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/mutex.h>
34 #include <linux/nsproxy.h>
35 #include <net/net_namespace.h>
36 #include <net/netns/generic.h>
38 #include <exec-env/rte_kni_common.h>
43 MODULE_LICENSE("Dual BSD/GPL");
44 MODULE_AUTHOR("Intel Corporation");
45 MODULE_DESCRIPTION("Kernel Module for managing kni devices");
47 #define KNI_RX_LOOP_NUM 1000
49 #define KNI_MAX_DEVICES 32
51 extern void kni_net_rx(struct kni_dev *kni);
52 extern void kni_net_init(struct net_device *dev);
53 extern void kni_net_config_lo_mode(char *lo_str);
54 extern void kni_net_poll_resp(struct kni_dev *kni);
55 extern void kni_set_ethtool_ops(struct net_device *netdev);
57 extern int ixgbe_kni_probe(struct pci_dev *pdev, struct net_device **lad_dev);
58 extern void ixgbe_kni_remove(struct pci_dev *pdev);
59 extern struct pci_device_id ixgbe_pci_tbl[];
60 extern int igb_kni_probe(struct pci_dev *pdev, struct net_device **lad_dev);
61 extern void igb_kni_remove(struct pci_dev *pdev);
62 extern struct pci_device_id igb_pci_tbl[];
64 static int kni_open(struct inode *inode, struct file *file);
65 static int kni_release(struct inode *inode, struct file *file);
66 static int kni_ioctl(struct inode *inode, unsigned int ioctl_num,
67 unsigned long ioctl_param);
68 static int kni_compat_ioctl(struct inode *inode, unsigned int ioctl_num,
69 unsigned long ioctl_param);
70 static int kni_dev_remove(struct kni_dev *dev);
72 static int __init kni_parse_kthread_mode(void);
74 /* KNI processing for single kernel thread mode */
75 static int kni_thread_single(void *unused);
76 /* KNI processing for multiple kernel thread mode */
77 static int kni_thread_multiple(void *param);
79 static struct file_operations kni_fops = {
82 .release = kni_release,
83 .unlocked_ioctl = (void *)kni_ioctl,
84 .compat_ioctl = (void *)kni_compat_ioctl,
87 static struct miscdevice kni_misc = {
88 .minor = MISC_DYNAMIC_MINOR,
94 static char *lo_mode = NULL;
96 /* Kernel thread mode */
97 static char *kthread_mode = NULL;
98 static unsigned multiple_kthread_on = 0;
100 #define KNI_DEV_IN_USE_BIT_NUM 0 /* Bit number for device in use */
102 static int kni_net_id;
105 unsigned long device_in_use; /* device in use flag */
106 struct mutex kni_kthread_lock;
107 struct task_struct *kni_kthread;
108 struct rw_semaphore kni_list_lock;
109 struct list_head kni_list_head;
112 static int __net_init kni_init_net(struct net *net)
114 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
115 struct kni_net *knet = net_generic(net, kni_net_id);
116 memset(knet, 0, sizeof(*knet));
118 struct kni_net *knet;
121 knet = kzalloc(sizeof(struct kni_net), GFP_KERNEL);
128 /* Clear the bit of device in use */
129 clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use);
131 mutex_init(&knet->kni_kthread_lock);
133 init_rwsem(&knet->kni_list_lock);
134 INIT_LIST_HEAD(&knet->kni_list_head);
136 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
139 ret = net_assign_generic(net, kni_net_id, knet);
147 static void __net_exit kni_exit_net(struct net *net)
149 struct kni_net *knet = net_generic(net, kni_net_id);
150 mutex_destroy(&knet->kni_kthread_lock);
151 #ifndef HAVE_SIMPLIFIED_PERNET_OPERATIONS
156 static struct pernet_operations kni_net_ops = {
157 .init = kni_init_net,
158 .exit = kni_exit_net,
159 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
161 .size = sizeof(struct kni_net),
170 KNI_PRINT("######## DPDK kni module loading ########\n");
172 if (kni_parse_kthread_mode() < 0) {
173 KNI_ERR("Invalid parameter for kthread_mode\n");
177 if (multiple_kthread_on == 0)
178 KNI_PRINT("Single kernel thread for all KNI devices\n");
180 KNI_PRINT("Multiple kernel thread mode enabled\n");
182 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
183 rc = register_pernet_subsys(&kni_net_ops);
