Implement PAPR virtual SCSI interface (ibmvscsi)

This patch implements the infrastructure and hypercalls necessary for
the PAPR specified Virtual SCSI interface.  This is the normal method
for providing (virtual) disks to PAPR partitions.

Signed-off-by: Ben Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David Gibson <dwg@au1.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>

5 files changed, 1455 insertions, 0 deletions

diff --git a/hw/ppc-viosrp.h b/hw/ppc-viosrp.h
new file mode 100644
index 0000000000..d8e365db1e
--- /dev/null
+++ b/hw/ppc-viosrp.h
@@ -0,0 +1,216 @@
+/*****************************************************************************/
+/* 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 PPC_VIOSRP_H
+#define PPC_VIOSRP_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;
+    uint8_t reserved[SRP_MAX_IU_LEN];
+};
+
+enum viosrp_crq_formats {
+    VIOSRP_SRP_FORMAT = 0x01,
+    VIOSRP_MAD_FORMAT = 0x02,
+    VIOSRP_OS400_FORMAT = 0x03,
+    VIOSRP_AIX_FORMAT = 0x04,
+    VIOSRP_LINUX_FORMAT = 0x06,
+    VIOSRP_INLINE_FORMAT = 0x07
+};
+
+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 {
+    uint8_t valid;        /* used by RPA */
+    uint8_t format;        /* SCSI vs out-of-band */
+    uint8_t reserved;
+    uint8_t status;        /* non-scsi failure? (e.g. DMA failure) */
+    uint16_t timeout;        /* in seconds */
+    uint16_t IU_length;        /* in bytes */
+    uint64_t 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_HOST_CONFIG_TYPE = 0x04,
+    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 {
+    uint32_t type;
+    uint16_t status;
+    uint16_t length;
+    uint64_t 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;
+    uint64_t buffer;
+    uint32_t port;
+};
+
+struct viosrp_error_log {
+    struct mad_common common;
+    uint64_t buffer;
+};
+
+struct viosrp_adapter_info {
+    struct mad_common common;
+    uint64_t buffer;
+};
+
+struct viosrp_host_config {
+    struct mad_common common;
+    uint64_t buffer;
+};
+
+struct viosrp_fast_fail {
+    struct mad_common common;
+};
+
+struct viosrp_capabilities {
+    struct mad_common common;
+    uint64_t buffer;
+};
+
+struct mad_capability_common {
+    uint32_t cap_type;
+    uint16_t length;
+    uint16_t server_support;
+};
+
+struct mad_reserve_cap {
+    struct mad_capability_common common;
+    uint32_t type;
+};
+
+struct mad_migration_cap {
+    struct mad_capability_common common;
+    uint32_t ecl;
+};
+
+struct capabilities {
+    uint32_t 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_host_config host_config;
+    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];
+    uint32_t partition_number;
+    uint32_t mad_version;
+    uint32_t os_type;
+    uint32_t port_max_txu[8];    /* per-port maximum transfer */
+};
+
+#endif
diff --git a/hw/spapr.c b/hw/spapr.c
index b432a9dcee..3bffaabe86 100644
--- a/hw/spapr.c
+++ b/hw/spapr.c
@@ -28,6 +28,7 @@
 #include "hw.h"
 #include "elf.h"
 #include "net.h"
+#include "blockdev.h"
 
 #include "hw/boards.h"
 #include "hw/ppc.h"
@@ -353,6 +354,12 @@ static void ppc_spapr_init(ram_addr_t ram_size,
         }
     }
 
+    for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
+        spapr_vscsi_create(spapr->vio_bus, 0x2000 + i,
+                           xics_find_qirq(spapr->icp, irq), irq);
+        irq++;
+    }
+
     if (kernel_filename) {
         uint64_t lowaddr = 0;
 
@@ -411,6 +418,7 @@ static QEMUMachine spapr_machine = {
     .max_cpus = MAX_CPUS,
     .no_vga = 1,
     .no_parallel = 1,
+    .use_scsi = 1,
 };
 
 static void spapr_machine_init(void)
diff --git a/hw/spapr_vio.h b/hw/spapr_vio.h
index ba16795c37..b7d0daab07 100644
--- a/hw/spapr_vio.h
+++ b/hw/spapr_vio.h
@@ -101,4 +101,7 @@ void spapr_vty_create(VIOsPAPRBus *bus,
 void spapr_vlan_create(VIOsPAPRBus *bus, uint32_t reg, NICInfo *nd,
                        qemu_irq qirq, uint32_t vio_irq_num);
 
+void spapr_vscsi_create(VIOsPAPRBus *bus, uint32_t reg,
+                        qemu_irq qirq, uint32_t vio_irq_num);
+
 #endif /* _HW_SPAPR_VIO_H */
diff --git a/hw/spapr_vscsi.c b/hw/spapr_vscsi.c
new file mode 100644
index 0000000000..e142dae624
--- /dev/null
+++ b/hw/spapr_vscsi.c
@@ -0,0 +1,988 @@
+/*
+ * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
+ *
+ * PAPR Virtual SCSI, aka ibmvscsi
+ *
+ * Copyright (c) 2010,2011 Benjamin Herrenschmidt, IBM Corporation.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ * TODO:
+ *
+ *  - Cleanups :-)
+ *  - Sort out better how to assign devices to VSCSI instances
+ *  - Fix residual counts
+ *  - Add indirect descriptors support
+ *  - Maybe do autosense (PAPR seems to mandate it, linux doesn't care)
+ */
+#include "hw.h"
+#include "scsi.h"
+#include "scsi-defs.h"
+#include "net.h" /* Remove that when we can */
+#include "srp.h"
+#include "hw/qdev.h"
+#include "hw/spapr.h"
+#include "hw/spapr_vio.h"
+#include "hw/ppc-viosrp.h"
+
+#include <libfdt.h>
+
+/*#define DEBUG_VSCSI*/
+
+#ifdef DEBUG_VSCSI
+#define dprintf(fmt, ...) \
+    do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
+#else
+#define dprintf(fmt, ...) \
+    do { } while (0)
+#endif
+
+/*
+ * Virtual SCSI device
+ */
+
+/* Random numbers */
+#define VSCSI_MAX_SECTORS       4096
+#define VSCSI_REQ_LIMIT         24
