1 /* Intel(R) Ethernet Switch Host Interface Driver
2 * Copyright(c) 2013 - 2017 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 #include <net/udp_tunnel.h>
26 * fm10k_setup_tx_resources - allocate Tx resources (Descriptors)
27 * @tx_ring: tx descriptor ring (for a specific queue) to setup
29 * Return 0 on success, negative on failure
31 int fm10k_setup_tx_resources(struct fm10k_ring *tx_ring)
33 struct device *dev = tx_ring->dev;
36 size = sizeof(struct fm10k_tx_buffer) * tx_ring->count;
38 tx_ring->tx_buffer = vzalloc(size);
39 if (!tx_ring->tx_buffer)
42 u64_stats_init(&tx_ring->syncp);
44 /* round up to nearest 4K */
45 tx_ring->size = tx_ring->count * sizeof(struct fm10k_tx_desc);
46 tx_ring->size = ALIGN(tx_ring->size, 4096);
48 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
49 &tx_ring->dma, GFP_KERNEL);
56 vfree(tx_ring->tx_buffer);
57 tx_ring->tx_buffer = NULL;
62 * fm10k_setup_all_tx_resources - allocate all queues Tx resources
63 * @interface: board private structure
65 * If this function returns with an error, then it's possible one or
66 * more of the rings is populated (while the rest are not). It is the
67 * callers duty to clean those orphaned rings.
69 * Return 0 on success, negative on failure
71 static int fm10k_setup_all_tx_resources(struct fm10k_intfc *interface)
75 for (i = 0; i < interface->num_tx_queues; i++) {
76 err = fm10k_setup_tx_resources(interface->tx_ring[i]);
80 netif_err(interface, probe, interface->netdev,
81 "Allocation for Tx Queue %u failed\n", i);
87 /* rewind the index freeing the rings as we go */
89 fm10k_free_tx_resources(interface->tx_ring[i]);
94 * fm10k_setup_rx_resources - allocate Rx resources (Descriptors)
95 * @rx_ring: rx descriptor ring (for a specific queue) to setup
97 * Returns 0 on success, negative on failure
99 int fm10k_setup_rx_resources(struct fm10k_ring *rx_ring)
101 struct device *dev = rx_ring->dev;
104 size = sizeof(struct fm10k_rx_buffer) * rx_ring->count;
106 rx_ring->rx_buffer = vzalloc(size);
107 if (!rx_ring->rx_buffer)
110 u64_stats_init(&rx_ring->syncp);
112 /* Round up to nearest 4K */
113 rx_ring->size = rx_ring->count * sizeof(union fm10k_rx_desc);
114 rx_ring->size = ALIGN(rx_ring->size, 4096);
116 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
117 &rx_ring->dma, GFP_KERNEL);
123 vfree(rx_ring->rx_buffer);
124 rx_ring->rx_buffer = NULL;
129 * fm10k_setup_all_rx_resources - allocate all queues Rx resources
130 * @interface: board private structure
132 * If this function returns with an error, then it's possible one or
133 * more of the rings is populated (while the rest are not). It is the
134 * callers duty to clean those orphaned rings.
136 * Return 0 on success, negative on failure
138 static int fm10k_setup_all_rx_resources(struct fm10k_intfc *interface)
142 for (i = 0; i < interface->num_rx_queues; i++) {
143 err = fm10k_setup_rx_resources(interface->rx_ring[i]);
147 netif_err(interface, probe, interface->netdev,
148 "Allocation for Rx Queue %u failed\n", i);
154 /* rewind the index freeing the rings as we go */
156 fm10k_free_rx_resources(interface->rx_ring[i]);
160 void fm10k_unmap_and_free_tx_resource(struct fm10k_ring *ring,
161 struct fm10k_tx_buffer *tx_buffer)
163 if (tx_buffer->skb) {
164 dev_kfree_skb_any(tx_buffer->skb);
165 if (dma_unmap_len(tx_buffer, len))
166 dma_unmap_single(ring->dev,
167 dma_unmap_addr(tx_buffer, dma),
168 dma_unmap_len(tx_buffer, len),
170 } else if (dma_unmap_len(tx_buffer, len)) {
171 dma_unmap_page(ring->dev,
172 dma_unmap_addr(tx_buffer, dma),
173 dma_unmap_len(tx_buffer, len),
176 tx_buffer->next_to_watch = NULL;
177 tx_buffer->skb = NULL;
178 dma_unmap_len_set(tx_buffer, len, 0);
179 /* tx_buffer must be completely set up in the transmit path */
183 * fm10k_clean_tx_ring - Free Tx Buffers
184 * @tx_ring: ring to be cleaned
186 static void fm10k_clean_tx_ring(struct fm10k_ring *tx_ring)
188 struct fm10k_tx_buffer *tx_buffer;
192 /* ring already cleared, nothing to do */
193 if (!tx_ring->tx_buffer)
196 /* Free all the Tx ring sk_buffs */
197 for (i = 0; i < tx_ring->count; i++) {
198 tx_buffer = &tx_ring->tx_buffer[i];
199 fm10k_unmap_and_free_tx_resource(tx_ring, tx_buffer);
202 /* reset BQL values */
203 netdev_tx_reset_queue(txring_txq(tx_ring));
205 size = sizeof(struct fm10k_tx_buffer) * tx_ring->count;
206 memset(tx_ring->tx_buffer, 0, size);
208 /* Zero out the descriptor ring */
209 memset(tx_ring->desc, 0, tx_ring->size);
213 * fm10k_free_tx_resources - Free Tx Resources per Queue
214 * @tx_ring: Tx descriptor ring for a specific queue
216 * Free all transmit software resources
218 void fm10k_free_tx_resources(struct fm10k_ring *tx_ring)
220 fm10k_clean_tx_ring(tx_ring);
222 vfree(tx_ring->tx_buffer);
223 tx_ring->tx_buffer = NULL;
225 /* if not set, then don't free */
229 dma_free_coherent(tx_ring->dev, tx_ring->size,
230 tx_ring->desc, tx_ring->dma);
231 tx_ring->desc = NULL;
235 * fm10k_clean_all_tx_rings - Free Tx Buffers for all queues
236 * @interface: board private structure
238 void fm10k_clean_all_tx_rings(struct fm10k_intfc *interface)
242 for (i = 0; i < interface->num_tx_queues; i++)
243 fm10k_clean_tx_ring(interface->tx_ring[i]);
247 * fm10k_free_all_tx_resources - Free Tx Resources for All Queues
248 * @interface: board private structure
250 * Free all transmit software resources
