[SCSI] scsi tgt: IBM eServer i/pSeries virtual SCSI target driver
authorFUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Thu, 30 Nov 2006 18:00:54 +0000 (03:00 +0900)
committerJames Bottomley <jejb@mulgrave.il.steeleye.com>
Sun, 3 Dec 2006 15:28:01 +0000 (09:28 -0600)
This is IBM Virtual SCSI target driver for tgt. The driver is based on
the original ibmvscsis driver:

http://lkml.org/lkml/2005/10/17/99

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Santiago Leon <santil@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
drivers/scsi/Kconfig
drivers/scsi/Makefile
drivers/scsi/ibmvscsi/Makefile
drivers/scsi/ibmvscsi/ibmvstgt.c [new file with mode: 0644]

index 5570cdbf8f111d889de5173a44124f0fed5c4df1..1b4bfaabcfbca4ea41d8bfb2c49a46ab41b8f8a8 100644 (file)
@@ -821,6 +821,20 @@ config SCSI_IBMVSCSI
          To compile this driver as a module, choose M here: the
          module will be called ibmvscsic.
 
+config SCSI_IBMVSCSIS
+       tristate "IBM Virtual SCSI Server support"
+       depends on PPC_PSERIES && SCSI_TGT && SCSI_SRP
+       help
+         This is the SRP target driver for IBM pSeries virtual environments.
+
+         The userspace component needed to initialize the driver and
+         documentation can be found:
+
+         http://stgt.berlios.de/
+
+         To compile this driver as a module, choose M here: the
+         module will be called ibmvstgt.
+
 config SCSI_INITIO
        tristate "Initio 9100U(W) support"
        depends on PCI && SCSI
index 00691c4c3a6f4fa3efa5fb58257f23d31c449878..bd7c9888f7f40ca31052e1fa62796fbb08fb9e37 100644 (file)
@@ -128,6 +128,7 @@ obj-$(CONFIG_SCSI_NSP32)    += nsp32.o
 obj-$(CONFIG_SCSI_IPR)         += ipr.o
 obj-$(CONFIG_SCSI_SRP)         += libsrp.o
 obj-$(CONFIG_SCSI_IBMVSCSI)    += ibmvscsi/
+obj-$(CONFIG_SCSI_IBMVSCSIS)   += ibmvscsi/
 obj-$(CONFIG_SCSI_HPTIOP)      += hptiop.o
 obj-$(CONFIG_SCSI_STEX)                += stex.o
 
index 4e247b6b8700b9b6c0d3195e83b0c2e4f7eeeeb6..6ac0633d5452af8c36a0683653485980335950e3 100644 (file)
@@ -3,3 +3,5 @@ obj-$(CONFIG_SCSI_IBMVSCSI)     += ibmvscsic.o
 ibmvscsic-y                    += ibmvscsi.o
 ibmvscsic-$(CONFIG_PPC_ISERIES)        += iseries_vscsi.o 
 ibmvscsic-$(CONFIG_PPC_PSERIES)        += rpa_vscsi.o 
+
+obj-$(CONFIG_SCSI_IBMVSCSIS)   += ibmvstgt.o
diff --git a/drivers/scsi/ibmvscsi/ibmvstgt.c b/drivers/scsi/ibmvscsi/ibmvstgt.c
new file mode 100644 (file)
index 0000000..0e74174
--- /dev/null
@@ -0,0 +1,958 @@
+/*
+ * IBM eServer i/pSeries 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) 2005-2006 FUJITA Tomonori <tomof@acm.org>
+ *
+ * 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
+ */
+#include <linux/interrupt.h>
+#include <linux/module.h>
+#include <scsi/scsi.h>
+#include <scsi/scsi_host.h>
+#include <scsi/scsi_tgt.h>
+#include <scsi/libsrp.h>
+#include <asm/hvcall.h>
+#include <asm/iommu.h>
+#include <asm/prom.h>
+#include <asm/vio.h>
+
+#include "ibmvscsi.h"
+
+#define        INITIAL_SRP_LIMIT       16
+#define        DEFAULT_MAX_SECTORS     512
+
+#define        TGT_NAME        "ibmvstgt"
+
+/*
+ * Hypervisor calls.
+ */
+#define h_copy_rdma(l, sa, sb, da, db) \
+                       plpar_hcall_norets(H_COPY_RDMA, l, sa, sb, da, db)
+#define h_send_crq(ua, l, h) \
+                       plpar_hcall_norets(H_SEND_CRQ, ua, l, h)
+#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);
+
+/* tmp - will replace with SCSI logging stuff */
+#define eprintk(fmt, args...)                                  \
+do {                                                           \
+       printk("%s(%d) " fmt, __FUNCTION__, __LINE__, ##args);  \
+} while (0)
+/* #define dprintk eprintk */
+#define dprintk(fmt, args...)
