IB/usnic: Add Cisco VIC low-level hardware driver
authorUpinder Malhi <umalhi@cisco.com>
Tue, 10 Sep 2013 03:38:16 +0000 (03:38 +0000)
committerRoland Dreier <roland@purestorage.com>
Tue, 14 Jan 2014 08:44:28 +0000 (00:44 -0800)
This adds a driver that allows userspace to use UD-like QPs over a
proprietary Cisco transport with Cisco's Virtual Interface Cards (VICs),
including VIC 1240 and 1280 cards.

Signed-off-by: Upinder Malhi <umalhi@cisco.com>
Signed-off-by: Christian Benvenuti <benve@cisco.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
30 files changed:
MAINTAINERS
drivers/infiniband/Kconfig
drivers/infiniband/Makefile
drivers/infiniband/hw/usnic/Kconfig [new file with mode: 0644]
drivers/infiniband/hw/usnic/Makefile [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_abi.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_common_pkt_hdr.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_common_util.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_debugfs.c [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_debugfs.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_fwd.c [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_fwd.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_ib.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_ib_main.c [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_ib_qp_grp.c [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_ib_qp_grp.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_ib_sysfs.c [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_ib_sysfs.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_ib_verbs.c [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_ib_verbs.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_log.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_transport.c [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_transport.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_uiom.c [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_uiom.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_uiom_interval_tree.c [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_uiom_interval_tree.h [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_vnic.c [new file with mode: 0644]
drivers/infiniband/hw/usnic/usnic_vnic.h [new file with mode: 0644]

index 31a046213274e06e63e26da0514ede7ebc7c4443..9bc170ee9a2a6bdac3c23be638a75bb338aa85e1 100644 (file)
@@ -2158,6 +2158,11 @@ M:       Nishank Trivedi <nistrive@cisco.com>
 S:     Supported
 F:     drivers/net/ethernet/cisco/enic/
 
+CISCO VIC LOW LATENCY NIC DRIVER
+M:      Upinder Malhi <umalhi@cisco.com>
+S:      Supported
+F:      drivers/infiniband/hw/usnic
+
 CIRRUS LOGIC EP93XX ETHERNET DRIVER
 M:     Hartley Sweeten <hsweeten@visionengravers.com>
 L:     netdev@vger.kernel.org
index 5ceda710f516bc4721e14961a504fa7e2c3054a9..10219ee92191bce2c7e8828f8b61dc5c419d0c35 100644 (file)
@@ -53,6 +53,7 @@ source "drivers/infiniband/hw/mlx4/Kconfig"
 source "drivers/infiniband/hw/mlx5/Kconfig"
 source "drivers/infiniband/hw/nes/Kconfig"
 source "drivers/infiniband/hw/ocrdma/Kconfig"
+source "drivers/infiniband/hw/usnic/Kconfig"
 
 source "drivers/infiniband/ulp/ipoib/Kconfig"
 
index 1fe69888515f32cdd5f20f20b9a1d9d5f79876b6..bf508b5550c452f32997a888badb4c885900cec9 100644 (file)
@@ -10,6 +10,7 @@ obj-$(CONFIG_MLX4_INFINIBAND)         += hw/mlx4/
 obj-$(CONFIG_MLX5_INFINIBAND)          += hw/mlx5/
 obj-$(CONFIG_INFINIBAND_NES)           += hw/nes/
 obj-$(CONFIG_INFINIBAND_OCRDMA)                += hw/ocrdma/
+obj-$(CONFIG_INFINIBAND_USNIC)         += hw/usnic/
 obj-$(CONFIG_INFINIBAND_IPOIB)         += ulp/ipoib/
 obj-$(CONFIG_INFINIBAND_SRP)           += ulp/srp/
 obj-$(CONFIG_INFINIBAND_SRPT)          += ulp/srpt/
diff --git a/drivers/infiniband/hw/usnic/Kconfig b/drivers/infiniband/hw/usnic/Kconfig
new file mode 100644 (file)
index 0000000..2cc8ba0
--- /dev/null
@@ -0,0 +1,10 @@
+config INFINIBAND_USNIC
+       tristate "Verbs support for Cisco VIC"
+       depends on NETDEVICES && ETHERNET && PCI && INTEL_IOMMU
+       select ENIC
+       select NET_VENDOR_CISCO
+       select PCI_IOV
+       select INFINIBAND_USER_ACCESS
+       ---help---
+         This is a low-level driver for Cisco's Virtual Interface
+         Cards (VICs), including the VIC 1240 and 1280 cards.
diff --git a/drivers/infiniband/hw/usnic/Makefile b/drivers/infiniband/hw/usnic/Makefile
new file mode 100644 (file)
index 0000000..99fb2db
--- /dev/null
@@ -0,0 +1,15 @@
+ccflags-y := -Idrivers/net/ethernet/cisco/enic
+
+obj-$(CONFIG_INFINIBAND_USNIC)+= usnic_verbs.o
+
+usnic_verbs-y=\
+usnic_fwd.o \
+usnic_transport.o \
+usnic_uiom.o \
+usnic_uiom_interval_tree.o \
+usnic_vnic.o \
+usnic_ib_main.o \
+usnic_ib_qp_grp.o \
+usnic_ib_sysfs.o \
+usnic_ib_verbs.o \
+usnic_debugfs.o \
diff --git a/drivers/infiniband/hw/usnic/usnic.h b/drivers/infiniband/hw/usnic/usnic.h
new file mode 100644 (file)
index 0000000..d741c76
--- /dev/null
@@ -0,0 +1,29 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_H_
+#define USNIC_H_
+
+#define DRV_NAME       "usnic_verbs"
+
+#define PCI_DEVICE_ID_CISCO_VIC_USPACE_NIC     0x00cf  /* User space NIC */
+
+#define DRV_VERSION    "1.0.2"
+#define DRV_RELDATE    "September 09, 2013"
+
+#endif /* USNIC_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_abi.h b/drivers/infiniband/hw/usnic/usnic_abi.h
new file mode 100644 (file)
index 0000000..510b7d7
--- /dev/null
@@ -0,0 +1,56 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+
+#ifndef USNIC_ABI_H
+#define USNIC_ABI_H
+
+/* ABI between userspace and kernel */
+#define USNIC_UVERBS_ABI_VERSION       2
+
+#define USNIC_QP_GRP_MAX_WQS           8
+#define USNIC_QP_GRP_MAX_RQS           8
+#define USNIC_QP_GRP_MAX_CQS           16
+
+enum usnic_transport_type {
+       USNIC_TRANSPORT_UNKNOWN         = 0,
+       USNIC_TRANSPORT_ROCE_CUSTOM     = 1,
+       USNIC_TRANSPORT_MAX             = 2,
+};
+
+/*TODO: Future - usnic_modify_qp needs to pass in generic filters */
+struct usnic_ib_create_qp_resp {
+       u32                             vfid;
+       u32                             qp_grp_id;
+       u64                             bar_bus_addr;
+       u32                             bar_len;
+/*
+ * WQ, RQ, CQ are explicity specified bc exposing a generic resources inteface
+ * expands the scope of ABI to many files.
+ */
+       u32                             wq_cnt;
+       u32                             rq_cnt;
+       u32                             cq_cnt;
+       u32                             wq_idx[USNIC_QP_GRP_MAX_WQS];
+       u32                             rq_idx[USNIC_QP_GRP_MAX_RQS];
+       u32                             cq_idx[USNIC_QP_GRP_MAX_CQS];
+       u32                             transport;
+       u32                             reserved[9];
+};
+
+#endif /* USNIC_ABI_H */
diff --git a/drivers/infiniband/hw/usnic/usnic_common_pkt_hdr.h b/drivers/infiniband/hw/usnic/usnic_common_pkt_hdr.h
new file mode 100644 (file)
index 0000000..3935672
--- /dev/null
@@ -0,0 +1,27 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_CMN_PKT_HDR_H
+#define USNIC_CMN_PKT_HDR_H
+
+#define USNIC_ROCE_ETHERTYPE           (0x8915)
+#define USNIC_ROCE_GRH_VER              (8)
+#define USNIC_PROTO_VER                 (1)
+#define USNIC_ROCE_GRH_VER_SHIFT        (4)
+
+#endif /* USNIC_COMMON_PKT_HDR_H */
diff --git a/drivers/infiniband/hw/usnic/usnic_common_util.h b/drivers/infiniband/hw/usnic/usnic_common_util.h
new file mode 100644 (file)
index 0000000..128550a
--- /dev/null
@@ -0,0 +1,51 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_CMN_UTIL_H
+#define USNIC_CMN_UTIL_H
+
+static inline void
+usnic_mac_to_gid(const char *const mac, char *raw_gid)
+{
+       raw_gid[0] = 0xfe;
+       raw_gid[1] = 0x80;
+       memset(&raw_gid[2], 0, 6);
+       raw_gid[8] = mac[0]^2;
+       raw_gid[9] = mac[1];
+       raw_gid[10] = mac[2];
+       raw_gid[11] = 0xff;
+       raw_gid[12] = 0xfe;
+       raw_gid[13] = mac[3];
+       raw_gid[14] = mac[4];
+       raw_gid[15] = mac[5];
+}
+
+static inline void
+usnic_write_gid_if_id_from_mac(char *mac, char *raw_gid)
+{
+       raw_gid[8] = mac[0]^2;
+       raw_gid[9] = mac[1];
+       raw_gid[10] = mac[2];
+       raw_gid[11] = 0xff;
+       raw_gid[12] = 0xfe;
+       raw_gid[13] = mac[3];
+       raw_gid[14] = mac[4];
+       raw_gid[15] = mac[5];
+}
+
+#endif /* USNIC_COMMON_UTIL_H */
diff --git a/drivers/infiniband/hw/usnic/usnic_debugfs.c b/drivers/infiniband/hw/usnic/usnic_debugfs.c
new file mode 100644 (file)
index 0000000..91386df
--- /dev/null
@@ -0,0 +1,71 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#include <linux/debugfs.h>
+#include <linux/module.h>
+
+#include "usnic.h"
+#include "usnic_log.h"
+#include "usnic_debugfs.h"
+
+static struct dentry *debugfs_root;
+
+static ssize_t usnic_debugfs_buildinfo_read(struct file *f, char __user *data,
+                                               size_t count, loff_t *ppos)
+{
+       char buf[500];
+       int res;
+
+       if (*ppos > 0)
+               return 0;
+
+       res = scnprintf(buf, sizeof(buf),
+                       "version:       %s\n"
+                       "build date:    %s\n",
+                       DRV_VERSION, DRV_RELDATE);
+
+       return simple_read_from_buffer(data, count, ppos, buf, res);
+}
+
+static const struct file_operations usnic_debugfs_buildinfo_ops = {
+       .owner = THIS_MODULE,
+       .open = simple_open,
+       .read = usnic_debugfs_buildinfo_read
+};
+
+void usnic_debugfs_init(void)
+{
+       debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
+       if (IS_ERR(debugfs_root)) {
+               usnic_err("Failed to create debugfs root dir, check if debugfs is enabled in kernel configuration\n");
+               debugfs_root = NULL;
+               return;
+       }
+
+       debugfs_create_file("build-info", S_IRUGO, debugfs_root,
+                               NULL, &usnic_debugfs_buildinfo_ops);
+}
+
+void usnic_debugfs_exit(void)
+{
+       if (!debugfs_root)
+               return;
+
+       debugfs_remove_recursive(debugfs_root);
+       debugfs_root = NULL;
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_debugfs.h b/drivers/infiniband/hw/usnic/usnic_debugfs.h
new file mode 100644 (file)
index 0000000..914a330
--- /dev/null
@@ -0,0 +1,25 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#ifndef USNIC_DEBUGFS_H_
+#define USNIC_DEBUGFS_H_
+
+void usnic_debugfs_init(void);
+
+void usnic_debugfs_exit(void);
+
+#endif /*!USNIC_DEBUGFS_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_fwd.c b/drivers/infiniband/hw/usnic/usnic_fwd.c
new file mode 100644 (file)
index 0000000..8e42216
--- /dev/null
@@ -0,0 +1,243 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/netdevice.h>
+#include <linux/pci.h>
+
+#include "enic_api.h"
+#include "usnic_common_pkt_hdr.h"
+#include "usnic_fwd.h"
+#include "usnic_log.h"
+
+struct usnic_fwd_dev *usnic_fwd_dev_alloc(struct pci_dev *pdev)
+{
+       struct usnic_fwd_dev *ufdev;
+
+       ufdev = kzalloc(sizeof(*ufdev), GFP_KERNEL);
+       if (!ufdev)
+               return NULL;
+
+       ufdev->pdev = pdev;
+       ufdev->netdev = pci_get_drvdata(pdev);
+       spin_lock_init(&ufdev->lock);
+
+       return ufdev;
+}
+
+void usnic_fwd_dev_free(struct usnic_fwd_dev *ufdev)
+{
+       kfree(ufdev);
+}
+
+static int usnic_fwd_devcmd(struct usnic_fwd_dev *ufdev, int vnic_idx,
+                               enum vnic_devcmd_cmd cmd, u64 *a0, u64 *a1)
+{
+       int status;
+       struct net_device *netdev = ufdev->netdev;
+
+       spin_lock(&ufdev->lock);
+       status = enic_api_devcmd_proxy_by_index(netdev,
+                       vnic_idx,
+                       cmd,
+                       a0, a1,
+                       1000);
+       spin_unlock(&ufdev->lock);
+       if (status) {
+               if (status == ERR_EINVAL && cmd == CMD_DEL_FILTER) {
+                       usnic_dbg("Dev %s vnic idx %u cmd %u already deleted",
+                                       netdev_name(netdev), vnic_idx, cmd);
+               } else {
+                       usnic_err("Dev %s vnic idx %u cmd %u failed with status %d\n",
+                                       netdev_name(netdev), vnic_idx, cmd,
+                                       status);
+               }
+       } else {
+               usnic_dbg("Dev %s vnic idx %u cmd %u success",
+                               netdev_name(netdev), vnic_idx,
+                               cmd);
+       }
+
+       return status;
+}
+
+int usnic_fwd_add_usnic_filter(struct usnic_fwd_dev *ufdev, int vnic_idx,
+                               int rq_idx, struct usnic_fwd_filter *fwd_filter,
+                               struct usnic_fwd_filter_hndl **filter_hndl)
+{
+       struct filter_tlv *tlv, *tlv_va;
+       struct filter *filter;
+       struct filter_action *action;
+       struct pci_dev *pdev;
+       struct usnic_fwd_filter_hndl *usnic_filter_hndl;
+       int status;
+       u64 a0, a1;
+       u64 tlv_size;
+       dma_addr_t tlv_pa;
+
+       pdev = ufdev->pdev;
+       tlv_size = (2*sizeof(struct filter_tlv) +
+               sizeof(struct filter) +
+               sizeof(struct filter_action));
+       tlv = pci_alloc_consistent(pdev, tlv_size, &tlv_pa);
+       if (!tlv) {
+               usnic_err("Failed to allocate memory\n");
+               return -ENOMEM;
+       }
+
+       usnic_filter_hndl = kzalloc(sizeof(*usnic_filter_hndl), GFP_ATOMIC);
+       if (!usnic_filter_hndl) {
+               usnic_err("Failed to allocate memory for hndl\n");
+               pci_free_consistent(pdev, tlv_size, tlv, tlv_pa);
+               return -ENOMEM;
+       }
+
+       tlv_va = tlv;
+       a0 = tlv_pa;
+       a1 = tlv_size;
+       memset(tlv, 0, tlv_size);
+       tlv->type = CLSF_TLV_FILTER;
+       tlv->length = sizeof(struct filter);
+       filter = (struct filter *)&tlv->val;
+       filter->type = FILTER_USNIC_ID;
+       filter->u.usnic.ethtype = USNIC_ROCE_ETHERTYPE;
+       filter->u.usnic.flags = FILTER_FIELD_USNIC_ETHTYPE |
+                                       FILTER_FIELD_USNIC_ID |
+                                       FILTER_FIELD_USNIC_PROTO;
+       filter->u.usnic.proto_version = (USNIC_ROCE_GRH_VER <<
+                                               USNIC_ROCE_GRH_VER_SHIFT)
+                                                       | USNIC_PROTO_VER;
+       filter->u.usnic.usnic_id = fwd_filter->port_num;
+       tlv = (struct filter_tlv *)((char *)tlv + sizeof(struct filter_tlv) +
+                       sizeof(struct filter));
+       tlv->type = CLSF_TLV_ACTION;
+       tlv->length = sizeof(struct filter_action);
+       action = (struct filter_action *)&tlv->val;
+       action->type = FILTER_ACTION_RQ_STEERING;
+       action->u.rq_idx = rq_idx;
+
+       status = usnic_fwd_devcmd(ufdev, vnic_idx, CMD_ADD_FILTER, &a0, &a1);
+       pci_free_consistent(pdev, tlv_size, tlv_va, tlv_pa);
+       if (status) {
+               usnic_err("VF %s Filter add failed with status:%d",
+                               pci_name(pdev),
+                               status);
+               kfree(usnic_filter_hndl);
+               return status;
+       } else {
+               usnic_dbg("VF %s FILTER ID:%u",
+                               pci_name(pdev),
+                               (u32)a0);
+       }
+
+       usnic_filter_hndl->type = FILTER_USNIC_ID;
+       usnic_filter_hndl->id = (u32)a0;
+       usnic_filter_hndl->vnic_idx = vnic_idx;
+       usnic_filter_hndl->ufdev = ufdev;
+       usnic_filter_hndl->filter = fwd_filter;
+       *filter_hndl = usnic_filter_hndl;
+
+       return status;
+}
+
+int usnic_fwd_del_filter(struct usnic_fwd_filter_hndl *filter_hndl)
+{
+       int status;
+       u64 a0, a1;
+       struct net_device *netdev;
+
+       netdev = filter_hndl->ufdev->netdev;
+       a0 = filter_hndl->id;
+
+       status = usnic_fwd_devcmd(filter_hndl->ufdev, filter_hndl->vnic_idx,
+                                       CMD_DEL_FILTER, &a0, &a1);
+       if (status) {
+               if (status == ERR_EINVAL) {
+                       usnic_dbg("Filter %u already deleted for VF Idx %u pf: %s status: %d",
+                                       filter_hndl->id, filter_hndl->vnic_idx,
+                                       netdev_name(netdev), status);
+                       status = 0;
+                       kfree(filter_hndl);
+               } else {
+                       usnic_err("PF %s VF Idx %u Filter: %u FILTER DELETE failed with status %d",
+                                       netdev_name(netdev),
+                                       filter_hndl->vnic_idx, filter_hndl->id,
+                                       status);
+               }
+       } else {
+               usnic_dbg("PF %s VF Idx %u Filter: %u FILTER DELETED",
+                               netdev_name(netdev), filter_hndl->vnic_idx,
+                               filter_hndl->id);
+               kfree(filter_hndl);
+       }
+
+       return status;
+}
+
+int usnic_fwd_enable_rq(struct usnic_fwd_dev *ufdev, int vnic_idx, int rq_idx)
+{
+       int status;
+       struct net_device *pf_netdev;
+       u64 a0, a1;
+
+       pf_netdev = ufdev->netdev;
+       a0 = rq_idx;
+       a1 = CMD_QP_RQWQ;
+
+       status = usnic_fwd_devcmd(ufdev, vnic_idx, CMD_QP_ENABLE, &a0, &a1);
+
+       if (status) {
+               usnic_err("PF %s VNIC Index %u RQ Index: %u ENABLE Failed with status %d",
+                               netdev_name(pf_netdev),
+                               vnic_idx,
+                               rq_idx,
+                               status);
+       } else {
+               usnic_dbg("PF %s VNIC Index %u RQ Index: %u ENABLED",
+                               netdev_name(pf_netdev),
+                               vnic_idx, rq_idx);
+       }
+
+       return status;
+}
+
+int usnic_fwd_disable_rq(struct usnic_fwd_dev *ufdev, int vnic_idx, int rq_idx)
+{
+       int status;
+       u64 a0, a1;
+       struct net_device *pf_netdev;
+
+       pf_netdev = ufdev->netdev;
+       a0 = rq_idx;
+       a1 = CMD_QP_RQWQ;
+
+       status = usnic_fwd_devcmd(ufdev, vnic_idx, CMD_QP_DISABLE, &a0, &a1);
+
+       if (status) {
+               usnic_err("PF %s VNIC Index %u RQ Index: %u DISABLE Failed with status %d",
+                               netdev_name(pf_netdev),
+                               vnic_idx,
+                               rq_idx,
+                               status);
+       } else {
+               usnic_dbg("PF %s VNIC Index %u RQ Index: %u DISABLED",
+                               netdev_name(pf_netdev),
+                               vnic_idx,
+                               rq_idx);
+       }
+
+       return status;
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_fwd.h b/drivers/infiniband/hw/usnic/usnic_fwd.h
new file mode 100644 (file)
index 0000000..6973901
--- /dev/null
@@ -0,0 +1,58 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_FWD_H_
+#define USNIC_FWD_H_
+
+#include <linux/if.h>
+#include <linux/pci.h>
+#include <linux/spinlock.h>
+
+#include "usnic_abi.h"
+#include "vnic_devcmd.h"
+
+struct usnic_fwd_dev {
+       struct pci_dev                  *pdev;
+       struct net_device               *netdev;
+       spinlock_t                      lock;
+};
+
+struct usnic_fwd_filter {
+       enum usnic_transport_type       transport;
+       u16                             port_num;
+};
+
+struct usnic_fwd_filter_hndl {
+       enum filter_type                type;
+       u32                             id;
+       u32                             vnic_idx;
+       struct usnic_fwd_dev            *ufdev;
+       struct list_head                link;
+       struct usnic_fwd_filter         *filter;
+};
+
+struct usnic_fwd_dev *usnic_fwd_dev_alloc(struct pci_dev *pdev);
+void usnic_fwd_dev_free(struct usnic_fwd_dev *ufdev);
+int usnic_fwd_add_usnic_filter(struct usnic_fwd_dev *ufdev, int vnic_idx,
+                               int rq_idx, struct usnic_fwd_filter *filter,
+                               struct usnic_fwd_filter_hndl **filter_hndl);
+int usnic_fwd_del_filter(struct usnic_fwd_filter_hndl *filter_hndl);
+int usnic_fwd_enable_rq(struct usnic_fwd_dev *ufdev, int vnic_idx, int rq_idx);
+int usnic_fwd_disable_rq(struct usnic_fwd_dev *ufdev, int vnic_idx, int rq_idx);
+
+#endif /* !USNIC_FWD_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_ib.h b/drivers/infiniband/hw/usnic/usnic_ib.h
new file mode 100644 (file)