252 static void fm10k_free_all_tx_resources(struct fm10k_intfc *interface)
254 int i = interface->num_tx_queues;
257 fm10k_free_tx_resources(interface->tx_ring[i]);
261 * fm10k_clean_rx_ring - Free Rx Buffers per Queue
262 * @rx_ring: ring to free buffers from
264 static void fm10k_clean_rx_ring(struct fm10k_ring *rx_ring)
269 if (!rx_ring->rx_buffer)
273 dev_kfree_skb(rx_ring->skb);
276 /* Free all the Rx ring sk_buffs */
277 for (i = 0; i < rx_ring->count; i++) {
278 struct fm10k_rx_buffer *buffer = &rx_ring->rx_buffer[i];
279 /* clean-up will only set page pointer to NULL */
283 dma_unmap_page(rx_ring->dev, buffer->dma,
284 PAGE_SIZE, DMA_FROM_DEVICE);
285 __free_page(buffer->page);
290 size = sizeof(struct fm10k_rx_buffer) * rx_ring->count;
291 memset(rx_ring->rx_buffer, 0, size);
293 /* Zero out the descriptor ring */
294 memset(rx_ring->desc, 0, rx_ring->size);
296 rx_ring->next_to_alloc = 0;
297 rx_ring->next_to_clean = 0;
298 rx_ring->next_to_use = 0;
302 * fm10k_free_rx_resources - Free Rx Resources
303 * @rx_ring: ring to clean the resources from
305 * Free all receive software resources
307 void fm10k_free_rx_resources(struct fm10k_ring *rx_ring)
309 fm10k_clean_rx_ring(rx_ring);
311 vfree(rx_ring->rx_buffer);
312 rx_ring->rx_buffer = NULL;
314 /* if not set, then don't free */
318 dma_free_coherent(rx_ring->dev, rx_ring->size,
319 rx_ring->desc, rx_ring->dma);
321 rx_ring->desc = NULL;
325 * fm10k_clean_all_rx_rings - Free Rx Buffers for all queues
326 * @interface: board private structure
328 void fm10k_clean_all_rx_rings(struct fm10k_intfc *interface)
332 for (i = 0; i < interface->num_rx_queues; i++)
333 fm10k_clean_rx_ring(interface->rx_ring[i]);
337 * fm10k_free_all_rx_resources - Free Rx Resources for All Queues
338 * @interface: board private structure
340 * Free all receive software resources
342 static void fm10k_free_all_rx_resources(struct fm10k_intfc *interface)
344 int i = interface->num_rx_queues;
347 fm10k_free_rx_resources(interface->rx_ring[i]);
351 * fm10k_request_glort_range - Request GLORTs for use in configuring rules
352 * @interface: board private structure
354 * This function allocates a range of glorts for this interface to use.
356 static void fm10k_request_glort_range(struct fm10k_intfc *interface)
358 struct fm10k_hw *hw = &interface->hw;
359 u16 mask = (~hw->mac.dglort_map) >> FM10K_DGLORTMAP_MASK_SHIFT;
361 /* establish GLORT base */
362 interface->glort = hw->mac.dglort_map & FM10K_DGLORTMAP_NONE;
363 interface->glort_count = 0;
365 /* nothing we can do until mask is allocated */
366 if (hw->mac.dglort_map == FM10K_DGLORTMAP_NONE)
369 /* we support 3 possible GLORT configurations.
370 * 1: VFs consume all but the last 1
371 * 2: VFs and PF split glorts with possible gap between
372 * 3: VFs allocated first 64, all others belong to PF
374 if (mask <= hw->iov.total_vfs) {
375 interface->glort_count = 1;
376 interface->glort += mask;
377 } else if (mask < 64) {
378 interface->glort_count = (mask + 1) / 2;
379 interface->glort += interface->glort_count;
381 interface->glort_count = mask - 63;
382 interface->glort += 64;
387 * fm10k_free_udp_port_info
388 * @interface: board private structure
390 * This function frees both geneve_port and vxlan_port structures
392 static void fm10k_free_udp_port_info(struct fm10k_intfc *interface)
394 struct fm10k_udp_port *port;
396 /* flush all entries from vxlan list */
397 port = list_first_entry_or_null(&interface->vxlan_port,
398 struct fm10k_udp_port, list);
400 list_del(&port->list);
402 port = list_first_entry_or_null(&interface->vxlan_port,
403 struct fm10k_udp_port,
407 /* flush all entries from geneve list */
408 port = list_first_entry_or_null(&interface->geneve_port,
409 struct fm10k_udp_port, list);
411 list_del(&port->list);
413 port = list_first_entry_or_null(&interface->vxlan_port,
414 struct fm10k_udp_port,
420 * fm10k_restore_udp_port_info
421 * @interface: board private structure
423 * This function restores the value in the tunnel_cfg register(s) after reset
425 static void fm10k_restore_udp_port_info(struct fm10k_intfc *interface)
427 struct fm10k_hw *hw = &interface->hw;
428 struct fm10k_udp_port *port;
430 /* only the PF supports configuring tunnels */
431 if (hw->mac.type != fm10k_mac_pf)
434 port = list_first_entry_or_null(&interface->vxlan_port,
435 struct fm10k_udp_port, list);
437 /* restore tunnel configuration register */
438 fm10k_write_reg(hw, FM10K_TUNNEL_CFG,
439 (port ? ntohs(port->port) : 0) |
440 (ETH_P_TEB << FM10K_TUNNEL_CFG_NVGRE_SHIFT));
442 port = list_first_entry_or_null(&interface->geneve_port,
443 struct fm10k_udp_port, list);
445 /* restore Geneve tunnel configuration register */
446 fm10k_write_reg(hw, FM10K_TUNNEL_CFG_GENEVE,
447 (port ? ntohs(port->port) : 0));
450 static struct fm10k_udp_port *
451 fm10k_remove_tunnel_port(struct list_head *ports,
452 struct udp_tunnel_info *ti)
454 struct fm10k_udp_port *port;
456 list_for_each_entry(port, ports, list) {
457 if ((port->port == ti->port) &&
458 (port->sa_family == ti->sa_family)) {
459 list_del(&port->list);
467 static void fm10k_insert_tunnel_port(struct list_head *ports,
468 struct udp_tunnel_info *ti)
470 struct fm10k_udp_port *port;
472 /* remove existing port entry from the list so that the newest items
473 * are always at the tail of the list.
