2 * Copyright (C) 2015 Netronome Systems, Inc.
4 * This software is dual licensed under the GNU General License Version 2,
5 * June 1991 as shown in the file COPYING in the top-level directory of this
6 * source tree or the BSD 2-Clause License provided below. You have the
7 * option to license this software under the complete terms of either license.
9 * The BSD 2-Clause License:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * 1. Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
19 * 2. Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 * Netronome network device driver: Common functions between PF and VF
37 * Authors: Jakub Kicinski <jakub.kicinski@netronome.com>
38 * Jason McMullan <jason.mcmullan@netronome.com>
39 * Rolf Neugebauer <rolf.neugebauer@netronome.com>
40 * Brad Petrus <brad.petrus@netronome.com>
41 * Chris Telfer <chris.telfer@netronome.com>
44 #include <linux/module.h>
45 #include <linux/kernel.h>
46 #include <linux/init.h>
48 #include <linux/netdevice.h>
49 #include <linux/etherdevice.h>
50 #include <linux/interrupt.h>
52 #include <linux/ipv6.h>
53 #include <linux/pci.h>
54 #include <linux/pci_regs.h>
55 #include <linux/msi.h>
56 #include <linux/ethtool.h>
57 #include <linux/log2.h>
58 #include <linux/if_vlan.h>
59 #include <linux/random.h>
61 #include <linux/ktime.h>
63 #include <net/pkt_cls.h>
64 #include <net/vxlan.h>
66 #include "nfp_net_ctrl.h"
70 * nfp_net_get_fw_version() - Read and parse the FW version
71 * @fw_ver: Output fw_version structure to read to
72 * @ctrl_bar: Mapped address of the control BAR
74 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
75 void __iomem *ctrl_bar)
79 reg = readl(ctrl_bar + NFP_NET_CFG_VERSION);
80 put_unaligned_le32(reg, fw_ver);
85 * Firmware reconfig may take a while so we have two versions of it -
86 * synchronous and asynchronous (posted). All synchronous callers are holding
87 * RTNL so we don't have to worry about serializing them.
89 static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update)
91 nn_writel(nn, NFP_NET_CFG_UPDATE, update);
92 /* ensure update is written before pinging HW */
94 nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1);
97 /* Pass 0 as update to run posted reconfigs. */
98 static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update)
100 update |= nn->reconfig_posted;
101 nn->reconfig_posted = 0;
103 nfp_net_reconfig_start(nn, update);
105 nn->reconfig_timer_active = true;
106 mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ);
109 static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check)
113 reg = nn_readl(nn, NFP_NET_CFG_UPDATE);
116 if (reg & NFP_NET_CFG_UPDATE_ERR) {
117 nn_err(nn, "Reconfig error: 0x%08x\n", reg);
119 } else if (last_check) {
120 nn_err(nn, "Reconfig timeout: 0x%08x\n", reg);
127 static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
129 bool timed_out = false;
131 /* Poll update field, waiting for NFP to ack the config */
132 while (!nfp_net_reconfig_check_done(nn, timed_out)) {
134 timed_out = time_is_before_eq_jiffies(deadline);
137 if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR)
140 return timed_out ? -EIO : 0;
143 static void nfp_net_reconfig_timer(unsigned long data)
145 struct nfp_net *nn = (void *)data;
147 spin_lock_bh(&nn->reconfig_lock);
149 nn->reconfig_timer_active = false;
151 /* If sync caller is present it will take over from us */
152 if (nn->reconfig_sync_present)
155 /* Read reconfig status and report errors */
156 nfp_net_reconfig_check_done(nn, true);
158 if (nn->reconfig_posted)
159 nfp_net_reconfig_start_async(nn, 0);
161 spin_unlock_bh(&nn->reconfig_lock);
165 * nfp_net_reconfig_post() - Post async reconfig request
166 * @nn: NFP Net device to reconfigure
167 * @update: The value for the update field in the BAR config
169 * Record FW reconfiguration request. Reconfiguration will be kicked off
170 * whenever reconfiguration machinery is idle. Multiple requests can be
173 static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update)
175 spin_lock_bh(&nn->reconfig_lock);
177 /* Sync caller will kick off async reconf when it's done, just post */
178 if (nn->reconfig_sync_present) {
179 nn->reconfig_posted |= update;
183 /* Opportunistically check if the previous command is done */
184 if (!nn->reconfig_timer_active ||
185 nfp_net_reconfig_check_done(nn, false))
186 nfp_net_reconfig_start_async(nn, update);
188 nn->reconfig_posted |= update;
190 spin_unlock_bh(&nn->reconfig_lock);
194 * nfp_net_reconfig() - Reconfigure the firmware
195 * @nn: NFP Net device to reconfigure
196 * @update: The value for the update field in the BAR config
198 * Write the update word to the BAR and ping the reconfig queue. The
199 * poll until the firmware has acknowledged the update by zeroing the
202 * Return: Negative errno on error, 0 on success
204 int nfp_net_reconfig(struct nfp_net *nn, u32 update)
206 bool cancelled_timer = false;
207 u32 pre_posted_requests;
210 spin_lock_bh(&nn->reconfig_lock);
212 nn->reconfig_sync_present = true;
214 if (nn->reconfig_timer_active) {
215 del_timer(&nn->reconfig_timer);
216 nn->reconfig_timer_active = false;
217 cancelled_timer = true;
219 pre_posted_requests = nn->reconfig_posted;
220 nn->reconfig_posted = 0;
222 spin_unlock_bh(&nn->reconfig_lock);
225 nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires);
227 /* Run the posted reconfigs which were issued before we started */
228 if (pre_posted_requests) {
229 nfp_net_reconfig_start(nn, pre_posted_requests);
230 nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
233 nfp_net_reconfig_start(nn, update);
234 ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
236 spin_lock_bh(&nn->reconfig_lock);
238 if (nn->reconfig_posted)
239 nfp_net_reconfig_start_async(nn, 0);
241 nn->reconfig_sync_present = false;
243 spin_unlock_bh(&nn->reconfig_lock);
248 /* Interrupt configuration and handling
252 * nfp_net_irq_unmask_msix() - Unmask MSI-X after automasking
253 * @nn: NFP Network structure
254 * @entry_nr: MSI-X table entry
256 * Clear the MSI-X table mask bit for the given entry bypassing Linux irq
257 * handling subsystem. Use *only* to reenable automasked vectors.
259 static void nfp_net_irq_unmask_msix(struct nfp_net *nn, unsigned int entry_nr)
261 struct list_head *msi_head = &nn->pdev->dev.msi_list;
262 struct msi_desc *entry;
265 /* All MSI-Xs have the same mask_base */
266 entry = list_first_entry(msi_head, struct msi_desc, list);
268 off = (PCI_MSIX_ENTRY_SIZE * entry_nr) +
269 PCI_MSIX_ENTRY_VECTOR_CTRL;
270 writel(0, entry->mask_base + off);
271 readl(entry->mask_base);
275 * nfp_net_irq_unmask() - Unmask automasked interrupt
276 * @nn: NFP Network structure
277 * @entry_nr: MSI-X table entry
279 * If MSI-X auto-masking is enabled clear the mask bit, otherwise
280 * clear the ICR for the entry.
282 static void nfp_net_irq_unmask(struct nfp_net *nn, unsigned int entry_nr)
284 if (nn->ctrl & NFP_NET_CFG_CTRL_MSIXAUTO) {
285 nfp_net_irq_unmask_msix(nn, entry_nr);
289 nn_writeb(nn, NFP_NET_CFG_ICR(entry_nr), NFP_NET_CFG_ICR_UNMASKED);
294 * nfp_net_msix_alloc() - Try to allocate MSI-X irqs
295 * @nn: NFP Network structure
296 * @nr_vecs: Number of MSI-X vectors to allocate
298 * For MSI-X we want at least NFP_NET_NON_Q_VECTORS + 1 vectors.
300 * Return: Number of MSI-X vectors obtained or 0 on error.
302 static int nfp_net_msix_alloc(struct nfp_net *nn, int nr_vecs)
304 struct pci_dev *pdev = nn->pdev;
308 for (i = 0; i < nr_vecs; i++)
309 nn->irq_entries[i].entry = i;
311 nvecs = pci_enable_msix_range(pdev, nn->irq_entries,
312 NFP_NET_NON_Q_VECTORS + 1, nr_vecs);
314 nn_warn(nn, "Failed to enable MSI-X. Wanted %d-%d (err=%d)\n",
315 NFP_NET_NON_Q_VECTORS + 1, nr_vecs, nvecs);
323 * nfp_net_irqs_wanted() - Work out how many interrupt vectors we want
324 * @nn: NFP Network structure
326 * We want a vector per CPU (or ring), whatever is smaller plus
327 * NFP_NET_NON_Q_VECTORS for LSC etc.
329 * Return: Number of interrupts wanted
331 static int nfp_net_irqs_wanted(struct nfp_net *nn)
336 ncpus = num_online_cpus();
338 vecs = max_t(int, nn->num_tx_rings, nn->num_rx_rings);
339 vecs = min_t(int, vecs, ncpus);
341 return vecs + NFP_NET_NON_Q_VECTORS;
345 * nfp_net_irqs_alloc() - allocates MSI-X irqs
346 * @nn: NFP Network structure
348 * Return: Number of irqs obtained or 0 on error.
350 int nfp_net_irqs_alloc(struct nfp_net *nn)
354 wanted_irqs = nfp_net_irqs_wanted(nn);
356 nn->num_irqs = nfp_net_msix_alloc(nn, wanted_irqs);
357 if (nn->num_irqs == 0) {
358 nn_err(nn, "Failed to allocate MSI-X IRQs\n");
362 nn->num_r_vecs = nn->num_irqs - NFP_NET_NON_Q_VECTORS;
364 if (nn->num_irqs < wanted_irqs)
365 nn_warn(nn, "Unable to allocate %d vectors. Got %d instead\n",
366 wanted_irqs, nn->num_irqs);
372 * nfp_net_irqs_disable() - Disable interrupts
373 * @nn: NFP Network structure
375 * Undoes what @nfp_net_irqs_alloc() does.
377 void nfp_net_irqs_disable(struct nfp_net *nn)
379 pci_disable_msix(nn->pdev);
383 * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings.
385 * @data: Opaque data structure
387 * Return: Indicate if the interrupt has been handled.
389 static irqreturn_t nfp_net_irq_rxtx(int irq, void *data)
391 struct nfp_net_r_vector *r_vec = data;
393 napi_schedule_irqoff(&r_vec->napi);
395 /* The FW auto-masks any interrupt, either via the MASK bit in
396 * the MSI-X table or via the per entry ICR field. So there
397 * is no need to disable interrupts here.
403 * nfp_net_read_link_status() - Reread link status from control BAR
404 * @nn: NFP Network structure
406 static void nfp_net_read_link_status(struct nfp_net *nn)
412 spin_lock_irqsave(&nn->link_status_lock, flags);
414 sts = nn_readl(nn, NFP_NET_CFG_STS);
415 link_up = !!(sts & NFP_NET_CFG_STS_LINK);
417 if (nn->link_up == link_up)
420 nn->link_up = link_up;
423 netif_carrier_on(nn->netdev);
424 netdev_info(nn->netdev, "NIC Link is Up\n");
426 netif_carrier_off(nn->netdev);
427 netdev_info(nn->netdev, "NIC Link is Down\n");
430 spin_unlock_irqrestore(&nn->link_status_lock, flags);
434 * nfp_net_irq_lsc() - Interrupt service routine for link state changes
436 * @data: Opaque data structure
438 * Return: Indicate if the interrupt has been handled.
440 static irqreturn_t nfp_net_irq_lsc(int irq, void *data)
442 struct nfp_net *nn = data;
444 nfp_net_read_link_status(nn);
446 nfp_net_irq_unmask(nn, NFP_NET_IRQ_LSC_IDX);
452 * nfp_net_irq_exn() - Interrupt service routine for exceptions
454 * @data: Opaque data structure
456 * Return: Indicate if the interrupt has been handled.
