Linux Virtualization - Apr 2019 - [RFC 3/3] RDMA/virtio-rdma: VirtIO rdma driver

Data center backends use more and more RDMA or RoCE devices and more and
more software runs in virtualized environment.
There is a need for a standard to enable RDMA/RoCE on Virtual Machines.

Virtio is the optimal solution since is the de-facto para-virtualizaton
technology and also because the Virtio specification
allows Hardware Vendors to support Virtio protocol natively in order to
achieve bare metal performance.

This RFC is an effort to addresses challenges in defining the RDMA/RoCE
Virtio Specification and a look forward on possible implementation
techniques.

Open issues/Todo list:
List is huge, this is only start point of the project.
Anyway, here is one example of item in the list:
- Multi VirtQ: Every QP has two rings and every CQ has one. This means that
  in order to support for example 32K QPs we will need 64K VirtQ. Not sure
  that this is reasonable so one option is to have one for all and
  multiplex the traffic on it. This is not good approach as by design it
  introducing an optional starvation. Another approach would be multi
  queues and round-robin (for example) between them.

Expectations from this posting:
In general, any comment is welcome, starting from hey, drop this as it is a
very bad idea, to yeah, go ahead, we really want it.
Idea here is that since it is not a minor effort i first want to know if
there is some sort interest in the community for such device.

The scope of the implementation is limited to probing the device and doing
some basic ibverbs commands. Data-path is not yet implemented. So with this
one can expect only that driver is (partialy) loaded and basic queries and
resource allocation is done.

One note regarding the patchset.
I know it is not standard to collaps patches from several repos as i did
here (qemu and linux) but decided to do it anyway so the whole picture can
be seen.

patch 1: virtio-net: Move some virtio-net-pci decl to include/hw/virtio
	This is a prelimenary patch just as a hack so i will not need to
	impelement new netdev
patch 2: hw/virtio-rdma: VirtIO rdma device
	The implementation of the device
patch 3: RDMA/virtio-rdma: VirtIO rdma driver
	The device driver

-- 
2.20.1

Signed-off-by: Yuval Shaia <yuval.shaia at oracle.com>
---
 hw/virtio/virtio-net-pci.c         | 18 ++-------------
 include/hw/virtio/virtio-net-pci.h | 35 ++++++++++++++++++++++++++++++
 2 files changed, 37 insertions(+), 16 deletions(-)
 create mode 100644 include/hw/virtio/virtio-net-pci.h

diff --git a/hw/virtio/virtio-net-pci.c b/hw/virtio/virtio-net-pci.c
index db07ab9e21..63617d5550 100644
--- a/hw/virtio/virtio-net-pci.c
+++ b/hw/virtio/virtio-net-pci.c
@@ -17,24 +17,10 @@
 
 #include "qemu/osdep.h"
 
-#include "hw/virtio/virtio-net.h"
+#include "hw/virtio/virtio-net-pci.h"
 #include "virtio-pci.h"
 #include "qapi/error.h"
 
-typedef struct VirtIONetPCI VirtIONetPCI;
-
-/*
- * virtio-net-pci: This extends VirtioPCIProxy.
- */
-#define TYPE_VIRTIO_NET_PCI "virtio-net-pci-base"
-#define VIRTIO_NET_PCI(obj) \
-        OBJECT_CHECK(VirtIONetPCI, (obj), TYPE_VIRTIO_NET_PCI)
-
-struct VirtIONetPCI {
-    VirtIOPCIProxy parent_obj;
-    VirtIONet vdev;
-};
-
 static Property virtio_net_properties[] = {
     DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
                     VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, true),
@@ -82,7 +68,7 @@ static void virtio_net_pci_instance_init(Object *obj)
 
 static const VirtioPCIDeviceTypeInfo virtio_net_pci_info = {
     .base_name             = TYPE_VIRTIO_NET_PCI,
-    .generic_name          = "virtio-net-pci",
+    .generic_name          = TYPE_VIRTIO_NET_PCI_GENERIC,
     .transitional_name     = "virtio-net-pci-transitional",
     .non_transitional_name = "virtio-net-pci-non-transitional",
     .instance_size = sizeof(VirtIONetPCI),
diff --git a/include/hw/virtio/virtio-net-pci.h
b/include/hw/virtio/virtio-net-pci.h
new file mode 100644
index 0000000000..f14e6ed992
--- /dev/null
+++ b/include/hw/virtio/virtio-net-pci.h
@@ -0,0 +1,35 @@
+/*
+ * PCI Virtio Network Device
+ *
+ * Copyright IBM, Corp. 2007
+ *
+ * Authors:
+ *  Anthony Liguori   <aliguori at us.ibm.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2.  See
+ * the COPYING file in the top-level directory.
+ *
+ */
+
+#ifndef QEMU_VIRTIO_NET_PCI_H
+#define QEMU_VIRTIO_NET_PCI_H
+
+#include "hw/virtio/virtio-net.h"
+#include "virtio-pci.h"
+
+typedef struct VirtIONetPCI VirtIONetPCI;
+
+/*
+ * virtio-net-pci: This extends VirtioPCIProxy.
+ */
+#define TYPE_VIRTIO_NET_PCI_GENERIC "virtio-net-pci"
+#define TYPE_VIRTIO_NET_PCI "virtio-net-pci-base"
+#define VIRTIO_NET_PCI(obj) \
+        OBJECT_CHECK(VirtIONetPCI, (obj), TYPE_VIRTIO_NET_PCI)
+
+struct VirtIONetPCI {
+    VirtIOPCIProxy parent_obj;
+    VirtIONet vdev;
+};
+
+#endif
-- 
2.20.1

Signed-off-by: Yuval Shaia <yuval.shaia at oracle.com>
---
 hw/Kconfig                                  |   1 +
 hw/rdma/Kconfig                             |   4 +
 hw/rdma/Makefile.objs                       |   2 +
 hw/rdma/virtio/virtio-rdma-ib.c             | 287 ++++++++++++++++++++
 hw/rdma/virtio/virtio-rdma-ib.h             |  93 +++++++
 hw/rdma/virtio/virtio-rdma-main.c           | 185 +++++++++++++
 hw/virtio/Makefile.objs                     |   1 +
 hw/virtio/virtio-rdma-pci.c                 | 108 ++++++++
 include/hw/pci/pci.h                        |   1 +
 include/hw/virtio/virtio-rdma.h             |  44 +++
 include/standard-headers/linux/virtio_ids.h |   1 +
 11 files changed, 727 insertions(+)
 create mode 100644 hw/rdma/Kconfig
 create mode 100644 hw/rdma/virtio/virtio-rdma-ib.c
 create mode 100644 hw/rdma/virtio/virtio-rdma-ib.h
 create mode 100644 hw/rdma/virtio/virtio-rdma-main.c
 create mode 100644 hw/virtio/virtio-rdma-pci.c
 create mode 100644 include/hw/virtio/virtio-rdma.h