+
+struct vio_port {
+       struct vio_dev *dma_dev;
+
+       struct crq_queue crq_queue;
+       struct work_struct crq_work;
+
+       unsigned long liobn;
+       unsigned long riobn;
+};
+
+static struct workqueue_struct *vtgtd;
+
+/*
+ * These are fixed for the system and come from the Open Firmware device tree.
+ * We just store them here to save getting them every time.
+ */
+static char system_id[64] = "";
+static char partition_name[97] = "UNKNOWN";
+static unsigned int partition_number = -1;
+
+static struct vio_port *target_to_port(struct srp_target *target)
+{
+       return (struct vio_port *) target->ldata;
+}
+
+static inline union viosrp_iu *vio_iu(struct iu_entry *iue)
+{
+       return (union viosrp_iu *) (iue->sbuf->buf);
+}
+
+static int send_iu(struct iu_entry *iue, uint64_t length, uint8_t format)
+{
+       struct srp_target *target = iue->target;
+       struct vio_port *vport = target_to_port(target);
+       long rc, rc1;
+       union {
+               struct viosrp_crq cooked;
+               uint64_t raw[2];
+       } crq;
+
+       /* First copy the SRP */
+       rc = h_copy_rdma(length, vport->liobn, iue->sbuf->dma,
+                        vport->riobn, iue->remote_token);
+
+       if (rc)
+               eprintk("Error %ld transferring data\n", rc);
+
+       crq.cooked.valid = 0x80;
+       crq.cooked.format = format;
+       crq.cooked.reserved = 0x00;
+       crq.cooked.timeout = 0x00;
+       crq.cooked.IU_length = length;
+       crq.cooked.IU_data_ptr = vio_iu(iue)->srp.rsp.tag;
+
+       if (rc == 0)
+               crq.cooked.status = 0x99;       /* Just needs to be non-zero */
+       else
+               crq.cooked.status = 0x00;
+
+       rc1 = h_send_crq(vport->dma_dev->unit_address, crq.raw[0], crq.raw[1]);
+
+       if (rc1) {
+               eprintk("%ld sending response\n", rc1);
+               return rc1;
+       }
+
+       return rc;
+}
+
+#define SRP_RSP_SENSE_DATA_LEN 18
+
+static int send_rsp(struct iu_entry *iue, struct scsi_cmnd *sc,
+                   unsigned char status, unsigned char asc)
+{
+       union viosrp_iu *iu = vio_iu(iue);
+       uint64_t tag = iu->srp.rsp.tag;
+
+       /* If the linked bit is on and status is good */
+       if (test_bit(V_LINKED, &iue->flags) && (status == NO_SENSE))
+               status = 0x10;
+
+       memset(iu, 0, sizeof(struct srp_rsp));
+       iu->srp.rsp.opcode = SRP_RSP;
+       iu->srp.rsp.req_lim_delta = 1;
+       iu->srp.rsp.tag = tag;
+
+       if (test_bit(V_DIOVER, &iue->flags))
+               iu->srp.rsp.flags |= SRP_RSP_FLAG_DIOVER;
+
+       iu->srp.rsp.data_in_res_cnt = 0;
+       iu->srp.rsp.data_out_res_cnt = 0;
+
+       iu->srp.rsp.flags &= ~SRP_RSP_FLAG_RSPVALID;
+
+       iu->srp.rsp.resp_data_len = 0;
+       iu->srp.rsp.status = status;
+       if (status) {
+               uint8_t *sense = iu->srp.rsp.data;
+
+               if (sc) {
+                       iu->srp.rsp.flags |= SRP_RSP_FLAG_SNSVALID;
+                       iu->srp.rsp.sense_data_len = SCSI_SENSE_BUFFERSIZE;
+                       memcpy(sense, sc->sense_buffer, SCSI_SENSE_BUFFERSIZE);
+               } else {
+                       iu->srp.rsp.status = SAM_STAT_CHECK_CONDITION;
+                       iu->srp.rsp.flags |= SRP_RSP_FLAG_SNSVALID;
+                       iu->srp.rsp.sense_data_len = SRP_RSP_SENSE_DATA_LEN;
+
+                       /* Valid bit and 'current errors' */
+                       sense[0] = (0x1 << 7 | 0x70);
+                       /* Sense key */
+                       sense[2] = status;
+                       /* Additional sense length */
+                       sense[7] = 0xa; /* 10 bytes */
+                       /* Additional sense code */
+                       sense[12] = asc;
+               }
+       }
+
+       send_iu(iue, sizeof(iu->srp.rsp) + SRP_RSP_SENSE_DATA_LEN,
+               VIOSRP_SRP_FORMAT);
+
+       return 0;
+}
+
+static void handle_cmd_queue(struct srp_target *target)
+{
+       struct Scsi_Host *shost = target->shost;
