1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright(c) 2010-2014 Intel Corporation.
6 #include <linux/version.h>
7 #include <linux/module.h>
8 #include <linux/miscdevice.h>
9 #include <linux/netdevice.h>
10 #include <linux/etherdevice.h>
11 #include <linux/pci.h>
12 #include <linux/kthread.h>
13 #include <linux/rwsem.h>
14 #include <linux/mutex.h>
15 #include <linux/nsproxy.h>
16 #include <net/net_namespace.h>
17 #include <net/netns/generic.h>
19 #include <exec-env/rte_kni_common.h>
24 MODULE_LICENSE("Dual BSD/GPL");
25 MODULE_AUTHOR("Intel Corporation");
26 MODULE_DESCRIPTION("Kernel Module for managing kni devices");
28 #define KNI_RX_LOOP_NUM 1000
30 #define KNI_MAX_DEVICES 32
32 extern const struct pci_device_id ixgbe_pci_tbl[];
33 extern const struct pci_device_id igb_pci_tbl[];
38 /* Kernel thread mode */
39 static char *kthread_mode;
40 static uint32_t multiple_kthread_on;
42 #define KNI_DEV_IN_USE_BIT_NUM 0 /* Bit number for device in use */
44 static int kni_net_id;
47 unsigned long device_in_use; /* device in use flag */
48 struct mutex kni_kthread_lock;
49 struct task_struct *kni_kthread;
50 struct rw_semaphore kni_list_lock;
51 struct list_head kni_list_head;
55 kni_init_net(struct net *net)
57 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
58 struct kni_net *knet = net_generic(net, kni_net_id);
60 memset(knet, 0, sizeof(*knet));
65 knet = kzalloc(sizeof(struct kni_net), GFP_KERNEL);
72 /* Clear the bit of device in use */
73 clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use);
75 mutex_init(&knet->kni_kthread_lock);
77 init_rwsem(&knet->kni_list_lock);
78 INIT_LIST_HEAD(&knet->kni_list_head);
80 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
83 ret = net_assign_generic(net, kni_net_id, knet);
91 static void __net_exit
92 kni_exit_net(struct net *net)
94 struct kni_net *knet __maybe_unused;
96 knet = net_generic(net, kni_net_id);
97 mutex_destroy(&knet->kni_kthread_lock);
99 #ifndef HAVE_SIMPLIFIED_PERNET_OPERATIONS
104 static struct pernet_operations kni_net_ops = {
105 .init = kni_init_net,
106 .exit = kni_exit_net,
107 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
109 .size = sizeof(struct kni_net),
114 kni_thread_single(void *data)
116 struct kni_net *knet = data;
120 while (!kthread_should_stop()) {
121 down_read(&knet->kni_list_lock);
122 for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
123 list_for_each_entry(dev, &knet->kni_list_head, list) {
125 kni_net_poll_resp(dev);
128 up_read(&knet->kni_list_lock);
129 #ifdef RTE_KNI_PREEMPT_DEFAULT
130 /* reschedule out for a while */
131 schedule_timeout_interruptible(
132 usecs_to_jiffies(KNI_KTHREAD_RESCHEDULE_INTERVAL));
140 kni_thread_multiple(void *param)
143 struct kni_dev *dev = param;
145 while (!kthread_should_stop()) {
146 for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
148 kni_net_poll_resp(dev);
150 #ifdef RTE_KNI_PREEMPT_DEFAULT
151 schedule_timeout_interruptible(
152 usecs_to_jiffies(KNI_KTHREAD_RESCHEDULE_INTERVAL));
160 kni_open(struct inode *inode, struct file *file)
162 struct net *net = current->nsproxy->net_ns;
163 struct kni_net *knet = net_generic(net, kni_net_id);
165 /* kni device can be opened by one user only per netns */
166 if (test_and_set_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use))