diff --git a/hw/Kconfig b/hw/Kconfig
index d5ecd02070..88b9f15007 100644
--- a/hw/Kconfig
+++ b/hw/Kconfig
@@ -26,6 +26,7 @@ source pci-bridge/Kconfig
 source pci-host/Kconfig
 source pcmcia/Kconfig
 source pci/Kconfig
+source rdma/Kconfig
 source scsi/Kconfig
 source sd/Kconfig
 source smbios/Kconfig
diff --git a/hw/rdma/Kconfig b/hw/rdma/Kconfig
new file mode 100644
index 0000000000..b10bd7182b
--- /dev/null
+++ b/hw/rdma/Kconfig
@@ -0,0 +1,4 @@
+config VIRTIO_RDMA
+    bool
+    default y
+    depends on VIRTIO
diff --git a/hw/rdma/Makefile.objs b/hw/rdma/Makefile.objs
index c354e60e5b..ed640882be 100644
--- a/hw/rdma/Makefile.objs
+++ b/hw/rdma/Makefile.objs
@@ -3,3 +3,5 @@ obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o
rdma.o
 obj-$(CONFIG_PCI) += vmw/pvrdma_dev_ring.o vmw/pvrdma_cmd.o \
                      vmw/pvrdma_qp_ops.o vmw/pvrdma_main.o
 endif
+obj-$(CONFIG_VIRTIO_RDMA) += virtio/virtio-rdma-main.o \
+                             virtio/virtio-rdma-ib.o
diff --git a/hw/rdma/virtio/virtio-rdma-ib.c b/hw/rdma/virtio/virtio-rdma-ib.c
new file mode 100644
index 0000000000..2590a831a2
--- /dev/null
+++ b/hw/rdma/virtio/virtio-rdma-ib.c
@@ -0,0 +1,287 @@
+/*
+ * Virtio RDMA Device - IB verbs
+ *
+ * Copyright (C) 2019 Oracle
+ *
+ * Authors:
+ *  Yuval Shaia <yuval.shaia at oracle.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2.  See
+ * the COPYING file in the top-level directory.
+ *
+ */
+
+#include <infiniband/verbs.h>
+
+#include "qemu/osdep.h"
+
+#include "virtio-rdma-ib.h"
+#include "../rdma_utils.h"
+#include "../rdma_rm.h"
+#include "../rdma_backend.h"
+
+int virtio_rdma_query_device(VirtIORdma *rdev, struct iovec *in,
+                             struct iovec *out)
+{
+    struct ibv_device_attr attr = {};
+    int offs;
+    size_t s;
+
+    addrconf_addr_eui48((unsigned char *)&attr.sys_image_guid,
+                        (const char *)&rdev->netdev->mac);
+
+    attr.max_mr_size = 4096;
+    attr.page_size_cap = 4096;
+    attr.vendor_id = 1;
+    attr.vendor_part_id = 1;
+    attr.hw_ver = VIRTIO_RDMA_HW_VER;
+    attr.max_qp = 1024;
+    attr.max_qp_wr = 1024;
+    attr.device_cap_flags = 0;
+    attr.max_sge = 64;
+    attr.max_sge_rd = 64;
+    attr.max_cq = 1024;
+    attr.max_cqe = 64;
+    attr.max_mr = 1024;
+    attr.max_pd = 1024;
+    attr.max_qp_rd_atom = 0;
+    attr.max_ee_rd_atom = 0;
+    attr.max_res_rd_atom = 0;
+    attr.max_qp_init_rd_atom = 0;
+    attr.max_ee_init_rd_atom = 0;
+    attr.atomic_cap = IBV_ATOMIC_NONE;
+    attr.max_ee = 0;
+    attr.max_rdd = 0;
+    attr.max_mw = 0;
+    attr.max_raw_ipv6_qp = 0;
+    attr.max_raw_ethy_qp = 0;
+    attr.max_mcast_grp = 0;
+    attr.max_mcast_qp_attach = 0;
+    attr.max_total_mcast_qp_attach = 0;
+    attr.max_ah = 1024;
+    attr.max_fmr = 0;
+    attr.max_map_per_fmr = 0;
+    attr.max_srq = 0;
+    attr.max_srq_wr = 0;
+    attr.max_srq_sge = 0;
+    attr.max_pkeys = 1;
+    attr.local_ca_ack_delay = 0;
+    attr.phys_port_cnt = VIRTIO_RDMA_PORT_CNT;
+
+    offs = offsetof(struct ibv_device_attr, sys_image_guid);
+    s = iov_from_buf(out, 1, 0, (void *)&attr + offs, sizeof(attr) - offs);
+
+    return s == sizeof(attr) - offs ? VIRTIO_RDMA_CTRL_OK :
+                                      VIRTIO_RDMA_CTRL_ERR;
+}
+
+int virtio_rdma_query_port(VirtIORdma *rdev, struct iovec *in,
+                           struct iovec *out)
+{
+    struct ibv_port_attr attr = {};
+    struct cmd_query_port cmd = {};
+    int offs;
+    size_t s;
+
+    s = iov_to_buf(in, 1, 0, &cmd, sizeof(cmd));
+    if (s != sizeof(cmd)) {
+        return VIRTIO_RDMA_CTRL_ERR;
+    }
+
+    if (cmd.port != 1) {
+        return VIRTIO_RDMA_CTRL_ERR;
+    }
+
+    attr.state = IBV_PORT_ACTIVE;
+    attr.max_mtu = attr.active_mtu = IBV_MTU_1024;
+    attr.gid_tbl_len = 256;
+    attr.port_cap_flags = 0;
+    attr.max_msg_sz = 1024;
+    attr.bad_pkey_cntr = 0;
+    attr.qkey_viol_cntr = 0;
+    attr.pkey_tbl_len = 1;
+    attr.lid = 0;
+    attr.sm_lid = 0;
+    attr.lmc = 0;
+    attr.max_vl_num = 1;
+    attr.sm_sl = 0;
+    attr.subnet_timeout = 0;
+    attr.init_type_reply = 0;
+    attr.active_width = 0;
+    attr.active_speed = 0;
+    attr.phys_state = 0;
+
+    offs = offsetof(struct ibv_port_attr, state);
+    s = iov_from_buf(out, 1, 0, (void *)&attr + offs, sizeof(attr) - offs);
+
+    return s == sizeof(attr) - offs ? VIRTIO_RDMA_CTRL_OK :
+                                      VIRTIO_RDMA_CTRL_ERR;
+}
+
+int virtio_rdma_create_cq(VirtIORdma *rdev, struct iovec *in,
+                          struct iovec *out)
+{
+    struct cmd_create_cq cmd = {};
+    struct rsp_create_cq rsp = {};
+    size_t s;
+    int rc;
+
+    s = iov_to_buf(in, 1, 0, &cmd, sizeof(cmd));
+    if (s != sizeof(cmd)) {
+        return VIRTIO_RDMA_CTRL_ERR;
+    }
+
+    /* TODO: Define MAX_CQE */
+#define MAX_CQE 1024
+    /* TODO: Check MAX_CQ */
+    if (cmd.cqe > MAX_CQE) {
+        return VIRTIO_RDMA_CTRL_ERR;
+    }
+
+    printf("%s: %d\n", __func__, cmd.cqe);
+
+    /* TODO: Create VirtQ */
+
+    rc = rdma_rm_alloc_cq(rdev->rdma_dev_res, rdev->backend_dev, cmd.cqe,
+                          &rsp.cqn, NULL);
+    if (rc) {
+        /* TODO: Destroy VirtQ */
+        return VIRTIO_RDMA_CTRL_ERR;
+    }
+
+    printf("%s: %d\n", __func__, rsp.cqn);
+
+    s = iov_from_buf(out, 1, 0, &rsp, sizeof(rsp));
+
+    return s == sizeof(rsp) ? VIRTIO_RDMA_CTRL_OK :
+                              VIRTIO_RDMA_CTRL_ERR;
+}
+
+int virtio_rdma_destroy_cq(VirtIORdma *rdev, struct iovec *in,
+                          struct iovec *out)
+{
+    struct cmd_destroy_cq cmd = {};
+    size_t s;
+
+    s = iov_to_buf(in, 1, 0, &cmd, sizeof(cmd));
+    if (s != sizeof(cmd)) {
+        return VIRTIO_RDMA_CTRL_ERR;
+    }
+
+    printf("%s: %d\n", __func__, cmd.cqn);
+
+    /* TODO: Destroy VirtQ */
+
+    rdma_rm_dealloc_cq(rdev->rdma_dev_res, cmd.cqn);
+
+    return VIRTIO_RDMA_CTRL_OK;
+}
+
+int virtio_rdma_create_pd(VirtIORdma *rdev, struct iovec *in,
+                          struct iovec *out)
+{
+    struct rsp_create_pd rsp = {};
+    size_t s;
+    int rc;
+
+    /* TODO: Check MAX_PD */
+
+    /* TODO: ctx */
+    rc = rdma_rm_alloc_pd(rdev->rdma_dev_res, rdev->backend_dev,
&rsp.pdn,
+                          0);
+    if (rc)
+        return VIRTIO_RDMA_CTRL_ERR;
+
+    printf("%s: %d\n", __func__, rsp.pdn);
+
+    s = iov_from_buf(out, 1, 0, &rsp, sizeof(rsp));
+
+    return s == sizeof(rsp) ? VIRTIO_RDMA_CTRL_OK :
+                              VIRTIO_RDMA_CTRL_ERR;
+}
+
+int virtio_rdma_destroy_pd(VirtIORdma *rdev, struct iovec *in,
+                          struct iovec *out)
+{
+    struct cmd_destroy_pd cmd = {};
+    size_t s;
+
+    s = iov_to_buf(in, 1, 0, &cmd, sizeof(cmd));
+    if (s != sizeof(cmd)) {
+        return VIRTIO_RDMA_CTRL_ERR;
+    }
+
+    printf("%s: %d\n", __func__, cmd.pdn);
+
+    rdma_rm_dealloc_cq(rdev->rdma_dev_res, cmd.pdn);
+
+    return VIRTIO_RDMA_CTRL_OK;
+}
+
+int virtio_rdma_get_dma_mr(VirtIORdma *rdev, struct iovec *in,
+                           struct iovec *out)
+{
+    struct cmd_get_dma_mr cmd = {};
+    struct rsp_get_dma_mr rsp = {};
+    size_t s;
+
+    s = iov_to_buf(in, 1, 0, &cmd, sizeof(cmd));
+    if (s != sizeof(cmd)) {
+        return VIRTIO_RDMA_CTRL_ERR;
+    }
+
+    /* TODO: Call rdma_rm_alloc_mr */
+    rsp.mrn = 0x10;
+    rsp.lkey = 0x11;
+    rsp.rkey = 0x12;
+    printf("%s: 0x%x\n", __func__, rsp.mrn);
+
+    s = iov_from_buf(out, 1, 0, &rsp, sizeof(rsp));
+
+    return s == sizeof(rsp) ? VIRTIO_RDMA_CTRL_OK :
+                              VIRTIO_RDMA_CTRL_ERR;
+}
+
+static void virtio_rdma_init_dev_caps(VirtIORdma *rdev)
+{
+    rdev->dev_attr.max_qp_wr = 1024;
+}
+
+int virtio_rdma_init_ib(VirtIORdma *rdev)
+{
+    int rc;
+
+    virtio_rdma_init_dev_caps(rdev);
+
+    rdev->rdma_dev_res = g_malloc0(sizeof(RdmaDeviceResources));
+    rdev->backend_dev = g_malloc0(sizeof(RdmaBackendDev));
+
+    rc = rdma_backend_init(rdev->backend_dev, NULL, rdev->rdma_dev_res,
+                           rdev->backend_device_name,
+                           rdev->backend_port_num, &rdev->dev_attr,
+                           &rdev->mad_chr);
+    if (rc) {
+        rdma_error_report("Fail to initialize backend device");
+        return rc;
+    }
+
+    rc = rdma_rm_init(rdev->rdma_dev_res, &rdev->dev_attr);
+    if (rc) {
+        rdma_error_report("Fail to initialize resource manager");
+        return rc;
+    }
+
+    /* rdma_backend_start(rdev->backend_dev); */
+
+    return 0;
+}
+
+void virtio_rdma_fini_ib(VirtIORdma *rdev)
+{
+    /* rdma_backend_stop(rdev->backend_dev); */
+    rdma_rm_fini(rdev->rdma_dev_res, rdev->backend_dev,
+                 rdev->backend_eth_device_name);
+    rdma_backend_fini(rdev->backend_dev);
+    g_free(rdev->rdma_dev_res);
+    g_free(rdev->backend_dev);
+}
diff --git a/hw/rdma/virtio/virtio-rdma-ib.h b/hw/rdma/virtio/virtio-rdma-ib.h
new file mode 100644
index 0000000000..c4bdc063ac
--- /dev/null
+++ b/hw/rdma/virtio/virtio-rdma-ib.h
@@ -0,0 +1,93 @@
+/*
+ * Virtio RDMA Device - IB verbs
+ *
+ * Copyright (C) 2019 Oracle
+ *
+ * Authors:
+ *  Yuval Shaia <yuval.shaia at oracle.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2.  See
+ * the COPYING file in the top-level directory.
+ *
+ */
+
+#include "qemu/osdep.h"
+#include "qemu/iov.h"
+#include "hw/virtio/virtio-rdma.h"
+
+/* TODO: Move to uapi header file */
+#define VIRTIO_RDMA_CTRL_OK    0
+#define VIRTIO_RDMA_CTRL_ERR   1
+
+enum {
+    VIRTIO_CMD_QUERY_DEVICE = 10,
+    VIRTIO_CMD_QUERY_PORT,
+    VIRTIO_CMD_CREATE_CQ,
+    VIRTIO_CMD_DESTROY_CQ,
+    VIRTIO_CMD_CREATE_PD,
+    VIRTIO_CMD_DESTROY_PD,
+    VIRTIO_CMD_GET_DMA_MR,
+};
+
+struct control_buf {
+    uint8_t cmd;
+    uint8_t status;
+};
+
+struct cmd_query_port {
+    uint8_t port;
+};
+
+struct cmd_create_cq {
+    uint32_t cqe;
+};
+
+struct rsp_create_cq {
+    uint32_t cqn;
+};
+
+struct cmd_destroy_cq {
+    uint32_t cqn;
+};
+
+struct rsp_create_pd {
+    uint32_t pdn;
+};
+
+struct cmd_destroy_pd {
+    uint32_t pdn;
+};
+
+struct cmd_get_dma_mr {
+    uint32_t pdn;
+    uint32_t access_flags;
+};
+
+struct rsp_get_dma_mr {
+    uint32_t mrn;
+    uint32_t lkey;
+    uint32_t rkey;
+};
+
+/* TODO: Move to uapi header file */
+
+#define VIRTIO_RDMA_PORT_CNT    1
+#define VIRTIO_RDMA_HW_VER      1
+
+int virtio_rdma_init_ib(VirtIORdma *rdev);
+void virtio_rdma_fini_ib(VirtIORdma *rdev);
+
+int virtio_rdma_query_device(VirtIORdma *rdev, struct iovec *in,
+                             struct iovec *out);
+int virtio_rdma_query_port(VirtIORdma *rdev, struct iovec *in,
+                           struct iovec *out);
+int virtio_rdma_create_cq(VirtIORdma *rdev, struct iovec *in,
+                          struct iovec *out);
+int virtio_rdma_destroy_cq(VirtIORdma *rdev, struct iovec *in,
+                          struct iovec *out);
+int virtio_rdma_create_pd(VirtIORdma *rdev, struct iovec *in,
+                          struct iovec *out);
+int virtio_rdma_destroy_pd(VirtIORdma *rdev, struct iovec *in,
+                          struct iovec *out);
+int virtio_rdma_get_dma_mr(VirtIORdma *rdev, struct iovec *in,
+                           struct iovec *out);
diff --git a/hw/rdma/virtio/virtio-rdma-main.c
b/hw/rdma/virtio/virtio-rdma-main.c
new file mode 100644
index 0000000000..54f75b14c0
--- /dev/null
+++ b/hw/rdma/virtio/virtio-rdma-main.c
@@ -0,0 +1,185 @@
+/*
+ * Virtio RDMA Device
+ *
+ * Copyright (C) 2019 Oracle
+ *
+ * Authors:
+ *  Yuval Shaia <yuval.shaia at oracle.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2.  See
+ * the COPYING file in the top-level directory.
+ *
+ */
+
+#include <infiniband/verbs.h>
+
+#include "qemu/osdep.h"
+#include "hw/virtio/virtio.h"
+#include "qemu/error-report.h"
+#include "hw/virtio/virtio-bus.h"
+#include "hw/virtio/virtio-rdma.h"
+#include "include/standard-headers/linux/virtio_ids.h"
+
+#include "virtio-rdma-ib.h"
+#include "../rdma_rm_defs.h"
+#include "../rdma_utils.h"
+
+static void virtio_rdma_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
+{
+    VirtIORdma *r = VIRTIO_RDMA(vdev);
+    struct control_buf cb;
+    VirtQueueElement *e;
+    size_t s;
+
+    virtio_queue_set_notification(vq, 0);
+
+    for (;;) {
+        e = virtqueue_pop(vq, sizeof(VirtQueueElement));
+        if (!e) {
+            break;
+        }
+
+        if (iov_size(e->in_sg, e->in_num) < sizeof(cb.status) ||
+            iov_size(e->out_sg, e->out_num) < sizeof(cb.cmd)) {
+            virtio_error(vdev, "Got invalid message size");
+            virtqueue_detach_element(vq, e, 0);
+            g_free(e);
+            break;
+        }
+
+        s = iov_to_buf(&e->out_sg[0], 1, 0, &cb.cmd,
sizeof(cb.cmd));
+        if (s != sizeof(cb.cmd)) {
+            cb.status = VIRTIO_RDMA_CTRL_ERR;
+        } else {
+            printf("cmd=%d\n", cb.cmd);
+            switch (cb.cmd) {
+            case VIRTIO_CMD_QUERY_DEVICE:
+                cb.status = virtio_rdma_query_device(r, &e->out_sg[1],
+                                                     &e->in_sg[0]);
+                break;
+            case VIRTIO_CMD_QUERY_PORT:
+                cb.status = virtio_rdma_query_port(r, &e->out_sg[1],
+                                                   &e->in_sg[0]);
+                break;
+            case VIRTIO_CMD_CREATE_CQ:
+                cb.status = virtio_rdma_create_cq(r, &e->out_sg[1],
+                                                  &e->in_sg[0]);
+                break;
+            case VIRTIO_CMD_DESTROY_CQ:
+                cb.status = virtio_rdma_destroy_cq(r, &e->out_sg[1],
+                                                  &e->in_sg[0]);
+                break;
+            case VIRTIO_CMD_CREATE_PD:
+                cb.status = virtio_rdma_create_pd(r, &e->out_sg[1],
+                                                  &e->in_sg[0]);
+                break;
+            case VIRTIO_CMD_DESTROY_PD:
+                cb.status = virtio_rdma_destroy_pd(r, &e->out_sg[1],
+                                                  &e->in_sg[0]);
+                break;
+            case VIRTIO_CMD_GET_DMA_MR:
+                cb.status = virtio_rdma_get_dma_mr(r, &e->out_sg[1],
+                                                  &e->in_sg[0]);
+                break;
+            default:
+                cb.status = VIRTIO_RDMA_CTRL_ERR;
+            }
+        }
+        printf("status=%d\n", cb.status);
+        s = iov_from_buf(&e->in_sg[1], 1, 0, &cb.status,
sizeof(cb.status));
+        assert(s == sizeof(cb.status));
+
+        virtqueue_push(vq, e, sizeof(cb.status));
+        virtio_notify(vdev, vq);
+    }
+
+    virtio_queue_set_notification(vq, 1);
+}
+
+static void virtio_rdma_device_realize(DeviceState *dev, Error **errp)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(dev);
+    VirtIORdma *r = VIRTIO_RDMA(dev);
+    int rc;
+
+    rc = virtio_rdma_init_ib(r);
+    if (rc) {
+        rdma_error_report("Fail to initialize IB layer");
+        return;
+    }
+
+    virtio_init(vdev, "virtio-rdma", VIRTIO_ID_RDMA, 1024);
+
+    r->ctrl_vq = virtio_add_queue(vdev, 64, virtio_rdma_handle_ctrl);
+}
+
+static void virtio_rdma_device_unrealize(DeviceState *dev, Error **errp)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(dev);
+    VirtIORdma *r = VIRTIO_RDMA(dev);
+
+    virtio_del_queue(vdev, 0);
+
+    virtio_cleanup(vdev);
+
+    virtio_rdma_fini_ib(r);
+}
+
+static uint64_t virtio_rdma_get_features(VirtIODevice *vdev, uint64_t features,
+                                        Error **errp)
+{
+    /* virtio_add_feature(&features, VIRTIO_NET_F_MAC); */
+
+    vdev->backend_features = features;
+
+    return features;
+}
+
+
+static Property virtio_rdma_dev_properties[] = {
+    DEFINE_PROP_STRING("netdev", VirtIORdma,
backend_eth_device_name),
+    DEFINE_PROP_STRING("ibdev",VirtIORdma, backend_device_name),
+    DEFINE_PROP_UINT8("ibport", VirtIORdma, backend_port_num, 1),
+    DEFINE_PROP_UINT64("dev-caps-max-mr-size", VirtIORdma,
dev_attr.max_mr_size,
+                       MAX_MR_SIZE),
+    DEFINE_PROP_INT32("dev-caps-max-qp", VirtIORdma, dev_attr.max_qp,
MAX_QP),
+    DEFINE_PROP_INT32("dev-caps-max-cq", VirtIORdma, dev_attr.max_cq,
MAX_CQ),
+    DEFINE_PROP_INT32("dev-caps-max-mr", VirtIORdma, dev_attr.max_mr,
MAX_MR),
+    DEFINE_PROP_INT32("dev-caps-max-pd", VirtIORdma, dev_attr.max_pd,
MAX_PD),
+    DEFINE_PROP_INT32("dev-caps-qp-rd-atom", VirtIORdma,
+                       dev_attr.max_qp_rd_atom, MAX_QP_RD_ATOM),
+    DEFINE_PROP_INT32("dev-caps-max-qp-init-rd-atom", VirtIORdma,
+                      dev_attr.max_qp_init_rd_atom, MAX_QP_INIT_RD_ATOM),
+    DEFINE_PROP_INT32("dev-caps-max-ah", VirtIORdma, dev_attr.max_ah,
MAX_AH),
+    DEFINE_PROP_CHR("mad-chardev", VirtIORdma, mad_chr),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static void virtio_rdma_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
+
+    set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
+    vdc->realize = virtio_rdma_device_realize;
+    vdc->unrealize = virtio_rdma_device_unrealize;
+    vdc->get_features = virtio_rdma_get_features;
+
+    dc->desc = "Virtio RDMA Device";
+    dc->props = virtio_rdma_dev_properties;
+    set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
+}
+
+static const TypeInfo virtio_rdma_info = {
+    .name = TYPE_VIRTIO_RDMA,
+    .parent = TYPE_VIRTIO_DEVICE,
+    .instance_size = sizeof(VirtIORdma),
+    .class_init = virtio_rdma_class_init,
+};
+
+static void virtio_register_types(void)
+{
+    type_register_static(&virtio_rdma_info);
+}
+
+type_init(virtio_register_types)
diff --git a/hw/virtio/Makefile.objs b/hw/virtio/Makefile.objs
index f2ab667a21..fd701feb9f 100644
--- a/hw/virtio/Makefile.objs
+++ b/hw/virtio/Makefile.objs
@@ -27,6 +27,7 @@ obj-$(CONFIG_VIRTIO_9P) += virtio-9p-pci.o
 obj-$(CONFIG_VIRTIO_SCSI) += virtio-scsi-pci.o
 obj-$(CONFIG_VIRTIO_BLK) += virtio-blk-pci.o
 obj-$(CONFIG_VIRTIO_NET) += virtio-net-pci.o
+obj-$(CONFIG_VIRTIO_RDMA) += virtio-rdma-pci.o
 obj-$(CONFIG_VIRTIO_SERIAL) += virtio-serial-pci.o
 endif
 else
diff --git a/hw/virtio/virtio-rdma-pci.c b/hw/virtio/virtio-rdma-pci.c
new file mode 100644
index 0000000000..36efce285b
--- /dev/null
+++ b/hw/virtio/virtio-rdma-pci.c
@@ -0,0 +1,108 @@
+/*
+ * Virtio rdma PCI Bindings
+ *
+ * Copyright (C) 2019 Oracle
+ *
+ * Authors:
+ *  Yuval Shaia <yuval.shaia at oracle.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2.  See
+ * the COPYING file in the top-level directory.
+ *
+ */
+
+#include "qemu/osdep.h"
+
+#include "hw/virtio/virtio-net-pci.h"
+#include "hw/virtio/virtio-rdma.h"
+#include "virtio-pci.h"
+#include "qapi/error.h"
+
+typedef struct VirtIORdmaPCI VirtIORdmaPCI;
+
+/*
+ * virtio-rdma-pci: This extends VirtioPCIProxy.
+ */
+#define TYPE_VIRTIO_RDMA_PCI "virtio-rdma-pci-base"
+#define VIRTIO_RDMA_PCI(obj) \
+        OBJECT_CHECK(VirtIORdmaPCI, (obj), TYPE_VIRTIO_RDMA_PCI)
+
+struct VirtIORdmaPCI {
+    VirtIOPCIProxy parent_obj;
+    VirtIORdma vdev;
+};
+
+static Property virtio_rdma_properties[] = {
+    DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
+                    VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, true),
+    DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 3),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static void virtio_rdma_pci_realize(VirtIOPCIProxy *vpci_dev, Error **errp)
+{
+    VirtIORdmaPCI *dev = VIRTIO_RDMA_PCI(vpci_dev);
+    DeviceState *vdev = DEVICE(&dev->vdev);
+    VirtIONetPCI *vnet_pci;
+    PCIDevice *func0;
+
+    qdev_set_parent_bus(vdev, BUS(&vpci_dev->bus));
+    object_property_set_bool(OBJECT(vdev), true, "realized", errp);
+
+    func0 = pci_get_function_0(&vpci_dev->pci_dev);
+    /* Break if not virtio device in slot 0 */
+    if (strcmp(object_get_typename(OBJECT(func0)),
+               TYPE_VIRTIO_NET_PCI_GENERIC)) {
+        error_setg(errp, "Device on %x.0 is type %s but must be %s",
+                   PCI_SLOT(vpci_dev->pci_dev.devfn),
+                   object_get_typename(OBJECT(func0)),
+                   TYPE_VIRTIO_NET_PCI_GENERIC);
+        return;
+    }
+    vnet_pci = VIRTIO_NET_PCI(func0);
+    dev->vdev.netdev = &vnet_pci->vdev;
+}
+
+static void virtio_rdma_pci_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
+    VirtioPCIClass *vpciklass = VIRTIO_PCI_CLASS(klass);
+
+    k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
+    k->device_id = PCI_DEVICE_ID_VIRTIO_RDMA;
+    k->revision = VIRTIO_PCI_ABI_VERSION;
+    k->class_id = PCI_CLASS_NETWORK_OTHER;
+    set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
+    dc->props = virtio_rdma_properties;
+    vpciklass->realize = virtio_rdma_pci_realize;
+}
+
+static void virtio_rdma_pci_instance_init(Object *obj)
+{
+    VirtIORdmaPCI *dev = VIRTIO_RDMA_PCI(obj);
+
+    virtio_instance_init_common(obj, &dev->vdev, sizeof(dev->vdev),
+                                TYPE_VIRTIO_RDMA);
+    /*
+    object_property_add_alias(obj, "bootindex",
OBJECT(&dev->vdev),
+                              "bootindex", &error_abort);
+    */
+}
+
+static const VirtioPCIDeviceTypeInfo virtio_rdma_pci_info = {
+    .base_name             = TYPE_VIRTIO_RDMA_PCI,
+    .generic_name          = "virtio-rdma-pci",
+    .transitional_name     = "virtio-rdma-pci-transitional",
+    .non_transitional_name = "virtio-rdma-pci-non-transitional",
+    .instance_size = sizeof(VirtIORdmaPCI),
+    .instance_init = virtio_rdma_pci_instance_init,
+    .class_init    = virtio_rdma_pci_class_init,
+};
+
+static void virtio_rdma_pci_register(void)
+{
+    virtio_pci_types_register(&virtio_rdma_pci_info);
+}
+
+type_init(virtio_rdma_pci_register)
diff --git a/include/hw/pci/pci.h b/include/hw/pci/pci.h
index d87f5f93e9..c2d34c382f 100644
--- a/include/hw/pci/pci.h
+++ b/include/hw/pci/pci.h
@@ -85,6 +85,7 @@ extern bool pci_available;
 #define PCI_DEVICE_ID_VIRTIO_RNG         0x1005
 #define PCI_DEVICE_ID_VIRTIO_9P          0x1009
 #define PCI_DEVICE_ID_VIRTIO_VSOCK       0x1012
+#define PCI_DEVICE_ID_VIRTIO_RDMA        0x1013
 
