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/vxlan.h>
65 #include "nfp_net_ctrl.h"
69 * nfp_net_get_fw_version() - Read and parse the FW version
70 * @fw_ver: Output fw_version structure to read to
71 * @ctrl_bar: Mapped address of the control BAR
73 void nfp_net_get_fw_version(struct nfp_net_fw_version
*fw_ver
,
74 void __iomem
*ctrl_bar
)
78 reg
= readl(ctrl_bar
+ NFP_NET_CFG_VERSION
);
79 put_unaligned_le32(reg
, fw_ver
);
84 * Firmware reconfig may take a while so we have two versions of it -
85 * synchronous and asynchronous (posted). All synchronous callers are holding
86 * RTNL so we don't have to worry about serializing them.
88 static void nfp_net_reconfig_start(struct nfp_net
*nn
, u32 update
)
90 nn_writel(nn
, NFP_NET_CFG_UPDATE
, update
);
91 /* ensure update is written before pinging HW */
93 nfp_qcp_wr_ptr_add(nn
->qcp_cfg
, 1);
96 /* Pass 0 as update to run posted reconfigs. */
97 static void nfp_net_reconfig_start_async(struct nfp_net
*nn
, u32 update
)
99 update
|= nn
->reconfig_posted
;
100 nn
->reconfig_posted
= 0;
102 nfp_net_reconfig_start(nn
, update
);
104 nn
->reconfig_timer_active
= true;
105 mod_timer(&nn
->reconfig_timer
, jiffies
+ NFP_NET_POLL_TIMEOUT
* HZ
);
108 static bool nfp_net_reconfig_check_done(struct nfp_net
*nn
, bool last_check
)
112 reg
= nn_readl(nn
, NFP_NET_CFG_UPDATE
);
115 if (reg
& NFP_NET_CFG_UPDATE_ERR
) {
116 nn_err(nn
, "Reconfig error: 0x%08x\n", reg
);
118 } else if (last_check
) {
119 nn_err(nn
, "Reconfig timeout: 0x%08x\n", reg
);
126 static int nfp_net_reconfig_wait(struct nfp_net
*nn
, unsigned long deadline
)
128 bool timed_out
= false;
130 /* Poll update field, waiting for NFP to ack the config */
131 while (!nfp_net_reconfig_check_done(nn
, timed_out
)) {
133 timed_out
= time_is_before_eq_jiffies(deadline
);
136 if (nn_readl(nn
, NFP_NET_CFG_UPDATE
) & NFP_NET_CFG_UPDATE_ERR
)
139 return timed_out
? -EIO
: 0;
142 static void nfp_net_reconfig_timer(unsigned long data
)
144 struct nfp_net
*nn
= (void *)data
;
146 spin_lock_bh(&nn
->reconfig_lock
);
148 nn
->reconfig_timer_active
= false;
150 /* If sync caller is present it will take over from us */
151 if (nn
->reconfig_sync_present
)
154 /* Read reconfig status and report errors */
155 nfp_net_reconfig_check_done(nn
, true);
157 if (nn
->reconfig_posted
)
158 nfp_net_reconfig_start_async(nn
, 0);
160 spin_unlock_bh(&nn
->reconfig_lock
);
164 * nfp_net_reconfig_post() - Post async reconfig request
165 * @nn: NFP Net device to reconfigure
166 * @update: The value for the update field in the BAR config
168 * Record FW reconfiguration request. Reconfiguration will be kicked off
169 * whenever reconfiguration machinery is idle. Multiple requests can be
172 static void nfp_net_reconfig_post(struct nfp_net
*nn
, u32 update
)
174 spin_lock_bh(&nn
->reconfig_lock
);
176 /* Sync caller will kick off async reconf when it's done, just post */
177 if (nn
->reconfig_sync_present
) {
178 nn
->reconfig_posted
|= update
;
182 /* Opportunistically check if the previous command is done */
183 if (!nn
->reconfig_timer_active
||
184 nfp_net_reconfig_check_done(nn
, false))
185 nfp_net_reconfig_start_async(nn
, update
);
187 nn
->reconfig_posted
|= update
;
189 spin_unlock_bh(&nn
->reconfig_lock
);
193 * nfp_net_reconfig() - Reconfigure the firmware
194 * @nn: NFP Net device to reconfigure
195 * @update: The value for the update field in the BAR config