475 port = fm10k_remove_tunnel_port(ports, ti);
477 port = kmalloc(sizeof(*port), GFP_ATOMIC);
480 port->port = ti->port;
481 port->sa_family = ti->sa_family;
484 list_add_tail(&port->list, ports);
488 * fm10k_udp_tunnel_add
489 * @netdev: network interface device structure
490 * @ti: Tunnel endpoint information
492 * This function is called when a new UDP tunnel port has been added.
493 * Due to hardware restrictions, only one port per type can be offloaded at
496 static void fm10k_udp_tunnel_add(struct net_device *dev,
497 struct udp_tunnel_info *ti)
499 struct fm10k_intfc *interface = netdev_priv(dev);
501 /* only the PF supports configuring tunnels */
502 if (interface->hw.mac.type != fm10k_mac_pf)
506 case UDP_TUNNEL_TYPE_VXLAN:
507 fm10k_insert_tunnel_port(&interface->vxlan_port, ti);
509 case UDP_TUNNEL_TYPE_GENEVE:
510 fm10k_insert_tunnel_port(&interface->geneve_port, ti);
516 fm10k_restore_udp_port_info(interface);
520 * fm10k_udp_tunnel_del
521 * @netdev: network interface device structure
522 * @ti: Tunnel endpoint information
524 * This function is called when a new UDP tunnel port is deleted. The freed
525 * port will be removed from the list, then we reprogram the offloaded port
526 * based on the head of the list.
528 static void fm10k_udp_tunnel_del(struct net_device *dev,
529 struct udp_tunnel_info *ti)
531 struct fm10k_intfc *interface = netdev_priv(dev);
532 struct fm10k_udp_port *port = NULL;
534 if (interface->hw.mac.type != fm10k_mac_pf)
538 case UDP_TUNNEL_TYPE_VXLAN:
539 port = fm10k_remove_tunnel_port(&interface->vxlan_port, ti);
541 case UDP_TUNNEL_TYPE_GENEVE:
542 port = fm10k_remove_tunnel_port(&interface->geneve_port, ti);
548 /* if we did remove a port we need to free its memory */
551 fm10k_restore_udp_port_info(interface);
555 * fm10k_open - Called when a network interface is made active
556 * @netdev: network interface device structure
558 * Returns 0 on success, negative value on failure
560 * The open entry point is called when a network interface is made
561 * active by the system (IFF_UP). At this point all resources needed
562 * for transmit and receive operations are allocated, the interrupt
563 * handler is registered with the OS, the watchdog timer is started,
564 * and the stack is notified that the interface is ready.
566 int fm10k_open(struct net_device *netdev)
568 struct fm10k_intfc *interface = netdev_priv(netdev);
571 /* allocate transmit descriptors */
572 err = fm10k_setup_all_tx_resources(interface);
576 /* allocate receive descriptors */
577 err = fm10k_setup_all_rx_resources(interface);
581 /* allocate interrupt resources */
582 err = fm10k_qv_request_irq(interface);
586 /* setup GLORT assignment for this port */
587 fm10k_request_glort_range(interface);
589 /* Notify the stack of the actual queue counts */
590 err = netif_set_real_num_tx_queues(netdev,
591 interface->num_tx_queues);
595 err = netif_set_real_num_rx_queues(netdev,
596 interface->num_rx_queues);
600 udp_tunnel_get_rx_info(netdev);
607 fm10k_qv_free_irq(interface);
609 fm10k_free_all_rx_resources(interface);
611 fm10k_free_all_tx_resources(interface);
617 * fm10k_close - Disables a network interface
618 * @netdev: network interface device structure
620 * Returns 0, this is not allowed to fail
622 * The close entry point is called when an interface is de-activated
623 * by the OS. The hardware is still under the drivers control, but
624 * needs to be disabled. A global MAC reset is issued to stop the
625 * hardware, and all transmit and receive resources are freed.