 #define PCI_VENDOR_ID_REDHAT             0x1b36
 #define PCI_DEVICE_ID_REDHAT_BRIDGE      0x0001
diff --git a/include/hw/virtio/virtio-rdma.h b/include/hw/virtio/virtio-rdma.h
new file mode 100644
index 0000000000..3c7534cd8a
--- /dev/null
+++ b/include/hw/virtio/virtio-rdma.h
@@ -0,0 +1,44 @@
+/*
+ * Virtio RDMA Device
+ *
+ * Copyright (C) 2019 Oracle
+ *
+ * Authors:
+ *  Yuval Shaia <yuval.shaia at oracle.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2.  See
+ * the COPYING file in the top-level directory.
+ *
+ */
+
+#ifndef QEMU_VIRTIO_RDMA_H
+#define QEMU_VIRTIO_RDMA_H
+
+#include <infiniband/verbs.h>
+
+#include "chardev/char-fe.h"
+#include "hw/virtio/virtio.h"
+#include "hw/virtio/virtio-net.h"
+
+#define TYPE_VIRTIO_RDMA "virtio-rdma-device"
+#define VIRTIO_RDMA(obj) \
+        OBJECT_CHECK(VirtIORdma, (obj), TYPE_VIRTIO_RDMA)
+
+typedef struct RdmaBackendDev RdmaBackendDev;
+typedef struct RdmaDeviceResources RdmaDeviceResources;
+struct ibv_device_attr;
+
+typedef struct VirtIORdma {
+    VirtIODevice parent_obj;
+    VirtQueue *ctrl_vq;
+    VirtIONet *netdev;
+    RdmaBackendDev *backend_dev;
+    RdmaDeviceResources *rdma_dev_res;
+    CharBackend mad_chr;
+    char *backend_eth_device_name;
+    char *backend_device_name;
+    uint8_t backend_port_num;
+    struct ibv_device_attr dev_attr;
+} VirtIORdma;
+
+#endif
diff --git a/include/standard-headers/linux/virtio_ids.h
b/include/standard-headers/linux/virtio_ids.h
index 6d5c3b2d4f..bd2c699450 100644
--- a/include/standard-headers/linux/virtio_ids.h
+++ b/include/standard-headers/linux/virtio_ids.h
@@ -43,5 +43,6 @@
 #define VIRTIO_ID_INPUT        18 /* virtio input */
 #define VIRTIO_ID_VSOCK        19 /* virtio vsock transport */
 #define VIRTIO_ID_CRYPTO       20 /* virtio crypto */
+#define VIRTIO_ID_RDMA         26 /* virtio crypto */
 
 #endif /* _LINUX_VIRTIO_IDS_H */
-- 
2.20.1

Signed-off-by: Yuval Shaia <yuval.shaia at oracle.com>
---
 drivers/infiniband/Kconfig                    |   1 +
 drivers/infiniband/hw/Makefile                |   1 +
 drivers/infiniband/hw/virtio/Kconfig          |   6 +
 drivers/infiniband/hw/virtio/Makefile         |   4 +
 drivers/infiniband/hw/virtio/virtio_rdma.h    |  40 +
 .../infiniband/hw/virtio/virtio_rdma_device.c |  59 ++
 .../infiniband/hw/virtio/virtio_rdma_device.h |  32 +
 drivers/infiniband/hw/virtio/virtio_rdma_ib.c | 711 ++++++++++++++++++
 drivers/infiniband/hw/virtio/virtio_rdma_ib.h |  48 ++
 .../infiniband/hw/virtio/virtio_rdma_main.c   | 149 ++++
 .../infiniband/hw/virtio/virtio_rdma_netdev.c |  44 ++
 .../infiniband/hw/virtio/virtio_rdma_netdev.h |  33 +
 include/uapi/linux/virtio_ids.h               |   1 +
 13 files changed, 1129 insertions(+)
 create mode 100644 drivers/infiniband/hw/virtio/Kconfig
 create mode 100644 drivers/infiniband/hw/virtio/Makefile
 create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma.h
 create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_device.c
 create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_device.h
 create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_ib.c
 create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_ib.h
 create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_main.c
 create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_netdev.c
 create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_netdev.h

diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig
index a1fb840de45d..218a47d4cecf 100644
--- a/drivers/infiniband/Kconfig
+++ b/drivers/infiniband/Kconfig
@@ -107,6 +107,7 @@ source "drivers/infiniband/hw/hfi1/Kconfig"
 source "drivers/infiniband/hw/qedr/Kconfig"
 source "drivers/infiniband/sw/rdmavt/Kconfig"
 source "drivers/infiniband/sw/rxe/Kconfig"
+source "drivers/infiniband/hw/virtio/Kconfig"
 endif
 