index 0000000..3511c85
--- /dev/null
@@ -0,0 +1,115 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_IB_H_
+#define USNIC_IB_H_
+
+#include <linux/iommu.h>
+#include <linux/netdevice.h>
+#include <linux/version.h>
+
+#include <rdma/ib_verbs.h>
+
+
+#include "usnic.h"
+#include "usnic_abi.h"
+#include "usnic_vnic.h"
+
+#define USNIC_IB_PORT_CNT              1
+#define USNIC_IB_NUM_COMP_VECTORS      1
+
+extern unsigned int usnic_ib_share_vf;
+
+struct usnic_ib_ucontext {
+       struct ib_ucontext              ibucontext;
+       /* Protected by usnic_ib_dev->usdev_lock */
+       struct list_head                qp_grp_list;
+       struct list_head                link;
+};
+
+struct usnic_ib_pd {
+       struct ib_pd                    ibpd;
+       struct usnic_uiom_pd            *umem_pd;
+};
+
+struct usnic_ib_mr {
+       struct ib_mr                    ibmr;
+       struct usnic_uiom_reg           *umem;
+};
+
+struct usnic_ib_dev {
+       struct ib_device                ib_dev;
+       struct pci_dev                  *pdev;
+       struct net_device               *netdev;
+       struct usnic_fwd_dev            *ufdev;
+       bool                            link_up;
+       struct list_head                ib_dev_link;
+       struct list_head                vf_dev_list;
+       struct list_head                ctx_list;
+       struct mutex                    usdev_lock;
+       char                            mac[ETH_ALEN];
+       unsigned int                    mtu;
+
+       /* provisioning information */
+       struct kref                     vf_cnt;
+       unsigned int                    vf_res_cnt[USNIC_VNIC_RES_TYPE_MAX];
+
+       /* sysfs vars for QPN reporting */
+       struct kobject *qpn_kobj;
+};
+
+struct usnic_ib_vf {
+       struct usnic_ib_dev             *pf;
+       spinlock_t                      lock;
+       struct usnic_vnic               *vnic;
+       unsigned int                    qp_grp_ref_cnt;
+       struct usnic_ib_pd              *pd;
+       struct list_head                link;
+};
+
+static inline
+struct usnic_ib_dev *to_usdev(struct ib_device *ibdev)
+{
+       return container_of(ibdev, struct usnic_ib_dev, ib_dev);
+}
+
+static inline
+struct usnic_ib_ucontext *to_ucontext(struct ib_ucontext *ibucontext)
+{
+       return container_of(ibucontext, struct usnic_ib_ucontext, ibucontext);
+}
+
+static inline
+struct usnic_ib_pd *to_upd(struct ib_pd *ibpd)
+{
+       return container_of(ibpd, struct usnic_ib_pd, ibpd);
+}
+
+static inline
+struct usnic_ib_ucontext *to_uucontext(struct ib_ucontext *ibucontext)
+{
+       return container_of(ibucontext, struct usnic_ib_ucontext, ibucontext);
+}
+
+static inline
+struct usnic_ib_mr *to_umr(struct ib_mr *ibmr)
+{
+       return container_of(ibmr, struct usnic_ib_mr, ibmr);
+}
+void usnic_ib_log_vf(struct usnic_ib_vf *vf);
+#endif /* USNIC_IB_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_main.c b/drivers/infiniband/hw/usnic/usnic_ib_main.c
new file mode 100644 (file)
index 0000000..dc09c12
--- /dev/null
@@ -0,0 +1,598 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * Author: Upinder Malhi <umalhi@cisco.com>
+ * Author: Anant Deepak <anadeepa@cisco.com>
+ * Author: Cesare Cantu' <cantuc@cisco.com>
+ * Author: Jeff Squyres <jsquyres@cisco.com>
+ * Author: Kiran Thirumalai <kithirum@cisco.com>
+ * Author: Xuyang Wang <xuywang@cisco.com>
+ * Author: Reese Faucette <rfaucett@cisco.com>
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/errno.h>
+#include <linux/pci.h>
+#include <linux/netdevice.h>
+
+#include <rdma/ib_user_verbs.h>
+#include <rdma/ib_addr.h>
+
+#include "usnic_abi.h"
+#include "usnic_common_util.h"
+#include "usnic_ib.h"
+#include "usnic_ib_qp_grp.h"
+#include "usnic_log.h"
+#include "usnic_fwd.h"
+#include "usnic_debugfs.h"
+#include "usnic_ib_verbs.h"
+#include "usnic_transport.h"
+#include "usnic_uiom.h"
+#include "usnic_ib_sysfs.h"
+
+unsigned int usnic_log_lvl = USNIC_LOG_LVL_ERR;
+unsigned int usnic_ib_share_vf = 1;
+
+static const char usnic_version[] =
+       DRV_NAME ": Cisco VIC (USNIC) Verbs Driver v"
+       DRV_VERSION " (" DRV_RELDATE ")\n";
+
+static DEFINE_MUTEX(usnic_ib_ibdev_list_lock);
+static LIST_HEAD(usnic_ib_ibdev_list);
+
+/* Callback dump funcs */
+static int usnic_ib_dump_vf_hdr(void *obj, char *buf, int buf_sz)
+{
+       struct usnic_ib_vf *vf = obj;
+       return scnprintf(buf, buf_sz, "PF: %s ", vf->pf->ib_dev.name);
+}
+/* End callback dump funcs */
+
+static void usnic_ib_dump_vf(struct usnic_ib_vf *vf, char *buf, int buf_sz)
+{
+       usnic_vnic_dump(vf->vnic, buf, buf_sz, vf,
+                       usnic_ib_dump_vf_hdr,
+                       usnic_ib_qp_grp_dump_hdr, usnic_ib_qp_grp_dump_rows);
+}
+
+void usnic_ib_log_vf(struct usnic_ib_vf *vf)
+{
+       char buf[1000];
+       usnic_ib_dump_vf(vf, buf, sizeof(buf));
+       usnic_dbg("%s\n", buf);
+}
+
+/* Start of netdev section */
+static inline const char *usnic_ib_netdev_event_to_string(unsigned long event)
+{
+       const char *event2str[] = {"NETDEV_NONE", "NETDEV_UP", "NETDEV_DOWN",
+               "NETDEV_REBOOT", "NETDEV_CHANGE",
+               "NETDEV_REGISTER", "NETDEV_UNREGISTER", "NETDEV_CHANGEMTU",
+               "NETDEV_CHANGEADDR", "NETDEV_GOING_DOWN", "NETDEV_FEAT_CHANGE",
+               "NETDEV_BONDING_FAILOVER", "NETDEV_PRE_UP",
+               "NETDEV_PRE_TYPE_CHANGE", "NETDEV_POST_TYPE_CHANGE",
+               "NETDEV_POST_INT", "NETDEV_UNREGISTER_FINAL", "NETDEV_RELEASE",
+               "NETDEV_NOTIFY_PEERS", "NETDEV_JOIN"
+       };
+
+       if (event >= ARRAY_SIZE(event2str))
+               return "UNKNOWN_NETDEV_EVENT";
+       else
+               return event2str[event];
+}
+
+static void usnic_ib_qp_grp_modify_active_to_err(struct usnic_ib_dev *us_ibdev)
+{
+       struct usnic_ib_ucontext *ctx;
+       struct usnic_ib_qp_grp *qp_grp;
+       enum ib_qp_state cur_state;
+       int status;
+
+       BUG_ON(!mutex_is_locked(&us_ibdev->usdev_lock));
+
+       list_for_each_entry(ctx, &us_ibdev->ctx_list, link) {
+               list_for_each_entry(qp_grp, &ctx->qp_grp_list, link) {
+                       cur_state = qp_grp->state;
+                       if (cur_state == IB_QPS_INIT ||
+                               cur_state == IB_QPS_RTR ||
+                               cur_state == IB_QPS_RTS) {
+                               status = usnic_ib_qp_grp_modify(qp_grp,
+                                                               IB_QPS_ERR,
+                                                               NULL);
+                               if (status) {
+                                       usnic_err("Failed to transistion qp grp %u from %s to %s\n",
+                                               qp_grp->grp_id,
+                                               usnic_ib_qp_grp_state_to_string
+                                               (cur_state),
+                                               usnic_ib_qp_grp_state_to_string
+                                               (IB_QPS_ERR));
+                               }
+                       }
+               }
+       }
+}
+
+static void usnic_ib_handle_usdev_event(struct usnic_ib_dev *us_ibdev,
+                                       unsigned long event)
+{
+       struct net_device *netdev;
+       struct ib_event ib_event;
+
+       memset(&ib_event, 0, sizeof(ib_event));
+
+       mutex_lock(&us_ibdev->usdev_lock);
+       netdev = us_ibdev->netdev;
+       switch (event) {
+       case NETDEV_REBOOT:
+               usnic_info("PF Reset on %s\n", us_ibdev->ib_dev.name);
+               usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
+               ib_event.event = IB_EVENT_PORT_ERR;
+               ib_event.device = &us_ibdev->ib_dev;
+               ib_event.element.port_num = 1;
+               ib_dispatch_event(&ib_event);
+               break;
+       case NETDEV_UP:
+       case NETDEV_DOWN:
+       case NETDEV_CHANGE:
+               if (!us_ibdev->link_up && netif_carrier_ok(netdev)) {
+                       us_ibdev->link_up = true;
+                       usnic_info("Link UP on %s\n", us_ibdev->ib_dev.name);
+                       ib_event.event = IB_EVENT_PORT_ACTIVE;
+                       ib_event.device = &us_ibdev->ib_dev;
+                       ib_event.element.port_num = 1;
+                       ib_dispatch_event(&ib_event);
+               } else if (us_ibdev->link_up && !netif_carrier_ok(netdev)) {
+                       us_ibdev->link_up = false;
+                       usnic_info("Link DOWN on %s\n", us_ibdev->ib_dev.name);
+                       usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
+                       ib_event.event = IB_EVENT_PORT_ERR;
+                       ib_event.device = &us_ibdev->ib_dev;
+                       ib_event.element.port_num = 1;
+                       ib_dispatch_event(&ib_event);
+               } else {
+                       usnic_dbg("Ignorning %s on %s\n",
+                                       usnic_ib_netdev_event_to_string(event),
+                                       us_ibdev->ib_dev.name);
+               }
+               break;
+       case NETDEV_CHANGEADDR:
+               if (!memcmp(us_ibdev->mac, netdev->dev_addr,
+                               sizeof(us_ibdev->mac))) {
+                       usnic_dbg("Ignorning addr change on %s\n",
+                                       us_ibdev->ib_dev.name);
+               } else {
+                       usnic_info(" %s old mac: %pM new mac: %pM\n",
+                                       us_ibdev->ib_dev.name,
+                                       us_ibdev->mac,
+                                       netdev->dev_addr);
+                       memcpy(us_ibdev->mac, netdev->dev_addr,
+                               sizeof(us_ibdev->mac));
+                       usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
+                       ib_event.event = IB_EVENT_GID_CHANGE;
+                       ib_event.device = &us_ibdev->ib_dev;
+                       ib_event.element.port_num = 1;
+                       ib_dispatch_event(&ib_event);
+               }
+
+               break;
+       case NETDEV_CHANGEMTU:
+               if (us_ibdev->mtu != netdev->mtu) {
+                       usnic_info("MTU Change on %s old: %u new: %u\n",
+                                       us_ibdev->ib_dev.name,
+                                       us_ibdev->mtu, netdev->mtu);
+                       us_ibdev->mtu = netdev->mtu;
+                       usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
+               } else {
+                       usnic_dbg("Ignoring MTU change on %s\n",
+                                       us_ibdev->ib_dev.name);
+               }
+               break;
+       default:
+               usnic_dbg("Ignorning event %s on %s",
+                               usnic_ib_netdev_event_to_string(event),
+                               us_ibdev->ib_dev.name);
+       }
+       mutex_unlock(&us_ibdev->usdev_lock);
+}
+
+static int usnic_ib_netdevice_event(struct notifier_block *notifier,
+                                       unsigned long event, void *ptr)
+{
+       struct usnic_ib_dev *us_ibdev;
+
+       struct net_device *netdev = netdev_notifier_info_to_dev(ptr);
+
+       mutex_lock(&usnic_ib_ibdev_list_lock);
+       list_for_each_entry(us_ibdev, &usnic_ib_ibdev_list, ib_dev_link) {
+               if (us_ibdev->netdev == netdev) {
+                       usnic_ib_handle_usdev_event(us_ibdev, event);
+                       break;
+               }
+       }
+       mutex_unlock(&usnic_ib_ibdev_list_lock);
+
+       return NOTIFY_DONE;
+}
+
+static struct notifier_block usnic_ib_netdevice_notifier = {
+       .notifier_call = usnic_ib_netdevice_event
+};
+/* End of netdev section */
+
+/* Start of PF discovery section */
+static void *usnic_ib_device_add(struct pci_dev *dev)
+{
+       struct usnic_ib_dev *us_ibdev;
+       union ib_gid gid;
+
+       usnic_dbg("\n");
+
+       us_ibdev = (struct usnic_ib_dev *)ib_alloc_device(sizeof(*us_ibdev));
+       if (IS_ERR_OR_NULL(us_ibdev)) {
+               usnic_err("Device %s context alloc failed\n",
+                               netdev_name(pci_get_drvdata(dev)));
+               return ERR_PTR(us_ibdev ? PTR_ERR(us_ibdev) : -EFAULT);
+       }
+
+       us_ibdev->ufdev = usnic_fwd_dev_alloc(dev);
+       if (IS_ERR_OR_NULL(us_ibdev->ufdev)) {
+               usnic_err("Failed to alloc ufdev for %s with err %ld\n",
+                               pci_name(dev), PTR_ERR(us_ibdev->ufdev));
+               goto err_dealloc;
+       }
+
+       mutex_init(&us_ibdev->usdev_lock);
+       INIT_LIST_HEAD(&us_ibdev->vf_dev_list);
+       INIT_LIST_HEAD(&us_ibdev->ctx_list);
+
+       us_ibdev->pdev = dev;
+       us_ibdev->netdev = pci_get_drvdata(dev);
+       us_ibdev->ib_dev.owner = THIS_MODULE;
+       us_ibdev->ib_dev.node_type = RDMA_NODE_USNIC;
+       us_ibdev->ib_dev.phys_port_cnt = USNIC_IB_PORT_CNT;
+       us_ibdev->ib_dev.num_comp_vectors = USNIC_IB_NUM_COMP_VECTORS;
+       us_ibdev->ib_dev.dma_device = &dev->dev;
+       us_ibdev->ib_dev.uverbs_abi_ver = USNIC_UVERBS_ABI_VERSION;
+       strlcpy(us_ibdev->ib_dev.name, "usnic_%d", IB_DEVICE_NAME_MAX);
+
+       us_ibdev->ib_dev.uverbs_cmd_mask =
+               (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
+               (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
+               (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
+               (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
+               (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
+               (1ull << IB_USER_VERBS_CMD_REG_MR) |
+               (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
+               (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
+               (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
+               (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
+               (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
+               (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
+               (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
+               (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
+               (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
+               (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
+               (1ull << IB_USER_VERBS_CMD_OPEN_QP);
+
+       us_ibdev->ib_dev.query_device = usnic_ib_query_device;
+       us_ibdev->ib_dev.query_port = usnic_ib_query_port;
+       us_ibdev->ib_dev.query_pkey = usnic_ib_query_pkey;
+       us_ibdev->ib_dev.query_gid = usnic_ib_query_gid;
+       us_ibdev->ib_dev.get_link_layer = usnic_ib_port_link_layer;
+       us_ibdev->ib_dev.alloc_pd = usnic_ib_alloc_pd;
+       us_ibdev->ib_dev.dealloc_pd = usnic_ib_dealloc_pd;
+       us_ibdev->ib_dev.create_qp = usnic_ib_create_qp;
+       us_ibdev->ib_dev.modify_qp = usnic_ib_modify_qp;
+       us_ibdev->ib_dev.query_qp = usnic_ib_query_qp;
+       us_ibdev->ib_dev.destroy_qp = usnic_ib_destroy_qp;
+       us_ibdev->ib_dev.create_cq = usnic_ib_create_cq;
+       us_ibdev->ib_dev.destroy_cq = usnic_ib_destroy_cq;
+       us_ibdev->ib_dev.reg_user_mr = usnic_ib_reg_mr;
+       us_ibdev->ib_dev.dereg_mr = usnic_ib_dereg_mr;
+       us_ibdev->ib_dev.alloc_ucontext = usnic_ib_alloc_ucontext;
+       us_ibdev->ib_dev.dealloc_ucontext = usnic_ib_dealloc_ucontext;
+       us_ibdev->ib_dev.mmap = usnic_ib_mmap;
+       us_ibdev->ib_dev.create_ah = usnic_ib_create_ah;
+       us_ibdev->ib_dev.destroy_ah = usnic_ib_destroy_ah;
+       us_ibdev->ib_dev.post_send = usnic_ib_post_send;
+       us_ibdev->ib_dev.post_recv = usnic_ib_post_recv;
+       us_ibdev->ib_dev.poll_cq = usnic_ib_poll_cq;
+       us_ibdev->ib_dev.req_notify_cq = usnic_ib_req_notify_cq;
+       us_ibdev->ib_dev.get_dma_mr = usnic_ib_get_dma_mr;
+
+
+       if (ib_register_device(&us_ibdev->ib_dev, NULL))
+               goto err_fwd_dealloc;
+
+       us_ibdev->link_up = netif_carrier_ok(us_ibdev->netdev);
+       us_ibdev->mtu = us_ibdev->netdev->mtu;
+       memcpy(&us_ibdev->mac, us_ibdev->netdev->dev_addr,
+               sizeof(us_ibdev->mac));
+       usnic_mac_to_gid(us_ibdev->netdev->perm_addr, &gid.raw[0]);
+       memcpy(&us_ibdev->ib_dev.node_guid, &gid.global.interface_id,
+               sizeof(gid.global.interface_id));
+       kref_init(&us_ibdev->vf_cnt);
+
+       usnic_info("Added ibdev: %s netdev: %s with mac %pM Link: %u MTU: %u\n",
+                       us_ibdev->ib_dev.name, netdev_name(us_ibdev->netdev),
+                       us_ibdev->mac, us_ibdev->link_up, us_ibdev->mtu);
+       return us_ibdev;
+
+err_fwd_dealloc:
+       usnic_fwd_dev_free(us_ibdev->ufdev);
+err_dealloc:
+       usnic_err("failed -- deallocing device\n");
+       ib_dealloc_device(&us_ibdev->ib_dev);
+       return NULL;
+}
+
+static void usnic_ib_device_remove(struct usnic_ib_dev *us_ibdev)
+{
+       usnic_info("Unregistering %s\n", us_ibdev->ib_dev.name);
+       usnic_ib_sysfs_unregister_usdev(us_ibdev);
+       usnic_fwd_dev_free(us_ibdev->ufdev);
+       ib_unregister_device(&us_ibdev->ib_dev);
+       ib_dealloc_device(&us_ibdev->ib_dev);
+}
+
+static void usnic_ib_undiscover_pf(struct kref *kref)
+{
+       struct usnic_ib_dev *us_ibdev, *tmp;
+       struct pci_dev *dev;
+       bool found = false;
+
+       dev = container_of(kref, struct usnic_ib_dev, vf_cnt)->pdev;
+       mutex_lock(&usnic_ib_ibdev_list_lock);
+       list_for_each_entry_safe(us_ibdev, tmp,
+                               &usnic_ib_ibdev_list, ib_dev_link) {
+               if (us_ibdev->pdev == dev) {
+                       list_del(&us_ibdev->ib_dev_link);
+                       usnic_ib_device_remove(us_ibdev);
+                       found = true;
+                       break;
+               }
+       }
+
+       WARN(!found, "Failed to remove PF %s\n", pci_name(dev));
+
+       mutex_unlock(&usnic_ib_ibdev_list_lock);
+}
+
+static struct usnic_ib_dev *usnic_ib_discover_pf(struct usnic_vnic *vnic)
+{
+       struct usnic_ib_dev *us_ibdev;
+       struct pci_dev *parent_pci, *vf_pci;
+       int err;
+
+       vf_pci = usnic_vnic_get_pdev(vnic);
+       parent_pci = pci_physfn(vf_pci);
+
+       BUG_ON(!parent_pci);
+
+       mutex_lock(&usnic_ib_ibdev_list_lock);
+       list_for_each_entry(us_ibdev, &usnic_ib_ibdev_list, ib_dev_link) {
+               if (us_ibdev->pdev == parent_pci) {
+                       kref_get(&us_ibdev->vf_cnt);
+                       goto out;
+               }
+       }
+
+       us_ibdev = usnic_ib_device_add(parent_pci);
+       if (IS_ERR_OR_NULL(us_ibdev)) {
+               us_ibdev = (us_ibdev) ? us_ibdev : ERR_PTR(-EFAULT);
+               goto out;
+       }
+
+       err = usnic_ib_sysfs_register_usdev(us_ibdev);
+       if (err) {
+               usnic_ib_device_remove(us_ibdev);
+               us_ibdev = ERR_PTR(err);
+               goto out;
+       }
+
+       list_add(&us_ibdev->ib_dev_link, &usnic_ib_ibdev_list);
+out:
+       mutex_unlock(&usnic_ib_ibdev_list_lock);
+       return us_ibdev;
+}
+/* End of PF discovery section */
+
+/* Start of PCI section */
+
+static DEFINE_PCI_DEVICE_TABLE(usnic_ib_pci_ids) = {
+       {PCI_DEVICE(PCI_VENDOR_ID_CISCO, PCI_DEVICE_ID_CISCO_VIC_USPACE_NIC)},
+       {0,}
+};
+
+static int usnic_ib_pci_probe(struct pci_dev *pdev,
+                               const struct pci_device_id *id)
+{
+       int err;
+       struct usnic_ib_dev *pf;
+       struct usnic_ib_vf *vf;
+       enum usnic_vnic_res_type res_type;
+
+       vf = kzalloc(sizeof(*vf), GFP_KERNEL);
+       if (!vf)
+               return -ENOMEM;
+
+       err = pci_enable_device(pdev);
+       if (err) {
+               usnic_err("Failed to enable %s with err %d\n",
+                               pci_name(pdev), err);
+               goto out_clean_vf;
+       }
+
+       err = pci_request_regions(pdev, DRV_NAME);
+       if (err) {
+               usnic_err("Failed to request region for %s with err %d\n",
+                               pci_name(pdev), err);
+               goto out_disable_device;
+       }
+
+       pci_set_master(pdev);
+       pci_set_drvdata(pdev, vf);
+
+       vf->vnic = usnic_vnic_alloc(pdev);
+       if (IS_ERR_OR_NULL(vf->vnic)) {
+               err = (vf->vnic ? PTR_ERR(vf->vnic) : -ENOMEM);
