4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of version 2 of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 * The full GNU General Public License is included in this distribution
19 * in the file called LICENSE.GPL.
21 * Contact Information:
25 #include <linux/version.h>
26 #include <linux/module.h>
27 #include <linux/miscdevice.h>
28 #include <linux/netdevice.h>
29 #include <linux/etherdevice.h>
30 #include <linux/pci.h>
31 #include <linux/kthread.h>
32 #include <linux/rwsem.h>
33 #include <linux/nsproxy.h>
34 #include <net/net_namespace.h>
35 #include <net/netns/generic.h>
37 #include <exec-env/rte_kni_common.h>
42 MODULE_LICENSE("Dual BSD/GPL");
43 MODULE_AUTHOR("Intel Corporation");
44 MODULE_DESCRIPTION("Kernel Module for managing kni devices");
46 #define KNI_RX_LOOP_NUM 1000
48 #define KNI_MAX_DEVICES 32
50 extern void kni_net_rx(struct kni_dev *kni);
51 extern void kni_net_init(struct net_device *dev);
52 extern void kni_net_config_lo_mode(char *lo_str);
53 extern void kni_net_poll_resp(struct kni_dev *kni);
54 extern void kni_set_ethtool_ops(struct net_device *netdev);
56 extern int ixgbe_kni_probe(struct pci_dev *pdev, struct net_device **lad_dev);
57 extern void ixgbe_kni_remove(struct pci_dev *pdev);
58 extern struct pci_device_id ixgbe_pci_tbl[];
59 extern int igb_kni_probe(struct pci_dev *pdev, struct net_device **lad_dev);
60 extern void igb_kni_remove(struct pci_dev *pdev);
62 static int kni_open(struct inode *inode, struct file *file);
63 static int kni_release(struct inode *inode, struct file *file);
64 static int kni_ioctl(struct inode *inode, unsigned int ioctl_num,
65 unsigned long ioctl_param);
66 static int kni_compat_ioctl(struct inode *inode, unsigned int ioctl_num,
67 unsigned long ioctl_param);
68 static int kni_dev_remove(struct kni_dev *dev);
70 static int __init kni_parse_kthread_mode(void);
72 /* KNI processing for single kernel thread mode */
73 static int kni_thread_single(void *unused);
74 /* KNI processing for multiple kernel thread mode */
75 static int kni_thread_multiple(void *param);
77 static struct file_operations kni_fops = {
80 .release = kni_release,
81 .unlocked_ioctl = (void *)kni_ioctl,
82 .compat_ioctl = (void *)kni_compat_ioctl,
85 static struct miscdevice kni_misc = {
86 .minor = MISC_DYNAMIC_MINOR,
92 static char *lo_mode = NULL;
94 /* Kernel thread mode */
95 static char *kthread_mode = NULL;
96 static unsigned multiple_kthread_on = 0;
98 #define KNI_DEV_IN_USE_BIT_NUM 0 /* Bit number for device in use */
100 static int kni_net_id;
103 unsigned long device_in_use; /* device in use flag */
104 struct task_struct *kni_kthread;
105 struct rw_semaphore kni_list_lock;
106 struct list_head kni_list_head;
109 static int __net_init kni_init_net(struct net *net)
111 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
112 struct kni_net *knet = net_generic(net, kni_net_id);
114 struct kni_net *knet;
117 knet = kmalloc(sizeof(struct kni_net), GFP_KERNEL);
124 /* Clear the bit of device in use */
125 clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use);
127 init_rwsem(&knet->kni_list_lock);
128 INIT_LIST_HEAD(&knet->kni_list_head);
130 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
133 ret = net_assign_generic(net, kni_net_id, knet);
141 static void __net_exit kni_exit_net(struct net *net)
143 #ifndef HAVE_SIMPLIFIED_PERNET_OPERATIONS
144 struct kni_net *knet = net_generic(net, kni_net_id);
150 static struct pernet_operations kni_net_ops = {
151 .init = kni_init_net,
152 .exit = kni_exit_net,
153 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
155 .size = sizeof(struct kni_net),
164 KNI_PRINT("######## DPDK kni module loading ########\n");
166 if (kni_parse_kthread_mode() < 0) {
167 KNI_ERR("Invalid parameter for kthread_mode\n");
171 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
172 rc = register_pernet_subsys(&kni_net_ops);
174 rc = register_pernet_gen_subsys(&kni_net_id, &kni_net_ops);
179 rc = misc_register(&kni_misc);
181 KNI_ERR("Misc registration failed\n");
185 /* Configure the lo mode according to the input parameter */
186 kni_net_config_lo_mode(lo_mode);
188 KNI_PRINT("######## DPDK kni module loaded ########\n");
193 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
194 unregister_pernet_subsys(&kni_net_ops);
196 register_pernet_gen_subsys(&kni_net_id, &kni_net_ops);