+       struct iu_entry *iue;
+       struct srp_cmd *cmd;
+       unsigned long flags;
+       int err;
+
+retry:
+       spin_lock_irqsave(&target->lock, flags);
+
+       list_for_each_entry(iue, &target->cmd_queue, ilist) {
+               if (!test_and_set_bit(V_FLYING, &iue->flags)) {
+                       spin_unlock_irqrestore(&target->lock, flags);
+                       cmd = iue->sbuf->buf;
+                       err = srp_cmd_queue(shost, cmd, iue, 0);
+                       if (err) {
+                               eprintk("cannot queue cmd %p %d\n", cmd, err);
+                               srp_iu_put(iue);
+                       }
+                       goto retry;
+               }
+       }
+
+       spin_unlock_irqrestore(&target->lock, flags);
+}
+
+static int ibmvstgt_rdma(struct scsi_cmnd *sc, struct scatterlist *sg, int nsg,
+                        struct srp_direct_buf *md, int nmd,
+                        enum dma_data_direction dir, unsigned int rest)
+{
+       struct iu_entry *iue = (struct iu_entry *) sc->SCp.ptr;
+       struct srp_target *target = iue->target;
+       struct vio_port *vport = target_to_port(target);
+       dma_addr_t token;
+       long err;
+       unsigned int done = 0;
+       int i, sidx, soff;
+
+       sidx = soff = 0;
+       token = sg_dma_address(sg + sidx);
+
+       for (i = 0; i < nmd && rest; i++) {
+               unsigned int mdone, mlen;
+
+               mlen = min(rest, md[i].len);
+               for (mdone = 0; mlen;) {
+                       int slen = min(sg_dma_len(sg + sidx) - soff, mlen);
+
+                       if (dir == DMA_TO_DEVICE)
+                               err = h_copy_rdma(slen,
+                                                 vport->riobn,
+                                                 md[i].va + mdone,
+                                                 vport->liobn,
+                                                 token + soff);
+                       else
+                               err = h_copy_rdma(slen,
+                                                 vport->liobn,
+                                                 token + soff,
+                                                 vport->riobn,
+                                                 md[i].va + mdone);
+
+                       if (err != H_SUCCESS) {
+                               eprintk("rdma error %d %d\n", dir, slen);
+                               goto out;
+                       }
+
+                       mlen -= slen;
+                       mdone += slen;
+                       soff += slen;
+                       done += slen;
+
+                       if (soff == sg_dma_len(sg + sidx)) {
+                               sidx++;
+                               soff = 0;
+                               token = sg_dma_address(sg + sidx);
+
+                               if (sidx > nsg) {
+                                       eprintk("out of sg %p %d %d\n",
+                                               iue, sidx, nsg);
+                                       goto out;
+                               }
+                       }
+               };
+
+               rest -= mlen;
+       }
+out:
+
+       return 0;
+}
+
+static int ibmvstgt_transfer_data(struct scsi_cmnd *sc,
+                                 void (*done)(struct scsi_cmnd *))
+{
+       struct iu_entry *iue = (struct iu_entry *) sc->SCp.ptr;
+       int err;
+
+       err = srp_transfer_data(sc, &vio_iu(iue)->srp.cmd, ibmvstgt_rdma, 1, 1);
+
+       done(sc);
+
+       return err;
+}
+
+static int ibmvstgt_cmd_done(struct scsi_cmnd *sc,
+                            void (*done)(struct scsi_cmnd *))
+{
+       unsigned long flags;
+       struct iu_entry *iue = (struct iu_entry *) sc->SCp.ptr;
+       struct srp_target *target = iue->target;
+
+       dprintk("%p %p %x\n", iue, target, vio_iu(iue)->srp.cmd.cdb[0]);
+
+       spin_lock_irqsave(&target->lock, flags);
+       list_del(&iue->ilist);
+       spin_unlock_irqrestore(&target->lock, flags);