+               usnic_err("Failed to alloc vnic for %s with err %d\n",
+                               pci_name(pdev), err);
+               goto out_release_regions;
+       }
+
+       pf = usnic_ib_discover_pf(vf->vnic);
+       if (IS_ERR_OR_NULL(pf)) {
+               usnic_err("Failed to discover pf of vnic %s with err%ld\n",
+                               pci_name(pdev), PTR_ERR(pf));
+               err = (pf ? PTR_ERR(pf) : -EFAULT);
+               goto out_clean_vnic;
+       }
+
+       vf->pf = pf;
+       spin_lock_init(&vf->lock);
+       mutex_lock(&pf->usdev_lock);
+       list_add_tail(&vf->link, &pf->vf_dev_list);
+       /*
+        * Save max settings (will be same for each VF, easier to re-write than
+        * to say "if (!set) { set_values(); set=1; }
+        */
+       for (res_type = USNIC_VNIC_RES_TYPE_EOL+1;
+                       res_type < USNIC_VNIC_RES_TYPE_MAX;
+                       res_type++) {
+               pf->vf_res_cnt[res_type] = usnic_vnic_res_cnt(vf->vnic,
+                                                               res_type);
+       }
+
+       mutex_unlock(&pf->usdev_lock);
+
+       usnic_info("Registering usnic VF %s into PF %s\n", pci_name(pdev),
+                       pf->ib_dev.name);
+       usnic_ib_log_vf(vf);
+       return 0;
+
+out_clean_vnic:
+       usnic_vnic_free(vf->vnic);
+out_release_regions:
+       pci_set_drvdata(pdev, NULL);
+       pci_clear_master(pdev);
+       pci_release_regions(pdev);
+out_disable_device:
+       pci_disable_device(pdev);
+out_clean_vf:
+       kfree(vf);
+       return err;
+}
+
+static void usnic_ib_pci_remove(struct pci_dev *pdev)
+{
+       struct usnic_ib_vf *vf = pci_get_drvdata(pdev);
+       struct usnic_ib_dev *pf = vf->pf;
+
+       mutex_lock(&pf->usdev_lock);
+       list_del(&vf->link);
+       mutex_unlock(&pf->usdev_lock);
+
+       kref_put(&pf->vf_cnt, usnic_ib_undiscover_pf);
+       usnic_vnic_free(vf->vnic);
+       pci_set_drvdata(pdev, NULL);
+       pci_clear_master(pdev);
+       pci_release_regions(pdev);
+       pci_disable_device(pdev);
+       kfree(vf);
+
+       usnic_info("Removed VF %s\n", pci_name(pdev));
+}
+
+/* PCI driver entry points */
+static struct pci_driver usnic_ib_pci_driver = {
+       .name = DRV_NAME,
+       .id_table = usnic_ib_pci_ids,
+       .probe = usnic_ib_pci_probe,
+       .remove = usnic_ib_pci_remove,
+};
+/* End of PCI section */
+
+/* Start of module section */
+static int __init usnic_ib_init(void)
+{
+       int err;
+
+       printk_once(KERN_INFO "%s", usnic_version);
+
+       err = usnic_uiom_init(DRV_NAME);
+       if (err) {
+               usnic_err("Unable to initalize umem with err %d\n", err);
+               return err;
+       }
+
+       if (pci_register_driver(&usnic_ib_pci_driver)) {
+               usnic_err("Unable to register with PCI\n");
+               goto out_umem_fini;
+       }
+
+       err = register_netdevice_notifier(&usnic_ib_netdevice_notifier);
+       if (err) {
+               usnic_err("Failed to register netdev notifier\n");
+               goto out_pci_unreg;
+       }
+
+       err = usnic_transport_init();
+       if (err) {
+               usnic_err("Failed to initialize transport\n");
+               goto out_unreg_netdev_notifier;
+       }
+
+       usnic_debugfs_init();
+
+       return 0;
+
+out_unreg_netdev_notifier:
+       unregister_netdevice_notifier(&usnic_ib_netdevice_notifier);
+out_pci_unreg:
+       pci_unregister_driver(&usnic_ib_pci_driver);
+out_umem_fini:
+       usnic_uiom_fini();
+
+       return err;
+}
+
+static void __exit usnic_ib_destroy(void)
+{
+       usnic_dbg("\n");
+       usnic_debugfs_exit();
+       usnic_transport_fini();
+       unregister_netdevice_notifier(&usnic_ib_netdevice_notifier);
+       pci_unregister_driver(&usnic_ib_pci_driver);
+       usnic_uiom_fini();
+}
+
+MODULE_DESCRIPTION("Cisco VIC (usNIC) Verbs Driver");
+MODULE_AUTHOR("Upinder Malhi <umalhi@cisco.com>");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_VERSION(DRV_VERSION);
+module_param(usnic_log_lvl, uint, S_IRUGO | S_IWUSR);
+module_param(usnic_ib_share_vf, uint, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(usnic_log_lvl, " Off=0, Err=1, Info=2, Debug=3");
+MODULE_PARM_DESC(usnic_ib_share_vf, "Off=0, On=1 VF sharing amongst QPs");
+MODULE_DEVICE_TABLE(pci, usnic_ib_pci_ids);
+
+module_init(usnic_ib_init);
+module_exit(usnic_ib_destroy);
+/* End of module section */
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_qp_grp.c b/drivers/infiniband/hw/usnic/usnic_ib_qp_grp.c
new file mode 100644 (file)
index 0000000..ca5fa6a
--- /dev/null
@@ -0,0 +1,541 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/errno.h>
+#include <linux/module.h>
+#include <linux/spinlock.h>
+
+#include "usnic_log.h"
+#include "usnic_vnic.h"
+#include "usnic_fwd.h"
+#include "usnic_uiom.h"
+#include "usnic_ib_qp_grp.h"
+#include "usnic_ib_sysfs.h"
+#include "usnic_transport.h"
+
+const char *usnic_ib_qp_grp_state_to_string(enum ib_qp_state state)
+{
+       switch (state) {
+       case IB_QPS_RESET:
+               return "Rst";
+       case IB_QPS_INIT:
+               return "Init";
+       case IB_QPS_RTR:
+               return "RTR";
+       case IB_QPS_RTS:
+               return "RTS";
+       case IB_QPS_SQD:
+               return "SQD";
+       case IB_QPS_SQE:
+               return "SQE";
+       case IB_QPS_ERR:
+               return "ERR";
+       default:
+               return "UNKOWN STATE";
+
+       }
+}
+
+int usnic_ib_qp_grp_dump_hdr(char *buf, int buf_sz)
+{
+       return scnprintf(buf, buf_sz, "|QPN\t|State\t|PID\t|VF Idx\t|Fil ID");
+}
+
+int usnic_ib_qp_grp_dump_rows(void *obj, char *buf, int buf_sz)
+{
+       struct usnic_ib_qp_grp *qp_grp = obj;
+       struct usnic_fwd_filter_hndl *default_filter_hndl;
+       if (obj) {
+               default_filter_hndl = list_first_entry(&qp_grp->filter_hndls,
+                                       struct usnic_fwd_filter_hndl, link);
+               return scnprintf(buf, buf_sz, "|%d\t|%s\t|%d\t|%hu\t|%d",
+                                       qp_grp->ibqp.qp_num,
+                                       usnic_ib_qp_grp_state_to_string(
+                                                       qp_grp->state),
+                                       qp_grp->owner_pid,
+                                       usnic_vnic_get_index(qp_grp->vf->vnic),
+                                       default_filter_hndl->id);
+       } else {
+               return scnprintf(buf, buf_sz, "|N/A\t|N/A\t|N/A\t|N/A\t|N/A");
+       }
+}
+
+static int add_fwd_filter(struct usnic_ib_qp_grp *qp_grp,
+                               struct usnic_fwd_filter *fwd_filter)
+{
+       struct usnic_fwd_filter_hndl *filter_hndl;
+       int status;
+       struct usnic_vnic_res_chunk *chunk;
+       int rq_idx;
+
+       WARN_ON(!spin_is_locked(&qp_grp->lock));
+
+       chunk = usnic_ib_qp_grp_get_chunk(qp_grp, USNIC_VNIC_RES_TYPE_RQ);
+       if (IS_ERR_OR_NULL(chunk) || chunk->cnt < 1) {
+               usnic_err("Failed to get RQ info for qp_grp %u\n",
+                               qp_grp->grp_id);
+               return -EFAULT;
+       }
+
+       rq_idx = chunk->res[0]->vnic_idx;
+
+       switch (qp_grp->transport) {
+       case USNIC_TRANSPORT_ROCE_CUSTOM:
+               status = usnic_fwd_add_usnic_filter(qp_grp->ufdev,
+                                       usnic_vnic_get_index(qp_grp->vf->vnic),
+                                       rq_idx,
+                                       fwd_filter,
+                                       &filter_hndl);
+               break;
+       default:
+               usnic_err("Unable to install filter for qp_grp %u for transport %d",
+                               qp_grp->grp_id, qp_grp->transport);
+               status = -EINVAL;
+       }
+
+       if (status)
+               return status;
+
+       list_add_tail(&filter_hndl->link, &qp_grp->filter_hndls);
+       return 0;
+}
+
+static int del_all_filters(struct usnic_ib_qp_grp *qp_grp)
+{
+       int err, status;
+       struct usnic_fwd_filter_hndl *filter_hndl, *tmp;
+
+       WARN_ON(!spin_is_locked(&qp_grp->lock));
+
+       status = 0;
+
+       list_for_each_entry_safe(filter_hndl, tmp,
+                                       &qp_grp->filter_hndls, link) {
+               list_del(&filter_hndl->link);
+               err = usnic_fwd_del_filter(filter_hndl);
+               if (err) {
+                       usnic_err("Failed to delete filter %u of qp_grp %d\n",
+                                       filter_hndl->id, qp_grp->grp_id);
+               }
+               status |= err;
+       }
+
+       return status;
+}
+
+static int enable_qp_grp(struct usnic_ib_qp_grp *qp_grp)
+{
+
+       int status;
+       int i, vnic_idx;
+       struct usnic_vnic_res_chunk *res_chunk;
+       struct usnic_vnic_res *res;
+
+       WARN_ON(!spin_is_locked(&qp_grp->lock));
+
+       vnic_idx = usnic_vnic_get_index(qp_grp->vf->vnic);
+
+       res_chunk = usnic_ib_qp_grp_get_chunk(qp_grp, USNIC_VNIC_RES_TYPE_RQ);
+       if (IS_ERR_OR_NULL(res_chunk)) {
+               usnic_err("Unable to get %s with err %ld\n",
+                       usnic_vnic_res_type_to_str(USNIC_VNIC_RES_TYPE_RQ),
+                       PTR_ERR(res_chunk));
+               return res_chunk ? PTR_ERR(res_chunk) : -ENOMEM;
+       }
+
+       for (i = 0; i < res_chunk->cnt; i++) {
+               res = res_chunk->res[i];
+               status = usnic_fwd_enable_rq(qp_grp->ufdev, vnic_idx,
+                                               res->vnic_idx);
+               if (status) {
+                       usnic_err("Failed to enable rq %d of %s:%d\n with err %d\n",
+                                       res->vnic_idx,
+                                       netdev_name(qp_grp->ufdev->netdev),
+                                       vnic_idx, status);
+                       goto out_err;
+               }
+       }
+
+       return 0;
+
+out_err:
+       for (i--; i >= 0; i--) {
+               res = res_chunk->res[i];
+               usnic_fwd_disable_rq(qp_grp->ufdev, vnic_idx,
+                                       res->vnic_idx);
+       }
+
+       return status;
+}
+
+static int disable_qp_grp(struct usnic_ib_qp_grp *qp_grp)
+{
+       int i, vnic_idx;
+       struct usnic_vnic_res_chunk *res_chunk;
+       struct usnic_vnic_res *res;
+       int status = 0;
+
+       WARN_ON(!spin_is_locked(&qp_grp->lock));
+       vnic_idx = usnic_vnic_get_index(qp_grp->vf->vnic);
+
+       res_chunk = usnic_ib_qp_grp_get_chunk(qp_grp, USNIC_VNIC_RES_TYPE_RQ);
+       if (IS_ERR_OR_NULL(res_chunk)) {
+               usnic_err("Unable to get %s with err %ld\n",
+                       usnic_vnic_res_type_to_str(USNIC_VNIC_RES_TYPE_RQ),
+                       PTR_ERR(res_chunk));
+               return res_chunk ? PTR_ERR(res_chunk) : -ENOMEM;
+       }
+
+       for (i = 0; i < res_chunk->cnt; i++) {
+               res = res_chunk->res[i];
+               status = usnic_fwd_disable_rq(qp_grp->ufdev, vnic_idx,
+                                               res->vnic_idx);
+               if (status) {
+                       usnic_err("Failed to disable rq %d of %s:%d\n with err %d\n",
+                                       res->vnic_idx,
+                                       netdev_name(qp_grp->ufdev->netdev),
+                                       vnic_idx, status);
+               }
+       }
+
+       return status;
+
+}
+
+int usnic_ib_qp_grp_modify(struct usnic_ib_qp_grp *qp_grp,
+                               enum ib_qp_state new_state,
+                               struct usnic_fwd_filter *fwd_filter)
+{
+       int status = 0;
+       int vnic_idx;
+       struct ib_event ib_event;
+       enum ib_qp_state old_state;
+
+       old_state = qp_grp->state;
+       vnic_idx = usnic_vnic_get_index(qp_grp->vf->vnic);
+
+       spin_lock(&qp_grp->lock);
+       switch (new_state) {
+       case IB_QPS_RESET:
+               switch (old_state) {
+               case IB_QPS_RESET:
+                       /* NO-OP */
+                       break;
+               case IB_QPS_INIT:
+                       status = del_all_filters(qp_grp);
+                       break;
+               case IB_QPS_RTR:
+               case IB_QPS_RTS:
+               case IB_QPS_ERR:
+                       status = disable_qp_grp(qp_grp);
+                       status &= del_all_filters(qp_grp);
+                       break;
+               default:
+                       status = -EINVAL;
+               }
+               break;
+       case IB_QPS_INIT:
+               switch (old_state) {
+               case IB_QPS_RESET:
+                       status = add_fwd_filter(qp_grp, fwd_filter);
+                       break;
+               case IB_QPS_INIT:
+                       status = add_fwd_filter(qp_grp, fwd_filter);
+                       break;
+               case IB_QPS_RTR:
+                       status = disable_qp_grp(qp_grp);
+                       break;
+               case IB_QPS_RTS:
+                       status = disable_qp_grp(qp_grp);
+                       break;
+               default:
+                       status = -EINVAL;
+               }
+               break;
+       case IB_QPS_RTR:
+               switch (old_state) {
+               case IB_QPS_INIT:
+                       status = enable_qp_grp(qp_grp);
+                       break;
+               default:
+                       status = -EINVAL;
+               }
+               break;
+       case IB_QPS_RTS:
+               switch (old_state) {
+               case IB_QPS_RTR:
+                       /* NO-OP FOR NOW */
+                       break;
+               default:
+                       status = -EINVAL;
+               }
+               break;
+       case IB_QPS_ERR:
+               ib_event.device = &qp_grp->vf->pf->ib_dev;
+               ib_event.element.qp = &qp_grp->ibqp;
+               ib_event.event = IB_EVENT_QP_FATAL;
+
+               switch (old_state) {
+               case IB_QPS_RESET:
+                       qp_grp->ibqp.event_handler(&ib_event,
+                                       qp_grp->ibqp.qp_context);
+                       break;
+               case IB_QPS_INIT:
+                       status = del_all_filters(qp_grp);
+                       qp_grp->ibqp.event_handler(&ib_event,
+                                       qp_grp->ibqp.qp_context);
+                       break;
+               case IB_QPS_RTR:
+               case IB_QPS_RTS:
+                       status = disable_qp_grp(qp_grp);
+                       status &= del_all_filters(qp_grp);
+                       qp_grp->ibqp.event_handler(&ib_event,
+                                       qp_grp->ibqp.qp_context);
+                       break;
+               default:
+                       status = -EINVAL;
+               }
+               break;
+       default:
+               status = -EINVAL;
+       }
+       spin_unlock(&qp_grp->lock);
+
+       if (!status) {
+               qp_grp->state = new_state;
+               usnic_info("Transistioned %u from %s to %s",
+               qp_grp->grp_id,
+               usnic_ib_qp_grp_state_to_string(old_state),
+               usnic_ib_qp_grp_state_to_string(new_state));
+       } else {
+               usnic_err("Failed to transistion %u from %s to %s",
+               qp_grp->grp_id,
+               usnic_ib_qp_grp_state_to_string(old_state),
+               usnic_ib_qp_grp_state_to_string(new_state));
+       }
+
+       return status;
+}
+
+static struct usnic_vnic_res_chunk**
+alloc_res_chunk_list(struct usnic_vnic *vnic,
+                       struct usnic_vnic_res_spec *res_spec, void *owner_obj)
+{
+       enum usnic_vnic_res_type res_type;
+       struct usnic_vnic_res_chunk **res_chunk_list;
+       int err, i, res_cnt, res_lst_sz;
+
+       for (res_lst_sz = 0;
+               res_spec->resources[res_lst_sz].type != USNIC_VNIC_RES_TYPE_EOL;
+               res_lst_sz++) {
+               /* Do Nothing */
+       }
+
+       res_chunk_list = kzalloc(sizeof(*res_chunk_list)*(res_lst_sz+1),
+                                       GFP_ATOMIC);
+       if (!res_chunk_list)
+               return ERR_PTR(-ENOMEM);
+
+       for (i = 0; res_spec->resources[i].type != USNIC_VNIC_RES_TYPE_EOL;
+               i++) {
+               res_type = res_spec->resources[i].type;
+               res_cnt = res_spec->resources[i].cnt;
+
+               res_chunk_list[i] = usnic_vnic_get_resources(vnic, res_type,
+                                       res_cnt, owner_obj);
+               if (IS_ERR_OR_NULL(res_chunk_list[i])) {
+                       err = (res_chunk_list[i] ?
+                                       PTR_ERR(res_chunk_list[i]) : -ENOMEM);
+                       usnic_err("Failed to get %s from %s with err %d\n",
+                               usnic_vnic_res_type_to_str(res_type),
+                               usnic_vnic_pci_name(vnic),
+                               err);
+                       goto out_free_res;
+               }
+       }
+
+       return res_chunk_list;
+
+out_free_res:
+       for (i--; i > 0; i--)
+               usnic_vnic_put_resources(res_chunk_list[i]);
+       kfree(res_chunk_list);
+       return ERR_PTR(err);
+}
+
+static void free_qp_grp_res(struct usnic_vnic_res_chunk **res_chunk_list)
+{
+       int i;
+       for (i = 0; res_chunk_list[i]; i++)
+               usnic_vnic_put_resources(res_chunk_list[i]);
+       kfree(res_chunk_list);
+}
+
+static int qp_grp_and_vf_bind(struct usnic_ib_vf *vf,
+                               struct usnic_ib_pd *pd,
+                               struct usnic_ib_qp_grp *qp_grp)
+{
+       int err;
+       struct pci_dev *pdev;
+
+       WARN_ON(!spin_is_locked(&vf->lock));
+
+       pdev = usnic_vnic_get_pdev(vf->vnic);
+       if (vf->qp_grp_ref_cnt == 0) {
+               err = usnic_uiom_attach_dev_to_pd(pd->umem_pd, &pdev->dev);
+               if (err) {
+                       usnic_err("Failed to attach %s to domain\n",
+                                       pci_name(pdev));
+                       return err;
+               }
+               vf->pd = pd;
+       }
+       vf->qp_grp_ref_cnt++;
+
+       WARN_ON(vf->pd != pd);
+       qp_grp->vf = vf;
+
+       return 0;
+}
+
+static void qp_grp_and_vf_unbind(struct usnic_ib_qp_grp *qp_grp)
+{
+       struct pci_dev *pdev;
+       struct usnic_ib_pd *pd;
+
+       WARN_ON(!spin_is_locked(&qp_grp->vf->lock));
+
+       pd = qp_grp->vf->pd;
+       pdev = usnic_vnic_get_pdev(qp_grp->vf->vnic);
+       if (--qp_grp->vf->qp_grp_ref_cnt == 0) {
+               qp_grp->vf->pd = NULL;
+               usnic_uiom_detach_dev_from_pd(pd->umem_pd, &pdev->dev);
+       }
+       qp_grp->vf = NULL;
+}
+
+static void log_spec(struct usnic_vnic_res_spec *res_spec)
+{
+       char buf[512];
+       usnic_vnic_spec_dump(buf, sizeof(buf), res_spec);
+       usnic_dbg("%s\n", buf);
+}
+
+struct usnic_ib_qp_grp *
+usnic_ib_qp_grp_create(struct usnic_fwd_dev *ufdev,
+                       struct usnic_ib_vf *vf,
+                       struct usnic_ib_pd *pd,
+                       struct usnic_vnic_res_spec *res_spec,
+                       enum usnic_transport_type transport)
+{
+       struct usnic_ib_qp_grp *qp_grp;
+       u16 port_num;
+       int err;
+
+       WARN_ON(!spin_is_locked(&vf->lock));
+
+       err = usnic_vnic_res_spec_satisfied(&min_transport_spec[transport],
+                                               res_spec);
+       if (err) {
+               usnic_err("Spec does not meet miniumum req for transport %d\n",
+                               transport);
+               log_spec(res_spec);
+               return ERR_PTR(err);
+       }
+
+       port_num = usnic_transport_rsrv_port(transport, 0);
+       if (!port_num) {
+               usnic_err("Unable to allocate port for %s\n",
+                               netdev_name(ufdev->netdev));
+               return ERR_PTR(-EINVAL);
+       }
+
+       qp_grp = kzalloc(sizeof(*qp_grp), GFP_ATOMIC);
+       if (!qp_grp) {
+               usnic_err("Unable to alloc qp_grp - Out of memory\n");
+               return NULL;
+       }
+
+       qp_grp->res_chunk_list = alloc_res_chunk_list(vf->vnic, res_spec,
+                                                       qp_grp);
+       if (IS_ERR_OR_NULL(qp_grp->res_chunk_list)) {
+               err = qp_grp->res_chunk_list ?