+
+#define SCSI_SENSE_BUF_SIZE     96
+#define SRP_RSP_SENSE_DATA_LEN  18
+
+typedef union vscsi_crq {
+    struct viosrp_crq s;
+    uint8_t raw[16];
+} vscsi_crq;
+
+typedef struct vscsi_req {
+    vscsi_crq               crq;
+    union viosrp_iu         iu;
+
+    /* SCSI request tracking */
+    SCSIDevice              *sdev;
+    uint32_t                qtag; /* qemu tag != srp tag */
+    int                     lun;
+    int                     active;
+    long                    data_len;
+    int                     writing;
+    int                     sensing;
+    int                     senselen;
+    uint8_t                 sense[SCSI_SENSE_BUF_SIZE];
+
+    /* RDMA related bits */
+    uint8_t                 dma_fmt;
+    struct srp_direct_buf   ext_desc;
+    struct srp_direct_buf   *cur_desc;
+    struct srp_indirect_buf *ind_desc;
+    int                     local_desc;
+    int                     total_desc;
+} vscsi_req;
+
+
+typedef struct {
+    VIOsPAPRDevice vdev;
+    SCSIBus bus;
+    vscsi_req reqs[VSCSI_REQ_LIMIT];
+} VSCSIState;
+
+/* XXX Debug only */
+static VSCSIState *dbg_vscsi_state;
+
+
+static struct vscsi_req *vscsi_get_req(VSCSIState *s)
+{
+    vscsi_req *req;
+    int i;
+
+    for (i = 0; i < VSCSI_REQ_LIMIT; i++) {
+        req = &s->reqs[i];
+        if (!req->active) {
+            memset(req, 0, sizeof(*req));
+            req->qtag = i;
+            req->active = 1;
+            return req;
+        }
+    }
+    return NULL;
+}
+
+static void vscsi_put_req(VSCSIState *s, vscsi_req *req)
+{
+    req->active = 0;
+}
+
+static vscsi_req *vscsi_find_req(VSCSIState *s, uint32_t tag)
+{
+    if (tag >= VSCSI_REQ_LIMIT || !s->reqs[tag].active) {
+        return NULL;
+    }
+    return &s->reqs[tag];
+}
+
+static void vscsi_decode_id_lun(uint64_t srp_lun, int *id, int *lun)
+{
+    /* XXX Figure that one out properly ! This is crackpot */
+    *id = (srp_lun >> 56) & 0x7f;
+    *lun = (srp_lun >> 48) & 0xff;
+}
+
+static int vscsi_send_iu(VSCSIState *s, vscsi_req *req,
+                         uint64_t length, uint8_t format)
+{
+    long rc, rc1;
+
+    /* First copy the SRP */
+    rc = spapr_tce_dma_write(&s->vdev, req->crq.s.IU_data_ptr,
+                             &req->iu, length);
+    if (rc) {
+        fprintf(stderr, "vscsi_send_iu: DMA write failure !\n");
+    }
+
+    req->crq.s.valid = 0x80;
+    req->crq.s.format = format;
+    req->crq.s.reserved = 0x00;
+    req->crq.s.timeout = cpu_to_be16(0x0000);
+    req->crq.s.IU_length = cpu_to_be16(length);
+    req->crq.s.IU_data_ptr = req->iu.srp.rsp.tag; /* right byte order */
+
+    if (rc == 0) {
+        req->crq.s.status = 0x99; /* Just needs to be non-zero */
+    } else {
+        req->crq.s.status = 0x00;
+    }
+
+    rc1 = spapr_vio_send_crq(&s->vdev, req->crq.raw);
+    if (rc1) {
+        fprintf(stderr, "vscsi_send_iu: Error sending response\n");
+        return rc1;
+    }
+
+    return rc;
+}
+
+static void vscsi_makeup_sense(VSCSIState *s, vscsi_req *req,
+                               uint8_t key, uint8_t asc, uint8_t ascq)
+{
+    req->senselen = SRP_RSP_SENSE_DATA_LEN;
+
+    /* Valid bit and 'current errors' */
+    req->sense[0] = (0x1 << 7 | 0x70);
+    /* Sense key */
+    req->sense[2] = key;
+    /* Additional sense length */
+    req->sense[7] = 0xa; /* 10 bytes */
+    /* Additional sense code */
+    req->sense[12] = asc;
+    req->sense[13] = ascq;
+}
+
+static int vscsi_send_rsp(VSCSIState *s, vscsi_req *req,
+                          uint8_t status, int32_t res_in, int32_t res_out)
+{
+    union viosrp_iu *iu = &req->iu;
+    uint64_t tag = iu->srp.rsp.tag;
+    int total_len = sizeof(iu->srp.rsp);
+
+    dprintf("VSCSI: Sending resp status: 0x%x, "
+            "res_in: %d, res_out: %d\n", status, res_in, res_out);
+
+    memset(iu, 0, sizeof(struct srp_rsp));
+    iu->srp.rsp.opcode = SRP_RSP;
+    iu->srp.rsp.req_lim_delta = cpu_to_be32(1);
+    iu->srp.rsp.tag = tag;
+
+    /* Handle residuals */
+    if (res_in < 0) {
+        iu->srp.rsp.flags |= SRP_RSP_FLAG_DIUNDER;
+        res_in = -res_in;
+    } else if (res_in) {
+        iu->srp.rsp.flags |= SRP_RSP_FLAG_DIOVER;
+    }
+    if (res_out < 0) {
+        iu->srp.rsp.flags |= SRP_RSP_FLAG_DOUNDER;
+        res_out = -res_out;
+    } else if (res_out) {
+        iu->srp.rsp.flags |= SRP_RSP_FLAG_DOOVER;
+    }
+    iu->srp.rsp.data_in_res_cnt = cpu_to_be32(res_in);
+    iu->srp.rsp.data_out_res_cnt = cpu_to_be32(res_out);
+
+    /* We don't do response data */
+    /* iu->srp.rsp.flags &= ~SRP_RSP_FLAG_RSPVALID; */
+    iu->srp.rsp.resp_data_len = cpu_to_be32(0);
+
+    /* Handle success vs. failure */
+    iu->srp.rsp.status = status;
+    if (status) {
+        iu->srp.rsp.sol_not = (iu->srp.cmd.sol_not & 0x04) >> 2;
+        if (req->senselen) {
+            req->iu.srp.rsp.flags |= SRP_RSP_FLAG_SNSVALID;
+            req->iu.srp.rsp.sense_data_len = cpu_to_be32(req->senselen);
+            memcpy(req->iu.srp.rsp.data, req->sense, req->senselen);
+            total_len += req->senselen;
+        }
+    } else {
+        iu->srp.rsp.sol_not = (iu->srp.cmd.sol_not & 0x02) >> 1;
+    }
+
+    vscsi_send_iu(s, req, total_len, VIOSRP_SRP_FORMAT);
+    return 0;
+}
+