+
+       if (sc->result != SAM_STAT_GOOD) {
+               eprintk("operation failed %p %d %x\n",
+                       iue, sc->result, vio_iu(iue)->srp.cmd.cdb[0]);
+               send_rsp(iue, sc, HARDWARE_ERROR, 0x00);
+       } else
+               send_rsp(iue, sc, NO_SENSE, 0x00);
+
+       done(sc);
+       srp_iu_put(iue);
+       return 0;
+}
+
+int send_adapter_info(struct iu_entry *iue,
+                     dma_addr_t remote_buffer, uint16_t length)
+{
+       struct srp_target *target = iue->target;
+       struct vio_port *vport = target_to_port(target);
+       struct Scsi_Host *shost = target->shost;
+       dma_addr_t data_token;
+       struct mad_adapter_info_data *info;
+       int err;
+
+       info = dma_alloc_coherent(target->dev, sizeof(*info), &data_token,
+                                 GFP_KERNEL);
+       if (!info) {
+               eprintk("bad dma_alloc_coherent %p\n", target);
+               return 1;
+       }
+
+       /* Get remote info */
+       err = h_copy_rdma(sizeof(*info), vport->riobn, remote_buffer,
+                         vport->liobn, data_token);
+       if (err == H_SUCCESS) {
+               dprintk("Client connect: %s (%d)\n",
+                       info->partition_name, info->partition_number);
+       }
+
+       memset(info, 0, sizeof(*info));
+
+       strcpy(info->srp_version, "16.a");
+       strncpy(info->partition_name, partition_name,
+               sizeof(info->partition_name));
+       info->partition_number = partition_number;
+       info->mad_version = 1;
+       info->os_type = 2;
+       info->port_max_txu[0] = shost->hostt->max_sectors << 9;
+
+       /* Send our info to remote */
+       err = h_copy_rdma(sizeof(*info), vport->liobn, data_token,
+                         vport->riobn, remote_buffer);
+
+       dma_free_coherent(target->dev, sizeof(*info), info, data_token);
+
+       if (err != H_SUCCESS) {
+               eprintk("Error sending adapter info %d\n", err);
+               return 1;
+       }
+
+       return 0;
+}
+
+static void process_login(struct iu_entry *iue)
+{
+       union viosrp_iu *iu = vio_iu(iue);
+       struct srp_login_rsp *rsp = &iu->srp.login_rsp;
+       uint64_t tag = iu->srp.rsp.tag;
+
+       /* TODO handle case that requested size is wrong and
+        * buffer format is wrong
+        */
+       memset(iu, 0, sizeof(struct srp_login_rsp));
+       rsp->opcode = SRP_LOGIN_RSP;
+       rsp->req_lim_delta = INITIAL_SRP_LIMIT;
+       rsp->tag = tag;
+       rsp->max_it_iu_len = sizeof(union srp_iu);
+       rsp->max_ti_iu_len = sizeof(union srp_iu);
+       /* direct and indirect */
+       rsp->buf_fmt = SRP_BUF_FORMAT_DIRECT | SRP_BUF_FORMAT_INDIRECT;
+
+       send_iu(iue, sizeof(*rsp), VIOSRP_SRP_FORMAT);
+}
+
+static inline void queue_cmd(struct iu_entry *iue)
+{
+       struct srp_target *target = iue->target;
+       unsigned long flags;
+
+       spin_lock_irqsave(&target->lock, flags);
+       list_add_tail(&iue->ilist, &target->cmd_queue);
+       spin_unlock_irqrestore(&target->lock, flags);
+}
+
+static int process_tsk_mgmt(struct iu_entry *iue)
+{
+       union viosrp_iu *iu = vio_iu(iue);
+       int fn;
+
+       dprintk("%p %u\n", iue, iu->srp.tsk_mgmt.tsk_mgmt_func);
+
+       switch (iu->srp.tsk_mgmt.tsk_mgmt_func) {
+       case SRP_TSK_ABORT_TASK:
+               fn = ABORT_TASK;
+               break;
+       case SRP_TSK_ABORT_TASK_SET:
+               fn = ABORT_TASK_SET;
+               break;
+       case SRP_TSK_CLEAR_TASK_SET:
+               fn = CLEAR_TASK_SET;
+               break;
+       case SRP_TSK_LUN_RESET:
+               fn = LOGICAL_UNIT_RESET;
+               break;
+       case SRP_TSK_CLEAR_ACA:
+               fn = CLEAR_ACA;
+               break;
+       default:
+               fn = 0;
+       }
+       if (fn)
+               scsi_tgt_tsk_mgmt_request(iue->target->shost, fn,
+                                         iu->srp.tsk_mgmt.task_tag,
+                                         (struct scsi_lun *) &iu->srp.tsk_mgmt.lun,