+                               PTR_ERR(qp_grp->res_chunk_list) : -ENOMEM;
+               usnic_err("Unable to alloc res for %d with err %d\n",
+                               qp_grp->grp_id, err);
+               goto out_free_port;
+       }
+
+       INIT_LIST_HEAD(&qp_grp->filter_hndls);
+       spin_lock_init(&qp_grp->lock);
+       qp_grp->ufdev = ufdev;
+       qp_grp->transport = transport;
+       qp_grp->filters[DFLT_FILTER_IDX].transport = transport;
+       qp_grp->filters[DFLT_FILTER_IDX].port_num = port_num;
+       qp_grp->state = IB_QPS_RESET;
+       qp_grp->owner_pid = current->pid;
+
+       /* qp_num is same as default filter port_num */
+       qp_grp->ibqp.qp_num = qp_grp->filters[DFLT_FILTER_IDX].port_num;
+       qp_grp->grp_id = qp_grp->ibqp.qp_num;
+
+       err = qp_grp_and_vf_bind(vf, pd, qp_grp);
+       if (err)
+               goto out_free_port;
+
+       usnic_ib_sysfs_qpn_add(qp_grp);
+
+       return qp_grp;
+
+out_free_port:
+       kfree(qp_grp);
+       usnic_transport_unrsrv_port(transport, port_num);
+
+       return ERR_PTR(err);
+}
+
+void usnic_ib_qp_grp_destroy(struct usnic_ib_qp_grp *qp_grp)
+{
+       u16 default_port_num;
+       enum usnic_transport_type transport;
+
+       WARN_ON(qp_grp->state != IB_QPS_RESET);
+       WARN_ON(!spin_is_locked(&qp_grp->vf->lock));
+
+       transport = qp_grp->filters[DFLT_FILTER_IDX].transport;
+       default_port_num = qp_grp->filters[DFLT_FILTER_IDX].port_num;
+
+       usnic_ib_sysfs_qpn_remove(qp_grp);
+       qp_grp_and_vf_unbind(qp_grp);
+       free_qp_grp_res(qp_grp->res_chunk_list);
+       kfree(qp_grp);
+       usnic_transport_unrsrv_port(transport, default_port_num);
+}
+
+struct usnic_vnic_res_chunk*
+usnic_ib_qp_grp_get_chunk(struct usnic_ib_qp_grp *qp_grp,
+                               enum usnic_vnic_res_type res_type)
+{
+       int i;
+
+       for (i = 0; qp_grp->res_chunk_list[i]; i++) {
+               if (qp_grp->res_chunk_list[i]->type == res_type)
+                       return qp_grp->res_chunk_list[i];
+       }
+
+       return ERR_PTR(-EINVAL);
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_qp_grp.h b/drivers/infiniband/hw/usnic/usnic_ib_qp_grp.h
new file mode 100644 (file)
index 0000000..6416a95
--- /dev/null
@@ -0,0 +1,95 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_IB_QP_GRP_H_
+#define USNIC_IB_QP_GRP_H_
+
+#include <rdma/ib_verbs.h>
+
+#include "usnic_ib.h"
+#include "usnic_abi.h"
+#include "usnic_fwd.h"
+#include "usnic_vnic.h"
+
+#define MAX_QP_GRP_FILTERS     10
+#define DFLT_FILTER_IDX                0
+
+/*
+ * The qp group struct represents all the hw resources needed to present a ib_qp
+ */
+struct usnic_ib_qp_grp {
+       struct ib_qp                            ibqp;
+       enum ib_qp_state                        state;
+       int                                     grp_id;
+
+       struct usnic_fwd_dev                    *ufdev;
+       short unsigned                          filter_cnt;
+       struct usnic_fwd_filter                 filters[MAX_QP_GRP_FILTERS];
+       struct list_head                        filter_hndls;
+       enum usnic_transport_type               transport;
+       struct usnic_ib_ucontext                *ctx;
+
+       struct usnic_vnic_res_chunk             **res_chunk_list;
+
+       pid_t                                   owner_pid;
+       struct usnic_ib_vf                      *vf;
+       struct list_head                        link;
+
+       spinlock_t                              lock;
+
+       struct kobject                          kobj;
+};
+
+static const struct
+usnic_vnic_res_spec min_transport_spec[USNIC_TRANSPORT_MAX] = {
+       { /*USNIC_TRANSPORT_UNKNOWN*/
+               .resources = {
+                       {.type = USNIC_VNIC_RES_TYPE_EOL,       .cnt = 0,},
+               },
+       },
+       { /*USNIC_TRANSPORT_ROCE_CUSTOM*/
+               .resources = {
+                       {.type = USNIC_VNIC_RES_TYPE_WQ,        .cnt = 1,},
+                       {.type = USNIC_VNIC_RES_TYPE_RQ,        .cnt = 1,},
+                       {.type = USNIC_VNIC_RES_TYPE_CQ,        .cnt = 1,},
+                       {.type = USNIC_VNIC_RES_TYPE_EOL,       .cnt = 0,},
+               },
+       },
+};
+
+const char *usnic_ib_qp_grp_state_to_string(enum ib_qp_state state);
+int usnic_ib_qp_grp_dump_hdr(char *buf, int buf_sz);
+int usnic_ib_qp_grp_dump_rows(void *obj, char *buf, int buf_sz);
+struct usnic_ib_qp_grp *
+usnic_ib_qp_grp_create(struct usnic_fwd_dev *ufdev, struct usnic_ib_vf *vf,
+                       struct usnic_ib_pd *pd,
+                       struct usnic_vnic_res_spec *res_spec,
+                       enum usnic_transport_type transport);
+void usnic_ib_qp_grp_destroy(struct usnic_ib_qp_grp *qp_grp);
+int usnic_ib_qp_grp_modify(struct usnic_ib_qp_grp *qp_grp,
+                               enum ib_qp_state new_state,
+                               struct usnic_fwd_filter *fwd_filter);
+struct usnic_vnic_res_chunk
+*usnic_ib_qp_grp_get_chunk(struct usnic_ib_qp_grp *qp_grp,
+                               enum usnic_vnic_res_type type);
+static inline
+struct usnic_ib_qp_grp *to_uqp_grp(struct ib_qp *ibqp)
+{
+       return container_of(ibqp, struct usnic_ib_qp_grp, ibqp);
+}
+#endif /* USNIC_IB_QP_GRP_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_sysfs.c b/drivers/infiniband/hw/usnic/usnic_ib_sysfs.c
new file mode 100644 (file)
index 0000000..bad985e
--- /dev/null
@@ -0,0 +1,351 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/errno.h>
+
+#include <rdma/ib_user_verbs.h>
+#include <rdma/ib_addr.h>
+
+#include "usnic_common_util.h"
+#include "usnic_ib.h"
+#include "usnic_ib_qp_grp.h"
+#include "usnic_vnic.h"
+#include "usnic_ib_verbs.h"
+#include "usnic_log.h"
+
+#define UPDATE_PTR_LEFT(N, P, L)                       \
+do {                                                   \
+       L -= (N);                                       \
+       P += (N);                                       \
+} while (0)
+
+static ssize_t usnic_ib_show_fw_ver(struct device *device,
+                                       struct device_attribute *attr,
+                                       char *buf)
+{
+       struct usnic_ib_dev *us_ibdev =
+               container_of(device, struct usnic_ib_dev, ib_dev.dev);
+       struct ethtool_drvinfo info;
+
+       mutex_lock(&us_ibdev->usdev_lock);
+       us_ibdev->netdev->ethtool_ops->get_drvinfo(us_ibdev->netdev, &info);
+       mutex_unlock(&us_ibdev->usdev_lock);
+
+       return scnprintf(buf, PAGE_SIZE, "%s\n", info.fw_version);
+}
+
+static ssize_t usnic_ib_show_board(struct device *device,
+                                       struct device_attribute *attr,
+                                       char *buf)
+{
+       struct usnic_ib_dev *us_ibdev =
+               container_of(device, struct usnic_ib_dev, ib_dev.dev);
+       unsigned short subsystem_device_id;
+
+       mutex_lock(&us_ibdev->usdev_lock);
+       subsystem_device_id = us_ibdev->pdev->subsystem_device;
+       mutex_unlock(&us_ibdev->usdev_lock);
+
+       return scnprintf(buf, PAGE_SIZE, "%hu\n", subsystem_device_id);
+}
+
+/*
+ * Report the configuration for this PF
+ */
+static ssize_t
+usnic_ib_show_config(struct device *device, struct device_attribute *attr,
+                       char *buf)
+{
+       struct usnic_ib_dev *us_ibdev;
+       char *ptr;
+       unsigned left;
+       unsigned n;
+       enum usnic_vnic_res_type res_type;
+
+       us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev);
+
+       /* Buffer space limit is 1 page */
+       ptr = buf;
+       left = PAGE_SIZE;
+
+       mutex_lock(&us_ibdev->usdev_lock);
+       if (atomic_read(&us_ibdev->vf_cnt.refcount) > 0) {
+               char *busname;
+
+               /*
+                * bus name seems to come with annoying prefix.
+                * Remove it if it is predictable
+                */
+               busname = us_ibdev->pdev->bus->name;
+               if (strncmp(busname, "PCI Bus ", 8) == 0)
+                       busname += 8;
+
+               n = scnprintf(ptr, left,
+                       "%s: %s:%d.%d, %s, %pM, %u VFs\n Per VF:",
+                       us_ibdev->ib_dev.name,
+                       busname,
+                       PCI_SLOT(us_ibdev->pdev->devfn),
+                       PCI_FUNC(us_ibdev->pdev->devfn),
+                       netdev_name(us_ibdev->netdev),
+                       us_ibdev->mac,
+                       atomic_read(&us_ibdev->vf_cnt.refcount));
+               UPDATE_PTR_LEFT(n, ptr, left);
+
+               for (res_type = USNIC_VNIC_RES_TYPE_EOL;
+                               res_type < USNIC_VNIC_RES_TYPE_MAX;
+                               res_type++) {
+                       if (us_ibdev->vf_res_cnt[res_type] == 0)
+                               continue;
+                       n = scnprintf(ptr, left, " %d %s%s",
+                               us_ibdev->vf_res_cnt[res_type],
+                               usnic_vnic_res_type_to_str(res_type),
+                               (res_type < (USNIC_VNIC_RES_TYPE_MAX - 1)) ?
+                                "," : "");
+                       UPDATE_PTR_LEFT(n, ptr, left);
+               }
+               n = scnprintf(ptr, left, "\n");
+               UPDATE_PTR_LEFT(n, ptr, left);
+       } else {
+               n = scnprintf(ptr, left, "%s: no VFs\n",
+                               us_ibdev->ib_dev.name);
+               UPDATE_PTR_LEFT(n, ptr, left);
+       }
+       mutex_unlock(&us_ibdev->usdev_lock);
+
+       return ptr - buf;
+}
+
+static ssize_t
+usnic_ib_show_iface(struct device *device, struct device_attribute *attr,
+                       char *buf)
+{
+       struct usnic_ib_dev *us_ibdev;
+
+       us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev);
+
+       return scnprintf(buf, PAGE_SIZE, "%s\n",
+                       netdev_name(us_ibdev->netdev));
+}
+
+static ssize_t
+usnic_ib_show_max_vf(struct device *device, struct device_attribute *attr,
+                       char *buf)
+{
+       struct usnic_ib_dev *us_ibdev;
+
+       us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev);
+
+       return scnprintf(buf, PAGE_SIZE, "%u\n",
+                       atomic_read(&us_ibdev->vf_cnt.refcount));
+}
+
+static ssize_t
+usnic_ib_show_qp_per_vf(struct device *device, struct device_attribute *attr,
+                       char *buf)
+{
+       struct usnic_ib_dev *us_ibdev;
+       int qp_per_vf;
+
+       us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev);
+       qp_per_vf = max(us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_WQ],
+                       us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_RQ]);
+
+       return scnprintf(buf, PAGE_SIZE,
+                               "%d\n", qp_per_vf);
+}
+
+static ssize_t
+usnic_ib_show_cq_per_vf(struct device *device, struct device_attribute *attr,
+                       char *buf)
+{
+       struct usnic_ib_dev *us_ibdev;
+
+       us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev);
+
+       return scnprintf(buf, PAGE_SIZE, "%d\n",
+                       us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_CQ]);
+}
+
+static DEVICE_ATTR(fw_ver, S_IRUGO, usnic_ib_show_fw_ver, NULL);
+static DEVICE_ATTR(board_id, S_IRUGO, usnic_ib_show_board, NULL);
+static DEVICE_ATTR(config, S_IRUGO, usnic_ib_show_config, NULL);
+static DEVICE_ATTR(iface, S_IRUGO, usnic_ib_show_iface, NULL);
+static DEVICE_ATTR(max_vf, S_IRUGO, usnic_ib_show_max_vf, NULL);
+static DEVICE_ATTR(qp_per_vf, S_IRUGO, usnic_ib_show_qp_per_vf, NULL);
+static DEVICE_ATTR(cq_per_vf, S_IRUGO, usnic_ib_show_cq_per_vf, NULL);
+
+static struct device_attribute *usnic_class_attributes[] = {
+       &dev_attr_fw_ver,
+       &dev_attr_board_id,
+       &dev_attr_config,
+       &dev_attr_iface,
+       &dev_attr_max_vf,
+       &dev_attr_qp_per_vf,
+       &dev_attr_cq_per_vf,
+};
+
+struct qpn_attribute {
+       struct attribute attr;
+       ssize_t (*show)(struct usnic_ib_qp_grp *, char *buf);
+};
+
+/*
+ * Definitions for supporting QPN entries in sysfs
+ */
+static ssize_t
+usnic_ib_qpn_attr_show(struct kobject *kobj, struct attribute *attr, char *buf)
+{
+       struct usnic_ib_qp_grp *qp_grp;
+       struct qpn_attribute *qpn_attr;
+
+       qp_grp = container_of(kobj, struct usnic_ib_qp_grp, kobj);
+       qpn_attr = container_of(attr, struct qpn_attribute, attr);
+
+       return qpn_attr->show(qp_grp, buf);
+}
+
+static const struct sysfs_ops usnic_ib_qpn_sysfs_ops = {
+       .show = usnic_ib_qpn_attr_show
+};
+
+#define QPN_ATTR_RO(NAME) \
+struct qpn_attribute qpn_attr_##NAME = __ATTR_RO(NAME)
+
+static ssize_t context_show(struct usnic_ib_qp_grp *qp_grp, char *buf)
+{
+       return scnprintf(buf, PAGE_SIZE, "0x%p\n", qp_grp->ctx);
+}
+
+static ssize_t summary_show(struct usnic_ib_qp_grp *qp_grp, char *buf)
+{
+       int i, j, n;
+       int left;
+       char *ptr;
+       struct usnic_vnic_res_chunk *res_chunk;
+       struct usnic_fwd_filter_hndl *default_filter_hndl;
+       struct usnic_vnic_res *vnic_res;
+
+       left = PAGE_SIZE;
+       ptr = buf;
+       default_filter_hndl = list_first_entry(&qp_grp->filter_hndls,
+                                       struct usnic_fwd_filter_hndl, link);
+
+       n = scnprintf(ptr, left,
+                       "QPN: %d State: (%s) PID: %u VF Idx: %hu Filter ID: 0x%x ",
+                       qp_grp->ibqp.qp_num,
+                       usnic_ib_qp_grp_state_to_string(qp_grp->state),
+                       qp_grp->owner_pid,
+                       usnic_vnic_get_index(qp_grp->vf->vnic),
+                       default_filter_hndl->id);
+       UPDATE_PTR_LEFT(n, ptr, left);
+
+       for (i = 0; qp_grp->res_chunk_list[i]; i++) {
+               res_chunk = qp_grp->res_chunk_list[i];
+               for (j = 0; j < res_chunk->cnt; j++) {
+                       vnic_res = res_chunk->res[j];
+                       n = scnprintf(ptr, left, "%s[%d] ",
+                               usnic_vnic_res_type_to_str(vnic_res->type),
+                               vnic_res->vnic_idx);
+                       UPDATE_PTR_LEFT(n, ptr, left);
+               }
+       }
+
+       n = scnprintf(ptr, left, "\n");
+       UPDATE_PTR_LEFT(n, ptr, left);
+
+       return ptr - buf;
+}
+
+static QPN_ATTR_RO(context);
+static QPN_ATTR_RO(summary);
+
+static struct attribute *usnic_ib_qpn_default_attrs[] = {
+       &qpn_attr_context.attr,
+       &qpn_attr_summary.attr,
+       NULL
+};
+
+static struct kobj_type usnic_ib_qpn_type = {
+       .sysfs_ops = &usnic_ib_qpn_sysfs_ops,
+       .default_attrs = usnic_ib_qpn_default_attrs
+};
+
+int usnic_ib_sysfs_register_usdev(struct usnic_ib_dev *us_ibdev)
+{
+       int i;
+       int err;
+       for (i = 0; i < ARRAY_SIZE(usnic_class_attributes); ++i) {
+               err = device_create_file(&us_ibdev->ib_dev.dev,
+                                               usnic_class_attributes[i]);
+               if (err) {
+                       usnic_err("Failed to create device file %d for %s eith err %d",
+                               i, us_ibdev->ib_dev.name, err);
+                       return -EINVAL;
+               }
+       }
+
+       /* create kernel object for looking at individual QPs */
+       kobject_get(&us_ibdev->ib_dev.dev.kobj);
+       us_ibdev->qpn_kobj = kobject_create_and_add("qpn",
+                       &us_ibdev->ib_dev.dev.kobj);
+       if (us_ibdev->qpn_kobj == NULL) {
+               kobject_put(&us_ibdev->ib_dev.dev.kobj);
+               return -ENOMEM;
+       }
+
+       return 0;
+}
+
+void usnic_ib_sysfs_unregister_usdev(struct usnic_ib_dev *us_ibdev)
+{
+       int i;
+       for (i = 0; i < ARRAY_SIZE(usnic_class_attributes); ++i) {
+               device_remove_file(&us_ibdev->ib_dev.dev,
+                                       usnic_class_attributes[i]);
+       }
+
+       kobject_put(us_ibdev->qpn_kobj);
+}
+
+void usnic_ib_sysfs_qpn_add(struct usnic_ib_qp_grp *qp_grp)
+{
+       struct usnic_ib_dev *us_ibdev;
+       int err;
+
+       us_ibdev = qp_grp->vf->pf;
+
+       err = kobject_init_and_add(&qp_grp->kobj, &usnic_ib_qpn_type,
+                       kobject_get(us_ibdev->qpn_kobj),
+                       "%d", qp_grp->grp_id);
+       if (err) {
+               kobject_put(us_ibdev->qpn_kobj);
+               return;
+       }
+}
+
+void usnic_ib_sysfs_qpn_remove(struct usnic_ib_qp_grp *qp_grp)
+{
+       struct usnic_ib_dev *us_ibdev;
+
+       us_ibdev = qp_grp->vf->pf;
+
+       kobject_put(&qp_grp->kobj);
+       kobject_put(us_ibdev->qpn_kobj);
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_sysfs.h b/drivers/infiniband/hw/usnic/usnic_ib_sysfs.h
new file mode 100644 (file)
index 0000000..0d09b49
--- /dev/null
@@ -0,0 +1,29 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_IB_SYSFS_H_
+#define USNIC_IB_SYSFS_H_
+
+#include "usnic_ib.h"
+
+int usnic_ib_sysfs_register_usdev(struct usnic_ib_dev *us_ibdev);
+void usnic_ib_sysfs_unregister_usdev(struct usnic_ib_dev *us_ibdev);
+void usnic_ib_sysfs_qpn_add(struct usnic_ib_qp_grp *qp_grp);
+void usnic_ib_sysfs_qpn_remove(struct usnic_ib_qp_grp *qp_grp);
+
+#endif /* !USNIC_IB_SYSFS_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_verbs.c b/drivers/infiniband/hw/usnic/usnic_ib_verbs.c
new file mode 100644 (file)
index 0000000..8f8dfa2
--- /dev/null
@@ -0,0 +1,736 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/errno.h>
+
+#include <rdma/ib_user_verbs.h>
+#include <rdma/ib_addr.h>
+
+#include "usnic_abi.h"
+#include "usnic_ib.h"
+#include "usnic_common_util.h"
+#include "usnic_ib_qp_grp.h"
+#include "usnic_fwd.h"
+#include "usnic_log.h"
+#include "usnic_uiom.h"
+#include "usnic_transport.h"
+
+#define USNIC_DEFAULT_TRANSPORT USNIC_TRANSPORT_ROCE_CUSTOM
+
+static void usnic_ib_fw_string_to_u64(char *fw_ver_str, u64 *fw_ver)
+{
+       *fw_ver = (u64) *fw_ver_str;
+}
+
+static int usnic_ib_fill_create_qp_resp(struct usnic_ib_qp_grp *qp_grp,
+                                       struct ib_udata *udata)
+{
+       struct usnic_ib_dev *us_ibdev;
+       struct usnic_ib_create_qp_resp resp;
+       struct pci_dev *pdev;
+       struct vnic_dev_bar *bar;
+       struct usnic_vnic_res_chunk *chunk;
+       int i, err;
+
+       memset(&resp, 0, sizeof(resp));
+
+       us_ibdev = qp_grp->vf->pf;
+       pdev = usnic_vnic_get_pdev(qp_grp->vf->vnic);
+       if (!pdev) {
+               usnic_err("Failed to get pdev of qp_grp %d\n",
+                               qp_grp->grp_id);
+               return -EFAULT;
+       }
+
+       bar = usnic_vnic_get_bar(qp_grp->vf->vnic, 0);
+       if (!bar) {
+               usnic_err("Failed to get bar0 of qp_grp %d vf %s",
+                               qp_grp->grp_id, pci_name(pdev));
+               return -EFAULT;
+       }
+
+       resp.vfid = usnic_vnic_get_index(qp_grp->vf->vnic);
+       resp.bar_bus_addr = bar->bus_addr;
+       resp.bar_len = bar->len;
+       resp.transport = qp_grp->transport;
+
+       chunk = usnic_ib_qp_grp_get_chunk(qp_grp, USNIC_VNIC_RES_TYPE_RQ);
+       if (IS_ERR_OR_NULL(chunk)) {
+               usnic_err("Failed to get chunk %s for qp_grp %d with err %ld\n",
+                       usnic_vnic_res_type_to_str(USNIC_VNIC_RES_TYPE_RQ),
+                       qp_grp->grp_id,
+                       PTR_ERR(chunk));
+               return chunk ? PTR_ERR(chunk) : -ENOMEM;
+       }
+
+       WARN_ON(chunk->type != USNIC_VNIC_RES_TYPE_RQ);
+       resp.rq_cnt = chunk->cnt;
+       for (i = 0; i < chunk->cnt; i++)
+               resp.rq_idx[i] = chunk->res[i]->vnic_idx;
+
+       chunk = usnic_ib_qp_grp_get_chunk(qp_grp, USNIC_VNIC_RES_TYPE_WQ);
+       if (IS_ERR_OR_NULL(chunk)) {
+               usnic_err("Failed to get chunk %s for qp_grp %d with err %ld\n",
+                       usnic_vnic_res_type_to_str(USNIC_VNIC_RES_TYPE_WQ),
+                       qp_grp->grp_id,
+                       PTR_ERR(chunk));
+               return chunk ? PTR_ERR(chunk) : -ENOMEM;
+       }
+
+       WARN_ON(chunk->type != USNIC_VNIC_RES_TYPE_WQ);
+       resp.wq_cnt = chunk->cnt;
+       for (i = 0; i < chunk->cnt; i++)
+               resp.wq_idx[i] = chunk->res[i]->vnic_idx;
+
+       chunk = usnic_ib_qp_grp_get_chunk(qp_grp, USNIC_VNIC_RES_TYPE_CQ);
+       if (IS_ERR_OR_NULL(chunk)) {
+               usnic_err("Failed to get chunk %s for qp_grp %d with err %ld\n",
+                       usnic_vnic_res_type_to_str(USNIC_VNIC_RES_TYPE_CQ),
+                       qp_grp->grp_id,
+                       PTR_ERR(chunk));
+               return chunk ? PTR_ERR(chunk) : -ENOMEM;
+       }
+
+       WARN_ON(chunk->type != USNIC_VNIC_RES_TYPE_CQ);
+       resp.cq_cnt = chunk->cnt;
+       for (i = 0; i < chunk->cnt; i++)
+               resp.cq_idx[i] = chunk->res[i]->vnic_idx;
+
+       err = ib_copy_to_udata(udata, &resp, sizeof(resp));
+       if (err) {
+               usnic_err("Failed to copy udata for %s", us_ibdev->ib_dev.name);
+               return err;