169 file->private_data = get_net(net);
170 pr_debug("/dev/kni opened\n");
176 kni_dev_remove(struct kni_dev *dev)
181 #ifdef RTE_KNI_KMOD_ETHTOOL
183 if (pci_match_id(ixgbe_pci_tbl, dev->pci_dev))
184 ixgbe_kni_remove(dev->pci_dev);
185 else if (pci_match_id(igb_pci_tbl, dev->pci_dev))
186 igb_kni_remove(dev->pci_dev);
191 unregister_netdev(dev->net_dev);
192 free_netdev(dev->net_dev);
195 kni_net_release_fifo_phy(dev);
201 kni_release(struct inode *inode, struct file *file)
203 struct net *net = file->private_data;
204 struct kni_net *knet = net_generic(net, kni_net_id);
205 struct kni_dev *dev, *n;
207 /* Stop kernel thread for single mode */
208 if (multiple_kthread_on == 0) {
209 mutex_lock(&knet->kni_kthread_lock);
210 /* Stop kernel thread */
211 if (knet->kni_kthread != NULL) {
212 kthread_stop(knet->kni_kthread);
213 knet->kni_kthread = NULL;
215 mutex_unlock(&knet->kni_kthread_lock);
218 down_write(&knet->kni_list_lock);
219 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
220 /* Stop kernel thread for multiple mode */
221 if (multiple_kthread_on && dev->pthread != NULL) {
222 kthread_stop(dev->pthread);
227 list_del(&dev->list);
229 up_write(&knet->kni_list_lock);
231 /* Clear the bit of device in use */
232 clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use);
235 pr_debug("/dev/kni closed\n");
241 kni_check_param(struct kni_dev *kni, struct rte_kni_device_info *dev)
246 /* Check if network name has been used */
247 if (!strncmp(kni->name, dev->name, RTE_KNI_NAMESIZE)) {
248 pr_err("KNI name %s duplicated\n", dev->name);
256 kni_run_thread(struct kni_net *knet, struct kni_dev *kni, uint8_t force_bind)
259 * Create a new kernel thread for multiple mode, set its core affinity,
260 * and finally wake it up.
262 if (multiple_kthread_on) {
263 kni->pthread = kthread_create(kni_thread_multiple,
264 (void *)kni, "kni_%s", kni->name);
265 if (IS_ERR(kni->pthread)) {
271 kthread_bind(kni->pthread, kni->core_id);
272 wake_up_process(kni->pthread);
274 mutex_lock(&knet->kni_kthread_lock);
276 if (knet->kni_kthread == NULL) {
277 knet->kni_kthread = kthread_create(kni_thread_single,
278 (void *)knet, "kni_single");
279 if (IS_ERR(knet->kni_kthread)) {
280 mutex_unlock(&knet->kni_kthread_lock);
286 kthread_bind(knet->kni_kthread, kni->core_id);
287 wake_up_process(knet->kni_kthread);
290 mutex_unlock(&knet->kni_kthread_lock);
297 kni_ioctl_create(struct net *net, uint32_t ioctl_num,
298 unsigned long ioctl_param)
300 struct kni_net *knet = net_generic(net, kni_net_id);
302 struct rte_kni_device_info dev_info;
303 struct net_device *net_dev = NULL;
304 struct kni_dev *kni, *dev, *n;
305 #ifdef RTE_KNI_KMOD_ETHTOOL
306 struct pci_dev *found_pci = NULL;
307 struct net_device *lad_dev = NULL;
308 struct pci_dev *pci = NULL;
311 pr_info("Creating kni...\n");
312 /* Check the buffer size, to avoid warning */
313 if (_IOC_SIZE(ioctl_num) > sizeof(dev_info))
316 /* Copy kni info from user space */
317 ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info));
319 pr_err("copy_from_user in kni_ioctl_create");
323 /* Check if name is zero-ended */
324 if (strnlen(dev_info.name, sizeof(dev_info.name)) == sizeof(dev_info.name)) {
325 pr_err("kni.name not zero-terminated");
330 * Check if the cpu core id is valid for binding.