+                                         iue);
+       else
+               send_rsp(iue, NULL, ILLEGAL_REQUEST, 0x20);
+
+       return !fn;
+}
+
+static int process_mad_iu(struct iu_entry *iue)
+{
+       union viosrp_iu *iu = vio_iu(iue);
+       struct viosrp_adapter_info *info;
+       struct viosrp_host_config *conf;
+
+       switch (iu->mad.empty_iu.common.type) {
+       case VIOSRP_EMPTY_IU_TYPE:
+               eprintk("%s\n", "Unsupported EMPTY MAD IU");
+               break;
+       case VIOSRP_ERROR_LOG_TYPE:
+               eprintk("%s\n", "Unsupported ERROR LOG MAD IU");
+               iu->mad.error_log.common.status = 1;
+               send_iu(iue, sizeof(iu->mad.error_log), VIOSRP_MAD_FORMAT);
+               break;
+       case VIOSRP_ADAPTER_INFO_TYPE:
+               info = &iu->mad.adapter_info;
+               info->common.status = send_adapter_info(iue, info->buffer,
+                                                       info->common.length);
+               send_iu(iue, sizeof(*info), VIOSRP_MAD_FORMAT);
+               break;
+       case VIOSRP_HOST_CONFIG_TYPE:
+               conf = &iu->mad.host_config;
+               conf->common.status = 1;
+               send_iu(iue, sizeof(*conf), VIOSRP_MAD_FORMAT);
+               break;
+       default:
+               eprintk("Unknown type %u\n", iu->srp.rsp.opcode);
+       }
+
+       return 1;
+}
+
+static int process_srp_iu(struct iu_entry *iue)
+{
+       union viosrp_iu *iu = vio_iu(iue);
+       int done = 1;
+       u8 opcode = iu->srp.rsp.opcode;
+
+       switch (opcode) {
+       case SRP_LOGIN_REQ:
+               process_login(iue);
+               break;
+       case SRP_TSK_MGMT:
+               done = process_tsk_mgmt(iue);
+               break;
+       case SRP_CMD:
+               queue_cmd(iue);
+               done = 0;
+               break;
+       case SRP_LOGIN_RSP:
+       case SRP_I_LOGOUT:
+       case SRP_T_LOGOUT:
+       case SRP_RSP:
+       case SRP_CRED_REQ:
+       case SRP_CRED_RSP:
+       case SRP_AER_REQ:
+       case SRP_AER_RSP:
+               eprintk("Unsupported type %u\n", opcode);
+               break;
+       default:
+               eprintk("Unknown type %u\n", opcode);
+       }
+
+       return done;
+}
+
+static void process_iu(struct viosrp_crq *crq, struct srp_target *target)
+{
+       struct vio_port *vport = target_to_port(target);
+       struct iu_entry *iue;
+       long err, done;
+
+       iue = srp_iu_get(target);
+       if (!iue) {
+               eprintk("Error getting IU from pool, %p\n", target);
+               return;
+       }
+
+       iue->remote_token = crq->IU_data_ptr;
+
+       err = h_copy_rdma(crq->IU_length, vport->riobn,
+                         iue->remote_token, vport->liobn, iue->sbuf->dma);
+
+       if (err != H_SUCCESS) {
+               eprintk("%ld transferring data error %p\n", err, iue);
+               done = 1;
+               goto out;
+       }
+
+       if (crq->format == VIOSRP_MAD_FORMAT)
+               done = process_mad_iu(iue);
+       else
+               done = process_srp_iu(iue);
+out:
+       if (done)
+               srp_iu_put(iue);
+}
+
+static irqreturn_t ibmvstgt_interrupt(int irq, void *data)
+{
+       struct srp_target *target = (struct srp_target *) data;
+       struct vio_port *vport = target_to_port(target);
+
+       vio_disable_interrupts(vport->dma_dev);
+       queue_work(vtgtd, &vport->crq_work);
+
+       return IRQ_HANDLED;
+}
+
+static int crq_queue_create(struct crq_queue *queue, struct srp_target *target)
+{
+       int err;
+       struct vio_port *vport = target_to_port(target);
+
+       queue->msgs = (struct viosrp_crq *) get_zeroed_page(GFP_KERNEL);
+       if (!queue->msgs)
+               goto malloc_failed;
+       queue->size = PAGE_SIZE / sizeof(*queue->msgs);
+
+       queue->msg_token = dma_map_single(target->dev, queue->msgs,
+                                         queue->size * sizeof(*queue->msgs),
+                                         DMA_BIDIRECTIONAL);