197 * Write the update word to the BAR and ping the reconfig queue. The
198 * poll until the firmware has acknowledged the update by zeroing the
201 * Return: Negative errno on error, 0 on success
203 int nfp_net_reconfig(struct nfp_net
*nn
, u32 update
)
205 bool cancelled_timer
= false;
206 u32 pre_posted_requests
;
209 spin_lock_bh(&nn
->reconfig_lock
);
211 nn
->reconfig_sync_present
= true;
213 if (nn
->reconfig_timer_active
) {
214 del_timer(&nn
->reconfig_timer
);
215 nn
->reconfig_timer_active
= false;
216 cancelled_timer
= true;
218 pre_posted_requests
= nn
->reconfig_posted
;
219 nn
->reconfig_posted
= 0;
221 spin_unlock_bh(&nn
->reconfig_lock
);
224 nfp_net_reconfig_wait(nn
, nn
->reconfig_timer
.expires
);
226 /* Run the posted reconfigs which were issued before we started */
227 if (pre_posted_requests
) {
228 nfp_net_reconfig_start(nn
, pre_posted_requests
);
229 nfp_net_reconfig_wait(nn
, jiffies
+ HZ
* NFP_NET_POLL_TIMEOUT
);
232 nfp_net_reconfig_start(nn
, update
);
233 ret
= nfp_net_reconfig_wait(nn
, jiffies
+ HZ
* NFP_NET_POLL_TIMEOUT
);
235 spin_lock_bh(&nn
->reconfig_lock
);
237 if (nn
->reconfig_posted
)
238 nfp_net_reconfig_start_async(nn
, 0);
240 nn
->reconfig_sync_present
= false;
242 spin_unlock_bh(&nn
->reconfig_lock
);
247 /* Interrupt configuration and handling
251 * nfp_net_irq_unmask_msix() - Unmask MSI-X after automasking
252 * @nn: NFP Network structure
253 * @entry_nr: MSI-X table entry
255 * Clear the MSI-X table mask bit for the given entry bypassing Linux irq
256 * handling subsystem. Use *only* to reenable automasked vectors.
258 static void nfp_net_irq_unmask_msix(struct nfp_net
*nn
, unsigned int entry_nr
)
260 struct list_head
*msi_head
= &nn
->pdev
->dev
.msi_list
;
261 struct msi_desc
*entry
;
264 /* All MSI-Xs have the same mask_base */
265 entry
= list_first_entry(msi_head
, struct msi_desc
, list
);
267 off
= (PCI_MSIX_ENTRY_SIZE
* entry_nr
) +
268 PCI_MSIX_ENTRY_VECTOR_CTRL
;
269 writel(0, entry
->mask_base
+ off
);
270 readl(entry
->mask_base
);
274 * nfp_net_irq_unmask() - Unmask automasked interrupt
275 * @nn: NFP Network structure
276 * @entry_nr: MSI-X table entry
278 * If MSI-X auto-masking is enabled clear the mask bit, otherwise
279 * clear the ICR for the entry.
281 static void nfp_net_irq_unmask(struct nfp_net
*nn
, unsigned int entry_nr
)
283 if (nn
->ctrl
& NFP_NET_CFG_CTRL_MSIXAUTO
) {
284 nfp_net_irq_unmask_msix(nn
, entry_nr
);
288 nn_writeb(nn
, NFP_NET_CFG_ICR(entry_nr
), NFP_NET_CFG_ICR_UNMASKED
);
293 * nfp_net_msix_alloc() - Try to allocate MSI-X irqs
294 * @nn: NFP Network structure
295 * @nr_vecs: Number of MSI-X vectors to allocate
297 * For MSI-X we want at least NFP_NET_NON_Q_VECTORS + 1 vectors.
299 * Return: Number of MSI-X vectors obtained or 0 on error.
301 static int nfp_net_msix_alloc(struct nfp_net
*nn
, int nr_vecs
)
303 struct pci_dev
*pdev
= nn
->pdev
;
307 for (i
= 0; i
< nr_vecs
; i
++)
308 nn
->irq_entries
[i
].entry
= i
;
310 nvecs
= pci_enable_msix_range(pdev
, nn
->irq_entries
,
311 NFP_NET_NON_Q_VECTORS
+ 1, nr_vecs
);
313 nn_warn(nn
, "Failed to enable MSI-X. Wanted %d-%d (err=%d)\n",
314 NFP_NET_NON_Q_VECTORS
+ 1, nr_vecs
, nvecs
);
322 * nfp_net_irqs_wanted() - Work out how many interrupt vectors we want
323 * @nn: NFP Network structure
325 * We want a vector per CPU (or ring), whatever is smaller plus
326 * NFP_NET_NON_Q_VECTORS for LSC etc.