+static inline void vscsi_swap_desc(struct srp_direct_buf *desc)
+{
+    desc->va = be64_to_cpu(desc->va);
+    desc->len = be32_to_cpu(desc->len);
+}
+
+static int vscsi_srp_direct_data(VSCSIState *s, vscsi_req *req,
+                                 uint8_t *buf, uint32_t len)
+{
+    struct srp_direct_buf *md = req->cur_desc;
+    uint32_t llen;
+    int rc;
+
+    dprintf("VSCSI: direct segment 0x%x bytes, va=0x%llx desc len=0x%x\n",
+            len, (unsigned long long)md->va, md->len);
+
+    llen = MIN(len, md->len);
+    if (llen) {
+        if (req->writing) { /* writing = to device = reading from memory */
+            rc = spapr_tce_dma_read(&s->vdev, md->va, buf, llen);
+        } else {
+            rc = spapr_tce_dma_write(&s->vdev, md->va, buf, llen);
+        }
+    }
+    md->len -= llen;
+    md->va += llen;
+
+    if (rc) {
+        return -1;
+    }
+    return llen;
+}
+
+static int vscsi_srp_indirect_data(VSCSIState *s, vscsi_req *req,
+                                   uint8_t *buf, uint32_t len)
+{
+    struct srp_direct_buf *td = &req->ind_desc->table_desc;
+    struct srp_direct_buf *md = req->cur_desc;
+    int rc = 0;
+    uint32_t llen, total = 0;
+
+    dprintf("VSCSI: indirect segment 0x%x bytes, td va=0x%llx len=0x%x\n",
+            len, (unsigned long long)td->va, td->len);
+
+    /* While we have data ... */
+    while (len) {
+        /* If we have a descriptor but it's empty, go fetch a new one */
+        if (md && md->len == 0) {
+            /* More local available, use one */
+            if (req->local_desc) {
+                md = ++req->cur_desc;
+                --req->local_desc;
+                --req->total_desc;
+                td->va += sizeof(struct srp_direct_buf);
+            } else {
+                md = req->cur_desc = NULL;
+            }
+        }
+        /* No descriptor at hand, fetch one */
+        if (!md) {
+            if (!req->total_desc) {
+                dprintf("VSCSI:   Out of descriptors !\n");
+                break;
+            }
+            md = req->cur_desc = &req->ext_desc;
+            dprintf("VSCSI:   Reading desc from 0x%llx\n",
+                    (unsigned long long)td->va);
+            rc = spapr_tce_dma_read(&s->vdev, td->va, md,
+                                    sizeof(struct srp_direct_buf));
+            if (rc) {
+                dprintf("VSCSI: tce_dma_read -> %d reading ext_desc\n", rc);
+                break;
+            }
+            vscsi_swap_desc(md);
+            td->va += sizeof(struct srp_direct_buf);
+            --req->total_desc;
+        }
+        dprintf("VSCSI:   [desc va=0x%llx,len=0x%x] remaining=0x%x\n",
+                (unsigned long long)md->va, md->len, len);
+
+        /* Perform transfer */
+        llen = MIN(len, md->len);
+        if (req->writing) { /* writing = to device = reading from memory */
+            rc = spapr_tce_dma_read(&s->vdev, md->va, buf, llen);
+        } else {
+            rc = spapr_tce_dma_write(&s->vdev, md->va, buf, llen);
+        }
+        if (rc) {
+            dprintf("VSCSI: tce_dma_r/w(%d) -> %d\n", req->writing, rc);
+            break;
+        }
+        dprintf("VSCSI:     data: %02x %02x %02x %02x...\n",
+                buf[0], buf[1], buf[2], buf[3]);
+
+        len -= llen;
+        buf += llen;
+        total += llen;
+        md->va += llen;
+        md->len -= llen;
+    }
+    return rc ? -1 : total;
+}
+
+static int vscsi_srp_transfer_data(VSCSIState *s, vscsi_req *req,
+                                   int writing, uint8_t *buf, uint32_t len)
+{
+    int err = 0;
+
+    switch (req->dma_fmt) {
+    case SRP_NO_DATA_DESC:
+        dprintf("VSCSI: no data desc transfer, skipping 0x%x bytes\n", len);
+        break;
+    case SRP_DATA_DESC_DIRECT:
+        err = vscsi_srp_direct_data(s, req, buf, len);
+        break;
+    case SRP_DATA_DESC_INDIRECT:
+        err = vscsi_srp_indirect_data(s, req, buf, len);
+        break;
+    }
+    return err;
+}
+
+/* Bits from linux srp */
+static int data_out_desc_size(struct srp_cmd *cmd)
+{
+    int size = 0;
+    uint8_t 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:
+        break;
+    }
+    return size;
+}
+
+static int vscsi_preprocess_desc(vscsi_req *req)
+{
+    struct srp_cmd *cmd = &req->iu.srp.cmd;
+    int offset, i;
+
+    offset = cmd->add_cdb_len & ~3;
+
+    if (req->writing) {
+        req->dma_fmt = cmd->buf_fmt >> 4;
+    } else {
+        offset += data_out_desc_size(cmd);
+        req->dma_fmt = cmd->buf_fmt & ((1U << 4) - 1);
+    }
+
+    switch (req->dma_fmt) {
+    case SRP_NO_DATA_DESC:
+        break;
+    case SRP_DATA_DESC_DIRECT:
+        req->cur_desc = (struct srp_direct_buf *)(cmd->add_data + offset);
+        req->total_desc = req->local_desc = 1;
+        vscsi_swap_desc(req->cur_desc);
+        dprintf("VSCSI: using direct RDMA %s, 0x%x bytes MD: 0x%llx\n",
+                req->writing ? "write" : "read",
+                req->cur_desc->len, (unsigned long long)req->cur_desc->va);
+        break;
+    case SRP_DATA_DESC_INDIRECT:
+        req->ind_desc = (struct srp_indirect_buf *)(cmd->add_data + offset);
+        vscsi_swap_desc(&req->ind_desc->table_desc);
+        req->total_desc = req->ind_desc->table_desc.len /
+            sizeof(struct srp_direct_buf);