+
+       if (dma_mapping_error(queue->msg_token))
+               goto map_failed;
+
+       err = h_reg_crq(vport->dma_dev->unit_address, queue->msg_token,
+                       PAGE_SIZE);
+
+       /* If the adapter was left active for some reason (like kexec)
+        * try freeing and re-registering
+        */
+       if (err == H_RESOURCE) {
+           do {
+               err = h_free_crq(vport->dma_dev->unit_address);
+           } while (err == H_BUSY || H_IS_LONG_BUSY(err));
+
+           err = h_reg_crq(vport->dma_dev->unit_address, queue->msg_token,
+                           PAGE_SIZE);
+       }
+
+       if (err != H_SUCCESS && err != 2) {
+               eprintk("Error 0x%x opening virtual adapter\n", err);
+               goto reg_crq_failed;
+       }
+
+       err = request_irq(vport->dma_dev->irq, &ibmvstgt_interrupt,
+                         SA_INTERRUPT, "ibmvstgt", target);
+       if (err)
+               goto req_irq_failed;
+
+       vio_enable_interrupts(vport->dma_dev);
+
+       h_send_crq(vport->dma_dev->unit_address, 0xC001000000000000, 0);
+
+       queue->cur = 0;
+       spin_lock_init(&queue->lock);
+
+       return 0;
+
+req_irq_failed:
+       do {
+               err = h_free_crq(vport->dma_dev->unit_address);
+       } while (err == H_BUSY || H_IS_LONG_BUSY(err));
+
+reg_crq_failed:
+       dma_unmap_single(target->dev, queue->msg_token,
+                        queue->size * sizeof(*queue->msgs), DMA_BIDIRECTIONAL);
+map_failed:
+       free_page((unsigned long) queue->msgs);
+
+malloc_failed:
+       return -ENOMEM;
+}
+
+static void crq_queue_destroy(struct srp_target *target)
+{
+       struct vio_port *vport = target_to_port(target);
+       struct crq_queue *queue = &vport->crq_queue;
+       int err;
+
+       free_irq(vport->dma_dev->irq, target);
+       do {
+               err = h_free_crq(vport->dma_dev->unit_address);
+       } while (err == H_BUSY || H_IS_LONG_BUSY(err));
+
+       dma_unmap_single(target->dev, queue->msg_token,
+                        queue->size * sizeof(*queue->msgs), DMA_BIDIRECTIONAL);
+
+       free_page((unsigned long) queue->msgs);
+}
+
+static void process_crq(struct viosrp_crq *crq,        struct srp_target *target)
+{
+       struct vio_port *vport = target_to_port(target);
+       dprintk("%x %x\n", crq->valid, crq->format);
+
+       switch (crq->valid) {
+       case 0xC0:
+               /* initialization */
+               switch (crq->format) {
+               case 0x01:
+                       h_send_crq(vport->dma_dev->unit_address,
+                                  0xC002000000000000, 0);
+                       break;
+               case 0x02:
+                       break;
+               default:
+                       eprintk("Unknown format %u\n", crq->format);
+               }
+               break;
+       case 0xFF:
+               /* transport event */
+               break;
+       case 0x80:
+               /* real payload */
+               switch (crq->format) {
+               case VIOSRP_SRP_FORMAT:
+               case VIOSRP_MAD_FORMAT:
+                       process_iu(crq, target);
+                       break;
+               case VIOSRP_OS400_FORMAT:
+               case VIOSRP_AIX_FORMAT:
+               case VIOSRP_LINUX_FORMAT:
+               case VIOSRP_INLINE_FORMAT:
+                       eprintk("Unsupported format %u\n", crq->format);
+                       break;
+               default:
+                       eprintk("Unknown format %u\n", crq->format);
+               }
+               break;
+       default:
+               eprintk("unknown message type 0x%02x!?\n", crq->valid);
+       }
+}
+
+static inline struct viosrp_crq *next_crq(struct crq_queue *queue)
+{
+       struct viosrp_crq *crq;
+       unsigned long flags;
+
+       spin_lock_irqsave(&queue->lock, flags);
+       crq = &queue->msgs[queue->cur];
+       if (crq->valid & 0x80) {
+               if (++queue->cur == queue->size)
+                       queue->cur = 0;