185 rc = register_pernet_gen_subsys(&kni_net_id, &kni_net_ops);
190 rc = misc_register(&kni_misc);
192 KNI_ERR("Misc registration failed\n");
196 /* Configure the lo mode according to the input parameter */
197 kni_net_config_lo_mode(lo_mode);
199 KNI_PRINT("######## DPDK kni module loaded ########\n");
204 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
205 unregister_pernet_subsys(&kni_net_ops);
207 unregister_pernet_gen_subsys(kni_net_id, &kni_net_ops);
215 misc_deregister(&kni_misc);
216 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
217 unregister_pernet_subsys(&kni_net_ops);
219 unregister_pernet_gen_subsys(kni_net_id, &kni_net_ops);
221 KNI_PRINT("####### DPDK kni module unloaded #######\n");
225 kni_parse_kthread_mode(void)
230 if (strcmp(kthread_mode, "single") == 0)
232 else if (strcmp(kthread_mode, "multiple") == 0)
233 multiple_kthread_on = 1;
241 kni_open(struct inode *inode, struct file *file)
243 struct net *net = current->nsproxy->net_ns;
244 struct kni_net *knet = net_generic(net, kni_net_id);
246 /* kni device can be opened by one user only per netns */
247 if (test_and_set_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use))
250 file->private_data = get_net(net);
251 KNI_PRINT("/dev/kni opened\n");
257 kni_release(struct inode *inode, struct file *file)
259 struct net *net = file->private_data;
260 struct kni_net *knet = net_generic(net, kni_net_id);
261 struct kni_dev *dev, *n;
263 /* Stop kernel thread for single mode */
264 if (multiple_kthread_on == 0) {
265 mutex_lock(&knet->kni_kthread_lock);
266 /* Stop kernel thread */
267 if (knet->kni_kthread != NULL) {
268 kthread_stop(knet->kni_kthread);
269 knet->kni_kthread = NULL;
271 mutex_unlock(&knet->kni_kthread_lock);
274 down_write(&knet->kni_list_lock);
275 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
276 /* Stop kernel thread for multiple mode */
277 if (multiple_kthread_on && dev->pthread != NULL) {
278 kthread_stop(dev->pthread);
283 kni_vhost_backend_release(dev);
286 list_del(&dev->list);
288 up_write(&knet->kni_list_lock);
290 /* Clear the bit of device in use */
291 clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use);
294 KNI_PRINT("/dev/kni closed\n");
300 kni_thread_single(void *data)
302 struct kni_net *knet = data;
306 while (!kthread_should_stop()) {
307 down_read(&knet->kni_list_lock);
308 for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
309 list_for_each_entry(dev, &knet->kni_list_head, list) {
311 kni_chk_vhost_rx(dev);
315 kni_net_poll_resp(dev);
318 up_read(&knet->kni_list_lock);
319 #ifdef RTE_KNI_PREEMPT_DEFAULT
320 /* reschedule out for a while */
321 schedule_timeout_interruptible(usecs_to_jiffies( \
322 KNI_KTHREAD_RESCHEDULE_INTERVAL));
330 kni_thread_multiple(void *param)
333 struct kni_dev *dev = (struct kni_dev *)param;
335 while (!kthread_should_stop()) {
336 for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
338 kni_chk_vhost_rx(dev);
342 kni_net_poll_resp(dev);
344 #ifdef RTE_KNI_PREEMPT_DEFAULT
345 schedule_timeout_interruptible(usecs_to_jiffies( \
346 KNI_KTHREAD_RESCHEDULE_INTERVAL));
354 kni_dev_remove(struct kni_dev *dev)
360 if (pci_match_id(ixgbe_pci_tbl, dev->pci_dev))
361 ixgbe_kni_remove(dev->pci_dev);
362 else if (pci_match_id(igb_pci_tbl, dev->pci_dev))
363 igb_kni_remove(dev->pci_dev);
367 unregister_netdev(dev->net_dev);
368 free_netdev(dev->net_dev);
375 kni_check_param(struct kni_dev *kni, struct rte_kni_device_info *dev)
380 /* Check if network name has been used */
381 if (!strncmp(kni->name, dev->name, RTE_KNI_NAMESIZE)) {
382 KNI_ERR("KNI name %s duplicated\n", dev->name);
390 kni_run_thread(struct kni_net *knet, struct kni_dev *kni, uint8_t force_bind)
393 * Create a new kernel thread for multiple mode, set its core affinity,
394 * and finally wake it up.