+       }
+
+       return 0;
+}
+
+static struct usnic_ib_qp_grp*
+find_free_vf_and_create_qp_grp(struct usnic_ib_dev *us_ibdev,
+                               struct usnic_ib_pd *pd,
+                               enum usnic_transport_type transport,
+                               struct usnic_vnic_res_spec *res_spec)
+{
+       struct usnic_ib_vf *vf;
+       struct usnic_vnic *vnic;
+       struct usnic_ib_qp_grp *qp_grp;
+       struct device *dev, **dev_list;
+       int i, found = 0;
+
+       BUG_ON(!mutex_is_locked(&us_ibdev->usdev_lock));
+
+       if (list_empty(&us_ibdev->vf_dev_list)) {
+               usnic_info("No vfs to allocate\n");
+               return NULL;
+       }
+
+       if (!us_ibdev->link_up) {
+               usnic_info("Cannot allocate qp b/c PF link is down\n");
+               return NULL;
+       }
+
+       if (usnic_ib_share_vf) {
+               /* Try to find resouces on a used vf which is in pd */
+               dev_list = usnic_uiom_get_dev_list(pd->umem_pd);
+               for (i = 0; dev_list[i]; i++) {
+                       dev = dev_list[i];
+                       vf = pci_get_drvdata(to_pci_dev(dev));
+                       spin_lock(&vf->lock);
+                       vnic = vf->vnic;
+                       if (!usnic_vnic_check_room(vnic, res_spec)) {
+                               usnic_dbg("Found used vnic %s from %s\n",
+                                               us_ibdev->ib_dev.name,
+                                               pci_name(usnic_vnic_get_pdev(
+                                                                       vnic)));
+                               found = 1;
+                               break;
+                       }
+                       spin_unlock(&vf->lock);
+
+               }
+               usnic_uiom_free_dev_list(dev_list);
+       }
+
+       if (!found) {
+               /* Try to find resources on an unused vf */
+               list_for_each_entry(vf, &us_ibdev->vf_dev_list, link) {
+                       spin_lock(&vf->lock);
+                       vnic = vf->vnic;
+                       if (vf->qp_grp_ref_cnt == 0 &&
+                               usnic_vnic_check_room(vnic, res_spec) == 0) {
+                               found = 1;
+                               break;
+                       }
+                       spin_unlock(&vf->lock);
+               }
+       }
+
+       if (!found) {
+               usnic_info("No free qp grp found on %s\n",
+                               us_ibdev->ib_dev.name);
+               return ERR_PTR(-ENOMEM);
+       }
+
+       qp_grp = usnic_ib_qp_grp_create(us_ibdev->ufdev, vf, pd, res_spec,
+                                               transport);
+       spin_unlock(&vf->lock);
+       if (IS_ERR_OR_NULL(qp_grp)) {
+               usnic_err("Failed to allocate qp_grp\n");
+               return ERR_PTR(qp_grp ? PTR_ERR(qp_grp) : -ENOMEM);
+       }
+
+       return qp_grp;
+}
+
+static void qp_grp_destroy(struct usnic_ib_qp_grp *qp_grp)
+{
+       struct usnic_ib_vf *vf = qp_grp->vf;
+
+       WARN_ON(qp_grp->state != IB_QPS_RESET);
+
+       spin_lock(&vf->lock);
+       usnic_ib_qp_grp_destroy(qp_grp);
+       spin_unlock(&vf->lock);
+}
+
+static void eth_speed_to_ib_speed(int speed, u8 *active_speed,
+                                       u8 *active_width)
+{
+       if (speed <= 10000) {
+               *active_width = IB_WIDTH_1X;
+               *active_speed = IB_SPEED_FDR10;
+       } else if (speed <= 20000) {
+               *active_width = IB_WIDTH_4X;
+               *active_speed = IB_SPEED_DDR;
+       } else if (speed <= 30000) {
+               *active_width = IB_WIDTH_4X;
+               *active_speed = IB_SPEED_QDR;
+       } else if (speed <= 40000) {
+               *active_width = IB_WIDTH_4X;
+               *active_speed = IB_SPEED_FDR10;
+       } else {
+               *active_width = IB_WIDTH_4X;
+               *active_speed = IB_SPEED_EDR;
+       }
+}
+
+/* Start of ib callback functions */
+
+enum rdma_link_layer usnic_ib_port_link_layer(struct ib_device *device,
+                                               u8 port_num)
+{
+       return IB_LINK_LAYER_ETHERNET;
+}
+
+int usnic_ib_query_device(struct ib_device *ibdev,
+                               struct ib_device_attr *props)
+{
+       struct usnic_ib_dev *us_ibdev = to_usdev(ibdev);
+       union ib_gid gid;
+       struct ethtool_drvinfo info;
+       struct ethtool_cmd cmd;
+       int qp_per_vf;
+
+       usnic_dbg("\n");
+       mutex_lock(&us_ibdev->usdev_lock);
+       us_ibdev->netdev->ethtool_ops->get_drvinfo(us_ibdev->netdev, &info);
+       us_ibdev->netdev->ethtool_ops->get_settings(us_ibdev->netdev, &cmd);
+       memset(props, 0, sizeof(*props));
+       usnic_mac_to_gid(us_ibdev->mac, &gid.raw[0]);
+       memcpy(&props->sys_image_guid, &gid.global.interface_id,
+               sizeof(gid.global.interface_id));
+       usnic_ib_fw_string_to_u64(&info.fw_version[0], &props->fw_ver);
+       props->max_mr_size = USNIC_UIOM_MAX_MR_SIZE;
+       props->page_size_cap = USNIC_UIOM_PAGE_SIZE;
+       props->vendor_id = PCI_VENDOR_ID_CISCO;
+       props->vendor_part_id = PCI_DEVICE_ID_CISCO_VIC_USPACE_NIC;
+       props->hw_ver = us_ibdev->pdev->subsystem_device;
+       qp_per_vf = max(us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_WQ],
+                       us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_RQ]);
+       props->max_qp = qp_per_vf *
+               atomic_read(&us_ibdev->vf_cnt.refcount);
+       props->device_cap_flags = IB_DEVICE_PORT_ACTIVE_EVENT |
+               IB_DEVICE_SYS_IMAGE_GUID | IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
+       props->max_cq = us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_CQ] *
+               atomic_read(&us_ibdev->vf_cnt.refcount);
+       props->max_pd = USNIC_UIOM_MAX_PD_CNT;
+       props->max_mr = USNIC_UIOM_MAX_MR_CNT;
+       props->local_ca_ack_delay = 0;
+       props->max_pkeys = 0;
+       props->atomic_cap = IB_ATOMIC_NONE;
+       props->masked_atomic_cap = props->atomic_cap;
+       props->max_qp_rd_atom = 0;
+       props->max_qp_init_rd_atom = 0;
+       props->max_res_rd_atom = 0;
+       props->max_srq = 0;
+       props->max_srq_wr = 0;
+       props->max_srq_sge = 0;
+       props->max_fast_reg_page_list_len = 0;
+       props->max_mcast_grp = 0;
+       props->max_mcast_qp_attach = 0;
+       props->max_total_mcast_qp_attach = 0;
+       props->max_map_per_fmr = 0;
+       /* Owned by Userspace
+        * max_qp_wr, max_sge, max_sge_rd, max_cqe */
+       mutex_unlock(&us_ibdev->usdev_lock);
+
+       return 0;
+}
+
+int usnic_ib_query_port(struct ib_device *ibdev, u8 port,
+                               struct ib_port_attr *props)
+{
+       struct usnic_ib_dev *us_ibdev = to_usdev(ibdev);
+       struct ethtool_cmd cmd;
+
+       usnic_dbg("\n");
+
+       mutex_lock(&us_ibdev->usdev_lock);
+       us_ibdev->netdev->ethtool_ops->get_settings(us_ibdev->netdev, &cmd);
+       memset(props, 0, sizeof(*props));
+
+       props->lid = 0;
+       props->lmc = 1;
+       props->sm_lid = 0;
+       props->sm_sl = 0;
+
+       if (us_ibdev->link_up) {
+               props->state = IB_PORT_ACTIVE;
+               props->phys_state = 5;
+       } else {
+               props->state = IB_PORT_DOWN;
+               props->phys_state = 3;
+       }
+
+       props->port_cap_flags = 0;
+       props->gid_tbl_len = 1;
+       props->pkey_tbl_len = 1;
+       props->bad_pkey_cntr = 0;
+       props->qkey_viol_cntr = 0;
+       eth_speed_to_ib_speed(cmd.speed, &props->active_speed,
+                               &props->active_width);
+       props->max_mtu = IB_MTU_4096;
+       props->active_mtu = iboe_get_mtu(us_ibdev->mtu);
+       /* Userspace will adjust for hdrs */
+       props->max_msg_sz = us_ibdev->mtu;
+       props->max_vl_num = 1;
+       mutex_unlock(&us_ibdev->usdev_lock);
+
+       return 0;
+}
+
+int usnic_ib_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
+                               int qp_attr_mask,
+                               struct ib_qp_init_attr *qp_init_attr)
+{
+       struct usnic_ib_qp_grp *qp_grp;
+       struct usnic_ib_vf *vf;
+       int err;
+
+       usnic_dbg("\n");
+
+       memset(qp_attr, 0, sizeof(*qp_attr));
+       memset(qp_init_attr, 0, sizeof(*qp_init_attr));
+
+       qp_grp = to_uqp_grp(qp);
+       vf = qp_grp->vf;
+       mutex_lock(&vf->pf->usdev_lock);
+       usnic_dbg("\n");
+       qp_attr->qp_state = qp_grp->state;
+       qp_attr->cur_qp_state = qp_grp->state;
+
+       switch (qp_grp->ibqp.qp_type) {
+       case IB_QPT_UD:
+               qp_attr->qkey = 0;
+               break;
+       default:
+               usnic_err("Unexpected qp_type %d\n", qp_grp->ibqp.qp_type);
+               err = -EINVAL;
+               goto err_out;
+       }
+
+       mutex_unlock(&vf->pf->usdev_lock);
+       return 0;
+
+err_out:
+       mutex_unlock(&vf->pf->usdev_lock);
+       return err;
+}
+
+int usnic_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
+                               union ib_gid *gid)
+{
+
+       struct usnic_ib_dev *us_ibdev = to_usdev(ibdev);
+       usnic_dbg("\n");
+
+       if (index > 1)
+               return -EINVAL;
+
+       mutex_lock(&us_ibdev->usdev_lock);
+       memset(&(gid->raw[0]), 0, sizeof(gid->raw));
+       usnic_mac_to_gid(us_ibdev->mac, &gid->raw[0]);
+       mutex_unlock(&us_ibdev->usdev_lock);
+
+       return 0;
+}
+
+int usnic_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+                               u16 *pkey)
+{
+       if (index > 1)
+               return -EINVAL;
+
+       *pkey = 0xffff;
+       return 0;
+}
+
+struct ib_pd *usnic_ib_alloc_pd(struct ib_device *ibdev,
+                                       struct ib_ucontext *context,
+                                       struct ib_udata *udata)
+{
+       struct usnic_ib_pd *pd;
+       void *umem_pd;
+
+       usnic_dbg("\n");
+
+       pd = kzalloc(sizeof(*pd), GFP_KERNEL);
+       if (!pd)
+               return ERR_PTR(-ENOMEM);
+
+       umem_pd = pd->umem_pd = usnic_uiom_alloc_pd();
+       if (IS_ERR_OR_NULL(umem_pd)) {
+               kfree(pd);
+               return ERR_PTR(umem_pd ? PTR_ERR(umem_pd) : -ENOMEM);
+       }
+
+       usnic_info("domain 0x%p allocated for context 0x%p and device %s\n",
+                       pd, context, ibdev->name);
+       return &pd->ibpd;
+}
+
+int usnic_ib_dealloc_pd(struct ib_pd *pd)
+{
+       usnic_info("freeing domain 0x%p\n", pd);
+
+       usnic_uiom_dealloc_pd((to_upd(pd))->umem_pd);
+       kfree(pd);
+       return 0;
+}
+
+struct ib_qp *usnic_ib_create_qp(struct ib_pd *pd,
+                                       struct ib_qp_init_attr *init_attr,
+                                       struct ib_udata *udata)
+{
+       int err;
+       struct usnic_ib_dev *us_ibdev;
+       struct usnic_ib_qp_grp *qp_grp;
+       struct usnic_ib_ucontext *ucontext;
+       int cq_cnt;
+       struct usnic_vnic_res_spec res_spec;
+
+       usnic_dbg("\n");
+
+       ucontext = to_uucontext(pd->uobject->context);
+       us_ibdev = to_usdev(pd->device);
+
+       if (init_attr->qp_type != IB_QPT_UD) {
+               usnic_err("%s asked to make a non-UD QP: %d\n",
+                               us_ibdev->ib_dev.name, init_attr->qp_type);
+               return ERR_PTR(-EINVAL);
+       }
+
+       mutex_lock(&us_ibdev->usdev_lock);
+       cq_cnt = (init_attr->send_cq == init_attr->recv_cq) ? 1 : 2,
+       res_spec = min_transport_spec[USNIC_DEFAULT_TRANSPORT];
+       usnic_vnic_res_spec_update(&res_spec, USNIC_VNIC_RES_TYPE_CQ, cq_cnt);
+       qp_grp = find_free_vf_and_create_qp_grp(us_ibdev, to_upd(pd),
+                                               USNIC_DEFAULT_TRANSPORT,
+                                               &res_spec);
+       if (IS_ERR_OR_NULL(qp_grp)) {
+               err = (qp_grp ? PTR_ERR(qp_grp) : -ENOMEM);
+               goto out_release_mutex;
+       }
+
+       err = usnic_ib_fill_create_qp_resp(qp_grp, udata);
+       if (err) {
+               err = -EBUSY;
+               goto out_release_qp_grp;
+       }
+
+       qp_grp->ctx = ucontext;
+       list_add_tail(&qp_grp->link, &ucontext->qp_grp_list);
+       usnic_ib_log_vf(qp_grp->vf);
+       mutex_unlock(&us_ibdev->usdev_lock);
+       return &qp_grp->ibqp;
+
+out_release_qp_grp:
+       qp_grp_destroy(qp_grp);
+out_release_mutex:
+       mutex_unlock(&us_ibdev->usdev_lock);
+       return ERR_PTR(err);
+}
+
+int usnic_ib_destroy_qp(struct ib_qp *qp)
+{
+       struct usnic_ib_qp_grp *qp_grp;
+       struct usnic_ib_vf *vf;
+
+       usnic_dbg("\n");
+
+       qp_grp = to_uqp_grp(qp);
+       vf = qp_grp->vf;
+       mutex_lock(&vf->pf->usdev_lock);
+       if (usnic_ib_qp_grp_modify(qp_grp, IB_QPS_RESET, NULL)) {
+               usnic_err("Failed to move qp grp %u to reset\n",
+                               qp_grp->grp_id);
+       }
+
+       list_del(&qp_grp->link);
+       qp_grp_destroy(qp_grp);
+       mutex_unlock(&vf->pf->usdev_lock);
+
+       return 0;
+}
+
+int usnic_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+                               int attr_mask, struct ib_udata *udata)
+{
+       struct usnic_ib_qp_grp *qp_grp;
+       int status;
+       usnic_dbg("\n");
+
+       qp_grp = to_uqp_grp(ibqp);
+
+       /* TODO: Future Support All States */
+       mutex_lock(&qp_grp->vf->pf->usdev_lock);
+       if ((attr_mask & IB_QP_STATE) && attr->qp_state == IB_QPS_INIT) {
+               status = usnic_ib_qp_grp_modify(qp_grp, IB_QPS_INIT,
+                                       &qp_grp->filters[DFLT_FILTER_IDX]);
+       } else if ((attr_mask & IB_QP_STATE) && attr->qp_state == IB_QPS_RTR) {
+               status = usnic_ib_qp_grp_modify(qp_grp, IB_QPS_RTR, NULL);
+       } else if ((attr_mask & IB_QP_STATE) && attr->qp_state == IB_QPS_RTS) {
+               status = usnic_ib_qp_grp_modify(qp_grp, IB_QPS_RTS, NULL);
+       } else {
+               usnic_err("Unexpected combination mask: %u state: %u\n",
+                               attr_mask & IB_QP_STATE, attr->qp_state);
+               status = -EINVAL;
+       }
+
+       mutex_unlock(&qp_grp->vf->pf->usdev_lock);
+       return status;
+}
+
+struct ib_cq *usnic_ib_create_cq(struct ib_device *ibdev, int entries,
+                                       int vector, struct ib_ucontext *context,
+                                       struct ib_udata *udata)
+{
+       struct ib_cq *cq;
+
+       usnic_dbg("\n");
+       cq = kzalloc(sizeof(*cq), GFP_KERNEL);
+       if (!cq)
+               return ERR_PTR(-EBUSY);
+
+       return cq;
+}
+
+int usnic_ib_destroy_cq(struct ib_cq *cq)
+{
+       usnic_dbg("\n");
+       kfree(cq);
+       return 0;
+}
+
+struct ib_mr *usnic_ib_reg_mr(struct ib_pd *pd, u64 start, u64 length,
+                                       u64 virt_addr, int access_flags,
+                                       struct ib_udata *udata)
+{
+       struct usnic_ib_mr *mr;
+       int err;
+
+       usnic_dbg("start 0x%llx va 0x%llx length 0x%llx\n", start,
+                       virt_addr, length);
+
+       mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+       if (IS_ERR_OR_NULL(mr))
+               return ERR_PTR(mr ? PTR_ERR(mr) : -ENOMEM);
+
+       mr->umem = usnic_uiom_reg_get(to_upd(pd)->umem_pd, start, length,
+                                       access_flags, 0);
+       if (IS_ERR_OR_NULL(mr->umem)) {
+               err = (mr->umem) ? PTR_ERR(mr->umem) : -EFAULT;
+               goto err_free;
+       }
+
+       mr->ibmr.lkey = mr->ibmr.rkey = 0;
+       return &mr->ibmr;
+
+err_free:
+       kfree(mr);
+       return ERR_PTR(err);
+}
+
+int usnic_ib_dereg_mr(struct ib_mr *ibmr)
+{
+       struct usnic_ib_mr *mr = to_umr(ibmr);
+
+       usnic_dbg("va 0x%lx length 0x%zx\n", mr->umem->va, mr->umem->length);
+
+       usnic_uiom_reg_release(mr->umem, ibmr->pd->uobject->context->closing);
+       kfree(mr);
+       return 0;
+}
+
+struct ib_ucontext *usnic_ib_alloc_ucontext(struct ib_device *ibdev,
+                                                       struct ib_udata *udata)
+{
+       struct usnic_ib_ucontext *context;
+       struct usnic_ib_dev *us_ibdev = to_usdev(ibdev);
+       usnic_dbg("\n");
+
+       context = kmalloc(sizeof(*context), GFP_KERNEL);
+       if (!context)
+               return ERR_PTR(-ENOMEM);
+
+       INIT_LIST_HEAD(&context->qp_grp_list);
+       mutex_lock(&us_ibdev->usdev_lock);
+       list_add_tail(&context->link, &us_ibdev->ctx_list);
+       mutex_unlock(&us_ibdev->usdev_lock);
+
+       return &context->ibucontext;
+}
+
+int usnic_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
+{
+       struct usnic_ib_ucontext *context = to_uucontext(ibcontext);
+       struct usnic_ib_dev *us_ibdev = to_usdev(ibcontext->device);
+       usnic_dbg("\n");
+
+       mutex_lock(&us_ibdev->usdev_lock);
+       BUG_ON(!list_empty(&context->qp_grp_list));
+       list_del(&context->link);
+       mutex_unlock(&us_ibdev->usdev_lock);
+       kfree(context);
+       return 0;
+}
+
+int usnic_ib_mmap(struct ib_ucontext *context,
+                               struct vm_area_struct *vma)
+{
+       struct usnic_ib_ucontext *uctx = to_ucontext(context);
+       struct usnic_ib_dev *us_ibdev;
+       struct usnic_ib_qp_grp *qp_grp;
+       struct usnic_ib_vf *vf;
+       struct vnic_dev_bar *bar;
+       dma_addr_t bus_addr;
+       unsigned int len;
+       unsigned int vfid;
+
+       usnic_dbg("\n");
+
+       us_ibdev = to_usdev(context->device);
+       vma->vm_flags |= VM_IO;
+       vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+       vfid = vma->vm_pgoff;
+       usnic_dbg("Page Offset %lu PAGE_SHIFT %u VFID %u\n",
+                       vma->vm_pgoff, PAGE_SHIFT, vfid);
+
+       mutex_lock(&us_ibdev->usdev_lock);
+       list_for_each_entry(qp_grp, &uctx->qp_grp_list, link) {
+               vf = qp_grp->vf;
+               if (usnic_vnic_get_index(vf->vnic) == vfid) {
+                       bar = usnic_vnic_get_bar(vf->vnic, 0);
+                       if ((vma->vm_end - vma->vm_start) != bar->len) {
+                               usnic_err("Bar0 Len %lu - Request map %lu\n",
+                                               bar->len,
+                                               vma->vm_end - vma->vm_start);
+                               mutex_unlock(&us_ibdev->usdev_lock);
+                               return -EINVAL;
+                       }
+                       bus_addr = bar->bus_addr;
+                       len = bar->len;
+                       usnic_dbg("bus: %pa vaddr: %p size: %ld\n",
+                                       &bus_addr, bar->vaddr, bar->len);
+                       mutex_unlock(&us_ibdev->usdev_lock);
+
+                       return remap_pfn_range(vma,
+                                               vma->vm_start,
+                                               bus_addr >> PAGE_SHIFT,
+                                               len, vma->vm_page_prot);
+               }
+       }
+
+       mutex_unlock(&us_ibdev->usdev_lock);
+       usnic_err("No VF %u found\n", vfid);
+       return -EINVAL;
+}
+
+/* In ib callbacks section -  Start of stub funcs */
+struct ib_ah *usnic_ib_create_ah(struct ib_pd *pd,
+                                       struct ib_ah_attr *ah_attr)
+{
+       usnic_dbg("\n");
+       return ERR_PTR(-EPERM);
+}
+
+int usnic_ib_destroy_ah(struct ib_ah *ah)
+{
+       usnic_dbg("\n");
+       return -EINVAL;
+}
+
+int usnic_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+                               struct ib_send_wr **bad_wr)
+{
+       usnic_dbg("\n");
+       return -EINVAL;
+}
+
+int usnic_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+                               struct ib_recv_wr **bad_wr)
+{
+       usnic_dbg("\n");
+       return -EINVAL;
+}
+
+int usnic_ib_poll_cq(struct ib_cq *ibcq, int num_entries,
+                               struct ib_wc *wc)
+{
+       usnic_dbg("\n");
+       return -EINVAL;
+}
+
+int usnic_ib_req_notify_cq(struct ib_cq *cq,
+                                       enum ib_cq_notify_flags flags)
+{
+       usnic_dbg("\n");
+       return -EINVAL;
+}
+
+struct ib_mr *usnic_ib_get_dma_mr(struct ib_pd *pd, int acc)
+{
+       usnic_dbg("\n");
+       return ERR_PTR(-ENOMEM);
+}
+
+
+/* In ib callbacks section - End of stub funcs */
+/* End of ib callbacks section */
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_verbs.h b/drivers/infiniband/hw/usnic/usnic_ib_verbs.h
new file mode 100644 (file)
index 0000000..bb864f5
--- /dev/null
@@ -0,0 +1,72 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_IB_VERBS_H_
+#define USNIC_IB_VERBS_H_
+
+#include "usnic_ib.h"
+
+enum rdma_link_layer usnic_ib_port_link_layer(struct ib_device *device,
+                                               u8 port_num);
+int usnic_ib_query_device(struct ib_device *ibdev,
+                               struct ib_device_attr *props);