627 int fm10k_close(struct net_device *netdev)
629 struct fm10k_intfc *interface = netdev_priv(netdev);
631 fm10k_down(interface);
633 fm10k_qv_free_irq(interface);
635 fm10k_free_udp_port_info(interface);
637 fm10k_free_all_tx_resources(interface);
638 fm10k_free_all_rx_resources(interface);
643 static netdev_tx_t fm10k_xmit_frame(struct sk_buff *skb, struct net_device *dev)
645 struct fm10k_intfc *interface = netdev_priv(dev);
646 unsigned int r_idx = skb->queue_mapping;
649 if ((skb->protocol == htons(ETH_P_8021Q)) &&
650 !skb_vlan_tag_present(skb)) {
651 /* FM10K only supports hardware tagging, any tags in frame
652 * are considered 2nd level or "outer" tags
654 struct vlan_hdr *vhdr;
657 /* make sure skb is not shared */
658 skb = skb_share_check(skb, GFP_ATOMIC);
662 /* make sure there is enough room to move the ethernet header */
663 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
666 /* verify the skb head is not shared */
667 err = skb_cow_head(skb, 0);
673 /* locate VLAN header */
674 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
676 /* pull the 2 key pieces of data out of it */
677 __vlan_hwaccel_put_tag(skb,
679 ntohs(vhdr->h_vlan_TCI));
680 proto = vhdr->h_vlan_encapsulated_proto;
681 skb->protocol = (ntohs(proto) >= 1536) ? proto :
684 /* squash it by moving the ethernet addresses up 4 bytes */
685 memmove(skb->data + VLAN_HLEN, skb->data, 12);
686 __skb_pull(skb, VLAN_HLEN);
687 skb_reset_mac_header(skb);
690 /* The minimum packet size for a single buffer is 17B so pad the skb
691 * in order to meet this minimum size requirement.
693 if (unlikely(skb->len < 17)) {
694 int pad_len = 17 - skb->len;
696 if (skb_pad(skb, pad_len))
698 __skb_put(skb, pad_len);
701 if (r_idx >= interface->num_tx_queues)
702 r_idx %= interface->num_tx_queues;
704 err = fm10k_xmit_frame_ring(skb, interface->tx_ring[r_idx]);
710 * fm10k_tx_timeout - Respond to a Tx Hang
711 * @netdev: network interface device structure
713 static void fm10k_tx_timeout(struct net_device *netdev)
715 struct fm10k_intfc *interface = netdev_priv(netdev);
716 bool real_tx_hang = false;
719 #define TX_TIMEO_LIMIT 16000
720 for (i = 0; i < interface->num_tx_queues; i++) {
721 struct fm10k_ring *tx_ring = interface->tx_ring[i];
723 if (check_for_tx_hang(tx_ring) && fm10k_check_tx_hang(tx_ring))
728 fm10k_tx_timeout_reset(interface);
730 netif_info(interface, drv, netdev,
731 "Fake Tx hang detected with timeout of %d seconds\n",
732 netdev->watchdog_timeo / HZ);
734 /* fake Tx hang - increase the kernel timeout */
735 if (netdev->watchdog_timeo < TX_TIMEO_LIMIT)
736 netdev->watchdog_timeo *= 2;
741 * fm10k_host_mbx_ready - Check PF interface's mailbox readiness
742 * @interface: board private structure
744 * This function checks if the PF interface's mailbox is ready before queueing
745 * mailbox messages for transmission. This will prevent filling the TX mailbox
746 * queue when the receiver is not ready. VF interfaces are exempt from this
747 * check since it will block all PF-VF mailbox messages from being sent from
748 * the VF to the PF at initialization.
750 static bool fm10k_host_mbx_ready(struct fm10k_intfc *interface)
752 struct fm10k_hw *hw = &interface->hw;
754 return (hw->mac.type == fm10k_mac_vf || interface->host_ready);
757 static int fm10k_uc_vlan_unsync(struct net_device *netdev,
758 const unsigned char *uc_addr)
760 struct fm10k_intfc *interface = netdev_priv(netdev);
761 struct fm10k_hw *hw = &interface->hw;
762 u16 glort = interface->glort;
763 u16 vid = interface->vid;
764 bool set = !!(vid / VLAN_N_VID);
765 int err = -EHOSTDOWN;
767 /* drop any leading bits on the VLAN ID */
768 vid &= VLAN_N_VID - 1;
770 if (fm10k_host_mbx_ready(interface))
771 err = hw->mac.ops.update_uc_addr(hw, glort, uc_addr,
777 /* return non-zero value as we are only doing a partial sync/unsync */
781 static int fm10k_mc_vlan_unsync(struct net_device *netdev,
782 const unsigned char *mc_addr)
784 struct fm10k_intfc *interface = netdev_priv(netdev);
785 struct fm10k_hw *hw = &interface->hw;
786 u16 glort = interface->glort;
787 u16 vid = interface->vid;
788 bool set = !!(vid / VLAN_N_VID);
789 int err = -EHOSTDOWN;
791 /* drop any leading bits on the VLAN ID */
792 vid &= VLAN_N_VID - 1;
794 if (fm10k_host_mbx_ready(interface))
795 err = hw->mac.ops.update_mc_addr(hw, glort, mc_addr, vid, set);
800 /* return non-zero value as we are only doing a partial sync/unsync */
804 static int fm10k_update_vid(struct net_device *netdev, u16 vid, bool set)
806 struct fm10k_intfc *interface = netdev_priv(netdev);
807 struct fm10k_hw *hw = &interface->hw;
811 /* updates do not apply to VLAN 0 */
815 if (vid >= VLAN_N_VID)
818 /* Verify we have permission to add VLANs */
819 if (hw->mac.vlan_override)
822 /* update active_vlans bitmask */
823 set_bit(vid, interface->active_vlans);
825 clear_bit(vid, interface->active_vlans);
827 /* disable the default VLAN ID on ring if we have an active VLAN */
828 for (i = 0; i < interface->num_rx_queues; i++) {
829 struct fm10k_ring *rx_ring = interface->rx_ring[i];
830 u16 rx_vid = rx_ring->vid & (VLAN_N_VID - 1);