458 static irqreturn_t nfp_net_irq_exn(int irq, void *data)
460 struct nfp_net *nn = data;
462 nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__);
463 /* XXX TO BE IMPLEMENTED */
468 * nfp_net_tx_ring_init() - Fill in the boilerplate for a TX ring
469 * @tx_ring: TX ring structure
470 * @r_vec: IRQ vector servicing this ring
474 nfp_net_tx_ring_init(struct nfp_net_tx_ring *tx_ring,
475 struct nfp_net_r_vector *r_vec, unsigned int idx)
477 struct nfp_net *nn = r_vec->nfp_net;
480 tx_ring->r_vec = r_vec;
482 tx_ring->qcidx = tx_ring->idx * nn->stride_tx;
483 tx_ring->qcp_q = nn->tx_bar + NFP_QCP_QUEUE_OFF(tx_ring->qcidx);
487 * nfp_net_rx_ring_init() - Fill in the boilerplate for a RX ring
488 * @rx_ring: RX ring structure
489 * @r_vec: IRQ vector servicing this ring
493 nfp_net_rx_ring_init(struct nfp_net_rx_ring *rx_ring,
494 struct nfp_net_r_vector *r_vec, unsigned int idx)
496 struct nfp_net *nn = r_vec->nfp_net;
499 rx_ring->r_vec = r_vec;
501 rx_ring->fl_qcidx = rx_ring->idx * nn->stride_rx;
502 rx_ring->rx_qcidx = rx_ring->fl_qcidx + (nn->stride_rx - 1);
504 rx_ring->qcp_fl = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->fl_qcidx);
505 rx_ring->qcp_rx = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->rx_qcidx);
509 * nfp_net_irqs_assign() - Assign IRQs and setup rvecs.
510 * @netdev: netdev structure
512 static void nfp_net_irqs_assign(struct net_device *netdev)
514 struct nfp_net *nn = netdev_priv(netdev);
515 struct nfp_net_r_vector *r_vec;
518 /* Assumes nn->num_tx_rings == nn->num_rx_rings */
519 if (nn->num_tx_rings > nn->num_r_vecs) {
520 nn_warn(nn, "More rings (%d) than vectors (%d).\n",
521 nn->num_tx_rings, nn->num_r_vecs);
522 nn->num_tx_rings = nn->num_r_vecs;
523 nn->num_rx_rings = nn->num_r_vecs;
526 nn->lsc_handler = nfp_net_irq_lsc;
527 nn->exn_handler = nfp_net_irq_exn;
529 for (r = 0; r < nn->num_r_vecs; r++) {
530 r_vec = &nn->r_vecs[r];
532 r_vec->handler = nfp_net_irq_rxtx;
533 r_vec->irq_idx = NFP_NET_NON_Q_VECTORS + r;
535 cpumask_set_cpu(r, &r_vec->affinity_mask);
540 * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN)
541 * @nn: NFP Network structure
542 * @ctrl_offset: Control BAR offset where IRQ configuration should be written
543 * @format: printf-style format to construct the interrupt name
544 * @name: Pointer to allocated space for interrupt name
545 * @name_sz: Size of space for interrupt name
546 * @vector_idx: Index of MSI-X vector used for this interrupt
547 * @handler: IRQ handler to register for this interrupt
550 nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset,
551 const char *format, char *name, size_t name_sz,
552 unsigned int vector_idx, irq_handler_t handler)
554 struct msix_entry *entry;
557 entry = &nn->irq_entries[vector_idx];
559 snprintf(name, name_sz, format, netdev_name(nn->netdev));
560 err = request_irq(entry->vector, handler, 0, name, nn);
562 nn_err(nn, "Failed to request IRQ %d (err=%d).\n",
566 nn_writeb(nn, ctrl_offset, vector_idx);
572 * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN)
573 * @nn: NFP Network structure
574 * @ctrl_offset: Control BAR offset where IRQ configuration should be written
575 * @vector_idx: Index of MSI-X vector used for this interrupt
577 static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset,
578 unsigned int vector_idx)
580 nn_writeb(nn, ctrl_offset, 0xff);
581 free_irq(nn->irq_entries[vector_idx].vector, nn);
586 * One queue controller peripheral queue is used for transmit. The
587 * driver en-queues packets for transmit by advancing the write
588 * pointer. The device indicates that packets have transmitted by
589 * advancing the read pointer. The driver maintains a local copy of
590 * the read and write pointer in @struct nfp_net_tx_ring. The driver
591 * keeps @wr_p in sync with the queue controller write pointer and can
592 * determine how many packets have been transmitted by comparing its
593 * copy of the read pointer @rd_p with the read pointer maintained by
594 * the queue controller peripheral.
598 * nfp_net_tx_full() - Check if the TX ring is full
599 * @tx_ring: TX ring to check
600 * @dcnt: Number of descriptors that need to be enqueued (must be >= 1)
602 * This function checks, based on the *host copy* of read/write
603 * pointer if a given TX ring is full. The real TX queue may have
604 * some newly made available slots.
606 * Return: True if the ring is full.
608 static inline int nfp_net_tx_full(struct nfp_net_tx_ring *tx_ring, int dcnt)
610 return (tx_ring->wr_p - tx_ring->rd_p) >= (tx_ring->cnt - dcnt);
613 /* Wrappers for deciding when to stop and restart TX queues */
614 static int nfp_net_tx_ring_should_wake(struct nfp_net_tx_ring *tx_ring)
616 return !nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS * 4);
619 static int nfp_net_tx_ring_should_stop(struct nfp_net_tx_ring *tx_ring)
621 return nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS + 1);
625 * nfp_net_tx_ring_stop() - stop tx ring
626 * @nd_q: netdev queue
627 * @tx_ring: driver tx queue structure
629 * Safely stop TX ring. Remember that while we are running .start_xmit()
630 * someone else may be cleaning the TX ring completions so we need to be
631 * extra careful here.
633 static void nfp_net_tx_ring_stop(struct netdev_queue *nd_q,
634 struct nfp_net_tx_ring *tx_ring)
636 netif_tx_stop_queue(nd_q);
638 /* We can race with the TX completion out of NAPI so recheck */
640 if (unlikely(nfp_net_tx_ring_should_wake(tx_ring)))
641 netif_tx_start_queue(nd_q);
645 * nfp_net_tx_tso() - Set up Tx descriptor for LSO
646 * @nn: NFP Net device
647 * @r_vec: per-ring structure
648 * @txbuf: Pointer to driver soft TX descriptor
649 * @txd: Pointer to HW TX descriptor
650 * @skb: Pointer to SKB
652 * Set up Tx descriptor for LSO, do nothing for non-LSO skbs.
653 * Return error on packet header greater than maximum supported LSO header size.
655 static void nfp_net_tx_tso(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
656 struct nfp_net_tx_buf *txbuf,
657 struct nfp_net_tx_desc *txd, struct sk_buff *skb)
662 if (!skb_is_gso(skb))
665 if (!skb->encapsulation)
666 hdrlen = skb_transport_offset(skb) + tcp_hdrlen(skb);
668 hdrlen = skb_inner_transport_header(skb) - skb->data +
669 inner_tcp_hdrlen(skb);
671 txbuf->pkt_cnt = skb_shinfo(skb)->gso_segs;
672 txbuf->real_len += hdrlen * (txbuf->pkt_cnt - 1);
674 mss = skb_shinfo(skb)->gso_size & PCIE_DESC_TX_MSS_MASK;
675 txd->l4_offset = hdrlen;
676 txd->mss = cpu_to_le16(mss);
677 txd->flags |= PCIE_DESC_TX_LSO;
679 u64_stats_update_begin(&r_vec->tx_sync);
681 u64_stats_update_end(&r_vec->tx_sync);
685 * nfp_net_tx_csum() - Set TX CSUM offload flags in TX descriptor
686 * @nn: NFP Net device
687 * @r_vec: per-ring structure
688 * @txbuf: Pointer to driver soft TX descriptor
689 * @txd: Pointer to TX descriptor
690 * @skb: Pointer to SKB
692 * This function sets the TX checksum flags in the TX descriptor based
693 * on the configuration and the protocol of the packet to be transmitted.
695 static void nfp_net_tx_csum(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
696 struct nfp_net_tx_buf *txbuf,
697 struct nfp_net_tx_desc *txd, struct sk_buff *skb)
699 struct ipv6hdr *ipv6h;
703 if (!(nn->ctrl & NFP_NET_CFG_CTRL_TXCSUM))
706 if (skb->ip_summed != CHECKSUM_PARTIAL)
709 txd->flags |= PCIE_DESC_TX_CSUM;
710 if (skb->encapsulation)
711 txd->flags |= PCIE_DESC_TX_ENCAP;
713 iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
714 ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
716 if (iph->version == 4) {
717 txd->flags |= PCIE_DESC_TX_IP4_CSUM;
718 l4_hdr = iph->protocol;
719 } else if (ipv6h->version == 6) {
720 l4_hdr = ipv6h->nexthdr;
722 nn_warn_ratelimit(nn, "partial checksum but ipv=%x!\n",
729 txd->flags |= PCIE_DESC_TX_TCP_CSUM;
732 txd->flags |= PCIE_DESC_TX_UDP_CSUM;
735 nn_warn_ratelimit(nn, "partial checksum but l4 proto=%x!\n",
740 u64_stats_update_begin(&r_vec->tx_sync);
741 if (skb->encapsulation)
742 r_vec->hw_csum_tx_inner += txbuf->pkt_cnt;
744 r_vec->hw_csum_tx += txbuf->pkt_cnt;
745 u64_stats_update_end(&r_vec->tx_sync);
749 * nfp_net_tx() - Main transmit entry point
750 * @skb: SKB to transmit
751 * @netdev: netdev structure
753 * Return: NETDEV_TX_OK on success.
