+               }
+               if (server_len == 0) {
+                       if (!sgp) {
+                               dev_err(&vscsi->dev, "rdma: ran out of scatter/gather list\n");
+                               rc = -EIO;
+                               break;
+                       }
+                       server_ioba = sg_dma_address(sgp);
+                       server_len = sg_dma_len(sgp);
+               }
+
+               buf_len = tx_len;
+
+               if (buf_len > client_len)
+                       buf_len = client_len;
+
+               if (buf_len > server_len)
+                       buf_len = server_len;
+
+               if (buf_len > max_vdma_size)
+                       buf_len = max_vdma_size;
+
+               if (dir == DMA_TO_DEVICE) {
+                       /* read from client */
+                       rc = h_copy_rdma(buf_len,
+                                        vscsi->dds.window[REMOTE].liobn,
+                                        client_ioba,
+                                        vscsi->dds.window[LOCAL].liobn,
+                                        server_ioba);
+               } else {
+                       /* write to client */
+                       struct srp_cmd *srp = (struct srp_cmd *)iue->sbuf->buf;
+
+                       if (!READ_CMD(srp->cdb))
+                               print_hex_dump_bytes(" data:", DUMP_PREFIX_NONE,
+                                                    sg_virt(sgp), buf_len);
+                       /* The h_copy_rdma will cause phyp, running in another
+                        * partition, to read memory, so we need to make sure
+                        * the data has been written out, hence these syncs.
+                        */
+                       /* ensure that everything is in memory */
+                       isync();
+                       /* ensure that memory has been made visible */
+                       dma_wmb();
+                       rc = h_copy_rdma(buf_len,
+                                        vscsi->dds.window[LOCAL].liobn,
+                                        server_ioba,
+                                        vscsi->dds.window[REMOTE].liobn,
+                                        client_ioba);
+               }
+               switch (rc) {
+               case H_SUCCESS:
+                       break;
+               case H_PERMISSION:
+               case H_SOURCE_PARM:
+               case H_DEST_PARM:
+                       if (connection_broken(vscsi)) {
+                               spin_lock_bh(&vscsi->intr_lock);
+                               vscsi->flags |=
+                                       (RESPONSE_Q_DOWN | CLIENT_FAILED);
+                               spin_unlock_bh(&vscsi->intr_lock);
+                       }
+                       dev_err(&vscsi->dev, "rdma: h_copy_rdma failed, rc %ld\n",
+                               rc);
+                       break;
+
+               default:
+                       dev_err(&vscsi->dev, "rdma: unknown error %ld from h_copy_rdma\n",
+                               rc);
+                       break;
+               }
+
+               if (!rc) {
+                       tx_len -= buf_len;
+                       if (tx_len) {
+                               client_len -= buf_len;
+                               if (client_len == 0)
+                                       md_idx++;
+                               else
+                                       client_ioba += buf_len;
+
+                               server_len -= buf_len;
+                               if (server_len == 0)
+                                       sgp = sg_next(sgp);
+                               else
+                                       server_ioba += buf_len;
+                       } else {
+                               break;
+                       }
+               }
+       } while (!rc);
+
+       return rc;
+}
+
+/**
+ * ibmvscsis_handle_crq() - Handle CRQ
+ * @data:      Pointer to our adapter structure
+ *
+ * Read the command elements from the command queue and copy the payloads
+ * associated with the command elements to local memory and execute the
+ * SRP requests.
+ *
+ * Note: this is an edge triggered interrupt. It can not be shared.
+ */
+static void ibmvscsis_handle_crq(unsigned long data)
+{
+       struct scsi_info *vscsi = (struct scsi_info *)data;
+       struct viosrp_crq *crq;
+       long rc;
+       bool ack = true;
+       volatile u8 valid;
+
+       spin_lock_bh(&vscsi->intr_lock);
+
+       pr_debug("got interrupt\n");
+
+       /*
+        * if we are in a path where we are waiting for all pending commands
+        * to complete because we received a transport event and anything in
+        * the command queue is for a new connection,  do nothing
+        */
+       if (TARGET_STOP(vscsi)) {
+               vio_enable_interrupts(vscsi->dma_dev);
+
+               pr_debug("handle_crq, don't process: flags 0x%x, state 0x%hx\n",
+                        vscsi->flags, vscsi->state);
+               spin_unlock_bh(&vscsi->intr_lock);
+               return;
+       }
+
+       rc = vscsi->flags & SCHEDULE_DISCONNECT;
+       crq = vscsi->cmd_q.base_addr + vscsi->cmd_q.index;
+       valid = crq->valid;
+       dma_rmb();
+
+       while (valid) {
+               /*
+                * These are edege triggered interrupts. After dropping out of
+                * the while loop, the code must check for work since an
+                * interrupt could be lost, and an elment be left on the queue,
+                * hence the label.
+                */
+cmd_work:
+               vscsi->cmd_q.index =
+                       (vscsi->cmd_q.index + 1) & vscsi->cmd_q.mask;
+
+               if (!rc) {
+                       rc = ibmvscsis_parse_command(vscsi, crq);
+               } else {
+                       if ((uint)crq->valid == VALID_TRANS_EVENT) {
+                               /*
+                                * must service the transport layer events even
+                                * in an error state, dont break out until all
+                                * the consecutive transport events have been
+                                * processed
+                                */
+                               rc = ibmvscsis_trans_event(vscsi, crq);
+                       } else if (vscsi->flags & TRANS_EVENT) {
+                               /*
+                                * if a tranport event has occurred leave
+                                * everything but transport events on the queue
+                                */
+                               pr_debug("handle_crq, ignoring\n");
+
+                               /*
+                                * need to decrement the queue index so we can
+                                * look at the elment again
+                                */
+                               if (vscsi->cmd_q.index)
+                                       vscsi->cmd_q.index -= 1;
+                               else
+                                       /*
+                                        * index is at 0 it just wrapped.
+                                        * have it index last element in q
+                                        */
+                                       vscsi->cmd_q.index = vscsi->cmd_q.mask;
+                               break;
+                       }
+               }
+
+               crq->valid = INVALIDATE_CMD_RESP_EL;
+
+               crq = vscsi->cmd_q.base_addr + vscsi->cmd_q.index;
+               valid = crq->valid;
+               dma_rmb();
+       }
+
+       if (!rc) {
+               if (ack) {
+                       vio_enable_interrupts(vscsi->dma_dev);
+                       ack = false;
+                       pr_debug("handle_crq, reenabling interrupts\n");
+               }
+               valid = crq->valid;
+               dma_rmb();
+               if (valid)
+                       goto cmd_work;
+       } else {
+               pr_debug("handle_crq, error: flags 0x%x, state 0x%hx, crq index 0x%x\n",
+                        vscsi->flags, vscsi->state, vscsi->cmd_q.index);
+       }
+
+       pr_debug("Leaving handle_crq: schedule_q empty %d, flags 0x%x, state 0x%hx\n",
+                (int)list_empty(&vscsi->schedule_q), vscsi->flags,
+                vscsi->state);
+
+       spin_unlock_bh(&vscsi->intr_lock);
+}
+
+static int ibmvscsis_probe(struct vio_dev *vdev,
+                          const struct vio_device_id *id)
+{
+       struct scsi_info *vscsi;
+       int rc = 0;
+       long hrc = 0;
+       char wq_name[24];
+
+       vscsi = kzalloc(sizeof(*vscsi), GFP_KERNEL);
+       if (!vscsi) {
+               rc = -ENOMEM;
+               pr_err("probe: allocation of adapter failed\n");
+               return rc;
+       }
+
+       vscsi->dma_dev = vdev;
+       vscsi->dev = vdev->dev;
+       INIT_LIST_HEAD(&vscsi->schedule_q);
+       INIT_LIST_HEAD(&vscsi->waiting_rsp);
+       INIT_LIST_HEAD(&vscsi->active_q);
+
+       snprintf(vscsi->tport.tport_name, 256, "%s", dev_name(&vdev->dev));
+
+       pr_debug("probe tport_name: %s\n", vscsi->tport.tport_name);
+
+       rc = read_dma_window(vscsi);
+       if (rc)
+               goto free_adapter;
+       pr_debug("Probe: liobn 0x%x, riobn 0x%x\n",
+                vscsi->dds.window[LOCAL].liobn,
+                vscsi->dds.window[REMOTE].liobn);
+
+       strcpy(vscsi->eye, "VSCSI ");
+       strncat(vscsi->eye, vdev->name, MAX_EYE);
+
+       vscsi->dds.unit_id = vdev->unit_address;
+
+       spin_lock_bh(&ibmvscsis_dev_lock);
+       list_add_tail(&vscsi->list, &ibmvscsis_dev_list);
+       spin_unlock_bh(&ibmvscsis_dev_lock);
+
+       /*
+        * TBD: How do we determine # of cmds to request?  Do we know how
+        * many "children" we have?
+        */
+       vscsi->request_limit = INITIAL_SRP_LIMIT;
+       rc = srp_target_alloc(&vscsi->target, &vdev->dev, vscsi->request_limit,
+                             SRP_MAX_IU_LEN);
+       if (rc)
+               goto rem_list;
+
+       vscsi->target.ldata = vscsi;
+
+       rc = ibmvscsis_alloc_cmds(vscsi, vscsi->request_limit);
+       if (rc) {
+               dev_err(&vscsi->dev, "alloc_cmds failed, rc %d, num %d\n",
+                       rc, vscsi->request_limit);
+               goto free_target;
+       }
+
+       /*
+        * Note: the lock is used in freeing timers, so must initialize
+        * first so that ordering in case of error is correct.
+        */
+       spin_lock_init(&vscsi->intr_lock);
+
+       rc = ibmvscsis_alloctimer(vscsi);
+       if (rc) {
+               dev_err(&vscsi->dev, "probe: alloctimer failed, rc %d\n", rc);
+               goto free_cmds;
+       }
+
+       rc = ibmvscsis_create_command_q(vscsi, 256);
+       if (rc) {
+               dev_err(&vscsi->dev, "probe: create_command_q failed, rc %d\n",
+                       rc);
+               goto free_timer;
+       }
+
+       vscsi->map_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
+       if (!vscsi->map_buf) {
+               rc = -ENOMEM;
+               dev_err(&vscsi->dev, "probe: allocating cmd buffer failed\n");
+               goto destroy_queue;
+       }
+
+       vscsi->map_ioba = dma_map_single(&vdev->dev, vscsi->map_buf, PAGE_SIZE,
+                                        DMA_BIDIRECTIONAL);
+       if (dma_mapping_error(&vdev->dev, vscsi->map_ioba)) {
+               dev_err(&vscsi->dev, "probe: error mapping command buffer\n");
+               goto free_buf;
+       }
+
+       hrc = h_vioctl(vscsi->dds.unit_id, H_GET_PARTNER_INFO,
+                      (u64)vscsi->map_ioba | ((u64)PAGE_SIZE << 32), 0, 0, 0,
+                      0);
+       if (hrc == H_SUCCESS)
+               vscsi->client_data.partition_number =
+                       be64_to_cpu(*(u64 *)vscsi->map_buf);
+       /*
+        * We expect the VIOCTL to fail if we're configured as "any