832 if (test_bit(rx_vid, interface->active_vlans))
833 rx_ring->vid |= FM10K_VLAN_CLEAR;
835 rx_ring->vid &= ~FM10K_VLAN_CLEAR;
838 /* Do not remove default VLAN ID related entries from VLAN and MAC
841 if (!set && vid == hw->mac.default_vid)
844 /* Do not throw an error if the interface is down. We will sync once
847 if (test_bit(__FM10K_DOWN, interface->state))
850 fm10k_mbx_lock(interface);
852 /* only need to update the VLAN if not in promiscuous mode */
853 if (!(netdev->flags & IFF_PROMISC)) {
854 err = hw->mac.ops.update_vlan(hw, vid, 0, set);
859 /* update our base MAC address if host's mailbox is ready */
860 if (fm10k_host_mbx_ready(interface))
861 err = hw->mac.ops.update_uc_addr(hw, interface->glort,
862 hw->mac.addr, vid, set, 0);
869 /* set VLAN ID prior to syncing/unsyncing the VLAN */
870 interface->vid = vid + (set ? VLAN_N_VID : 0);
872 /* Update the unicast and multicast address list to add/drop VLAN */
873 __dev_uc_unsync(netdev, fm10k_uc_vlan_unsync);
874 __dev_mc_unsync(netdev, fm10k_mc_vlan_unsync);
877 fm10k_mbx_unlock(interface);
882 static int fm10k_vlan_rx_add_vid(struct net_device *netdev,
883 __always_unused __be16 proto, u16 vid)
885 /* update VLAN and address table based on changes */
886 return fm10k_update_vid(netdev, vid, true);
889 static int fm10k_vlan_rx_kill_vid(struct net_device *netdev,
890 __always_unused __be16 proto, u16 vid)
892 /* update VLAN and address table based on changes */
893 return fm10k_update_vid(netdev, vid, false);
896 static u16 fm10k_find_next_vlan(struct fm10k_intfc *interface, u16 vid)
898 struct fm10k_hw *hw = &interface->hw;
899 u16 default_vid = hw->mac.default_vid;
900 u16 vid_limit = vid < default_vid ? default_vid : VLAN_N_VID;
902 vid = find_next_bit(interface->active_vlans, vid_limit, ++vid);
907 static void fm10k_clear_unused_vlans(struct fm10k_intfc *interface)
909 struct fm10k_hw *hw = &interface->hw;
912 /* loop through and find any gaps in the table */
913 for (vid = 0, prev_vid = 0;
914 prev_vid < VLAN_N_VID;
915 prev_vid = vid + 1, vid = fm10k_find_next_vlan(interface, vid)) {
919 /* send request to clear multiple bits at a time */
920 prev_vid += (vid - prev_vid - 1) << FM10K_VLAN_LENGTH_SHIFT;
921 hw->mac.ops.update_vlan(hw, prev_vid, 0, false);
925 static int __fm10k_uc_sync(struct net_device *dev,
926 const unsigned char *addr, bool sync)
928 struct fm10k_intfc *interface = netdev_priv(dev);
929 struct fm10k_hw *hw = &interface->hw;
930 u16 vid, glort = interface->glort;
933 if (!is_valid_ether_addr(addr))
934 return -EADDRNOTAVAIL;
936 /* update table with current entries if host's mailbox is ready */
937 if (!fm10k_host_mbx_ready(interface))
940 for (vid = hw->mac.default_vid ? fm10k_find_next_vlan(interface, 0) : 1;
942 vid = fm10k_find_next_vlan(interface, vid)) {
943 err = hw->mac.ops.update_uc_addr(hw, glort, addr,
952 static int fm10k_uc_sync(struct net_device *dev,
953 const unsigned char *addr)
955 return __fm10k_uc_sync(dev, addr, true);
958 static int fm10k_uc_unsync(struct net_device *dev,
959 const unsigned char *addr)
961 return __fm10k_uc_sync(dev, addr, false);
964 static int fm10k_set_mac(struct net_device *dev, void *p)
966 struct fm10k_intfc *interface = netdev_priv(dev);
967 struct fm10k_hw *hw = &interface->hw;
968 struct sockaddr *addr = p;
971 if (!is_valid_ether_addr(addr->sa_data))
972 return -EADDRNOTAVAIL;
974 if (dev->flags & IFF_UP) {
975 /* setting MAC address requires mailbox */
976 fm10k_mbx_lock(interface);
978 err = fm10k_uc_sync(dev, addr->sa_data);
980 fm10k_uc_unsync(dev, hw->mac.addr);
982 fm10k_mbx_unlock(interface);
986 ether_addr_copy(dev->dev_addr, addr->sa_data);
987 ether_addr_copy(hw->mac.addr, addr->sa_data);
988 dev->addr_assign_type &= ~NET_ADDR_RANDOM;
991 /* if we had a mailbox error suggest trying again */
992 return err ? -EAGAIN : 0;
995 static int __fm10k_mc_sync(struct net_device *dev,
996 const unsigned char *addr, bool sync)
998 struct fm10k_intfc *interface = netdev_priv(dev);
999 struct fm10k_hw *hw = &interface->hw;
1000 u16 vid, glort = interface->glort;
1002 /* update table with current entries if host's mailbox is ready */
1003 if (!fm10k_host_mbx_ready(interface))
1006 for (vid = hw->mac.default_vid ? fm10k_find_next_vlan(interface, 0) : 1;
1008 vid = fm10k_find_next_vlan(interface, vid)) {
1009 hw->mac.ops.update_mc_addr(hw, glort, addr, vid, sync);
1015 static int fm10k_mc_sync(struct net_device *dev,
1016 const unsigned char *addr)
1018 return __fm10k_mc_sync(dev, addr, true);
1021 static int fm10k_mc_unsync(struct net_device *dev,
1022 const unsigned char *addr)
1024 return __fm10k_mc_sync(dev, addr, false);
1027 static void fm10k_set_rx_mode(struct net_device *dev)
1029 struct fm10k_intfc *interface = netdev_priv(dev);
1030 struct fm10k_hw *hw = &interface->hw;
1033 /* no need to update the harwdare if we are not running */
1034 if (!(dev->flags & IFF_UP))
1037 /* determine new mode based on flags */
1038 xcast_mode = (dev->flags & IFF_PROMISC) ? FM10K_XCAST_MODE_PROMISC :
1039 (dev->flags & IFF_ALLMULTI) ? FM10K_XCAST_MODE_ALLMULTI :
1040 (dev->flags & (IFF_BROADCAST | IFF_MULTICAST)) ?