755 static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
757 struct nfp_net *nn = netdev_priv(netdev);
758 const struct skb_frag_struct *frag;
759 struct nfp_net_r_vector *r_vec;
760 struct nfp_net_tx_desc *txd, txdg;
761 struct nfp_net_tx_buf *txbuf;
762 struct nfp_net_tx_ring *tx_ring;
763 struct netdev_queue *nd_q;
770 qidx = skb_get_queue_mapping(skb);
771 tx_ring = &nn->tx_rings[qidx];
772 r_vec = tx_ring->r_vec;
773 nd_q = netdev_get_tx_queue(nn->netdev, qidx);
775 nr_frags = skb_shinfo(skb)->nr_frags;
777 if (unlikely(nfp_net_tx_full(tx_ring, nr_frags + 1))) {
778 nn_warn_ratelimit(nn, "TX ring %d busy. wrp=%u rdp=%u\n",
779 qidx, tx_ring->wr_p, tx_ring->rd_p);
780 netif_tx_stop_queue(nd_q);
781 u64_stats_update_begin(&r_vec->tx_sync);
783 u64_stats_update_end(&r_vec->tx_sync);
784 return NETDEV_TX_BUSY;
787 /* Start with the head skbuf */
788 dma_addr = dma_map_single(&nn->pdev->dev, skb->data, skb_headlen(skb),
790 if (dma_mapping_error(&nn->pdev->dev, dma_addr))
793 wr_idx = tx_ring->wr_p % tx_ring->cnt;
795 /* Stash the soft descriptor of the head then initialize it */
796 txbuf = &tx_ring->txbufs[wr_idx];
798 txbuf->dma_addr = dma_addr;
801 txbuf->real_len = skb->len;
803 /* Build TX descriptor */
804 txd = &tx_ring->txds[wr_idx];
805 txd->offset_eop = (nr_frags == 0) ? PCIE_DESC_TX_EOP : 0;
806 txd->dma_len = cpu_to_le16(skb_headlen(skb));
807 nfp_desc_set_dma_addr(txd, dma_addr);
808 txd->data_len = cpu_to_le16(skb->len);
814 nfp_net_tx_tso(nn, r_vec, txbuf, txd, skb);
816 nfp_net_tx_csum(nn, r_vec, txbuf, txd, skb);
818 if (skb_vlan_tag_present(skb) && nn->ctrl & NFP_NET_CFG_CTRL_TXVLAN) {
819 txd->flags |= PCIE_DESC_TX_VLAN;
820 txd->vlan = cpu_to_le16(skb_vlan_tag_get(skb));
825 /* all descs must match except for in addr, length and eop */
828 for (f = 0; f < nr_frags; f++) {
829 frag = &skb_shinfo(skb)->frags[f];
830 fsize = skb_frag_size(frag);
832 dma_addr = skb_frag_dma_map(&nn->pdev->dev, frag, 0,
833 fsize, DMA_TO_DEVICE);
834 if (dma_mapping_error(&nn->pdev->dev, dma_addr))
837 wr_idx = (wr_idx + 1) % tx_ring->cnt;
838 tx_ring->txbufs[wr_idx].skb = skb;
839 tx_ring->txbufs[wr_idx].dma_addr = dma_addr;
840 tx_ring->txbufs[wr_idx].fidx = f;
842 txd = &tx_ring->txds[wr_idx];
844 txd->dma_len = cpu_to_le16(fsize);
845 nfp_desc_set_dma_addr(txd, dma_addr);
847 (f == nr_frags - 1) ? PCIE_DESC_TX_EOP : 0;
850 u64_stats_update_begin(&r_vec->tx_sync);
852 u64_stats_update_end(&r_vec->tx_sync);
855 netdev_tx_sent_queue(nd_q, txbuf->real_len);
857 skb_tx_timestamp(skb);
859 tx_ring->wr_p += nr_frags + 1;
860 if (nfp_net_tx_ring_should_stop(tx_ring))
861 nfp_net_tx_ring_stop(nd_q, tx_ring);
863 tx_ring->wr_ptr_add += nr_frags + 1;
864 if (!skb->xmit_more || netif_xmit_stopped(nd_q)) {
865 /* force memory write before we let HW know */
867 nfp_qcp_wr_ptr_add(tx_ring->qcp_q, tx_ring->wr_ptr_add);
868 tx_ring->wr_ptr_add = 0;
875 frag = &skb_shinfo(skb)->frags[f];
876 dma_unmap_page(&nn->pdev->dev,
877 tx_ring->txbufs[wr_idx].dma_addr,
878 skb_frag_size(frag), DMA_TO_DEVICE);
879 tx_ring->txbufs[wr_idx].skb = NULL;
880 tx_ring->txbufs[wr_idx].dma_addr = 0;
881 tx_ring->txbufs[wr_idx].fidx = -2;
884 wr_idx += tx_ring->cnt;
886 dma_unmap_single(&nn->pdev->dev, tx_ring->txbufs[wr_idx].dma_addr,
887 skb_headlen(skb), DMA_TO_DEVICE);
888 tx_ring->txbufs[wr_idx].skb = NULL;
889 tx_ring->txbufs[wr_idx].dma_addr = 0;
890 tx_ring->txbufs[wr_idx].fidx = -2;
892 nn_warn_ratelimit(nn, "Failed to map DMA TX buffer\n");
893 u64_stats_update_begin(&r_vec->tx_sync);
895 u64_stats_update_end(&r_vec->tx_sync);
896 dev_kfree_skb_any(skb);
901 * nfp_net_tx_complete() - Handled completed TX packets
902 * @tx_ring: TX ring structure
904 * Return: Number of completed TX descriptors
906 static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring)
908 struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
909 struct nfp_net *nn = r_vec->nfp_net;
910 const struct skb_frag_struct *frag;
911 struct netdev_queue *nd_q;
912 u32 done_pkts = 0, done_bytes = 0;
919 /* Work out how many descriptors have been transmitted */
920 qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
922 if (qcp_rd_p == tx_ring->qcp_rd_p)
925 if (qcp_rd_p > tx_ring->qcp_rd_p)
926 todo = qcp_rd_p - tx_ring->qcp_rd_p;
928 todo = qcp_rd_p + tx_ring->cnt - tx_ring->qcp_rd_p;
931 idx = tx_ring->rd_p % tx_ring->cnt;
934 skb = tx_ring->txbufs[idx].skb;
938 nr_frags = skb_shinfo(skb)->nr_frags;
939 fidx = tx_ring->txbufs[idx].fidx;
943 dma_unmap_single(&nn->pdev->dev,
944 tx_ring->txbufs[idx].dma_addr,
945 skb_headlen(skb), DMA_TO_DEVICE);
947 done_pkts += tx_ring->txbufs[idx].pkt_cnt;
948 done_bytes += tx_ring->txbufs[idx].real_len;
951 frag = &skb_shinfo(skb)->frags[fidx];
952 dma_unmap_page(&nn->pdev->dev,
953 tx_ring->txbufs[idx].dma_addr,
954 skb_frag_size(frag), DMA_TO_DEVICE);
957 /* check for last gather fragment */
958 if (fidx == nr_frags - 1)
959 dev_kfree_skb_any(skb);
961 tx_ring->txbufs[idx].dma_addr = 0;
962 tx_ring->txbufs[idx].skb = NULL;
963 tx_ring->txbufs[idx].fidx = -2;
966 tx_ring->qcp_rd_p = qcp_rd_p;
968 u64_stats_update_begin(&r_vec->tx_sync);
969 r_vec->tx_bytes += done_bytes;
970 r_vec->tx_pkts += done_pkts;
971 u64_stats_update_end(&r_vec->tx_sync);
973 nd_q = netdev_get_tx_queue(nn->netdev, tx_ring->idx);
974 netdev_tx_completed_queue(nd_q, done_pkts, done_bytes);
975 if (nfp_net_tx_ring_should_wake(tx_ring)) {
976 /* Make sure TX thread will see updated tx_ring->rd_p */
979 if (unlikely(netif_tx_queue_stopped(nd_q)))
980 netif_tx_wake_queue(nd_q);
983 WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
984 "TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
985 tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
989 * nfp_net_tx_ring_reset() - Free any untransmitted buffers and reset pointers
990 * @nn: NFP Net device
991 * @tx_ring: TX ring structure
993 * Assumes that the device is stopped, must be idempotent.
996 nfp_net_tx_ring_reset(struct nfp_net *nn, struct nfp_net_tx_ring *tx_ring)
998 const struct skb_frag_struct *frag;
999 struct netdev_queue *nd_q;
1000 struct pci_dev *pdev = nn->pdev;
1002 while (tx_ring->rd_p != tx_ring->wr_p) {
1003 int nr_frags, fidx, idx;
1004 struct sk_buff *skb;
1006 idx = tx_ring->rd_p % tx_ring->cnt;
1007 skb = tx_ring->txbufs[idx].skb;
1008 nr_frags = skb_shinfo(skb)->nr_frags;
1009 fidx = tx_ring->txbufs[idx].fidx;
1013 dma_unmap_single(&pdev->dev,
1014 tx_ring->txbufs[idx].dma_addr,
1015 skb_headlen(skb), DMA_TO_DEVICE);
1017 /* unmap fragment */
1018 frag = &skb_shinfo(skb)->frags[fidx];
1019 dma_unmap_page(&pdev->dev,
1020 tx_ring->txbufs[idx].dma_addr,
1021 skb_frag_size(frag), DMA_TO_DEVICE);
1024 /* check for last gather fragment */
1025 if (fidx == nr_frags - 1)
1026 dev_kfree_skb_any(skb);
1028 tx_ring->txbufs[idx].dma_addr = 0;
1029 tx_ring->txbufs[idx].skb = NULL;
1030 tx_ring->txbufs[idx].fidx = -2;
1032 tx_ring->qcp_rd_p++;
1036 memset(tx_ring->txds, 0, sizeof(*tx_ring->txds) * tx_ring->cnt);
1039 tx_ring->qcp_rd_p = 0;
1040 tx_ring->wr_ptr_add = 0;
1042 nd_q = netdev_get_tx_queue(nn->netdev, tx_ring->idx);
1043 netdev_tx_reset_queue(nd_q);
1046 static void nfp_net_tx_timeout(struct net_device *netdev)
1048 struct nfp_net *nn = netdev_priv(netdev);
1051 for (i = 0; i < nn->num_tx_rings; i++) {
1052 if (!netif_tx_queue_stopped(netdev_get_tx_queue(netdev, i)))
1054 nn_warn(nn, "TX timeout on ring: %d\n", i);
1056 nn_warn(nn, "TX watchdog timeout\n");
1059 /* Receive processing
1063 * nfp_net_rx_space() - return the number of free slots on the RX ring
1064 * @rx_ring: RX ring structure
1066 * Make sure we leave at least one slot free.
1068 * Return: True if there is space on the RX ring
1070 static inline int nfp_net_rx_space(struct nfp_net_rx_ring *rx_ring)
1072 return (rx_ring->cnt - 1) - (rx_ring->wr_p - rx_ring->rd_p);
1076 * nfp_net_rx_alloc_one() - Allocate and map skb for RX
1077 * @rx_ring: RX ring structure of the skb
1078 * @dma_addr: Pointer to storage for DMA address (output param)
1079 * @fl_bufsz: size of freelist buffers
1081 * This function will allcate a new skb, map it for DMA.
1083 * Return: allocated skb or NULL on failure.
1085 static struct sk_buff *
1086 nfp_net_rx_alloc_one(struct nfp_net_rx_ring *rx_ring, dma_addr_t *dma_addr,
1087 unsigned int fl_bufsz)
1089 struct nfp_net *nn = rx_ring->r_vec->nfp_net;
1090 struct sk_buff *skb;
1092 skb = netdev_alloc_skb(nn->netdev, fl_bufsz);
1094 nn_warn_ratelimit(nn, "Failed to alloc receive SKB\n");
1098 *dma_addr = dma_map_single(&nn->pdev->dev, skb->data,
1099 fl_bufsz, DMA_FROM_DEVICE);
1100 if (dma_mapping_error(&nn->pdev->dev, *dma_addr)) {
1101 dev_kfree_skb_any(skb);
1102 nn_warn_ratelimit(nn, "Failed to map DMA RX buffer\n");
1110 * nfp_net_rx_give_one() - Put mapped skb on the software and hardware rings
1111 * @rx_ring: RX ring structure
1112 * @skb: Skb to put on rings
1113 * @dma_addr: DMA address of skb mapping
1115 static void nfp_net_rx_give_one(struct nfp_net_rx_ring *rx_ring,
1116 struct sk_buff *skb, dma_addr_t dma_addr)
1118 unsigned int wr_idx;
1120 wr_idx = rx_ring->wr_p % rx_ring->cnt;
1122 /* Stash SKB and DMA address away */
1123 rx_ring->rxbufs[wr_idx].skb = skb;
1124 rx_ring->rxbufs[wr_idx].dma_addr = dma_addr;
1126 /* Fill freelist descriptor */
1127 rx_ring->rxds[wr_idx].fld.reserved = 0;
1128 rx_ring->rxds[wr_idx].fld.meta_len_dd = 0;
1129 nfp_desc_set_dma_addr(&rx_ring->rxds[wr_idx].fld, dma_addr);
1132 rx_ring->wr_ptr_add++;
1133 if (rx_ring->wr_ptr_add >= NFP_NET_FL_BATCH) {
1134 /* Update write pointer of the freelist queue. Make
1135 * sure all writes are flushed before telling the hardware.
1138 nfp_qcp_wr_ptr_add(rx_ring->qcp_fl, rx_ring->wr_ptr_add);
1139 rx_ring->wr_ptr_add = 0;
1144 * nfp_net_rx_ring_reset() - Reflect in SW state of freelist after disable
1145 * @rx_ring: RX ring structure
1147 * Assumes that the device is stopped, must be idempotent.
