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);
61 extern struct pci_device_id igb_pci_tbl[];
63 static int kni_open(struct inode *inode, struct file *file);
64 static int kni_release(struct inode *inode, struct file *file);
65 static int kni_ioctl(struct inode *inode, unsigned int ioctl_num,
66 unsigned long ioctl_param);
67 static int kni_compat_ioctl(struct inode *inode, unsigned int ioctl_num,
68 unsigned long ioctl_param);
69 static int kni_dev_remove(struct kni_dev *dev);
71 static int __init kni_parse_kthread_mode(void);
73 /* KNI processing for single kernel thread mode */
74 static int kni_thread_single(void *unused);
75 /* KNI processing for multiple kernel thread mode */
76 static int kni_thread_multiple(void *param);
78 static struct file_operations kni_fops = {
81 .release = kni_release,
82 .unlocked_ioctl = (void *)kni_ioctl,
83 .compat_ioctl = (void *)kni_compat_ioctl,
86 static struct miscdevice kni_misc = {
87 .minor = MISC_DYNAMIC_MINOR,
93 static char *lo_mode = NULL;
95 /* Kernel thread mode */
96 static char *kthread_mode = NULL;
97 static unsigned multiple_kthread_on = 0;
99 #define KNI_DEV_IN_USE_BIT_NUM 0 /* Bit number for device in use */
101 static int kni_net_id;
104 unsigned long device_in_use; /* device in use flag */
105 struct task_struct *kni_kthread;
106 struct rw_semaphore kni_list_lock;
107 struct list_head kni_list_head;
110 static int __net_init kni_init_net(struct net *net)
112 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
113 struct kni_net *knet = net_generic(net, kni_net_id);
115 struct kni_net *knet;
118 knet = kmalloc(sizeof(struct kni_net), GFP_KERNEL);
125 /* Clear the bit of device in use */
126 clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use);
128 init_rwsem(&knet->kni_list_lock);
129 INIT_LIST_HEAD(&knet->kni_list_head);
131 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
134 ret = net_assign_generic(net, kni_net_id, knet);
142 static void __net_exit kni_exit_net(struct net *net)
144 #ifndef HAVE_SIMPLIFIED_PERNET_OPERATIONS
145 struct kni_net *knet = net_generic(net, kni_net_id);
151 static struct pernet_operations kni_net_ops = {
152 .init = kni_init_net,
153 .exit = kni_exit_net,
154 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
156 .size = sizeof(struct kni_net),
165 KNI_PRINT("######## DPDK kni module loading ########\n");
167 if (kni_parse_kthread_mode() < 0) {
168 KNI_ERR("Invalid parameter for kthread_mode\n");
172 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
173 rc = register_pernet_subsys(&kni_net_ops);
175 rc = register_pernet_gen_subsys(&kni_net_id, &kni_net_ops);
180 rc = misc_register(&kni_misc);
182 KNI_ERR("Misc registration failed\n");
186 /* Configure the lo mode according to the input parameter */
187 kni_net_config_lo_mode(lo_mode);
189 KNI_PRINT("######## DPDK kni module loaded ########\n");
194 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
195 unregister_pernet_subsys(&kni_net_ops);
197 unregister_pernet_gen_subsys(kni_net_id, &kni_net_ops);
205 misc_deregister(&kni_misc);
206 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
207 unregister_pernet_subsys(&kni_net_ops);
209 unregister_pernet_gen_subsys(kni_net_id, &kni_net_ops);
211 KNI_PRINT("####### DPDK kni module unloaded #######\n");
215 kni_parse_kthread_mode(void)
220 if (strcmp(kthread_mode, "single") == 0)
222 else if (strcmp(kthread_mode, "multiple") == 0)
223 multiple_kthread_on = 1;
231 kni_open(struct inode *inode, struct file *file)
233 struct net *net = current->nsproxy->net_ns;
234 struct kni_net *knet = net_generic(net, kni_net_id);