+        * client can connect" and the client isn't activated yet.
+        * We'll make the call again when he sends an init msg.
+        */
+       pr_debug("probe hrc %ld, client partition num %d\n",
+                hrc, vscsi->client_data.partition_number);
+
+       tasklet_init(&vscsi->work_task, ibmvscsis_handle_crq,
+                    (unsigned long)vscsi);
+
+       init_completion(&vscsi->wait_idle);
+
+       snprintf(wq_name, 24, "ibmvscsis%s", dev_name(&vdev->dev));
+       vscsi->work_q = create_workqueue(wq_name);
+       if (!vscsi->work_q) {
+               rc = -ENOMEM;
+               dev_err(&vscsi->dev, "create_workqueue failed\n");
+               goto unmap_buf;
+       }
+
+       rc = request_irq(vdev->irq, ibmvscsis_interrupt, 0, "ibmvscsis", vscsi);
+       if (rc) {
+               rc = -EPERM;
+               dev_err(&vscsi->dev, "probe: request_irq failed, rc %d\n", rc);
+               goto destroy_WQ;
+       }
+
+       spin_lock_bh(&vscsi->intr_lock);
+       vio_enable_interrupts(vdev);
+       if (rc) {
+               dev_err(&vscsi->dev, "enabling interrupts failed, rc %d\n", rc);
+               rc = -ENODEV;
+               spin_unlock_bh(&vscsi->intr_lock);
+               goto free_irq;
+       }
+
+       if (ibmvscsis_check_q(vscsi)) {
+               rc = ERROR;
+               dev_err(&vscsi->dev, "probe: check_q failed, rc %d\n", rc);
+               spin_unlock_bh(&vscsi->intr_lock);
+               goto disable_interrupt;
+       }
+       spin_unlock_bh(&vscsi->intr_lock);
+
+       dev_set_drvdata(&vdev->dev, vscsi);
+
+       return 0;
+
+disable_interrupt:
+       vio_disable_interrupts(vdev);
+free_irq:
+       free_irq(vdev->irq, vscsi);
+destroy_WQ:
+       destroy_workqueue(vscsi->work_q);
+unmap_buf:
+       dma_unmap_single(&vdev->dev, vscsi->map_ioba, PAGE_SIZE,
+                        DMA_BIDIRECTIONAL);
+free_buf:
+       kfree(vscsi->map_buf);
+destroy_queue:
+       tasklet_kill(&vscsi->work_task);
+       ibmvscsis_unregister_command_q(vscsi);
+       ibmvscsis_destroy_command_q(vscsi);
+free_timer:
+       ibmvscsis_freetimer(vscsi);
+free_cmds:
+       ibmvscsis_free_cmds(vscsi);
+free_target:
+       srp_target_free(&vscsi->target);
+rem_list:
+       spin_lock_bh(&ibmvscsis_dev_lock);
+       list_del(&vscsi->list);
+       spin_unlock_bh(&ibmvscsis_dev_lock);
+free_adapter:
+       kfree(vscsi);
+
+       return rc;
+}
+
+static int ibmvscsis_remove(struct vio_dev *vdev)
+{
+       struct scsi_info *vscsi = dev_get_drvdata(&vdev->dev);
+
+       pr_debug("remove (%s)\n", dev_name(&vscsi->dma_dev->dev));
+
+       /*
+        * TBD: Need to handle if there are commands on the waiting_rsp q
+        *      Actually, can there still be cmds outstanding to tcm?
+        */
+
+       vio_disable_interrupts(vdev);
+       free_irq(vdev->irq, vscsi);
+       destroy_workqueue(vscsi->work_q);
+       dma_unmap_single(&vdev->dev, vscsi->map_ioba, PAGE_SIZE,
+                        DMA_BIDIRECTIONAL);
+       kfree(vscsi->map_buf);
+       tasklet_kill(&vscsi->work_task);
+       ibmvscsis_unregister_command_q(vscsi);
+       ibmvscsis_destroy_command_q(vscsi);
+       ibmvscsis_freetimer(vscsi);
+       ibmvscsis_free_cmds(vscsi);
+       srp_target_free(&vscsi->target);
+       spin_lock_bh(&ibmvscsis_dev_lock);
+       list_del(&vscsi->list);
+       spin_unlock_bh(&ibmvscsis_dev_lock);
+       kfree(vscsi);
+
+       return 0;
+}
+
+static ssize_t system_id_show(struct device *dev,
+                             struct device_attribute *attr, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%s\n", system_id);
+}
+
+static ssize_t partition_number_show(struct device *dev,
+                                    struct device_attribute *attr, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%x\n", partition_number);
+}
+
+static ssize_t unit_address_show(struct device *dev,
+                                struct device_attribute *attr, char *buf)
+{
+       struct scsi_info *vscsi = container_of(dev, struct scsi_info, dev);
+
+       return snprintf(buf, PAGE_SIZE, "%x\n", vscsi->dma_dev->unit_address);
+}
+
+static int ibmvscsis_get_system_info(void)
+{
+       struct device_node *rootdn, *vdevdn;
+       const char *id, *model, *name;
+       const uint *num;
+
+       rootdn = of_find_node_by_path("/");
+       if (!rootdn)
+               return -ENOENT;
+
+       model = of_get_property(rootdn, "model", NULL);
+       id = of_get_property(rootdn, "system-id", NULL);
+       if (model && id)
+               snprintf(system_id, sizeof(system_id), "%s-%s", model, id);
+
+       name = of_get_property(rootdn, "ibm,partition-name", NULL);
+       if (name)
+               strncpy(partition_name, name, sizeof(partition_name));
+
+       num = of_get_property(rootdn, "ibm,partition-no", NULL);
+       if (num)
+               partition_number = *num;
+
+       of_node_put(rootdn);
+
+       vdevdn = of_find_node_by_path("/vdevice");
+       if (vdevdn) {
+               const uint *mvds;
+
+               mvds = of_get_property(vdevdn, "ibm,max-virtual-dma-size",
+                                      NULL);
+               if (mvds)
+                       max_vdma_size = *mvds;
+               of_node_put(vdevdn);
+       }
+
+       return 0;
+}
+
+static char *ibmvscsis_get_fabric_name(void)
+{
+       return "ibmvscsis";
+}
+
+static char *ibmvscsis_get_fabric_wwn(struct se_portal_group *se_tpg)
+{
+       struct ibmvscsis_tport *tport =
+               container_of(se_tpg, struct ibmvscsis_tport, se_tpg);
+
+       return tport->tport_name;
+}
+
+static u16 ibmvscsis_get_tag(struct se_portal_group *se_tpg)
+{
+       struct ibmvscsis_tport *tport =
+               container_of(se_tpg, struct ibmvscsis_tport, se_tpg);
+
+       return tport->tport_tpgt;
+}
+
+static u32 ibmvscsis_get_default_depth(struct se_portal_group *se_tpg)
+{
+       return 1;
+}
+
+static int ibmvscsis_check_true(struct se_portal_group *se_tpg)
+{
+       return 1;
+}
+
+static int ibmvscsis_check_false(struct se_portal_group *se_tpg)
+{
+       return 0;
+}
+
+static u32 ibmvscsis_tpg_get_inst_index(struct se_portal_group *se_tpg)
+{
+       return 1;
+}
+
+static int ibmvscsis_check_stop_free(struct se_cmd *se_cmd)
+{
+       return target_put_sess_cmd(se_cmd);
+}
+
+static void ibmvscsis_release_cmd(struct se_cmd *se_cmd)
+{
+       struct ibmvscsis_cmd *cmd = container_of(se_cmd, struct ibmvscsis_cmd,
+                                                se_cmd);
+       struct scsi_info *vscsi = cmd->adapter;
+
+       pr_debug("release_cmd %p, flags %d\n", se_cmd, cmd->flags);
+
+       spin_lock_bh(&vscsi->intr_lock);
+       /* Remove from active_q */
+       list_del(&cmd->list);
+       list_add_tail(&cmd->list, &vscsi->waiting_rsp);
+       ibmvscsis_send_messages(vscsi);
+       spin_unlock_bh(&vscsi->intr_lock);
+}
+
+static u32 ibmvscsis_sess_get_index(struct se_session *se_sess)
+{
+       return 0;
+}
+
+static int ibmvscsis_write_pending(struct se_cmd *se_cmd)
+{
+       struct ibmvscsis_cmd *cmd = container_of(se_cmd, struct ibmvscsis_cmd,
+                                                se_cmd);
+       struct iu_entry *iue = cmd->iue;
+       int rc;
+
+       pr_debug("write_pending, se_cmd %p, length 0x%x\n",
+                se_cmd, se_cmd->data_length);
+
+       rc = srp_transfer_data(cmd, &vio_iu(iue)->srp.cmd, ibmvscsis_rdma,
+                              1, 1);
+       if (rc) {
+               pr_err("srp_transfer_data() failed: %d\n", rc);
+               return -EAGAIN;
+       }
+       /*
+        * We now tell TCM to add this WRITE CDB directly into the TCM storage
+        * object execution queue.
+        */
+       target_execute_cmd(se_cmd);
+       return 0;
+}
+
+static int ibmvscsis_write_pending_status(struct se_cmd *se_cmd)
+{
+       return 0;
+}
+
+static void ibmvscsis_set_default_node_attrs(struct se_node_acl *nacl)
+{
+}
+
+static int ibmvscsis_get_cmd_state(struct se_cmd *se_cmd)
+{
+       return 0;
+}
+
+static int ibmvscsis_queue_data_in(struct se_cmd *se_cmd)
+{
+       struct ibmvscsis_cmd *cmd = container_of(se_cmd, struct ibmvscsis_cmd,
+                                                se_cmd);
+       struct iu_entry *iue = cmd->iue;
+       struct scsi_info *vscsi = cmd->adapter;
+       char *sd;
+       uint len = 0;
+       int rc;
+
+       pr_debug("queue_data_in, se_cmd %p, length 0x%x\n",
+                se_cmd, se_cmd->data_length);
+
+       rc = srp_transfer_data(cmd, &vio_iu(iue)->srp.cmd, ibmvscsis_rdma, 1,
+                              1);
+       if (rc) {
+               pr_err("srp_transfer_data failed: %d\n", rc);
+               sd = se_cmd->sense_buffer;
+               se_cmd->scsi_sense_length = 18;
+               memset(se_cmd->sense_buffer, 0, se_cmd->scsi_sense_length);
+               /* Logical Unit Communication Time-out asc/ascq = 0x0801 */
+               scsi_build_sense_buffer(0, se_cmd->sense_buffer, MEDIUM_ERROR,
+                                       0x08, 0x01);
+       }
+
+       srp_build_response(vscsi, cmd, &len);
+       cmd->rsp.format = SRP_FORMAT;
+       cmd->rsp.len = len;
+
+       return 0;
+}
+
+static int ibmvscsis_queue_status(struct se_cmd *se_cmd)
+{
+       struct ibmvscsis_cmd *cmd = container_of(se_cmd, struct ibmvscsis_cmd,
+                                                se_cmd);
+       struct scsi_info *vscsi = cmd->adapter;
+       uint len;
+
+       pr_debug("queue_status %p\n", se_cmd);
+
+       srp_build_response(vscsi, cmd, &len);
+       cmd->rsp.format = SRP_FORMAT;
+       cmd->rsp.len = len;
+
+       return 0;
+}
+
+static void ibmvscsis_queue_tm_rsp(struct se_cmd *se_cmd)
+{
+       struct ibmvscsis_cmd *cmd = container_of(se_cmd, struct ibmvscsis_cmd,
+                                                se_cmd);
+       struct scsi_info *vscsi = cmd->adapter;
+       uint len;
+
+       pr_debug("queue_tm_rsp %p, status %d\n",
+                se_cmd, (int)se_cmd->se_tmr_req->response);
+
+       srp_build_response(vscsi, cmd, &len);
+       cmd->rsp.format = SRP_FORMAT;
+       cmd->rsp.len = len;
+}
+
+static void ibmvscsis_aborted_task(struct se_cmd *se_cmd)
+{
+       /* TBD: What (if anything) should we do here? */
+       pr_debug("ibmvscsis_aborted_task %p\n", se_cmd);
+}
+
+static struct se_wwn *ibmvscsis_make_tport(struct target_fabric_configfs *tf,
+                                          struct config_group *group,
+                                          const char *name)
+{
+       struct ibmvscsis_tport *tport;
+
+       tport = ibmvscsis_lookup_port(name);
+       if (tport) {
+               tport->tport_proto_id = SCSI_PROTOCOL_SRP;
+               pr_debug("make_tport(%s), pointer:%p, tport_id:%x\n",
+                        name, tport, tport->tport_proto_id);
+               return &tport->tport_wwn;
+       }
+
+       return ERR_PTR(-EINVAL);
+}
+
+static void ibmvscsis_drop_tport(struct se_wwn *wwn)
+{
+       struct ibmvscsis_tport *tport = container_of(wwn,
+                                                    struct ibmvscsis_tport,
+                                                    tport_wwn);
+
+       pr_debug("drop_tport(%s)\n",
+                config_item_name(&tport->tport_wwn.wwn_group.cg_item));
+}
+
+static struct se_portal_group *ibmvscsis_make_tpg(struct se_wwn *wwn,
+                                                 struct config_group *group,
+                                                 const char *name)
+{
+       struct ibmvscsis_tport *tport =
+               container_of(wwn, struct ibmvscsis_tport, tport_wwn);
+       int rc;
+
+       tport->releasing = false;
+
+       rc = core_tpg_register(&tport->tport_wwn, &tport->se_tpg,
+                              tport->tport_proto_id);
+       if (rc)
+               return ERR_PTR(rc);
+
+       return &tport->se_tpg;
+}
+
+static void ibmvscsis_drop_tpg(struct se_portal_group *se_tpg)
+{
+       struct ibmvscsis_tport *tport = container_of(se_tpg,
+                                                    struct ibmvscsis_tport,