+        req->local_desc = req->writing ? cmd->data_out_desc_cnt :
+            cmd->data_in_desc_cnt;
+        for (i = 0; i < req->local_desc; i++) {
+            vscsi_swap_desc(&req->ind_desc->desc_list[i]);
+        }
+        req->cur_desc = req->local_desc ? &req->ind_desc->desc_list[0] : NULL;
+        dprintf("VSCSI: using indirect RDMA %s, 0x%x bytes %d descs "
+                "(%d local) VA: 0x%llx\n",
+                req->writing ? "read" : "write",
+                be32_to_cpu(req->ind_desc->len),
+                req->total_desc, req->local_desc,
+                (unsigned long long)req->ind_desc->table_desc.va);
+        break;
+    default:
+        fprintf(stderr,
+                "vscsi_preprocess_desc: Unknown format %x\n", req->dma_fmt);
+        return -1;
+    }
+
+    return 0;
+}
+
+static void vscsi_send_request_sense(VSCSIState *s, vscsi_req *req)
+{
+    SCSIDevice *sdev = req->sdev;
+    uint8_t *cdb = req->iu.srp.cmd.cdb;
+    int n;
+
+    cdb[0] = 3;
+    cdb[1] = 0;
+    cdb[2] = 0;
+    cdb[3] = 0;
+    cdb[4] = 96;
+    cdb[5] = 0;
+    req->sensing = 1;
+    n = sdev->info->send_command(sdev, req->qtag, cdb, req->lun);
+    dprintf("VSCSI: Queued request sense tag 0x%x\n", req->qtag);
+    if (n < 0) {
+        fprintf(stderr, "VSCSI: REQUEST_SENSE wants write data !?!?!?\n");
+        sdev->info->cancel_io(sdev, req->qtag);
+        vscsi_makeup_sense(s, req, HARDWARE_ERROR, 0, 0);
+        vscsi_send_rsp(s, req, CHECK_CONDITION, 0, 0);
+        vscsi_put_req(s, req);
+        return;
+    } else if (n == 0) {
+        return;
+    }
+    sdev->info->read_data(sdev, req->qtag);
+}
+
+/* Callback to indicate that the SCSI layer has completed a transfer.  */
+static void vscsi_command_complete(SCSIBus *bus, int reason, uint32_t tag,
+                                   uint32_t arg)
+{
+    VSCSIState *s = DO_UPCAST(VSCSIState, vdev.qdev, bus->qbus.parent);
+    vscsi_req *req = vscsi_find_req(s, tag);
+    SCSIDevice *sdev;
+    uint8_t *buf;
+    int32_t res_in = 0, res_out = 0;
+    int len, rc = 0;
+
+    dprintf("VSCSI: SCSI cmd complete, r=0x%x tag=0x%x arg=0x%x, req=%p\n",
+            reason, tag, arg, req);
+    if (req == NULL) {
+        fprintf(stderr, "VSCSI: Can't find request for tag 0x%x\n", tag);
+        return;
+    }
+    sdev = req->sdev;
+
+    if (req->sensing) {
+        if (reason == SCSI_REASON_DONE) {
+            dprintf("VSCSI: Sense done !\n");
+            vscsi_send_rsp(s, req, CHECK_CONDITION, 0, 0);
+            vscsi_put_req(s, req);
+        } else {
+            uint8_t *buf = sdev->info->get_buf(sdev, tag);
+
+            len = MIN(arg, SCSI_SENSE_BUF_SIZE);
+            dprintf("VSCSI: Sense data, %d bytes:\n", len);
+            dprintf("       %02x  %02x  %02x  %02x  %02x  %02x  %02x  %02x\n",
+                    buf[0], buf[1], buf[2], buf[3],
+                    buf[4], buf[5], buf[6], buf[7]);
+            dprintf("       %02x  %02x  %02x  %02x  %02x  %02x  %02x  %02x\n",
+                    buf[8], buf[9], buf[10], buf[11],
+                    buf[12], buf[13], buf[14], buf[15]);
+            memcpy(req->sense, buf, len);
+            req->senselen = len;
+            sdev->info->read_data(sdev, req->qtag);
+        }
+        return;
+    }
+
+    if (reason == SCSI_REASON_DONE) {
+        dprintf("VSCSI: Command complete err=%d\n", arg);
+        if (arg == 0) {
+            /* We handle overflows, not underflows for normal commands,
+             * but hopefully nobody cares
+             */
+            if (req->writing) {
+                res_out = req->data_len;
+            } else {
+                res_in = req->data_len;
+            }
+            vscsi_send_rsp(s, req, 0, res_in, res_out);
+        } else if (arg == CHECK_CONDITION) {
+            dprintf("VSCSI: Got CHECK_CONDITION, requesting sense...\n");
+            vscsi_send_request_sense(s, req);
+            return;
+        } else {
+            vscsi_send_rsp(s, req, arg, 0, 0);
+        }
+        vscsi_put_req(s, req);
+        return;
+    }
+
+    /* "arg" is how much we have read for reads and how much we want
+     * to write for writes (ie, how much is to be DMA'd)
+     */
+    if (arg) {
+        buf = sdev->info->get_buf(sdev, tag);
+        rc = vscsi_srp_transfer_data(s, req, req->writing, buf, arg);
+    }
+    if (rc < 0) {
+        fprintf(stderr, "VSCSI: RDMA error rc=%d!\n", rc);
+        sdev->info->cancel_io(sdev, req->qtag);
+        vscsi_makeup_sense(s, req, HARDWARE_ERROR, 0, 0);
+        vscsi_send_rsp(s, req, CHECK_CONDITION, 0, 0);
+        vscsi_put_req(s, req);
+        return;
+    }
+
+    /* Start next chunk */
+    req->data_len -= rc;
+    if (req->writing) {
+        sdev->info->write_data(sdev, req->qtag);
+    } else {
+        sdev->info->read_data(sdev, req->qtag);
+    }
+}
+
+static void vscsi_process_login(VSCSIState *s, vscsi_req *req)
+{
+    union viosrp_iu *iu = &req->iu;
+    struct srp_login_rsp *rsp = &iu->srp.login_rsp;
+    uint64_t tag = iu->srp.rsp.tag;
+
+    dprintf("VSCSI: Got login, sendin response !\n");
+
+    /* 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;
+    /* Don't advertise quite as many request as we support to
+     * keep room for management stuff etc...