1149 static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring)
1151 unsigned int wr_idx, last_idx;
1153 /* wr_p == rd_p means ring was never fed FL bufs. RX rings are always
1154 * kept at cnt - 1 FL bufs.
1156 if (rx_ring->wr_p == 0 && rx_ring->rd_p == 0)
1159 /* Move the empty entry to the end of the list */
1160 wr_idx = rx_ring->wr_p % rx_ring->cnt;
1161 last_idx = rx_ring->cnt - 1;
1162 rx_ring->rxbufs[wr_idx].dma_addr = rx_ring->rxbufs[last_idx].dma_addr;
1163 rx_ring->rxbufs[wr_idx].skb = rx_ring->rxbufs[last_idx].skb;
1164 rx_ring->rxbufs[last_idx].dma_addr = 0;
1165 rx_ring->rxbufs[last_idx].skb = NULL;
1167 memset(rx_ring->rxds, 0, sizeof(*rx_ring->rxds) * rx_ring->cnt);
1170 rx_ring->wr_ptr_add = 0;
1174 * nfp_net_rx_ring_bufs_free() - Free any buffers currently on the RX ring
1175 * @nn: NFP Net device
1176 * @rx_ring: RX ring to remove buffers from
1178 * Assumes that the device is stopped and buffers are in [0, ring->cnt - 1)
1179 * entries. After device is disabled nfp_net_rx_ring_reset() must be called
1180 * to restore required ring geometry.
1183 nfp_net_rx_ring_bufs_free(struct nfp_net *nn, struct nfp_net_rx_ring *rx_ring)
1185 struct pci_dev *pdev = nn->pdev;
1188 for (i = 0; i < rx_ring->cnt - 1; i++) {
1189 /* NULL skb can only happen when initial filling of the ring
1190 * fails to allocate enough buffers and calls here to free
1191 * already allocated ones.
1193 if (!rx_ring->rxbufs[i].skb)
1196 dma_unmap_single(&pdev->dev, rx_ring->rxbufs[i].dma_addr,
1197 rx_ring->bufsz, DMA_FROM_DEVICE);
1198 dev_kfree_skb_any(rx_ring->rxbufs[i].skb);
1199 rx_ring->rxbufs[i].dma_addr = 0;
1200 rx_ring->rxbufs[i].skb = NULL;
1205 * nfp_net_rx_ring_bufs_alloc() - Fill RX ring with buffers (don't give to FW)
1206 * @nn: NFP Net device
1207 * @rx_ring: RX ring to remove buffers from
1210 nfp_net_rx_ring_bufs_alloc(struct nfp_net *nn, struct nfp_net_rx_ring *rx_ring)
1212 struct nfp_net_rx_buf *rxbufs;
1215 rxbufs = rx_ring->rxbufs;
1217 for (i = 0; i < rx_ring->cnt - 1; i++) {
1219 nfp_net_rx_alloc_one(rx_ring, &rxbufs[i].dma_addr,
1221 if (!rxbufs[i].skb) {
1222 nfp_net_rx_ring_bufs_free(nn, rx_ring);
1231 * nfp_net_rx_ring_fill_freelist() - Give buffers from the ring to FW
1232 * @rx_ring: RX ring to fill
1234 static void nfp_net_rx_ring_fill_freelist(struct nfp_net_rx_ring *rx_ring)
1238 for (i = 0; i < rx_ring->cnt - 1; i++)
1239 nfp_net_rx_give_one(rx_ring, rx_ring->rxbufs[i].skb,
1240 rx_ring->rxbufs[i].dma_addr);
1244 * nfp_net_rx_csum_has_errors() - group check if rxd has any csum errors
1245 * @flags: RX descriptor flags field in CPU byte order
1247 static int nfp_net_rx_csum_has_errors(u16 flags)
1249 u16 csum_all_checked, csum_all_ok;
1251 csum_all_checked = flags & __PCIE_DESC_RX_CSUM_ALL;
1252 csum_all_ok = flags & __PCIE_DESC_RX_CSUM_ALL_OK;
1254 return csum_all_checked != (csum_all_ok << PCIE_DESC_RX_CSUM_OK_SHIFT);
1258 * nfp_net_rx_csum() - set SKB checksum field based on RX descriptor flags
1259 * @nn: NFP Net device
1260 * @r_vec: per-ring structure
1261 * @rxd: Pointer to RX descriptor
1262 * @skb: Pointer to SKB
1264 static void nfp_net_rx_csum(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
1265 struct nfp_net_rx_desc *rxd, struct sk_buff *skb)
1267 skb_checksum_none_assert(skb);
1269 if (!(nn->netdev->features & NETIF_F_RXCSUM))
1272 if (nfp_net_rx_csum_has_errors(le16_to_cpu(rxd->rxd.flags))) {
1273 u64_stats_update_begin(&r_vec->rx_sync);
1274 r_vec->hw_csum_rx_error++;
1275 u64_stats_update_end(&r_vec->rx_sync);
1279 /* Assume that the firmware will never report inner CSUM_OK unless outer
1280 * L4 headers were successfully parsed. FW will always report zero UDP
1281 * checksum as CSUM_OK.
1283 if (rxd->rxd.flags & PCIE_DESC_RX_TCP_CSUM_OK ||
1284 rxd->rxd.flags & PCIE_DESC_RX_UDP_CSUM_OK) {
1285 __skb_incr_checksum_unnecessary(skb);
1286 u64_stats_update_begin(&r_vec->rx_sync);
1287 r_vec->hw_csum_rx_ok++;
1288 u64_stats_update_end(&r_vec->rx_sync);
1291 if (rxd->rxd.flags & PCIE_DESC_RX_I_TCP_CSUM_OK ||
1292 rxd->rxd.flags & PCIE_DESC_RX_I_UDP_CSUM_OK) {
1293 __skb_incr_checksum_unnecessary(skb);
1294 u64_stats_update_begin(&r_vec->rx_sync);
1295 r_vec->hw_csum_rx_inner_ok++;
1296 u64_stats_update_end(&r_vec->rx_sync);
1300 static void nfp_net_set_hash(struct net_device *netdev, struct sk_buff *skb,
1301 unsigned int type, __be32 *hash)
1303 if (!(netdev->features & NETIF_F_RXHASH))
1307 case NFP_NET_RSS_IPV4:
1308 case NFP_NET_RSS_IPV6:
1309 case NFP_NET_RSS_IPV6_EX:
1310 skb_set_hash(skb, get_unaligned_be32(hash), PKT_HASH_TYPE_L3);
1313 skb_set_hash(skb, get_unaligned_be32(hash), PKT_HASH_TYPE_L4);
1319 nfp_net_set_hash_desc(struct net_device *netdev, struct sk_buff *skb,
1320 struct nfp_net_rx_desc *rxd)
1322 struct nfp_net_rx_hash *rx_hash;
1324 if (!(rxd->rxd.flags & PCIE_DESC_RX_RSS))
1327 rx_hash = (struct nfp_net_rx_hash *)(skb->data - sizeof(*rx_hash));
1329 nfp_net_set_hash(netdev, skb, get_unaligned_be32(&rx_hash->hash_type),
1334 nfp_net_parse_meta(struct net_device *netdev, struct sk_buff *skb,
1337 u8 *data = skb->data - meta_len;
1340 meta_info = get_unaligned_be32(data);
1344 switch (meta_info & NFP_NET_META_FIELD_MASK) {
1345 case NFP_NET_META_HASH:
1346 meta_info >>= NFP_NET_META_FIELD_SIZE;
1347 nfp_net_set_hash(netdev, skb,
1348 meta_info & NFP_NET_META_FIELD_MASK,
1352 case NFP_NET_META_MARK:
1353 skb->mark = get_unaligned_be32(data);
1360 meta_info >>= NFP_NET_META_FIELD_SIZE;
1367 * nfp_net_rx() - receive up to @budget packets on @rx_ring
1368 * @rx_ring: RX ring to receive from
1369 * @budget: NAPI budget
1371 * Note, this function is separated out from the napi poll function to
1372 * more cleanly separate packet receive code from other bookkeeping
1373 * functions performed in the napi poll function.
1375 * There are differences between the NFP-3200 firmware and the
1376 * NFP-6000 firmware. The NFP-3200 firmware uses a dedicated RX queue
1377 * to indicate that new packets have arrived. The NFP-6000 does not
1378 * have this queue and uses the DD bit in the RX descriptor. This
1379 * method cannot be used on the NFP-3200 as it causes a race
1380 * condition: The RX ring write pointer on the NFP-3200 is updated
1381 * after packets (and descriptors) have been DMAed. If the DD bit is
1382 * used and subsequently the read pointer is updated this may lead to
1383 * the RX queue to underflow (if the firmware has not yet update the
1384 * write pointer). Therefore we use slightly ugly conditional code
1385 * below to handle the differences. We may, in the future update the
1386 * NFP-3200 firmware to behave the same as the firmware on the
1389 * Return: Number of packets received.
1391 static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget)
1393 struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
1394 struct nfp_net *nn = r_vec->nfp_net;
1395 unsigned int data_len, meta_len;
1396 int avail = 0, pkts_polled = 0;
1397 struct sk_buff *skb, *new_skb;
1398 struct nfp_net_rx_desc *rxd;
1399 dma_addr_t new_dma_addr;
1403 if (nn->is_nfp3200) {
1404 /* Work out how many packets arrived */
1405 qcp_wr_p = nfp_qcp_wr_ptr_read(rx_ring->qcp_rx);
1406 idx = rx_ring->rd_p % rx_ring->cnt;
1408 if (qcp_wr_p == idx)
1409 /* No new packets */
1413 avail = qcp_wr_p - idx;
1415 avail = qcp_wr_p + rx_ring->cnt - idx;
1420 while (avail > 0 && pkts_polled < budget) {
1421 idx = rx_ring->rd_p % rx_ring->cnt;
1423 rxd = &rx_ring->rxds[idx];
1424 if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD)) {
1426 nn_dbg(nn, "RX descriptor not valid (DD)%d:%u rxd[0]=%#x rxd[1]=%#x\n",
1428 rxd->vals[0], rxd->vals[1]);
1431 /* Memory barrier to ensure that we won't do other reads
1432 * before the DD bit.
1440 skb = rx_ring->rxbufs[idx].skb;
1442 new_skb = nfp_net_rx_alloc_one(rx_ring, &new_dma_addr,
1445 nfp_net_rx_give_one(rx_ring, rx_ring->rxbufs[idx].skb,
1446 rx_ring->rxbufs[idx].dma_addr);
1447 u64_stats_update_begin(&r_vec->rx_sync);
1449 u64_stats_update_end(&r_vec->rx_sync);
1453 dma_unmap_single(&nn->pdev->dev,
1454 rx_ring->rxbufs[idx].dma_addr,
1455 nn->fl_bufsz, DMA_FROM_DEVICE);
1457 nfp_net_rx_give_one(rx_ring, new_skb, new_dma_addr);
1460 * <-- [rx_offset] -->
1461 * ---------------------------------------------------------
1462 * | [XX] | metadata | packet | XXXX |
1463 * ---------------------------------------------------------
1464 * <---------------- data_len --------------->
1466 * The rx_offset is fixed for all packets, the meta_len can vary
1467 * on a packet by packet basis. If rx_offset is set to zero
1468 * (_RX_OFFSET_DYNAMIC) metadata starts at the beginning of the
1469 * buffer and is immediately followed by the packet (no [XX]).