332 if (dev_info.force_bind && !cpu_online(dev_info.core_id)) {
333 pr_err("cpu %u is not online\n", dev_info.core_id);
337 /* Check if it has been created */
338 down_read(&knet->kni_list_lock);
339 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
340 if (kni_check_param(dev, &dev_info) < 0) {
341 up_read(&knet->kni_list_lock);
345 up_read(&knet->kni_list_lock);
347 net_dev = alloc_netdev(sizeof(struct kni_dev), dev_info.name,
352 if (net_dev == NULL) {
353 pr_err("error allocating device \"%s\"\n", dev_info.name);
357 dev_net_set(net_dev, net);
359 kni = netdev_priv(net_dev);
361 kni->net_dev = net_dev;
362 kni->group_id = dev_info.group_id;
363 kni->core_id = dev_info.core_id;
364 strncpy(kni->name, dev_info.name, RTE_KNI_NAMESIZE);
366 /* Translate user space info into kernel space info */
367 kni->tx_q = phys_to_virt(dev_info.tx_phys);
368 kni->rx_q = phys_to_virt(dev_info.rx_phys);
369 kni->alloc_q = phys_to_virt(dev_info.alloc_phys);
370 kni->free_q = phys_to_virt(dev_info.free_phys);
372 kni->req_q = phys_to_virt(dev_info.req_phys);
373 kni->resp_q = phys_to_virt(dev_info.resp_phys);
374 kni->sync_va = dev_info.sync_va;
375 kni->sync_kva = phys_to_virt(dev_info.sync_phys);
377 kni->mbuf_size = dev_info.mbuf_size;
379 pr_debug("tx_phys: 0x%016llx, tx_q addr: 0x%p\n",
380 (unsigned long long) dev_info.tx_phys, kni->tx_q);
381 pr_debug("rx_phys: 0x%016llx, rx_q addr: 0x%p\n",
382 (unsigned long long) dev_info.rx_phys, kni->rx_q);
383 pr_debug("alloc_phys: 0x%016llx, alloc_q addr: 0x%p\n",
384 (unsigned long long) dev_info.alloc_phys, kni->alloc_q);
385 pr_debug("free_phys: 0x%016llx, free_q addr: 0x%p\n",
386 (unsigned long long) dev_info.free_phys, kni->free_q);
387 pr_debug("req_phys: 0x%016llx, req_q addr: 0x%p\n",
388 (unsigned long long) dev_info.req_phys, kni->req_q);
389 pr_debug("resp_phys: 0x%016llx, resp_q addr: 0x%p\n",
390 (unsigned long long) dev_info.resp_phys, kni->resp_q);
391 pr_debug("mbuf_size: %u\n", kni->mbuf_size);
393 pr_debug("PCI: %02x:%02x.%02x %04x:%04x\n",
399 #ifdef RTE_KNI_KMOD_ETHTOOL
400 pci = pci_get_device(dev_info.vendor_id, dev_info.device_id, NULL);
402 /* Support Ethtool */
404 pr_debug("pci_bus: %02x:%02x:%02x\n",
406 PCI_SLOT(pci->devfn),
407 PCI_FUNC(pci->devfn));
409 if ((pci->bus->number == dev_info.bus) &&
410 (PCI_SLOT(pci->devfn) == dev_info.devid) &&
411 (PCI_FUNC(pci->devfn) == dev_info.function)) {
414 if (pci_match_id(ixgbe_pci_tbl, found_pci))
415 ret = ixgbe_kni_probe(found_pci, &lad_dev);
416 else if (pci_match_id(igb_pci_tbl, found_pci))
417 ret = igb_kni_probe(found_pci, &lad_dev);
421 pr_debug("PCI found: pci=0x%p, lad_dev=0x%p\n",
424 kni->lad_dev = lad_dev;
425 kni_set_ethtool_ops(kni->net_dev);
427 pr_err("Device not supported by ethtool");
431 kni->pci_dev = found_pci;
432 kni->device_id = dev_info.device_id;
435 pci = pci_get_device(dev_info.vendor_id,
436 dev_info.device_id, pci);
443 ether_addr_copy(net_dev->dev_addr, kni->lad_dev->dev_addr);
445 /* if user has provided a valid mac address */
446 if (is_valid_ether_addr((unsigned char *)(dev_info.mac_addr)))
447 memcpy(net_dev->dev_addr, dev_info.mac_addr, ETH_ALEN);
450 * Generate random mac address. eth_random_addr() is the
451 * newer version of generating mac address in kernel.