 source "drivers/infiniband/ulp/ipoib/Kconfig"
diff --git a/drivers/infiniband/hw/Makefile b/drivers/infiniband/hw/Makefile
index e4f31c1be8f7..10ffb2c421e4 100644
--- a/drivers/infiniband/hw/Makefile
+++ b/drivers/infiniband/hw/Makefile
@@ -14,3 +14,4 @@ obj-$(CONFIG_INFINIBAND_HFI1)		+= hfi1/
 obj-$(CONFIG_INFINIBAND_HNS)		+= hns/
 obj-$(CONFIG_INFINIBAND_QEDR)		+= qedr/
 obj-$(CONFIG_INFINIBAND_BNXT_RE)	+= bnxt_re/
+obj-$(CONFIG_INFINIBAND_VIRTIO_RDMA)	+= virtio/
diff --git a/drivers/infiniband/hw/virtio/Kconfig
b/drivers/infiniband/hw/virtio/Kconfig
new file mode 100644
index 000000000000..92e41691cf5d
--- /dev/null
+++ b/drivers/infiniband/hw/virtio/Kconfig
@@ -0,0 +1,6 @@
+config INFINIBAND_VIRTIO_RDMA
+	tristate "VirtIO Paravirtualized RDMA Driver"
+	depends on NETDEVICES && ETHERNET && PCI && INET
+	---help---
+	  This driver provides low-level support for VirtIO Paravirtual
+	  RDMA adapter.
diff --git a/drivers/infiniband/hw/virtio/Makefile
b/drivers/infiniband/hw/virtio/Makefile
new file mode 100644
index 000000000000..fb637e467167
--- /dev/null
+++ b/drivers/infiniband/hw/virtio/Makefile
@@ -0,0 +1,4 @@
+obj-$(CONFIG_INFINIBAND_VIRTIO_RDMA) += virtio_rdma.o
+
+virtio_rdma-y := virtio_rdma_main.o virtio_rdma_device.o virtio_rdma_ib.o \
+		 virtio_rdma_netdev.o
diff --git a/drivers/infiniband/hw/virtio/virtio_rdma.h
b/drivers/infiniband/hw/virtio/virtio_rdma.h
new file mode 100644
index 000000000000..7896a2dfb812
--- /dev/null
+++ b/drivers/infiniband/hw/virtio/virtio_rdma.h
@@ -0,0 +1,40 @@
+/*
+ * Virtio RDMA device: Driver main data types
+ *
+ * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
+ */
+
+#ifndef __VIRTIO_RDMA__
+#define __VIRTIO_RDMA__
+
+#include <linux/virtio.h>
+#include <rdma/ib_verbs.h>
+
+struct virtio_rdma_info {
+	struct ib_device ib_dev;
+	struct virtio_device *vdev;
+	struct virtqueue *ctrl_vq;
+	wait_queue_head_t acked; /* arm on send to host, release on recv */
+	struct net_device *netdev;
+};
+
+static inline struct virtio_rdma_info *to_vdev(struct ib_device *ibdev)
+{
+	return container_of(ibdev, struct virtio_rdma_info, ib_dev);
+}
+
+#endif
diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_device.c
b/drivers/infiniband/hw/virtio/virtio_rdma_device.c
new file mode 100644
index 000000000000..ae41e530644f
--- /dev/null
+++ b/drivers/infiniband/hw/virtio/virtio_rdma_device.c
@@ -0,0 +1,59 @@
+/*
+ * Virtio RDMA device: Device related functions and data
+ *
+ * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
+ */
+
+#include <linux/virtio_config.h>
+
+#include "virtio_rdma.h"
+
+static void rdma_ctrl_ack(struct virtqueue *vq)
+{
+	struct virtio_rdma_info *dev = vq->vdev->priv;
+
+	wake_up(&dev->acked);
+
+	printk("%s\n", __func__);
+}
+
+int init_device(struct virtio_rdma_info *dev)
+{
+#define TMP_MAX_VQ 1
+	int rc;
+	struct virtqueue *vqs[TMP_MAX_VQ];
+	vq_callback_t *cbs[TMP_MAX_VQ];
+	const char *names[TMP_MAX_VQ];
+
+	names[0] = "ctrl";
+	cbs[0] = rdma_ctrl_ack;
+	cbs[0] = NULL;
+
+	rc = virtio_find_vqs(dev->vdev, TMP_MAX_VQ, vqs, cbs, names, NULL);
+	if (rc)
+		return rc;
+
+	dev->ctrl_vq = vqs[0];
+
+	return 0;
+}
+
+void fini_device(struct virtio_rdma_info *dev)
+{
+	dev->vdev->config->reset(dev->vdev);
+	dev->vdev->config->del_vqs(dev->vdev);
+}
diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_device.h
b/drivers/infiniband/hw/virtio/virtio_rdma_device.h
new file mode 100644
index 000000000000..d9b1240daf92
--- /dev/null
+++ b/drivers/infiniband/hw/virtio/virtio_rdma_device.h
@@ -0,0 +1,32 @@
+/*
+ * Virtio RDMA device: Device related functions and data
+ *
+ * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
+ */
+
+#ifndef __VIRTIO_RDMA_DEVICE__
+#define __VIRTIO_RDMA_DEVICE__
+
+#define VIRTIO_RDMA_BOARD_ID	1
+#define VIRTIO_RDMA_HW_NAME	"virtio-rdma"
+#define VIRTIO_RDMA_HW_REV	1
+#define VIRTIO_RDMA_DRIVER_VER	"1.0"
+
+int init_device(struct virtio_rdma_info *dev);
+void fini_device(struct virtio_rdma_info *dev);
+
+#endif
diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_ib.c
b/drivers/infiniband/hw/virtio/virtio_rdma_ib.c
new file mode 100644
index 000000000000..02bf4a332611
--- /dev/null
+++ b/drivers/infiniband/hw/virtio/virtio_rdma_ib.c
@@ -0,0 +1,711 @@
+/*
+ * Virtio RDMA device: IB related functions and data
+ *
+ * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
+ */
+
+#include <linux/scatterlist.h>
+#include <linux/virtio.h>
+#include <rdma/ib_mad.h>
+
+#include "virtio_rdma.h"
+#include "virtio_rdma_device.h"
+#include "virtio_rdma_ib.h"
+
+/* TODO: Move to uapi header file */
+
+/*
+ * Control virtqueue data structures
+ *
+ * The control virtqueue expects a header in the first sg entry
+ * and an ack/status response in the last entry.  Data for the
+ * command goes in between.
+ */
+
+#define VIRTIO_RDMA_CTRL_OK	0
+#define VIRTIO_RDMA_CTRL_ERR	1
+
+struct control_buf {
+	__u8 cmd;
+	__u8 status;
+};
+
+enum {
+	VIRTIO_CMD_QUERY_DEVICE = 10,
+	VIRTIO_CMD_QUERY_PORT,
+	VIRTIO_CMD_CREATE_CQ,
+	VIRTIO_CMD_DESTROY_CQ,
+	VIRTIO_CMD_CREATE_PD,
+	VIRTIO_CMD_DESTROY_PD,
+	VIRTIO_CMD_GET_DMA_MR,
+};
+
+struct cmd_query_port {
+	__u8 port;
+};
+
+struct cmd_create_cq {
+	__u32 cqe;
+};
+
+struct rsp_create_cq {
+	__u32 cqn;
+};
+
+struct cmd_destroy_cq {
+	__u32 cqn;
+};
+
+struct rsp_create_pd {
+	__u32 pdn;
+};
+
+struct cmd_destroy_pd {
+	__u32 pdn;
+};
+
+struct cmd_get_dma_mr {
+	__u32 pdn;
+	__u32 access_flags;
+};
+
+struct rsp_get_dma_mr {
+	__u32 mrn;
+	__u32 lkey;
+	__u32 rkey;
+};
+
+/* TODO: Move to uapi header file */
+
+struct virtio_rdma_ib_cq {
+	struct ib_cq ibcq;
+	u32 cq_handle;
+};
+
+/* TODO: For the scope fof the RFC i'm utilizing ib*_*_attr structures */
+
+static int virtio_rdma_exec_cmd(struct virtio_rdma_info *di, int cmd,
+				struct scatterlist *in, struct scatterlist *out)
+{
+	struct scatterlist *sgs[4], hdr, status;
+	struct control_buf *ctrl;
+	unsigned tmp;
+	int rc;
+
+	ctrl = kmalloc(sizeof(*ctrl), GFP_ATOMIC);
+	ctrl->cmd = cmd;
+	ctrl->status = ~0;
+
+	sg_init_one(&hdr, &ctrl->cmd, sizeof(ctrl->cmd));
+	sgs[0] = &hdr;
+	sgs[1] = in;
+	sgs[2] = out;
+	sg_init_one(&status, &ctrl->status, sizeof(ctrl->status));
+	sgs[3] = &status;
+
+	rc = virtqueue_add_sgs(di->ctrl_vq, sgs, 2, 2, di, GFP_ATOMIC);
+	if (rc)
+		goto out;
+
+	if (unlikely(!virtqueue_kick(di->ctrl_vq))) {
+		goto out_with_status;
+	}
+
+	/* Spin for a response, the kick causes an ioport write, trapping
+	 * into the hypervisor, so the request should be handled
+	 * immediately */
+	while (!virtqueue_get_buf(di->ctrl_vq, &tmp) &&
+	       !virtqueue_is_broken(di->ctrl_vq))
+		cpu_relax();
+
+out_with_status:
+	printk("%s: cmd %d, status %d\n", __func__, ctrl->cmd,
ctrl->status);
+	rc = ctrl->status == VIRTIO_RDMA_CTRL_OK ? 0 : 1;
+
+out:
+	kfree(ctrl);
+
+	return rc;
+}
+
+static int virtio_rdma_port_immutable(struct ib_device *ibdev, u8 port_num,
+				      struct ib_port_immutable *immutable)
+{
+	struct ib_port_attr attr;
+	int rc;
+
+	rc = ib_query_port(ibdev, port_num, &attr);
+	if (rc)
+		return rc;
+
+	immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
+	immutable->pkey_tbl_len = attr.pkey_tbl_len;
+	immutable->gid_tbl_len = attr.gid_tbl_len;
+	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
+
+	return 0;
+}
+
+static int virtio_rdma_query_device(struct ib_device *ibdev,
+				    struct ib_device_attr *props,
+				    struct ib_udata *uhw)
+{
+	struct scatterlist data;
+	int offs;
+	int rc;
+
+	if (uhw->inlen || uhw->outlen)
+		return -EINVAL;
+
+	/* We start with sys_image_guid because of inconsistency beween ib_
+	 * and ibv_ */
+	offs = offsetof(struct ib_device_attr, sys_image_guid);
+	sg_init_one(&data, (void *)props + offs, sizeof(*props) - offs);
+
+	rc = virtio_rdma_exec_cmd(to_vdev(ibdev), VIRTIO_CMD_QUERY_DEVICE, NULL,
+				  &data);
+
+	printk("%s: sys_image_guid 0x%llx\n", __func__,
+	       be64_to_cpu(props->sys_image_guid));
+
+	return rc;
+}
+
+static int virtio_rdma_query_port(struct ib_device *ibdev, u8 port,
+				  struct ib_port_attr *props)
+{
+	struct scatterlist in, out;
+	struct cmd_query_port *cmd;
+	int offs;
+	int rc;
+
+	cmd = kmalloc(sizeof(*cmd), GFP_ATOMIC);
+	if (!cmd)
+		return -ENOMEM;
+
+	/* We start with state because of inconsistency beween ib and ibv */
+	offs = offsetof(struct ib_port_attr, state);
+	sg_init_one(&out, (void *)props + offs, sizeof(*props) - offs);
+
+	cmd->port = port;
+	sg_init_one(&in, cmd, sizeof(*cmd));
+	printk("%s: port %d\n", __func__, cmd->port);
+
+	rc = virtio_rdma_exec_cmd(to_vdev(ibdev), VIRTIO_CMD_QUERY_PORT, &in,
+				  &out);
+
+	printk("%s: gid_tbl_len %d\n", __func__, props->gid_tbl_len);
+
+	kfree(cmd);
+
+	return rc;
+}
+
+static struct net_device *virtio_rdma_get_netdev(struct ib_device *ibdev,
+						 u8 port_num)
+{
+	struct virtio_rdma_info *ri = to_vdev(ibdev);
+
+	printk("%s:\n", __func__);
+
+	return ri->netdev;
+}
+
+struct ib_cq *virtio_rdma_create_cq(struct ib_device *ibdev,
+				    const struct ib_cq_init_attr *attr,
+				    struct ib_ucontext *context,
+				    struct ib_udata *udata)
+{
+	struct scatterlist in, out;
+	struct virtio_rdma_ib_cq *vcq;
+	struct cmd_create_cq *cmd;
+	struct rsp_create_cq *rsp;
+	struct ib_cq *cq = NULL;
+	int rc;
+
+	/* TODO: Check MAX_CQ */
+
+	cmd = kmalloc(sizeof(*cmd), GFP_ATOMIC);
+	if (!cmd)
+		return ERR_PTR(-ENOMEM);
+
+	rsp = kmalloc(sizeof(*rsp), GFP_ATOMIC);
+	if (!rsp) {
+		kfree(cmd);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	vcq = kzalloc(sizeof(*vcq), GFP_KERNEL);
+	if (!vcq)
+		goto out;
+
+	cmd->cqe = attr->cqe;
+	sg_init_one(&in, cmd, sizeof(*cmd));
+	printk("%s: cqe %d\n", __func__, cmd->cqe);
+
+	sg_init_one(&out, rsp, sizeof(*rsp));
+
+	rc = virtio_rdma_exec_cmd(to_vdev(ibdev), VIRTIO_CMD_CREATE_CQ, &in,
+				  &out);
+	if (rc)
+		goto out_err;
+
+	printk("%s: cqn 0x%x\n", __func__, rsp->cqn);
+	vcq->cq_handle = rsp->cqn;
+	vcq->ibcq.cqe = attr->cqe;
+	cq = &vcq->ibcq;
+
+	goto out;
+
+out_err:
+	kfree(vcq);
+	return ERR_PTR(rc);
+
+out:
+	kfree(rsp);
+	kfree(cmd);
+	return cq;
+}
+
+int virtio_rdma_destroy_cq(struct ib_cq *cq)
+{
+	struct virtio_rdma_ib_cq *vcq;
+	struct scatterlist in;
+	struct cmd_destroy_cq *cmd;
+	int rc;
+
+	cmd = kmalloc(sizeof(*cmd), GFP_ATOMIC);
+	if (!cmd)
+		return -ENOMEM;
+
+	vcq = container_of(cq, struct virtio_rdma_ib_cq, ibcq);
+
+	cmd->cqn = vcq->cq_handle;
+	sg_init_one(&in, cmd, sizeof(*cmd));
+
+	rc = virtio_rdma_exec_cmd(to_vdev(cq->device), VIRTIO_CMD_DESTROY_CQ,
+				  &in, NULL);
+
+	kfree(cmd);
+
+	kfree(vcq);
+
+	return rc;
+}
+
+int virtio_rdma_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context,
+			 struct ib_udata *udata)
+{
+	struct virtio_rdma_pd *pd = to_vpd(ibpd);
+	struct ib_device *ibdev = ibpd->device;
+	struct rsp_create_pd *rsp;
+	struct scatterlist out;
+	int rc;
+
+	/* TODO: Check MAX_PD */
+
+	rsp = kmalloc(sizeof(*rsp), GFP_ATOMIC);
+	if (!rsp)
+		return -ENOMEM;
+
+	sg_init_one(&out, rsp, sizeof(*rsp));
+
+	rc = virtio_rdma_exec_cmd(to_vdev(ibdev), VIRTIO_CMD_CREATE_PD, NULL,
+				  &out);
+	if (rc)
+		goto out;
+
+	pd->pd_handle = rsp->pdn;
+
+	printk("%s: pd_handle=%d\n", __func__, pd->pd_handle);
+
+out:
+	kfree(rsp);
+
+	printk("%s: rc=%d\n", __func__, rc);
+	return rc;
+}
+
+void virtio_rdma_dealloc_pd(struct ib_pd *pd)
+{
+	struct virtio_rdma_pd *vpd = to_vpd(pd);
+	struct ib_device *ibdev = pd->device;
+	struct cmd_destroy_pd *cmd;
+	struct scatterlist in;
+
+	printk("%s:\n", __func__);
+
+	cmd = kmalloc(sizeof(*cmd), GFP_ATOMIC);
+	if (!cmd)
+		return;
+
+	cmd->pdn = vpd->pd_handle;
+	sg_init_one(&in, cmd, sizeof(*cmd));
+
+	virtio_rdma_exec_cmd(to_vdev(ibdev), VIRTIO_CMD_DESTROY_PD, &in, NULL);
+
+	kfree(cmd);
+}
+
+struct ib_mr *virtio_rdma_get_dma_mr(struct ib_pd *pd, int acc)
+
+{
+	struct virtio_rdma_user_mr *mr;
+	struct scatterlist in, out;
+	struct cmd_get_dma_mr *cmd = NULL;
+	struct rsp_get_dma_mr *rsp = NULL;
+	int rc;
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	cmd = kmalloc(sizeof(*cmd), GFP_ATOMIC);
+	if (!cmd) {
+		kfree(mr);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	rsp = kmalloc(sizeof(*rsp), GFP_ATOMIC);
+	if (!cmd) {
+		kfree(mr);
+		kfree(cmd);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	cmd->pdn = to_vpd(pd)->pd_handle;
+	cmd->access_flags = acc;
+	sg_init_one(&in, cmd, sizeof(*cmd));
+
+	sg_init_one(&out, rsp, sizeof(*rsp));
+
+	rc = virtio_rdma_exec_cmd(to_vdev(pd->device), VIRTIO_CMD_GET_DMA_MR,
+				  &in, &out);
+	if (rc) {
+		kfree(mr);
+		kfree(cmd);
+		return ERR_PTR(rc);
+	}
+
+	mr->mr_handle = rsp->mrn;
+	mr->ibmr.lkey = rsp->lkey;
+	mr->ibmr.rkey = rsp->rkey;
+
+	printk("%s: mr_handle=0x%x\n", __func__, mr->mr_handle);
+
+	kfree(cmd);
+	kfree(rsp);
+
+	return &mr->ibmr;
+}
+
+struct ib_qp *virtio_rdma_create_qp(struct ib_pd *pd,
+				    struct ib_qp_init_attr *init_attr,
+				    struct ib_udata *udata)
+{
+	/* struct pvrdma_dev *dev = to_vdev(pd->device); */
+	struct virtio_rdma_qp *qp;
+
+	printk("%s:\n", __func__);
+
+	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+	if (!qp)
+		return ERR_PTR(-ENOMEM);
+
+	return &qp->ibqp;
+}
+
+int virtio_rdma_query_gid(struct ib_device *ibdev, u8 port, int index,
+			  union ib_gid *gid)
+{
+	memset(gid, 0, sizeof(union ib_gid));
+
+	printk("%s: port %d, index %d\n", __func__, port, index);
+
+	return 0;
+}
+
+static int virtio_rdma_add_gid(const struct ib_gid_attr *attr, void **context)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+struct ib_mr *virtio_rdma_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
+				   u32 max_num_sg)
+{
+	printk("%s: mr_type %d, max_num_sg %d\n", __func__, mr_type,
+	       max_num_sg);
+
+	return NULL;
+}
+
+int virtio_rdma_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata
*udata)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+struct ib_ah *virtio_rdma_create_ah(struct ib_pd *pd,
+				    struct rdma_ah_attr *ah_attr, u32 flags,
+				    struct ib_udata *udata)
+{
+	printk("%s:\n", __func__);
+
+	return NULL;
+}
+
+void virtio_rdma_dealloc_ucontext(struct ib_ucontext *ibcontext)
+
+{
+}
+
+static int virtio_rdma_del_gid(const struct ib_gid_attr *attr, void **context)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+int virtio_rdma_dereg_mr(struct ib_mr *ibmr)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+int virtio_rdma_destroy_ah(struct ib_ah *ah, u32 flags)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+struct virtio_rdma_cq {
+	struct ib_cq ibcq;
+};
+
+int virtio_rdma_destroy_qp(struct ib_qp *qp)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+static void virtio_rdma_get_fw_ver_str(struct ib_device *device, char *str)
+{
+	printk("%s:\n", __func__);
+}
+
+enum rdma_link_layer virtio_rdma_port_link_layer(struct ib_device *ibdev,
+						 u8 port)
+{
+	return IB_LINK_LAYER_ETHERNET;
+}
+
+int virtio_rdma_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
+			  int sg_nents, unsigned int *sg_offset)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+int virtio_rdma_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vma)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+int virtio_rdma_modify_port(struct ib_device *ibdev, u8 port, int mask,
+			    struct ib_port_modify *props)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+int virtio_rdma_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+			  int attr_mask, struct ib_udata *udata)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+int virtio_rdma_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+int virtio_rdma_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
+			  const struct ib_recv_wr **bad_wr)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+int virtio_rdma_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
+			  const struct ib_send_wr **bad_wr)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+int virtio_rdma_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+			   u16 *pkey)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+int virtio_rdma_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+			 int attr_mask, struct ib_qp_init_attr *init_attr)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+struct ib_mr *virtio_rdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+				      u64 virt_addr, int access_flags,
+				      struct ib_udata *udata)
+{
+	printk("%s:\n", __func__);
+
+	return NULL;
+}
+
+int virtio_rdma_req_notify_cq(struct ib_cq *ibcq,
+			      enum ib_cq_notify_flags notify_flags)
+{
+	printk("%s:\n", __func__);
+
+	return 0;
+}
+
+static const struct ib_device_ops virtio_rdma_dev_ops = {
+	.get_port_immutable = virtio_rdma_port_immutable,
+	.query_device = virtio_rdma_query_device,
+	.query_port = virtio_rdma_query_port,
+	.get_netdev = virtio_rdma_get_netdev,
+	.create_cq = virtio_rdma_create_cq,
+	.destroy_cq = virtio_rdma_destroy_cq,
+	.alloc_pd = virtio_rdma_alloc_pd,
+	.dealloc_pd = virtio_rdma_dealloc_pd,
+	.get_dma_mr = virtio_rdma_get_dma_mr,
+	.create_qp = virtio_rdma_create_qp,
+	.query_gid = virtio_rdma_query_gid,
+	.add_gid = virtio_rdma_add_gid,
+	.alloc_mr = virtio_rdma_alloc_mr,
+	.alloc_ucontext = virtio_rdma_alloc_ucontext,
+	.create_ah = virtio_rdma_create_ah,
+	.dealloc_ucontext = virtio_rdma_dealloc_ucontext,
+	.del_gid = virtio_rdma_del_gid,
+	.dereg_mr = virtio_rdma_dereg_mr,
+	.destroy_ah = virtio_rdma_destroy_ah,
+	.destroy_qp = virtio_rdma_destroy_qp,
+	.get_dev_fw_str = virtio_rdma_get_fw_ver_str,
+	.get_link_layer = virtio_rdma_port_link_layer,
+	.get_port_immutable = virtio_rdma_port_immutable,
+	.map_mr_sg = virtio_rdma_map_mr_sg,
+	.mmap = virtio_rdma_mmap,
+	.modify_port = virtio_rdma_modify_port,
+	.modify_qp = virtio_rdma_modify_qp,
+	.poll_cq = virtio_rdma_poll_cq,
+	.post_recv = virtio_rdma_post_recv,
+	.post_send = virtio_rdma_post_send,
+	.query_device = virtio_rdma_query_device,
+	.query_pkey = virtio_rdma_query_pkey,
+	.query_port = virtio_rdma_query_port,
+	.query_qp = virtio_rdma_query_qp,
+	.reg_user_mr = virtio_rdma_reg_user_mr,
+	.req_notify_cq = virtio_rdma_req_notify_cq,
+	INIT_RDMA_OBJ_SIZE(ib_pd, virtio_rdma_pd, ibpd),
+};
+
+static ssize_t hca_type_show(struct device *device,
+			     struct device_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%s-%s\n", VIRTIO_RDMA_HW_NAME,
+		       VIRTIO_RDMA_DRIVER_VER);
+}
+static DEVICE_ATTR_RO(hca_type);
+
+static ssize_t hw_rev_show(struct device *device,
+			   struct device_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%d\n", VIRTIO_RDMA_HW_REV);
+}
+static DEVICE_ATTR_RO(hw_rev);
+
+static ssize_t board_id_show(struct device *device,
+			     struct device_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%d\n", VIRTIO_RDMA_BOARD_ID);
+}
+static DEVICE_ATTR_RO(board_id);
+
+static struct attribute *virtio_rdmaa_class_attributes[] = {
+	&dev_attr_hw_rev.attr,
+	&dev_attr_hca_type.attr,
+	&dev_attr_board_id.attr,
+	NULL,
+};
+
+static const struct attribute_group virtio_rdmaa_attr_group = {
+	.attrs = virtio_rdmaa_class_attributes,
+};
+
+int init_ib(struct virtio_rdma_info *ri)
+{
+	int rc;
+
+	ri->ib_dev.owner = THIS_MODULE;
+	ri->ib_dev.num_comp_vectors = 1;
+	ri->ib_dev.dev.parent = &ri->vdev->dev;
+	ri->ib_dev.node_type = RDMA_NODE_IB_CA;
+	ri->ib_dev.phys_port_cnt = 1;
+	ri->ib_dev.uverbs_cmd_mask +		(1ull <<
IB_USER_VERBS_CMD_QUERY_DEVICE)	|
+		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
+		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
+		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
+		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
+		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD);
+
+	rdma_set_device_sysfs_group(&ri->ib_dev, &virtio_rdmaa_attr_group);
+
+	ib_set_device_ops(&ri->ib_dev, &virtio_rdma_dev_ops);
+
+	rc = ib_register_device(&ri->ib_dev, "virtio_rdma%d");
+
+	return rc;
+}
+
+void fini_ib(struct virtio_rdma_info *ri)
+{
+	ib_unregister_device(&ri->ib_dev);
+}
diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_ib.h
b/drivers/infiniband/hw/virtio/virtio_rdma_ib.h
new file mode 100644
index 000000000000..7b82a60581ff
--- /dev/null
+++ b/drivers/infiniband/hw/virtio/virtio_rdma_ib.h
@@ -0,0 +1,48 @@
+/*
+ * Virtio RDMA device: IB related functions and data
+ *
+ * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
+ */
+
+#ifndef __VIRTIO_RDMA_IB__
+#define __VIRTIO_RDMA_IB__
+
+#include <rdma/ib_verbs.h>
+
+struct virtio_rdma_pd {
+	struct ib_pd ibpd;
+	u32 pd_handle;
+};
+
+struct virtio_rdma_user_mr {
+	struct ib_mr ibmr;
+	u32 mr_handle;
+};
+
+struct virtio_rdma_qp {
+	struct ib_qp ibqp;
+};
+
+static inline struct virtio_rdma_pd *to_vpd(struct ib_pd *ibpd)
+{
+	return container_of(ibpd, struct virtio_rdma_pd, ibpd);
+}
+
+int init_ib(struct virtio_rdma_info *ri);
+void fini_ib(struct virtio_rdma_info *ri);
+
+#endif
diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_main.c
b/drivers/infiniband/hw/virtio/virtio_rdma_main.c
new file mode 100644
index 000000000000..811533d63160
--- /dev/null
+++ b/drivers/infiniband/hw/virtio/virtio_rdma_main.c
@@ -0,0 +1,149 @@
+/*
+ * Virtio RDMA device
+ *
+ * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
+ */
+
+#include <linux/err.h>
+#include <linux/scatterlist.h>
+#include <linux/spinlock.h>
+#include <linux/virtio.h>
+#include <linux/module.h>
+#include <uapi/linux/virtio_ids.h>
+
+#include "virtio_rdma.h"
+#include "virtio_rdma_device.h"
+#include "virtio_rdma_ib.h"
+#include "virtio_rdma_netdev.h"
+
+/* TODO:
+ * - How to hook to unload driver, we need to undo all the stuff with did
+ *   for all the devices that probed
+ * -
+ */
+
+static int virtio_rdma_probe(struct virtio_device *vdev)
+{
+	struct virtio_rdma_info *ri;
+	int rc = -EIO;
+
+	ri = ib_alloc_device(virtio_rdma_info, ib_dev);
+	if (!ri) {
+		pr_err("Fail to allocate IB device\n");
+		rc = -ENOMEM;
+		goto out;
+	}
+	vdev->priv = ri;
+
+	ri->vdev = vdev;
+
+	rc = init_device(ri);
+	if (rc) {
+		pr_err("Fail to connect to device\n");
+		goto out_dealloc_ib_device;
+	}
+
+	rc = init_netdev(ri);
+	if (rc) {
+		pr_err("Fail to connect to NetDev layer\n");
+		goto out_fini_device;
+	}
+
+	rc = init_ib(ri);
+	if (rc) {
+		pr_err("Fail to connect to IB layer\n");
+		goto out_fini_netdev;
+	}
+
+	pr_info("VirtIO RDMA device %d probed\n", vdev->index);
+
+	goto out;
+
+out_fini_netdev:
+	fini_netdev(ri);
+
+out_fini_device:
+	fini_device(ri);
+
+out_dealloc_ib_device:
+	ib_dealloc_device(&ri->ib_dev);
+
+	vdev->priv = NULL;
+
+out:
+	return rc;
+}
+
+static void virtio_rdma_remove(struct virtio_device *vdev)
+{
+	struct virtio_rdma_info *ri = vdev->priv;
+
+	if (!ri)
+		return;
+
+	vdev->priv = NULL;
+
+	fini_ib(ri);
+
+	fini_netdev(ri);
+
+	fini_device(ri);
+
+	ib_dealloc_device(&ri->ib_dev);
+
+	pr_info("VirtIO RDMA device %d removed\n", vdev->index);
+}
+
+static struct virtio_device_id id_table[] = {
+	{ VIRTIO_ID_RDMA, VIRTIO_DEV_ANY_ID },
+	{ 0 },
+};
+
+static struct virtio_driver virtio_rdma_driver = {
+	.driver.name	= KBUILD_MODNAME,
+	.driver.owner	= THIS_MODULE,
+	.id_table	= id_table,
+	.probe		= virtio_rdma_probe,
+	.remove		= virtio_rdma_remove,
+};
+
+static int __init virtio_rdma_init(void)
+{
+	int rc;
+
+	rc = register_virtio_driver(&virtio_rdma_driver);
+	if (rc) {
+		pr_err("%s: Fail to register virtio driver (%d)\n", __func__,
+		       rc);
+		return rc;
+	}
+
+	return 0;
+}
+
+static void __exit virtio_rdma_fini(void)
+{
+	unregister_virtio_driver(&virtio_rdma_driver);
+}
+
+module_init(virtio_rdma_init);
+module_exit(virtio_rdma_fini);
+
+MODULE_DEVICE_TABLE(virtio, id_table);
+MODULE_AUTHOR("Yuval Shaia");
+MODULE_DESCRIPTION("Virtio RDMA driver");
+MODULE_LICENSE("Dual BSD/GPL");
diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_netdev.c
b/drivers/infiniband/hw/virtio/virtio_rdma_netdev.c
new file mode 100644
index 000000000000..001f30b3e0b9
--- /dev/null
+++ b/drivers/infiniband/hw/virtio/virtio_rdma_netdev.c
@@ -0,0 +1,44 @@
+/*
+ * Virtio RDMA device
+ *
+ * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
+ */
+
+#include "virtio_rdma_netdev.h"
+
+int init_netdev(struct virtio_rdma_info *ri)
+{
+	struct net_device *dev;
+	struct virtio_rdma_netdev_info *vrndi;
+
+	dev = alloc_etherdev(sizeof(struct virtio_rdma_netdev_info));
+	if (!dev) {
+		return -ENOMEM;
+	}
+
+	SET_NETDEV_DEV(dev, &ri->vdev->dev);
+	vrndi = netdev_priv(dev);
+	vrndi->ri = ri;
+	ri->netdev = dev;
+
+	return 0;
+}
+
+void fini_netdev(struct virtio_rdma_info *ri)
+{
+	unregister_netdev(ri->netdev);
+}
diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_netdev.h
b/drivers/infiniband/hw/virtio/virtio_rdma_netdev.h
new file mode 100644
index 000000000000..e7e5d276d8ec
--- /dev/null
+++ b/drivers/infiniband/hw/virtio/virtio_rdma_netdev.h
@@ -0,0 +1,33 @@
+/*
+ * Virtio RDMA device: Netdev related functions and data
+ *
+ * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
+ */
+
+#ifndef __VIRTIO_RDMA_NETDEV__
+#define __VIRTIO_RDMA_NETDEV__
+
+#include "virtio_rdma.h"
+
+struct virtio_rdma_netdev_info {
+	struct virtio_rdma_info *ri;
+};
+
+int init_netdev(struct virtio_rdma_info *ri);
+void fini_netdev(struct virtio_rdma_info *ri);
+
+#endif
diff --git a/include/uapi/linux/virtio_ids.h b/include/uapi/linux/virtio_ids.h
index 6d5c3b2d4f4d..288ee6fec8d3 100644
--- a/include/uapi/linux/virtio_ids.h
+++ b/include/uapi/linux/virtio_ids.h
@@ -43,5 +43,6 @@
 #define VIRTIO_ID_INPUT        18 /* virtio input */
 #define VIRTIO_ID_VSOCK        19 /* virtio vsock transport */
 #define VIRTIO_ID_CRYPTO       20 /* virtio crypto */
+#define VIRTIO_ID_RDMA         26 /* RDMA */
 