+     */
+    rsp->req_lim_delta = cpu_to_be32(VSCSI_REQ_LIMIT-2);
+    rsp->tag = tag;
+    rsp->max_it_iu_len = cpu_to_be32(sizeof(union srp_iu));
+    rsp->max_ti_iu_len = cpu_to_be32(sizeof(union srp_iu));
+    /* direct and indirect */
+    rsp->buf_fmt = cpu_to_be16(SRP_BUF_FORMAT_DIRECT | SRP_BUF_FORMAT_INDIRECT);
+
+    vscsi_send_iu(s, req, sizeof(*rsp), VIOSRP_SRP_FORMAT);
+}
+
+static void vscsi_inquiry_no_target(VSCSIState *s, vscsi_req *req)
+{
+    uint8_t *cdb = req->iu.srp.cmd.cdb;
+    uint8_t resp_data[36];
+    int rc, len, alen;
+
+    /* We dont do EVPD. Also check that page_code is 0 */
+    if ((cdb[1] & 0x01) || (cdb[1] & 0x01) || cdb[2] != 0) {
+        /* Send INVALID FIELD IN CDB */
+        vscsi_makeup_sense(s, req, ILLEGAL_REQUEST, 0x24, 0);
+        vscsi_send_rsp(s, req, CHECK_CONDITION, 0, 0);
+        return;
+    }
+    alen = cdb[3];
+    alen = (alen << 8) | cdb[4];
+    len = MIN(alen, 36);
+
+    /* Fake up inquiry using PQ=3 */
+    memset(resp_data, 0, 36);
+    resp_data[0] = 0x7f;   /* Not capable of supporting a device here */
+    resp_data[2] = 0x06;   /* SPS-4 */
+    resp_data[3] = 0x02;   /* Resp data format */
+    resp_data[4] = 36 - 5; /* Additional length */
+    resp_data[7] = 0x10;   /* Sync transfers */
+    memcpy(&resp_data[16], "QEMU EMPTY      ", 16);
+    memcpy(&resp_data[8], "QEMU    ", 8);
+
+    req->writing = 0;
+    vscsi_preprocess_desc(req);
+    rc = vscsi_srp_transfer_data(s, req, 0, resp_data, len);
+    if (rc < 0) {
+        vscsi_makeup_sense(s, req, HARDWARE_ERROR, 0, 0);
+        vscsi_send_rsp(s, req, CHECK_CONDITION, 0, 0);
+    } else {
+        vscsi_send_rsp(s, req, 0, 36 - rc, 0);
+    }
+}
+
+static int vscsi_queue_cmd(VSCSIState *s, vscsi_req *req)
+{
+    union srp_iu *srp = &req->iu.srp;
+    SCSIDevice *sdev;
+    int n, id, lun;
+
+    vscsi_decode_id_lun(be64_to_cpu(srp->cmd.lun), &id, &lun);
+
+    /* Qemu vs. linux issue with LUNs to be sorted out ... */
+    sdev = (id < 8 && lun < 16) ? s->bus.devs[id] : NULL;
+    if (!sdev) {
+        dprintf("VSCSI: Command for id %d with no drive\n", id);
+        if (srp->cmd.cdb[0] == INQUIRY) {
+            vscsi_inquiry_no_target(s, req);
+        } else {
+            vscsi_makeup_sense(s, req, ILLEGAL_REQUEST, 0x24, 0x00);
+            vscsi_send_rsp(s, req, CHECK_CONDITION, 0, 0);
+        } return 1;
+    }
+
+    req->sdev = sdev;
+    req->lun = lun;
+    n = sdev->info->send_command(sdev, req->qtag, srp->cmd.cdb, lun);
+
+    dprintf("VSCSI: Queued command tag 0x%x CMD 0x%x ID %d LUN %d ret: %d\n",
+            req->qtag, srp->cmd.cdb[0], id, lun, n);
+
+    if (n) {
+        /* Transfer direction must be set before preprocessing the
+         * descriptors
+         */
+        req->writing = (n < 1);
+
+        /* Preprocess RDMA descriptors */
+        vscsi_preprocess_desc(req);
+    }
+
+    /* Get transfer direction and initiate transfer */
+    if (n > 0) {
+        req->data_len = n;
+        sdev->info->read_data(sdev, req->qtag);
+    } else if (n < 0) {
+        req->data_len = -n;
+        sdev->info->write_data(sdev, req->qtag);
+    }
+    /* Don't touch req here, it may have been recycled already */
+
+    return 0;
+}
+
+static int vscsi_process_tsk_mgmt(VSCSIState *s, vscsi_req *req)
+{
+    union viosrp_iu *iu = &req->iu;
+    int fn;
+
+    fprintf(stderr, "vscsi_process_tsk_mgmt %02x\n",
+            iu->srp.tsk_mgmt.tsk_mgmt_func);
+
+    switch (iu->srp.tsk_mgmt.tsk_mgmt_func) {
+#if 0 /* We really don't deal with these for now */