+int usnic_ib_query_port(struct ib_device *ibdev, u8 port,
+                               struct ib_port_attr *props);
+int usnic_ib_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
+                               int qp_attr_mask,
+                               struct ib_qp_init_attr *qp_init_attr);
+int usnic_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
+                               union ib_gid *gid);
+int usnic_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+                               u16 *pkey);
+struct ib_pd *usnic_ib_alloc_pd(struct ib_device *ibdev,
+                               struct ib_ucontext *context,
+                               struct ib_udata *udata);
+int usnic_ib_dealloc_pd(struct ib_pd *pd);
+struct ib_qp *usnic_ib_create_qp(struct ib_pd *pd,
+                                       struct ib_qp_init_attr *init_attr,
+                                       struct ib_udata *udata);
+int usnic_ib_destroy_qp(struct ib_qp *qp);
+int usnic_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+                               int attr_mask, struct ib_udata *udata);
+struct ib_cq *usnic_ib_create_cq(struct ib_device *ibdev, int entries,
+                                       int vector, struct ib_ucontext *context,
+                                       struct ib_udata *udata);
+int usnic_ib_destroy_cq(struct ib_cq *cq);
+struct ib_mr *usnic_ib_reg_mr(struct ib_pd *pd, u64 start, u64 length,
+                               u64 virt_addr, int access_flags,
+                               struct ib_udata *udata);
+int usnic_ib_dereg_mr(struct ib_mr *ibmr);
+struct ib_ucontext *usnic_ib_alloc_ucontext(struct ib_device *ibdev,
+                                               struct ib_udata *udata);
+int usnic_ib_dealloc_ucontext(struct ib_ucontext *ibcontext);
+int usnic_ib_mmap(struct ib_ucontext *context,
+                       struct vm_area_struct *vma);
+struct ib_ah *usnic_ib_create_ah(struct ib_pd *pd,
+                                       struct ib_ah_attr *ah_attr);
+int usnic_ib_destroy_ah(struct ib_ah *ah);
+int usnic_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+                       struct ib_send_wr **bad_wr);
+int usnic_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+                       struct ib_recv_wr **bad_wr);
+int usnic_ib_poll_cq(struct ib_cq *ibcq, int num_entries,
+                       struct ib_wc *wc);
+int usnic_ib_req_notify_cq(struct ib_cq *cq,
+                               enum ib_cq_notify_flags flags);
+struct ib_mr *usnic_ib_get_dma_mr(struct ib_pd *pd, int acc);
+#endif /* !USNIC_IB_VERBS_H */
diff --git a/drivers/infiniband/hw/usnic/usnic_log.h b/drivers/infiniband/hw/usnic/usnic_log.h
new file mode 100644 (file)
index 0000000..75777a6
--- /dev/null
@@ -0,0 +1,58 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_LOG_H_
+#define USNIC_LOG_H_
+
+#include "usnic.h"
+
+extern unsigned int usnic_log_lvl;
+
+#define USNIC_LOG_LVL_NONE             (0)
+#define USNIC_LOG_LVL_ERR              (1)
+#define USNIC_LOG_LVL_INFO             (2)
+#define USNIC_LOG_LVL_DBG              (3)
+
+#define usnic_printk(lvl, args...) \
+       do { \
+               printk(lvl "%s:%s:%d: ", DRV_NAME, __func__, \
+                               __LINE__); \
+               printk(args); \
+       } while (0)
+
+#define usnic_dbg(args...) \
+       do { \
+               if (unlikely(usnic_log_lvl >= USNIC_LOG_LVL_DBG)) { \
+                       usnic_printk(KERN_INFO, args); \
+       } \
+} while (0)
+
+#define usnic_info(args...) \
+do { \
+       if (usnic_log_lvl >= USNIC_LOG_LVL_INFO) { \
+                       usnic_printk(KERN_INFO, args); \
+       } \
+} while (0)
+
+#define usnic_err(args...) \
+       do { \
+               if (usnic_log_lvl >= USNIC_LOG_LVL_ERR) { \
+                       usnic_printk(KERN_ERR, args); \
+               } \
+       } while (0)
+#endif /* !USNIC_LOG_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_transport.c b/drivers/infiniband/hw/usnic/usnic_transport.c
new file mode 100644 (file)
index 0000000..723bd6c
--- /dev/null
@@ -0,0 +1,125 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/bitmap.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+
+#include "usnic_transport.h"
+#include "usnic_log.h"
+
+/* ROCE */
+static unsigned long *roce_bitmap;
+static u16 roce_next_port = 1;
+#define ROCE_BITMAP_SZ ((1 << (8 /*CHAR_BIT*/ * sizeof(u16)))/8 /*CHAR BIT*/)
+static DEFINE_SPINLOCK(roce_bitmap_lock);
+
+static const char *transport_to_str(enum usnic_transport_type type)
+{
+       switch (type) {
+       case USNIC_TRANSPORT_UNKNOWN:
+               return "Unknown";
+       case USNIC_TRANSPORT_ROCE_CUSTOM:
+               return "roce custom";
+       case USNIC_TRANSPORT_MAX:
+               return "Max?";
+       default:
+               return "Not known";
+       }
+}
+
+/*
+ * reserve a port number.  if "0" specified, we will try to pick one
+ * starting at roce_next_port.  roce_next_port will take on the values
+ * 1..4096
+ */
+u16 usnic_transport_rsrv_port(enum usnic_transport_type type, u16 port_num)
+{
+       if (type == USNIC_TRANSPORT_ROCE_CUSTOM) {
+               spin_lock(&roce_bitmap_lock);
+               if (!port_num) {
+                       port_num = bitmap_find_next_zero_area(roce_bitmap,
+                                               ROCE_BITMAP_SZ,
+                                               roce_next_port /* start */,
+                                               1 /* nr */,
+                                               0 /* align */);
+                       roce_next_port = (port_num & 4095) + 1;
+               } else if (test_bit(port_num, roce_bitmap)) {
+                       usnic_err("Failed to allocate port for %s\n",
+                                       transport_to_str(type));
+                       spin_unlock(&roce_bitmap_lock);
+                       goto out_fail;
+               }
+               bitmap_set(roce_bitmap, port_num, 1);
+               spin_unlock(&roce_bitmap_lock);
+       } else {
+               usnic_err("Failed to allocate port - transport %s unsupported\n",
+                               transport_to_str(type));
+               goto out_fail;
+       }
+
+       usnic_dbg("Allocating port %hu for %s\n", port_num,
+                       transport_to_str(type));
+       return port_num;
+
+out_fail:
+       return 0;
+}
+
+void usnic_transport_unrsrv_port(enum usnic_transport_type type, u16 port_num)
+{
+       if (type == USNIC_TRANSPORT_ROCE_CUSTOM) {
+               spin_lock(&roce_bitmap_lock);
+               if (!port_num) {
+                       usnic_err("Unreserved unvalid port num 0 for %s\n",
+                                       transport_to_str(type));
+                       goto out_roce_custom;
+               }
+
+               if (!test_bit(port_num, roce_bitmap)) {
+                       usnic_err("Unreserving invalid %hu for %s\n",
+                                       port_num,
+                                       transport_to_str(type));
+                       goto out_roce_custom;
+               }
+               bitmap_clear(roce_bitmap, port_num, 1);
+               usnic_dbg("Freeing port %hu for %s\n", port_num,
+                               transport_to_str(type));
+out_roce_custom:
+               spin_unlock(&roce_bitmap_lock);
+       } else {
+               usnic_err("Freeing invalid port %hu for %d\n", port_num, type);
+       }
+}
+
+int usnic_transport_init(void)
+{
+       roce_bitmap = kzalloc(ROCE_BITMAP_SZ, GFP_KERNEL);
+       if (!roce_bitmap) {
+               usnic_err("Failed to allocate bit map");
+               return -ENOMEM;
+       }
+
+       /* Do not ever allocate bit 0, hence set it here */
+       bitmap_set(roce_bitmap, 0, 1);
+       return 0;
+}
+
+void usnic_transport_fini(void)
+{
+       kfree(roce_bitmap);
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_transport.h b/drivers/infiniband/hw/usnic/usnic_transport.h
new file mode 100644 (file)
index 0000000..091fdaf
--- /dev/null
@@ -0,0 +1,28 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_TRANSPORT_H_
+#define USNIC_TRANSPORT_H_
+
+#include "usnic_abi.h"
+
+u16 usnic_transport_rsrv_port(enum usnic_transport_type type, u16 port_num);
+void usnic_transport_unrsrv_port(enum usnic_transport_type type, u16 port_num);
+int usnic_transport_init(void);
+void usnic_transport_fini(void);
+#endif /* !USNIC_TRANSPORT_H */
diff --git a/drivers/infiniband/hw/usnic/usnic_uiom.c b/drivers/infiniband/hw/usnic/usnic_uiom.c
new file mode 100644 (file)
index 0000000..c841a75
--- /dev/null
@@ -0,0 +1,603 @@
+/*
+ * Copyright (c) 2005 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
+ * Copyright (c) 2013 Cisco Systems.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/mm.h>
+#include <linux/dma-mapping.h>
+#include <linux/sched.h>
+#include <linux/hugetlb.h>
+#include <linux/dma-attrs.h>
+#include <linux/iommu.h>
+#include <linux/workqueue.h>
+#include <linux/list.h>
+#include <linux/pci.h>
+
+#include "usnic_log.h"
+#include "usnic_uiom.h"
+#include "usnic_uiom_interval_tree.h"
+
+static struct workqueue_struct *usnic_uiom_wq;
+
+#define USNIC_UIOM_PAGE_CHUNK                                          \
+       ((PAGE_SIZE - offsetof(struct usnic_uiom_chunk, page_list))     /\
+       ((void *) &((struct usnic_uiom_chunk *) 0)->page_list[1] -      \
+       (void *) &((struct usnic_uiom_chunk *) 0)->page_list[0]))
+
+static void usnic_uiom_reg_account(struct work_struct *work)
+{
+       struct usnic_uiom_reg *umem = container_of(work,
+                                               struct usnic_uiom_reg, work);
+
+       down_write(&umem->mm->mmap_sem);
+       umem->mm->locked_vm -= umem->diff;
+       up_write(&umem->mm->mmap_sem);
+       mmput(umem->mm);
+       kfree(umem);
+}
+
+static int usnic_uiom_dma_fault(struct iommu_domain *domain,
+                               struct device *dev,
+                               unsigned long iova, int flags,
+                               void *token)
+{
+       usnic_err("Device %s iommu fault domain 0x%pK va 0x%lx flags 0x%x\n",
+               dev_name(dev),
+               domain, iova, flags);
+       return -ENOSYS;
+}
+
+static void usnic_uiom_put_pages(struct list_head *chunk_list, int dirty)
+{
+       struct usnic_uiom_chunk *chunk, *tmp;
+       struct page *page;
+       int i;
+       dma_addr_t pa;
+
+       list_for_each_entry_safe(chunk, tmp, chunk_list, list) {
+               for (i = 0; i < chunk->nents; i++) {
+                       page = sg_page(&chunk->page_list[i]);
+                       pa = sg_phys(&chunk->page_list[i]);
+                       if (dirty)
+                               set_page_dirty_lock(page);
+                       put_page(page);
+                       usnic_dbg("pa: %pa\n", &pa);
+               }
+               kfree(chunk);
+       }
+}
+
+static int usnic_uiom_get_pages(unsigned long addr, size_t size, int writable,
+                               int dmasync, struct list_head *chunk_list)
+{
+       struct page **page_list;
+       struct usnic_uiom_chunk *chunk;
+       unsigned long locked;
+       unsigned long lock_limit;
+       unsigned long cur_base;
+       unsigned long npages;
+       int ret;
+       int off;
+       int i;
+       int flags;
+       dma_addr_t pa;
+       DEFINE_DMA_ATTRS(attrs);
+
+       if (dmasync)
+               dma_set_attr(DMA_ATTR_WRITE_BARRIER, &attrs);
+
+       if (!can_do_mlock())
+               return -EPERM;
+
+       INIT_LIST_HEAD(chunk_list);
+
+       page_list = (struct page **) __get_free_page(GFP_KERNEL);
+       if (!page_list)
+               return -ENOMEM;
+
+       npages = PAGE_ALIGN(size + (addr & ~PAGE_MASK)) >> PAGE_SHIFT;
+
+       down_write(&current->mm->mmap_sem);
+
+       locked = npages + current->mm->locked_vm;
+       lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
+
+       if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) {
+               ret = -ENOMEM;
+               goto out;
+       }
+
+       flags = IOMMU_READ | IOMMU_CACHE;
+       flags |= (writable) ? IOMMU_WRITE : 0;
+       cur_base = addr & PAGE_MASK;
+       ret = 0;
+
+       while (npages) {
+               ret = get_user_pages(current, current->mm, cur_base,
+                                       min_t(unsigned long, npages,
+                                       PAGE_SIZE / sizeof(struct page *)),
+                                       1, !writable, page_list, NULL);
+
+               if (ret < 0)
+                       goto out;
+
+               npages -= ret;
+               off = 0;
+
+               while (ret) {
+                       chunk = kmalloc(sizeof(*chunk) +
+                                       sizeof(struct scatterlist) *
+                                       min_t(int, ret, USNIC_UIOM_PAGE_CHUNK),
+                                       GFP_KERNEL);
+                       if (!chunk) {
+                               ret = -ENOMEM;
+                               goto out;
+                       }
+
+                       chunk->nents = min_t(int, ret, USNIC_UIOM_PAGE_CHUNK);
+                       sg_init_table(chunk->page_list, chunk->nents);
+                       for (i = 0; i < chunk->nents; ++i) {
+                               sg_set_page(&chunk->page_list[i],
+                                                       page_list[i + off],
+                                                       PAGE_SIZE, 0);
+                               pa = sg_phys(&chunk->page_list[i]);
+                               usnic_dbg("va: 0x%lx pa: %pa\n",
+                                               cur_base + i*PAGE_SIZE, &pa);
+                       }
+                       cur_base += chunk->nents * PAGE_SIZE;
+                       ret -= chunk->nents;
+                       off += chunk->nents;
+                       list_add_tail(&chunk->list, chunk_list);
+               }
+
+               ret = 0;
+       }
+
+out:
+       if (ret < 0)
+               usnic_uiom_put_pages(chunk_list, 0);
+       else
+               current->mm->locked_vm = locked;
+
+       up_write(&current->mm->mmap_sem);
+       free_page((unsigned long) page_list);
+       return ret;
+}
+
+static void usnic_uiom_unmap_sorted_intervals(struct list_head *intervals,
+                                               struct usnic_uiom_pd *pd)
+{
+       struct usnic_uiom_interval_node *interval, *tmp;
+       long unsigned va, size;
+
+       list_for_each_entry_safe(interval, tmp, intervals, link) {
+               va = interval->start << PAGE_SHIFT;
+               size = ((interval->last - interval->start) + 1) << PAGE_SHIFT;
+               while (size > 0) {
+                       /* Workaround for RH 970401 */
+                       usnic_dbg("va 0x%lx size 0x%lx", va, PAGE_SIZE);
+                       iommu_unmap(pd->domain, va, PAGE_SIZE);
+                       va += PAGE_SIZE;
+                       size -= PAGE_SIZE;
+               }
+       }
+}
+
+static void __usnic_uiom_reg_release(struct usnic_uiom_pd *pd,
+                                       struct usnic_uiom_reg *uiomr,
+                                       int dirty)
+{
+       int npages;
+       unsigned long vpn_start, vpn_last;
+       struct usnic_uiom_interval_node *interval, *tmp;
+       int writable = 0;
+       LIST_HEAD(rm_intervals);
+
+       npages = PAGE_ALIGN(uiomr->length + uiomr->offset) >> PAGE_SHIFT;
+       vpn_start = (uiomr->va & PAGE_MASK) >> PAGE_SHIFT;
+       vpn_last = vpn_start + npages - 1;
+
+       spin_lock(&pd->lock);
+       usnic_uiom_remove_interval(&pd->rb_root, vpn_start,
+                                       vpn_last, &rm_intervals);
+       usnic_uiom_unmap_sorted_intervals(&rm_intervals, pd);
+
+       list_for_each_entry_safe(interval, tmp, &rm_intervals, link) {
+               if (interval->flags & IOMMU_WRITE)
+                       writable = 1;
+               list_del(&interval->link);
+               kfree(interval);
+       }
+
+       usnic_uiom_put_pages(&uiomr->chunk_list, dirty & writable);
+       spin_unlock(&pd->lock);
+}
+
+static int usnic_uiom_map_sorted_intervals(struct list_head *intervals,
+                                               struct usnic_uiom_reg *uiomr)
+{
+       int i, err;
+       size_t size;
+       struct usnic_uiom_chunk *chunk;
+       struct usnic_uiom_interval_node *interval_node;
+       dma_addr_t pa;
+       dma_addr_t pa_start = 0;
+       dma_addr_t pa_end = 0;
+       long int va_start = -EINVAL;
+       struct usnic_uiom_pd *pd = uiomr->pd;
+       long int va = uiomr->va & PAGE_MASK;
+       int flags = IOMMU_READ | IOMMU_CACHE;
+
+       flags |= (uiomr->writable) ? IOMMU_WRITE : 0;
+       chunk = list_first_entry(&uiomr->chunk_list, struct usnic_uiom_chunk,
+                                                                       list);
+       list_for_each_entry(interval_node, intervals, link) {
+iter_chunk:
+               for (i = 0; i < chunk->nents; i++, va += PAGE_SIZE) {
+                       pa = sg_phys(&chunk->page_list[i]);
+                       if ((va >> PAGE_SHIFT) < interval_node->start)
+                               continue;
+
+                       if ((va >> PAGE_SHIFT) == interval_node->start) {
+                               /* First page of the interval */
+                               va_start = va;
+                               pa_start = pa;
+                               pa_end = pa;
+                       }
+
+                       WARN_ON(va_start == -EINVAL);
+
+                       if ((pa_end + PAGE_SIZE != pa) &&
+                                       (pa != pa_start)) {
+                               /* PAs are not contiguous */
+                               size = pa_end - pa_start + PAGE_SIZE;
+                               usnic_dbg("va 0x%lx pa %pa size 0x%zx flags 0x%x",
+                                       va_start, &pa_start, size, flags);
+                               err = iommu_map(pd->domain, va_start, pa_start,
+                                                       size, flags);
+                               if (err) {
+                                       usnic_err("Failed to map va 0x%lx pa 0x%pa size 0x%zx with err %d\n",
+                                               va_start, &pa_start, size, err);
+                                       goto err_out;
+                               }
+                               va_start = va;
+                               pa_start = pa;
+                               pa_end = pa;
+                       }
+
+                       if ((va >> PAGE_SHIFT) == interval_node->last) {
+                               /* Last page of the interval */
+                               size = pa - pa_start + PAGE_SIZE;
+                               usnic_dbg("va 0x%lx pa %pa size 0x%zx flags 0x%x\n",
+                                       va_start, &pa_start, size, flags);
+                               err = iommu_map(pd->domain, va_start, pa_start,
+                                               size, flags);
+                               if (err) {
+                                       usnic_err("Failed to map va 0x%lx pa %pa size 0x%zx with err %d\n",
+                                               va_start, &pa_start, size, err);
+                                       goto err_out;
+                               }
+                               break;
+                       }
+
+                       if (pa != pa_start)
+                               pa_end += PAGE_SIZE;
+               }
+
+               if (i == chunk->nents) {
+                       /*
+                        * Hit last entry of the chunk,
+                        * hence advance to next chunk
+                        */
+                       chunk = list_first_entry(&chunk->list,
+                                                       struct usnic_uiom_chunk,
+                                                       list);
+                       goto iter_chunk;
+               }
+       }
+
+       return 0;
+
+err_out:
+       usnic_uiom_unmap_sorted_intervals(intervals, pd);
+       return err;
+}
+
+struct usnic_uiom_reg *usnic_uiom_reg_get(struct usnic_uiom_pd *pd,
+                                               unsigned long addr, size_t size,
+                                               int writable, int dmasync)
+{
+       struct usnic_uiom_reg *uiomr;
+       unsigned long va_base, vpn_start, vpn_last;
+       unsigned long npages;
+       int offset, err;
+       LIST_HEAD(sorted_diff_intervals);
+
+       /*
+        * Intel IOMMU map throws an error if a translation entry is
+        * changed from read to write.  This module may not unmap
+        * and then remap the entry after fixing the permission
+        * b/c this open up a small windows where hw DMA may page fault
+        * Hence, make all entries to be writable.