236 /* kni device can be opened by one user only per netns */
237 if (test_and_set_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use))
240 /* Create kernel thread for single mode */
241 if (multiple_kthread_on == 0) {
242 KNI_PRINT("Single kernel thread for all KNI devices\n");
243 /* Create kernel thread for RX */
244 knet->kni_kthread = kthread_run(kni_thread_single, (void *)knet,
246 if (IS_ERR(knet->kni_kthread)) {
247 KNI_ERR("Unable to create kernel threaed\n");
248 return PTR_ERR(knet->kni_kthread);
251 KNI_PRINT("Multiple kernel thread mode enabled\n");
253 file->private_data = get_net(net);
254 KNI_PRINT("/dev/kni opened\n");
260 kni_release(struct inode *inode, struct file *file)
262 struct net *net = file->private_data;
263 struct kni_net *knet = net_generic(net, kni_net_id);
264 struct kni_dev *dev, *n;
266 /* Stop kernel thread for single mode */
267 if (multiple_kthread_on == 0) {
268 /* Stop kernel thread */
269 kthread_stop(knet->kni_kthread);
270 knet->kni_kthread = NULL;
273 down_write(&knet->kni_list_lock);
274 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
275 /* Stop kernel thread for multiple mode */
276 if (multiple_kthread_on && dev->pthread != NULL) {
277 kthread_stop(dev->pthread);
282 kni_vhost_backend_release(dev);
285 list_del(&dev->list);
287 up_write(&knet->kni_list_lock);
289 /* Clear the bit of device in use */
290 clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use);
293 KNI_PRINT("/dev/kni closed\n");
299 kni_thread_single(void *data)
301 struct kni_net *knet = data;
305 while (!kthread_should_stop()) {
306 down_read(&knet->kni_list_lock);
307 for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
308 list_for_each_entry(dev, &knet->kni_list_head, list) {
310 kni_chk_vhost_rx(dev);
314 kni_net_poll_resp(dev);
317 up_read(&knet->kni_list_lock);
318 #ifdef RTE_KNI_PREEMPT_DEFAULT
319 /* reschedule out for a while */
320 schedule_timeout_interruptible(usecs_to_jiffies( \
321 KNI_KTHREAD_RESCHEDULE_INTERVAL));
329 kni_thread_multiple(void *param)
332 struct kni_dev *dev = (struct kni_dev *)param;
334 while (!kthread_should_stop()) {
335 for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
337 kni_chk_vhost_rx(dev);
341 kni_net_poll_resp(dev);
343 #ifdef RTE_KNI_PREEMPT_DEFAULT
344 schedule_timeout_interruptible(usecs_to_jiffies( \
345 KNI_KTHREAD_RESCHEDULE_INTERVAL));
353 kni_dev_remove(struct kni_dev *dev)
359 if (pci_match_id(ixgbe_pci_tbl, dev->pci_dev))
360 ixgbe_kni_remove(dev->pci_dev);
361 else if (pci_match_id(igb_pci_tbl, dev->pci_dev))
362 igb_kni_remove(dev->pci_dev);
366 unregister_netdev(dev->net_dev);
367 free_netdev(dev->net_dev);
374 kni_check_param(struct kni_dev *kni, struct rte_kni_device_info *dev)
379 /* Check if network name has been used */
380 if (!strncmp(kni->name, dev->name, RTE_KNI_NAMESIZE)) {
381 KNI_ERR("KNI name %s duplicated\n", dev->name);
389 kni_ioctl_create(struct net *net,
390 unsigned int ioctl_num, unsigned long ioctl_param)
392 struct kni_net *knet = net_generic(net, kni_net_id);
394 struct rte_kni_device_info dev_info;
395 struct pci_dev *pci = NULL;
396 struct pci_dev *found_pci = NULL;
397 struct net_device *net_dev = NULL;
398 struct net_device *lad_dev = NULL;
399 struct kni_dev *kni, *dev, *n;
401 printk(KERN_INFO "KNI: Creating kni...\n");
402 /* Check the buffer size, to avoid warning */
403 if (_IOC_SIZE(ioctl_num) > sizeof(dev_info))
406 /* Copy kni info from user space */
407 ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info));
409 KNI_ERR("copy_from_user in kni_ioctl_create");
414 * Check if the cpu core id is valid for binding,
415 * for multiple kernel thread mode.