+    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;
+#endif
+    default:
+        fn = 0;
+    }
+    if (fn) {
+        /* XXX Send/Handle target task management */
+        ;
+    } else {
+        vscsi_makeup_sense(s, req, ILLEGAL_REQUEST, 0x20, 0);
+        vscsi_send_rsp(s, req, CHECK_CONDITION, 0, 0);
+    }
+    return !fn;
+}
+
+static int vscsi_handle_srp_req(VSCSIState *s, vscsi_req *req)
+{
+    union srp_iu *srp = &req->iu.srp;
+    int done = 1;
+    uint8_t opcode = srp->rsp.opcode;
+
+    switch (opcode) {
+    case SRP_LOGIN_REQ:
+        vscsi_process_login(s, req);
+        break;
+    case SRP_TSK_MGMT:
+        done = vscsi_process_tsk_mgmt(s, req);
+        break;
+    case SRP_CMD:
+        done = vscsi_queue_cmd(s, req);
+        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:
+        fprintf(stderr, "VSCSI: Unsupported opcode %02x\n", opcode);
+        break;
+    default:
+        fprintf(stderr, "VSCSI: Unknown type %02x\n", opcode);
+    }
+
+    return done;
+}
+
+static int vscsi_send_adapter_info(VSCSIState *s, vscsi_req *req)
+{
+    struct viosrp_adapter_info *sinfo;
+    struct mad_adapter_info_data info;
+    int rc;
+
+    sinfo = &req->iu.mad.adapter_info;
+
+#if 0 /* What for ? */
+    rc = spapr_tce_dma_read(&s->vdev, be64_to_cpu(sinfo->buffer),
+                            &info, be16_to_cpu(sinfo->common.length));
+    if (rc) {
+        fprintf(stderr, "vscsi_send_adapter_info: DMA read failure !\n");
+    }
+#endif
+    memset(&info, 0, sizeof(info));
+    strcpy(info.srp_version, SRP_VERSION);
+    strncpy(info.partition_name, "qemu", sizeof("qemu"));
+    info.partition_number = cpu_to_be32(0);
+    info.mad_version = cpu_to_be32(1);
+    info.os_type = cpu_to_be32(2);
+    info.port_max_txu[0] = cpu_to_be32(VSCSI_MAX_SECTORS << 9);
+
+    rc = spapr_tce_dma_write(&s->vdev, be64_to_cpu(sinfo->buffer),
+                             &info, be16_to_cpu(sinfo->common.length));
+    if (rc)  {
+        fprintf(stderr, "vscsi_send_adapter_info: DMA write failure !\n");
+    }
+
+    sinfo->common.status = rc ? cpu_to_be32(1) : 0;
+
+    return vscsi_send_iu(s, req, sizeof(*sinfo), VIOSRP_MAD_FORMAT);
+}
+
+static int vscsi_handle_mad_req(VSCSIState *s, vscsi_req *req)
+{
+    union mad_iu *mad = &req->iu.mad;
+
+    switch (be32_to_cpu(mad->empty_iu.common.type)) {
+    case VIOSRP_EMPTY_IU_TYPE:
+        fprintf(stderr, "Unsupported EMPTY MAD IU\n");
+        break;
+    case VIOSRP_ERROR_LOG_TYPE:
+        fprintf(stderr, "Unsupported ERROR LOG MAD IU\n");
+        mad->error_log.common.status = cpu_to_be16(1);
+        vscsi_send_iu(s, req, sizeof(mad->error_log), VIOSRP_MAD_FORMAT);
+        break;
+    case VIOSRP_ADAPTER_INFO_TYPE:
+        vscsi_send_adapter_info(s, req);
+        break;
+    case VIOSRP_HOST_CONFIG_TYPE:
+        mad->host_config.common.status = cpu_to_be16(1);
+        vscsi_send_iu(s, req, sizeof(mad->host_config), VIOSRP_MAD_FORMAT);
+        break;
+    default:
+        fprintf(stderr, "VSCSI: Unknown MAD type %02x\n",
+                be32_to_cpu(mad->empty_iu.common.type));
+    }
+
+    return 1;
+}
+
+static void vscsi_got_payload(VSCSIState *s, vscsi_crq *crq)
+{
+    vscsi_req *req;
+    int done;
+
+    req = vscsi_get_req(s);
+    if (req == NULL) {
+        fprintf(stderr, "VSCSI: Failed to get a request !\n");
+        return;
+    }
+
+    /* We only support a limited number of descriptors, we know
+     * the ibmvscsi driver uses up to 10 max, so it should fit
+     * in our 256 bytes IUs. If not we'll have to increase the size
+     * of the structure.