+        */
+       writable = 1;
+
+       va_base = addr & PAGE_MASK;
+       offset = addr & ~PAGE_MASK;
+       npages = PAGE_ALIGN(size + offset) >> PAGE_SHIFT;
+       vpn_start = (addr & PAGE_MASK) >> PAGE_SHIFT;
+       vpn_last = vpn_start + npages - 1;
+
+       uiomr = kmalloc(sizeof(*uiomr), GFP_KERNEL);
+       if (!uiomr)
+               return ERR_PTR(-ENOMEM);
+
+       uiomr->va = va_base;
+       uiomr->offset = offset;
+       uiomr->length = size;
+       uiomr->writable = writable;
+       uiomr->pd = pd;
+
+       err = usnic_uiom_get_pages(addr, size, writable, dmasync,
+                                       &uiomr->chunk_list);
+       if (err) {
+               usnic_err("Failed get_pages vpn [0x%lx,0x%lx] err %d\n",
+                               vpn_start, vpn_last, err);
+               goto out_free_uiomr;
+       }
+
+       spin_lock(&pd->lock);
+       err = usnic_uiom_get_intervals_diff(vpn_start, vpn_last,
+                                               (writable) ? IOMMU_WRITE : 0,
+                                               IOMMU_WRITE,
+                                               &pd->rb_root,
+                                               &sorted_diff_intervals);
+       if (err) {
+               usnic_err("Failed disjoint interval vpn [0x%lx,0x%lx] err %d\n",
+                                               vpn_start, vpn_last, err);
+               goto out_put_pages;
+       }
+
+       err = usnic_uiom_map_sorted_intervals(&sorted_diff_intervals, uiomr);
+       if (err) {
+               usnic_err("Failed map interval vpn [0x%lx,0x%lx] err %d\n",
+                                               vpn_start, vpn_last, err);
+               goto out_put_intervals;
+
+       }
+
+       err = usnic_uiom_insert_interval(&pd->rb_root, vpn_start, vpn_last,
+                                       (writable) ? IOMMU_WRITE : 0);
+       if (err) {
+               usnic_err("Failed insert interval vpn [0x%lx,0x%lx] err %d\n",
+                                               vpn_start, vpn_last, err);
+               goto out_unmap_intervals;
+       }
+
+       usnic_uiom_put_interval_set(&sorted_diff_intervals);
+       spin_unlock(&pd->lock);
+
+       return uiomr;
+
+out_unmap_intervals:
+       usnic_uiom_unmap_sorted_intervals(&sorted_diff_intervals, pd);
+out_put_intervals:
+       usnic_uiom_put_interval_set(&sorted_diff_intervals);
+out_put_pages:
+       usnic_uiom_put_pages(&uiomr->chunk_list, 0);
+       spin_unlock(&pd->lock);
+out_free_uiomr:
+       kfree(uiomr);
+       return ERR_PTR(err);
+}
+
+void usnic_uiom_reg_release(struct usnic_uiom_reg *uiomr, int closing)
+{
+       struct mm_struct *mm;
+       unsigned long diff;
+
+       __usnic_uiom_reg_release(uiomr->pd, uiomr, 1);
+
+       mm = get_task_mm(current);
+       if (!mm) {
+               kfree(uiomr);
+               return;
+       }
+
+       diff = PAGE_ALIGN(uiomr->length + uiomr->offset) >> PAGE_SHIFT;
+
+       /*
+        * We may be called with the mm's mmap_sem already held.  This
+        * can happen when a userspace munmap() is the call that drops
+        * the last reference to our file and calls our release
+        * method.  If there are memory regions to destroy, we'll end
+        * up here and not be able to take the mmap_sem.  In that case
+        * we defer the vm_locked accounting to the system workqueue.
+        */
+       if (closing) {
+               if (!down_write_trylock(&mm->mmap_sem)) {
+                       INIT_WORK(&uiomr->work, usnic_uiom_reg_account);
+                       uiomr->mm = mm;
+                       uiomr->diff = diff;
+
+                       queue_work(usnic_uiom_wq, &uiomr->work);
+                       return;
+               }
+       } else
+               down_write(&mm->mmap_sem);
+
+       current->mm->locked_vm -= diff;
+       up_write(&mm->mmap_sem);
+       mmput(mm);
+       kfree(uiomr);
+}
+
+struct usnic_uiom_pd *usnic_uiom_alloc_pd(void)
+{
+       struct usnic_uiom_pd *pd;
+       void *domain;
+
+       pd = kzalloc(sizeof(*pd), GFP_KERNEL);
+       if (!pd)
+               return ERR_PTR(-ENOMEM);
+
+       pd->domain = domain = iommu_domain_alloc(&pci_bus_type);
+       if (IS_ERR_OR_NULL(domain)) {
+               usnic_err("Failed to allocate IOMMU domain with err %ld\n",
+                               PTR_ERR(pd->domain));
+               kfree(pd);
+               return ERR_PTR(domain ? PTR_ERR(domain) : -ENOMEM);
+       }
+
+       iommu_set_fault_handler(pd->domain, usnic_uiom_dma_fault, NULL);
+
+       spin_lock_init(&pd->lock);
+       INIT_LIST_HEAD(&pd->devs);
+
+       return pd;
+}
+
+void usnic_uiom_dealloc_pd(struct usnic_uiom_pd *pd)
+{
+       iommu_domain_free(pd->domain);
+       kfree(pd);
+}
+
+int usnic_uiom_attach_dev_to_pd(struct usnic_uiom_pd *pd, struct device *dev)
+{
+       struct usnic_uiom_dev *uiom_dev;
+       int err;
+
+       uiom_dev = kzalloc(sizeof(*uiom_dev), GFP_KERNEL);
+       if (!uiom_dev)
+               return -ENOMEM;
+       uiom_dev->dev = dev;
+
+       err = iommu_attach_device(pd->domain, dev);
+       if (err)
+               goto out_free_dev;
+
+       if (!iommu_domain_has_cap(pd->domain, IOMMU_CAP_CACHE_COHERENCY)) {
+               usnic_err("IOMMU of %s does not support cache coherency\n",
+                               dev_name(dev));
+               err = -EINVAL;
+               goto out_detach_device;
+       }
+
+       spin_lock(&pd->lock);
+       list_add_tail(&uiom_dev->link, &pd->devs);
+       pd->dev_cnt++;
+       spin_unlock(&pd->lock);
+
+       return 0;
+
+out_detach_device:
+       iommu_detach_device(pd->domain, dev);
+out_free_dev:
+       kfree(uiom_dev);
+       return err;
+}
+
+void usnic_uiom_detach_dev_from_pd(struct usnic_uiom_pd *pd, struct device *dev)
+{
+       struct usnic_uiom_dev *uiom_dev;
+       int found = 0;
+
+       spin_lock(&pd->lock);
+       list_for_each_entry(uiom_dev, &pd->devs, link) {
+               if (uiom_dev->dev == dev) {
+                       found = 1;
+                       break;
+               }
+       }
+
+       if (!found) {
+               usnic_err("Unable to free dev %s - not found\n",
+                               dev_name(dev));
+               spin_unlock(&pd->lock);
+               return;
+       }
+
+       list_del(&uiom_dev->link);
+       pd->dev_cnt--;
+       spin_unlock(&pd->lock);
+
+       return iommu_detach_device(pd->domain, dev);
+}
+
+struct device **usnic_uiom_get_dev_list(struct usnic_uiom_pd *pd)
+{
+       struct usnic_uiom_dev *uiom_dev;
+       struct device **devs;
+       int i = 0;
+
+       spin_lock(&pd->lock);
+       devs = kcalloc(pd->dev_cnt + 1, sizeof(*devs), GFP_ATOMIC);
+       if (!devs) {
+               devs = ERR_PTR(-ENOMEM);
+               goto out;
+       }
+
+       list_for_each_entry(uiom_dev, &pd->devs, link) {
+               devs[i++] = uiom_dev->dev;
+       }
+out:
+       spin_unlock(&pd->lock);
+       return devs;
+}
+
+void usnic_uiom_free_dev_list(struct device **devs)
+{
+       kfree(devs);
+}
+
+int usnic_uiom_init(char *drv_name)
+{
+       if (!iommu_present(&pci_bus_type)) {
+               usnic_err("IOMMU required but not present or enabled.  USNIC QPs will not function w/o enabling IOMMU\n");
+               return -EPERM;
+       }
+
+       usnic_uiom_wq = create_workqueue(drv_name);
+       if (!usnic_uiom_wq) {
+               usnic_err("Unable to alloc wq for drv %s\n", drv_name);
+               return -ENOMEM;
+       }
+
+       return 0;
+}
+
+void usnic_uiom_fini(void)
+{
+       flush_workqueue(usnic_uiom_wq);
+       destroy_workqueue(usnic_uiom_wq);
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_uiom.h b/drivers/infiniband/hw/usnic/usnic_uiom.h
new file mode 100644 (file)
index 0000000..7044099
--- /dev/null
@@ -0,0 +1,80 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_UIOM_H_
+#define USNIC_UIOM_H_
+
+#include <linux/list.h>
+#include <linux/scatterlist.h>
+
+#include "usnic_uiom_interval_tree.h"
+
+#define USNIC_UIOM_READ                        (1)
+#define USNIC_UIOM_WRITE               (2)
+
+#define USNIC_UIOM_MAX_PD_CNT          (1000)
+#define USNIC_UIOM_MAX_MR_CNT          (1000000)
+#define USNIC_UIOM_MAX_MR_SIZE         (~0UL)
+#define USNIC_UIOM_PAGE_SIZE           (PAGE_SIZE)
+
+struct usnic_uiom_dev {
+       struct device                   *dev;
+       struct list_head                link;
+};
+
+struct usnic_uiom_pd {
+       struct iommu_domain             *domain;
+       spinlock_t                      lock;
+       struct rb_root                  rb_root;
+       struct list_head                devs;
+       int                             dev_cnt;
+};
+
+struct usnic_uiom_reg {
+       struct usnic_uiom_pd            *pd;
+       unsigned long                   va;
+       size_t                          length;
+       int                             offset;
+       int                             page_size;
+       int                             writable;
+       struct list_head                chunk_list;
+       struct work_struct              work;
+       struct mm_struct                *mm;
+       unsigned long                   diff;
+};
+
+struct usnic_uiom_chunk {
+       struct list_head                list;
+       int                             nents;
+       struct scatterlist              page_list[0];
+};
+
+struct usnic_uiom_pd *usnic_uiom_alloc_pd(void);
+void usnic_uiom_dealloc_pd(struct usnic_uiom_pd *pd);
+int usnic_uiom_attach_dev_to_pd(struct usnic_uiom_pd *pd, struct device *dev);
+void usnic_uiom_detach_dev_from_pd(struct usnic_uiom_pd *pd,
+                                       struct device *dev);
+struct device **usnic_uiom_get_dev_list(struct usnic_uiom_pd *pd);
+void usnic_uiom_free_dev_list(struct device **devs);
+struct usnic_uiom_reg *usnic_uiom_reg_get(struct usnic_uiom_pd *pd,
+                                               unsigned long addr, size_t size,
+                                               int access, int dmasync);
+void usnic_uiom_reg_release(struct usnic_uiom_reg *uiomr, int closing);
+int usnic_uiom_init(char *drv_name);
+void usnic_uiom_fini(void);
+#endif /* USNIC_UIOM_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_uiom_interval_tree.c b/drivers/infiniband/hw/usnic/usnic_uiom_interval_tree.c
new file mode 100644 (file)
index 0000000..7e1dafc
--- /dev/null
@@ -0,0 +1,237 @@
+#include <linux/init.h>
+#include <linux/list.h>
+#include <linux/slab.h>
+#include <linux/list_sort.h>
+#include <linux/version.h>
+
+#include <linux/interval_tree_generic.h>
+#include "usnic_uiom_interval_tree.h"
+
+#define START(node) ((node)->start)
+#define LAST(node) ((node)->last)
+
+#define MAKE_NODE(node, start, end, ref_cnt, flags, err, err_out)      \
+               do {                                                    \
+                       node = usnic_uiom_interval_node_alloc(start,    \
+                                       end, ref_cnt, flags);           \
+                               if (!node) {                            \
+                                       err = -ENOMEM;                  \
+                                       goto err_out;                   \
+                               }                                       \
+               } while (0)
+
+#define MARK_FOR_ADD(node, list) (list_add_tail(&node->link, list))
+
+#define MAKE_NODE_AND_APPEND(node, start, end, ref_cnt, flags, err,    \
+                               err_out, list)                          \
+                               do {                                    \
+                                       MAKE_NODE(node, start, end,     \
+                                               ref_cnt, flags, err,    \
+                                               err_out);               \
+                                       MARK_FOR_ADD(node, list);       \
+                               } while (0)
+
+#define FLAGS_EQUAL(flags1, flags2, mask)                              \
+                       (((flags1) & (mask)) == ((flags2) & (mask)))
+
+static struct usnic_uiom_interval_node*
+usnic_uiom_interval_node_alloc(long int start, long int last, int ref_cnt,
+                               int flags)
+{
+       struct usnic_uiom_interval_node *interval = kzalloc(sizeof(*interval),
+                                                               GFP_ATOMIC);
+       if (!interval)
+               return NULL;
+
+       interval->start = start;
+       interval->last = last;
+       interval->flags = flags;
+       interval->ref_cnt = ref_cnt;
+
+       return interval;
+}
+
+static int interval_cmp(void *priv, struct list_head *a, struct list_head *b)
+{
+       struct usnic_uiom_interval_node *node_a, *node_b;
+
+       node_a = list_entry(a, struct usnic_uiom_interval_node, link);
+       node_b = list_entry(b, struct usnic_uiom_interval_node, link);
+
+       /* long to int */
+       if (node_a->start < node_b->start)
+               return -1;
+       else if (node_a->start > node_b->start)
+               return 1;
+
+       return 0;
+}
+
+static void
+find_intervals_intersection_sorted(struct rb_root *root, unsigned long start,
+                                       unsigned long last,
+                                       struct list_head *list)
+{
+       struct usnic_uiom_interval_node *node;
+
+       INIT_LIST_HEAD(list);
+
+       for (node = usnic_uiom_interval_tree_iter_first(root, start, last);
+               node;
+               node = usnic_uiom_interval_tree_iter_next(node, start, last))
+               list_add_tail(&node->link, list);
+
+       list_sort(NULL, list, interval_cmp);
+}
+
+int usnic_uiom_get_intervals_diff(unsigned long start, unsigned long last,
+                                       int flags, int flag_mask,
+                                       struct rb_root *root,
+                                       struct list_head *diff_set)
+{
+       struct usnic_uiom_interval_node *interval, *tmp;
+       int err = 0;
+       long int pivot = start;
+       LIST_HEAD(intersection_set);
+
+       INIT_LIST_HEAD(diff_set);
+
+       find_intervals_intersection_sorted(root, start, last,
+                                               &intersection_set);
+
+       list_for_each_entry(interval, &intersection_set, link) {
+               if (pivot < interval->start) {
+                       MAKE_NODE_AND_APPEND(tmp, pivot, interval->start - 1,
+                                               1, flags, err, err_out,
+                                               diff_set);
+                       pivot = interval->start;
+               }
+
+               /*
+                * Invariant: Set [start, pivot] is either in diff_set or root,
+                * but not in both.
+                */
+
+               if (pivot > interval->last) {
+                       continue;
+               } else if (pivot <= interval->last &&
+                               FLAGS_EQUAL(interval->flags, flags,
+                               flag_mask)) {
+                       pivot = interval->last + 1;
+               }
+       }
+
+       if (pivot <= last)
+               MAKE_NODE_AND_APPEND(tmp, pivot, last, 1, flags, err, err_out,
+                                       diff_set);
+
+       return 0;
+
+err_out:
+       list_for_each_entry_safe(interval, tmp, diff_set, link) {
+               list_del(&interval->link);
+               kfree(interval);
+       }
+
+       return err;
+}
+
+void usnic_uiom_put_interval_set(struct list_head *intervals)
+{
+       struct usnic_uiom_interval_node *interval, *tmp;
+       list_for_each_entry_safe(interval, tmp, intervals, link)
+               kfree(interval);
+}
+
+int usnic_uiom_insert_interval(struct rb_root *root, unsigned long start,
+                               unsigned long last, int flags)
+{
+       struct usnic_uiom_interval_node *interval, *tmp;
+       unsigned long istart, ilast;
+       int iref_cnt, iflags;
+       unsigned long lpivot = start;
+       int err = 0;
+       LIST_HEAD(to_add);
+       LIST_HEAD(intersection_set);
+
+       find_intervals_intersection_sorted(root, start, last,
+                                               &intersection_set);
+
+       list_for_each_entry(interval, &intersection_set, link) {
+               /*
+                * Invariant - lpivot is the left edge of next interval to be
+                * inserted
+                */
+               istart = interval->start;
+               ilast = interval->last;
+               iref_cnt = interval->ref_cnt;
+               iflags = interval->flags;
+
+               if (istart < lpivot) {
+                       MAKE_NODE_AND_APPEND(tmp, istart, lpivot - 1, iref_cnt,
+                                               iflags, err, err_out, &to_add);
+               } else if (istart > lpivot) {
+                       MAKE_NODE_AND_APPEND(tmp, lpivot, istart - 1, 1, flags,
+                                               err, err_out, &to_add);
+                       lpivot = istart;
+               } else {
+                       lpivot = istart;
+               }
+
+               if (ilast > last) {
+                       MAKE_NODE_AND_APPEND(tmp, lpivot, last, iref_cnt + 1,
+                                               iflags | flags, err, err_out,
+                                               &to_add);
+                       MAKE_NODE_AND_APPEND(tmp, last + 1, ilast, iref_cnt,
+                                               iflags, err, err_out, &to_add);
+               } else {
+                       MAKE_NODE_AND_APPEND(tmp, lpivot, ilast, iref_cnt + 1,
+                                               iflags | flags, err, err_out,
+                                               &to_add);
+               }
+
+               lpivot = ilast + 1;
+       }
+
+       if (lpivot <= last)
+               MAKE_NODE_AND_APPEND(tmp, lpivot, last, 1, flags, err, err_out,
+                                       &to_add);
+
+       list_for_each_entry_safe(interval, tmp, &intersection_set, link) {
+               usnic_uiom_interval_tree_remove(interval, root);
+               kfree(interval);
+       }
+
+       list_for_each_entry(interval, &to_add, link)
+               usnic_uiom_interval_tree_insert(interval, root);
+
+       return 0;
+
+err_out:
+       list_for_each_entry_safe(interval, tmp, &to_add, link)
+               kfree(interval);
+
+       return err;
+}
+
+void usnic_uiom_remove_interval(struct rb_root *root, unsigned long start,
+                               unsigned long last, struct list_head *removed)
+{
+       struct usnic_uiom_interval_node *interval;
+
+       for (interval = usnic_uiom_interval_tree_iter_first(root, start, last);
+                       interval;
+                       interval = usnic_uiom_interval_tree_iter_next(interval,
+                                                                       start,
+                                                                       last)) {
+               if (--interval->ref_cnt == 0)
+                       list_add_tail(&interval->link, removed);
+       }
+
+       list_for_each_entry(interval, removed, link)
+               usnic_uiom_interval_tree_remove(interval, root);
+}
+
+INTERVAL_TREE_DEFINE(struct usnic_uiom_interval_node, rb,
+                       unsigned long, __subtree_last,
+                       START, LAST, , usnic_uiom_interval_tree)
diff --git a/drivers/infiniband/hw/usnic/usnic_uiom_interval_tree.h b/drivers/infiniband/hw/usnic/usnic_uiom_interval_tree.h
new file mode 100644 (file)
index 0000000..030ba6e
--- /dev/null
@@ -0,0 +1,74 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_UIOM_INTERVAL_TREE_H_
+#define USNIC_UIOM_INTERVAL_TREE_H_
+
+#include <linux/version.h>
+#include <linux/rbtree.h>
+
+struct usnic_uiom_interval_node {
+       struct rb_node                  rb;
+       struct list_head                link;
+       unsigned long                   start;
+       unsigned long                   last;
+       unsigned long                   __subtree_last;
+       unsigned int                    ref_cnt;
+       int                             flags;
+};
+
+extern void
+usnic_uiom_interval_tree_insert(struct usnic_uiom_interval_node *node,
+                                       struct rb_root *root);
+extern void
+usnic_uiom_interval_tree_remove(struct usnic_uiom_interval_node *node,
+                                       struct rb_root *root);
+extern struct usnic_uiom_interval_node *
+usnic_uiom_interval_tree_iter_first(struct rb_root *root,
+                                       unsigned long start,
+                                       unsigned long last);
+extern struct usnic_uiom_interval_node *
+usnic_uiom_interval_tree_iter_next(struct usnic_uiom_interval_node *node,
+                       unsigned long start, unsigned long last);
+/*
+ * Inserts {start...last} into {root}.  If there are overlaps,
+ * nodes will be broken up and merged
+ */
+int usnic_uiom_insert_interval(struct rb_root *root,
+                               unsigned long start, unsigned long last,
+                               int flags);
+/*
+ * Removed {start...last} from {root}.  The nodes removed are returned in
+ * 'removed.' The caller is responsibile for freeing memory of nodes in
+ * 'removed.'
