1 /* Intel Ethernet Switch Host Interface Driver
2 * Copyright(c) 2013 - 2014 Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * The full GNU General Public License is included in this distribution in
14 * the file called "COPYING".
16 * Contact Information:
17 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
18 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
22 #include <linux/vmalloc.h>
23 #if IS_ENABLED(CONFIG_FM10K_VXLAN)
24 #include <net/vxlan.h>
25 #endif /* CONFIG_FM10K_VXLAN */
28 * fm10k_setup_tx_resources - allocate Tx resources (Descriptors)
29 * @tx_ring: tx descriptor ring (for a specific queue) to setup
31 * Return 0 on success, negative on failure
33 int fm10k_setup_tx_resources(struct fm10k_ring *tx_ring)
35 struct device *dev = tx_ring->dev;
38 size = sizeof(struct fm10k_tx_buffer) * tx_ring->count;
40 tx_ring->tx_buffer = vzalloc(size);
41 if (!tx_ring->tx_buffer)
44 u64_stats_init(&tx_ring->syncp);
46 /* round up to nearest 4K */
47 tx_ring->size = tx_ring->count * sizeof(struct fm10k_tx_desc);
48 tx_ring->size = ALIGN(tx_ring->size, 4096);
50 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
51 &tx_ring->dma, GFP_KERNEL);
58 vfree(tx_ring->tx_buffer);
59 tx_ring->tx_buffer = NULL;
64 * fm10k_setup_all_tx_resources - allocate all queues Tx resources
65 * @interface: board private structure
67 * If this function returns with an error, then it's possible one or
68 * more of the rings is populated (while the rest are not). It is the
69 * callers duty to clean those orphaned rings.
71 * Return 0 on success, negative on failure
73 static int fm10k_setup_all_tx_resources(struct fm10k_intfc *interface)
77 for (i = 0; i < interface->num_tx_queues; i++) {
78 err = fm10k_setup_tx_resources(interface->tx_ring[i]);
82 netif_err(interface, probe, interface->netdev,
83 "Allocation for Tx Queue %u failed\n", i);
89 /* rewind the index freeing the rings as we go */
91 fm10k_free_tx_resources(interface->tx_ring[i]);
96 * fm10k_setup_rx_resources - allocate Rx resources (Descriptors)
97 * @rx_ring: rx descriptor ring (for a specific queue) to setup
99 * Returns 0 on success, negative on failure
101 int fm10k_setup_rx_resources(struct fm10k_ring *rx_ring)
103 struct device *dev = rx_ring->dev;
106 size = sizeof(struct fm10k_rx_buffer) * rx_ring->count;
108 rx_ring->rx_buffer = vzalloc(size);
109 if (!rx_ring->rx_buffer)
112 u64_stats_init(&rx_ring->syncp);
114 /* Round up to nearest 4K */
115 rx_ring->size = rx_ring->count * sizeof(union fm10k_rx_desc);
116 rx_ring->size = ALIGN(rx_ring->size, 4096);
118 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
119 &rx_ring->dma, GFP_KERNEL);
125 vfree(rx_ring->rx_buffer);
126 rx_ring->rx_buffer = NULL;
131 * fm10k_setup_all_rx_resources - allocate all queues Rx resources
132 * @interface: board private structure
134 * If this function returns with an error, then it's possible one or
135 * more of the rings is populated (while the rest are not). It is the
136 * callers duty to clean those orphaned rings.
138 * Return 0 on success, negative on failure
140 static int fm10k_setup_all_rx_resources(struct fm10k_intfc *interface)
144 for (i = 0; i < interface->num_rx_queues; i++) {
145 err = fm10k_setup_rx_resources(interface->rx_ring[i]);
149 netif_err(interface, probe, interface->netdev,
150 "Allocation for Rx Queue %u failed\n", i);
156 /* rewind the index freeing the rings as we go */
158 fm10k_free_rx_resources(interface->rx_ring[i]);
162 void fm10k_unmap_and_free_tx_resource(struct fm10k_ring *ring,
163 struct fm10k_tx_buffer *tx_buffer)
165 if (tx_buffer->skb) {
166 dev_kfree_skb_any(tx_buffer->skb);
167 if (dma_unmap_len(tx_buffer, len))
168 dma_unmap_single(ring->dev,
169 dma_unmap_addr(tx_buffer, dma),
170 dma_unmap_len(tx_buffer, len),
172 } else if (dma_unmap_len(tx_buffer, len)) {
173 dma_unmap_page(ring->dev,
174 dma_unmap_addr(tx_buffer, dma),
175 dma_unmap_len(tx_buffer, len),
178 tx_buffer->next_to_watch = NULL;
179 tx_buffer->skb = NULL;
180 dma_unmap_len_set(tx_buffer, len, 0);
181 /* tx_buffer must be completely set up in the transmit path */
185 * fm10k_clean_tx_ring - Free Tx Buffers
186 * @tx_ring: ring to be cleaned
188 static void fm10k_clean_tx_ring(struct fm10k_ring *tx_ring)