417 if (multiple_kthread_on && dev_info.force_bind &&
418 !cpu_online(dev_info.core_id)) {
419 KNI_ERR("cpu %u is not online\n", dev_info.core_id);
423 /* Check if it has been created */
424 down_read(&knet->kni_list_lock);
425 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
426 if (kni_check_param(dev, &dev_info) < 0) {
427 up_read(&knet->kni_list_lock);
431 up_read(&knet->kni_list_lock);
433 net_dev = alloc_netdev(sizeof(struct kni_dev), dev_info.name,
434 #ifdef NET_NAME_UNKNOWN
438 if (net_dev == NULL) {
439 KNI_ERR("error allocating device \"%s\"\n", dev_info.name);
443 dev_net_set(net_dev, net);
445 kni = netdev_priv(net_dev);
447 kni->net_dev = net_dev;
448 kni->group_id = dev_info.group_id;
449 kni->core_id = dev_info.core_id;
450 strncpy(kni->name, dev_info.name, RTE_KNI_NAMESIZE);
452 /* Translate user space info into kernel space info */
453 kni->tx_q = phys_to_virt(dev_info.tx_phys);
454 kni->rx_q = phys_to_virt(dev_info.rx_phys);
455 kni->alloc_q = phys_to_virt(dev_info.alloc_phys);
456 kni->free_q = phys_to_virt(dev_info.free_phys);
458 kni->req_q = phys_to_virt(dev_info.req_phys);
459 kni->resp_q = phys_to_virt(dev_info.resp_phys);
460 kni->sync_va = dev_info.sync_va;
461 kni->sync_kva = phys_to_virt(dev_info.sync_phys);
463 kni->mbuf_kva = phys_to_virt(dev_info.mbuf_phys);
464 kni->mbuf_va = dev_info.mbuf_va;
467 kni->vhost_queue = NULL;
468 kni->vq_status = BE_STOP;
470 kni->mbuf_size = dev_info.mbuf_size;
472 KNI_PRINT("tx_phys: 0x%016llx, tx_q addr: 0x%p\n",
473 (unsigned long long) dev_info.tx_phys, kni->tx_q);
474 KNI_PRINT("rx_phys: 0x%016llx, rx_q addr: 0x%p\n",
475 (unsigned long long) dev_info.rx_phys, kni->rx_q);
476 KNI_PRINT("alloc_phys: 0x%016llx, alloc_q addr: 0x%p\n",
477 (unsigned long long) dev_info.alloc_phys, kni->alloc_q);
478 KNI_PRINT("free_phys: 0x%016llx, free_q addr: 0x%p\n",
479 (unsigned long long) dev_info.free_phys, kni->free_q);
480 KNI_PRINT("req_phys: 0x%016llx, req_q addr: 0x%p\n",
481 (unsigned long long) dev_info.req_phys, kni->req_q);
482 KNI_PRINT("resp_phys: 0x%016llx, resp_q addr: 0x%p\n",
483 (unsigned long long) dev_info.resp_phys, kni->resp_q);
484 KNI_PRINT("mbuf_phys: 0x%016llx, mbuf_kva: 0x%p\n",
485 (unsigned long long) dev_info.mbuf_phys, kni->mbuf_kva);
486 KNI_PRINT("mbuf_va: 0x%p\n", dev_info.mbuf_va);
487 KNI_PRINT("mbuf_size: %u\n", kni->mbuf_size);
489 KNI_DBG("PCI: %02x:%02x.%02x %04x:%04x\n",
496 pci = pci_get_device(dev_info.vendor_id, dev_info.device_id, NULL);
498 /* Support Ethtool */
500 KNI_PRINT("pci_bus: %02x:%02x:%02x \n",
502 PCI_SLOT(pci->devfn),
503 PCI_FUNC(pci->devfn));
505 if ((pci->bus->number == dev_info.bus) &&
506 (PCI_SLOT(pci->devfn) == dev_info.devid) &&
507 (PCI_FUNC(pci->devfn) == dev_info.function)) {
510 if (pci_match_id(ixgbe_pci_tbl, found_pci))
511 ret = ixgbe_kni_probe(found_pci, &lad_dev);
512 else if (pci_match_id(igb_pci_tbl, found_pci))
513 ret = igb_kni_probe(found_pci, &lad_dev);
517 KNI_DBG("PCI found: pci=0x%p, lad_dev=0x%p\n",
520 kni->lad_dev = lad_dev;
521 kni_set_ethtool_ops(kni->net_dev);
523 KNI_ERR("Device not supported by ethtool");
527 kni->pci_dev = found_pci;
528 kni->device_id = dev_info.device_id;
531 pci = pci_get_device(dev_info.vendor_id,
532 dev_info.device_id, pci);
538 memcpy(net_dev->dev_addr, kni->lad_dev->dev_addr, ETH_ALEN);
541 * Generate random mac address. eth_random_addr() is the newer
542 * version of generating mac address in linux kernel.