204 misc_deregister(&kni_misc);
205 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
206 unregister_pernet_subsys(&kni_net_ops);
208 register_pernet_gen_subsys(&kni_net_id, &kni_net_ops);
210 KNI_PRINT("####### DPDK kni module unloaded #######\n");
214 kni_parse_kthread_mode(void)
219 if (strcmp(kthread_mode, "single") == 0)
221 else if (strcmp(kthread_mode, "multiple") == 0)
222 multiple_kthread_on = 1;
230 kni_open(struct inode *inode, struct file *file)
232 struct net *net = current->nsproxy->net_ns;
233 struct kni_net *knet = net_generic(net, kni_net_id);
235 /* kni device can be opened by one user only per netns */
236 if (test_and_set_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use))
239 /* Create kernel thread for single mode */
240 if (multiple_kthread_on == 0) {
241 KNI_PRINT("Single kernel thread for all KNI devices\n");
242 /* Create kernel thread for RX */
243 knet->kni_kthread = kthread_run(kni_thread_single, (void *)knet,
245 if (IS_ERR(knet->kni_kthread)) {
246 KNI_ERR("Unable to create kernel threaed\n");
247 return PTR_ERR(knet->kni_kthread);
250 KNI_PRINT("Multiple kernel thread mode enabled\n");
252 file->private_data = get_net(net);
253 KNI_PRINT("/dev/kni opened\n");
259 kni_release(struct inode *inode, struct file *file)
261 struct net *net = file->private_data;
262 struct kni_net *knet = net_generic(net, kni_net_id);
263 struct kni_dev *dev, *n;
265 /* Stop kernel thread for single mode */
266 if (multiple_kthread_on == 0) {
267 /* Stop kernel thread */
268 kthread_stop(knet->kni_kthread);
269 knet->kni_kthread = NULL;
272 down_write(&knet->kni_list_lock);
273 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
274 /* Stop kernel thread for multiple mode */
275 if (multiple_kthread_on && dev->pthread != NULL) {
276 kthread_stop(dev->pthread);
281 kni_vhost_backend_release(dev);
284 list_del(&dev->list);
286 up_write(&knet->kni_list_lock);
288 /* Clear the bit of device in use */
289 clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use);
292 KNI_PRINT("/dev/kni closed\n");
298 kni_thread_single(void *data)
300 struct kni_net *knet = data;
304 while (!kthread_should_stop()) {
305 down_read(&knet->kni_list_lock);
306 for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
307 list_for_each_entry(dev, &knet->kni_list_head, list) {
309 kni_chk_vhost_rx(dev);
313 kni_net_poll_resp(dev);
316 up_read(&knet->kni_list_lock);
317 #ifdef RTE_KNI_PREEMPT_DEFAULT
318 /* reschedule out for a while */
319 schedule_timeout_interruptible(usecs_to_jiffies( \
320 KNI_KTHREAD_RESCHEDULE_INTERVAL));
328 kni_thread_multiple(void *param)
331 struct kni_dev *dev = (struct kni_dev *)param;
333 while (!kthread_should_stop()) {
334 for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
336 kni_chk_vhost_rx(dev);
340 kni_net_poll_resp(dev);
342 #ifdef RTE_KNI_PREEMPT_DEFAULT
343 schedule_timeout_interruptible(usecs_to_jiffies( \
344 KNI_KTHREAD_RESCHEDULE_INTERVAL));
352 kni_dev_remove(struct kni_dev *dev)
357 if (pci_match_id(ixgbe_pci_tbl, dev->pci_dev))
358 ixgbe_kni_remove(dev->pci_dev);
360 switch (dev->device_id) {
361 #define RTE_PCI_DEV_ID_DECL_IGB(vend, dev) case (dev):
362 #include <rte_pci_dev_ids.h>
363 igb_kni_remove(dev->pci_dev);
370 unregister_netdev(dev->net_dev);
371 free_netdev(dev->net_dev);
378 kni_check_param(struct kni_dev *kni, struct rte_kni_device_info *dev)
383 /* Check if network name has been used */
384 if (!strncmp(kni->name, dev->name, RTE_KNI_NAMESIZE)) {
385 KNI_ERR("KNI name %s duplicated\n", dev->name);
393 kni_ioctl_create(struct net *net,
394 unsigned int ioctl_num, unsigned long ioctl_param)
396 struct kni_net *knet = net_generic(net, kni_net_id);
398 struct rte_kni_device_info dev_info;
399 struct pci_dev *pci = NULL;
400 struct pci_dev *found_pci = NULL;
401 struct net_device *net_dev = NULL;
402 struct net_device *lad_dev = NULL;
403 struct kni_dev *kni, *dev, *n;
405 printk(KERN_INFO "KNI: Creating kni...\n");
406 /* Check the buffer size, to avoid warning */
407 if (_IOC_SIZE(ioctl_num) > sizeof(dev_info))
410 /* Copy kni info from user space */
411 ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info));
413 KNI_ERR("copy_from_user in kni_ioctl_create");
418 * Check if the cpu core id is valid for binding,
419 * for multiple kernel thread mode.