190 struct fm10k_tx_buffer *tx_buffer;
194 /* ring already cleared, nothing to do */
195 if (!tx_ring->tx_buffer)
198 /* Free all the Tx ring sk_buffs */
199 for (i = 0; i < tx_ring->count; i++) {
200 tx_buffer = &tx_ring->tx_buffer[i];
201 fm10k_unmap_and_free_tx_resource(tx_ring, tx_buffer);
204 /* reset BQL values */
205 netdev_tx_reset_queue(txring_txq(tx_ring));
207 size = sizeof(struct fm10k_tx_buffer) * tx_ring->count;
208 memset(tx_ring->tx_buffer, 0, size);
210 /* Zero out the descriptor ring */
211 memset(tx_ring->desc, 0, tx_ring->size);
215 * fm10k_free_tx_resources - Free Tx Resources per Queue
216 * @tx_ring: Tx descriptor ring for a specific queue
218 * Free all transmit software resources
220 void fm10k_free_tx_resources(struct fm10k_ring *tx_ring)
222 fm10k_clean_tx_ring(tx_ring);
224 vfree(tx_ring->tx_buffer);
225 tx_ring->tx_buffer = NULL;
227 /* if not set, then don't free */
231 dma_free_coherent(tx_ring->dev, tx_ring->size,
232 tx_ring->desc, tx_ring->dma);
233 tx_ring->desc = NULL;
237 * fm10k_clean_all_tx_rings - Free Tx Buffers for all queues
238 * @interface: board private structure
240 void fm10k_clean_all_tx_rings(struct fm10k_intfc *interface)
244 for (i = 0; i < interface->num_tx_queues; i++)
245 fm10k_clean_tx_ring(interface->tx_ring[i]);
247 /* remove any stale timestamp buffers and free them */
248 skb_queue_purge(&interface->ts_tx_skb_queue);
252 * fm10k_free_all_tx_resources - Free Tx Resources for All Queues
253 * @interface: board private structure
255 * Free all transmit software resources
257 static void fm10k_free_all_tx_resources(struct fm10k_intfc *interface)
259 int i = interface->num_tx_queues;
262 fm10k_free_tx_resources(interface->tx_ring[i]);
266 * fm10k_clean_rx_ring - Free Rx Buffers per Queue
267 * @rx_ring: ring to free buffers from
269 static void fm10k_clean_rx_ring(struct fm10k_ring *rx_ring)
274 if (!rx_ring->rx_buffer)
278 dev_kfree_skb(rx_ring->skb);
281 /* Free all the Rx ring sk_buffs */
282 for (i = 0; i < rx_ring->count; i++) {
283 struct fm10k_rx_buffer *buffer = &rx_ring->rx_buffer[i];
284 /* clean-up will only set page pointer to NULL */
288 dma_unmap_page(rx_ring->dev, buffer->dma,
289 PAGE_SIZE, DMA_FROM_DEVICE);
290 __free_page(buffer->page);
295 size = sizeof(struct fm10k_rx_buffer) * rx_ring->count;
296 memset(rx_ring->rx_buffer, 0, size);
298 /* Zero out the descriptor ring */
299 memset(rx_ring->desc, 0, rx_ring->size);
301 rx_ring->next_to_alloc = 0;
302 rx_ring->next_to_clean = 0;
303 rx_ring->next_to_use = 0;
307 * fm10k_free_rx_resources - Free Rx Resources
308 * @rx_ring: ring to clean the resources from
310 * Free all receive software resources
312 void fm10k_free_rx_resources(struct fm10k_ring *rx_ring)
314 fm10k_clean_rx_ring(rx_ring);
316 vfree(rx_ring->rx_buffer);
317 rx_ring->rx_buffer = NULL;
319 /* if not set, then don't free */
323 dma_free_coherent(rx_ring->dev, rx_ring->size,
324 rx_ring->desc, rx_ring->dma);
326 rx_ring->desc = NULL;
330 * fm10k_clean_all_rx_rings - Free Rx Buffers for all queues
331 * @interface: board private structure
333 void fm10k_clean_all_rx_rings(struct fm10k_intfc *interface)
337 for (i = 0; i < interface->num_rx_queues; i++)
338 fm10k_clean_rx_ring(interface->rx_ring[i]);
342 * fm10k_free_all_rx_resources - Free Rx Resources for All Queues
343 * @interface: board private structure
345 * Free all receive software resources
347 static void fm10k_free_all_rx_resources(struct fm10k_intfc *interface)
349 int i = interface->num_rx_queues;
352 fm10k_free_rx_resources(interface->rx_ring[i]);
356 * fm10k_request_glort_range - Request GLORTs for use in configuring rules
357 * @interface: board private structure
359 * This function allocates a range of glorts for this interface to use.
361 static void fm10k_request_glort_range(struct fm10k_intfc *interface)
363 struct fm10k_hw *hw = &interface->hw;
364 u16 mask = (~hw->mac.dglort_map) >> FM10K_DGLORTMAP_MASK_SHIFT;
366 /* establish GLORT base */
367 interface->glort = hw->mac.dglort_map & FM10K_DGLORTMAP_NONE;
368 interface->glort_count = 0;
370 /* nothing we can do until mask is allocated */
371 if (hw->mac.dglort_map == FM10K_DGLORTMAP_NONE)
374 /* we support 3 possible GLORT configurations.