544 random_ether_addr(net_dev->dev_addr);
546 ret = register_netdev(net_dev);
548 KNI_ERR("error %i registering device \"%s\"\n",
552 free_netdev(net_dev);
561 * Create a new kernel thread for multiple mode, set its core affinity,
562 * and finally wake it up.
564 if (multiple_kthread_on) {
565 kni->pthread = kthread_create(kni_thread_multiple,
567 "kni_%s", kni->name);
568 if (IS_ERR(kni->pthread)) {
572 if (dev_info.force_bind)
573 kthread_bind(kni->pthread, kni->core_id);
574 wake_up_process(kni->pthread);
577 down_write(&knet->kni_list_lock);
578 list_add(&kni->list, &knet->kni_list_head);
579 up_write(&knet->kni_list_lock);
585 kni_ioctl_release(struct net *net,
586 unsigned int ioctl_num, unsigned long ioctl_param)
588 struct kni_net *knet = net_generic(net, kni_net_id);
590 struct kni_dev *dev, *n;
591 struct rte_kni_device_info dev_info;
593 if (_IOC_SIZE(ioctl_num) > sizeof(dev_info))
596 ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info));
598 KNI_ERR("copy_from_user in kni_ioctl_release");
602 /* Release the network device according to its name */
603 if (strlen(dev_info.name) == 0)
606 down_write(&knet->kni_list_lock);
607 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
608 if (strncmp(dev->name, dev_info.name, RTE_KNI_NAMESIZE) != 0)
611 if (multiple_kthread_on && dev->pthread != NULL) {
612 kthread_stop(dev->pthread);
617 kni_vhost_backend_release(dev);
620 list_del(&dev->list);
624 up_write(&knet->kni_list_lock);
625 printk(KERN_INFO "KNI: %s release kni named %s\n",
626 (ret == 0 ? "Successfully" : "Unsuccessfully"), dev_info.name);
632 kni_ioctl(struct inode *inode,
633 unsigned int ioctl_num,
634 unsigned long ioctl_param)
637 struct net *net = current->nsproxy->net_ns;
639 KNI_DBG("IOCTL num=0x%0x param=0x%0lx\n", ioctl_num, ioctl_param);
642 * Switch according to the ioctl called
644 switch (_IOC_NR(ioctl_num)) {
645 case _IOC_NR(RTE_KNI_IOCTL_TEST):
646 /* For test only, not used */
648 case _IOC_NR(RTE_KNI_IOCTL_CREATE):
649 ret = kni_ioctl_create(net, ioctl_num, ioctl_param);
651 case _IOC_NR(RTE_KNI_IOCTL_RELEASE):
652 ret = kni_ioctl_release(net, ioctl_num, ioctl_param);
655 KNI_DBG("IOCTL default\n");
663 kni_compat_ioctl(struct inode *inode,
664 unsigned int ioctl_num,
665 unsigned long ioctl_param)
667 /* 32 bits app on 64 bits OS to be supported later */
668 KNI_PRINT("Not implemented.\n");
673 module_init(kni_init);
674 module_exit(kni_exit);
676 module_param(lo_mode, charp, S_IRUGO | S_IWUSR);
677 MODULE_PARM_DESC(lo_mode,
678 "KNI loopback mode (default=lo_mode_none):\n"
679 " lo_mode_none Kernel loopback disabled\n"
680 " lo_mode_fifo Enable kernel loopback with fifo\n"
681 " lo_mode_fifo_skb Enable kernel loopback with fifo and skb buffer\n"
685 module_param(kthread_mode, charp, S_IRUGO);
686 MODULE_PARM_DESC(kthread_mode,
687 "Kernel thread mode (default=single):\n"
688 " single Single kernel thread mode enabled.\n"
689 " multiple Multiple kernel thread mode enabled.\n"