396 if (multiple_kthread_on) {
397 kni->pthread = kthread_create(kni_thread_multiple,
398 (void *)kni, "kni_%s", kni->name);
399 if (IS_ERR(kni->pthread)) {
405 kthread_bind(kni->pthread, kni->core_id);
406 wake_up_process(kni->pthread);
408 mutex_lock(&knet->kni_kthread_lock);
410 if (knet->kni_kthread == NULL) {
411 knet->kni_kthread = kthread_create(kni_thread_single,
412 (void *)knet, "kni_single");
413 if (IS_ERR(knet->kni_kthread)) {
414 mutex_unlock(&knet->kni_kthread_lock);
420 kthread_bind(knet->kni_kthread, kni->core_id);
421 wake_up_process(knet->kni_kthread);
424 mutex_unlock(&knet->kni_kthread_lock);
431 kni_ioctl_create(struct net *net,
432 unsigned int ioctl_num, unsigned long ioctl_param)
434 struct kni_net *knet = net_generic(net, kni_net_id);
436 struct rte_kni_device_info dev_info;
437 struct pci_dev *pci = NULL;
438 struct pci_dev *found_pci = NULL;
439 struct net_device *net_dev = NULL;
440 struct net_device *lad_dev = NULL;
441 struct kni_dev *kni, *dev, *n;
443 printk(KERN_INFO "KNI: Creating kni...\n");
444 /* Check the buffer size, to avoid warning */
445 if (_IOC_SIZE(ioctl_num) > sizeof(dev_info))
448 /* Copy kni info from user space */
449 ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info));
451 KNI_ERR("copy_from_user in kni_ioctl_create");
456 * Check if the cpu core id is valid for binding.
458 if (dev_info.force_bind && !cpu_online(dev_info.core_id)) {
459 KNI_ERR("cpu %u is not online\n", dev_info.core_id);
463 /* Check if it has been created */
464 down_read(&knet->kni_list_lock);
465 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
466 if (kni_check_param(dev, &dev_info) < 0) {
467 up_read(&knet->kni_list_lock);
471 up_read(&knet->kni_list_lock);
473 net_dev = alloc_netdev(sizeof(struct kni_dev), dev_info.name,
474 #ifdef NET_NAME_UNKNOWN
478 if (net_dev == NULL) {
479 KNI_ERR("error allocating device \"%s\"\n", dev_info.name);
483 dev_net_set(net_dev, net);
485 kni = netdev_priv(net_dev);
487 kni->net_dev = net_dev;
488 kni->group_id = dev_info.group_id;
489 kni->core_id = dev_info.core_id;
490 strncpy(kni->name, dev_info.name, RTE_KNI_NAMESIZE);
492 /* Translate user space info into kernel space info */
493 kni->tx_q = phys_to_virt(dev_info.tx_phys);
494 kni->rx_q = phys_to_virt(dev_info.rx_phys);
495 kni->alloc_q = phys_to_virt(dev_info.alloc_phys);
496 kni->free_q = phys_to_virt(dev_info.free_phys);
498 kni->req_q = phys_to_virt(dev_info.req_phys);
499 kni->resp_q = phys_to_virt(dev_info.resp_phys);
500 kni->sync_va = dev_info.sync_va;
501 kni->sync_kva = phys_to_virt(dev_info.sync_phys);
504 kni->vhost_queue = NULL;
505 kni->vq_status = BE_STOP;
507 kni->mbuf_size = dev_info.mbuf_size;
509 KNI_PRINT("tx_phys: 0x%016llx, tx_q addr: 0x%p\n",
510 (unsigned long long) dev_info.tx_phys, kni->tx_q);
511 KNI_PRINT("rx_phys: 0x%016llx, rx_q addr: 0x%p\n",
512 (unsigned long long) dev_info.rx_phys, kni->rx_q);
513 KNI_PRINT("alloc_phys: 0x%016llx, alloc_q addr: 0x%p\n",
514 (unsigned long long) dev_info.alloc_phys, kni->alloc_q);
515 KNI_PRINT("free_phys: 0x%016llx, free_q addr: 0x%p\n",
516 (unsigned long long) dev_info.free_phys, kni->free_q);
517 KNI_PRINT("req_phys: 0x%016llx, req_q addr: 0x%p\n",
518 (unsigned long long) dev_info.req_phys, kni->req_q);
519 KNI_PRINT("resp_phys: 0x%016llx, resp_q addr: 0x%p\n",
520 (unsigned long long) dev_info.resp_phys, kni->resp_q);
521 KNI_PRINT("mbuf_size: %u\n", kni->mbuf_size);
523 KNI_DBG("PCI: %02x:%02x.%02x %04x:%04x\n",
530 pci = pci_get_device(dev_info.vendor_id, dev_info.device_id, NULL);
532 /* Support Ethtool */
534 KNI_PRINT("pci_bus: %02x:%02x:%02x \n",
536 PCI_SLOT(pci->devfn),
537 PCI_FUNC(pci->devfn));
539 if ((pci->bus->number == dev_info.bus) &&
540 (PCI_SLOT(pci->devfn) == dev_info.devid) &&
541 (PCI_FUNC(pci->devfn) == dev_info.function)) {
544 if (pci_match_id(ixgbe_pci_tbl, found_pci))
545 ret = ixgbe_kni_probe(found_pci, &lad_dev);
546 else if (pci_match_id(igb_pci_tbl, found_pci))
547 ret = igb_kni_probe(found_pci, &lad_dev);
551 KNI_DBG("PCI found: pci=0x%p, lad_dev=0x%p\n",
554 kni->lad_dev = lad_dev;
555 kni_set_ethtool_ops(kni->net_dev);
557 KNI_ERR("Device not supported by ethtool");
561 kni->pci_dev = found_pci;
562 kni->device_id = dev_info.device_id;
565 pci = pci_get_device(dev_info.vendor_id,
566 dev_info.device_id, pci);
572 memcpy(net_dev->dev_addr, kni->lad_dev->dev_addr, ETH_ALEN);
575 * Generate random mac address. eth_random_addr() is the newer
576 * version of generating mac address in linux kernel.