375 * 1: VFs consume all but the last 1
376 * 2: VFs and PF split glorts with possible gap between
377 * 3: VFs allocated first 64, all others belong to PF
379 if (mask <= hw->iov.total_vfs) {
380 interface->glort_count = 1;
381 interface->glort += mask;
382 } else if (mask < 64) {
383 interface->glort_count = (mask + 1) / 2;
384 interface->glort += interface->glort_count;
386 interface->glort_count = mask - 63;
387 interface->glort += 64;
392 * fm10k_del_vxlan_port_all
393 * @interface: board private structure
395 * This function frees the entire vxlan_port list
397 static void fm10k_del_vxlan_port_all(struct fm10k_intfc *interface)
399 struct fm10k_vxlan_port *vxlan_port;
401 /* flush all entries from list */
402 vxlan_port = list_first_entry_or_null(&interface->vxlan_port,
403 struct fm10k_vxlan_port, list);
405 list_del(&vxlan_port->list);
407 vxlan_port = list_first_entry_or_null(&interface->vxlan_port,
408 struct fm10k_vxlan_port,
414 * fm10k_restore_vxlan_port
415 * @interface: board private structure
417 * This function restores the value in the tunnel_cfg register after reset
419 static void fm10k_restore_vxlan_port(struct fm10k_intfc *interface)
421 struct fm10k_hw *hw = &interface->hw;
422 struct fm10k_vxlan_port *vxlan_port;
424 /* only the PF supports configuring tunnels */
425 if (hw->mac.type != fm10k_mac_pf)
428 vxlan_port = list_first_entry_or_null(&interface->vxlan_port,
429 struct fm10k_vxlan_port, list);
431 /* restore tunnel configuration register */
432 fm10k_write_reg(hw, FM10K_TUNNEL_CFG,
433 (vxlan_port ? ntohs(vxlan_port->port) : 0) |
434 (ETH_P_TEB << FM10K_TUNNEL_CFG_NVGRE_SHIFT));
438 * fm10k_add_vxlan_port
439 * @netdev: network interface device structure
440 * @sa_family: Address family of new port
441 * @port: port number used for VXLAN
443 * This funciton is called when a new VXLAN interface has added a new port
444 * number to the range that is currently in use for VXLAN. The new port
445 * number is always added to the tail so that the port number list should
446 * match the order in which the ports were allocated. The head of the list
447 * is always used as the VXLAN port number for offloads.
449 static void fm10k_add_vxlan_port(struct net_device *dev,
450 sa_family_t sa_family, __be16 port) {
451 struct fm10k_intfc *interface = netdev_priv(dev);
452 struct fm10k_vxlan_port *vxlan_port;
454 /* only the PF supports configuring tunnels */
455 if (interface->hw.mac.type != fm10k_mac_pf)
458 /* existing ports are pulled out so our new entry is always last */
459 fm10k_vxlan_port_for_each(vxlan_port, interface) {
460 if ((vxlan_port->port == port) &&
461 (vxlan_port->sa_family == sa_family)) {
462 list_del(&vxlan_port->list);
467 /* allocate memory to track ports */
468 vxlan_port = kmalloc(sizeof(*vxlan_port), GFP_ATOMIC);
471 vxlan_port->port = port;
472 vxlan_port->sa_family = sa_family;
475 /* add new port value to list */
476 list_add_tail(&vxlan_port->list, &interface->vxlan_port);
478 fm10k_restore_vxlan_port(interface);
482 * fm10k_del_vxlan_port
483 * @netdev: network interface device structure
484 * @sa_family: Address family of freed port
485 * @port: port number used for VXLAN
487 * This funciton is called when a new VXLAN interface has freed a port
488 * number from the range that is currently in use for VXLAN. The freed
489 * port is removed from the list and the new head is used to determine
490 * the port number for offloads.
492 static void fm10k_del_vxlan_port(struct net_device *dev,
493 sa_family_t sa_family, __be16 port) {
494 struct fm10k_intfc *interface = netdev_priv(dev);
495 struct fm10k_vxlan_port *vxlan_port;
497 if (interface->hw.mac.type != fm10k_mac_pf)
500 /* find the port in the list and free it */
501 fm10k_vxlan_port_for_each(vxlan_port, interface) {
502 if ((vxlan_port->port == port) &&
503 (vxlan_port->sa_family == sa_family)) {
504 list_del(&vxlan_port->list);
510 fm10k_restore_vxlan_port(interface);
514 * fm10k_open - Called when a network interface is made active
515 * @netdev: network interface device structure
517 * Returns 0 on success, negative value on failure
519 * The open entry point is called when a network interface is made
520 * active by the system (IFF_UP). At this point all resources needed
521 * for transmit and receive operations are allocated, the interrupt
522 * handler is registered with the OS, the watchdog timer is started,
523 * and the stack is notified that the interface is ready.
525 int fm10k_open(struct net_device *netdev)
527 struct fm10k_intfc *interface = netdev_priv(netdev);
530 /* allocate transmit descriptors */
531 err = fm10k_setup_all_tx_resources(interface);
535 /* allocate receive descriptors */
536 err = fm10k_setup_all_rx_resources(interface);
540 /* allocate interrupt resources */
541 err = fm10k_qv_request_irq(interface);
545 /* setup GLORT assignment for this port */
546 fm10k_request_glort_range(interface);
548 /* Notify the stack of the actual queue counts */
549 err = netif_set_real_num_tx_queues(netdev,
550 interface->num_tx_queues);
554 err = netif_set_real_num_rx_queues(netdev,
555 interface->num_rx_queues);
559 #if IS_ENABLED(CONFIG_FM10K_VXLAN)
560 /* update VXLAN port configuration */
561 vxlan_get_rx_port(netdev);
569 fm10k_qv_free_irq(interface);
571 fm10k_free_all_rx_resources(interface);
573 fm10k_free_all_tx_resources(interface);
579 * fm10k_close - Disables a network interface
580 * @netdev: network interface device structure
582 * Returns 0, this is not allowed to fail
584 * The close entry point is called when an interface is de-activated
585 * by the OS. The hardware is still under the drivers control, but
586 * needs to be disabled. A global MAC reset is issued to stop the
587 * hardware, and all transmit and receive resources are freed.