1471 meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
1472 data_len = le16_to_cpu(rxd->rxd.data_len);
1474 if (nn->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1475 skb_reserve(skb, meta_len);
1477 skb_reserve(skb, nn->rx_offset);
1478 skb_put(skb, data_len - meta_len);
1481 u64_stats_update_begin(&r_vec->rx_sync);
1483 r_vec->rx_bytes += skb->len;
1484 u64_stats_update_end(&r_vec->rx_sync);
1486 if (nn->fw_ver.major <= 3) {
1487 nfp_net_set_hash_desc(nn->netdev, skb, rxd);
1488 } else if (meta_len) {
1491 end = nfp_net_parse_meta(nn->netdev, skb, meta_len);
1492 if (unlikely(end != skb->data)) {
1493 u64_stats_update_begin(&r_vec->rx_sync);
1495 u64_stats_update_end(&r_vec->rx_sync);
1497 dev_kfree_skb_any(skb);
1498 nn_warn_ratelimit(nn, "invalid RX packet metadata\n");
1503 skb_record_rx_queue(skb, rx_ring->idx);
1504 skb->protocol = eth_type_trans(skb, nn->netdev);
1506 nfp_net_rx_csum(nn, r_vec, rxd, skb);
1508 if (rxd->rxd.flags & PCIE_DESC_RX_VLAN)
1509 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1510 le16_to_cpu(rxd->rxd.vlan));
1512 napi_gro_receive(&rx_ring->r_vec->napi, skb);
1516 nfp_qcp_rd_ptr_add(rx_ring->qcp_rx, pkts_polled);
1522 * nfp_net_poll() - napi poll function
1523 * @napi: NAPI structure
1524 * @budget: NAPI budget
1526 * Return: number of packets polled.
1528 static int nfp_net_poll(struct napi_struct *napi, int budget)
1530 struct nfp_net_r_vector *r_vec =
1531 container_of(napi, struct nfp_net_r_vector, napi);
1532 struct nfp_net_rx_ring *rx_ring = r_vec->rx_ring;
1533 struct nfp_net_tx_ring *tx_ring = r_vec->tx_ring;
1534 struct nfp_net *nn = r_vec->nfp_net;
1535 struct netdev_queue *txq;
1536 unsigned int pkts_polled;
1538 tx_ring = &nn->tx_rings[rx_ring->idx];
1539 txq = netdev_get_tx_queue(nn->netdev, tx_ring->idx);
1540 nfp_net_tx_complete(tx_ring);
1542 pkts_polled = nfp_net_rx(rx_ring, budget);
1544 if (pkts_polled < budget) {
1545 napi_complete_done(napi, pkts_polled);
1546 nfp_net_irq_unmask(nn, r_vec->irq_idx);
1552 /* Setup and Configuration
1556 * nfp_net_tx_ring_free() - Free resources allocated to a TX ring
1557 * @tx_ring: TX ring to free
1559 static void nfp_net_tx_ring_free(struct nfp_net_tx_ring *tx_ring)
1561 struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1562 struct nfp_net *nn = r_vec->nfp_net;
1563 struct pci_dev *pdev = nn->pdev;
1565 kfree(tx_ring->txbufs);
1568 dma_free_coherent(&pdev->dev, tx_ring->size,
1569 tx_ring->txds, tx_ring->dma);
1572 tx_ring->txbufs = NULL;
1573 tx_ring->txds = NULL;
1579 * nfp_net_tx_ring_alloc() - Allocate resource for a TX ring
1580 * @tx_ring: TX Ring structure to allocate
1581 * @cnt: Ring buffer count
1583 * Return: 0 on success, negative errno otherwise.
1585 static int nfp_net_tx_ring_alloc(struct nfp_net_tx_ring *tx_ring, u32 cnt)
1587 struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1588 struct nfp_net *nn = r_vec->nfp_net;
1589 struct pci_dev *pdev = nn->pdev;
1594 tx_ring->size = sizeof(*tx_ring->txds) * tx_ring->cnt;
1595 tx_ring->txds = dma_zalloc_coherent(&pdev->dev, tx_ring->size,
1596 &tx_ring->dma, GFP_KERNEL);
1600 sz = sizeof(*tx_ring->txbufs) * tx_ring->cnt;
1601 tx_ring->txbufs = kzalloc(sz, GFP_KERNEL);
1602 if (!tx_ring->txbufs)
1605 netif_set_xps_queue(nn->netdev, &r_vec->affinity_mask, tx_ring->idx);
1607 nn_dbg(nn, "TxQ%02d: QCidx=%02d cnt=%d dma=%#llx host=%p\n",
1608 tx_ring->idx, tx_ring->qcidx,
1609 tx_ring->cnt, (unsigned long long)tx_ring->dma, tx_ring->txds);
1614 nfp_net_tx_ring_free(tx_ring);
1618 static struct nfp_net_tx_ring *
1619 nfp_net_shadow_tx_rings_prepare(struct nfp_net *nn, u32 buf_cnt)
1621 struct nfp_net_tx_ring *rings;
1624 rings = kcalloc(nn->num_tx_rings, sizeof(*rings), GFP_KERNEL);
1628 for (r = 0; r < nn->num_tx_rings; r++) {
1629 nfp_net_tx_ring_init(&rings[r], nn->tx_rings[r].r_vec, r);
1631 if (nfp_net_tx_ring_alloc(&rings[r], buf_cnt))
1639 nfp_net_tx_ring_free(&rings[r]);
1644 static struct nfp_net_tx_ring *
1645 nfp_net_shadow_tx_rings_swap(struct nfp_net *nn, struct nfp_net_tx_ring *rings)
1647 struct nfp_net_tx_ring *old = nn->tx_rings;
1650 for (r = 0; r < nn->num_tx_rings; r++)
1651 old[r].r_vec->tx_ring = &rings[r];
1653 nn->tx_rings = rings;
1658 nfp_net_shadow_tx_rings_free(struct nfp_net *nn, struct nfp_net_tx_ring *rings)
1665 for (r = 0; r < nn->num_tx_rings; r++)
1666 nfp_net_tx_ring_free(&rings[r]);
1672 * nfp_net_rx_ring_free() - Free resources allocated to a RX ring
1673 * @rx_ring: RX ring to free
1675 static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring)
1677 struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
1678 struct nfp_net *nn = r_vec->nfp_net;
1679 struct pci_dev *pdev = nn->pdev;
1681 kfree(rx_ring->rxbufs);
1684 dma_free_coherent(&pdev->dev, rx_ring->size,
1685 rx_ring->rxds, rx_ring->dma);
1688 rx_ring->rxbufs = NULL;
1689 rx_ring->rxds = NULL;
1695 * nfp_net_rx_ring_alloc() - Allocate resource for a RX ring
1696 * @rx_ring: RX ring to allocate
1697 * @fl_bufsz: Size of buffers to allocate
1698 * @cnt: Ring buffer count
1700 * Return: 0 on success, negative errno otherwise.
1703 nfp_net_rx_ring_alloc(struct nfp_net_rx_ring *rx_ring, unsigned int fl_bufsz,
1706 struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
1707 struct nfp_net *nn = r_vec->nfp_net;
1708 struct pci_dev *pdev = nn->pdev;
1712 rx_ring->bufsz = fl_bufsz;
1714 rx_ring->size = sizeof(*rx_ring->rxds) * rx_ring->cnt;
1715 rx_ring->rxds = dma_zalloc_coherent(&pdev->dev, rx_ring->size,
1716 &rx_ring->dma, GFP_KERNEL);
1720 sz = sizeof(*rx_ring->rxbufs) * rx_ring->cnt;
1721 rx_ring->rxbufs = kzalloc(sz, GFP_KERNEL);
1722 if (!rx_ring->rxbufs)
1725 nn_dbg(nn, "RxQ%02d: FlQCidx=%02d RxQCidx=%02d cnt=%d dma=%#llx host=%p\n",
1726 rx_ring->idx, rx_ring->fl_qcidx, rx_ring->rx_qcidx,
1727 rx_ring->cnt, (unsigned long long)rx_ring->dma, rx_ring->rxds);
1732 nfp_net_rx_ring_free(rx_ring);
1736 static struct nfp_net_rx_ring *
1737 nfp_net_shadow_rx_rings_prepare(struct nfp_net *nn, unsigned int fl_bufsz,
1740 struct nfp_net_rx_ring *rings;
1743 rings = kcalloc(nn->num_rx_rings, sizeof(*rings), GFP_KERNEL);
1747 for (r = 0; r < nn->num_rx_rings; r++) {
1748 nfp_net_rx_ring_init(&rings[r], nn->rx_rings[r].r_vec, r);
1750 if (nfp_net_rx_ring_alloc(&rings[r], fl_bufsz, buf_cnt))
1753 if (nfp_net_rx_ring_bufs_alloc(nn, &rings[r]))
1761 nfp_net_rx_ring_bufs_free(nn, &rings[r]);
1763 nfp_net_rx_ring_free(&rings[r]);
1769 static struct nfp_net_rx_ring *
1770 nfp_net_shadow_rx_rings_swap(struct nfp_net *nn, struct nfp_net_rx_ring *rings)
1772 struct nfp_net_rx_ring *old = nn->rx_rings;
1775 for (r = 0; r < nn->num_rx_rings; r++)
1776 old[r].r_vec->rx_ring = &rings[r];
1778 nn->rx_rings = rings;
1783 nfp_net_shadow_rx_rings_free(struct nfp_net *nn, struct nfp_net_rx_ring *rings)
1790 for (r = 0; r < nn->num_r_vecs; r++) {
1791 nfp_net_rx_ring_bufs_free(nn, &rings[r]);
1792 nfp_net_rx_ring_free(&rings[r]);
1799 nfp_net_prepare_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
1802 struct msix_entry *entry = &nn->irq_entries[r_vec->irq_idx];
1805 r_vec->tx_ring = &nn->tx_rings[idx];
1806 nfp_net_tx_ring_init(r_vec->tx_ring, r_vec, idx);
1808 r_vec->rx_ring = &nn->rx_rings[idx];
1809 nfp_net_rx_ring_init(r_vec->rx_ring, r_vec, idx);
1811 snprintf(r_vec->name, sizeof(r_vec->name),
1812 "%s-rxtx-%d", nn->netdev->name, idx);
1813 err = request_irq(entry->vector, r_vec->handler, 0, r_vec->name, r_vec);
1815 nn_err(nn, "Error requesting IRQ %d\n", entry->vector);
1818 disable_irq(entry->vector);
1821 netif_napi_add(nn->netdev, &r_vec->napi,
1822 nfp_net_poll, NAPI_POLL_WEIGHT);
1824 irq_set_affinity_hint(entry->vector, &r_vec->affinity_mask);
1826 nn_dbg(nn, "RV%02d: irq=%03d/%03d\n", idx, entry->vector, entry->entry);
1832 nfp_net_cleanup_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec)
1834 struct msix_entry *entry = &nn->irq_entries[r_vec->irq_idx];
1836 irq_set_affinity_hint(entry->vector, NULL);
1837 netif_napi_del(&r_vec->napi);
1838 free_irq(entry->vector, r_vec);
1842 * nfp_net_rss_write_itbl() - Write RSS indirection table to device
1843 * @nn: NFP Net device to reconfigure
1845 void nfp_net_rss_write_itbl(struct nfp_net *nn)
1849 for (i = 0; i < NFP_NET_CFG_RSS_ITBL_SZ; i += 4)
1850 nn_writel(nn, NFP_NET_CFG_RSS_ITBL + i,
1851 get_unaligned_le32(nn->rss_itbl + i));
1855 * nfp_net_rss_write_key() - Write RSS hash key to device
1856 * @nn: NFP Net device to reconfigure
1858 void nfp_net_rss_write_key(struct nfp_net *nn)
1862 for (i = 0; i < NFP_NET_CFG_RSS_KEY_SZ; i += 4)
1863 nn_writel(nn, NFP_NET_CFG_RSS_KEY + i,
1864 get_unaligned_le32(nn->rss_key + i));
1868 * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW
1869 * @nn: NFP Net device to reconfigure
1871 void nfp_net_coalesce_write_cfg(struct nfp_net *nn)
1877 /* Compute factor used to convert coalesce '_usecs' parameters to
1878 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp
1881 factor = nn->me_freq_mhz / 16;
1883 /* copy RX interrupt coalesce parameters */
1884 value = (nn->rx_coalesce_max_frames << 16) |
1885 (factor * nn->rx_coalesce_usecs);
1886 for (i = 0; i < nn->num_r_vecs; i++)
1887 nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(i), value);
1889 /* copy TX interrupt coalesce parameters */
1890 value = (nn->tx_coalesce_max_frames << 16) |
1891 (factor * nn->tx_coalesce_usecs);
1892 for (i = 0; i < nn->num_r_vecs; i++)
1893 nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(i), value);
1897 * nfp_net_write_mac_addr() - Write mac address to the device control BAR
1898 * @nn: NFP Net device to reconfigure
1900 * Writes the MAC address from the netdev to the device control BAR. Does not
1901 * perform the required reconfig. We do a bit of byte swapping dance because
1904 static void nfp_net_write_mac_addr(struct nfp_net *nn)
1906 nn_writel(nn, NFP_NET_CFG_MACADDR + 0,
1907 get_unaligned_be32(nn->netdev->dev_addr));
1908 /* We can't do writew for NFP-3200 compatibility */
1909 nn_writel(nn, NFP_NET_CFG_MACADDR + 4,
1910 get_unaligned_be16(nn->netdev->dev_addr + 4) << 16);
1913 static void nfp_net_vec_clear_ring_data(struct nfp_net *nn, unsigned int idx)
1915 nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), 0);
1916 nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), 0);
1917 nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), 0);
1919 nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), 0);
1920 nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), 0);
1921 nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), 0);
1925 * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP
1926 * @nn: NFP Net device to reconfigure
1928 * Warning: must be fully idempotent.
