vfio: fix typo in doc for device setup
[dpdk.git] / lib / librte_eal / linuxapp / kni / kni_misc.c
index fd75e01..59d15ca 100644 (file)
@@ -1,37 +1,43 @@
 /*-
  * GPL LICENSE SUMMARY
- * 
- *   Copyright(c) 2010-2013 Intel Corporation. All rights reserved.
- * 
+ *
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *
  *   This program is free software; you can redistribute it and/or modify
  *   it under the terms of version 2 of the GNU General Public License as
  *   published by the Free Software Foundation.
- * 
+ *
  *   This program is distributed in the hope that it will be useful, but
  *   WITHOUT ANY WARRANTY; without even the implied warranty of
  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  *   General Public License for more details.
- * 
+ *
  *   You should have received a copy of the GNU General Public License
  *   along with this program; if not, write to the Free Software
  *   Foundation, Inc., 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  *   The full GNU General Public License is included in this distribution
  *   in the file called LICENSE.GPL.
- * 
+ *
  *   Contact Information:
  *   Intel Corporation
  */
 
+#include <linux/version.h>
 #include <linux/module.h>
 #include <linux/miscdevice.h>
 #include <linux/netdevice.h>
+#include <linux/etherdevice.h>
 #include <linux/pci.h>
 #include <linux/kthread.h>
 #include <linux/rwsem.h>
+#include <linux/nsproxy.h>
+#include <net/net_namespace.h>
+#include <net/netns/generic.h>
 
 #include <exec-env/rte_kni_common.h>
+
+#include "compat.h"
 #include "kni_dev.h"
-#include <rte_config.h>
 
 MODULE_LICENSE("Dual BSD/GPL");
 MODULE_AUTHOR("Intel Corporation");
@@ -58,9 +64,14 @@ static int kni_ioctl(struct inode *inode, unsigned int ioctl_num,
                                        unsigned long ioctl_param);
 static int kni_compat_ioctl(struct inode *inode, unsigned int ioctl_num,
                                                unsigned long ioctl_param);
+static int kni_dev_remove(struct kni_dev *dev);
+
+static int __init kni_parse_kthread_mode(void);
 
-/* kni kernel thread for rx */
-static int kni_thread(void *unused);
+/* KNI processing for single kernel thread mode */
+static int kni_thread_single(void *unused);
+/* KNI processing for multiple kernel thread mode */
+static int kni_thread_multiple(void *param);
 
 static struct file_operations kni_fops = {
        .owner = THIS_MODULE,
@@ -79,29 +90,96 @@ static struct miscdevice kni_misc = {
 /* loopback mode */
 static char *lo_mode = NULL;
 
+/* Kernel thread mode */
+static char *kthread_mode = NULL;
+static unsigned multiple_kthread_on = 0;
+
 #define KNI_DEV_IN_USE_BIT_NUM 0 /* Bit number for device in use */
 
-static volatile unsigned long device_in_use; /* device in use flag */
-static struct task_struct *kni_kthread;
+static int kni_net_id;
+
+struct kni_net {
+       unsigned long device_in_use; /* device in use flag */
+       struct task_struct *kni_kthread;
+       struct rw_semaphore kni_list_lock;
+       struct list_head kni_list_head;
+};
+
+static int __net_init kni_init_net(struct net *net)
+{
+#ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
+       struct kni_net *knet = net_generic(net, kni_net_id);
+#else
+       struct kni_net *knet;
+       int ret;
+
+       knet = kmalloc(sizeof(struct kni_net), GFP_KERNEL);
+       if (!knet) {
+               ret = -ENOMEM;
+               return ret;
+       }
+#endif
+
+       /* Clear the bit of device in use */
+       clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use);
+
+       init_rwsem(&knet->kni_list_lock);
+       INIT_LIST_HEAD(&knet->kni_list_head);
+
+#ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
+       return 0;
+#else
+       ret = net_assign_generic(net, kni_net_id, knet);
+       if (ret < 0)
+               kfree(knet);
+
+       return ret;
+#endif
+}
 