421 if (multiple_kthread_on && dev_info.force_bind &&
422 !cpu_online(dev_info.core_id)) {
423 KNI_ERR("cpu %u is not online\n", dev_info.core_id);
427 /* Check if it has been created */
428 down_read(&knet->kni_list_lock);
429 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
430 if (kni_check_param(dev, &dev_info) < 0) {
431 up_read(&knet->kni_list_lock);
435 up_read(&knet->kni_list_lock);
437 net_dev = alloc_netdev(sizeof(struct kni_dev), dev_info.name,
438 #ifdef NET_NAME_UNKNOWN
442 if (net_dev == NULL) {
443 KNI_ERR("error allocating device \"%s\"\n", dev_info.name);
447 dev_net_set(net_dev, net);
449 kni = netdev_priv(net_dev);
451 kni->net_dev = net_dev;
452 kni->group_id = dev_info.group_id;
453 kni->core_id = dev_info.core_id;
454 strncpy(kni->name, dev_info.name, RTE_KNI_NAMESIZE);
456 /* Translate user space info into kernel space info */
457 kni->tx_q = phys_to_virt(dev_info.tx_phys);
458 kni->rx_q = phys_to_virt(dev_info.rx_phys);
459 kni->alloc_q = phys_to_virt(dev_info.alloc_phys);
460 kni->free_q = phys_to_virt(dev_info.free_phys);
462 kni->req_q = phys_to_virt(dev_info.req_phys);
463 kni->resp_q = phys_to_virt(dev_info.resp_phys);
464 kni->sync_va = dev_info.sync_va;
465 kni->sync_kva = phys_to_virt(dev_info.sync_phys);
467 kni->mbuf_kva = phys_to_virt(dev_info.mbuf_phys);
468 kni->mbuf_va = dev_info.mbuf_va;
471 kni->vhost_queue = NULL;
472 kni->vq_status = BE_STOP;
474 kni->mbuf_size = dev_info.mbuf_size;
476 KNI_PRINT("tx_phys: 0x%016llx, tx_q addr: 0x%p\n",
477 (unsigned long long) dev_info.tx_phys, kni->tx_q);
478 KNI_PRINT("rx_phys: 0x%016llx, rx_q addr: 0x%p\n",
479 (unsigned long long) dev_info.rx_phys, kni->rx_q);
480 KNI_PRINT("alloc_phys: 0x%016llx, alloc_q addr: 0x%p\n",
481 (unsigned long long) dev_info.alloc_phys, kni->alloc_q);
482 KNI_PRINT("free_phys: 0x%016llx, free_q addr: 0x%p\n",
483 (unsigned long long) dev_info.free_phys, kni->free_q);
484 KNI_PRINT("req_phys: 0x%016llx, req_q addr: 0x%p\n",
485 (unsigned long long) dev_info.req_phys, kni->req_q);
486 KNI_PRINT("resp_phys: 0x%016llx, resp_q addr: 0x%p\n",
487 (unsigned long long) dev_info.resp_phys, kni->resp_q);
488 KNI_PRINT("mbuf_phys: 0x%016llx, mbuf_kva: 0x%p\n",
489 (unsigned long long) dev_info.mbuf_phys, kni->mbuf_kva);
490 KNI_PRINT("mbuf_va: 0x%p\n", dev_info.mbuf_va);
491 KNI_PRINT("mbuf_size: %u\n", kni->mbuf_size);
493 KNI_DBG("PCI: %02x:%02x.%02x %04x:%04x\n",
500 pci = pci_get_device(dev_info.vendor_id, dev_info.device_id, NULL);
502 /* Support Ethtool */
504 KNI_PRINT("pci_bus: %02x:%02x:%02x \n",
506 PCI_SLOT(pci->devfn),
507 PCI_FUNC(pci->devfn));
509 if ((pci->bus->number == dev_info.bus) &&
510 (PCI_SLOT(pci->devfn) == dev_info.devid) &&
511 (PCI_FUNC(pci->devfn) == dev_info.function)) {
514 if (pci_match_id(ixgbe_pci_tbl, found_pci))
515 ret = ixgbe_kni_probe(found_pci, &lad_dev);
517 switch (dev_info.device_id) {
518 #define RTE_PCI_DEV_ID_DECL_IGB(vend, dev) case (dev):
519 #include <rte_pci_dev_ids.h>
520 ret = igb_kni_probe(found_pci, &lad_dev);
527 KNI_DBG("PCI found: pci=0x%p, lad_dev=0x%p\n",
530 kni->lad_dev = lad_dev;
531 kni_set_ethtool_ops(kni->net_dev);
533 KNI_ERR("Device not supported by ethtool");
537 kni->pci_dev = found_pci;
538 kni->device_id = dev_info.device_id;
541 pci = pci_get_device(dev_info.vendor_id,
542 dev_info.device_id, pci);
548 memcpy(net_dev->dev_addr, kni->lad_dev->dev_addr, ETH_ALEN);
551 * Generate random mac address. eth_random_addr() is the newer
552 * version of generating mac address in linux kernel.