+                                                    se_tpg);
+
+       tport->releasing = true;
+       tport->enabled = false;
+
+       /*
+        * Release the virtual I_T Nexus for this ibmvscsis TPG
+        */
+       ibmvscsis_drop_nexus(tport);
+       /*
+        * Deregister the se_tpg from TCM..
+        */
+       core_tpg_deregister(se_tpg);
+}
+
+static ssize_t ibmvscsis_wwn_version_show(struct config_item *item,
+                                         char *page)
+{
+       return scnprintf(page, PAGE_SIZE, "%s\n", IBMVSCSIS_VERSION);
+}
+CONFIGFS_ATTR_RO(ibmvscsis_wwn_, version);
+
+static struct configfs_attribute *ibmvscsis_wwn_attrs[] = {
+       &ibmvscsis_wwn_attr_version,
+       NULL,
+};
+
+static ssize_t ibmvscsis_tpg_enable_show(struct config_item *item,
+                                        char *page)
+{
+       struct se_portal_group *se_tpg = to_tpg(item);
+       struct ibmvscsis_tport *tport = container_of(se_tpg,
+                                                    struct ibmvscsis_tport,
+                                                    se_tpg);
+
+       return snprintf(page, PAGE_SIZE, "%d\n", (tport->enabled) ? 1 : 0);
+}
+
+static ssize_t ibmvscsis_tpg_enable_store(struct config_item *item,
+                                         const char *page, size_t count)
+{
+       struct se_portal_group *se_tpg = to_tpg(item);
+       struct ibmvscsis_tport *tport = container_of(se_tpg,
+                                                    struct ibmvscsis_tport,
+                                                    se_tpg);
+       struct scsi_info *vscsi = container_of(tport, struct scsi_info, tport);
+       unsigned long tmp;
+       int rc;
+       long lrc;
+
+       rc = kstrtoul(page, 0, &tmp);
+       if (rc < 0) {
+               pr_err("Unable to extract srpt_tpg_store_enable\n");
+               return -EINVAL;
+       }
+
+       if ((tmp != 0) && (tmp != 1)) {
+               pr_err("Illegal value for srpt_tpg_store_enable\n");
+               return -EINVAL;
+       }
+
+       if (tmp) {
+               tport->enabled = true;
+               spin_lock_bh(&vscsi->intr_lock);
+               lrc = ibmvscsis_enable_change_state(vscsi);
+               if (lrc)
+                       pr_err("enable_change_state failed, rc %ld state %d\n",
+                              lrc, vscsi->state);
+               spin_unlock_bh(&vscsi->intr_lock);
+       } else {
+               tport->enabled = false;
+       }
+
+       pr_debug("tpg_enable_store, state %d\n", vscsi->state);
+
+       return count;
+}
+CONFIGFS_ATTR(ibmvscsis_tpg_, enable);
+
+static struct configfs_attribute *ibmvscsis_tpg_attrs[] = {
+       &ibmvscsis_tpg_attr_enable,
+       NULL,
+};
+
+static const struct target_core_fabric_ops ibmvscsis_ops = {
+       .module                         = THIS_MODULE,
+       .name                           = "ibmvscsis",
+       .get_fabric_name                = ibmvscsis_get_fabric_name,
+       .tpg_get_wwn                    = ibmvscsis_get_fabric_wwn,
+       .tpg_get_tag                    = ibmvscsis_get_tag,
+       .tpg_get_default_depth          = ibmvscsis_get_default_depth,
+       .tpg_check_demo_mode            = ibmvscsis_check_true,
+       .tpg_check_demo_mode_cache      = ibmvscsis_check_true,
+       .tpg_check_demo_mode_write_protect = ibmvscsis_check_false,
+       .tpg_check_prod_mode_write_protect = ibmvscsis_check_false,
+       .tpg_get_inst_index             = ibmvscsis_tpg_get_inst_index,
+       .check_stop_free                = ibmvscsis_check_stop_free,
+       .release_cmd                    = ibmvscsis_release_cmd,
+       .sess_get_index                 = ibmvscsis_sess_get_index,
+       .write_pending                  = ibmvscsis_write_pending,
+       .write_pending_status           = ibmvscsis_write_pending_status,
+       .set_default_node_attributes    = ibmvscsis_set_default_node_attrs,
+       .get_cmd_state                  = ibmvscsis_get_cmd_state,
+       .queue_data_in                  = ibmvscsis_queue_data_in,
+       .queue_status                   = ibmvscsis_queue_status,
+       .queue_tm_rsp                   = ibmvscsis_queue_tm_rsp,
+       .aborted_task                   = ibmvscsis_aborted_task,
+       /*
+        * Setup function pointers for logic in target_core_fabric_configfs.c
+        */
+       .fabric_make_wwn                = ibmvscsis_make_tport,
+       .fabric_drop_wwn                = ibmvscsis_drop_tport,
+       .fabric_make_tpg                = ibmvscsis_make_tpg,
+       .fabric_drop_tpg                = ibmvscsis_drop_tpg,
+
+       .tfc_wwn_attrs                  = ibmvscsis_wwn_attrs,
+       .tfc_tpg_base_attrs             = ibmvscsis_tpg_attrs,
+};
+
+static void ibmvscsis_dev_release(struct device *dev) {};
+
+static struct class_attribute ibmvscsis_class_attrs[] = {
+       __ATTR_NULL,
+};
+
+static struct device_attribute dev_attr_system_id =
+       __ATTR(system_id, S_IRUGO, system_id_show, NULL);
+
+static struct device_attribute dev_attr_partition_number =
+       __ATTR(partition_number, S_IRUGO, partition_number_show, NULL);
+
+static struct device_attribute dev_attr_unit_address =
+       __ATTR(unit_address, S_IRUGO, unit_address_show, NULL);
+
+static struct attribute *ibmvscsis_dev_attrs[] = {
+       &dev_attr_system_id.attr,
+       &dev_attr_partition_number.attr,
+       &dev_attr_unit_address.attr,
+};
+ATTRIBUTE_GROUPS(ibmvscsis_dev);
+
+static struct class ibmvscsis_class = {
+       .name           = "ibmvscsis",
+       .dev_release    = ibmvscsis_dev_release,
+       .class_attrs    = ibmvscsis_class_attrs,
+       .dev_groups     = ibmvscsis_dev_groups,
+};
+
+static struct vio_device_id ibmvscsis_device_table[] = {
+       { "v-scsi-host", "IBM,v-scsi-host" },
+       { "", "" }
+};
+MODULE_DEVICE_TABLE(vio, ibmvscsis_device_table);
+
+static struct vio_driver ibmvscsis_driver = {
+       .name = "ibmvscsis",
+       .id_table = ibmvscsis_device_table,
+       .probe = ibmvscsis_probe,
+       .remove = ibmvscsis_remove,
+};
+
+/*
+ * ibmvscsis_init() - Kernel Module initialization
+ *
+ * Note: vio_register_driver() registers callback functions, and at least one
+ * of those callback functions calls TCM - Linux IO Target Subsystem, thus
+ * the SCSI Target template must be registered before vio_register_driver()
+ * is called.
+ */
+static int __init ibmvscsis_init(void)
+{
+       int rc = 0;
+
+       rc = ibmvscsis_get_system_info();
+       if (rc) {
+               pr_err("rc %d from get_system_info\n", rc);
+               goto out;
+       }
+
+       rc = class_register(&ibmvscsis_class);
+       if (rc) {
+               pr_err("failed class register\n");
+               goto out;
+       }
+
+       rc = target_register_template(&ibmvscsis_ops);
+       if (rc) {
+               pr_err("rc %d from target_register_template\n", rc);
+               goto unregister_class;
+       }
+
+       rc = vio_register_driver(&ibmvscsis_driver);
+       if (rc) {
+               pr_err("rc %d from vio_register_driver\n", rc);
+               goto unregister_target;
+       }
+
+       return 0;
+
+unregister_target:
+       target_unregister_template(&ibmvscsis_ops);
+unregister_class:
+       class_unregister(&ibmvscsis_class);
+out:
+       return rc;
+}
+
+static void __exit ibmvscsis_exit(void)
+{
+       pr_info("Unregister IBM virtual SCSI host driver\n");
+       vio_unregister_driver(&ibmvscsis_driver);
+       target_unregister_template(&ibmvscsis_ops);
+       class_unregister(&ibmvscsis_class);
+}
+
+MODULE_DESCRIPTION("IBMVSCSIS fabric driver");
+MODULE_AUTHOR("Bryant G. Ly and Michael Cyr");
+MODULE_LICENSE("GPL");
+MODULE_VERSION(IBMVSCSIS_VERSION);
+module_init(ibmvscsis_init);
+module_exit(ibmvscsis_exit);
diff --git a/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.h b/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.h
new file mode 100644 (file)
index 0000000..981a0c9
--- /dev/null
@@ -0,0 +1,346 @@
+/*******************************************************************************
+ * IBM Virtual SCSI Target Driver
+ * Copyright (C) 2003-2005 Dave Boutcher (boutcher@us.ibm.com) IBM Corp.
+ *                        Santiago Leon (santil@us.ibm.com) IBM Corp.
+ *                        Linda Xie (lxie@us.ibm.com) IBM Corp.
+ *
+ * Copyright (C) 2005-2011 FUJITA Tomonori <tomof@acm.org>
+ * Copyright (C) 2010 Nicholas A. Bellinger <nab@kernel.org>
+ * Copyright (C) 2016 Bryant G. Ly <bryantly@linux.vnet.ibm.com> IBM Corp.
+ *
+ * Authors: Bryant G. Ly <bryantly@linux.vnet.ibm.com>
+ * Authors: Michael Cyr <mikecyr@linux.vnet.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ ****************************************************************************/
+
+#ifndef __H_IBMVSCSI_TGT
+#define __H_IBMVSCSI_TGT
+
+#include "libsrp.h"
+
+#define SYS_ID_NAME_LEN                64
+#define PARTITION_NAMELEN      96
+#define IBMVSCSIS_NAMELEN       32
+
+#define MSG_HI  0
+#define MSG_LOW 1
+
+#define MAX_CMD_Q_PAGES       4
+#define CRQ_PER_PAGE          (PAGE_SIZE / sizeof(struct viosrp_crq))
+/* in terms of number of elements */
+#define DEFAULT_CMD_Q_SIZE    CRQ_PER_PAGE
+#define MAX_CMD_Q_SIZE        (DEFAULT_CMD_Q_SIZE * MAX_CMD_Q_PAGES)
+
+#define SRP_VIOLATION           0x102  /* general error code */
+
+/*
+ * SRP buffer formats defined as of 16.a supported by this driver.
+ */
+#define SUPPORTED_FORMATS  ((SRP_DATA_DESC_DIRECT << 1) | \
+                           (SRP_DATA_DESC_INDIRECT << 1))
+
+#define SCSI_LUN_ADDR_METHOD_FLAT      1
+
+struct dma_window {
+       u32 liobn;      /* Unique per vdevice */
+       u64 tce_base;   /* Physical location of the TCE table */
+       u64 tce_size;   /* Size of the TCE table in bytes */
+};
+
+struct target_dds {
+       u64 unit_id;                /* 64 bit will force alignment */
+#define NUM_DMA_WINDOWS 2
+#define LOCAL  0
+#define REMOTE 1
+       struct dma_window  window[NUM_DMA_WINDOWS];
+
+       /* root node property "ibm,partition-no" */
+       uint partition_num;
+       char partition_name[PARTITION_NAMELEN];
+};
+
+#define MAX_NUM_PORTS        1
+#define MAX_H_COPY_RDMA      (128 * 1024)
+
+#define MAX_EYE   64
+
+/* Return codes */
+#define ADAPT_SUCCESS            0L
+/* choose error codes that do not conflict with PHYP */
+#define ERROR                   -40L
+
+struct format_code {
+       u8 reserved;
+       u8 buffers;
+};
+
+struct client_info {
+#define SRP_VERSION "16.a"
+       char srp_version[8];
+       /* root node property ibm,partition-name */
+       char partition_name[PARTITION_NAMELEN];
+       /* root node property ibm,partition-no */
+       u32 partition_number;
+       /* initially 1 */
+       u32 mad_version;
+       u32 os_type;
+};
+
+/*
+ * Changing this constant changes the number of seconds to wait before
+ * considering the client will never service its queue again.
+ */
+#define SECONDS_TO_CONSIDER_FAILED 30
+/*
+ * These constants set the polling period used to determine if the client
+ * has freed at least one element in the response queue.
+ */
+#define WAIT_SECONDS 1
+#define WAIT_NANO_SECONDS 5000
+#define MAX_TIMER_POPS ((1000000 / WAIT_NANO_SECONDS) * \
+                       SECONDS_TO_CONSIDER_FAILED)
+/*
+ * general purpose timer control block
+ * which can be used for multiple functions
+ */
+struct timer_cb {
+       struct hrtimer timer;
+       /*
+        * how long has it been since the client
+        * serviced the queue. The variable is incrmented
+        * in the service_wait_q routine and cleared
+        * in send messages
+        */
+       int timer_pops;
+       /* the timer is started */
+       bool started;
+};
+
+struct cmd_queue {
+       /* kva */
+       struct viosrp_crq *base_addr;
+       dma_addr_t crq_token;
+       /* used to maintain index */
+       uint mask;
+       /* current element */
+       uint index;
+       int size;
+};
+
+#define SCSOLNT_RESP_SHIFT     1
+#define UCSOLNT_RESP_SHIFT     2
+
+#define SCSOLNT         BIT(SCSOLNT_RESP_SHIFT)