 #endif /* _LINUX_VIRTIO_IDS_H */
-- 
2.17.0

On Thu, 11 Apr 2019 14:01:54 +0300
Yuval Shaia <yuval.shaia at oracle.com> wrote:
> Data center backends use more and more RDMA or RoCE devices and more and
> more software runs in virtualized environment.
> There is a need for a standard to enable RDMA/RoCE on Virtual Machines.
> 
> Virtio is the optimal solution since is the de-facto para-virtualizaton
> technology and also because the Virtio specification
> allows Hardware Vendors to support Virtio protocol natively in order to
> achieve bare metal performance.
> 
> This RFC is an effort to addresses challenges in defining the RDMA/RoCE
> Virtio Specification and a look forward on possible implementation
> techniques.
> 
> Open issues/Todo list:
> List is huge, this is only start point of the project.
> Anyway, here is one example of item in the list:
> - Multi VirtQ: Every QP has two rings and every CQ has one. This means that
>   in order to support for example 32K QPs we will need 64K VirtQ. Not sure
>   that this is reasonable so one option is to have one for all and
>   multiplex the traffic on it. This is not good approach as by design it
>   introducing an optional starvation. Another approach would be multi
>   queues and round-robin (for example) between them.
> 
> Expectations from this posting:
> In general, any comment is welcome, starting from hey, drop this as it is a
> very bad idea, to yeah, go ahead, we really want it.
> Idea here is that since it is not a minor effort i first want to know if
> there is some sort interest in the community for such device.
My first reaction is: Sounds sensible, but it would be good to have a
spec for this :)

You'll need a spec if you want this to go forward anyway, so at least a
sketch would be good to answer questions such as how many virtqueues
you use for which purpose, what is actually put on the virtqueues,
whether there are negotiable features, and what the expectations for
the device and the driver are. It also makes it easier to understand
how this is supposed to work in practice.

If folks agree that this sounds useful, the next step would be to
reserve an id for the device type.
> 
> The scope of the implementation is limited to probing the device and doing
> some basic ibverbs commands. Data-path is not yet implemented. So with this
> one can expect only that driver is (partialy) loaded and basic queries and
> resource allocation is done.
> 
> One note regarding the patchset.
> I know it is not standard to collaps patches from several repos as i did
> here (qemu and linux) but decided to do it anyway so the whole picture can
> be seen.
> 
> patch 1: virtio-net: Move some virtio-net-pci decl to include/hw/virtio
> 	This is a prelimenary patch just as a hack so i will not need to
> 	impelement new netdev
> patch 2: hw/virtio-rdma: VirtIO rdma device
> 	The implementation of the device
> patch 3: RDMA/virtio-rdma: VirtIO rdma driver
> 	The device driver
>

On Thu, Apr 11, 2019 at 07:02:15PM +0200, Cornelia Huck
wrote:
> On Thu, 11 Apr 2019 14:01:54 +0300
> Yuval Shaia <yuval.shaia at oracle.com> wrote:
> 
> > Data center backends use more and more RDMA or RoCE devices and more
and
> > more software runs in virtualized environment.
> > There is a need for a standard to enable RDMA/RoCE on Virtual
Machines.
> > 
> > Virtio is the optimal solution since is the de-facto
para-virtualizaton
> > technology and also because the Virtio specification
> > allows Hardware Vendors to support Virtio protocol natively in order
to
> > achieve bare metal performance.
> > 
> > This RFC is an effort to addresses challenges in defining the
RDMA/RoCE
> > Virtio Specification and a look forward on possible implementation
> > techniques.
> > 
> > Open issues/Todo list:
> > List is huge, this is only start point of the project.
> > Anyway, here is one example of item in the list:
> > - Multi VirtQ: Every QP has two rings and every CQ has one. This means
that
> >   in order to support for example 32K QPs we will need 64K VirtQ. Not
sure
> >   that this is reasonable so one option is to have one for all and
> >   multiplex the traffic on it. This is not good approach as by design
it
> >   introducing an optional starvation. Another approach would be multi
> >   queues and round-robin (for example) between them.
> > 
> > Expectations from this posting:
> > In general, any comment is welcome, starting from hey, drop this as it
is a
> > very bad idea, to yeah, go ahead, we really want it.
> > Idea here is that since it is not a minor effort i first want to know
if
> > there is some sort interest in the community for such device.
> 
> My first reaction is: Sounds sensible, but it would be good to have a
> spec for this :)
I'm unclear why you'd want to have a virtio queue for anything other
that some kind of command channel.