1930 static void nfp_net_clear_config_and_disable(struct nfp_net *nn)
1932 u32 new_ctrl, update;
1936 new_ctrl = nn->ctrl;
1937 new_ctrl &= ~NFP_NET_CFG_CTRL_ENABLE;
1938 update = NFP_NET_CFG_UPDATE_GEN;
1939 update |= NFP_NET_CFG_UPDATE_MSIX;
1940 update |= NFP_NET_CFG_UPDATE_RING;
1942 if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
1943 new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG;
1945 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
1946 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
1948 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
1949 err = nfp_net_reconfig(nn, update);
1951 nn_err(nn, "Could not disable device: %d\n", err);
1953 for (r = 0; r < nn->num_r_vecs; r++) {
1954 nfp_net_rx_ring_reset(nn->r_vecs[r].rx_ring);
1955 nfp_net_tx_ring_reset(nn, nn->r_vecs[r].tx_ring);
1956 nfp_net_vec_clear_ring_data(nn, r);
1959 nn->ctrl = new_ctrl;
1963 nfp_net_vec_write_ring_data(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
1966 /* Write the DMA address, size and MSI-X info to the device */
1967 nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), r_vec->rx_ring->dma);
1968 nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), ilog2(r_vec->rx_ring->cnt));
1969 nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), r_vec->irq_idx);
1971 nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), r_vec->tx_ring->dma);
1972 nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), ilog2(r_vec->tx_ring->cnt));
1973 nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), r_vec->irq_idx);
1976 static int __nfp_net_set_config_and_enable(struct nfp_net *nn)
1978 u32 new_ctrl, update = 0;
1982 new_ctrl = nn->ctrl;
1984 if (nn->cap & NFP_NET_CFG_CTRL_RSS) {
1985 nfp_net_rss_write_key(nn);
1986 nfp_net_rss_write_itbl(nn);
1987 nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg);
1988 update |= NFP_NET_CFG_UPDATE_RSS;
1991 if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) {
1992 nfp_net_coalesce_write_cfg(nn);
1994 new_ctrl |= NFP_NET_CFG_CTRL_IRQMOD;
1995 update |= NFP_NET_CFG_UPDATE_IRQMOD;
1998 for (r = 0; r < nn->num_r_vecs; r++)
1999 nfp_net_vec_write_ring_data(nn, &nn->r_vecs[r], r);
2001 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, nn->num_tx_rings == 64 ?
2002 0xffffffffffffffffULL : ((u64)1 << nn->num_tx_rings) - 1);
2004 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, nn->num_rx_rings == 64 ?
2005 0xffffffffffffffffULL : ((u64)1 << nn->num_rx_rings) - 1);
2007 nfp_net_write_mac_addr(nn);
2009 nn_writel(nn, NFP_NET_CFG_MTU, nn->netdev->mtu);
2010 nn_writel(nn, NFP_NET_CFG_FLBUFSZ, nn->fl_bufsz);
2013 new_ctrl |= NFP_NET_CFG_CTRL_ENABLE;
2014 update |= NFP_NET_CFG_UPDATE_GEN;
2015 update |= NFP_NET_CFG_UPDATE_MSIX;
2016 update |= NFP_NET_CFG_UPDATE_RING;
2017 if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2018 new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG;
2020 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2021 err = nfp_net_reconfig(nn, update);
2023 nn->ctrl = new_ctrl;
2025 for (r = 0; r < nn->num_r_vecs; r++)
2026 nfp_net_rx_ring_fill_freelist(nn->r_vecs[r].rx_ring);
2028 /* Since reconfiguration requests while NFP is down are ignored we
2029 * have to wipe the entire VXLAN configuration and reinitialize it.
2031 if (nn->ctrl & NFP_NET_CFG_CTRL_VXLAN) {
2032 memset(&nn->vxlan_ports, 0, sizeof(nn->vxlan_ports));
2033 memset(&nn->vxlan_usecnt, 0, sizeof(nn->vxlan_usecnt));
2034 udp_tunnel_get_rx_info(nn->netdev);
2041 * nfp_net_set_config_and_enable() - Write control BAR and enable NFP
2042 * @nn: NFP Net device to reconfigure
2044 static int nfp_net_set_config_and_enable(struct nfp_net *nn)
2048 err = __nfp_net_set_config_and_enable(nn);
2050 nfp_net_clear_config_and_disable(nn);
2056 * nfp_net_open_stack() - Start the device from stack's perspective
2057 * @nn: NFP Net device to reconfigure
2059 static void nfp_net_open_stack(struct nfp_net *nn)
2063 for (r = 0; r < nn->num_r_vecs; r++) {
2064 napi_enable(&nn->r_vecs[r].napi);
2065 enable_irq(nn->irq_entries[nn->r_vecs[r].irq_idx].vector);
2068 netif_tx_wake_all_queues(nn->netdev);
2070 enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2071 nfp_net_read_link_status(nn);
2074 static int nfp_net_netdev_open(struct net_device *netdev)
2076 struct nfp_net *nn = netdev_priv(netdev);
2079 if (nn->ctrl & NFP_NET_CFG_CTRL_ENABLE) {
2080 nn_err(nn, "Dev is already enabled: 0x%08x\n", nn->ctrl);
2084 /* Step 1: Allocate resources for rings and the like
2085 * - Request interrupts
2086 * - Allocate RX and TX ring resources
2087 * - Setup initial RSS table
2089 err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn",
2090 nn->exn_name, sizeof(nn->exn_name),
2091 NFP_NET_IRQ_EXN_IDX, nn->exn_handler);
2094 err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc",
2095 nn->lsc_name, sizeof(nn->lsc_name),
2096 NFP_NET_IRQ_LSC_IDX, nn->lsc_handler);
2099 disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2101 nn->rx_rings = kcalloc(nn->num_rx_rings, sizeof(*nn->rx_rings),
2103 if (!nn->rx_rings) {
2107 nn->tx_rings = kcalloc(nn->num_tx_rings, sizeof(*nn->tx_rings),
2109 if (!nn->tx_rings) {
2111 goto err_free_rx_rings;
2114 for (r = 0; r < nn->num_r_vecs; r++) {
2115 err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2117 goto err_free_prev_vecs;
2119 err = nfp_net_tx_ring_alloc(nn->r_vecs[r].tx_ring, nn->txd_cnt);
2121 goto err_cleanup_vec_p;
2123 err = nfp_net_rx_ring_alloc(nn->r_vecs[r].rx_ring,
2124 nn->fl_bufsz, nn->rxd_cnt);
2126 goto err_free_tx_ring_p;
2128 err = nfp_net_rx_ring_bufs_alloc(nn, nn->r_vecs[r].rx_ring);
2130 goto err_flush_rx_ring_p;
2133 err = netif_set_real_num_tx_queues(netdev, nn->num_tx_rings);
2135 goto err_free_rings;
2137 err = netif_set_real_num_rx_queues(netdev, nn->num_rx_rings);
2139 goto err_free_rings;
2141 /* Step 2: Configure the NFP
2142 * - Enable rings from 0 to tx_rings/rx_rings - 1.