+#define UCSOLNT         BIT(UCSOLNT_RESP_SHIFT)
+
+enum cmd_type {
+       SCSI_CDB        = 0x01,
+       TASK_MANAGEMENT = 0x02,
+       /* MAD or addressed to port 0 */
+       ADAPTER_MAD     = 0x04,
+       UNSET_TYPE      = 0x08,
+};
+
+struct iu_rsp {
+       u8 format;
+       u8 sol_not;
+       u16 len;
+       /* tag is just to help client identify cmd, so don't translate be/le */
+       u64 tag;
+};
+
+struct ibmvscsis_cmd {
+       struct list_head list;
+       /* Used for TCM Core operations */
+       struct se_cmd se_cmd;
+       struct iu_entry *iue;
+       struct iu_rsp rsp;
+       struct work_struct work;
+       struct scsi_info *adapter;
+       /* Sense buffer that will be mapped into outgoing status */
+       unsigned char sense_buf[TRANSPORT_SENSE_BUFFER];
+       u64 init_time;
+#define CMD_FAST_FAIL  BIT(0)
+       u32 flags;
+       char type;
+};
+
+struct ibmvscsis_nexus {
+       struct se_session *se_sess;
+};
+
+struct ibmvscsis_tport {
+       /* SCSI protocol the tport is providing */
+       u8 tport_proto_id;
+       /* ASCII formatted WWPN for SRP Target port */
+       char tport_name[IBMVSCSIS_NAMELEN];
+       /* Returned by ibmvscsis_make_tport() */
+       struct se_wwn tport_wwn;
+       /* Returned by ibmvscsis_make_tpg() */
+       struct se_portal_group se_tpg;
+       /* ibmvscsis port target portal group tag for TCM */
+       u16 tport_tpgt;
+       /* Pointer to TCM session for I_T Nexus */
+       struct ibmvscsis_nexus *ibmv_nexus;
+       bool enabled;
+       bool releasing;
+};
+
+struct scsi_info {
+       struct list_head list;
+       char eye[MAX_EYE];
+
+       /* commands waiting for space on repsonse queue */
+       struct list_head waiting_rsp;
+#define NO_QUEUE                    0x00
+#define WAIT_ENABLED                0X01
+       /* driver has received an initialize command */
+#define PART_UP_WAIT_ENAB           0x02
+#define WAIT_CONNECTION             0x04
+       /* have established a connection */
+#define CONNECTED                   0x08
+       /* at least one port is processing SRP IU */
+#define SRP_PROCESSING              0x10
+       /* remove request received */
+#define UNCONFIGURING               0x20
+       /* disconnect by letting adapter go idle, no error */
+#define WAIT_IDLE                   0x40
+       /* disconnecting to clear an error */
+#define ERR_DISCONNECT              0x80
+       /* disconnect to clear error state, then come back up */
+#define ERR_DISCONNECT_RECONNECT    0x100
+       /* disconnected after clearing an error */
+#define ERR_DISCONNECTED            0x200
+       /* A series of errors caused unexpected errors */
+#define UNDEFINED                   0x400
+       u16  state;
+       int fast_fail;
+       struct target_dds dds;
+       char *cmd_pool;
+       /* list of free commands */
+       struct list_head free_cmd;
+       /* command elements ready for scheduler */
+       struct list_head schedule_q;
+       /* commands sent to TCM */
+       struct list_head active_q;
+       caddr_t *map_buf;
+       /* ioba of map buffer */
+       dma_addr_t map_ioba;
+       /* allowable number of outstanding SRP requests */
+       int request_limit;
+       /* extra credit */
+       int credit;
+       /* outstanding transactions against credit limit */
+       int debit;
+
+       /* allow only one outstanding mad request */
+#define PROCESSING_MAD                0x00002
+       /* Waiting to go idle */
+#define WAIT_FOR_IDLE                0x00004
+       /* H_REG_CRQ called */
+#define CRQ_CLOSED                    0x00010
+       /* detected that client has failed */
+#define CLIENT_FAILED                 0x00040
+       /* detected that transport event occurred */
+#define TRANS_EVENT                   0x00080
+       /* don't attempt to send anything to the client */
+#define RESPONSE_Q_DOWN               0x00100
+       /* request made to schedule disconnect handler */
+#define SCHEDULE_DISCONNECT           0x00400
+       /* disconnect handler is scheduled */
+#define DISCONNECT_SCHEDULED          0x00800
+       u32 flags;
+       /* adapter lock */
+       spinlock_t intr_lock;
+       /* information needed to manage command queue */
+       struct cmd_queue cmd_q;
+       /* used in hcall to copy response back into srp buffer */
+       u64  empty_iu_id;
+       /* used in crq, to tag what iu the response is for */
+       u64  empty_iu_tag;
+       uint new_state;
+       /* control block for the response queue timer */
+       struct timer_cb rsp_q_timer;
+       /* keep last client to enable proper accounting */
+       struct client_info client_data;
+       /* what can this client do */
+       u32 client_cap;
+       /*
+        * The following two fields capture state and flag changes that
+        * can occur when the lock is given up.  In the orginal design,
+        * the lock was held during calls into phyp;
+        * however, phyp did not meet PAPR architecture.  This is
+        * a work around.
+        */
+       u16  phyp_acr_state;
+       u32 phyp_acr_flags;
+
+       struct workqueue_struct *work_q;
+       struct completion wait_idle;
+       struct device dev;
+       struct vio_dev *dma_dev;
+       struct srp_target target;
+       struct ibmvscsis_tport tport;
+       struct tasklet_struct work_task;
+       struct work_struct proc_work;
+};
+
+/*
+ * Provide a constant that allows software to detect the adapter is
+ * disconnecting from the client from one of several states.
+ */
+#define IS_DISCONNECTING (UNCONFIGURING | ERR_DISCONNECT_RECONNECT | \
+                         ERR_DISCONNECT)
+
+/*
+ * Provide a constant that can be used with interrupt handling that
+ * essentially lets the interrupt handler know that all requests should
+ * be thrown out,
+ */
+#define DONT_PROCESS_STATE (IS_DISCONNECTING | UNDEFINED | \
+                           ERR_DISCONNECTED  | WAIT_IDLE)
+
+/*
+ * If any of these flag bits are set then do not allow the interrupt
+ * handler to schedule the off level handler.
+ */
+#define BLOCK (DISCONNECT_SCHEDULED)
+
+/* State and transition events that stop the interrupt handler */
+#define TARGET_STOP(VSCSI) (long)(((VSCSI)->state & DONT_PROCESS_STATE) | \
+                                 ((VSCSI)->flags & BLOCK))
+
+/* flag bit that are not reset during disconnect */
+#define PRESERVE_FLAG_FIELDS 0
+
+#define vio_iu(IUE) ((union viosrp_iu *)((IUE)->sbuf->buf))
+
+#define READ_CMD(cdb)  (((cdb)[0] & 0x1F) == 8)
+#define WRITE_CMD(cdb) (((cdb)[0] & 0x1F) == 0xA)
+
+#ifndef H_GET_PARTNER_INFO
+#define H_GET_PARTNER_INFO      0x0000000000000008LL
+#endif
+
+#define h_copy_rdma(l, sa, sb, da, db) \
+               plpar_hcall_norets(H_COPY_RDMA, l, sa, sb, da, db)
+#define h_vioctl(u, o, a, u1, u2, u3, u4) \
+               plpar_hcall_norets(H_VIOCTL, u, o, a, u1, u2)
+#define h_reg_crq(ua, tok, sz) \
+               plpar_hcall_norets(H_REG_CRQ, ua, tok, sz)
+#define h_free_crq(ua) \
+               plpar_hcall_norets(H_FREE_CRQ, ua)
+#define h_send_crq(ua, d1, d2) \
+               plpar_hcall_norets(H_SEND_CRQ, ua, d1, d2)
+
+#endif
diff --git a/drivers/scsi/ibmvscsi_tgt/libsrp.c b/drivers/scsi/ibmvscsi_tgt/libsrp.c
new file mode 100644 (file)
index 0000000..5a4cc28
--- /dev/null
@@ -0,0 +1,427 @@
+/*******************************************************************************
+ * SCSI RDMA Protocol lib functions
+ *
+ * Copyright (C) 2006 FUJITA Tomonori <tomof@acm.org>
+ * Copyright (C) 2016 Bryant G. Ly <bryantly@linux.vnet.ibm.com> IBM Corp.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ ***********************************************************************/
+
+#define pr_fmt(fmt)    "libsrp: " fmt
+
+#include <linux/printk.h>
+#include <linux/err.h>
+#include <linux/slab.h>
+#include <linux/kfifo.h>
+#include <linux/scatterlist.h>
+#include <linux/dma-mapping.h>
+#include <linux/module.h>
+#include <scsi/srp.h>
+#include <target/target_core_base.h>
+#include "libsrp.h"
+#include "ibmvscsi_tgt.h"
+
+static int srp_iu_pool_alloc(struct srp_queue *q, size_t max,
+                            struct srp_buf **ring)
+{
+       struct iu_entry *iue;
+       int i;
+
+       q->pool = kcalloc(max, sizeof(struct iu_entry *), GFP_KERNEL);
+       if (!q->pool)
+               return -ENOMEM;
+       q->items = kcalloc(max, sizeof(struct iu_entry), GFP_KERNEL);
+       if (!q->items)
+               goto free_pool;
+
+       spin_lock_init(&q->lock);
+       kfifo_init(&q->queue, (void *)q->pool, max * sizeof(void *));
+
+       for (i = 0, iue = q->items; i < max; i++) {
+               kfifo_in(&q->queue, (void *)&iue, sizeof(void *));
+               iue->sbuf = ring[i];
+               iue++;
+       }
+       return 0;
+
+free_pool:
+       kfree(q->pool);
+       return -ENOMEM;
+}
+
+static void srp_iu_pool_free(struct srp_queue *q)
+{
+       kfree(q->items);
+       kfree(q->pool);
+}
+
+static struct srp_buf **srp_ring_alloc(struct device *dev,
+                                      size_t max, size_t size)
+{
+       struct srp_buf **ring;
+       int i;
+
+       ring = kcalloc(max, sizeof(struct srp_buf *), GFP_KERNEL);
+       if (!ring)
+               return NULL;
+
+       for (i = 0; i < max; i++) {
+               ring[i] = kzalloc(sizeof(*ring[i]), GFP_KERNEL);
+               if (!ring[i])
+                       goto out;
+               ring[i]->buf = dma_alloc_coherent(dev, size, &ring[i]->dma,
+                                                 GFP_KERNEL);
+               if (!ring[i]->buf)
+                       goto out;
+       }
+       return ring;
+
+out:
+       for (i = 0; i < max && ring[i]; i++) {
+               if (ring[i]->buf) {
+                       dma_free_coherent(dev, size, ring[i]->buf,
+                                         ring[i]->dma);
+               }
+               kfree(ring[i]);
+       }
+       kfree(ring);
+
+       return NULL;
+}
+
+static void srp_ring_free(struct device *dev, struct srp_buf **ring,
+                         size_t max, size_t size)
+{
+       int i;
+
+       for (i = 0; i < max; i++) {
+               dma_free_coherent(dev, size, ring[i]->buf, ring[i]->dma);
+               kfree(ring[i]);
+       }
+       kfree(ring);
+}
+
+int srp_target_alloc(struct srp_target *target, struct device *dev,
+                    size_t nr, size_t iu_size)
+{
+       int err;
+
+       spin_lock_init(&target->lock);
+
+       target->dev = dev;
+
+       target->srp_iu_size = iu_size;
+       target->rx_ring_size = nr;
+       target->rx_ring = srp_ring_alloc(target->dev, nr, iu_size);
+       if (!target->rx_ring)
+               return -ENOMEM;
+       err = srp_iu_pool_alloc(&target->iu_queue, nr, target->rx_ring);
+       if (err)
+               goto free_ring;
+
+       dev_set_drvdata(target->dev, target);
+       return 0;
+
+free_ring:
+       srp_ring_free(target->dev, target->rx_ring, nr, iu_size);
+       return -ENOMEM;
+}
+
+void srp_target_free(struct srp_target *target)
+{
+       dev_set_drvdata(target->dev, NULL);
+       srp_ring_free(target->dev, target->rx_ring, target->rx_ring_size,
+                     target->srp_iu_size);
+       srp_iu_pool_free(&target->iu_queue);
+}
+
+struct iu_entry *srp_iu_get(struct srp_target *target)
+{
+       struct iu_entry *iue = NULL;
+
+       if (kfifo_out_locked(&target->iu_queue.queue, (void *)&iue,
+                            sizeof(void *),
+                            &target->iu_queue.lock) != sizeof(void *)) {
+               WARN_ONCE(1, "unexpected fifo state");
+               return NULL;
+       }
+       if (!iue)
+               return iue;
+       iue->target = target;
+       iue->flags = 0;
+       return iue;
+}
+
+void srp_iu_put(struct iu_entry *iue)
+{
+       kfifo_in_locked(&iue->target->iu_queue.queue, (void *)&iue,
+                       sizeof(void *), &iue->target->iu_queue.lock);
+}
+
+static int srp_direct_data(struct ibmvscsis_cmd *cmd, struct srp_direct_buf *md,
+                          enum dma_data_direction dir, srp_rdma_t rdma_io,