589 int fm10k_close(struct net_device *netdev)
591 struct fm10k_intfc *interface = netdev_priv(netdev);
593 fm10k_down(interface);
595 fm10k_qv_free_irq(interface);
597 fm10k_del_vxlan_port_all(interface);
599 fm10k_free_all_tx_resources(interface);
600 fm10k_free_all_rx_resources(interface);
605 static netdev_tx_t fm10k_xmit_frame(struct sk_buff *skb, struct net_device *dev)
607 struct fm10k_intfc *interface = netdev_priv(dev);
608 unsigned int r_idx = skb->queue_mapping;
611 if ((skb->protocol == htons(ETH_P_8021Q)) &&
612 !skb_vlan_tag_present(skb)) {
613 /* FM10K only supports hardware tagging, any tags in frame
614 * are considered 2nd level or "outer" tags
616 struct vlan_hdr *vhdr;
619 /* make sure skb is not shared */
620 skb = skb_share_check(skb, GFP_ATOMIC);
624 /* make sure there is enough room to move the ethernet header */
625 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
628 /* verify the skb head is not shared */
629 err = skb_cow_head(skb, 0);
633 /* locate vlan header */
634 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
636 /* pull the 2 key pieces of data out of it */
637 __vlan_hwaccel_put_tag(skb,
639 ntohs(vhdr->h_vlan_TCI));
640 proto = vhdr->h_vlan_encapsulated_proto;
641 skb->protocol = (ntohs(proto) >= 1536) ? proto :
644 /* squash it by moving the ethernet addresses up 4 bytes */
645 memmove(skb->data + VLAN_HLEN, skb->data, 12);
646 __skb_pull(skb, VLAN_HLEN);
647 skb_reset_mac_header(skb);
650 /* The minimum packet size for a single buffer is 17B so pad the skb
651 * in order to meet this minimum size requirement.
653 if (unlikely(skb->len < 17)) {
654 int pad_len = 17 - skb->len;
656 if (skb_pad(skb, pad_len))
658 __skb_put(skb, pad_len);
661 /* prepare packet for hardware time stamping */
662 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))
663 fm10k_ts_tx_enqueue(interface, skb);
665 if (r_idx >= interface->num_tx_queues)
666 r_idx %= interface->num_tx_queues;
668 err = fm10k_xmit_frame_ring(skb, interface->tx_ring[r_idx]);
673 static int fm10k_change_mtu(struct net_device *dev, int new_mtu)
675 if (new_mtu < 68 || new_mtu > FM10K_MAX_JUMBO_FRAME_SIZE)
684 * fm10k_tx_timeout - Respond to a Tx Hang
685 * @netdev: network interface device structure
687 static void fm10k_tx_timeout(struct net_device *netdev)
689 struct fm10k_intfc *interface = netdev_priv(netdev);
690 bool real_tx_hang = false;
693 #define TX_TIMEO_LIMIT 16000
694 for (i = 0; i < interface->num_tx_queues; i++) {
695 struct fm10k_ring *tx_ring = interface->tx_ring[i];
697 if (check_for_tx_hang(tx_ring) && fm10k_check_tx_hang(tx_ring))
702 fm10k_tx_timeout_reset(interface);
704 netif_info(interface, drv, netdev,
705 "Fake Tx hang detected with timeout of %d seconds\n",
706 netdev->watchdog_timeo/HZ);
708 /* fake Tx hang - increase the kernel timeout */
709 if (netdev->watchdog_timeo < TX_TIMEO_LIMIT)
710 netdev->watchdog_timeo *= 2;
714 static int fm10k_uc_vlan_unsync(struct net_device *netdev,
715 const unsigned char *uc_addr)
717 struct fm10k_intfc *interface = netdev_priv(netdev);
718 struct fm10k_hw *hw = &interface->hw;
719 u16 glort = interface->glort;
720 u16 vid = interface->vid;
721 bool set = !!(vid / VLAN_N_VID);
724 /* drop any leading bits on the VLAN ID */
725 vid &= VLAN_N_VID - 1;
727 err = hw->mac.ops.update_uc_addr(hw, glort, uc_addr, vid, set, 0);
731 /* return non-zero value as we are only doing a partial sync/unsync */
735 static int fm10k_mc_vlan_unsync(struct net_device *netdev,
736 const unsigned char *mc_addr)
738 struct fm10k_intfc *interface = netdev_priv(netdev);
739 struct fm10k_hw *hw = &interface->hw;
740 u16 glort = interface->glort;
741 u16 vid = interface->vid;
742 bool set = !!(vid / VLAN_N_VID);
745 /* drop any leading bits on the VLAN ID */
746 vid &= VLAN_N_VID - 1;
748 err = hw->mac.ops.update_mc_addr(hw, glort, mc_addr, vid, set);
752 /* return non-zero value as we are only doing a partial sync/unsync */
756 static int fm10k_update_vid(struct net_device *netdev, u16 vid, bool set)
758 struct fm10k_intfc *interface = netdev_priv(netdev);
759 struct fm10k_hw *hw = &interface->hw;
762 /* updates do not apply to VLAN 0 */
766 if (vid >= VLAN_N_VID)
769 /* Verify we have permission to add VLANs */
770 if (hw->mac.vlan_override)
773 /* if default VLAN is already present do nothing */
774 if (vid == hw->mac.default_vid)
777 /* update active_vlans bitmask */
778 set_bit(vid, interface->active_vlans);
780 clear_bit(vid, interface->active_vlans);
782 fm10k_mbx_lock(interface);