+       } else
+               crq = NULL;
+       spin_unlock_irqrestore(&queue->lock, flags);
+
+       return crq;
+}
+
+static void handle_crq(void *data)
+{
+       struct srp_target *target = (struct srp_target *) data;
+       struct vio_port *vport = target_to_port(target);
+       struct viosrp_crq *crq;
+       int done = 0;
+
+       while (!done) {
+               while ((crq = next_crq(&vport->crq_queue)) != NULL) {
+                       process_crq(crq, target);
+                       crq->valid = 0x00;
+               }
+
+               vio_enable_interrupts(vport->dma_dev);
+
+               crq = next_crq(&vport->crq_queue);
+               if (crq) {
+                       vio_disable_interrupts(vport->dma_dev);
+                       process_crq(crq, target);
+                       crq->valid = 0x00;
+               } else
+                       done = 1;
+       }
+
+       handle_cmd_queue(target);
+}
+
+
+static int ibmvstgt_eh_abort_handler(struct scsi_cmnd *sc)
+{
+       unsigned long flags;
+       struct iu_entry *iue = (struct iu_entry *) sc->SCp.ptr;
+       struct srp_target *target = iue->target;
+
+       dprintk("%p %p %x\n", iue, target, vio_iu(iue)->srp.cmd.cdb[0]);
+
+       spin_lock_irqsave(&target->lock, flags);
+       list_del(&iue->ilist);
+       spin_unlock_irqrestore(&target->lock, flags);
+
+       srp_iu_put(iue);
+
+       return 0;
+}
+
+static int ibmvstgt_tsk_mgmt_response(u64 mid, int result)
+{
+       struct iu_entry *iue = (struct iu_entry *) ((void *) mid);
+       union viosrp_iu *iu = vio_iu(iue);
+       unsigned char status, asc;
+
+       eprintk("%p %d\n", iue, result);
+       status = NO_SENSE;
+       asc = 0;
+
+       switch (iu->srp.tsk_mgmt.tsk_mgmt_func) {
+       case SRP_TSK_ABORT_TASK:
+               asc = 0x14;
+               if (result)
+                       status = ABORTED_COMMAND;
+               break;
+       default:
+               break;
+       }
+
+       send_rsp(iue, NULL, status, asc);
+       srp_iu_put(iue);
+
+       return 0;
+}
+
+static ssize_t system_id_show(struct class_device *cdev, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%s\n", system_id);
+}
+
+static ssize_t partition_number_show(struct class_device *cdev, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%x\n", partition_number);
+}
+
+static ssize_t unit_address_show(struct class_device *cdev, char *buf)
+{
+       struct Scsi_Host *shost = class_to_shost(cdev);
+       struct srp_target *target = host_to_srp_target(shost);
+       struct vio_port *vport = target_to_port(target);
+       return snprintf(buf, PAGE_SIZE, "%x\n", vport->dma_dev->unit_address);
+}
+
+static CLASS_DEVICE_ATTR(system_id, S_IRUGO, system_id_show, NULL);
+static CLASS_DEVICE_ATTR(partition_number, S_IRUGO, partition_number_show, NULL);
+static CLASS_DEVICE_ATTR(unit_address, S_IRUGO, unit_address_show, NULL);
+
+static struct class_device_attribute *ibmvstgt_attrs[] = {
+       &class_device_attr_system_id,
+       &class_device_attr_partition_number,
+       &class_device_attr_unit_address,
+       NULL,
+};
+
+static struct scsi_host_template ibmvstgt_sht = {
+       .name                   = TGT_NAME,
+       .module                 = THIS_MODULE,
+       .can_queue              = INITIAL_SRP_LIMIT,
+       .sg_tablesize           = SG_ALL,
+       .use_clustering         = DISABLE_CLUSTERING,
+       .max_sectors            = DEFAULT_MAX_SECTORS,
+       .transfer_response      = ibmvstgt_cmd_done,
+       .transfer_data          = ibmvstgt_transfer_data,
+       .eh_abort_handler       = ibmvstgt_eh_abort_handler,
+       .tsk_mgmt_response      = ibmvstgt_tsk_mgmt_response,
+       .shost_attrs            = ibmvstgt_attrs,
+       .proc_name              = TGT_NAME,
+};
+
+static int ibmvstgt_probe(struct vio_dev *dev, const struct vio_device_id *id)
+{
+       struct Scsi_Host *shost;
+       struct srp_target *target;
+       struct vio_port *vport;
+       unsigned int *dma, dma_size;
+       int err = -ENOMEM;