328 * Return: Number of interrupts wanted
330 static int nfp_net_irqs_wanted(struct nfp_net
*nn
)
335 ncpus
= num_online_cpus();
337 vecs
= max_t(int, nn
->num_tx_rings
, nn
->num_rx_rings
);
338 vecs
= min_t(int, vecs
, ncpus
);
340 return vecs
+ NFP_NET_NON_Q_VECTORS
;
344 * nfp_net_irqs_alloc() - allocates MSI-X irqs
345 * @nn: NFP Network structure
347 * Return: Number of irqs obtained or 0 on error.
349 int nfp_net_irqs_alloc(struct nfp_net
*nn
)
353 wanted_irqs
= nfp_net_irqs_wanted(nn
);
355 nn
->num_irqs
= nfp_net_msix_alloc(nn
, wanted_irqs
);
356 if (nn
->num_irqs
== 0) {
357 nn_err(nn
, "Failed to allocate MSI-X IRQs\n");
361 nn
->num_r_vecs
= nn
->num_irqs
- NFP_NET_NON_Q_VECTORS
;
363 if (nn
->num_irqs
< wanted_irqs
)
364 nn_warn(nn
, "Unable to allocate %d vectors. Got %d instead\n",
365 wanted_irqs
, nn
->num_irqs
);
371 * nfp_net_irqs_disable() - Disable interrupts
372 * @nn: NFP Network structure
374 * Undoes what @nfp_net_irqs_alloc() does.
376 void nfp_net_irqs_disable(struct nfp_net
*nn
)
378 pci_disable_msix(nn
->pdev
);
382 * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings.
384 * @data: Opaque data structure
386 * Return: Indicate if the interrupt has been handled.
388 static irqreturn_t
nfp_net_irq_rxtx(int irq
, void *data
)
390 struct nfp_net_r_vector
*r_vec
= data
;
392 napi_schedule_irqoff(&r_vec
->napi
);
394 /* The FW auto-masks any interrupt, either via the MASK bit in
395 * the MSI-X table or via the per entry ICR field. So there
396 * is no need to disable interrupts here.
402 * nfp_net_read_link_status() - Reread link status from control BAR
403 * @nn: NFP Network structure
405 static void nfp_net_read_link_status(struct nfp_net
*nn
)
411 spin_lock_irqsave(&nn
->link_status_lock
, flags
);
413 sts
= nn_readl(nn
, NFP_NET_CFG_STS
);
414 link_up
= !!(sts
& NFP_NET_CFG_STS_LINK
);
416 if (nn
->link_up
== link_up
)
419 nn
->link_up
= link_up
;
422 netif_carrier_on(nn
->netdev
);
423 netdev_info(nn
->netdev
, "NIC Link is Up\n");
425 netif_carrier_off(nn
->netdev
);
426 netdev_info(nn
->netdev
, "NIC Link is Down\n");
429 spin_unlock_irqrestore(&nn
->link_status_lock
, flags
);
433 * nfp_net_irq_lsc() - Interrupt service routine for link state changes
435 * @data: Opaque data structure
437 * Return: Indicate if the interrupt has been handled.
439 static irqreturn_t
nfp_net_irq_lsc(int irq
, void *data
)
441 struct nfp_net
*nn
= data
;
443 nfp_net_read_link_status(nn
);
445 nfp_net_irq_unmask(nn
, NFP_NET_IRQ_LSC_IDX
);
451 * nfp_net_irq_exn() - Interrupt service routine for exceptions
453 * @data: Opaque data structure
455 * Return: Indicate if the interrupt has been handled.