+     */
+    if (crq->s.IU_length > sizeof(union viosrp_iu)) {
+        fprintf(stderr, "VSCSI: SRP IU too long (%d bytes) !\n",
+                crq->s.IU_length);
+        return;
+    }
+
+    /* XXX Handle failure differently ? */
+    if (spapr_tce_dma_read(&s->vdev, crq->s.IU_data_ptr, &req->iu,
+                           crq->s.IU_length)) {
+        fprintf(stderr, "vscsi_got_payload: DMA read failure !\n");
+        qemu_free(req);
+    }
+    memcpy(&req->crq, crq, sizeof(vscsi_crq));
+
+    if (crq->s.format == VIOSRP_MAD_FORMAT) {
+        done = vscsi_handle_mad_req(s, req);
+    } else {
+        done = vscsi_handle_srp_req(s, req);
+    }
+
+    if (done) {
+        vscsi_put_req(s, req);
+    }
+}
+
+
+static int vscsi_do_crq(struct VIOsPAPRDevice *dev, uint8_t *crq_data)
+{
+    VSCSIState *s = DO_UPCAST(VSCSIState, vdev, dev);
+    vscsi_crq crq;
+
+    memcpy(crq.raw, crq_data, 16);
+    crq.s.timeout = be16_to_cpu(crq.s.timeout);
+    crq.s.IU_length = be16_to_cpu(crq.s.IU_length);
+    crq.s.IU_data_ptr = be64_to_cpu(crq.s.IU_data_ptr);
+
+    dprintf("VSCSI: do_crq %02x %02x ...\n", crq.raw[0], crq.raw[1]);
+
+    switch (crq.s.valid) {
+    case 0xc0: /* Init command/response */
+
+        /* Respond to initialization request */
+        if (crq.s.format == 0x01) {
+            memset(crq.raw, 0, 16);
+            crq.s.valid = 0xc0;
+            crq.s.format = 0x02;
+            spapr_vio_send_crq(dev, crq.raw);
+        }
+
+        /* Note that in hotplug cases, we might get a 0x02
+         * as a result of us emitting the init request
+         */
+
+        break;
+    case 0xff: /* Link event */
+
+        /* Not handled for now */
+
+        break;
+    case 0x80: /* Payloads */
+        switch (crq.s.format) {
+        case VIOSRP_SRP_FORMAT: /* AKA VSCSI request */
+        case VIOSRP_MAD_FORMAT: /* AKA VSCSI response */
+            vscsi_got_payload(s, &crq);
+            break;
+        case VIOSRP_OS400_FORMAT:
+        case VIOSRP_AIX_FORMAT:
+        case VIOSRP_LINUX_FORMAT:
+        case VIOSRP_INLINE_FORMAT:
+            fprintf(stderr, "vscsi_do_srq: Unsupported payload format %02x\n",
+                    crq.s.format);
+            break;
+        default:
+            fprintf(stderr, "vscsi_do_srq: Unknown payload format %02x\n",
+                    crq.s.format);
+        }
+        break;
+    default:
+        fprintf(stderr, "vscsi_do_crq: unknown CRQ %02x %02x ...\n",
+                crq.raw[0], crq.raw[1]);
+    };
+
+    return 0;
+}
+
+static int spapr_vscsi_init(VIOsPAPRDevice *dev)
+{
+    VSCSIState *s = DO_UPCAST(VSCSIState, vdev, dev);
+    int i;
+
+    dbg_vscsi_state = s;
+
+    /* Initialize qemu request tags */
+    memset(s->reqs, 0, sizeof(s->reqs));
+    for (i = 0; i < VSCSI_REQ_LIMIT; i++) {
+        s->reqs[i].qtag = i;
+    }
+
+    dev->crq.SendFunc = vscsi_do_crq;
+
+    scsi_bus_new(&s->bus, &dev->qdev, 1, VSCSI_REQ_LIMIT,
+                 vscsi_command_complete);
+    if (!dev->qdev.hotplugged) {
+        scsi_bus_legacy_handle_cmdline(&s->bus);
+    }
+
+    return 0;
+}
+
+void spapr_vscsi_create(VIOsPAPRBus *bus, uint32_t reg,
+                        qemu_irq qirq, uint32_t vio_irq_num)
+{
+    DeviceState *dev;
+    VIOsPAPRDevice *sdev;
+
+    dev = qdev_create(&bus->bus, "spapr-vscsi");
+    qdev_prop_set_uint32(dev, "reg", reg);
+
+    qdev_init_nofail(dev);
+
+    sdev = (VIOsPAPRDevice *)dev;
+    sdev->qirq = qirq;
+    sdev->vio_irq_num = vio_irq_num;
+}
+
+static int spapr_vscsi_devnode(VIOsPAPRDevice *dev, void *fdt, int node_off)
+{
+    int ret;
+
+    ret = fdt_setprop_cell(fdt, node_off, "#address-cells", 2);
+    if (ret < 0) {
+        return ret;
+    }
+
+    ret = fdt_setprop_cell(fdt, node_off, "#size-cells", 0);
+    if (ret < 0) {
+        return ret;
+    }
+
+    return 0;
+}
+
+static VIOsPAPRDeviceInfo spapr_vscsi = {
+    .init = spapr_vscsi_init,
+    .devnode = spapr_vscsi_devnode,
+    .dt_name = "v-scsi",
+    .dt_type = "vscsi",
+    .dt_compatible = "IBM,v-scsi",
+    .signal_mask = 0x00000001,
+    .qdev.name = "spapr-vscsi",
+    .qdev.size = sizeof(VSCSIState),
+    .qdev.props = (Property[]) {
+        DEFINE_PROP_UINT32("reg", VIOsPAPRDevice, reg, 0x2000),
+        DEFINE_PROP_UINT32("dma-window", VIOsPAPRDevice,
+                           rtce_window_size, 0x10000000),
+        DEFINE_PROP_END_OF_LIST(),
+    },
+};
+
+static void spapr_vscsi_register(void)
+{
+    spapr_vio_bus_register_withprop(&spapr_vscsi);
+}
+device_init(spapr_vscsi_register);
diff --git a/hw/srp.h b/hw/srp.h
new file mode 100644
index 0000000000..afcd135c8f
--- /dev/null
+++ b/hw/srp.h
@@ -0,0 +1,240 @@
+/*
+ * Copyright (c) 2005 Cisco Systems.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef SCSI_SRP_H
+#define SCSI_SRP_H
+
+/*
+ * Structures and constants for the SCSI RDMA Protocol (SRP) as
+ * defined by the INCITS T10 committee.  This file was written using
+ * draft Revision 16a of the SRP standard.