1041 FM10K_XCAST_MODE_MULTI : FM10K_XCAST_MODE_NONE;
1043 fm10k_mbx_lock(interface);
1045 /* update xcast mode first, but only if it changed */
1046 if (interface->xcast_mode != xcast_mode) {
1047 /* update VLAN table */
1048 if (xcast_mode == FM10K_XCAST_MODE_PROMISC)
1049 hw->mac.ops.update_vlan(hw, FM10K_VLAN_ALL, 0, true);
1050 if (interface->xcast_mode == FM10K_XCAST_MODE_PROMISC)
1051 fm10k_clear_unused_vlans(interface);
1053 /* update xcast mode if host's mailbox is ready */
1054 if (fm10k_host_mbx_ready(interface))
1055 hw->mac.ops.update_xcast_mode(hw, interface->glort,
1058 /* record updated xcast mode state */
1059 interface->xcast_mode = xcast_mode;
1062 /* synchronize all of the addresses */
1063 __dev_uc_sync(dev, fm10k_uc_sync, fm10k_uc_unsync);
1064 __dev_mc_sync(dev, fm10k_mc_sync, fm10k_mc_unsync);
1066 fm10k_mbx_unlock(interface);
1069 void fm10k_restore_rx_state(struct fm10k_intfc *interface)
1071 struct net_device *netdev = interface->netdev;
1072 struct fm10k_hw *hw = &interface->hw;
1076 /* record glort for this interface */
1077 glort = interface->glort;
1079 /* convert interface flags to xcast mode */
1080 if (netdev->flags & IFF_PROMISC)
1081 xcast_mode = FM10K_XCAST_MODE_PROMISC;
1082 else if (netdev->flags & IFF_ALLMULTI)
1083 xcast_mode = FM10K_XCAST_MODE_ALLMULTI;
1084 else if (netdev->flags & (IFF_BROADCAST | IFF_MULTICAST))
1085 xcast_mode = FM10K_XCAST_MODE_MULTI;
1087 xcast_mode = FM10K_XCAST_MODE_NONE;
1089 fm10k_mbx_lock(interface);
1091 /* Enable logical port if host's mailbox is ready */
1092 if (fm10k_host_mbx_ready(interface))
1093 hw->mac.ops.update_lport_state(hw, glort,
1094 interface->glort_count, true);
1096 /* update VLAN table */
1097 hw->mac.ops.update_vlan(hw, FM10K_VLAN_ALL, 0,
1098 xcast_mode == FM10K_XCAST_MODE_PROMISC);
1100 /* Add filter for VLAN 0 */
1101 hw->mac.ops.update_vlan(hw, 0, 0, true);
1103 /* update table with current entries */
1104 for (vid = hw->mac.default_vid ? fm10k_find_next_vlan(interface, 0) : 1;
1106 vid = fm10k_find_next_vlan(interface, vid)) {
1107 hw->mac.ops.update_vlan(hw, vid, 0, true);
1109 /* Update unicast entries if host's mailbox is ready */
1110 if (fm10k_host_mbx_ready(interface))
1111 hw->mac.ops.update_uc_addr(hw, glort, hw->mac.addr,
1115 /* update xcast mode before synchronizing addresses if host's mailbox
1118 if (fm10k_host_mbx_ready(interface))
1119 hw->mac.ops.update_xcast_mode(hw, glort, xcast_mode);
1121 /* synchronize all of the addresses */
1122 __dev_uc_sync(netdev, fm10k_uc_sync, fm10k_uc_unsync);
1123 __dev_mc_sync(netdev, fm10k_mc_sync, fm10k_mc_unsync);
1125 fm10k_mbx_unlock(interface);
1127 /* record updated xcast mode state */
1128 interface->xcast_mode = xcast_mode;
1130 /* Restore tunnel configuration */
1131 fm10k_restore_udp_port_info(interface);
1134 void fm10k_reset_rx_state(struct fm10k_intfc *interface)
1136 struct net_device *netdev = interface->netdev;
1137 struct fm10k_hw *hw = &interface->hw;
1139 fm10k_mbx_lock(interface);
1141 /* clear the logical port state on lower device if host's mailbox is
1144 if (fm10k_host_mbx_ready(interface))
1145 hw->mac.ops.update_lport_state(hw, interface->glort,
1146 interface->glort_count, false);
1148 fm10k_mbx_unlock(interface);
1150 /* reset flags to default state */
1151 interface->xcast_mode = FM10K_XCAST_MODE_NONE;
1153 /* clear the sync flag since the lport has been dropped */
1154 __dev_uc_unsync(netdev, NULL);
1155 __dev_mc_unsync(netdev, NULL);
1159 * fm10k_get_stats64 - Get System Network Statistics
1160 * @netdev: network interface device structure
1161 * @stats: storage space for 64bit statistics
1163 * Obtain 64bit statistics in a way that is safe for both 32bit and 64bit
1166 static void fm10k_get_stats64(struct net_device *netdev,
1167 struct rtnl_link_stats64 *stats)
1169 struct fm10k_intfc *interface = netdev_priv(netdev);
1170 struct fm10k_ring *ring;
1171 unsigned int start, i;
1176 for (i = 0; i < interface->num_rx_queues; i++) {
1177 ring = READ_ONCE(interface->rx_ring[i]);
1183 start = u64_stats_fetch_begin_irq(&ring->syncp);
1184 packets = ring->stats.packets;
1185 bytes = ring->stats.bytes;
1186 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1188 stats->rx_packets += packets;
1189 stats->rx_bytes += bytes;
1192 for (i = 0; i < interface->num_tx_queues; i++) {
1193 ring = READ_ONCE(interface->tx_ring[i]);
1199 start = u64_stats_fetch_begin_irq(&ring->syncp);
1200 packets = ring->stats.packets;
1201 bytes = ring->stats.bytes;
1202 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1204 stats->tx_packets += packets;
1205 stats->tx_bytes += bytes;
1210 /* following stats updated by fm10k_service_task() */
1211 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1214 int fm10k_setup_tc(struct net_device *dev, u8 tc)
1216 struct fm10k_intfc *interface = netdev_priv(dev);
1219 /* Currently only the PF supports priority classes */
1220 if (tc && (interface->hw.mac.type != fm10k_mac_pf))
1223 /* Hardware supports up to 8 traffic classes */
1227 /* Hardware has to reinitialize queues to match packet
1228 * buffer alignment. Unfortunately, the hardware is not
1229 * flexible enough to do this dynamically.