-/* kni list lock */
-static DECLARE_RWSEM(kni_list_lock);
+static void __net_exit kni_exit_net(struct net *net)
+{
+#ifndef HAVE_SIMPLIFIED_PERNET_OPERATIONS
+       struct kni_net *knet = net_generic(net, kni_net_id);
+
+       kfree(knet);
+#endif
+}
 
-/* kni list */
-static struct list_head kni_list_head = LIST_HEAD_INIT(kni_list_head);
+static struct pernet_operations kni_net_ops = {
+       .init = kni_init_net,
+       .exit = kni_exit_net,
+#ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
+       .id   = &kni_net_id,
+       .size = sizeof(struct kni_net),
+#endif
+};
 
 static int __init
 kni_init(void)
 {
+       int rc;
+
        KNI_PRINT("######## DPDK kni module loading ########\n");
 
-       if (misc_register(&kni_misc) != 0) {
-               KNI_ERR("Misc registration failed\n");
-               return 1;
+       if (kni_parse_kthread_mode() < 0) {
+               KNI_ERR("Invalid parameter for kthread_mode\n");
+               return -EINVAL;
        }
 
-       /* Clear the bit of device in use */
-       clear_bit(KNI_DEV_IN_USE_BIT_NUM, &device_in_use);
+#ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
+       rc = register_pernet_subsys(&kni_net_ops);
+#else
+       rc = register_pernet_gen_subsys(&kni_net_id, &kni_net_ops);
+#endif
+       if (rc)
+               return -EPERM;
+
+       rc = misc_register(&kni_misc);
+       if (rc != 0) {
+               KNI_ERR("Misc registration failed\n");
+               goto out;
+       }
 
        /* Configure the lo mode according to the input parameter */
        kni_net_config_lo_mode(lo_mode);
@@ -109,29 +187,68 @@ kni_init(void)
        KNI_PRINT("######## DPDK kni module loaded  ########\n");
 
        return 0;
+
+out:
+#ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
+       unregister_pernet_subsys(&kni_net_ops);
+#else
+       register_pernet_gen_subsys(&kni_net_id, &kni_net_ops);
+#endif
+       return rc;
 }
 
 static void __exit
 kni_exit(void)
 {
        misc_deregister(&kni_misc);
+#ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
+       unregister_pernet_subsys(&kni_net_ops);
+#else
+       register_pernet_gen_subsys(&kni_net_id, &kni_net_ops);
+#endif
        KNI_PRINT("####### DPDK kni module unloaded  #######\n");
 }
 
+static int __init
+kni_parse_kthread_mode(void)
+{
+       if (!kthread_mode)
+               return 0;
+
+       if (strcmp(kthread_mode, "single") == 0)
+               return 0;
+       else if (strcmp(kthread_mode, "multiple") == 0)
+               multiple_kthread_on = 1;
+       else
+               return -1;
+
+       return 0;
+}
+
 static int
 kni_open(struct inode *inode, struct file *file)
 {
-       /* kni device can be opened by one user only, test and set bit */
-       if (test_and_set_bit(KNI_DEV_IN_USE_BIT_NUM, &device_in_use))
+       struct net *net = current->nsproxy->net_ns;
+       struct kni_net *knet = net_generic(net, kni_net_id);
+
+       /* kni device can be opened by one user only per netns */
+       if (test_and_set_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use))
                return -EBUSY;
 