784 /* only need to update the VLAN if not in promiscuous mode */
785 if (!(netdev->flags & IFF_PROMISC)) {
786 err = hw->mac.ops.update_vlan(hw, vid, 0, set);
791 /* update our base MAC address */
792 err = hw->mac.ops.update_uc_addr(hw, interface->glort, hw->mac.addr,
797 /* set vid prior to syncing/unsyncing the VLAN */
798 interface->vid = vid + (set ? VLAN_N_VID : 0);
800 /* Update the unicast and multicast address list to add/drop VLAN */
801 __dev_uc_unsync(netdev, fm10k_uc_vlan_unsync);
802 __dev_mc_unsync(netdev, fm10k_mc_vlan_unsync);
805 fm10k_mbx_unlock(interface);
810 static int fm10k_vlan_rx_add_vid(struct net_device *netdev,
811 __always_unused __be16 proto, u16 vid)
813 /* update VLAN and address table based on changes */
814 return fm10k_update_vid(netdev, vid, true);
817 static int fm10k_vlan_rx_kill_vid(struct net_device *netdev,
818 __always_unused __be16 proto, u16 vid)
820 /* update VLAN and address table based on changes */
821 return fm10k_update_vid(netdev, vid, false);
824 static u16 fm10k_find_next_vlan(struct fm10k_intfc *interface, u16 vid)
826 struct fm10k_hw *hw = &interface->hw;
827 u16 default_vid = hw->mac.default_vid;
828 u16 vid_limit = vid < default_vid ? default_vid : VLAN_N_VID;
830 vid = find_next_bit(interface->active_vlans, vid_limit, ++vid);
835 static void fm10k_clear_unused_vlans(struct fm10k_intfc *interface)
837 struct fm10k_hw *hw = &interface->hw;
840 /* loop through and find any gaps in the table */
841 for (vid = 0, prev_vid = 0;
842 prev_vid < VLAN_N_VID;
843 prev_vid = vid + 1, vid = fm10k_find_next_vlan(interface, vid)) {
847 /* send request to clear multiple bits at a time */
848 prev_vid += (vid - prev_vid - 1) << FM10K_VLAN_LENGTH_SHIFT;
849 hw->mac.ops.update_vlan(hw, prev_vid, 0, false);
853 static int __fm10k_uc_sync(struct net_device *dev,
854 const unsigned char *addr, bool sync)
856 struct fm10k_intfc *interface = netdev_priv(dev);
857 struct fm10k_hw *hw = &interface->hw;
858 u16 vid, glort = interface->glort;
861 if (!is_valid_ether_addr(addr))
862 return -EADDRNOTAVAIL;
864 /* update table with current entries */
865 for (vid = hw->mac.default_vid ? fm10k_find_next_vlan(interface, 0) : 0;
867 vid = fm10k_find_next_vlan(interface, vid)) {
868 err = hw->mac.ops.update_uc_addr(hw, glort, addr,
877 static int fm10k_uc_sync(struct net_device *dev,
878 const unsigned char *addr)
880 return __fm10k_uc_sync(dev, addr, true);
883 static int fm10k_uc_unsync(struct net_device *dev,
884 const unsigned char *addr)
886 return __fm10k_uc_sync(dev, addr, false);
889 static int fm10k_set_mac(struct net_device *dev, void *p)
891 struct fm10k_intfc *interface = netdev_priv(dev);
892 struct fm10k_hw *hw = &interface->hw;
893 struct sockaddr *addr = p;
896 if (!is_valid_ether_addr(addr->sa_data))
897 return -EADDRNOTAVAIL;
899 if (dev->flags & IFF_UP) {
900 /* setting MAC address requires mailbox */
901 fm10k_mbx_lock(interface);
903 err = fm10k_uc_sync(dev, addr->sa_data);
905 fm10k_uc_unsync(dev, hw->mac.addr);
907 fm10k_mbx_unlock(interface);
911 ether_addr_copy(dev->dev_addr, addr->sa_data);
912 ether_addr_copy(hw->mac.addr, addr->sa_data);
913 dev->addr_assign_type &= ~NET_ADDR_RANDOM;
916 /* if we had a mailbox error suggest trying again */
917 return err ? -EAGAIN : 0;
920 static int __fm10k_mc_sync(struct net_device *dev,
921 const unsigned char *addr, bool sync)
923 struct fm10k_intfc *interface = netdev_priv(dev);
924 struct fm10k_hw *hw = &interface->hw;
925 u16 vid, glort = interface->glort;
928 if (!is_multicast_ether_addr(addr))
929 return -EADDRNOTAVAIL;
931 /* update table with current entries */
932 for (vid = hw->mac.default_vid ? fm10k_find_next_vlan(interface, 0) : 0;
934 vid = fm10k_find_next_vlan(interface, vid)) {
935 err = hw->mac.ops.update_mc_addr(hw, glort, addr, vid, sync);
943 static int fm10k_mc_sync(struct net_device *dev,
944 const unsigned char *addr)
946 return __fm10k_mc_sync(dev, addr, true);
949 static int fm10k_mc_unsync(struct net_device *dev,
950 const unsigned char *addr)
952 return __fm10k_mc_sync(dev, addr, false);
955 static void fm10k_set_rx_mode(struct net_device *dev)
957 struct fm10k_intfc *interface = netdev_priv(dev);
958 struct fm10k_hw *hw = &interface->hw;
961 /* no need to update the harwdare if we are not running */
962 if (!(dev->flags & IFF_UP))
965 /* determine new mode based on flags */
966 xcast_mode = (dev->flags & IFF_PROMISC) ? FM10K_XCAST_MODE_PROMISC :
967 (dev->flags & IFF_ALLMULTI) ? FM10K_XCAST_MODE_ALLMULTI :
968 (dev->flags & (IFF_BROADCAST | IFF_MULTICAST)) ?