457 static irqreturn_t
nfp_net_irq_exn(int irq
, void *data
)
459 struct nfp_net
*nn
= data
;
461 nn_err(nn
, "%s: UNIMPLEMENTED.\n", __func__
);
462 /* XXX TO BE IMPLEMENTED */
467 * nfp_net_tx_ring_init() - Fill in the boilerplate for a TX ring
468 * @tx_ring: TX ring structure
469 * @r_vec: IRQ vector servicing this ring
473 nfp_net_tx_ring_init(struct nfp_net_tx_ring
*tx_ring
,
474 struct nfp_net_r_vector
*r_vec
, unsigned int idx
)
476 struct nfp_net
*nn
= r_vec
->nfp_net
;
479 tx_ring
->r_vec
= r_vec
;
481 tx_ring
->qcidx
= tx_ring
->idx
* nn
->stride_tx
;
482 tx_ring
->qcp_q
= nn
->tx_bar
+ NFP_QCP_QUEUE_OFF(tx_ring
->qcidx
);
486 * nfp_net_rx_ring_init() - Fill in the boilerplate for a RX ring
487 * @rx_ring: RX ring structure
488 * @r_vec: IRQ vector servicing this ring
492 nfp_net_rx_ring_init(struct nfp_net_rx_ring
*rx_ring
,
493 struct nfp_net_r_vector
*r_vec
, unsigned int idx
)
495 struct nfp_net
*nn
= r_vec
->nfp_net
;
498 rx_ring
->r_vec
= r_vec
;
500 rx_ring
->fl_qcidx
= rx_ring
->idx
* nn
->stride_rx
;
501 rx_ring
->rx_qcidx
= rx_ring
->fl_qcidx
+ (nn
->stride_rx
- 1);
503 rx_ring
->qcp_fl
= nn
->rx_bar
+ NFP_QCP_QUEUE_OFF(rx_ring
->fl_qcidx
);
504 rx_ring
->qcp_rx
= nn
->rx_bar
+ NFP_QCP_QUEUE_OFF(rx_ring
->rx_qcidx
);
508 * nfp_net_irqs_assign() - Assign IRQs and setup rvecs.
509 * @netdev: netdev structure
511 static void nfp_net_irqs_assign(struct net_device
*netdev
)
513 struct nfp_net
*nn
= netdev_priv(netdev
);
514 struct nfp_net_r_vector
*r_vec
;
517 /* Assumes nn->num_tx_rings == nn->num_rx_rings */
518 if (nn
->num_tx_rings
> nn
->num_r_vecs
) {
519 nn_warn(nn
, "More rings (%d) than vectors (%d).\n",
520 nn
->num_tx_rings
, nn
->num_r_vecs
);
521 nn
->num_tx_rings
= nn
->num_r_vecs
;
522 nn
->num_rx_rings
= nn
->num_r_vecs
;
525 nn
->lsc_handler
= nfp_net_irq_lsc
;
526 nn
->exn_handler
= nfp_net_irq_exn
;
528 for (r
= 0; r
< nn
->num_r_vecs
; r
++) {
529 r_vec
= &nn
->r_vecs
[r
];
531 r_vec
->handler
= nfp_net_irq_rxtx
;
532 r_vec
->irq_idx
= NFP_NET_NON_Q_VECTORS
+ r
;
534 cpumask_set_cpu(r
, &r_vec
->affinity_mask
);
539 * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN)
540 * @nn: NFP Network structure
541 * @ctrl_offset: Control BAR offset where IRQ configuration should be written
542 * @format: printf-style format to construct the interrupt name
543 * @name: Pointer to allocated space for interrupt name
544 * @name_sz: Size of space for interrupt name
545 * @vector_idx: Index of MSI-X vector used for this interrupt
546 * @handler: IRQ handler to register for this interrupt
549 nfp_net_aux_irq_request(struct nfp_net
*nn
, u32 ctrl_offset
,
550 const char *format
, char *name
, size_t name_sz
,
551 unsigned int vector_idx
, irq_handler_t handler
)
553 struct msix_entry
*entry
;
556 entry
= &nn
->irq_entries
[vector_idx
];
558 snprintf(name
, name_sz
, format
, netdev_name(nn
->netdev
));
559 err
= request_irq(entry
->vector
, handler
, 0, name
, nn
);
561 nn_err(nn
, "Failed to request IRQ %d (err=%d).\n",
565 nn_writeb(nn
, ctrl_offset
, vector_idx
);
571 * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN)
572 * @nn: NFP Network structure
573 * @ctrl_offset: Control BAR offset where IRQ configuration should be written
574 * @vector_idx: Index of MSI-X vector used for this interrupt
576 static void nfp_net_aux_irq_free(struct nfp_net
*nn
, u32 ctrl_offset
,
577 unsigned int vector_idx
)
579 nn_writeb(nn
, ctrl_offset
, 0xff);
580 free_irq(nn
->irq_entries
[vector_idx
].vector
, nn
);
585 * One queue controller peripheral queue is used for transmit. The
586 * driver en-queues packets for transmit by advancing the write
587 * pointer. The device indicates that packets have transmitted by
588 * advancing the read pointer. The driver maintains a local copy of
589 * the read and write pointer in @struct nfp_net_tx_ring. The driver
590 * keeps @wr_p in sync with the queue controller write pointer and can
591 * determine how many packets have been transmitted by comparing its
592 * copy of the read pointer @rd_p with the read pointer maintained by
593 * the queue controller peripheral.