-       /* Create kernel thread for RX */
-       kni_kthread = kthread_run(kni_thread, NULL, "kni_thread");
-       if (IS_ERR(kni_kthread)) {
-               KNI_ERR("Unable to create kernel threaed\n");
-               return PTR_ERR(kni_kthread);
-       }
+       /* Create kernel thread for single mode */
+       if (multiple_kthread_on == 0) {
+               KNI_PRINT("Single kernel thread for all KNI devices\n");
+               /* Create kernel thread for RX */
+               knet->kni_kthread = kthread_run(kni_thread_single, (void *)knet,
+                                               "kni_single");
+               if (IS_ERR(knet->kni_kthread)) {
+                       KNI_ERR("Unable to create kernel threaed\n");
+                       return PTR_ERR(knet->kni_kthread);
+               }
+       } else
+               KNI_PRINT("Multiple kernel thread mode enabled\n");
 
+       file->private_data = get_net(net);
        KNI_PRINT("/dev/kni opened\n");
 
        return 0;
@@ -140,72 +257,143 @@ kni_open(struct inode *inode, struct file *file)
 static int
 kni_release(struct inode *inode, struct file *file)
 {
+       struct net *net = file->private_data;
+       struct kni_net *knet = net_generic(net, kni_net_id);
        struct kni_dev *dev, *n;
 
-       KNI_PRINT("Stopping KNI thread...");
-
-       /* Stop kernel thread */
-       kthread_stop(kni_kthread);
-       kni_kthread = NULL;
+       /* Stop kernel thread for single mode */
+       if (multiple_kthread_on == 0) {
+               /* Stop kernel thread */
+               kthread_stop(knet->kni_kthread);
+               knet->kni_kthread = NULL;
+       }
 
-       down_write(&kni_list_lock);
-       list_for_each_entry_safe(dev, n, &kni_list_head, list) {
-               /* Call the remove part to restore pci dev */
-               switch (dev->device_id) {
-               #define RTE_PCI_DEV_ID_DECL_IGB(vend, dev) case (dev):
-               #include <rte_pci_dev_ids.h>
-                       igb_kni_remove(dev->pci_dev);
-                       break;
-               #define RTE_PCI_DEV_ID_DECL_IXGBE(vend, dev) case (dev):
-               #include <rte_pci_dev_ids.h>
-                       ixgbe_kni_remove(dev->pci_dev);
-                       break;
-               default:
-                       break;
+       down_write(&knet->kni_list_lock);
+       list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
+               /* Stop kernel thread for multiple mode */
+               if (multiple_kthread_on && dev->pthread != NULL) {
+                       kthread_stop(dev->pthread);
+                       dev->pthread = NULL;
                }
-               unregister_netdev(dev->net_dev);
-               free_netdev(dev->net_dev);
+
+#ifdef RTE_KNI_VHOST
+               kni_vhost_backend_release(dev);
+#endif
+               kni_dev_remove(dev);
                list_del(&dev->list);
        }
-       up_write(&kni_list_lock);
+       up_write(&knet->kni_list_lock);
 
        /* Clear the bit of device in use */
-       clear_bit(KNI_DEV_IN_USE_BIT_NUM, &device_in_use);
+       clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use);
 
+       put_net(net);
        KNI_PRINT("/dev/kni closed\n");
 
        return 0;
 }
 
 static int
-kni_thread(void *unused)
+kni_thread_single(void *data)
 {
+       struct kni_net *knet = data;
        int j;
-       struct kni_dev *dev, *n;
+       struct kni_dev *dev;
 
-       KNI_PRINT("Kernel thread for KNI started\n");
        while (!kthread_should_stop()) {
-               down_read(&kni_list_lock);
+               down_read(&knet->kni_list_lock);
                for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
-                       list_for_each_entry_safe(dev, n,
-                                       &kni_list_head, list) {
+                       list_for_each_entry(dev, &knet->kni_list_head, list) {
+#ifdef RTE_KNI_VHOST
+                               kni_chk_vhost_rx(dev);
+#else
                                kni_net_rx(dev);
+#endif
                                kni_net_poll_resp(dev);
                        }
                }
-               up_read(&kni_list_lock);
+               up_read(&knet->kni_list_lock);
+#ifdef RTE_KNI_PREEMPT_DEFAULT
                /* reschedule out for a while */
                schedule_timeout_interruptible(usecs_to_jiffies( \
                                KNI_KTHREAD_RESCHEDULE_INTERVAL));
+#endif
        }
-       KNI_PRINT("Kernel thread for KNI stopped\n");
 