554 random_ether_addr(net_dev->dev_addr);
556 ret = register_netdev(net_dev);
558 KNI_ERR("error %i registering device \"%s\"\n",
569 * Create a new kernel thread for multiple mode, set its core affinity,
570 * and finally wake it up.
572 if (multiple_kthread_on) {
573 kni->pthread = kthread_create(kni_thread_multiple,
575 "kni_%s", kni->name);
576 if (IS_ERR(kni->pthread)) {
580 if (dev_info.force_bind)
581 kthread_bind(kni->pthread, kni->core_id);
582 wake_up_process(kni->pthread);
585 down_write(&knet->kni_list_lock);
586 list_add(&kni->list, &knet->kni_list_head);
587 up_write(&knet->kni_list_lock);
593 kni_ioctl_release(struct net *net,
594 unsigned int ioctl_num, unsigned long ioctl_param)
596 struct kni_net *knet = net_generic(net, kni_net_id);
598 struct kni_dev *dev, *n;
599 struct rte_kni_device_info dev_info;
601 if (_IOC_SIZE(ioctl_num) > sizeof(dev_info))
604 ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info));
606 KNI_ERR("copy_from_user in kni_ioctl_release");
610 /* Release the network device according to its name */
611 if (strlen(dev_info.name) == 0)
614 down_write(&knet->kni_list_lock);
615 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
616 if (strncmp(dev->name, dev_info.name, RTE_KNI_NAMESIZE) != 0)
619 if (multiple_kthread_on && dev->pthread != NULL) {
620 kthread_stop(dev->pthread);
625 kni_vhost_backend_release(dev);
628 list_del(&dev->list);
632 up_write(&knet->kni_list_lock);
633 printk(KERN_INFO "KNI: %s release kni named %s\n",
634 (ret == 0 ? "Successfully" : "Unsuccessfully"), dev_info.name);
640 kni_ioctl(struct inode *inode,
641 unsigned int ioctl_num,
642 unsigned long ioctl_param)
645 struct net *net = current->nsproxy->net_ns;
647 KNI_DBG("IOCTL num=0x%0x param=0x%0lx\n", ioctl_num, ioctl_param);
650 * Switch according to the ioctl called
652 switch (_IOC_NR(ioctl_num)) {
653 case _IOC_NR(RTE_KNI_IOCTL_TEST):
654 /* For test only, not used */
656 case _IOC_NR(RTE_KNI_IOCTL_CREATE):
657 ret = kni_ioctl_create(net, ioctl_num, ioctl_param);
659 case _IOC_NR(RTE_KNI_IOCTL_RELEASE):
660 ret = kni_ioctl_release(net, ioctl_num, ioctl_param);
663 KNI_DBG("IOCTL default\n");
671 kni_compat_ioctl(struct inode *inode,
672 unsigned int ioctl_num,
673 unsigned long ioctl_param)
675 /* 32 bits app on 64 bits OS to be supported later */
676 KNI_PRINT("Not implemented.\n");
681 module_init(kni_init);
682 module_exit(kni_exit);
684 module_param(lo_mode, charp, S_IRUGO | S_IWUSR);
685 MODULE_PARM_DESC(lo_mode,
686 "KNI loopback mode (default=lo_mode_none):\n"
687 " lo_mode_none Kernel loopback disabled\n"
688 " lo_mode_fifo Enable kernel loopback with fifo\n"
689 " lo_mode_fifo_skb Enable kernel loopback with fifo and skb buffer\n"
693 module_param(kthread_mode, charp, S_IRUGO);
694 MODULE_PARM_DESC(kthread_mode,
695 "Kernel thread mode (default=single):\n"
696 " single Single kernel thread mode enabled.\n"
697 " multiple Multiple kernel thread mode enabled.\n"