+                          int dma_map, int ext_desc)
+{
+       struct iu_entry *iue = NULL;
+       struct scatterlist *sg = NULL;
+       int err, nsg = 0, len;
+
+       if (dma_map) {
+               iue = cmd->iue;
+               sg = cmd->se_cmd.t_data_sg;
+               nsg = dma_map_sg(iue->target->dev, sg, cmd->se_cmd.t_data_nents,
+                                DMA_BIDIRECTIONAL);
+               if (!nsg) {
+                       pr_err("fail to map %p %d\n", iue,
+                              cmd->se_cmd.t_data_nents);
+                       return 0;
+               }
+               len = min(cmd->se_cmd.data_length, be32_to_cpu(md->len));
+       } else {
+               len = be32_to_cpu(md->len);
+       }
+
+       err = rdma_io(cmd, sg, nsg, md, 1, dir, len);
+
+       if (dma_map)
+               dma_unmap_sg(iue->target->dev, sg, nsg, DMA_BIDIRECTIONAL);
+
+       return err;
+}
+
+static int srp_indirect_data(struct ibmvscsis_cmd *cmd, struct srp_cmd *srp_cmd,
+                            struct srp_indirect_buf *id,
+                            enum dma_data_direction dir, srp_rdma_t rdma_io,
+                            int dma_map, int ext_desc)
+{
+       struct iu_entry *iue = NULL;
+       struct srp_direct_buf *md = NULL;
+       struct scatterlist dummy, *sg = NULL;
+       dma_addr_t token = 0;
+       int err = 0;
+       int nmd, nsg = 0, len;
+
+       if (dma_map || ext_desc) {
+               iue = cmd->iue;
+               sg = cmd->se_cmd.t_data_sg;
+       }
+
+       nmd = be32_to_cpu(id->table_desc.len) / sizeof(struct srp_direct_buf);
+
+       if ((dir == DMA_FROM_DEVICE && nmd == srp_cmd->data_in_desc_cnt) ||
+           (dir == DMA_TO_DEVICE && nmd == srp_cmd->data_out_desc_cnt)) {
+               md = &id->desc_list[0];
+               goto rdma;
+       }
+
+       if (ext_desc && dma_map) {
+               md = dma_alloc_coherent(iue->target->dev,
+                                       be32_to_cpu(id->table_desc.len),
+                                       &token, GFP_KERNEL);
+               if (!md) {
+                       pr_err("Can't get dma memory %u\n",
+                              be32_to_cpu(id->table_desc.len));
+                       return -ENOMEM;
+               }
+
+               sg_init_one(&dummy, md, be32_to_cpu(id->table_desc.len));
+               sg_dma_address(&dummy) = token;
+               sg_dma_len(&dummy) = be32_to_cpu(id->table_desc.len);
+               err = rdma_io(cmd, &dummy, 1, &id->table_desc, 1, DMA_TO_DEVICE,
+                             be32_to_cpu(id->table_desc.len));
+               if (err) {
+                       pr_err("Error copying indirect table %d\n", err);
+                       goto free_mem;
+               }
+       } else {
+               pr_err("This command uses external indirect buffer\n");
+               return -EINVAL;
+       }
+
+rdma:
+       if (dma_map) {
+               nsg = dma_map_sg(iue->target->dev, sg, cmd->se_cmd.t_data_nents,
+                                DMA_BIDIRECTIONAL);
+               if (!nsg) {
+                       pr_err("fail to map %p %d\n", iue,
+                              cmd->se_cmd.t_data_nents);
+                       err = -EIO;
+                       goto free_mem;
+               }
+               len = min(cmd->se_cmd.data_length, be32_to_cpu(id->len));
+       } else {
+               len = be32_to_cpu(id->len);
+       }
+
+       err = rdma_io(cmd, sg, nsg, md, nmd, dir, len);
+
+       if (dma_map)
+               dma_unmap_sg(iue->target->dev, sg, nsg, DMA_BIDIRECTIONAL);
+
+free_mem:
+       if (token && dma_map) {
+               dma_free_coherent(iue->target->dev,
+                                 be32_to_cpu(id->table_desc.len), md, token);
+       }
+       return err;
+}
+
+static int data_out_desc_size(struct srp_cmd *cmd)
+{
+       int size = 0;
+       u8 fmt = cmd->buf_fmt >> 4;
+
+       switch (fmt) {
+       case SRP_NO_DATA_DESC:
+               break;
+       case SRP_DATA_DESC_DIRECT:
+               size = sizeof(struct srp_direct_buf);
+               break;
+       case SRP_DATA_DESC_INDIRECT:
+               size = sizeof(struct srp_indirect_buf) +
+                       sizeof(struct srp_direct_buf) * cmd->data_out_desc_cnt;
+               break;
+       default:
+               pr_err("client error. Invalid data_out_format %x\n", fmt);
+               break;
+       }
+       return size;
+}
+
+/*
+ * TODO: this can be called multiple times for a single command if it
+ * has very long data.
+ */
+int srp_transfer_data(struct ibmvscsis_cmd *cmd, struct srp_cmd *srp_cmd,
+                     srp_rdma_t rdma_io, int dma_map, int ext_desc)
+{
+       struct srp_direct_buf *md;
+       struct srp_indirect_buf *id;
+       enum dma_data_direction dir;
+       int offset, err = 0;
+       u8 format;
+
+       if (!cmd->se_cmd.t_data_nents)
+               return 0;
+
+       offset = srp_cmd->add_cdb_len & ~3;
+
+       dir = srp_cmd_direction(srp_cmd);
+       if (dir == DMA_FROM_DEVICE)
+               offset += data_out_desc_size(srp_cmd);
+
+       if (dir == DMA_TO_DEVICE)
+               format = srp_cmd->buf_fmt >> 4;
+       else
+               format = srp_cmd->buf_fmt & ((1U << 4) - 1);
+
+       switch (format) {
+       case SRP_NO_DATA_DESC:
+               break;
+       case SRP_DATA_DESC_DIRECT:
+               md = (struct srp_direct_buf *)(srp_cmd->add_data + offset);
+               err = srp_direct_data(cmd, md, dir, rdma_io, dma_map, ext_desc);
+               break;
+       case SRP_DATA_DESC_INDIRECT:
+               id = (struct srp_indirect_buf *)(srp_cmd->add_data + offset);
+               err = srp_indirect_data(cmd, srp_cmd, id, dir, rdma_io, dma_map,
+                                       ext_desc);
+               break;
+       default:
+               pr_err("Unknown format %d %x\n", dir, format);
+               err = -EINVAL;
+       }
+
+       return err;
+}
+
+u64 srp_data_length(struct srp_cmd *cmd, enum dma_data_direction dir)
+{
+       struct srp_direct_buf *md;
+       struct srp_indirect_buf *id;
+       u64 len = 0;
+       uint offset = cmd->add_cdb_len & ~3;
+       u8 fmt;
+
+       if (dir == DMA_TO_DEVICE) {
+               fmt = cmd->buf_fmt >> 4;
+       } else {
+               fmt = cmd->buf_fmt & ((1U << 4) - 1);
+               offset += data_out_desc_size(cmd);
+       }
+
+       switch (fmt) {
+       case SRP_NO_DATA_DESC:
+               break;
+       case SRP_DATA_DESC_DIRECT:
+               md = (struct srp_direct_buf *)(cmd->add_data + offset);
+               len = be32_to_cpu(md->len);
+               break;
+       case SRP_DATA_DESC_INDIRECT:
+               id = (struct srp_indirect_buf *)(cmd->add_data + offset);
+               len = be32_to_cpu(id->len);
+               break;
+       default:
+               pr_err("invalid data format %x\n", fmt);
+               break;
+       }
+       return len;
+}
+
+int srp_get_desc_table(struct srp_cmd *srp_cmd, enum dma_data_direction *dir,
+                      u64 *data_len)
+{
+       struct srp_indirect_buf *idb;
+       struct srp_direct_buf *db;
+       uint add_cdb_offset;
+       int rc;
+
+       /*
+        * The pointer computations below will only be compiled correctly
+        * if srp_cmd::add_data is declared as s8*, u8*, s8[] or u8[], so check
+        * whether srp_cmd::add_data has been declared as a byte pointer.
+        */
+       BUILD_BUG_ON(!__same_type(srp_cmd->add_data[0], (s8)0)
+                    && !__same_type(srp_cmd->add_data[0], (u8)0));
+
+       BUG_ON(!dir);
+       BUG_ON(!data_len);
+
+       rc = 0;
+       *data_len = 0;
+
+       *dir = DMA_NONE;
+
+       if (srp_cmd->buf_fmt & 0xf)
+               *dir = DMA_FROM_DEVICE;
+       else if (srp_cmd->buf_fmt >> 4)
+               *dir = DMA_TO_DEVICE;
+
+       add_cdb_offset = srp_cmd->add_cdb_len & ~3;
+       if (((srp_cmd->buf_fmt & 0xf) == SRP_DATA_DESC_DIRECT) ||
+           ((srp_cmd->buf_fmt >> 4) == SRP_DATA_DESC_DIRECT)) {
+               db = (struct srp_direct_buf *)(srp_cmd->add_data
+                                              + add_cdb_offset);
+               *data_len = be32_to_cpu(db->len);
+       } else if (((srp_cmd->buf_fmt & 0xf) == SRP_DATA_DESC_INDIRECT) ||
+                  ((srp_cmd->buf_fmt >> 4) == SRP_DATA_DESC_INDIRECT)) {
+               idb = (struct srp_indirect_buf *)(srp_cmd->add_data
+                                                 + add_cdb_offset);
+
+               *data_len = be32_to_cpu(idb->len);
+       }
+       return rc;
+}
+
+MODULE_DESCRIPTION("SCSI RDMA Protocol lib functions");
+MODULE_AUTHOR("FUJITA Tomonori");
+MODULE_LICENSE("GPL");
diff --git a/drivers/scsi/ibmvscsi_tgt/libsrp.h b/drivers/scsi/ibmvscsi_tgt/libsrp.h
new file mode 100644 (file)
index 0000000..4696f33
--- /dev/null
@@ -0,0 +1,123 @@
+#ifndef __LIBSRP_H__
+#define __LIBSRP_H__
+
+#include <linux/list.h>
+#include <linux/kfifo.h>
+#include <scsi/srp.h>
+
+enum srp_valid {
+       INVALIDATE_CMD_RESP_EL = 0,
+       VALID_CMD_RESP_EL = 0x80,
+       VALID_INIT_MSG = 0xC0,
+       VALID_TRANS_EVENT = 0xFF
+};
+
+enum srp_format {
+       SRP_FORMAT = 1,
+       MAD_FORMAT = 2,
+       OS400_FORMAT = 3,
+       AIX_FORMAT = 4,
+       LINUX_FORMAT = 5,
+       MESSAGE_IN_CRQ = 6
+};
+
+enum srp_init_msg {
+       INIT_MSG = 1,
+       INIT_COMPLETE_MSG = 2
+};
+
+enum srp_trans_event {
+       UNUSED_FORMAT = 0,
+       PARTNER_FAILED = 1,
+       PARTNER_DEREGISTER = 2,
+       MIGRATED = 6
+};
+
+enum srp_status {
+       HEADER_DESCRIPTOR = 0xF1,
+       PING = 0xF5,
+       PING_RESPONSE = 0xF6
+};
+
+enum srp_mad_version {
+       MAD_VERSION_1 = 1
+};
+
+enum srp_os_type {
+       OS400 = 1,
+       LINUX = 2,
+       AIX = 3,
+       OFW = 4
+};
+
+enum srp_task_attributes {
+       SRP_SIMPLE_TASK = 0,
+       SRP_HEAD_TASK = 1,
+       SRP_ORDERED_TASK = 2,
+       SRP_ACA_TASK = 4
+};
+
+enum {
+       SRP_TASK_MANAGEMENT_FUNCTION_COMPLETE           = 0,
+       SRP_REQUEST_FIELDS_INVALID                      = 2,
+       SRP_TASK_MANAGEMENT_FUNCTION_NOT_SUPPORTED      = 4,
+       SRP_TASK_MANAGEMENT_FUNCTION_FAILED             = 5
+};
+
+struct srp_buf {
+       dma_addr_t dma;
+       void *buf;
+};
+
+struct srp_queue {
+       void *pool;
+       void *items;
+       struct kfifo queue;
+       spinlock_t lock;
+};
+
+struct srp_target {
+       struct device *dev;
+
+       spinlock_t lock;
+       struct list_head cmd_queue;
+
+       size_t srp_iu_size;
+       struct srp_queue iu_queue;
+       size_t rx_ring_size;
+       struct srp_buf **rx_ring;
+
+       void *ldata;
+};
+
+struct iu_entry {
+       struct srp_target *target;
+
+       struct list_head ilist;
+       dma_addr_t remote_token;
+       unsigned long flags;
+
+       struct srp_buf *sbuf;
+       u16 iu_len;
+};
+
+struct ibmvscsis_cmd;
+
+typedef int (srp_rdma_t)(struct ibmvscsis_cmd *, struct scatterlist *, int,
+                        struct srp_direct_buf *, int,
+                        enum dma_data_direction, unsigned int);
+int srp_target_alloc(struct srp_target *, struct device *, size_t, size_t);
+void srp_target_free(struct srp_target *);
+struct iu_entry *srp_iu_get(struct srp_target *);
+void srp_iu_put(struct iu_entry *);
+int srp_transfer_data(struct ibmvscsis_cmd *, struct srp_cmd *,
+                     srp_rdma_t, int, int);
+u64 srp_data_length(struct srp_cmd *cmd, enum dma_data_direction dir);
+int srp_get_desc_table(struct srp_cmd *srp_cmd, enum dma_data_direction *dir,
+                      u64 *data_len);
+static inline int srp_cmd_direction(struct srp_cmd *cmd)
+{
+       return (cmd->buf_fmt >> 4) ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
+}
+
+#endif
index 50f3d3a0dd7b93d4e8789913997ebd67eeb70425..39b928c2849d71c47390ef55002dc767f7ae4e12 100644 (file)
@@ -492,7 +492,8 @@ void iscsit_aborted_task(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
        bool scsi_cmd = (cmd->iscsi_opcode == ISCSI_OP_SCSI_CMD);
 