578 random_ether_addr(net_dev->dev_addr);
580 ret = register_netdev(net_dev);
582 KNI_ERR("error %i registering device \"%s\"\n",
586 free_netdev(net_dev);
594 ret = kni_run_thread(knet, kni, dev_info.force_bind);
598 down_write(&knet->kni_list_lock);
599 list_add(&kni->list, &knet->kni_list_head);
600 up_write(&knet->kni_list_lock);
606 kni_ioctl_release(struct net *net,
607 unsigned int ioctl_num, unsigned long ioctl_param)
609 struct kni_net *knet = net_generic(net, kni_net_id);
611 struct kni_dev *dev, *n;
612 struct rte_kni_device_info dev_info;
614 if (_IOC_SIZE(ioctl_num) > sizeof(dev_info))
617 ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info));
619 KNI_ERR("copy_from_user in kni_ioctl_release");
623 /* Release the network device according to its name */
624 if (strlen(dev_info.name) == 0)
627 down_write(&knet->kni_list_lock);
628 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
629 if (strncmp(dev->name, dev_info.name, RTE_KNI_NAMESIZE) != 0)
632 if (multiple_kthread_on && dev->pthread != NULL) {
633 kthread_stop(dev->pthread);
638 kni_vhost_backend_release(dev);
641 list_del(&dev->list);
645 up_write(&knet->kni_list_lock);
646 printk(KERN_INFO "KNI: %s release kni named %s\n",
647 (ret == 0 ? "Successfully" : "Unsuccessfully"), dev_info.name);
653 kni_ioctl(struct inode *inode,
654 unsigned int ioctl_num,
655 unsigned long ioctl_param)
658 struct net *net = current->nsproxy->net_ns;
660 KNI_DBG("IOCTL num=0x%0x param=0x%0lx\n", ioctl_num, ioctl_param);
663 * Switch according to the ioctl called
665 switch (_IOC_NR(ioctl_num)) {
666 case _IOC_NR(RTE_KNI_IOCTL_TEST):
667 /* For test only, not used */
669 case _IOC_NR(RTE_KNI_IOCTL_CREATE):
670 ret = kni_ioctl_create(net, ioctl_num, ioctl_param);
672 case _IOC_NR(RTE_KNI_IOCTL_RELEASE):
673 ret = kni_ioctl_release(net, ioctl_num, ioctl_param);
676 KNI_DBG("IOCTL default\n");
684 kni_compat_ioctl(struct inode *inode,
685 unsigned int ioctl_num,
686 unsigned long ioctl_param)
688 /* 32 bits app on 64 bits OS to be supported later */
689 KNI_PRINT("Not implemented.\n");
694 module_init(kni_init);
695 module_exit(kni_exit);
697 module_param(lo_mode, charp, S_IRUGO | S_IWUSR);
698 MODULE_PARM_DESC(lo_mode,
699 "KNI loopback mode (default=lo_mode_none):\n"
700 " lo_mode_none Kernel loopback disabled\n"
701 " lo_mode_fifo Enable kernel loopback with fifo\n"
702 " lo_mode_fifo_skb Enable kernel loopback with fifo and skb buffer\n"
706 module_param(kthread_mode, charp, S_IRUGO);
707 MODULE_PARM_DESC(kthread_mode,
708 "Kernel thread mode (default=single):\n"
709 " single Single kernel thread mode enabled.\n"
710 " multiple Multiple kernel thread mode enabled.\n"