1231 if (netif_running(dev))
1234 fm10k_mbx_free_irq(interface);
1236 fm10k_clear_queueing_scheme(interface);
1238 /* we expect the prio_tc map to be repopulated later */
1239 netdev_reset_tc(dev);
1240 netdev_set_num_tc(dev, tc);
1242 err = fm10k_init_queueing_scheme(interface);
1244 goto err_queueing_scheme;
1246 err = fm10k_mbx_request_irq(interface);
1250 err = netif_running(dev) ? fm10k_open(dev) : 0;
1254 /* flag to indicate SWPRI has yet to be updated */
1255 set_bit(FM10K_FLAG_SWPRI_CONFIG, interface->flags);
1259 fm10k_mbx_free_irq(interface);
1261 fm10k_clear_queueing_scheme(interface);
1262 err_queueing_scheme:
1263 netif_device_detach(dev);
1268 static int __fm10k_setup_tc(struct net_device *dev, enum tc_setup_type type,
1269 u32 handle, u32 chain_index, __be16 proto,
1270 struct tc_to_netdev *tc)
1272 if (type != TC_SETUP_MQPRIO)
1275 tc->mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
1277 return fm10k_setup_tc(dev, tc->mqprio->num_tc);
1280 static void fm10k_assign_l2_accel(struct fm10k_intfc *interface,
1281 struct fm10k_l2_accel *l2_accel)
1283 struct fm10k_ring *ring;
1286 for (i = 0; i < interface->num_rx_queues; i++) {
1287 ring = interface->rx_ring[i];
1288 rcu_assign_pointer(ring->l2_accel, l2_accel);
1291 interface->l2_accel = l2_accel;
1294 static void *fm10k_dfwd_add_station(struct net_device *dev,
1295 struct net_device *sdev)
1297 struct fm10k_intfc *interface = netdev_priv(dev);
1298 struct fm10k_l2_accel *l2_accel = interface->l2_accel;
1299 struct fm10k_l2_accel *old_l2_accel = NULL;
1300 struct fm10k_dglort_cfg dglort = { 0 };
1301 struct fm10k_hw *hw = &interface->hw;
1305 /* allocate l2 accel structure if it is not available */
1307 /* verify there is enough free GLORTs to support l2_accel */
1308 if (interface->glort_count < 7)
1309 return ERR_PTR(-EBUSY);
1311 size = offsetof(struct fm10k_l2_accel, macvlan[7]);
1312 l2_accel = kzalloc(size, GFP_KERNEL);
1314 return ERR_PTR(-ENOMEM);
1317 l2_accel->dglort = interface->glort;
1319 /* update pointers */
1320 fm10k_assign_l2_accel(interface, l2_accel);
1321 /* do not expand if we are at our limit */
1322 } else if ((l2_accel->count == FM10K_MAX_STATIONS) ||
1323 (l2_accel->count == (interface->glort_count - 1))) {
1324 return ERR_PTR(-EBUSY);
1325 /* expand if we have hit the size limit */
1326 } else if (l2_accel->count == l2_accel->size) {
1327 old_l2_accel = l2_accel;
1328 size = offsetof(struct fm10k_l2_accel,
1329 macvlan[(l2_accel->size * 2) + 1]);
1330 l2_accel = kzalloc(size, GFP_KERNEL);
1332 return ERR_PTR(-ENOMEM);
1334 memcpy(l2_accel, old_l2_accel,
1335 offsetof(struct fm10k_l2_accel,
1336 macvlan[old_l2_accel->size]));
1338 l2_accel->size = (old_l2_accel->size * 2) + 1;
1340 /* update pointers */
1341 fm10k_assign_l2_accel(interface, l2_accel);
1342 kfree_rcu(old_l2_accel, rcu);
1345 /* add macvlan to accel table, and record GLORT for position */
1346 for (i = 0; i < l2_accel->size; i++) {
1347 if (!l2_accel->macvlan[i])
1351 /* record station */
1352 l2_accel->macvlan[i] = sdev;
1355 /* configure default DGLORT mapping for RSS/DCB */
1356 dglort.idx = fm10k_dglort_pf_rss;
1357 dglort.inner_rss = 1;
1358 dglort.rss_l = fls(interface->ring_feature[RING_F_RSS].mask);
1359 dglort.pc_l = fls(interface->ring_feature[RING_F_QOS].mask);
1360 dglort.glort = interface->glort;
1361 dglort.shared_l = fls(l2_accel->size);
1362 hw->mac.ops.configure_dglort_map(hw, &dglort);
1364 /* Add rules for this specific dglort to the switch */
1365 fm10k_mbx_lock(interface);
1367 glort = l2_accel->dglort + 1 + i;
1369 if (fm10k_host_mbx_ready(interface)) {
1370 hw->mac.ops.update_xcast_mode(hw, glort,
1371 FM10K_XCAST_MODE_MULTI);
1372 hw->mac.ops.update_uc_addr(hw, glort, sdev->dev_addr,
1376 fm10k_mbx_unlock(interface);
1381 static void fm10k_dfwd_del_station(struct net_device *dev, void *priv)
1383 struct fm10k_intfc *interface = netdev_priv(dev);
1384 struct fm10k_l2_accel *l2_accel = READ_ONCE(interface->l2_accel);
1385 struct fm10k_dglort_cfg dglort = { 0 };
1386 struct fm10k_hw *hw = &interface->hw;
1387 struct net_device *sdev = priv;