        spin_lock_bh(&conn->cmd_lock);
-       if (!list_empty(&cmd->i_conn_node))
+       if (!list_empty(&cmd->i_conn_node) &&
+           !(cmd->se_cmd.transport_state & CMD_T_FABRIC_STOP))
                list_del_init(&cmd->i_conn_node);
        spin_unlock_bh(&conn->cmd_lock);
 
@@ -4034,6 +4035,7 @@ int iscsi_target_rx_thread(void *arg)
 
 static void iscsit_release_commands_from_conn(struct iscsi_conn *conn)
 {
+       LIST_HEAD(tmp_list);
        struct iscsi_cmd *cmd = NULL, *cmd_tmp = NULL;
        struct iscsi_session *sess = conn->sess;
        /*
@@ -4042,18 +4044,26 @@ static void iscsit_release_commands_from_conn(struct iscsi_conn *conn)
         * has been reset -> returned sleeping pre-handler state.
         */
        spin_lock_bh(&conn->cmd_lock);
-       list_for_each_entry_safe(cmd, cmd_tmp, &conn->conn_cmd_list, i_conn_node) {
+       list_splice_init(&conn->conn_cmd_list, &tmp_list);
 
+       list_for_each_entry(cmd, &tmp_list, i_conn_node) {
+               struct se_cmd *se_cmd = &cmd->se_cmd;
+
+               if (se_cmd->se_tfo != NULL) {
+                       spin_lock(&se_cmd->t_state_lock);
+                       se_cmd->transport_state |= CMD_T_FABRIC_STOP;
+                       spin_unlock(&se_cmd->t_state_lock);
+               }
+       }
+       spin_unlock_bh(&conn->cmd_lock);
+
+       list_for_each_entry_safe(cmd, cmd_tmp, &tmp_list, i_conn_node) {
                list_del_init(&cmd->i_conn_node);
-               spin_unlock_bh(&conn->cmd_lock);
 