+ */
+void usnic_uiom_remove_interval(struct rb_root *root,
+                               unsigned long start, unsigned long last,
+                               struct list_head *removed);
+/*
+ * Returns {start...last} - {root} (relative complement of {start...last} in
+ * {root}) in diff_set sorted ascendingly
+ */
+int usnic_uiom_get_intervals_diff(unsigned long start,
+                                       unsigned long last, int flags,
+                                       int flag_mask,
+                                       struct rb_root *root,
+                                       struct list_head *diff_set);
+/* Call this to free diff_set returned by usnic_uiom_get_intervals_diff */
+void usnic_uiom_put_interval_set(struct list_head *intervals);
+#endif /* USNIC_UIOM_INTERVAL_TREE_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_vnic.c b/drivers/infiniband/hw/usnic/usnic_vnic.c
new file mode 100644 (file)
index 0000000..656b88c
--- /dev/null
@@ -0,0 +1,467 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/errno.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+
+#include "usnic_ib.h"
+#include "vnic_resource.h"
+#include "usnic_log.h"
+#include "usnic_vnic.h"
+
+struct usnic_vnic {
+       struct vnic_dev                 *vdev;
+       struct vnic_dev_bar             bar[PCI_NUM_RESOURCES];
+       struct usnic_vnic_res_chunk     chunks[USNIC_VNIC_RES_TYPE_MAX];
+       spinlock_t                      res_lock;
+};
+
+static enum vnic_res_type _to_vnic_res_type(enum usnic_vnic_res_type res_type)
+{
+#define DEFINE_USNIC_VNIC_RES_AT(usnic_vnic_res_t, vnic_res_type, desc, val) \
+               vnic_res_type,
+#define DEFINE_USNIC_VNIC_RES(usnic_vnic_res_t, vnic_res_type, desc) \
+               vnic_res_type,
+       static enum vnic_res_type usnic_vnic_type_2_vnic_type[] = {
+                                               USNIC_VNIC_RES_TYPES};
+#undef DEFINE_USNIC_VNIC_RES
+#undef DEFINE_USNIC_VNIC_RES_AT
+
+       if (res_type >= USNIC_VNIC_RES_TYPE_MAX)
+               return RES_TYPE_MAX;
+
+       return usnic_vnic_type_2_vnic_type[res_type];
+}
+
+const char *usnic_vnic_res_type_to_str(enum usnic_vnic_res_type res_type)
+{
+#define DEFINE_USNIC_VNIC_RES_AT(usnic_vnic_res_t, vnic_res_type, desc, val) \
+               desc,
+#define DEFINE_USNIC_VNIC_RES(usnic_vnic_res_t, vnic_res_type, desc) \
+               desc,
+       static const char * const usnic_vnic_res_type_desc[] = {
+                                               USNIC_VNIC_RES_TYPES};
+#undef DEFINE_USNIC_VNIC_RES
+#undef DEFINE_USNIC_VNIC_RES_AT
+
+       if (res_type >= USNIC_VNIC_RES_TYPE_MAX)
+               return "unknown";
+
+       return usnic_vnic_res_type_desc[res_type];
+
+}
+
+const char *usnic_vnic_pci_name(struct usnic_vnic *vnic)
+{
+       return pci_name(usnic_vnic_get_pdev(vnic));
+}
+
+int usnic_vnic_dump(struct usnic_vnic *vnic, char *buf,
+                       int buf_sz,
+                       void *hdr_obj,
+                       int (*printtitle)(void *, char*, int),
+                       int (*printcols)(char *, int),
+                       int (*printrow)(void *, char *, int))
+{
+       struct usnic_vnic_res_chunk *chunk;
+       struct usnic_vnic_res *res;
+       struct vnic_dev_bar *bar0;
+       int i, j, offset;
+
+       offset = 0;
+       bar0 = usnic_vnic_get_bar(vnic, 0);
+       offset += scnprintf(buf + offset, buf_sz - offset,
+                       "VF:%hu BAR0 bus_addr=%pa vaddr=0x%p size=%ld ",
+                       usnic_vnic_get_index(vnic),
+                       &bar0->bus_addr,
+                       bar0->vaddr, bar0->len);
+       if (printtitle)
+               offset += printtitle(hdr_obj, buf + offset, buf_sz - offset);
+       offset += scnprintf(buf + offset, buf_sz - offset, "\n");
+       offset += scnprintf(buf + offset, buf_sz - offset,
+                       "|RES\t|CTRL_PIN\t\t|IN_USE\t");
+       if (printcols)
+               offset += printcols(buf + offset, buf_sz - offset);
+       offset += scnprintf(buf + offset, buf_sz - offset, "\n");
+
+       spin_lock(&vnic->res_lock);
+       for (i = 0; i < ARRAY_SIZE(vnic->chunks); i++) {
+               chunk = &vnic->chunks[i];
+               for (j = 0; j < chunk->cnt; j++) {
+                       res = chunk->res[j];
+                       offset += scnprintf(buf + offset, buf_sz - offset,
+                                       "|%s[%u]\t|0x%p\t|%u\t",
+                                       usnic_vnic_res_type_to_str(res->type),
+                                       res->vnic_idx, res->ctrl, !!res->owner);
+                       if (printrow) {
+                               offset += printrow(res->owner, buf + offset,
+                                                       buf_sz - offset);
+                       }
+                       offset += scnprintf(buf + offset, buf_sz - offset,
+                                               "\n");
+               }
+       }
+       spin_unlock(&vnic->res_lock);
+       return offset;
+}
+
+void usnic_vnic_res_spec_update(struct usnic_vnic_res_spec *spec,
+                               enum usnic_vnic_res_type trgt_type,
+                               u16 cnt)
+{
+       int i;
+
+       for (i = 0; i < USNIC_VNIC_RES_TYPE_MAX; i++) {
+               if (spec->resources[i].type == trgt_type) {
+                       spec->resources[i].cnt = cnt;
+                       return;
+               }
+       }
+
+       WARN_ON(1);
+}
+
+int usnic_vnic_res_spec_satisfied(const struct usnic_vnic_res_spec *min_spec,
+                                       struct usnic_vnic_res_spec *res_spec)
+{
+       int found, i, j;
+
+       for (i = 0; i < USNIC_VNIC_RES_TYPE_MAX; i++) {
+               found = 0;
+
+               for (j = 0; j < USNIC_VNIC_RES_TYPE_MAX; j++) {
+                       if (res_spec->resources[i].type !=
+                               min_spec->resources[i].type)
+                               continue;
+                       found = 1;
+                       if (min_spec->resources[i].cnt >
+                                       res_spec->resources[i].cnt)
+                               return -EINVAL;
+                       break;
+               }
+
+               if (!found)
+                       return -EINVAL;
+       }
+       return 0;
+}
+
+int usnic_vnic_spec_dump(char *buf, int buf_sz,
+                               struct usnic_vnic_res_spec *res_spec)
+{
+       enum usnic_vnic_res_type res_type;
+       int res_cnt;
+       int i;
+       int offset = 0;
+
+       for (i = 0; i < USNIC_VNIC_RES_TYPE_MAX; i++) {
+               res_type = res_spec->resources[i].type;
+               res_cnt = res_spec->resources[i].cnt;
+               offset += scnprintf(buf + offset, buf_sz - offset,
+                               "Res: %s Cnt: %d ",
+                               usnic_vnic_res_type_to_str(res_type),
+                               res_cnt);
+       }
+
+       return offset;
+}
+
+int usnic_vnic_check_room(struct usnic_vnic *vnic,
+                               struct usnic_vnic_res_spec *res_spec)
+{
+       int i;
+       enum usnic_vnic_res_type res_type;
+       int res_cnt;
+
+       for (i = 0; i < USNIC_VNIC_RES_TYPE_MAX; i++) {
+               res_type = res_spec->resources[i].type;
+               res_cnt = res_spec->resources[i].cnt;
+
+               if (res_type == USNIC_VNIC_RES_TYPE_EOL)
+                       break;
+
+               if (res_cnt > usnic_vnic_res_free_cnt(vnic, res_type))
+                       return -EBUSY;
+       }
+
+       return 0;
+}
+
+int usnic_vnic_res_cnt(struct usnic_vnic *vnic,
+                               enum usnic_vnic_res_type type)
+{
+       return vnic->chunks[type].cnt;
+}
+
+int usnic_vnic_res_free_cnt(struct usnic_vnic *vnic,
+                               enum usnic_vnic_res_type type)
+{
+       return vnic->chunks[type].free_cnt;
+}
+
+struct usnic_vnic_res_chunk *
+usnic_vnic_get_resources(struct usnic_vnic *vnic, enum usnic_vnic_res_type type,
+                               int cnt, void *owner)
+{
+       struct usnic_vnic_res_chunk *src, *ret;
+       struct usnic_vnic_res *res;
+       int i;
+
+       if (usnic_vnic_res_free_cnt(vnic, type) < cnt || cnt < 1 || !owner)
+               return ERR_PTR(-EINVAL);
+
+       ret = kzalloc(sizeof(*ret), GFP_ATOMIC);
+       if (!ret) {
+               usnic_err("Failed to allocate chunk for %s - Out of memory\n",
+                               usnic_vnic_pci_name(vnic));
+               return ERR_PTR(-ENOMEM);
+       }
+
+       ret->res = kzalloc(sizeof(*(ret->res))*cnt, GFP_ATOMIC);
+       if (!ret->res) {
+               usnic_err("Failed to allocate resources for %s. Out of memory\n",
+                               usnic_vnic_pci_name(vnic));
+               kfree(ret);
+               return ERR_PTR(-ENOMEM);
+       }
+
+       spin_lock(&vnic->res_lock);
+       src = &vnic->chunks[type];
+       for (i = 0; i < src->cnt && ret->cnt < cnt; i++) {
+               res = src->res[i];
+               if (!res->owner) {
+                       src->free_cnt--;
+                       res->owner = owner;
+                       ret->res[ret->cnt++] = res;
+               }
+       }
+
+       spin_unlock(&vnic->res_lock);
+       ret->type = type;
+       ret->vnic = vnic;
+       WARN_ON(ret->cnt != cnt);
+
+       return ret;
+}
+
+void usnic_vnic_put_resources(struct usnic_vnic_res_chunk *chunk)
+{
+
+       struct usnic_vnic_res *res;
+       int i;
+       struct usnic_vnic *vnic = chunk->vnic;
+
+       spin_lock(&vnic->res_lock);
+       while ((i = --chunk->cnt) >= 0) {
+               res = chunk->res[i];
+               chunk->res[i] = NULL;
+               res->owner = NULL;
+               vnic->chunks[res->type].free_cnt++;
+       }
+       spin_unlock(&vnic->res_lock);
+
+       kfree(chunk->res);
+       kfree(chunk);
+}
+
+u16 usnic_vnic_get_index(struct usnic_vnic *vnic)
+{
+       return usnic_vnic_get_pdev(vnic)->devfn - 1;
+}
+
+static int usnic_vnic_alloc_res_chunk(struct usnic_vnic *vnic,
+                                       enum usnic_vnic_res_type type,
+                                       struct usnic_vnic_res_chunk *chunk)
+{
+       int cnt, err, i;
+       struct usnic_vnic_res *res;
+
+       cnt = vnic_dev_get_res_count(vnic->vdev, _to_vnic_res_type(type));
+       if (cnt < 1)
+               return -EINVAL;
+
+       chunk->cnt = chunk->free_cnt = cnt;
+       chunk->res = kzalloc(sizeof(*(chunk->res))*cnt, GFP_KERNEL);
+       if (!chunk->res)
+               return -ENOMEM;
+
+       for (i = 0; i < cnt; i++) {
+               res = kzalloc(sizeof(*res), GFP_KERNEL);
+               if (!res) {
+                       err = -ENOMEM;
+                       goto fail;
+               }
+               res->type = type;
+               res->vnic_idx = i;
+               res->vnic = vnic;
+               res->ctrl = vnic_dev_get_res(vnic->vdev,
+                                               _to_vnic_res_type(type), i);
+               chunk->res[i] = res;
+       }
+
+       chunk->vnic = vnic;
+       return 0;
+fail:
+       for (i--; i >= 0; i--)
+               kfree(chunk->res[i]);
+       kfree(chunk->res);
+       return err;
+}
+
+static void usnic_vnic_free_res_chunk(struct usnic_vnic_res_chunk *chunk)
+{
+       int i;
+       for (i = 0; i < chunk->cnt; i++)
+               kfree(chunk->res[i]);
+       kfree(chunk->res);
+}
+
+static int usnic_vnic_discover_resources(struct pci_dev *pdev,
+                                               struct usnic_vnic *vnic)
+{
+       enum usnic_vnic_res_type res_type;
+       int i;
+       int err = 0;
+
+       for (i = 0; i < ARRAY_SIZE(vnic->bar); i++) {
+               if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
+                       continue;
+               vnic->bar[i].len = pci_resource_len(pdev, i);
+               vnic->bar[i].vaddr = pci_iomap(pdev, i, vnic->bar[i].len);
+               if (!vnic->bar[i].vaddr) {
+                       usnic_err("Cannot memory-map BAR %d, aborting\n",
+                                       i);
+                       err = -ENODEV;
+                       goto out_clean_bar;
+               }
+               vnic->bar[i].bus_addr = pci_resource_start(pdev, i);
+       }
+
+       vnic->vdev = vnic_dev_register(NULL, pdev, pdev, vnic->bar,
+                       ARRAY_SIZE(vnic->bar));
+       if (!vnic->vdev) {
+               usnic_err("Failed to register device %s\n",
+                               pci_name(pdev));
+               err = -EINVAL;
+               goto out_clean_bar;
+       }
+
+       for (res_type = USNIC_VNIC_RES_TYPE_EOL + 1;
+                       res_type < USNIC_VNIC_RES_TYPE_MAX; res_type++) {
+               err = usnic_vnic_alloc_res_chunk(vnic, res_type,
+                                               &vnic->chunks[res_type]);
+               if (err) {
+                       usnic_err("Failed to alloc res %s with err %d\n",
+                                       usnic_vnic_res_type_to_str(res_type),
+                                       err);
+                       goto out_clean_chunks;
+               }
+       }
+
+       return 0;
+
+out_clean_chunks:
+       for (res_type--; res_type > USNIC_VNIC_RES_TYPE_EOL; res_type--)
+               usnic_vnic_free_res_chunk(&vnic->chunks[res_type]);
+       vnic_dev_unregister(vnic->vdev);
+out_clean_bar:
+       for (i = 0; i < ARRAY_SIZE(vnic->bar); i++) {
+               if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
+                       continue;
+               if (!vnic->bar[i].vaddr)
+                       break;
+
+               iounmap(vnic->bar[i].vaddr);
+       }
+
+       return err;
+}
+
+struct pci_dev *usnic_vnic_get_pdev(struct usnic_vnic *vnic)
+{
+       return vnic_dev_get_pdev(vnic->vdev);
+}
+
+struct vnic_dev_bar *usnic_vnic_get_bar(struct usnic_vnic *vnic,
+                               int bar_num)
+{
+       return (bar_num < ARRAY_SIZE(vnic->bar)) ? &vnic->bar[bar_num] : NULL;
+}
+
+static void usnic_vnic_release_resources(struct usnic_vnic *vnic)
+{
+       int i;
+       struct pci_dev *pdev;
+       enum usnic_vnic_res_type res_type;
+
+       pdev = usnic_vnic_get_pdev(vnic);
+
+       for (res_type = USNIC_VNIC_RES_TYPE_EOL + 1;
+                       res_type < USNIC_VNIC_RES_TYPE_MAX; res_type++)
+               usnic_vnic_free_res_chunk(&vnic->chunks[res_type]);
+
+       vnic_dev_unregister(vnic->vdev);
+
+       for (i = 0; i < ARRAY_SIZE(vnic->bar); i++) {
+               if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
+                       continue;
+               iounmap(vnic->bar[i].vaddr);
+       }
+}
+
+struct usnic_vnic *usnic_vnic_alloc(struct pci_dev *pdev)
+{
+       struct usnic_vnic *vnic;
+       int err = 0;
+
+       if (!pci_is_enabled(pdev)) {
+               usnic_err("PCI dev %s is disabled\n", pci_name(pdev));
+               return ERR_PTR(-EINVAL);
+       }
+
+       vnic = kzalloc(sizeof(*vnic), GFP_KERNEL);
+       if (!vnic) {
+               usnic_err("Failed to alloc vnic for %s - out of memory\n",
+                               pci_name(pdev));
+               return ERR_PTR(-ENOMEM);
+       }
+
+       spin_lock_init(&vnic->res_lock);
+
+       err = usnic_vnic_discover_resources(pdev, vnic);
+       if (err) {
+               usnic_err("Failed to discover %s resources with err %d\n",
+                               pci_name(pdev), err);
+               goto out_free_vnic;
+       }
+
+       usnic_dbg("Allocated vnic for %s\n", usnic_vnic_pci_name(vnic));
+
+       return vnic;
+
+out_free_vnic:
+       kfree(vnic);
+
+       return ERR_PTR(err);
+}
+
+void usnic_vnic_free(struct usnic_vnic *vnic)
+{
+       usnic_vnic_release_resources(vnic);
+       kfree(vnic);
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_vnic.h b/drivers/infiniband/hw/usnic/usnic_vnic.h
new file mode 100644 (file)
index 0000000..14d931a
--- /dev/null
@@ -0,0 +1,103 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_VNIC_H_
+#define USNIC_VNIC_H_
+
+#include <linux/pci.h>
+
+#include "vnic_dev.h"
+
+/*                      =USNIC_VNIC_RES_TYPE= =VNIC_RES=   =DESC= */
+#define USNIC_VNIC_RES_TYPES \
+       DEFINE_USNIC_VNIC_RES_AT(EOL, RES_TYPE_EOL, "EOL", 0) \
+       DEFINE_USNIC_VNIC_RES(WQ, RES_TYPE_WQ, "WQ") \
+       DEFINE_USNIC_VNIC_RES(RQ, RES_TYPE_RQ, "RQ") \
+       DEFINE_USNIC_VNIC_RES(CQ, RES_TYPE_CQ, "CQ") \
+       DEFINE_USNIC_VNIC_RES(INTR, RES_TYPE_INTR_CTRL, "INT") \
+       DEFINE_USNIC_VNIC_RES(MAX, RES_TYPE_MAX, "MAX")\
+
+#define DEFINE_USNIC_VNIC_RES_AT(usnic_vnic_res_t, vnic_res_type, desc, val) \
+       USNIC_VNIC_RES_TYPE_##usnic_vnic_res_t = val,
+#define DEFINE_USNIC_VNIC_RES(usnic_vnic_res_t, vnic_res_type, desc) \
+       USNIC_VNIC_RES_TYPE_##usnic_vnic_res_t,
+enum usnic_vnic_res_type {
+       USNIC_VNIC_RES_TYPES
+};
+#undef DEFINE_USNIC_VNIC_RES
+#undef DEFINE_USNIC_VNIC_RES_AT
+
+struct usnic_vnic_res {
+       enum usnic_vnic_res_type        type;
+       unsigned int                    vnic_idx;
+       struct usnic_vnic               *vnic;
+       void __iomem                    *ctrl;
+       void                            *owner;
+};
+
+struct usnic_vnic_res_chunk {
+       enum usnic_vnic_res_type        type;
+       int                             cnt;
+       int                             free_cnt;
+       struct usnic_vnic_res           **res;
+       struct usnic_vnic               *vnic;
+};
+
+struct usnic_vnic_res_desc {
+       enum usnic_vnic_res_type        type;
+       uint16_t                        cnt;
+};
+
+struct usnic_vnic_res_spec {
+       struct usnic_vnic_res_desc resources[USNIC_VNIC_RES_TYPE_MAX];
+};
+
+const char *usnic_vnic_res_type_to_str(enum usnic_vnic_res_type res_type);
+const char *usnic_vnic_pci_name(struct usnic_vnic *vnic);
+int usnic_vnic_dump(struct usnic_vnic *vnic, char *buf, int buf_sz,
+                       void *hdr_obj,
+                       int (*printtitle)(void *, char*, int),
+                       int (*printcols)(char *, int),
+                       int (*printrow)(void *, char *, int));
+void usnic_vnic_res_spec_update(struct usnic_vnic_res_spec *spec,
+                               enum usnic_vnic_res_type trgt_type,
+                               u16 cnt);
+int usnic_vnic_res_spec_satisfied(const struct usnic_vnic_res_spec *min_spec,
+                                       struct usnic_vnic_res_spec *res_spec);
+int usnic_vnic_spec_dump(char *buf, int buf_sz,
+                               struct usnic_vnic_res_spec *res_spec);
+int usnic_vnic_check_room(struct usnic_vnic *vnic,
+                               struct usnic_vnic_res_spec *res_spec);
+int usnic_vnic_res_cnt(struct usnic_vnic *vnic,
+                               enum usnic_vnic_res_type type);
+int usnic_vnic_res_free_cnt(struct usnic_vnic *vnic,
+                               enum usnic_vnic_res_type type);
+struct usnic_vnic_res_chunk *
+usnic_vnic_get_resources(struct usnic_vnic *vnic,
+                               enum usnic_vnic_res_type type,
+                               int cnt,
+                               void *owner);
+void usnic_vnic_put_resources(struct usnic_vnic_res_chunk *chunk);
+struct pci_dev *usnic_vnic_get_pdev(struct usnic_vnic *vnic);
+struct vnic_dev_bar *usnic_vnic_get_bar(struct usnic_vnic *vnic,
+                               int bar_num);
+struct usnic_vnic *usnic_vnic_alloc(struct pci_dev *pdev);
+void usnic_vnic_free(struct usnic_vnic *vnic);
+u16 usnic_vnic_get_index(struct usnic_vnic *vnic);
+
+#endif /*!USNIC_VNIC_H_*/