I'm not sure a QP or CQ benefits from this??

Jason

On 2019/4/11 19:01, Yuval Shaia wrote:
> Signed-off-by: Yuval Shaia <yuval.shaia at oracle.com>
> ---
>   drivers/infiniband/Kconfig                    |   1 +
>   drivers/infiniband/hw/Makefile                |   1 +
>   drivers/infiniband/hw/virtio/Kconfig          |   6 +
>   drivers/infiniband/hw/virtio/Makefile         |   4 +
>   drivers/infiniband/hw/virtio/virtio_rdma.h    |  40 +
>   .../infiniband/hw/virtio/virtio_rdma_device.c |  59 ++
>   .../infiniband/hw/virtio/virtio_rdma_device.h |  32 +
>   drivers/infiniband/hw/virtio/virtio_rdma_ib.c | 711 ++++++++++++++++++
>   drivers/infiniband/hw/virtio/virtio_rdma_ib.h |  48 ++
>   .../infiniband/hw/virtio/virtio_rdma_main.c   | 149 ++++
>   .../infiniband/hw/virtio/virtio_rdma_netdev.c |  44 ++
>   .../infiniband/hw/virtio/virtio_rdma_netdev.h |  33 +
>   include/uapi/linux/virtio_ids.h               |   1 +
>   13 files changed, 1129 insertions(+)
>   create mode 100644 drivers/infiniband/hw/virtio/Kconfig
>   create mode 100644 drivers/infiniband/hw/virtio/Makefile
>   create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma.h
>   create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_device.c
>   create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_device.h
>   create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_ib.c
>   create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_ib.h
>   create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_main.c
>   create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_netdev.c
>   create mode 100644 drivers/infiniband/hw/virtio/virtio_rdma_netdev.h
>
> diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig
> index a1fb840de45d..218a47d4cecf 100644
> --- a/drivers/infiniband/Kconfig
> +++ b/drivers/infiniband/Kconfig
> @@ -107,6 +107,7 @@ source "drivers/infiniband/hw/hfi1/Kconfig"
>   source "drivers/infiniband/hw/qedr/Kconfig"
>   source "drivers/infiniband/sw/rdmavt/Kconfig"
>   source "drivers/infiniband/sw/rxe/Kconfig"
> +source "drivers/infiniband/hw/virtio/Kconfig"
>   endif
>   
>   source "drivers/infiniband/ulp/ipoib/Kconfig"
> diff --git a/drivers/infiniband/hw/Makefile
b/drivers/infiniband/hw/Makefile
> index e4f31c1be8f7..10ffb2c421e4 100644
> --- a/drivers/infiniband/hw/Makefile
> +++ b/drivers/infiniband/hw/Makefile
> @@ -14,3 +14,4 @@ obj-$(CONFIG_INFINIBAND_HFI1)		+= hfi1/
>   obj-$(CONFIG_INFINIBAND_HNS)		+= hns/
>   obj-$(CONFIG_INFINIBAND_QEDR)		+= qedr/
>   obj-$(CONFIG_INFINIBAND_BNXT_RE)	+= bnxt_re/
> +obj-$(CONFIG_INFINIBAND_VIRTIO_RDMA)	+= virtio/
> diff --git a/drivers/infiniband/hw/virtio/Kconfig
b/drivers/infiniband/hw/virtio/Kconfig
> new file mode 100644
> index 000000000000..92e41691cf5d
> --- /dev/null
> +++ b/drivers/infiniband/hw/virtio/Kconfig
> @@ -0,0 +1,6 @@
> +config INFINIBAND_VIRTIO_RDMA
> +	tristate "VirtIO Paravirtualized RDMA Driver"
> +	depends on NETDEVICES && ETHERNET && PCI && INET
> +	---help---
> +	  This driver provides low-level support for VirtIO Paravirtual
> +	  RDMA adapter.
> diff --git a/drivers/infiniband/hw/virtio/Makefile
b/drivers/infiniband/hw/virtio/Makefile
> new file mode 100644
> index 000000000000..fb637e467167
> --- /dev/null
> +++ b/drivers/infiniband/hw/virtio/Makefile
> @@ -0,0 +1,4 @@
> +obj-$(CONFIG_INFINIBAND_VIRTIO_RDMA) += virtio_rdma.o
> +
> +virtio_rdma-y := virtio_rdma_main.o virtio_rdma_device.o virtio_rdma_ib.o
\
> +		 virtio_rdma_netdev.o
> diff --git a/drivers/infiniband/hw/virtio/virtio_rdma.h
b/drivers/infiniband/hw/virtio/virtio_rdma.h
> new file mode 100644
> index 000000000000..7896a2dfb812
> --- /dev/null
> +++ b/drivers/infiniband/hw/virtio/virtio_rdma.h
> @@ -0,0 +1,40 @@
> +/*
> + * Virtio RDMA device: Driver main data types
> + *
> + * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301
USA
> + */
> +
> +#ifndef __VIRTIO_RDMA__
> +#define __VIRTIO_RDMA__
> +
> +#include <linux/virtio.h>
> +#include <rdma/ib_verbs.h>
> +
> +struct virtio_rdma_info {
> +	struct ib_device ib_dev;
> +	struct virtio_device *vdev;
> +	struct virtqueue *ctrl_vq;
> +	wait_queue_head_t acked; /* arm on send to host, release on recv */
> +	struct net_device *netdev;
> +};
> +
> +static inline struct virtio_rdma_info *to_vdev(struct ib_device *ibdev)
> +{
> +	return container_of(ibdev, struct virtio_rdma_info, ib_dev);
> +}
> +
> +#endif
> diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_device.c
b/drivers/infiniband/hw/virtio/virtio_rdma_device.c
> new file mode 100644
> index 000000000000..ae41e530644f
> --- /dev/null
> +++ b/drivers/infiniband/hw/virtio/virtio_rdma_device.c
> @@ -0,0 +1,59 @@
> +/*
> + * Virtio RDMA device: Device related functions and data
> + *
> + * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301
USA
> + */
> +
> +#include <linux/virtio_config.h>
> +
> +#include "virtio_rdma.h"
> +
> +static void rdma_ctrl_ack(struct virtqueue *vq)
> +{
> +	struct virtio_rdma_info *dev = vq->vdev->priv;
> +
> +	wake_up(&dev->acked);
> +
> +	printk("%s\n", __func__);
Cool:-)

this line should be for debug?