597 * nfp_net_tx_full() - Check if the TX ring is full
598 * @tx_ring: TX ring to check
599 * @dcnt: Number of descriptors that need to be enqueued (must be >= 1)
601 * This function checks, based on the *host copy* of read/write
602 * pointer if a given TX ring is full. The real TX queue may have
603 * some newly made available slots.
605 * Return: True if the ring is full.
607 static inline int nfp_net_tx_full(struct nfp_net_tx_ring
*tx_ring
, int dcnt
)
609 return (tx_ring
->wr_p
- tx_ring
->rd_p
) >= (tx_ring
->cnt
- dcnt
);
612 /* Wrappers for deciding when to stop and restart TX queues */
613 static int nfp_net_tx_ring_should_wake(struct nfp_net_tx_ring
*tx_ring
)
615 return !nfp_net_tx_full(tx_ring
, MAX_SKB_FRAGS
* 4);
618 static int nfp_net_tx_ring_should_stop(struct nfp_net_tx_ring
*tx_ring
)
620 return nfp_net_tx_full(tx_ring
, MAX_SKB_FRAGS
+ 1);
624 * nfp_net_tx_ring_stop() - stop tx ring
625 * @nd_q: netdev queue
626 * @tx_ring: driver tx queue structure
628 * Safely stop TX ring. Remember that while we are running .start_xmit()
629 * someone else may be cleaning the TX ring completions so we need to be
630 * extra careful here.
632 static void nfp_net_tx_ring_stop(struct netdev_queue
*nd_q
,
633 struct nfp_net_tx_ring
*tx_ring
)
635 netif_tx_stop_queue(nd_q
);
637 /* We can race with the TX completion out of NAPI so recheck */
639 if (unlikely(nfp_net_tx_ring_should_wake(tx_ring
)))
640 netif_tx_start_queue(nd_q
);
644 * nfp_net_tx_tso() - Set up Tx descriptor for LSO
645 * @nn: NFP Net device
646 * @r_vec: per-ring structure
647 * @txbuf: Pointer to driver soft TX descriptor
648 * @txd: Pointer to HW TX descriptor
649 * @skb: Pointer to SKB
651 * Set up Tx descriptor for LSO, do nothing for non-LSO skbs.
652 * Return error on packet header greater than maximum supported LSO header size.
654 static void nfp_net_tx_tso(struct nfp_net
*nn
, struct nfp_net_r_vector
*r_vec
,
655 struct nfp_net_tx_buf
*txbuf
,
656 struct nfp_net_tx_desc
*txd
, struct sk_buff
*skb
)
661 if (!skb_is_gso(skb
))
664 if (!skb
->encapsulation
)
665 hdrlen
= skb_transport_offset(skb
) + tcp_hdrlen(skb
);
667 hdrlen
= skb_inner_transport_header(skb
) - skb
->data
+
668 inner_tcp_hdrlen(skb
);
670 txbuf
->pkt_cnt
= skb_shinfo(skb
)->gso_segs
;
671 txbuf
->real_len
+= hdrlen
* (txbuf
->pkt_cnt
- 1);
673 mss
= skb_shinfo(skb
)->gso_size
& PCIE_DESC_TX_MSS_MASK
;
674 txd
->l4_offset
= hdrlen
;
675 txd
->mss
= cpu_to_le16(mss
);
676 txd
->flags
|= PCIE_DESC_TX_LSO
;
678 u64_stats_update_begin(&r_vec
->tx_sync
);
680 u64_stats_update_end(&r_vec
->tx_sync
);
684 * nfp_net_tx_csum() - Set TX CSUM offload flags in TX descriptor
685 * @nn: NFP Net device
686 * @r_vec: per-ring structure
687 * @txbuf: Pointer to driver soft TX descriptor
688 * @txd: Pointer to TX descriptor
689 * @skb: Pointer to SKB
691 * This function sets the TX checksum flags in the TX descriptor based
692 * on the configuration and the protocol of the packet to be transmitted.