2143 * - Write MAC address (in case it changed)
2145 * - Set the Freelist buffer size
2148 err = nfp_net_set_config_and_enable(nn);
2150 goto err_free_rings;
2152 /* Step 3: Enable for kernel
2153 * - put some freelist descriptors on each RX ring
2154 * - enable NAPI on each ring
2155 * - enable all TX queues
2158 nfp_net_open_stack(nn);
2166 nfp_net_rx_ring_bufs_free(nn, nn->r_vecs[r].rx_ring);
2167 err_flush_rx_ring_p:
2168 nfp_net_rx_ring_free(nn->r_vecs[r].rx_ring);
2170 nfp_net_tx_ring_free(nn->r_vecs[r].tx_ring);
2172 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2174 kfree(nn->tx_rings);
2176 kfree(nn->rx_rings);
2178 nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2180 nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2185 * nfp_net_close_stack() - Quiescent the stack (part of close)
2186 * @nn: NFP Net device to reconfigure
2188 static void nfp_net_close_stack(struct nfp_net *nn)
2192 disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2193 netif_carrier_off(nn->netdev);
2194 nn->link_up = false;
2196 for (r = 0; r < nn->num_r_vecs; r++) {
2197 disable_irq(nn->irq_entries[nn->r_vecs[r].irq_idx].vector);
2198 napi_disable(&nn->r_vecs[r].napi);
2201 netif_tx_disable(nn->netdev);
2205 * nfp_net_close_free_all() - Free all runtime resources
2206 * @nn: NFP Net device to reconfigure
2208 static void nfp_net_close_free_all(struct nfp_net *nn)
2212 for (r = 0; r < nn->num_r_vecs; r++) {
2213 nfp_net_rx_ring_bufs_free(nn, nn->r_vecs[r].rx_ring);
2214 nfp_net_rx_ring_free(nn->r_vecs[r].rx_ring);
2215 nfp_net_tx_ring_free(nn->r_vecs[r].tx_ring);
2216 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2219 kfree(nn->rx_rings);
2220 kfree(nn->tx_rings);
2222 nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2223 nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2227 * nfp_net_netdev_close() - Called when the device is downed
2228 * @netdev: netdev structure
2230 static int nfp_net_netdev_close(struct net_device *netdev)
2232 struct nfp_net *nn = netdev_priv(netdev);
2234 if (!(nn->ctrl & NFP_NET_CFG_CTRL_ENABLE)) {
2235 nn_err(nn, "Dev is not up: 0x%08x\n", nn->ctrl);
2239 /* Step 1: Disable RX and TX rings from the Linux kernel perspective
2241 nfp_net_close_stack(nn);
2245 nfp_net_clear_config_and_disable(nn);
2247 /* Step 3: Free resources
2249 nfp_net_close_free_all(nn);
2251 nn_dbg(nn, "%s down", netdev->name);
2255 static void nfp_net_set_rx_mode(struct net_device *netdev)
2257 struct nfp_net *nn = netdev_priv(netdev);
2260 new_ctrl = nn->ctrl;
2262 if (netdev->flags & IFF_PROMISC) {
2263 if (nn->cap & NFP_NET_CFG_CTRL_PROMISC)
2264 new_ctrl |= NFP_NET_CFG_CTRL_PROMISC;
2266 nn_warn(nn, "FW does not support promiscuous mode\n");
2268 new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC;
2271 if (new_ctrl == nn->ctrl)
2274 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2275 nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN);
2277 nn->ctrl = new_ctrl;
2280 static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu)
2282 unsigned int old_mtu, old_fl_bufsz, new_fl_bufsz;
2283 struct nfp_net *nn = netdev_priv(netdev);
2284 struct nfp_net_rx_ring *tmp_rings;
2287 if (new_mtu < 68 || new_mtu > nn->max_mtu) {
2288 nn_err(nn, "New MTU (%d) is not valid\n", new_mtu);
2292 old_mtu = netdev->mtu;
2293 old_fl_bufsz = nn->fl_bufsz;
2294 new_fl_bufsz = NFP_NET_MAX_PREPEND + ETH_HLEN + VLAN_HLEN * 2 + new_mtu;
2296 if (!netif_running(netdev)) {
2297 netdev->mtu = new_mtu;
2298 nn->fl_bufsz = new_fl_bufsz;
2302 /* Prepare new rings */
2303 tmp_rings = nfp_net_shadow_rx_rings_prepare(nn, new_fl_bufsz,
2308 /* Stop device, swap in new rings, try to start the firmware */
2309 nfp_net_close_stack(nn);
2310 nfp_net_clear_config_and_disable(nn);
2312 tmp_rings = nfp_net_shadow_rx_rings_swap(nn, tmp_rings);
2314 netdev->mtu = new_mtu;
2315 nn->fl_bufsz = new_fl_bufsz;
2317 err = nfp_net_set_config_and_enable(nn);
2319 const int err_new = err;
2321 /* Try with old configuration and old rings */
2322 tmp_rings = nfp_net_shadow_rx_rings_swap(nn, tmp_rings);
2324 netdev->mtu = old_mtu;
2325 nn->fl_bufsz = old_fl_bufsz;
2327 err = __nfp_net_set_config_and_enable(nn);
2329 nn_err(nn, "Can't restore MTU - FW communication failed (%d,%d)\n",
2333 nfp_net_shadow_rx_rings_free(nn, tmp_rings);
2335 nfp_net_open_stack(nn);
2340 int nfp_net_set_ring_size(struct nfp_net *nn, u32 rxd_cnt, u32 txd_cnt)
2342 struct nfp_net_tx_ring *tx_rings = NULL;
2343 struct nfp_net_rx_ring *rx_rings = NULL;
2344 u32 old_rxd_cnt, old_txd_cnt;
2347 if (!netif_running(nn->netdev)) {
2348 nn->rxd_cnt = rxd_cnt;
2349 nn->txd_cnt = txd_cnt;
2353 old_rxd_cnt = nn->rxd_cnt;
2354 old_txd_cnt = nn->txd_cnt;
2356 /* Prepare new rings */
2357 if (nn->rxd_cnt != rxd_cnt) {
2358 rx_rings = nfp_net_shadow_rx_rings_prepare(nn, nn->fl_bufsz,
2363 if (nn->txd_cnt != txd_cnt) {
2364 tx_rings = nfp_net_shadow_tx_rings_prepare(nn, txd_cnt);
2366 nfp_net_shadow_rx_rings_free(nn, rx_rings);
2371 /* Stop device, swap in new rings, try to start the firmware */
2372 nfp_net_close_stack(nn);
2373 nfp_net_clear_config_and_disable(nn);
2376 rx_rings = nfp_net_shadow_rx_rings_swap(nn, rx_rings);
2378 tx_rings = nfp_net_shadow_tx_rings_swap(nn, tx_rings);
2380 nn->rxd_cnt = rxd_cnt;
2381 nn->txd_cnt = txd_cnt;
2383 err = nfp_net_set_config_and_enable(nn);
2385 const int err_new = err;
2387 /* Try with old configuration and old rings */
2389 rx_rings = nfp_net_shadow_rx_rings_swap(nn, rx_rings);
2391 tx_rings = nfp_net_shadow_tx_rings_swap(nn, tx_rings);
2393 nn->rxd_cnt = old_rxd_cnt;
2394 nn->txd_cnt = old_txd_cnt;
2396 err = __nfp_net_set_config_and_enable(nn);
2398 nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n",
2402 nfp_net_shadow_rx_rings_free(nn, rx_rings);
2403 nfp_net_shadow_tx_rings_free(nn, tx_rings);
2405 nfp_net_open_stack(nn);
2410 static struct rtnl_link_stats64 *nfp_net_stat64(struct net_device *netdev,
2411 struct rtnl_link_stats64 *stats)
2413 struct nfp_net *nn = netdev_priv(netdev);
2416 for (r = 0; r < nn->num_r_vecs; r++) {
2417 struct nfp_net_r_vector *r_vec = &nn->r_vecs[r];
2422 start = u64_stats_fetch_begin(&r_vec->rx_sync);
2423 data[0] = r_vec->rx_pkts;
2424 data[1] = r_vec->rx_bytes;
2425 data[2] = r_vec->rx_drops;
2426 } while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
2427 stats->rx_packets += data[0];
2428 stats->rx_bytes += data[1];
2429 stats->rx_dropped += data[2];
2432 start = u64_stats_fetch_begin(&r_vec->tx_sync);
2433 data[0] = r_vec->tx_pkts;
2434 data[1] = r_vec->tx_bytes;
2435 data[2] = r_vec->tx_errors;
2436 } while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
2437 stats->tx_packets += data[0];
2438 stats->tx_bytes += data[1];
2439 stats->tx_errors += data[2];
2445 static bool nfp_net_ebpf_capable(struct nfp_net *nn)
2447 if (nn->cap & NFP_NET_CFG_CTRL_BPF &&
2448 nn_readb(nn, NFP_NET_CFG_BPF_ABI) == NFP_NET_BPF_ABI)
2454 nfp_net_setup_tc(struct net_device *netdev, u32 handle, __be16 proto,
2455 struct tc_to_netdev *tc)
2457 struct nfp_net *nn = netdev_priv(netdev);
2459 if (TC_H_MAJ(handle) != TC_H_MAJ(TC_H_INGRESS))
2461 if (proto != htons(ETH_P_ALL))
2464 if (tc->type == TC_SETUP_CLSBPF && nfp_net_ebpf_capable(nn))
2465 return nfp_net_bpf_offload(nn, handle, proto, tc->cls_bpf);
2470 static int nfp_net_set_features(struct net_device *netdev,
2471 netdev_features_t features)
2473 netdev_features_t changed = netdev->features ^ features;
2474 struct nfp_net *nn = netdev_priv(netdev);
2478 /* Assume this is not called with features we have not advertised */
2480 new_ctrl = nn->ctrl;
2482 if (changed & NETIF_F_RXCSUM) {
2483 if (features & NETIF_F_RXCSUM)
2484 new_ctrl |= NFP_NET_CFG_CTRL_RXCSUM;
2486 new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM;
2489 if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
2490 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
2491 new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
2493 new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM;
2496 if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
2497 if (features & (NETIF_F_TSO | NETIF_F_TSO6))
2498 new_ctrl |= NFP_NET_CFG_CTRL_LSO;
2500 new_ctrl &= ~NFP_NET_CFG_CTRL_LSO;
2503 if (changed & NETIF_F_HW_VLAN_CTAG_RX) {
2504 if (features & NETIF_F_HW_VLAN_CTAG_RX)
2505 new_ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
2507 new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN;
2510 if (changed & NETIF_F_HW_VLAN_CTAG_TX) {
2511 if (features & NETIF_F_HW_VLAN_CTAG_TX)
2512 new_ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
2514 new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN;
2517 if (changed & NETIF_F_SG) {
2518 if (features & NETIF_F_SG)
2519 new_ctrl |= NFP_NET_CFG_CTRL_GATHER;
2521 new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER;
2524 if (changed & NETIF_F_HW_TC && nn->ctrl & NFP_NET_CFG_CTRL_BPF) {
2525 nn_err(nn, "Cannot disable HW TC offload while in use\n");
2529 nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n",
2530 netdev->features, features, changed);
2532 if (new_ctrl == nn->ctrl)
2535 nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->ctrl, new_ctrl);
2536 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2537 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
2541 nn->ctrl = new_ctrl;
2546 static netdev_features_t
2547 nfp_net_features_check(struct sk_buff *skb, struct net_device *dev,
2548 netdev_features_t features)
2552 /* We can't do TSO over double tagged packets (802.1AD) */
2553 features &= vlan_features_check(skb, features);
2555 if (!skb->encapsulation)
2558 /* Ensure that inner L4 header offset fits into TX descriptor field */
2559 if (skb_is_gso(skb)) {
2562 hdrlen = skb_inner_transport_header(skb) - skb->data +
2563 inner_tcp_hdrlen(skb);
2565 if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ))
2566 features &= ~NETIF_F_GSO_MASK;
2569 /* VXLAN/GRE check */
2570 switch (vlan_get_protocol(skb)) {
2571 case htons(ETH_P_IP):
2572 l4_hdr = ip_hdr(skb)->protocol;
2574 case htons(ETH_P_IPV6):
2575 l4_hdr = ipv6_hdr(skb)->nexthdr;
2578 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2581 if (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
2582 skb->inner_protocol != htons(ETH_P_TEB) ||
2583 (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) ||
2584 (l4_hdr == IPPROTO_UDP &&
2585 (skb_inner_mac_header(skb) - skb_transport_header(skb) !=
2586 sizeof(struct udphdr) + sizeof(struct vxlanhdr))))
2587 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2593 * nfp_net_set_vxlan_port() - set vxlan port in SW and reconfigure HW
2594 * @nn: NFP Net device to reconfigure
2595 * @idx: Index into the port table where new port should be written
2596 * @port: UDP port to configure (pass zero to remove VXLAN port)
2598 static void nfp_net_set_vxlan_port(struct nfp_net *nn, int idx, __be16 port)
2602 nn->vxlan_ports[idx] = port;
2604 if (!(nn->ctrl & NFP_NET_CFG_CTRL_VXLAN))
2607 BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1);
2608 for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2)
2609 nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(port),
2610 be16_to_cpu(nn->vxlan_ports[i + 1]) << 16 |
2611 be16_to_cpu(nn->vxlan_ports[i]));
2613 nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_VXLAN);
2617 * nfp_net_find_vxlan_idx() - find table entry of the port or a free one
2618 * @nn: NFP Network structure
2619 * @port: UDP port to look for
2621 * Return: if the port is already in the table -- it's position;
2622 * if the port is not in the table -- free position to use;
2623 * if the table is full -- -ENOSPC.