        return 0;
 }
 
 static int
-kni_ioctl_create(unsigned int ioctl_num, unsigned long ioctl_param)
+kni_thread_multiple(void *param)
 {
+       int j;
+       struct kni_dev *dev = (struct kni_dev *)param;
+
+       while (!kthread_should_stop()) {
+               for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
+#ifdef RTE_KNI_VHOST
+                       kni_chk_vhost_rx(dev);
+#else
+                       kni_net_rx(dev);
+#endif
+                       kni_net_poll_resp(dev);
+               }
+#ifdef RTE_KNI_PREEMPT_DEFAULT
+               schedule_timeout_interruptible(usecs_to_jiffies( \
+                               KNI_KTHREAD_RESCHEDULE_INTERVAL));
+#endif
+       }
+
+       return 0;
+}
+
+static int
+kni_dev_remove(struct kni_dev *dev)
+{
+       if (!dev)
+               return -ENODEV;
+
+       switch (dev->device_id) {
+       #define RTE_PCI_DEV_ID_DECL_IGB(vend, dev) case (dev):
+       #include <rte_pci_dev_ids.h>
+               igb_kni_remove(dev->pci_dev);
+               break;
+       #define RTE_PCI_DEV_ID_DECL_IXGBE(vend, dev) case (dev):
+       #include <rte_pci_dev_ids.h>
+               ixgbe_kni_remove(dev->pci_dev);
+               break;
+       default:
+               break;
+       }
+
+       if (dev->net_dev) {
+               unregister_netdev(dev->net_dev);
+               free_netdev(dev->net_dev);
+       }
+
+       return 0;
+}
+
+static int
+kni_check_param(struct kni_dev *kni, struct rte_kni_device_info *dev)
+{
+       if (!kni || !dev)
+               return -1;
+
+       /* Check if network name has been used */
+       if (!strncmp(kni->name, dev->name, RTE_KNI_NAMESIZE)) {
+               KNI_ERR("KNI name %s duplicated\n", dev->name);
+               return -1;
+       }
+
+       return 0;
+}
+
+static int
+kni_ioctl_create(struct net *net,
+               unsigned int ioctl_num, unsigned long ioctl_param)
+{
+       struct kni_net *knet = net_generic(net, kni_net_id);
        int ret;
        struct rte_kni_device_info dev_info;
        struct pci_dev *pci = NULL;
@@ -226,29 +414,44 @@ kni_ioctl_create(unsigned int ioctl_num, unsigned long ioctl_param)
                return -EIO;
        }
 
+       /**
+        * Check if the cpu core id is valid for binding,
+        * for multiple kernel thread mode.
+        */
+       if (multiple_kthread_on && dev_info.force_bind &&
+                               !cpu_online(dev_info.core_id)) {
+               KNI_ERR("cpu %u is not online\n", dev_info.core_id);
+               return -EINVAL;
+       }
+
        /* Check if it has been created */
-       down_read(&kni_list_lock);
-       list_for_each_entry_safe(dev, n, &kni_list_head, list) {
-               if (dev->port_id == dev_info.port_id) {
-                       up_read(&kni_list_lock);
-                       KNI_ERR("Port %d has already been created\n",
-                                               dev_info.port_id);
+       down_read(&knet->kni_list_lock);
+       list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
+               if (kni_check_param(dev, &dev_info) < 0) {
+                       up_read(&knet->kni_list_lock);
                        return -EINVAL;
                }
        }
-       up_read(&kni_list_lock);
+       up_read(&knet->kni_list_lock);
 
        net_dev = alloc_netdev(sizeof(struct kni_dev), dev_info.name,
+#ifdef NET_NAME_UNKNOWN
+                                                       NET_NAME_UNKNOWN,
+#endif
                                                        kni_net_init);
        if (net_dev == NULL) {
                KNI_ERR("error allocating device \"%s\"\n", dev_info.name);
                return -EBUSY;
        }
 