+
+       vport = kzalloc(sizeof(struct vio_port), GFP_KERNEL);
+       if (!vport)
+               return err;
+       shost = scsi_host_alloc(&ibmvstgt_sht, sizeof(struct srp_target));
+       if (!shost)
+               goto free_vport;
+       err = scsi_tgt_alloc_queue(shost);
+       if (err)
+               goto put_host;
+
+       target = host_to_srp_target(shost);
+       target->shost = shost;
+       vport->dma_dev = dev;
+       target->ldata = vport;
+       err = srp_target_alloc(target, &dev->dev, INITIAL_SRP_LIMIT,
+                              SRP_MAX_IU_LEN);
+       if (err)
+               goto put_host;
+
+       dma = (unsigned int *) vio_get_attribute(dev, "ibm,my-dma-window",
+                                                &dma_size);
+       if (!dma || dma_size != 40) {
+               eprintk("Couldn't get window property %d\n", dma_size);
+               err = -EIO;
+               goto free_srp_target;
+       }
+       vport->liobn = dma[0];
+       vport->riobn = dma[5];
+
+       INIT_WORK(&vport->crq_work, handle_crq, target);
+
+       err = crq_queue_create(&vport->crq_queue, target);
+       if (err)
+               goto free_srp_target;
+
+       err = scsi_add_host(shost, target->dev);
+       if (err)
+               goto destroy_queue;
+       return 0;
+
+destroy_queue:
+       crq_queue_destroy(target);
+free_srp_target:
+       srp_target_free(target);
+put_host:
+       scsi_host_put(shost);
+free_vport:
+       kfree(vport);
+       return err;
+}
+
+static int ibmvstgt_remove(struct vio_dev *dev)
+{
+       struct srp_target *target = (struct srp_target *) dev->dev.driver_data;
+       struct Scsi_Host *shost = target->shost;
+       struct vio_port *vport = target->ldata;
+
+       crq_queue_destroy(target);
+       scsi_remove_host(shost);
+       scsi_tgt_free_queue(shost);
+       srp_target_free(target);
+       kfree(vport);
+       scsi_host_put(shost);
+       return 0;
+}
+
+static struct vio_device_id ibmvstgt_device_table[] __devinitdata = {
+       {"v-scsi-host", "IBM,v-scsi-host"},
+       {"",""}
+};
+
+MODULE_DEVICE_TABLE(vio, ibmvstgt_device_table);
+
+static struct vio_driver ibmvstgt_driver = {
+       .id_table = ibmvstgt_device_table,
+       .probe = ibmvstgt_probe,
+       .remove = ibmvstgt_remove,
+       .driver = {
+               .name = "ibmvscsis",
+               .owner = THIS_MODULE,
+       }
+};
+
+static int get_system_info(void)
+{
+       struct device_node *rootdn;
+       const char *id, *model, *name;
+       unsigned int *num;
+
+       rootdn = find_path_device("/");
+       if (!rootdn)
+               return -ENOENT;
+
+       model = get_property(rootdn, "model", NULL);
+       id = get_property(rootdn, "system-id", NULL);
+       if (model && id)
+               snprintf(system_id, sizeof(system_id), "%s-%s", model, id);
+
+       name = get_property(rootdn, "ibm,partition-name", NULL);
+       if (name)
+               strncpy(partition_name, name, sizeof(partition_name));
+
+       num = (unsigned int *) get_property(rootdn, "ibm,partition-no", NULL);
+       if (num)
+               partition_number = *num;
+
+       return 0;
+}
+
+static int ibmvstgt_init(void)
+{
+       int err = -ENOMEM;
+
+       printk("IBM eServer i/pSeries Virtual SCSI Target Driver\n");
+
+       vtgtd = create_workqueue("ibmvtgtd");
+       if (!vtgtd)
+               return err;
+
+       err = get_system_info();
+       if (err)
+               goto destroy_wq;
+
+       err = vio_register_driver(&ibmvstgt_driver);
+       if (err)
+               goto destroy_wq;
+
+       return 0;
+
+destroy_wq:
+       destroy_workqueue(vtgtd);
+       return err;
+}
+
+static void ibmvstgt_exit(void)
+{
+       printk("Unregister IBM virtual SCSI driver\n");
+
+       destroy_workqueue(vtgtd);
+       vio_unregister_driver(&ibmvstgt_driver);
+}
+
+MODULE_DESCRIPTION("IBM Virtual SCSI Target");
+MODULE_AUTHOR("Santiago Leon");
+MODULE_LICENSE("GPL");
+
+module_init(ibmvstgt_init);
+module_exit(ibmvstgt_exit);