969 FM10K_XCAST_MODE_MULTI : FM10K_XCAST_MODE_NONE;
971 fm10k_mbx_lock(interface);
973 /* synchronize all of the addresses */
974 if (xcast_mode != FM10K_XCAST_MODE_PROMISC) {
975 __dev_uc_sync(dev, fm10k_uc_sync, fm10k_uc_unsync);
976 if (xcast_mode != FM10K_XCAST_MODE_ALLMULTI)
977 __dev_mc_sync(dev, fm10k_mc_sync, fm10k_mc_unsync);
980 /* if we aren't changing modes there is nothing to do */
981 if (interface->xcast_mode != xcast_mode) {
982 /* update VLAN table */
983 if (xcast_mode == FM10K_XCAST_MODE_PROMISC)
984 hw->mac.ops.update_vlan(hw, FM10K_VLAN_ALL, 0, true);
985 if (interface->xcast_mode == FM10K_XCAST_MODE_PROMISC)
986 fm10k_clear_unused_vlans(interface);
988 /* update xcast mode */
989 hw->mac.ops.update_xcast_mode(hw, interface->glort, xcast_mode);
991 /* record updated xcast mode state */
992 interface->xcast_mode = xcast_mode;
995 fm10k_mbx_unlock(interface);
998 void fm10k_restore_rx_state(struct fm10k_intfc *interface)
1000 struct net_device *netdev = interface->netdev;
1001 struct fm10k_hw *hw = &interface->hw;
1005 /* restore our address if perm_addr is set */
1006 if (hw->mac.type == fm10k_mac_vf) {
1007 if (is_valid_ether_addr(hw->mac.perm_addr)) {
1008 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
1009 ether_addr_copy(netdev->perm_addr, hw->mac.perm_addr);
1010 ether_addr_copy(netdev->dev_addr, hw->mac.perm_addr);
1011 netdev->addr_assign_type &= ~NET_ADDR_RANDOM;
1014 if (hw->mac.vlan_override)
1015 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
1017 netdev->features |= NETIF_F_HW_VLAN_CTAG_RX;
1020 /* record glort for this interface */
1021 glort = interface->glort;
1023 /* convert interface flags to xcast mode */
1024 if (netdev->flags & IFF_PROMISC)
1025 xcast_mode = FM10K_XCAST_MODE_PROMISC;
1026 else if (netdev->flags & IFF_ALLMULTI)
1027 xcast_mode = FM10K_XCAST_MODE_ALLMULTI;
1028 else if (netdev->flags & (IFF_BROADCAST | IFF_MULTICAST))
1029 xcast_mode = FM10K_XCAST_MODE_MULTI;
1031 xcast_mode = FM10K_XCAST_MODE_NONE;
1033 fm10k_mbx_lock(interface);
1035 /* Enable logical port */
1036 hw->mac.ops.update_lport_state(hw, glort, interface->glort_count, true);
1038 /* update VLAN table */
1039 hw->mac.ops.update_vlan(hw, FM10K_VLAN_ALL, 0,
1040 xcast_mode == FM10K_XCAST_MODE_PROMISC);
1042 /* Add filter for VLAN 0 */
1043 hw->mac.ops.update_vlan(hw, 0, 0, true);
1045 /* update table with current entries */
1046 for (vid = hw->mac.default_vid ? fm10k_find_next_vlan(interface, 0) : 0;
1048 vid = fm10k_find_next_vlan(interface, vid)) {
1049 hw->mac.ops.update_vlan(hw, vid, 0, true);
1050 hw->mac.ops.update_uc_addr(hw, glort, hw->mac.addr,
1054 /* synchronize all of the addresses */
1055 if (xcast_mode != FM10K_XCAST_MODE_PROMISC) {
1056 __dev_uc_sync(netdev, fm10k_uc_sync, fm10k_uc_unsync);
1057 if (xcast_mode != FM10K_XCAST_MODE_ALLMULTI)
1058 __dev_mc_sync(netdev, fm10k_mc_sync, fm10k_mc_unsync);
1061 /* update xcast mode */
1062 hw->mac.ops.update_xcast_mode(hw, glort, xcast_mode);
1064 fm10k_mbx_unlock(interface);
1066 /* record updated xcast mode state */
1067 interface->xcast_mode = xcast_mode;
1069 /* Restore tunnel configuration */
1070 fm10k_restore_vxlan_port(interface);
1073 void fm10k_reset_rx_state(struct fm10k_intfc *interface)
1075 struct net_device *netdev = interface->netdev;
1076 struct fm10k_hw *hw = &interface->hw;
1078 fm10k_mbx_lock(interface);
1080 /* clear the logical port state on lower device */
1081 hw->mac.ops.update_lport_state(hw, interface->glort,
1082 interface->glort_count, false);
1084 fm10k_mbx_unlock(interface);
1086 /* reset flags to default state */
1087 interface->xcast_mode = FM10K_XCAST_MODE_NONE;
1089 /* clear the sync flag since the lport has been dropped */
1090 __dev_uc_unsync(netdev, NULL);
1091 __dev_mc_unsync(netdev, NULL);
1095 * fm10k_get_stats64 - Get System Network Statistics
1096 * @netdev: network interface device structure
1097 * @stats: storage space for 64bit statistics
1099 * Returns 64bit statistics, for use in the ndo_get_stats64 callback. This
1100 * function replaces fm10k_get_stats for kernels which support it.