+       dev_net_set(net_dev, net);
+
        kni = netdev_priv(net_dev);
 
        kni->net_dev = net_dev;
-       kni->port_id = dev_info.port_id;
+       kni->group_id = dev_info.group_id;
+       kni->core_id = dev_info.core_id;
+       strncpy(kni->name, dev_info.name, RTE_KNI_NAMESIZE);
 
        /* Translate user space info into kernel space info */
        kni->tx_q = phys_to_virt(dev_info.tx_phys);
@@ -264,34 +467,44 @@ kni_ioctl_create(unsigned int ioctl_num, unsigned long ioctl_param)
        kni->mbuf_kva = phys_to_virt(dev_info.mbuf_phys);
        kni->mbuf_va = dev_info.mbuf_va;
 
+#ifdef RTE_KNI_VHOST
+       kni->vhost_queue = NULL;
+       kni->vq_status = BE_STOP;
+#endif
        kni->mbuf_size = dev_info.mbuf_size;
 
-       KNI_PRINT("tx_phys:          0x%016llx, tx_q addr:          0x%p\n",
-                                               (unsigned long long) dev_info.tx_phys, kni->tx_q);
-       KNI_PRINT("rx_phys:          0x%016llx, rx_q addr:          0x%p\n",
-                                               (unsigned long long) dev_info.rx_phys, kni->rx_q);
-       KNI_PRINT("alloc_phys:       0x%016llx, alloc_q addr:       0x%p\n",
-                                       (unsigned long long) dev_info.alloc_phys, kni->alloc_q);
-       KNI_PRINT("free_phys:        0x%016llx, free_q addr:        0x%p\n",
-                                       (unsigned long long) dev_info.free_phys, kni->free_q);
-       KNI_PRINT("req_phys:         0x%016llx, req_q addr:         0x%p\n",
-                                       (unsigned long long) dev_info.req_phys, kni->req_q);
-       KNI_PRINT("resp_phys:        0x%016llx, resp_q addr:        0x%p\n",
-                                       (unsigned long long) dev_info.resp_phys, kni->resp_q);
-       KNI_PRINT("mbuf_phys:        0x%016llx, mbuf_kva:           0x%p\n",
-                                       (unsigned long long) dev_info.mbuf_phys, kni->mbuf_kva);
-       KNI_PRINT("mbuf_va:          0x%p\n", dev_info.mbuf_va);
-       KNI_PRINT("mbuf_size:        %u\n", kni->mbuf_size);
-
-       KNI_DBG("PCI: %02x:%02x.%02x %04x:%04x\n", dev_info.bus, dev_info.devid,
-                       dev_info.function, dev_info.vendor_id, dev_info.device_id);
+       KNI_PRINT("tx_phys:      0x%016llx, tx_q addr:      0x%p\n",
+               (unsigned long long) dev_info.tx_phys, kni->tx_q);
+       KNI_PRINT("rx_phys:      0x%016llx, rx_q addr:      0x%p\n",
+               (unsigned long long) dev_info.rx_phys, kni->rx_q);
+       KNI_PRINT("alloc_phys:   0x%016llx, alloc_q addr:   0x%p\n",
+               (unsigned long long) dev_info.alloc_phys, kni->alloc_q);
+       KNI_PRINT("free_phys:    0x%016llx, free_q addr:    0x%p\n",
+               (unsigned long long) dev_info.free_phys, kni->free_q);
+       KNI_PRINT("req_phys:     0x%016llx, req_q addr:     0x%p\n",
+               (unsigned long long) dev_info.req_phys, kni->req_q);
+       KNI_PRINT("resp_phys:    0x%016llx, resp_q addr:    0x%p\n",
+               (unsigned long long) dev_info.resp_phys, kni->resp_q);
+       KNI_PRINT("mbuf_phys:    0x%016llx, mbuf_kva:       0x%p\n",
+               (unsigned long long) dev_info.mbuf_phys, kni->mbuf_kva);
+       KNI_PRINT("mbuf_va:      0x%p\n", dev_info.mbuf_va);
+       KNI_PRINT("mbuf_size:    %u\n", kni->mbuf_size);
+
+       KNI_DBG("PCI: %02x:%02x.%02x %04x:%04x\n",
+                                       dev_info.bus,
+                                       dev_info.devid,
+                                       dev_info.function,
+                                       dev_info.vendor_id,
+                                       dev_info.device_id);
 