Zhu Yanjun
> +}
> +
> +int init_device(struct virtio_rdma_info *dev)
> +{
> +#define TMP_MAX_VQ 1
> +	int rc;
> +	struct virtqueue *vqs[TMP_MAX_VQ];
> +	vq_callback_t *cbs[TMP_MAX_VQ];
> +	const char *names[TMP_MAX_VQ];
> +
> +	names[0] = "ctrl";
> +	cbs[0] = rdma_ctrl_ack;
> +	cbs[0] = NULL;
> +
> +	rc = virtio_find_vqs(dev->vdev, TMP_MAX_VQ, vqs, cbs, names, NULL);
> +	if (rc)
> +		return rc;
> +
> +	dev->ctrl_vq = vqs[0];
> +
> +	return 0;
> +}
> +
> +void fini_device(struct virtio_rdma_info *dev)
> +{
> +	dev->vdev->config->reset(dev->vdev);
> +	dev->vdev->config->del_vqs(dev->vdev);
> +}
> diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_device.h
b/drivers/infiniband/hw/virtio/virtio_rdma_device.h
> new file mode 100644
> index 000000000000..d9b1240daf92
> --- /dev/null
> +++ b/drivers/infiniband/hw/virtio/virtio_rdma_device.h
> @@ -0,0 +1,32 @@
> +/*
> + * Virtio RDMA device: Device related functions and data
> + *
> + * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301
USA
> + */
> +
> +#ifndef __VIRTIO_RDMA_DEVICE__
> +#define __VIRTIO_RDMA_DEVICE__
> +
> +#define VIRTIO_RDMA_BOARD_ID	1
> +#define VIRTIO_RDMA_HW_NAME	"virtio-rdma"
> +#define VIRTIO_RDMA_HW_REV	1
> +#define VIRTIO_RDMA_DRIVER_VER	"1.0"
> +
> +int init_device(struct virtio_rdma_info *dev);
> +void fini_device(struct virtio_rdma_info *dev);
> +
> +#endif
> diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_ib.c
b/drivers/infiniband/hw/virtio/virtio_rdma_ib.c
> new file mode 100644
> index 000000000000..02bf4a332611
> --- /dev/null
> +++ b/drivers/infiniband/hw/virtio/virtio_rdma_ib.c
> @@ -0,0 +1,711 @@
> +/*
> + * Virtio RDMA device: IB related functions and data
> + *
> + * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301
USA
> + */
> +
> +#include <linux/scatterlist.h>
> +#include <linux/virtio.h>
> +#include <rdma/ib_mad.h>
> +
> +#include "virtio_rdma.h"
> +#include "virtio_rdma_device.h"
> +#include "virtio_rdma_ib.h"
> +
> +/* TODO: Move to uapi header file */
> +
> +/*
> + * Control virtqueue data structures
> + *
> + * The control virtqueue expects a header in the first sg entry
> + * and an ack/status response in the last entry.  Data for the
> + * command goes in between.
> + */
> +
> +#define VIRTIO_RDMA_CTRL_OK	0
> +#define VIRTIO_RDMA_CTRL_ERR	1
> +
> +struct control_buf {
> +	__u8 cmd;
> +	__u8 status;
> +};
> +
> +enum {
> +	VIRTIO_CMD_QUERY_DEVICE = 10,
> +	VIRTIO_CMD_QUERY_PORT,
> +	VIRTIO_CMD_CREATE_CQ,
> +	VIRTIO_CMD_DESTROY_CQ,
> +	VIRTIO_CMD_CREATE_PD,
> +	VIRTIO_CMD_DESTROY_PD,
> +	VIRTIO_CMD_GET_DMA_MR,
> +};
> +
> +struct cmd_query_port {
> +	__u8 port;
> +};
> +
> +struct cmd_create_cq {
> +	__u32 cqe;
> +};
> +
> +struct rsp_create_cq {
> +	__u32 cqn;
> +};
> +
> +struct cmd_destroy_cq {
> +	__u32 cqn;
> +};
> +
> +struct rsp_create_pd {
> +	__u32 pdn;
> +};
> +
> +struct cmd_destroy_pd {
> +	__u32 pdn;
> +};
> +
> +struct cmd_get_dma_mr {
> +	__u32 pdn;
> +	__u32 access_flags;
> +};
> +
> +struct rsp_get_dma_mr {
> +	__u32 mrn;
> +	__u32 lkey;
> +	__u32 rkey;
> +};
> +
> +/* TODO: Move to uapi header file */
> +
> +struct virtio_rdma_ib_cq {
> +	struct ib_cq ibcq;
> +	u32 cq_handle;
> +};
> +
> +/* TODO: For the scope fof the RFC i'm utilizing ib*_*_attr structures
*/
> +
> +static int virtio_rdma_exec_cmd(struct virtio_rdma_info *di, int cmd,
> +				struct scatterlist *in, struct scatterlist *out)
> +{
> +	struct scatterlist *sgs[4], hdr, status;
> +	struct control_buf *ctrl;
> +	unsigned tmp;
> +	int rc;
> +
> +	ctrl = kmalloc(sizeof(*ctrl), GFP_ATOMIC);
> +	ctrl->cmd = cmd;
> +	ctrl->status = ~0;
> +
> +	sg_init_one(&hdr, &ctrl->cmd, sizeof(ctrl->cmd));
> +	sgs[0] = &hdr;
> +	sgs[1] = in;
> +	sgs[2] = out;
> +	sg_init_one(&status, &ctrl->status, sizeof(ctrl->status));
> +	sgs[3] = &status;
> +
> +	rc = virtqueue_add_sgs(di->ctrl_vq, sgs, 2, 2, di, GFP_ATOMIC);
> +	if (rc)
> +		goto out;
> +
> +	if (unlikely(!virtqueue_kick(di->ctrl_vq))) {
> +		goto out_with_status;
> +	}
> +
> +	/* Spin for a response, the kick causes an ioport write, trapping
> +	 * into the hypervisor, so the request should be handled
> +	 * immediately */
> +	while (!virtqueue_get_buf(di->ctrl_vq, &tmp) &&
> +	       !virtqueue_is_broken(di->ctrl_vq))
> +		cpu_relax();
> +
> +out_with_status:
> +	printk("%s: cmd %d, status %d\n", __func__, ctrl->cmd,
ctrl->status);
> +	rc = ctrl->status == VIRTIO_RDMA_CTRL_OK ? 0 : 1;
> +
> +out:
> +	kfree(ctrl);
> +
> +	return rc;
> +}
> +
> +static int virtio_rdma_port_immutable(struct ib_device *ibdev, u8
port_num,
> +				      struct ib_port_immutable *immutable)
> +{
> +	struct ib_port_attr attr;
> +	int rc;
> +
> +	rc = ib_query_port(ibdev, port_num, &attr);
> +	if (rc)
> +		return rc;
> +
> +	immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
> +	immutable->pkey_tbl_len = attr.pkey_tbl_len;
> +	immutable->gid_tbl_len = attr.gid_tbl_len;
> +	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
> +
> +	return 0;
> +}
> +
> +static int virtio_rdma_query_device(struct ib_device *ibdev,
> +				    struct ib_device_attr *props,
> +				    struct ib_udata *uhw)
> +{
> +	struct scatterlist data;
> +	int offs;
> +	int rc;
> +
> +	if (uhw->inlen || uhw->outlen)
> +		return -EINVAL;
> +
> +	/* We start with sys_image_guid because of inconsistency beween ib_
> +	 * and ibv_ */
> +	offs = offsetof(struct ib_device_attr, sys_image_guid);
> +	sg_init_one(&data, (void *)props + offs, sizeof(*props) - offs);
> +
> +	rc = virtio_rdma_exec_cmd(to_vdev(ibdev), VIRTIO_CMD_QUERY_DEVICE, NULL,
> +				  &data);
> +
> +	printk("%s: sys_image_guid 0x%llx\n", __func__,
> +	       be64_to_cpu(props->sys_image_guid));
> +
> +	return rc;
> +}
> +
> +static int virtio_rdma_query_port(struct ib_device *ibdev, u8 port,
> +				  struct ib_port_attr *props)
> +{
> +	struct scatterlist in, out;
> +	struct cmd_query_port *cmd;
> +	int offs;
> +	int rc;
> +
> +	cmd = kmalloc(sizeof(*cmd), GFP_ATOMIC);
> +	if (!cmd)
> +		return -ENOMEM;
> +
> +	/* We start with state because of inconsistency beween ib and ibv */
> +	offs = offsetof(struct ib_port_attr, state);
> +	sg_init_one(&out, (void *)props + offs, sizeof(*props) - offs);
> +
> +	cmd->port = port;
> +	sg_init_one(&in, cmd, sizeof(*cmd));
> +	printk("%s: port %d\n", __func__, cmd->port);
> +
> +	rc = virtio_rdma_exec_cmd(to_vdev(ibdev), VIRTIO_CMD_QUERY_PORT, &in,
> +				  &out);
> +
> +	printk("%s: gid_tbl_len %d\n", __func__,
props->gid_tbl_len);
> +
> +	kfree(cmd);
> +
> +	return rc;
> +}
> +
> +static struct net_device *virtio_rdma_get_netdev(struct ib_device *ibdev,
> +						 u8 port_num)
> +{
> +	struct virtio_rdma_info *ri = to_vdev(ibdev);
> +
> +	printk("%s:\n", __func__);
> +
> +	return ri->netdev;
> +}
> +
> +struct ib_cq *virtio_rdma_create_cq(struct ib_device *ibdev,
> +				    const struct ib_cq_init_attr *attr,
> +				    struct ib_ucontext *context,
> +				    struct ib_udata *udata)
> +{
> +	struct scatterlist in, out;
> +	struct virtio_rdma_ib_cq *vcq;
> +	struct cmd_create_cq *cmd;
> +	struct rsp_create_cq *rsp;
> +	struct ib_cq *cq = NULL;
> +	int rc;
> +
> +	/* TODO: Check MAX_CQ */
> +
> +	cmd = kmalloc(sizeof(*cmd), GFP_ATOMIC);
> +	if (!cmd)
> +		return ERR_PTR(-ENOMEM);
> +
> +	rsp = kmalloc(sizeof(*rsp), GFP_ATOMIC);
> +	if (!rsp) {
> +		kfree(cmd);
> +		return ERR_PTR(-ENOMEM);
> +	}
> +
> +	vcq = kzalloc(sizeof(*vcq), GFP_KERNEL);
> +	if (!vcq)
> +		goto out;
> +
> +	cmd->cqe = attr->cqe;
> +	sg_init_one(&in, cmd, sizeof(*cmd));
> +	printk("%s: cqe %d\n", __func__, cmd->cqe);
> +
> +	sg_init_one(&out, rsp, sizeof(*rsp));
> +
> +	rc = virtio_rdma_exec_cmd(to_vdev(ibdev), VIRTIO_CMD_CREATE_CQ, &in,
> +				  &out);
> +	if (rc)
> +		goto out_err;
> +
> +	printk("%s: cqn 0x%x\n", __func__, rsp->cqn);
> +	vcq->cq_handle = rsp->cqn;
> +	vcq->ibcq.cqe = attr->cqe;
> +	cq = &vcq->ibcq;
> +
> +	goto out;
> +
> +out_err:
> +	kfree(vcq);
> +	return ERR_PTR(rc);
> +
> +out:
> +	kfree(rsp);
> +	kfree(cmd);
> +	return cq;
> +}
> +
> +int virtio_rdma_destroy_cq(struct ib_cq *cq)
> +{
> +	struct virtio_rdma_ib_cq *vcq;
> +	struct scatterlist in;
> +	struct cmd_destroy_cq *cmd;
> +	int rc;
> +
> +	cmd = kmalloc(sizeof(*cmd), GFP_ATOMIC);
> +	if (!cmd)
> +		return -ENOMEM;
> +
> +	vcq = container_of(cq, struct virtio_rdma_ib_cq, ibcq);
> +
> +	cmd->cqn = vcq->cq_handle;
> +	sg_init_one(&in, cmd, sizeof(*cmd));
> +
> +	rc = virtio_rdma_exec_cmd(to_vdev(cq->device), VIRTIO_CMD_DESTROY_CQ,
> +				  &in, NULL);
> +
> +	kfree(cmd);
> +
> +	kfree(vcq);
> +
> +	return rc;
> +}
> +
> +int virtio_rdma_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context,
> +			 struct ib_udata *udata)
> +{
> +	struct virtio_rdma_pd *pd = to_vpd(ibpd);
> +	struct ib_device *ibdev = ibpd->device;
> +	struct rsp_create_pd *rsp;
> +	struct scatterlist out;
> +	int rc;
> +
> +	/* TODO: Check MAX_PD */
> +
> +	rsp = kmalloc(sizeof(*rsp), GFP_ATOMIC);
> +	if (!rsp)
> +		return -ENOMEM;
> +
> +	sg_init_one(&out, rsp, sizeof(*rsp));
> +
> +	rc = virtio_rdma_exec_cmd(to_vdev(ibdev), VIRTIO_CMD_CREATE_PD, NULL,
> +				  &out);
> +	if (rc)
> +		goto out;
> +
> +	pd->pd_handle = rsp->pdn;
> +
> +	printk("%s: pd_handle=%d\n", __func__, pd->pd_handle);
> +
> +out:
> +	kfree(rsp);
> +
> +	printk("%s: rc=%d\n", __func__, rc);
> +	return rc;
> +}
> +
> +void virtio_rdma_dealloc_pd(struct ib_pd *pd)
> +{
> +	struct virtio_rdma_pd *vpd = to_vpd(pd);
> +	struct ib_device *ibdev = pd->device;
> +	struct cmd_destroy_pd *cmd;
> +	struct scatterlist in;
> +
> +	printk("%s:\n", __func__);
> +
> +	cmd = kmalloc(sizeof(*cmd), GFP_ATOMIC);
> +	if (!cmd)
> +		return;
> +
> +	cmd->pdn = vpd->pd_handle;
> +	sg_init_one(&in, cmd, sizeof(*cmd));
> +
> +	virtio_rdma_exec_cmd(to_vdev(ibdev), VIRTIO_CMD_DESTROY_PD, &in,
NULL);
> +
> +	kfree(cmd);
> +}
> +
> +struct ib_mr *virtio_rdma_get_dma_mr(struct ib_pd *pd, int acc)
> +
> +{
> +	struct virtio_rdma_user_mr *mr;
> +	struct scatterlist in, out;
> +	struct cmd_get_dma_mr *cmd = NULL;
> +	struct rsp_get_dma_mr *rsp = NULL;
> +	int rc;
> +
> +	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
> +	if (!mr)
> +		return ERR_PTR(-ENOMEM);
> +
> +	cmd = kmalloc(sizeof(*cmd), GFP_ATOMIC);
> +	if (!cmd) {
> +		kfree(mr);
> +		return ERR_PTR(-ENOMEM);
> +	}
> +
> +	rsp = kmalloc(sizeof(*rsp), GFP_ATOMIC);
> +	if (!cmd) {
> +		kfree(mr);
> +		kfree(cmd);
> +		return ERR_PTR(-ENOMEM);
> +	}
> +
> +	cmd->pdn = to_vpd(pd)->pd_handle;
> +	cmd->access_flags = acc;
> +	sg_init_one(&in, cmd, sizeof(*cmd));
> +
> +	sg_init_one(&out, rsp, sizeof(*rsp));
> +
> +	rc = virtio_rdma_exec_cmd(to_vdev(pd->device), VIRTIO_CMD_GET_DMA_MR,
> +				  &in, &out);
> +	if (rc) {
> +		kfree(mr);
> +		kfree(cmd);
> +		return ERR_PTR(rc);
> +	}
> +
> +	mr->mr_handle = rsp->mrn;
> +	mr->ibmr.lkey = rsp->lkey;
> +	mr->ibmr.rkey = rsp->rkey;
> +
> +	printk("%s: mr_handle=0x%x\n", __func__, mr->mr_handle);
> +
> +	kfree(cmd);
> +	kfree(rsp);
> +
> +	return &mr->ibmr;
> +}
> +
> +struct ib_qp *virtio_rdma_create_qp(struct ib_pd *pd,
> +				    struct ib_qp_init_attr *init_attr,
> +				    struct ib_udata *udata)
> +{
> +	/* struct pvrdma_dev *dev = to_vdev(pd->device); */
> +	struct virtio_rdma_qp *qp;
> +
> +	printk("%s:\n", __func__);
> +
> +	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
> +	if (!qp)
> +		return ERR_PTR(-ENOMEM);
> +
> +	return &qp->ibqp;
> +}
> +
> +int virtio_rdma_query_gid(struct ib_device *ibdev, u8 port, int index,
> +			  union ib_gid *gid)
> +{
> +	memset(gid, 0, sizeof(union ib_gid));
> +
> +	printk("%s: port %d, index %d\n", __func__, port, index);
> +
> +	return 0;
> +}
> +
> +static int virtio_rdma_add_gid(const struct ib_gid_attr *attr, void
**context)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +struct ib_mr *virtio_rdma_alloc_mr(struct ib_pd *pd, enum ib_mr_type
mr_type,
> +				   u32 max_num_sg)
> +{
> +	printk("%s: mr_type %d, max_num_sg %d\n", __func__, mr_type,
> +	       max_num_sg);
> +
> +	return NULL;
> +}
> +
> +int virtio_rdma_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata
*udata)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +struct ib_ah *virtio_rdma_create_ah(struct ib_pd *pd,
> +				    struct rdma_ah_attr *ah_attr, u32 flags,
> +				    struct ib_udata *udata)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return NULL;
> +}
> +
> +void virtio_rdma_dealloc_ucontext(struct ib_ucontext *ibcontext)
> +
> +{
> +}
> +
> +static int virtio_rdma_del_gid(const struct ib_gid_attr *attr, void
**context)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +int virtio_rdma_dereg_mr(struct ib_mr *ibmr)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +int virtio_rdma_destroy_ah(struct ib_ah *ah, u32 flags)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +struct virtio_rdma_cq {
> +	struct ib_cq ibcq;
> +};
> +
> +int virtio_rdma_destroy_qp(struct ib_qp *qp)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +static void virtio_rdma_get_fw_ver_str(struct ib_device *device, char
*str)
> +{
> +	printk("%s:\n", __func__);
> +}
> +
> +enum rdma_link_layer virtio_rdma_port_link_layer(struct ib_device *ibdev,
> +						 u8 port)
> +{
> +	return IB_LINK_LAYER_ETHERNET;
> +}
> +
> +int virtio_rdma_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
> +			  int sg_nents, unsigned int *sg_offset)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +int virtio_rdma_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct
*vma)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +int virtio_rdma_modify_port(struct ib_device *ibdev, u8 port, int mask,
> +			    struct ib_port_modify *props)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +int virtio_rdma_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
> +			  int attr_mask, struct ib_udata *udata)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +int virtio_rdma_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc
*wc)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +int virtio_rdma_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
> +			  const struct ib_recv_wr **bad_wr)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +int virtio_rdma_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
> +			  const struct ib_send_wr **bad_wr)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +int virtio_rdma_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
> +			   u16 *pkey)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +int virtio_rdma_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
> +			 int attr_mask, struct ib_qp_init_attr *init_attr)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +struct ib_mr *virtio_rdma_reg_user_mr(struct ib_pd *pd, u64 start, u64
length,
> +				      u64 virt_addr, int access_flags,
> +				      struct ib_udata *udata)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return NULL;
> +}
> +
> +int virtio_rdma_req_notify_cq(struct ib_cq *ibcq,
> +			      enum ib_cq_notify_flags notify_flags)
> +{
> +	printk("%s:\n", __func__);
> +
> +	return 0;
> +}
> +
> +static const struct ib_device_ops virtio_rdma_dev_ops = {
> +	.get_port_immutable = virtio_rdma_port_immutable,
> +	.query_device = virtio_rdma_query_device,
> +	.query_port = virtio_rdma_query_port,
> +	.get_netdev = virtio_rdma_get_netdev,
> +	.create_cq = virtio_rdma_create_cq,
> +	.destroy_cq = virtio_rdma_destroy_cq,
> +	.alloc_pd = virtio_rdma_alloc_pd,
> +	.dealloc_pd = virtio_rdma_dealloc_pd,
> +	.get_dma_mr = virtio_rdma_get_dma_mr,
> +	.create_qp = virtio_rdma_create_qp,
> +	.query_gid = virtio_rdma_query_gid,
> +	.add_gid = virtio_rdma_add_gid,
> +	.alloc_mr = virtio_rdma_alloc_mr,
> +	.alloc_ucontext = virtio_rdma_alloc_ucontext,
> +	.create_ah = virtio_rdma_create_ah,
> +	.dealloc_ucontext = virtio_rdma_dealloc_ucontext,
> +	.del_gid = virtio_rdma_del_gid,
> +	.dereg_mr = virtio_rdma_dereg_mr,
> +	.destroy_ah = virtio_rdma_destroy_ah,
> +	.destroy_qp = virtio_rdma_destroy_qp,
> +	.get_dev_fw_str = virtio_rdma_get_fw_ver_str,
> +	.get_link_layer = virtio_rdma_port_link_layer,
> +	.get_port_immutable = virtio_rdma_port_immutable,
> +	.map_mr_sg = virtio_rdma_map_mr_sg,
> +	.mmap = virtio_rdma_mmap,
> +	.modify_port = virtio_rdma_modify_port,
> +	.modify_qp = virtio_rdma_modify_qp,
> +	.poll_cq = virtio_rdma_poll_cq,
> +	.post_recv = virtio_rdma_post_recv,
> +	.post_send = virtio_rdma_post_send,
> +	.query_device = virtio_rdma_query_device,
> +	.query_pkey = virtio_rdma_query_pkey,
> +	.query_port = virtio_rdma_query_port,
> +	.query_qp = virtio_rdma_query_qp,
> +	.reg_user_mr = virtio_rdma_reg_user_mr,
> +	.req_notify_cq = virtio_rdma_req_notify_cq,
> +	INIT_RDMA_OBJ_SIZE(ib_pd, virtio_rdma_pd, ibpd),
> +};
> +
> +static ssize_t hca_type_show(struct device *device,
> +			     struct device_attribute *attr, char *buf)
> +{
> +	return sprintf(buf, "%s-%s\n", VIRTIO_RDMA_HW_NAME,
> +		       VIRTIO_RDMA_DRIVER_VER);
> +}
> +static DEVICE_ATTR_RO(hca_type);
> +
> +static ssize_t hw_rev_show(struct device *device,
> +			   struct device_attribute *attr, char *buf)
> +{
> +	return sprintf(buf, "%d\n", VIRTIO_RDMA_HW_REV);
> +}
> +static DEVICE_ATTR_RO(hw_rev);
> +
> +static ssize_t board_id_show(struct device *device,
> +			     struct device_attribute *attr, char *buf)
> +{
> +	return sprintf(buf, "%d\n", VIRTIO_RDMA_BOARD_ID);
> +}
> +static DEVICE_ATTR_RO(board_id);
> +
> +static struct attribute *virtio_rdmaa_class_attributes[] = {
> +	&dev_attr_hw_rev.attr,
> +	&dev_attr_hca_type.attr,
> +	&dev_attr_board_id.attr,
> +	NULL,
> +};
> +
> +static const struct attribute_group virtio_rdmaa_attr_group = {
> +	.attrs = virtio_rdmaa_class_attributes,
> +};
> +
> +int init_ib(struct virtio_rdma_info *ri)
> +{
> +	int rc;
> +
> +	ri->ib_dev.owner = THIS_MODULE;
> +	ri->ib_dev.num_comp_vectors = 1;
> +	ri->ib_dev.dev.parent = &ri->vdev->dev;
> +	ri->ib_dev.node_type = RDMA_NODE_IB_CA;
> +	ri->ib_dev.phys_port_cnt = 1;
> +	ri->ib_dev.uverbs_cmd_mask > +		(1ull <<
IB_USER_VERBS_CMD_QUERY_DEVICE)	|
> +		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
> +		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
> +		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
> +		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
> +		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD);
> +
> +	rdma_set_device_sysfs_group(&ri->ib_dev,
&virtio_rdmaa_attr_group);
> +
> +	ib_set_device_ops(&ri->ib_dev, &virtio_rdma_dev_ops);
> +
> +	rc = ib_register_device(&ri->ib_dev, "virtio_rdma%d");
> +
> +	return rc;
> +}
> +
> +void fini_ib(struct virtio_rdma_info *ri)
> +{
> +	ib_unregister_device(&ri->ib_dev);
> +}
> diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_ib.h
b/drivers/infiniband/hw/virtio/virtio_rdma_ib.h
> new file mode 100644
> index 000000000000..7b82a60581ff
> --- /dev/null
> +++ b/drivers/infiniband/hw/virtio/virtio_rdma_ib.h
> @@ -0,0 +1,48 @@
> +/*
> + * Virtio RDMA device: IB related functions and data
> + *
> + * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301
USA
> + */
> +
> +#ifndef __VIRTIO_RDMA_IB__
> +#define __VIRTIO_RDMA_IB__
> +
> +#include <rdma/ib_verbs.h>
> +
> +struct virtio_rdma_pd {
> +	struct ib_pd ibpd;
> +	u32 pd_handle;
> +};
> +
> +struct virtio_rdma_user_mr {
> +	struct ib_mr ibmr;
> +	u32 mr_handle;
> +};
> +
> +struct virtio_rdma_qp {
> +	struct ib_qp ibqp;
> +};
> +
> +static inline struct virtio_rdma_pd *to_vpd(struct ib_pd *ibpd)
> +{
> +	return container_of(ibpd, struct virtio_rdma_pd, ibpd);
> +}
> +
> +int init_ib(struct virtio_rdma_info *ri);
> +void fini_ib(struct virtio_rdma_info *ri);
> +
> +#endif
> diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_main.c
b/drivers/infiniband/hw/virtio/virtio_rdma_main.c
> new file mode 100644
> index 000000000000..811533d63160
> --- /dev/null
> +++ b/drivers/infiniband/hw/virtio/virtio_rdma_main.c
> @@ -0,0 +1,149 @@
> +/*
> + * Virtio RDMA device
> + *
> + * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301
USA
> + */
> +
> +#include <linux/err.h>
> +#include <linux/scatterlist.h>
> +#include <linux/spinlock.h>
> +#include <linux/virtio.h>
> +#include <linux/module.h>
> +#include <uapi/linux/virtio_ids.h>
> +
> +#include "virtio_rdma.h"
> +#include "virtio_rdma_device.h"
> +#include "virtio_rdma_ib.h"
> +#include "virtio_rdma_netdev.h"
> +
> +/* TODO:
> + * - How to hook to unload driver, we need to undo all the stuff with did
> + *   for all the devices that probed
> + * -
> + */
> +
> +static int virtio_rdma_probe(struct virtio_device *vdev)
> +{
> +	struct virtio_rdma_info *ri;
> +	int rc = -EIO;
> +
> +	ri = ib_alloc_device(virtio_rdma_info, ib_dev);
> +	if (!ri) {
> +		pr_err("Fail to allocate IB device\n");
> +		rc = -ENOMEM;
> +		goto out;
> +	}
> +	vdev->priv = ri;
> +
> +	ri->vdev = vdev;
> +
> +	rc = init_device(ri);
> +	if (rc) {
> +		pr_err("Fail to connect to device\n");
> +		goto out_dealloc_ib_device;
> +	}
> +
> +	rc = init_netdev(ri);
> +	if (rc) {
> +		pr_err("Fail to connect to NetDev layer\n");
> +		goto out_fini_device;
> +	}
> +
> +	rc = init_ib(ri);
> +	if (rc) {
> +		pr_err("Fail to connect to IB layer\n");
> +		goto out_fini_netdev;
> +	}
> +
> +	pr_info("VirtIO RDMA device %d probed\n", vdev->index);
> +
> +	goto out;
> +
> +out_fini_netdev:
> +	fini_netdev(ri);
> +
> +out_fini_device:
> +	fini_device(ri);
> +
> +out_dealloc_ib_device:
> +	ib_dealloc_device(&ri->ib_dev);
> +
> +	vdev->priv = NULL;
> +
> +out:
> +	return rc;
> +}
> +
> +static void virtio_rdma_remove(struct virtio_device *vdev)
> +{
> +	struct virtio_rdma_info *ri = vdev->priv;
> +
> +	if (!ri)
> +		return;
> +
> +	vdev->priv = NULL;
> +
> +	fini_ib(ri);
> +
> +	fini_netdev(ri);
> +
> +	fini_device(ri);
> +
> +	ib_dealloc_device(&ri->ib_dev);
> +
> +	pr_info("VirtIO RDMA device %d removed\n", vdev->index);
> +}
> +
> +static struct virtio_device_id id_table[] = {
> +	{ VIRTIO_ID_RDMA, VIRTIO_DEV_ANY_ID },
> +	{ 0 },
> +};
> +
> +static struct virtio_driver virtio_rdma_driver = {
> +	.driver.name	= KBUILD_MODNAME,
> +	.driver.owner	= THIS_MODULE,
> +	.id_table	= id_table,
> +	.probe		= virtio_rdma_probe,
> +	.remove		= virtio_rdma_remove,
> +};
> +
> +static int __init virtio_rdma_init(void)
> +{
> +	int rc;
> +
> +	rc = register_virtio_driver(&virtio_rdma_driver);
> +	if (rc) {
> +		pr_err("%s: Fail to register virtio driver (%d)\n", __func__,
> +		       rc);
> +		return rc;
> +	}
> +
> +	return 0;
> +}
> +
> +static void __exit virtio_rdma_fini(void)
> +{
> +	unregister_virtio_driver(&virtio_rdma_driver);
> +}
> +
> +module_init(virtio_rdma_init);
> +module_exit(virtio_rdma_fini);
> +
> +MODULE_DEVICE_TABLE(virtio, id_table);
> +MODULE_AUTHOR("Yuval Shaia");
> +MODULE_DESCRIPTION("Virtio RDMA driver");
> +MODULE_LICENSE("Dual BSD/GPL");
> diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_netdev.c
b/drivers/infiniband/hw/virtio/virtio_rdma_netdev.c
> new file mode 100644
> index 000000000000..001f30b3e0b9
> --- /dev/null
> +++ b/drivers/infiniband/hw/virtio/virtio_rdma_netdev.c
> @@ -0,0 +1,44 @@
> +/*
> + * Virtio RDMA device
> + *
> + * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301
USA
> + */
> +
> +#include "virtio_rdma_netdev.h"
> +
> +int init_netdev(struct virtio_rdma_info *ri)
> +{
> +	struct net_device *dev;
> +	struct virtio_rdma_netdev_info *vrndi;
> +
> +	dev = alloc_etherdev(sizeof(struct virtio_rdma_netdev_info));
> +	if (!dev) {
> +		return -ENOMEM;
> +	}
> +
> +	SET_NETDEV_DEV(dev, &ri->vdev->dev);
> +	vrndi = netdev_priv(dev);
> +	vrndi->ri = ri;
> +	ri->netdev = dev;
> +
> +	return 0;
> +}
> +
> +void fini_netdev(struct virtio_rdma_info *ri)
> +{
> +	unregister_netdev(ri->netdev);
> +}
> diff --git a/drivers/infiniband/hw/virtio/virtio_rdma_netdev.h
b/drivers/infiniband/hw/virtio/virtio_rdma_netdev.h
> new file mode 100644
> index 000000000000..e7e5d276d8ec
> --- /dev/null
> +++ b/drivers/infiniband/hw/virtio/virtio_rdma_netdev.h
> @@ -0,0 +1,33 @@
> +/*
> + * Virtio RDMA device: Netdev related functions and data
> + *
> + * Copyright (C) 2019 Yuval Shaia Oracle 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301
USA
> + */
> +
> +#ifndef __VIRTIO_RDMA_NETDEV__
> +#define __VIRTIO_RDMA_NETDEV__
> +
> +#include "virtio_rdma.h"
> +
> +struct virtio_rdma_netdev_info {
> +	struct virtio_rdma_info *ri;
> +};
> +
> +int init_netdev(struct virtio_rdma_info *ri);
> +void fini_netdev(struct virtio_rdma_info *ri);
> +
> +#endif
> diff --git a/include/uapi/linux/virtio_ids.h
b/include/uapi/linux/virtio_ids.h
> index 6d5c3b2d4f4d..288ee6fec8d3 100644
> --- a/include/uapi/linux/virtio_ids.h
> +++ b/include/uapi/linux/virtio_ids.h
> @@ -43,5 +43,6 @@
>   #define VIRTIO_ID_INPUT        18 /* virtio input */
>   #define VIRTIO_ID_VSOCK        19 /* virtio vsock transport */
>   #define VIRTIO_ID_CRYPTO       20 /* virtio crypto */
> +#define VIRTIO_ID_RDMA         26 /* RDMA */
>   
>   #endif /* _LINUX_VIRTIO_IDS_H */