1394 /* search table for matching interface */
1395 for (i = 0; i < l2_accel->size; i++) {
1396 if (l2_accel->macvlan[i] == sdev)
1400 /* exit if macvlan not found */
1401 if (i == l2_accel->size)
1404 /* Remove any rules specific to this dglort */
1405 fm10k_mbx_lock(interface);
1407 glort = l2_accel->dglort + 1 + i;
1409 if (fm10k_host_mbx_ready(interface)) {
1410 hw->mac.ops.update_xcast_mode(hw, glort,
1411 FM10K_XCAST_MODE_NONE);
1412 hw->mac.ops.update_uc_addr(hw, glort, sdev->dev_addr,
1416 fm10k_mbx_unlock(interface);
1418 /* record removal */
1419 l2_accel->macvlan[i] = NULL;
1422 /* configure default DGLORT mapping for RSS/DCB */
1423 dglort.idx = fm10k_dglort_pf_rss;
1424 dglort.inner_rss = 1;
1425 dglort.rss_l = fls(interface->ring_feature[RING_F_RSS].mask);
1426 dglort.pc_l = fls(interface->ring_feature[RING_F_QOS].mask);
1427 dglort.glort = interface->glort;
1428 dglort.shared_l = fls(l2_accel->size);
1429 hw->mac.ops.configure_dglort_map(hw, &dglort);
1431 /* If table is empty remove it */
1432 if (l2_accel->count == 0) {
1433 fm10k_assign_l2_accel(interface, NULL);
1434 kfree_rcu(l2_accel, rcu);
1438 static netdev_features_t fm10k_features_check(struct sk_buff *skb,
1439 struct net_device *dev,
1440 netdev_features_t features)
1442 if (!skb->encapsulation || fm10k_tx_encap_offload(skb))
1445 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
1448 static const struct net_device_ops fm10k_netdev_ops = {
1449 .ndo_open = fm10k_open,
1450 .ndo_stop = fm10k_close,
1451 .ndo_validate_addr = eth_validate_addr,
1452 .ndo_start_xmit = fm10k_xmit_frame,
1453 .ndo_set_mac_address = fm10k_set_mac,
1454 .ndo_tx_timeout = fm10k_tx_timeout,
1455 .ndo_vlan_rx_add_vid = fm10k_vlan_rx_add_vid,
1456 .ndo_vlan_rx_kill_vid = fm10k_vlan_rx_kill_vid,
1457 .ndo_set_rx_mode = fm10k_set_rx_mode,
1458 .ndo_get_stats64 = fm10k_get_stats64,
1459 .ndo_setup_tc = __fm10k_setup_tc,
1460 .ndo_set_vf_mac = fm10k_ndo_set_vf_mac,
1461 .ndo_set_vf_vlan = fm10k_ndo_set_vf_vlan,
1462 .ndo_set_vf_rate = fm10k_ndo_set_vf_bw,
1463 .ndo_get_vf_config = fm10k_ndo_get_vf_config,
1464 .ndo_udp_tunnel_add = fm10k_udp_tunnel_add,
1465 .ndo_udp_tunnel_del = fm10k_udp_tunnel_del,
1466 .ndo_dfwd_add_station = fm10k_dfwd_add_station,
1467 .ndo_dfwd_del_station = fm10k_dfwd_del_station,
1468 #ifdef CONFIG_NET_POLL_CONTROLLER
1469 .ndo_poll_controller = fm10k_netpoll,
1471 .ndo_features_check = fm10k_features_check,
1474 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
1476 struct net_device *fm10k_alloc_netdev(const struct fm10k_info *info)
1478 netdev_features_t hw_features;
1479 struct fm10k_intfc *interface;
1480 struct net_device *dev;
1482 dev = alloc_etherdev_mq(sizeof(struct fm10k_intfc), MAX_QUEUES);
1486 /* set net device and ethtool ops */
1487 dev->netdev_ops = &fm10k_netdev_ops;
1488 fm10k_set_ethtool_ops(dev);
1490 /* configure default debug level */
1491 interface = netdev_priv(dev);
1492 interface->msg_enable = BIT(DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
1494 /* configure default features */
1495 dev->features |= NETIF_F_IP_CSUM |
1504 /* Only the PF can support VXLAN and NVGRE tunnel offloads */
1505 if (info->mac == fm10k_mac_pf) {
1506 dev->hw_enc_features = NETIF_F_IP_CSUM |
1510 NETIF_F_GSO_UDP_TUNNEL |
1514 dev->features |= NETIF_F_GSO_UDP_TUNNEL;
1517 /* all features defined to this point should be changeable */
1518 hw_features = dev->features;
1520 /* allow user to enable L2 forwarding acceleration */
1521 hw_features |= NETIF_F_HW_L2FW_DOFFLOAD;
1523 /* configure VLAN features */
1524 dev->vlan_features |= dev->features;
1526 /* we want to leave these both on as we cannot disable VLAN tag
1527 * insertion or stripping on the hardware since it is contained
1528 * in the FTAG and not in the frame itself.
1530 dev->features |= NETIF_F_HW_VLAN_CTAG_TX |
1531 NETIF_F_HW_VLAN_CTAG_RX |
1532 NETIF_F_HW_VLAN_CTAG_FILTER;
1534 dev->priv_flags |= IFF_UNICAST_FLT;
1536 dev->hw_features |= hw_features;
1538 /* MTU range: 68 - 15342 */
1539 dev->min_mtu = ETH_MIN_MTU;
1540 dev->max_mtu = FM10K_MAX_JUMBO_FRAME_SIZE;