2625 static int nfp_net_find_vxlan_idx(struct nfp_net *nn, __be16 port)
2627 int i, free_idx = -ENOSPC;
2629 for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i++) {
2630 if (nn->vxlan_ports[i] == port)
2632 if (!nn->vxlan_usecnt[i])
2639 static void nfp_net_add_vxlan_port(struct net_device *netdev,
2640 struct udp_tunnel_info *ti)
2642 struct nfp_net *nn = netdev_priv(netdev);
2645 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2648 idx = nfp_net_find_vxlan_idx(nn, ti->port);
2652 if (!nn->vxlan_usecnt[idx]++)
2653 nfp_net_set_vxlan_port(nn, idx, ti->port);
2656 static void nfp_net_del_vxlan_port(struct net_device *netdev,
2657 struct udp_tunnel_info *ti)
2659 struct nfp_net *nn = netdev_priv(netdev);
2662 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2665 idx = nfp_net_find_vxlan_idx(nn, ti->port);
2666 if (idx == -ENOSPC || !nn->vxlan_usecnt[idx])
2669 if (!--nn->vxlan_usecnt[idx])
2670 nfp_net_set_vxlan_port(nn, idx, 0);
2673 static const struct net_device_ops nfp_net_netdev_ops = {
2674 .ndo_open = nfp_net_netdev_open,
2675 .ndo_stop = nfp_net_netdev_close,
2676 .ndo_start_xmit = nfp_net_tx,
2677 .ndo_get_stats64 = nfp_net_stat64,
2678 .ndo_setup_tc = nfp_net_setup_tc,
2679 .ndo_tx_timeout = nfp_net_tx_timeout,
2680 .ndo_set_rx_mode = nfp_net_set_rx_mode,
2681 .ndo_change_mtu = nfp_net_change_mtu,
2682 .ndo_set_mac_address = eth_mac_addr,
2683 .ndo_set_features = nfp_net_set_features,
2684 .ndo_features_check = nfp_net_features_check,
2685 .ndo_udp_tunnel_add = nfp_net_add_vxlan_port,
2686 .ndo_udp_tunnel_del = nfp_net_del_vxlan_port,
2690 * nfp_net_info() - Print general info about the NIC
2691 * @nn: NFP Net device to reconfigure
2693 void nfp_net_info(struct nfp_net *nn)
2695 nn_info(nn, "Netronome %s %sNetdev: TxQs=%d/%d RxQs=%d/%d\n",
2696 nn->is_nfp3200 ? "NFP-32xx" : "NFP-6xxx",
2697 nn->is_vf ? "VF " : "",
2698 nn->num_tx_rings, nn->max_tx_rings,
2699 nn->num_rx_rings, nn->max_rx_rings);
2700 nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n",
2701 nn->fw_ver.resv, nn->fw_ver.class,
2702 nn->fw_ver.major, nn->fw_ver.minor,
2704 nn_info(nn, "CAP: %#x %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2706 nn->cap & NFP_NET_CFG_CTRL_PROMISC ? "PROMISC " : "",
2707 nn->cap & NFP_NET_CFG_CTRL_L2BC ? "L2BCFILT " : "",
2708 nn->cap & NFP_NET_CFG_CTRL_L2MC ? "L2MCFILT " : "",
2709 nn->cap & NFP_NET_CFG_CTRL_RXCSUM ? "RXCSUM " : "",
2710 nn->cap & NFP_NET_CFG_CTRL_TXCSUM ? "TXCSUM " : "",
2711 nn->cap & NFP_NET_CFG_CTRL_RXVLAN ? "RXVLAN " : "",
2712 nn->cap & NFP_NET_CFG_CTRL_TXVLAN ? "TXVLAN " : "",
2713 nn->cap & NFP_NET_CFG_CTRL_SCATTER ? "SCATTER " : "",
2714 nn->cap & NFP_NET_CFG_CTRL_GATHER ? "GATHER " : "",
2715 nn->cap & NFP_NET_CFG_CTRL_LSO ? "TSO " : "",
2716 nn->cap & NFP_NET_CFG_CTRL_RSS ? "RSS " : "",
2717 nn->cap & NFP_NET_CFG_CTRL_L2SWITCH ? "L2SWITCH " : "",
2718 nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "",
2719 nn->cap & NFP_NET_CFG_CTRL_IRQMOD ? "IRQMOD " : "",
2720 nn->cap & NFP_NET_CFG_CTRL_VXLAN ? "VXLAN " : "",
2721 nn->cap & NFP_NET_CFG_CTRL_NVGRE ? "NVGRE " : "",
2722 nfp_net_ebpf_capable(nn) ? "BPF " : "");
2726 * nfp_net_netdev_alloc() - Allocate netdev and related structure
2728 * @max_tx_rings: Maximum number of TX rings supported by device
2729 * @max_rx_rings: Maximum number of RX rings supported by device
2731 * This function allocates a netdev device and fills in the initial
2732 * part of the @struct nfp_net structure.
2734 * Return: NFP Net device structure, or ERR_PTR on error.
2736 struct nfp_net *nfp_net_netdev_alloc(struct pci_dev *pdev,
2737 int max_tx_rings, int max_rx_rings)
2739 struct net_device *netdev;
2743 netdev = alloc_etherdev_mqs(sizeof(struct nfp_net),
2744 max_tx_rings, max_rx_rings);
2746 return ERR_PTR(-ENOMEM);
2748 SET_NETDEV_DEV(netdev, &pdev->dev);
2749 nn = netdev_priv(netdev);
2751 nn->netdev = netdev;
2754 nn->max_tx_rings = max_tx_rings;
2755 nn->max_rx_rings = max_rx_rings;
2757 nqs = netif_get_num_default_rss_queues();
2758 nn->num_tx_rings = min_t(int, nqs, max_tx_rings);
2759 nn->num_rx_rings = min_t(int, nqs, max_rx_rings);
2761 nn->txd_cnt = NFP_NET_TX_DESCS_DEFAULT;
2762 nn->rxd_cnt = NFP_NET_RX_DESCS_DEFAULT;
2764 spin_lock_init(&nn->reconfig_lock);
2765 spin_lock_init(&nn->rx_filter_lock);
2766 spin_lock_init(&nn->link_status_lock);
2768 setup_timer(&nn->reconfig_timer,
2769 nfp_net_reconfig_timer, (unsigned long)nn);
2770 setup_timer(&nn->rx_filter_stats_timer,
2771 nfp_net_filter_stats_timer, (unsigned long)nn);
2777 * nfp_net_netdev_free() - Undo what @nfp_net_netdev_alloc() did
2778 * @nn: NFP Net device to reconfigure
2780 void nfp_net_netdev_free(struct nfp_net *nn)
2782 free_netdev(nn->netdev);
2786 * nfp_net_rss_init() - Set the initial RSS parameters
2787 * @nn: NFP Net device to reconfigure
2789 static void nfp_net_rss_init(struct nfp_net *nn)
2793 netdev_rss_key_fill(nn->rss_key, NFP_NET_CFG_RSS_KEY_SZ);
2795 for (i = 0; i < sizeof(nn->rss_itbl); i++)
2797 ethtool_rxfh_indir_default(i, nn->num_rx_rings);
2799 /* Enable IPv4/IPv6 TCP by default */
2800 nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP |
2801 NFP_NET_CFG_RSS_IPV6_TCP |
2802 NFP_NET_CFG_RSS_TOEPLITZ |
2803 NFP_NET_CFG_RSS_MASK;
2807 * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters
2808 * @nn: NFP Net device to reconfigure
2810 static void nfp_net_irqmod_init(struct nfp_net *nn)
2812 nn->rx_coalesce_usecs = 50;
2813 nn->rx_coalesce_max_frames = 64;
2814 nn->tx_coalesce_usecs = 50;
2815 nn->tx_coalesce_max_frames = 64;
2819 * nfp_net_netdev_init() - Initialise/finalise the netdev structure
2820 * @netdev: netdev structure
2822 * Return: 0 on success or negative errno on error.
2824 int nfp_net_netdev_init(struct net_device *netdev)
2826 struct nfp_net *nn = netdev_priv(netdev);
2829 /* Get some of the read-only fields from the BAR */
2830 nn->cap = nn_readl(nn, NFP_NET_CFG_CAP);
2831 nn->max_mtu = nn_readl(nn, NFP_NET_CFG_MAX_MTU);
2833 nfp_net_write_mac_addr(nn);
2835 /* Set default MTU and Freelist buffer size */
2836 if (nn->max_mtu < NFP_NET_DEFAULT_MTU)
2837 netdev->mtu = nn->max_mtu;
2839 netdev->mtu = NFP_NET_DEFAULT_MTU;
2840 nn->fl_bufsz = NFP_NET_DEFAULT_RX_BUFSZ;
2842 /* Advertise/enable offloads based on capabilities
2844 * Note: netdev->features show the currently enabled features
2845 * and netdev->hw_features advertises which features are
2846 * supported. By default we enable most features.
2848 netdev->hw_features = NETIF_F_HIGHDMA;
2849 if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM) {
2850 netdev->hw_features |= NETIF_F_RXCSUM;
2851 nn->ctrl |= NFP_NET_CFG_CTRL_RXCSUM;
2853 if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) {
2854 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2855 nn->ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
2857 if (nn->cap & NFP_NET_CFG_CTRL_GATHER) {
2858 netdev->hw_features |= NETIF_F_SG;
2859 nn->ctrl |= NFP_NET_CFG_CTRL_GATHER;
2861 if ((nn->cap & NFP_NET_CFG_CTRL_LSO) && nn->fw_ver.major > 2) {
2862 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
2863 nn->ctrl |= NFP_NET_CFG_CTRL_LSO;
2865 if (nn->cap & NFP_NET_CFG_CTRL_RSS) {
2866 netdev->hw_features |= NETIF_F_RXHASH;
2867 nfp_net_rss_init(nn);
2868 nn->ctrl |= NFP_NET_CFG_CTRL_RSS;
2870 if (nn->cap & NFP_NET_CFG_CTRL_VXLAN &&
2871 nn->cap & NFP_NET_CFG_CTRL_NVGRE) {
2872 if (nn->cap & NFP_NET_CFG_CTRL_LSO)
2873 netdev->hw_features |= NETIF_F_GSO_GRE |
2874 NETIF_F_GSO_UDP_TUNNEL;
2875 nn->ctrl |= NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE;
2877 netdev->hw_enc_features = netdev->hw_features;
2880 netdev->vlan_features = netdev->hw_features;
2882 if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN) {
2883 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
2884 nn->ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
2886 if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN) {
2887 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
2888 nn->ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
2891 netdev->features = netdev->hw_features;
2893 if (nfp_net_ebpf_capable(nn))
2894 netdev->hw_features |= NETIF_F_HW_TC;
2896 /* Advertise but disable TSO by default. */
2897 netdev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
2899 /* Allow L2 Broadcast and Multicast through by default, if supported */
2900 if (nn->cap & NFP_NET_CFG_CTRL_L2BC)
2901 nn->ctrl |= NFP_NET_CFG_CTRL_L2BC;
2902 if (nn->cap & NFP_NET_CFG_CTRL_L2MC)
2903 nn->ctrl |= NFP_NET_CFG_CTRL_L2MC;
2905 /* Allow IRQ moderation, if supported */
2906 if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) {
2907 nfp_net_irqmod_init(nn);
2908 nn->ctrl |= NFP_NET_CFG_CTRL_IRQMOD;
2911 /* On NFP-3200 enable MSI-X auto-masking, if supported and the
2912 * interrupts are not shared.
2914 if (nn->is_nfp3200 && nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO)
2915 nn->ctrl |= NFP_NET_CFG_CTRL_MSIXAUTO;
2917 /* On NFP4000/NFP6000, determine RX packet/metadata boundary offset */
2918 if (nn->fw_ver.major >= 2)
2919 nn->rx_offset = nn_readl(nn, NFP_NET_CFG_RX_OFFSET);
2921 nn->rx_offset = NFP_NET_RX_OFFSET;
2923 /* Stash the re-configuration queue away. First odd queue in TX Bar */
2924 nn->qcp_cfg = nn->tx_bar + NFP_QCP_QUEUE_ADDR_SZ;
2926 /* Make sure the FW knows the netdev is supposed to be disabled here */
2927 nn_writel(nn, NFP_NET_CFG_CTRL, 0);
2928 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
2929 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
2930 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_RING |
2931 NFP_NET_CFG_UPDATE_GEN);
2935 /* Finalise the netdev setup */
2936 ether_setup(netdev);
2937 netdev->netdev_ops = &nfp_net_netdev_ops;
2938 netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000);
2939 netif_carrier_off(netdev);
2941 nfp_net_set_ethtool_ops(netdev);
2942 nfp_net_irqs_assign(netdev);
2944 return register_netdev(netdev);
2948 * nfp_net_netdev_clean() - Undo what nfp_net_netdev_init() did.
2949 * @netdev: netdev structure
2951 void nfp_net_netdev_clean(struct net_device *netdev)
2953 unregister_netdev(netdev);