453 random_ether_addr(net_dev->dev_addr);
457 net_dev->mtu = dev_info.mtu;
459 ret = register_netdev(net_dev);
461 pr_err("error %i registering device \"%s\"\n",
465 free_netdev(net_dev);
469 ret = kni_run_thread(knet, kni, dev_info.force_bind);
473 down_write(&knet->kni_list_lock);
474 list_add(&kni->list, &knet->kni_list_head);
475 up_write(&knet->kni_list_lock);
481 kni_ioctl_release(struct net *net, uint32_t ioctl_num,
482 unsigned long ioctl_param)
484 struct kni_net *knet = net_generic(net, kni_net_id);
486 struct kni_dev *dev, *n;
487 struct rte_kni_device_info dev_info;
489 if (_IOC_SIZE(ioctl_num) > sizeof(dev_info))
492 ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info));
494 pr_err("copy_from_user in kni_ioctl_release");
498 /* Release the network device according to its name */
499 if (strlen(dev_info.name) == 0)
502 down_write(&knet->kni_list_lock);
503 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
504 if (strncmp(dev->name, dev_info.name, RTE_KNI_NAMESIZE) != 0)
507 if (multiple_kthread_on && dev->pthread != NULL) {
508 kthread_stop(dev->pthread);
513 list_del(&dev->list);
517 up_write(&knet->kni_list_lock);
518 pr_info("%s release kni named %s\n",
519 (ret == 0 ? "Successfully" : "Unsuccessfully"), dev_info.name);
525 kni_ioctl(struct inode *inode, uint32_t ioctl_num, unsigned long ioctl_param)
528 struct net *net = current->nsproxy->net_ns;
530 pr_debug("IOCTL num=0x%0x param=0x%0lx\n", ioctl_num, ioctl_param);
533 * Switch according to the ioctl called
535 switch (_IOC_NR(ioctl_num)) {
536 case _IOC_NR(RTE_KNI_IOCTL_TEST):
537 /* For test only, not used */
539 case _IOC_NR(RTE_KNI_IOCTL_CREATE):
540 ret = kni_ioctl_create(net, ioctl_num, ioctl_param);
542 case _IOC_NR(RTE_KNI_IOCTL_RELEASE):
543 ret = kni_ioctl_release(net, ioctl_num, ioctl_param);
546 pr_debug("IOCTL default\n");
554 kni_compat_ioctl(struct inode *inode, uint32_t ioctl_num,
555 unsigned long ioctl_param)
557 /* 32 bits app on 64 bits OS to be supported later */
558 pr_debug("Not implemented.\n");
563 static const struct file_operations kni_fops = {
564 .owner = THIS_MODULE,
566 .release = kni_release,
567 .unlocked_ioctl = (void *)kni_ioctl,
568 .compat_ioctl = (void *)kni_compat_ioctl,
571 static struct miscdevice kni_misc = {
572 .minor = MISC_DYNAMIC_MINOR,
578 kni_parse_kthread_mode(void)
583 if (strcmp(kthread_mode, "single") == 0)
585 else if (strcmp(kthread_mode, "multiple") == 0)
586 multiple_kthread_on = 1;
598 if (kni_parse_kthread_mode() < 0) {
599 pr_err("Invalid parameter for kthread_mode\n");
603 if (multiple_kthread_on == 0)
604 pr_debug("Single kernel thread for all KNI devices\n");
606 pr_debug("Multiple kernel thread mode enabled\n");
608 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
609 rc = register_pernet_subsys(&kni_net_ops);
611 rc = register_pernet_gen_subsys(&kni_net_id, &kni_net_ops);
616 rc = misc_register(&kni_misc);
618 pr_err("Misc registration failed\n");
622 /* Configure the lo mode according to the input parameter */
623 kni_net_config_lo_mode(lo_mode);
628 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
629 unregister_pernet_subsys(&kni_net_ops);
631 unregister_pernet_gen_subsys(kni_net_id, &kni_net_ops);
639 misc_deregister(&kni_misc);
640 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
641 unregister_pernet_subsys(&kni_net_ops);
643 unregister_pernet_gen_subsys(kni_net_id, &kni_net_ops);
647 module_init(kni_init);
648 module_exit(kni_exit);
650 module_param(lo_mode, charp, S_IRUGO | S_IWUSR);
651 MODULE_PARM_DESC(lo_mode,
652 "KNI loopback mode (default=lo_mode_none):\n"
653 " lo_mode_none Kernel loopback disabled\n"
654 " lo_mode_fifo Enable kernel loopback with fifo\n"
655 " lo_mode_fifo_skb Enable kernel loopback with fifo and skb buffer\n"
659 module_param(kthread_mode, charp, S_IRUGO);
660 MODULE_PARM_DESC(kthread_mode,
661 "Kernel thread mode (default=single):\n"
662 " single Single kernel thread mode enabled.\n"
663 " multiple Multiple kernel thread mode enabled.\n"