                iscsit_increment_maxcmdsn(cmd, sess);
-
                iscsit_free_cmd(cmd, true);
 
-               spin_lock_bh(&conn->cmd_lock);
        }
-       spin_unlock_bh(&conn->cmd_lock);
 }
 
 static void iscsit_stop_timers_for_cmds(
index b5212f0f9571b5ce85c5c2413baac8f7d900b58a..adf419fa429189ceca94d04782221fc34de23b6c 100644 (file)
@@ -1371,8 +1371,9 @@ static int __iscsi_target_login_thread(struct iscsi_np *np)
        }
        login->zero_tsih = zero_tsih;
 
-       conn->sess->se_sess->sup_prot_ops =
-               conn->conn_transport->iscsit_get_sup_prot_ops(conn);
+       if (conn->sess)
+               conn->sess->se_sess->sup_prot_ops =
+                       conn->conn_transport->iscsit_get_sup_prot_ops(conn);
 
        tpg = conn->tpg;
        if (!tpg) {
index a4046ca6e60da85d5e0f489ae58981fd5ba45426..6b423485c5d6b4f6e8e54a332f1dd62eda0325d9 100644 (file)
@@ -821,13 +821,15 @@ struct se_device *target_alloc_device(struct se_hba *hba, const char *name)
  * in ATA and we need to set TPE=1
  */
 bool target_configure_unmap_from_queue(struct se_dev_attrib *attrib,
-                                      struct request_queue *q, int block_size)
+                                      struct request_queue *q)
 {
+       int block_size = queue_logical_block_size(q);
+
        if (!blk_queue_discard(q))
                return false;
 
-       attrib->max_unmap_lba_count = (q->limits.max_discard_sectors << 9) /
-                                                               block_size;
+       attrib->max_unmap_lba_count =
+               q->limits.max_discard_sectors >> (ilog2(block_size) - 9);
        /*
         * Currently hardcoded to 1 in Linux/SCSI code..
         */
index 75f0f08b2a34f32d0b2bfdd35fa40f027e74ae75..d545993df18be9ede3253861e24c25726d9a8e27 100644 (file)
@@ -161,8 +161,7 @@ static int fd_configure_device(struct se_device *dev)
                        dev_size, div_u64(dev_size, fd_dev->fd_block_size),
                        fd_dev->fd_block_size);
 
-               if (target_configure_unmap_from_queue(&dev->dev_attrib, q,
-                                                     fd_dev->fd_block_size))
+               if (target_configure_unmap_from_queue(&dev->dev_attrib, q))
                        pr_debug("IFILE: BLOCK Discard support available,"
                                 " disabled by default\n");
                /*
@@ -523,7 +522,7 @@ fd_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
         */
        if (cmd->data_length > FD_MAX_BYTES) {
                pr_err("FILEIO: Not able to process I/O of %u bytes due to"
-                      "FD_MAX_BYTES: %u iovec count limitiation\n",
+                      "FD_MAX_BYTES: %u iovec count limitation\n",
                        cmd->data_length, FD_MAX_BYTES);
                return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
        }
index 22af12f8b8eb7ff13d079fb3291c680bb3d6dec0..47cf6c977367ed2b10314a0156f2b5deaaa1ca59 100644 (file)
@@ -121,8 +121,7 @@ static int iblock_configure_device(struct se_device *dev)
        dev->dev_attrib.hw_max_sectors = queue_max_hw_sectors(q);
        dev->dev_attrib.hw_queue_depth = q->nr_requests;
 
-       if (target_configure_unmap_from_queue(&dev->dev_attrib, q,
-                                             dev->dev_attrib.hw_block_size))
+       if (target_configure_unmap_from_queue(&dev->dev_attrib, q))
                pr_debug("IBLOCK: BLOCK Discard support available,"
                         " disabled by default\n");
 
index fc91e85f54ba683dd298afc6db1d907192cef598..e2c970a9d61c32c7a95d034f889992a7560fdb64 100644 (file)
@@ -146,6 +146,7 @@ sense_reason_t      target_cmd_size_check(struct se_cmd *cmd, unsigned int size);
 void   target_qf_do_work(struct work_struct *work);
 bool   target_check_wce(struct se_device *dev);
 bool   target_check_fua(struct se_device *dev);
+void   __target_execute_cmd(struct se_cmd *, bool);
 
 /* target_core_stat.c */
 void   target_stat_setup_dev_default_groups(struct se_device *);
index a9057aa07176ff240748372215333dd531296ca2..04f616b3ba0a848a80d4a70c084c1b45d406c168 100644 (file)
@@ -602,7 +602,7 @@ static sense_reason_t compare_and_write_callback(struct se_cmd *cmd, bool succes
        cmd->transport_state |= CMD_T_ACTIVE|CMD_T_BUSY|CMD_T_SENT;
        spin_unlock_irq(&cmd->t_state_lock);
 
-       __target_execute_cmd(cmd);
+       __target_execute_cmd(cmd, false);
 
        kfree(buf);
        return ret;
index 5ab3967dda43ee989a6a4e5ccf1a553e67a7a0db..6094a6beddde9fb5d045644b6b11b32e6bd149c1 100644 (file)
@@ -754,7 +754,15 @@ EXPORT_SYMBOL(target_complete_cmd);
 
 void target_complete_cmd_with_length(struct se_cmd *cmd, u8 scsi_status, int length)
 {
-       if (scsi_status == SAM_STAT_GOOD && length < cmd->data_length) {
+       if (scsi_status != SAM_STAT_GOOD) {
+               return;
+       }
+
+       /*
+        * Calculate new residual count based upon length of SCSI data
+        * transferred.
+        */
+       if (length < cmd->data_length) {
                if (cmd->se_cmd_flags & SCF_UNDERFLOW_BIT) {
                        cmd->residual_count += cmd->data_length - length;
                } else {
@@ -763,6 +771,12 @@ void target_complete_cmd_with_length(struct se_cmd *cmd, u8 scsi_status, int len
                }
 
                cmd->data_length = length;
+       } else if (length > cmd->data_length) {
+               cmd->se_cmd_flags |= SCF_OVERFLOW_BIT;
+               cmd->residual_count = length - cmd->data_length;
+       } else {
+               cmd->se_cmd_flags &= ~(SCF_OVERFLOW_BIT | SCF_UNDERFLOW_BIT);
+               cmd->residual_count = 0;
        }
 
        target_complete_cmd(cmd, scsi_status);
@@ -1303,23 +1317,6 @@ target_setup_cmd_from_cdb(struct se_cmd *cmd, unsigned char *cdb)
 
        trace_target_sequencer_start(cmd);
 
-       /*
-        * Check for an existing UNIT ATTENTION condition
-        */
-       ret = target_scsi3_ua_check(cmd);
-       if (ret)
-               return ret;
-
-       ret = target_alua_state_check(cmd);
-       if (ret)
-               return ret;
-
-       ret = target_check_reservation(cmd);
-       if (ret) {
-               cmd->scsi_status = SAM_STAT_RESERVATION_CONFLICT;
-               return ret;
-       }
-
        ret = dev->transport->parse_cdb(cmd);
        if (ret == TCM_UNSUPPORTED_SCSI_OPCODE)
                pr_warn_ratelimited("%s/%s: Unsupported SCSI Opcode 0x%02x, sending CHECK_CONDITION.\n",
@@ -1761,20 +1758,45 @@ queue_full:
 }
 EXPORT_SYMBOL(transport_generic_request_failure);
 
-void __target_execute_cmd(struct se_cmd *cmd)
+void __target_execute_cmd(struct se_cmd *cmd, bool do_checks)
 {
        sense_reason_t ret;
 
-       if (cmd->execute_cmd) {
-               ret = cmd->execute_cmd(cmd);
-               if (ret) {
-                       spin_lock_irq(&cmd->t_state_lock);
-                       cmd->transport_state &= ~(CMD_T_BUSY|CMD_T_SENT);
-                       spin_unlock_irq(&cmd->t_state_lock);
+       if (!cmd->execute_cmd) {
+               ret = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
+               goto err;
+       }
+       if (do_checks) {
+               /*
+                * Check for an existing UNIT ATTENTION condition after
+                * target_handle_task_attr() has done SAM task attr
+                * checking, and possibly have already defered execution
+                * out to target_restart_delayed_cmds() context.
+                */
+               ret = target_scsi3_ua_check(cmd);
+               if (ret)
+                       goto err;
 
-                       transport_generic_request_failure(cmd, ret);
+               ret = target_alua_state_check(cmd);
+               if (ret)
+                       goto err;
+
+               ret = target_check_reservation(cmd);
+               if (ret) {
+                       cmd->scsi_status = SAM_STAT_RESERVATION_CONFLICT;
+                       goto err;
                }
        }
+
+       ret = cmd->execute_cmd(cmd);
+       if (!ret)
+               return;
+err:
+       spin_lock_irq(&cmd->t_state_lock);
+       cmd->transport_state &= ~(CMD_T_BUSY|CMD_T_SENT);
+       spin_unlock_irq(&cmd->t_state_lock);
+
+       transport_generic_request_failure(cmd, ret);
 }
 
 static int target_write_prot_action(struct se_cmd *cmd)
@@ -1819,6 +1841,8 @@ static bool target_handle_task_attr(struct se_cmd *cmd)
        if (dev->transport->transport_flags & TRANSPORT_FLAG_PASSTHROUGH)
                return false;
 
+       cmd->se_cmd_flags |= SCF_TASK_ATTR_SET;
+
        /*
         * Check for the existence of HEAD_OF_QUEUE, and if true return 1
         * to allow the passed struct se_cmd list of tasks to the front of the list.
@@ -1899,7 +1923,7 @@ void target_execute_cmd(struct se_cmd *cmd)
                return;
        }
 
-       __target_execute_cmd(cmd);
+       __target_execute_cmd(cmd, true);
 }
 EXPORT_SYMBOL(target_execute_cmd);
 
@@ -1923,7 +1947,7 @@ static void target_restart_delayed_cmds(struct se_device *dev)
                list_del(&cmd->se_delayed_node);
                spin_unlock(&dev->delayed_cmd_lock);
 
-               __target_execute_cmd(cmd);
+               __target_execute_cmd(cmd, true);
 
                if (cmd->sam_task_attr == TCM_ORDERED_TAG)
                        break;
@@ -1941,6 +1965,9 @@ static void transport_complete_task_attr(struct se_cmd *cmd)
        if (dev->transport->transport_flags & TRANSPORT_FLAG_PASSTHROUGH)
                return;
 
+       if (!(cmd->se_cmd_flags & SCF_TASK_ATTR_SET))
+               goto restart;
+
        if (cmd->sam_task_attr == TCM_SIMPLE_TAG) {
                atomic_dec_mb(&dev->simple_cmds);
                dev->dev_cur_ordered_id++;
@@ -1957,7 +1984,7 @@ static void transport_complete_task_attr(struct se_cmd *cmd)
                pr_debug("Incremented dev_cur_ordered_id: %u for ORDERED\n",
                         dev->dev_cur_ordered_id);
        }
-
+restart:
        target_restart_delayed_cmds(dev);
 }
 