+ */
+
+enum {
+
+    SRP_LOGIN_REQ = 0x00,
+    SRP_TSK_MGMT  = 0x01,
+    SRP_CMD       = 0x02,
+    SRP_I_LOGOUT  = 0x03,
+    SRP_LOGIN_RSP = 0xc0,
+    SRP_RSP       = 0xc1,
+    SRP_LOGIN_REJ = 0xc2,
+    SRP_T_LOGOUT  = 0x80,
+    SRP_CRED_REQ  = 0x81,
+    SRP_AER_REQ   = 0x82,
+    SRP_CRED_RSP  = 0x41,
+    SRP_AER_RSP   = 0x42
+};
+
+enum {
+    SRP_BUF_FORMAT_DIRECT   = 1 << 1,
+    SRP_BUF_FORMAT_INDIRECT = 1 << 2
+};
+
+enum {
+    SRP_NO_DATA_DESC       = 0,
+    SRP_DATA_DESC_DIRECT   = 1,
+    SRP_DATA_DESC_INDIRECT = 2
+};
+
+enum {
+    SRP_TSK_ABORT_TASK     = 0x01,
+    SRP_TSK_ABORT_TASK_SET = 0x02,
+    SRP_TSK_CLEAR_TASK_SET = 0x04,
+    SRP_TSK_LUN_RESET      = 0x08,
+    SRP_TSK_CLEAR_ACA      = 0x40
+};
+
+enum srp_login_rej_reason {
+    SRP_LOGIN_REJ_UNABLE_ESTABLISH_CHANNEL   = 0x00010000,
+    SRP_LOGIN_REJ_INSUFFICIENT_RESOURCES     = 0x00010001,
+    SRP_LOGIN_REJ_REQ_IT_IU_LENGTH_TOO_LARGE = 0x00010002,
+    SRP_LOGIN_REJ_UNABLE_ASSOCIATE_CHANNEL   = 0x00010003,
+    SRP_LOGIN_REJ_UNSUPPORTED_DESCRIPTOR_FMT = 0x00010004,
+    SRP_LOGIN_REJ_MULTI_CHANNEL_UNSUPPORTED  = 0x00010005,
+    SRP_LOGIN_REJ_CHANNEL_LIMIT_REACHED      = 0x00010006
+};
+
+enum {
+    SRP_REV10_IB_IO_CLASS  = 0xff00,
+    SRP_REV16A_IB_IO_CLASS = 0x0100
+};
+
+struct srp_direct_buf {
+    uint64_t    va;
+    uint32_t    key;
+    uint32_t    len;
+};
+
+/*
+ * We need the packed attribute because the SRP spec puts the list of
+ * descriptors at an offset of 20, which is not aligned to the size of
+ * struct srp_direct_buf.  The whole structure must be packed to avoid
+ * having the 20-byte structure padded to 24 bytes on 64-bit architectures.
+ */
+struct srp_indirect_buf {
+    struct srp_direct_buf    table_desc;
+    uint32_t                 len;
+    struct srp_direct_buf    desc_list[0];
+} __attribute__((packed));
+
+enum {
+    SRP_MULTICHAN_SINGLE = 0,
+    SRP_MULTICHAN_MULTI  = 1
+};
+
+struct srp_login_req {
+    uint8_t    opcode;
+    uint8_t    reserved1[7];
+    uint64_t   tag;
+    uint32_t   req_it_iu_len;
+    uint8_t    reserved2[4];
+    uint16_t   req_buf_fmt;
+    uint8_t    req_flags;
+    uint8_t    reserved3[5];
+    uint8_t    initiator_port_id[16];
+    uint8_t    target_port_id[16];
+};
+
+/*
+ * The SRP spec defines the size of the LOGIN_RSP structure to be 52
+ * bytes, so it needs to be packed to avoid having it padded to 56
+ * bytes on 64-bit architectures.
+ */
+struct srp_login_rsp {
+    uint8_t    opcode;
+    uint8_t    reserved1[3];
+    uint32_t   req_lim_delta;
+    uint64_t   tag;
+    uint32_t   max_it_iu_len;
+    uint32_t   max_ti_iu_len;
+    uint16_t   buf_fmt;
+    uint8_t    rsp_flags;
+    uint8_t    reserved2[25];
+} __attribute__((packed));
+
+struct srp_login_rej {
+    uint8_t    opcode;
+    uint8_t    reserved1[3];
+    uint32_t   reason;
+    uint64_t   tag;
+    uint8_t    reserved2[8];
+    uint16_t   buf_fmt;
+    uint8_t    reserved3[6];
+};
+
+struct srp_i_logout {
+    uint8_t    opcode;
+    uint8_t    reserved[7];
+    uint64_t   tag;
+};
+
+struct srp_t_logout {
+    uint8_t    opcode;
+    uint8_t    sol_not;
+    uint8_t    reserved[2];
+    uint32_t   reason;
+    uint64_t   tag;
+};
+
+/*
+ * We need the packed attribute because the SRP spec only aligns the
+ * 8-byte LUN field to 4 bytes.
+ */
+struct srp_tsk_mgmt {
+    uint8_t    opcode;
+    uint8_t    sol_not;
+    uint8_t    reserved1[6];
+    uint64_t   tag;
+    uint8_t    reserved2[4];
+    uint64_t   lun __attribute__((packed));
+    uint8_t    reserved3[2];
+    uint8_t    tsk_mgmt_func;
+    uint8_t    reserved4;
+    uint64_t   task_tag;
+    uint8_t    reserved5[8];
+};
+
+/*
+ * We need the packed attribute because the SRP spec only aligns the
+ * 8-byte LUN field to 4 bytes.
+ */
+struct srp_cmd {
+    uint8_t    opcode;
+    uint8_t    sol_not;
+    uint8_t    reserved1[3];
+    uint8_t    buf_fmt;
+    uint8_t    data_out_desc_cnt;
+    uint8_t    data_in_desc_cnt;
+    uint64_t   tag;
+    uint8_t    reserved2[4];
+    uint64_t   lun __attribute__((packed));
+    uint8_t    reserved3;
+    uint8_t    task_attr;
+    uint8_t    reserved4;
+    uint8_t    add_cdb_len;
+    uint8_t    cdb[16];
+    uint8_t    add_data[0];
+};
+
+enum {
+    SRP_RSP_FLAG_RSPVALID = 1 << 0,
+    SRP_RSP_FLAG_SNSVALID = 1 << 1,
+    SRP_RSP_FLAG_DOOVER   = 1 << 2,
+    SRP_RSP_FLAG_DOUNDER  = 1 << 3,
+    SRP_RSP_FLAG_DIOVER   = 1 << 4,
+    SRP_RSP_FLAG_DIUNDER  = 1 << 5
+};
+
+/*
+ * The SRP spec defines the size of the RSP structure to be 36 bytes,
+ * so it needs to be packed to avoid having it padded to 40 bytes on
+ * 64-bit architectures.
+ */
+struct srp_rsp {
+    uint8_t    opcode;
+    uint8_t    sol_not;
+    uint8_t    reserved1[2];
+    uint32_t   req_lim_delta;
+    uint64_t   tag;
+    uint8_t    reserved2[2];
+    uint8_t    flags;
+    uint8_t    status;
+    uint32_t   data_out_res_cnt;
+    uint32_t   data_in_res_cnt;
+    uint32_t   sense_data_len;
+    uint32_t   resp_data_len;
+    uint8_t    data[0];
+} __attribute__((packed));
+
+#endif /* SCSI_SRP_H */