On Thu, Apr 11, 2019 at 07:02:15PM +0200, Cornelia Huck
wrote:
> On Thu, 11 Apr 2019 14:01:54 +0300
> Yuval Shaia <yuval.shaia at oracle.com> wrote:
> 
> > Data center backends use more and more RDMA or RoCE devices and more
and
> > more software runs in virtualized environment.
> > There is a need for a standard to enable RDMA/RoCE on Virtual
Machines.
> > 
> > Virtio is the optimal solution since is the de-facto
para-virtualizaton
> > technology and also because the Virtio specification
> > allows Hardware Vendors to support Virtio protocol natively in order
to
> > achieve bare metal performance.
> > 
> > This RFC is an effort to addresses challenges in defining the
RDMA/RoCE
> > Virtio Specification and a look forward on possible implementation
> > techniques.
> > 
> > Open issues/Todo list:
> > List is huge, this is only start point of the project.
> > Anyway, here is one example of item in the list:
> > - Multi VirtQ: Every QP has two rings and every CQ has one. This means
that
> >   in order to support for example 32K QPs we will need 64K VirtQ. Not
sure
> >   that this is reasonable so one option is to have one for all and
> >   multiplex the traffic on it. This is not good approach as by design
it
> >   introducing an optional starvation. Another approach would be multi
> >   queues and round-robin (for example) between them.
> > 
> > Expectations from this posting:
> > In general, any comment is welcome, starting from hey, drop this as it
is a
> > very bad idea, to yeah, go ahead, we really want it.
> > Idea here is that since it is not a minor effort i first want to know
if
> > there is some sort interest in the community for such device.
> 
> My first reaction is: Sounds sensible, but it would be good to have a
> spec for this :)
> 
> You'll need a spec if you want this to go forward anyway, so at least a
> sketch would be good to answer questions such as how many virtqueues
> you use for which purpose, what is actually put on the virtqueues,
> whether there are negotiable features, and what the expectations for
> the device and the driver are. It also makes it easier to understand
> how this is supposed to work in practice.
> 
> If folks agree that this sounds useful, the next step would be to
> reserve an id for the device type.
Thanks for the tips, will sure do that, it is that first i wanted to make
sure there is a use case here.

Waiting for any feedback from the community.
> 
> > 
> > The scope of the implementation is limited to probing the device and
doing
> > some basic ibverbs commands. Data-path is not yet implemented. So with
this
> > one can expect only that driver is (partialy) loaded and basic queries
and
> > resource allocation is done.
> > 
> > One note regarding the patchset.
> > I know it is not standard to collaps patches from several repos as i
did
> > here (qemu and linux) but decided to do it anyway so the whole picture
can
> > be seen.
> > 
> > patch 1: virtio-net: Move some virtio-net-pci decl to
include/hw/virtio
> > 	This is a prelimenary patch just as a hack so i will not need to
> > 	impelement new netdev
> > patch 2: hw/virtio-rdma: VirtIO rdma device
> > 	The implementation of the device
> > patch 3: RDMA/virtio-rdma: VirtIO rdma driver
> > 	The device driver
> > 
>

On 4/11/19 4:01 AM, Yuval Shaia wrote:
> +++ b/drivers/infiniband/hw/virtio/Kconfig
> @@ -0,0 +1,6 @@
> +config INFINIBAND_VIRTIO_RDMA
> +	tristate "VirtIO Paravirtualized RDMA Driver"
> +	depends on NETDEVICES && ETHERNET && PCI && INET
> +	---help---
> +	  This driver provides low-level support for VirtIO Paravirtual
> +	  RDMA adapter.
Does this driver really depend on Ethernet, or does it also work with
Ethernet support disabled?
> +static inline struct virtio_rdma_info *to_vdev(struct ib_device *ibdev)
> +{
> +	return container_of(ibdev, struct virtio_rdma_info, ib_dev);
> +}
Is it really worth to introduce this function? Have you considered to
use container_of(ibdev, struct virtio_rdma_info, ib_dev) directly instead
of to_vdev()?
> +static void rdma_ctrl_ack(struct virtqueue *vq)
> +{
> +	struct virtio_rdma_info *dev = vq->vdev->priv;
> +
> +	wake_up(&dev->acked);
> +
> +	printk("%s\n", __func__);
> +}
Should that printk() be changed into pr_debug()? The same comment holds for
all other printk() calls.
> +#define VIRTIO_RDMA_BOARD_ID	1
> +#define VIRTIO_RDMA_HW_NAME	"virtio-rdma"
> +#define VIRTIO_RDMA_HW_REV	1
> +#define VIRTIO_RDMA_DRIVER_VER	"1.0"
Is a driver version number useful in an upstream driver?
> +struct ib_cq *virtio_rdma_create_cq(struct ib_device *ibdev,
> +				    const struct ib_cq_init_attr *attr,
> +				    struct ib_ucontext *context,
> +				    struct ib_udata *udata)
> +{
> +	struct scatterlist in, out;
> +	struct virtio_rdma_ib_cq *vcq;
> +	struct cmd_create_cq *cmd;
> +	struct rsp_create_cq *rsp;
> +	struct ib_cq *cq = NULL;
> +	int rc;
> +
> +	/* TODO: Check MAX_CQ */
> +
> +	cmd = kmalloc(sizeof(*cmd), GFP_ATOMIC);
> +	if (!cmd)
> +		return ERR_PTR(-ENOMEM);
> +
> +	rsp = kmalloc(sizeof(*rsp), GFP_ATOMIC);
> +	if (!rsp) {
> +		kfree(cmd);
> +		return ERR_PTR(-ENOMEM);
> +	}
> +
> +	vcq = kzalloc(sizeof(*vcq), GFP_KERNEL);
> +	if (!vcq)
> +		goto out;
Are you sure that you want to mix GFP_ATOMIC and GFP_KERNEL in a single
function?

Thanks,

Bart.

On Thu, Apr 11, 2019 at 02:01:56PM +0300, Yuval Shaia
wrote:
> diff --git a/include/standard-headers/linux/virtio_ids.h
b/include/standard-headers/linux/virtio_ids.h
> index 6d5c3b2d4f..bd2c699450 100644
> --- a/include/standard-headers/linux/virtio_ids.h
> +++ b/include/standard-headers/linux/virtio_ids.h
> @@ -43,5 +43,6 @@
>  #define VIRTIO_ID_INPUT        18 /* virtio input */
>  #define VIRTIO_ID_VSOCK        19 /* virtio vsock transport */
>  #define VIRTIO_ID_CRYPTO       20 /* virtio crypto */
> +#define VIRTIO_ID_RDMA         26 /* virtio crypto */
>  
>  #endif /* _LINUX_VIRTIO_IDS_H */

Please reserve an ID with the virtio TC.
Process:
- create a github issue
- post patch with Fixes: tag
- wait a bit
- if no comments on either ask for a vote
> -- 
> 2.20.1