@@ -2557,15 +2584,10 @@ static void target_release_cmd_kref(struct kref *kref)
        bool fabric_stop;
 
        spin_lock_irqsave(&se_sess->sess_cmd_lock, flags);
-       if (list_empty(&se_cmd->se_cmd_list)) {
-               spin_unlock_irqrestore(&se_sess->sess_cmd_lock, flags);
-               target_free_cmd_mem(se_cmd);
-               se_cmd->se_tfo->release_cmd(se_cmd);
-               return;
-       }
 
        spin_lock(&se_cmd->t_state_lock);
-       fabric_stop = (se_cmd->transport_state & CMD_T_FABRIC_STOP);
+       fabric_stop = (se_cmd->transport_state & CMD_T_FABRIC_STOP) &&
+                     (se_cmd->transport_state & CMD_T_ABORTED);
        spin_unlock(&se_cmd->t_state_lock);
 
        if (se_cmd->cmd_wait_set || fabric_stop) {
index f5186a74439950631c48f9735e3d9fbed1699c30..6ffbb603d9122a0259daa69db5bcca03ba891aa5 100644 (file)
@@ -91,6 +91,7 @@ static void ft_tport_delete(struct ft_tport *tport)
 
        ft_sess_delete_all(tport);
        lport = tport->lport;
+       lport->service_params &= ~FCP_SPPF_TARG_FCN;
        BUG_ON(tport != lport->prov[FC_TYPE_FCP]);
        RCU_INIT_POINTER(lport->prov[FC_TYPE_FCP], NULL);
 
@@ -110,6 +111,7 @@ void ft_lport_add(struct fc_lport *lport, void *arg)
 {
        mutex_lock(&ft_lport_lock);
        ft_tport_get(lport);
+       lport->service_params |= FCP_SPPF_TARG_FCN;
        mutex_unlock(&ft_lport_lock);
 }
 
diff --git a/include/scsi/viosrp.h b/include/scsi/viosrp.h
new file mode 100644 (file)
index 0000000..974e07b
--- /dev/null
@@ -0,0 +1,220 @@
+/*****************************************************************************/
+/* srp.h -- SCSI RDMA Protocol definitions                                   */
+/*                                                                           */
+/* Written By: Colin Devilbis, IBM Corporation                               */
+/*                                                                           */
+/* Copyright (C) 2003 IBM Corporation                                        */
+/*                                                                           */
+/* This program is free software; you can redistribute it and/or modify      */
+/* it under the terms of the GNU General Public License as published by      */
+/* the Free Software Foundation; either version 2 of the License, or         */
+/* (at your option) any later version.                                       */
+/*                                                                           */
+/* This program is distributed in the hope that it will be useful,           */
+/* but WITHOUT ANY WARRANTY; without even the implied warranty of            */
+/* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the             */
+/* GNU General Public License for more details.                              */
+/*                                                                           */
+/* This file contains structures and definitions for IBM RPA (RS/6000        */
+/* platform architecture) implementation of the SRP (SCSI RDMA Protocol)     */
+/* standard.  SRP is used on IBM iSeries and pSeries platforms to send SCSI  */
+/* commands between logical partitions.                                      */
+/*                                                                           */
+/* SRP Information Units (IUs) are sent on a "Command/Response Queue" (CRQ)  */
+/* between partitions.  The definitions in this file are architected,        */
+/* and cannot be changed without breaking compatibility with other versions  */
+/* of Linux and other operating systems (AIX, OS/400) that talk this protocol*/
+/* between logical partitions                                                */
+/*****************************************************************************/
+#ifndef VIOSRP_H
+#define VIOSRP_H
+#include <scsi/srp.h>
+
+#define SRP_VERSION "16.a"
+#define SRP_MAX_IU_LEN 256
+#define SRP_MAX_LOC_LEN 32
+
+union srp_iu {
+       struct srp_login_req login_req;
+       struct srp_login_rsp login_rsp;
+       struct srp_login_rej login_rej;
+       struct srp_i_logout i_logout;
+       struct srp_t_logout t_logout;
+       struct srp_tsk_mgmt tsk_mgmt;
+       struct srp_cmd cmd;
+       struct srp_rsp rsp;
+       u8 reserved[SRP_MAX_IU_LEN];
+};
+
+enum viosrp_crq_headers {
+       VIOSRP_CRQ_FREE = 0x00,
+       VIOSRP_CRQ_CMD_RSP = 0x80,
+       VIOSRP_CRQ_INIT_RSP = 0xC0,
+       VIOSRP_CRQ_XPORT_EVENT = 0xFF
+};
+
+enum viosrp_crq_init_formats {
+       VIOSRP_CRQ_INIT = 0x01,
+       VIOSRP_CRQ_INIT_COMPLETE = 0x02
+};
+
+enum viosrp_crq_formats {
+       VIOSRP_SRP_FORMAT = 0x01,
+       VIOSRP_MAD_FORMAT = 0x02,
+       VIOSRP_OS400_FORMAT = 0x03,
+       VIOSRP_AIX_FORMAT = 0x04,
+       VIOSRP_LINUX_FORMAT = 0x05,
+       VIOSRP_INLINE_FORMAT = 0x06
+};
+
+enum viosrp_crq_status {
+       VIOSRP_OK = 0x0,
+       VIOSRP_NONRECOVERABLE_ERR = 0x1,
+       VIOSRP_VIOLATES_MAX_XFER = 0x2,
+       VIOSRP_PARTNER_PANIC = 0x3,
+       VIOSRP_DEVICE_BUSY = 0x8,
+       VIOSRP_ADAPTER_FAIL = 0x10,
+       VIOSRP_OK2 = 0x99,
+};
+
+struct viosrp_crq {
+       u8 valid;               /* used by RPA */
+       u8 format;              /* SCSI vs out-of-band */
+       u8 reserved;
+       u8 status;              /* non-scsi failure? (e.g. DMA failure) */
+       __be16 timeout;         /* in seconds */
+       __be16 IU_length;               /* in bytes */
+       __be64 IU_data_ptr;     /* the TCE for transferring data */
+};
+
+/* MADs are Management requests above and beyond the IUs defined in the SRP
+ * standard.
+ */
+enum viosrp_mad_types {
+       VIOSRP_EMPTY_IU_TYPE = 0x01,
+       VIOSRP_ERROR_LOG_TYPE = 0x02,
+       VIOSRP_ADAPTER_INFO_TYPE = 0x03,
+       VIOSRP_CAPABILITIES_TYPE = 0x05,
+       VIOSRP_ENABLE_FAST_FAIL = 0x08,
+};
+
+enum viosrp_mad_status {
+       VIOSRP_MAD_SUCCESS = 0x00,
+       VIOSRP_MAD_NOT_SUPPORTED = 0xF1,
+       VIOSRP_MAD_FAILED = 0xF7,
+};
+
+enum viosrp_capability_type {
+       MIGRATION_CAPABILITIES = 0x01,
+       RESERVATION_CAPABILITIES = 0x02,
+};
+
+enum viosrp_capability_support {
+       SERVER_DOES_NOT_SUPPORTS_CAP = 0x0,
+       SERVER_SUPPORTS_CAP = 0x01,
+       SERVER_CAP_DATA = 0x02,
+};
+
+enum viosrp_reserve_type {
+       CLIENT_RESERVE_SCSI_2 = 0x01,
+};
+
+enum viosrp_capability_flag {
+       CLIENT_MIGRATED = 0x01,
+       CLIENT_RECONNECT = 0x02,
+       CAP_LIST_SUPPORTED = 0x04,
+       CAP_LIST_DATA = 0x08,
+};
+
+/*
+ * Common MAD header
+ */
+struct mad_common {
+       __be32 type;
+       __be16 status;
+       __be16 length;
+       __be64 tag;
+};
+
+/*
+ * All SRP (and MAD) requests normally flow from the
+ * client to the server.  There is no way for the server to send
+ * an asynchronous message back to the client.  The Empty IU is used
+ * to hang out a meaningless request to the server so that it can respond
+ * asynchrouously with something like a SCSI AER
+ */
+struct viosrp_empty_iu {
+       struct mad_common common;
+       __be64 buffer;
+       __be32 port;
+};
+
+struct viosrp_error_log {
+       struct mad_common common;
+       __be64 buffer;
+};
+
+struct viosrp_adapter_info {
+       struct mad_common common;
+       __be64 buffer;
+};
+
+struct viosrp_fast_fail {
+       struct mad_common common;
+};
+
+struct viosrp_capabilities {
+       struct mad_common common;
+       __be64 buffer;
+};
+
+struct mad_capability_common {
+       __be32 cap_type;
+       __be16 length;
+       __be16 server_support;
+};
+
+struct mad_reserve_cap {
+       struct mad_capability_common common;
+       __be32 type;
+};
+
+struct mad_migration_cap {
+       struct mad_capability_common common;
+       __be32 ecl;
+};
+
+struct capabilities {
+       __be32 flags;
+       char name[SRP_MAX_LOC_LEN];
+       char loc[SRP_MAX_LOC_LEN];
+       struct mad_migration_cap migration;
+       struct mad_reserve_cap reserve;
+};
+
+union mad_iu {
+       struct viosrp_empty_iu empty_iu;
+       struct viosrp_error_log error_log;
+       struct viosrp_adapter_info adapter_info;
+       struct viosrp_fast_fail fast_fail;
+       struct viosrp_capabilities capabilities;
+};
+
+union viosrp_iu {
+       union srp_iu srp;
+       union mad_iu mad;
+};
+
+struct mad_adapter_info_data {
+       char srp_version[8];
+       char partition_name[96];
+       __be32 partition_number;
+#define SRP_MAD_VERSION_1 1
+       __be32 mad_version;
+#define SRP_MAD_OS_LINUX 2
+#define SRP_MAD_OS_AIX 3
+       __be32 os_type;
+       __be32 port_max_txu[8]; /* per-port maximum transfer */
+};
+
+#endif
index d8ab5101fad5533876339f788b75d855d382bc4f..f6f3bc52c1ac2e21611ba7be2a274c7cb442166d 100644 (file)
@@ -95,6 +95,6 @@ sense_reason_t passthrough_parse_cdb(struct se_cmd *cmd,
 bool target_sense_desc_format(struct se_device *dev);
 sector_t target_to_linux_sector(struct se_device *dev, sector_t lb);
 bool target_configure_unmap_from_queue(struct se_dev_attrib *attrib,
-                                      struct request_queue *q, int block_size);
+                                      struct request_queue *q);
 
 #endif /* TARGET_CORE_BACKEND_H */
index b316b44d03f3667fea87daef334a3e5a5060f5bc..fb8e3b6febdff7f5fdb3f0335455b9e074647c40 100644 (file)
@@ -142,6 +142,7 @@ enum se_cmd_flags_table {
        SCF_PASSTHROUGH_PROT_SG_TO_MEM_NOALLOC = 0x00200000,
        SCF_ACK_KREF                    = 0x00400000,
        SCF_USE_CPUID                   = 0x00800000,
+       SCF_TASK_ATTR_SET               = 0x01000000,
 };
 
 /*
index de44462a7680c2071a4f37088e7241209de117f3..5cd6faa6e0d166ed07444cf5e3735e2626483205 100644 (file)
@@ -163,7 +163,6 @@ int core_tmr_alloc_req(struct se_cmd *, void *, u8, gfp_t);
 void   core_tmr_release_req(struct se_tmr_req *);
 int    transport_generic_handle_tmr(struct se_cmd *);
 void   transport_generic_request_failure(struct se_cmd *, sense_reason_t);
-void   __target_execute_cmd(struct se_cmd *);
 int    transport_lookup_tmr_lun(struct se_cmd *, u64);
 void   core_allocate_nexus_loss_ua(struct se_node_acl *acl);