1102 static struct rtnl_link_stats64 *fm10k_get_stats64(struct net_device *netdev,
1103 struct rtnl_link_stats64 *stats)
1105 struct fm10k_intfc *interface = netdev_priv(netdev);
1106 struct fm10k_ring *ring;
1107 unsigned int start, i;
1112 for (i = 0; i < interface->num_rx_queues; i++) {
1113 ring = ACCESS_ONCE(interface->rx_ring[i]);
1119 start = u64_stats_fetch_begin_irq(&ring->syncp);
1120 packets = ring->stats.packets;
1121 bytes = ring->stats.bytes;
1122 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1124 stats->rx_packets += packets;
1125 stats->rx_bytes += bytes;
1128 for (i = 0; i < interface->num_tx_queues; i++) {
1129 ring = ACCESS_ONCE(interface->rx_ring[i]);
1135 start = u64_stats_fetch_begin_irq(&ring->syncp);
1136 packets = ring->stats.packets;
1137 bytes = ring->stats.bytes;
1138 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1140 stats->tx_packets += packets;
1141 stats->tx_bytes += bytes;
1146 /* following stats updated by fm10k_service_task() */
1147 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1152 int fm10k_setup_tc(struct net_device *dev, u8 tc)
1154 struct fm10k_intfc *interface = netdev_priv(dev);
1156 /* Currently only the PF supports priority classes */
1157 if (tc && (interface->hw.mac.type != fm10k_mac_pf))
1160 /* Hardware supports up to 8 traffic classes */
1164 /* Hardware has to reinitialize queues to match packet
1165 * buffer alignment. Unfortunately, the hardware is not
1166 * flexible enough to do this dynamically.
1168 if (netif_running(dev))
1171 fm10k_mbx_free_irq(interface);
1173 fm10k_clear_queueing_scheme(interface);
1175 /* we expect the prio_tc map to be repopulated later */
1176 netdev_reset_tc(dev);
1177 netdev_set_num_tc(dev, tc);
1179 fm10k_init_queueing_scheme(interface);
1181 fm10k_mbx_request_irq(interface);
1183 if (netif_running(dev))
1186 /* flag to indicate SWPRI has yet to be updated */
1187 interface->flags |= FM10K_FLAG_SWPRI_CONFIG;
1192 static int fm10k_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
1196 return fm10k_get_ts_config(netdev, ifr);
1198 return fm10k_set_ts_config(netdev, ifr);
1204 static void fm10k_assign_l2_accel(struct fm10k_intfc *interface,
1205 struct fm10k_l2_accel *l2_accel)
1207 struct fm10k_ring *ring;
1210 for (i = 0; i < interface->num_rx_queues; i++) {
1211 ring = interface->rx_ring[i];
1212 rcu_assign_pointer(ring->l2_accel, l2_accel);
1215 interface->l2_accel = l2_accel;
1218 static void *fm10k_dfwd_add_station(struct net_device *dev,
1219 struct net_device *sdev)
1221 struct fm10k_intfc *interface = netdev_priv(dev);
1222 struct fm10k_l2_accel *l2_accel = interface->l2_accel;
1223 struct fm10k_l2_accel *old_l2_accel = NULL;
1224 struct fm10k_dglort_cfg dglort = { 0 };
1225 struct fm10k_hw *hw = &interface->hw;
1229 /* allocate l2 accel structure if it is not available */
1231 /* verify there is enough free GLORTs to support l2_accel */
1232 if (interface->glort_count < 7)
1233 return ERR_PTR(-EBUSY);
1235 size = offsetof(struct fm10k_l2_accel, macvlan[7]);
1236 l2_accel = kzalloc(size, GFP_KERNEL);
1238 return ERR_PTR(-ENOMEM);
1241 l2_accel->dglort = interface->glort;
1243 /* update pointers */
1244 fm10k_assign_l2_accel(interface, l2_accel);
1245 /* do not expand if we are at our limit */
1246 } else if ((l2_accel->count == FM10K_MAX_STATIONS) ||
1247 (l2_accel->count == (interface->glort_count - 1))) {
1248 return ERR_PTR(-EBUSY);
1249 /* expand if we have hit the size limit */
1250 } else if (l2_accel->count == l2_accel->size) {
1251 old_l2_accel = l2_accel;
1252 size = offsetof(struct fm10k_l2_accel,
1253 macvlan[(l2_accel->size * 2) + 1]);
1254 l2_accel = kzalloc(size, GFP_KERNEL);
1256 return ERR_PTR(-ENOMEM);
1258 memcpy(l2_accel, old_l2_accel,
1259 offsetof(struct fm10k_l2_accel,
1260 macvlan[old_l2_accel->size]));
1262 l2_accel->size = (old_l2_accel->size * 2) + 1;
1264 /* update pointers */
1265 fm10k_assign_l2_accel(interface, l2_accel);
1266 kfree_rcu(old_l2_accel, rcu);
1269 /* add macvlan to accel table, and record GLORT for position */
1270 for (i = 0; i < l2_accel->size; i++) {
1271 if (!l2_accel->macvlan[i])
1275 /* record station */
1276 l2_accel->macvlan[i] = sdev;
1279 /* configure default DGLORT mapping for RSS/DCB */
1280 dglort.idx = fm10k_dglort_pf_rss;
1281 dglort.inner_rss = 1;
1282 dglort.rss_l = fls(interface->ring_feature[RING_F_RSS].mask);
1283 dglort.pc_l = fls(interface->ring_feature[RING_F_QOS].mask);
1284 dglort.glort = interface->glort;