        pci = pci_get_device(dev_info.vendor_id, dev_info.device_id, NULL);
 
        /* Support Ethtool */
        while (pci) {
-               KNI_PRINT("pci_bus: %02x:%02x:%02x \n", pci->bus->number,
-                               PCI_SLOT(pci->devfn), PCI_FUNC(pci->devfn));
+               KNI_PRINT("pci_bus: %02x:%02x:%02x \n",
+                                       pci->bus->number,
+                                       PCI_SLOT(pci->devfn),
+                                       PCI_FUNC(pci->devfn));
 
                if ((pci->bus->number == dev_info.bus) &&
                        (PCI_SLOT(pci->devfn) == dev_info.devid) &&
@@ -332,65 +545,94 @@ kni_ioctl_create(unsigned int ioctl_num, unsigned long ioctl_param)
        if (pci)
                pci_dev_put(pci);
 
+       if (kni->lad_dev)
+               memcpy(net_dev->dev_addr, kni->lad_dev->dev_addr, ETH_ALEN);
+       else
+               /*
+                * Generate random mac address. eth_random_addr() is the newer
+                * version of generating mac address in linux kernel.
+                */
+               random_ether_addr(net_dev->dev_addr);
+
        ret = register_netdev(net_dev);
        if (ret) {
                KNI_ERR("error %i registering device \"%s\"\n",
                                        ret, dev_info.name);
-               free_netdev(net_dev);
+               kni_dev_remove(kni);
                return -ENODEV;
        }
 
-       down_write(&kni_list_lock);
-       list_add(&kni->list, &kni_list_head);
-       up_write(&kni_list_lock);
-       printk(KERN_INFO "KNI: Successfully create kni for port %d\n",
-                                               dev_info.port_id);
+#ifdef RTE_KNI_VHOST
+       kni_vhost_init(kni);
+#endif
+
+       /**
+        * Create a new kernel thread for multiple mode, set its core affinity,
+        * and finally wake it up.
+        */
+       if (multiple_kthread_on) {
+               kni->pthread = kthread_create(kni_thread_multiple,
+                                             (void *)kni,
+                                             "kni_%s", kni->name);
+               if (IS_ERR(kni->pthread)) {
+                       kni_dev_remove(kni);
+                       return -ECANCELED;
+               }
+               if (dev_info.force_bind)
+                       kthread_bind(kni->pthread, kni->core_id);
+               wake_up_process(kni->pthread);
+       }
+
+       down_write(&knet->kni_list_lock);
+       list_add(&kni->list, &knet->kni_list_head);
+       up_write(&knet->kni_list_lock);
 
        return 0;
 }
 
 static int
-kni_ioctl_release(unsigned int ioctl_num, unsigned long ioctl_param)
+kni_ioctl_release(struct net *net,
+               unsigned int ioctl_num, unsigned long ioctl_param)
 {
+       struct kni_net *knet = net_generic(net, kni_net_id);
        int ret = -EINVAL;
-       uint8_t port_id;
        struct kni_dev *dev, *n;
+       struct rte_kni_device_info dev_info;
 
-       if (_IOC_SIZE(ioctl_num) > sizeof(port_id))
+       if (_IOC_SIZE(ioctl_num) > sizeof(dev_info))
                        return -EINVAL;
 
-       ret = copy_from_user(&port_id, (void *)ioctl_param, sizeof(port_id));
+       ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info));
        if (ret) {
                KNI_ERR("copy_from_user in kni_ioctl_release");
                return -EIO;
        }
 
-       down_write(&kni_list_lock);
-       list_for_each_entry_safe(dev, n, &kni_list_head, list) {
-               if (dev->port_id != port_id)
+       /* Release the network device according to its name */
+       if (strlen(dev_info.name) == 0)
+               return ret;
+
+       down_write(&knet->kni_list_lock);
+       list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
+               if (strncmp(dev->name, dev_info.name, RTE_KNI_NAMESIZE) != 0)
                        continue;
 
-               switch (dev->device_id) {
-               #define RTE_PCI_DEV_ID_DECL_IGB(vend, dev) case (dev):
-               #include <rte_pci_dev_ids.h>
-                       igb_kni_remove(dev->pci_dev);
-                       break;
-               #define RTE_PCI_DEV_ID_DECL_IXGBE(vend, dev) case (dev):
-               #include <rte_pci_dev_ids.h>
-                       ixgbe_kni_remove(dev->pci_dev);
-                       break;
-               default:
-                       break;
+               if (multiple_kthread_on && dev->pthread != NULL) {
+                       kthread_stop(dev->pthread);
+                       dev->pthread = NULL;
                }
-               unregister_netdev(dev->net_dev);
-               free_netdev(dev->net_dev);
+
+#ifdef RTE_KNI_VHOST
+               kni_vhost_backend_release(dev);
+#endif
+               kni_dev_remove(dev);
                list_del(&dev->list);
                ret = 0;
                break;
        }
-       up_write(&kni_list_lock);
-       printk(KERN_INFO "KNI: %s release kni for port %d\n",
-               (ret == 0 ? "Successfully" : "Unsuccessfully"), port_id);
+       up_write(&knet->kni_list_lock);
+       printk(KERN_INFO "KNI: %s release kni named %s\n",
+               (ret == 0 ? "Successfully" : "Unsuccessfully"), dev_info.name);
 
        return ret;
 }
@@ -401,8 +643,9 @@ kni_ioctl(struct inode *inode,
        unsigned long ioctl_param)
 {
        int ret = -EINVAL;
+       struct net *net = current->nsproxy->net_ns;
 
-       KNI_DBG("IOCTL num=0x%0x param=0x%0lx \n", ioctl_num, ioctl_param);
+       KNI_DBG("IOCTL num=0x%0x param=0x%0lx\n", ioctl_num, ioctl_param);
 
        /*
         * Switch according to the ioctl called
@@ -412,13 +655,13 @@ kni_ioctl(struct inode *inode,
                /* For test only, not used */
                break;
        case _IOC_NR(RTE_KNI_IOCTL_CREATE):
-               ret = kni_ioctl_create(ioctl_num, ioctl_param);
+               ret = kni_ioctl_create(net, ioctl_num, ioctl_param);
                break;
        case _IOC_NR(RTE_KNI_IOCTL_RELEASE):
-               ret = kni_ioctl_release(ioctl_num, ioctl_param);
+               ret = kni_ioctl_release(net, ioctl_num, ioctl_param);
                break;
        default:
-               KNI_DBG("IOCTL default \n");
+               KNI_DBG("IOCTL default\n");
                break;
        }
 
@@ -448,3 +691,10 @@ MODULE_PARM_DESC(lo_mode,
 "\n"
 );
 
+module_param(kthread_mode, charp, S_IRUGO);
+MODULE_PARM_DESC(kthread_mode,
+"Kernel thread mode (default=single):\n"
+"    single    Single kernel thread mode enabled.\n"
+"    multiple  Multiple kernel thread mode enabled.\n"
+"\n"
+);