1285 dglort.shared_l = fls(l2_accel->size);
1286 hw->mac.ops.configure_dglort_map(hw, &dglort);
1288 /* Add rules for this specific dglort to the switch */
1289 fm10k_mbx_lock(interface);
1291 glort = l2_accel->dglort + 1 + i;
1292 hw->mac.ops.update_xcast_mode(hw, glort, FM10K_XCAST_MODE_MULTI);
1293 hw->mac.ops.update_uc_addr(hw, glort, sdev->dev_addr, 0, true, 0);
1295 fm10k_mbx_unlock(interface);
1300 static void fm10k_dfwd_del_station(struct net_device *dev, void *priv)
1302 struct fm10k_intfc *interface = netdev_priv(dev);
1303 struct fm10k_l2_accel *l2_accel = ACCESS_ONCE(interface->l2_accel);
1304 struct fm10k_dglort_cfg dglort = { 0 };
1305 struct fm10k_hw *hw = &interface->hw;
1306 struct net_device *sdev = priv;
1313 /* search table for matching interface */
1314 for (i = 0; i < l2_accel->size; i++) {
1315 if (l2_accel->macvlan[i] == sdev)
1319 /* exit if macvlan not found */
1320 if (i == l2_accel->size)
1323 /* Remove any rules specific to this dglort */
1324 fm10k_mbx_lock(interface);
1326 glort = l2_accel->dglort + 1 + i;
1327 hw->mac.ops.update_xcast_mode(hw, glort, FM10K_XCAST_MODE_NONE);
1328 hw->mac.ops.update_uc_addr(hw, glort, sdev->dev_addr, 0, false, 0);
1330 fm10k_mbx_unlock(interface);
1332 /* record removal */
1333 l2_accel->macvlan[i] = NULL;
1336 /* configure default DGLORT mapping for RSS/DCB */
1337 dglort.idx = fm10k_dglort_pf_rss;
1338 dglort.inner_rss = 1;
1339 dglort.rss_l = fls(interface->ring_feature[RING_F_RSS].mask);
1340 dglort.pc_l = fls(interface->ring_feature[RING_F_QOS].mask);
1341 dglort.glort = interface->glort;
1343 dglort.shared_l = fls(l2_accel->size);
1344 hw->mac.ops.configure_dglort_map(hw, &dglort);
1346 /* If table is empty remove it */
1347 if (l2_accel->count == 0) {
1348 fm10k_assign_l2_accel(interface, NULL);
1349 kfree_rcu(l2_accel, rcu);
1353 static netdev_features_t fm10k_features_check(struct sk_buff *skb,
1354 struct net_device *dev,
1355 netdev_features_t features)
1357 if (!skb->encapsulation || fm10k_tx_encap_offload(skb))
1360 return features & ~(NETIF_F_ALL_CSUM | NETIF_F_GSO_MASK);
1363 static const struct net_device_ops fm10k_netdev_ops = {
1364 .ndo_open = fm10k_open,
1365 .ndo_stop = fm10k_close,
1366 .ndo_validate_addr = eth_validate_addr,
1367 .ndo_start_xmit = fm10k_xmit_frame,
1368 .ndo_set_mac_address = fm10k_set_mac,
1369 .ndo_change_mtu = fm10k_change_mtu,
1370 .ndo_tx_timeout = fm10k_tx_timeout,
1371 .ndo_vlan_rx_add_vid = fm10k_vlan_rx_add_vid,
1372 .ndo_vlan_rx_kill_vid = fm10k_vlan_rx_kill_vid,
1373 .ndo_set_rx_mode = fm10k_set_rx_mode,
1374 .ndo_get_stats64 = fm10k_get_stats64,
1375 .ndo_setup_tc = fm10k_setup_tc,
1376 .ndo_set_vf_mac = fm10k_ndo_set_vf_mac,
1377 .ndo_set_vf_vlan = fm10k_ndo_set_vf_vlan,
1378 .ndo_set_vf_rate = fm10k_ndo_set_vf_bw,
1379 .ndo_get_vf_config = fm10k_ndo_get_vf_config,
1380 .ndo_add_vxlan_port = fm10k_add_vxlan_port,
1381 .ndo_del_vxlan_port = fm10k_del_vxlan_port,
1382 .ndo_do_ioctl = fm10k_ioctl,
1383 .ndo_dfwd_add_station = fm10k_dfwd_add_station,
1384 .ndo_dfwd_del_station = fm10k_dfwd_del_station,
1385 .ndo_features_check = fm10k_features_check,
1388 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
1390 struct net_device *fm10k_alloc_netdev(void)
1392 struct fm10k_intfc *interface;
1393 struct net_device *dev;
1395 dev = alloc_etherdev_mq(sizeof(struct fm10k_intfc), MAX_QUEUES);
1399 /* set net device and ethtool ops */
1400 dev->netdev_ops = &fm10k_netdev_ops;
1401 fm10k_set_ethtool_ops(dev);
1403 /* configure default debug level */
1404 interface = netdev_priv(dev);
1405 interface->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
1407 /* configure default features */
1408 dev->features |= NETIF_F_IP_CSUM |
1414 NETIF_F_GSO_UDP_TUNNEL |
1418 /* all features defined to this point should be changeable */
1419 dev->hw_features |= dev->features;
1421 /* allow user to enable L2 forwarding acceleration */
1422 dev->hw_features |= NETIF_F_HW_L2FW_DOFFLOAD;
1424 /* configure VLAN features */
1425 dev->vlan_features |= dev->features;
1427 /* configure tunnel offloads */
1428 dev->hw_enc_features |= NETIF_F_IP_CSUM |
1432 NETIF_F_GSO_UDP_TUNNEL |
1435 /* we want to leave these both on as we cannot disable VLAN tag
1436 * insertion or stripping on the hardware since it is contained
1437 * in the FTAG and not in the frame itself.
1439 dev->features |= NETIF_F_HW_VLAN_CTAG_TX |
1440 NETIF_F_HW_VLAN_CTAG_RX |
1441 NETIF_F_HW_VLAN_CTAG_FILTER;
1443 dev->priv_flags |= IFF_UNICAST_FLT;