694 static void nfp_net_tx_csum(struct nfp_net
*nn
, struct nfp_net_r_vector
*r_vec
,
695 struct nfp_net_tx_buf
*txbuf
,
696 struct nfp_net_tx_desc
*txd
, struct sk_buff
*skb
)
698 struct ipv6hdr
*ipv6h
;
702 if (!(nn
->ctrl
& NFP_NET_CFG_CTRL_TXCSUM
))
705 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
708 txd
->flags
|= PCIE_DESC_TX_CSUM
;
709 if (skb
->encapsulation
)
710 txd
->flags
|= PCIE_DESC_TX_ENCAP
;
712 iph
= skb
->encapsulation
? inner_ip_hdr(skb
) : ip_hdr(skb
);
713 ipv6h
= skb
->encapsulation
? inner_ipv6_hdr(skb
) : ipv6_hdr(skb
);
715 if (iph
->version
== 4) {
716 txd
->flags
|= PCIE_DESC_TX_IP4_CSUM
;
717 l4_hdr
= iph
->protocol
;
718 } else if (ipv6h
->version
== 6) {
719 l4_hdr
= ipv6h
->nexthdr
;
721 nn_warn_ratelimit(nn
, "partial checksum but ipv=%x!\n",
728 txd
->flags
|= PCIE_DESC_TX_TCP_CSUM
;
731 txd
->flags
|= PCIE_DESC_TX_UDP_CSUM
;
734 nn_warn_ratelimit(nn
, "partial checksum but l4 proto=%x!\n",
739 u64_stats_update_begin(&r_vec
->tx_sync
);
740 if (skb
->encapsulation
)
741 r_vec
->hw_csum_tx_inner
+= txbuf
->pkt_cnt
;
743 r_vec
->hw_csum_tx
+= txbuf
->pkt_cnt
;
744 u64_stats_update_end(&r_vec
->tx_sync
);
748 * nfp_net_tx() - Main transmit entry point
749 * @skb: SKB to transmit
750 * @netdev: netdev structure
752 * Return: NETDEV_TX_OK on success.
754 static int nfp_net_tx(struct sk_buff
*skb
, struct net_device
*netdev
)
756 struct nfp_net
*nn
= netdev_priv(netdev
);
757 const struct skb_frag_struct
*frag
;
758 struct nfp_net_r_vector
*r_vec
;
759 struct nfp_net_tx_desc
*txd
, txdg
;
760 struct nfp_net_tx_buf
*txbuf
;
761 struct nfp_net_tx_ring
*tx_ring
;
762 struct netdev_queue
*nd_q
;
769 qidx
= skb_get_queue_mapping(skb
);
770 tx_ring
= &nn
->tx_rings
[qidx
];
771 r_vec
= tx_ring
->r_vec
;
772 nd_q
= netdev_get_tx_queue(nn
->netdev
, qidx
);
774 nr_frags
= skb_shinfo(skb
)->nr_frags
;
776 if (unlikely(nfp_net_tx_full(tx_ring
, nr_frags
+ 1))) {
777 nn_warn_ratelimit(nn
, "TX ring %d busy. wrp=%u rdp=%u\n",
778 qidx
, tx_ring
->wr_p
, tx_ring
->rd_p
);
779 netif_tx_stop_queue(nd_q
);
780 u64_stats_update_begin(&r_vec
->tx_sync
);
782 u64_stats_update_end(&r_vec
->tx_sync
);
783 return NETDEV_TX_BUSY
;
786 /* Start with the head skbuf */
787 dma_addr
= dma_map_single(&nn
->pdev
->dev
, skb
->data
, skb_headlen(skb
),
789 if (dma_mapping_error(&nn
->pdev
->dev
, dma_addr
))
792 wr_idx
= tx_ring
->wr_p
% tx_ring
->cnt
;
794 /* Stash the soft descriptor of the head then initialize it */
795 txbuf
= &tx_ring
->txbufs
[wr_idx
];
797 txbuf
->dma_addr
= dma_addr
;
800 txbuf
->real_len
= skb
->len
;
802 /* Build TX descriptor */
803 txd
= &tx_ring
->txds
[wr_idx
];
804 txd
->offset_eop
= (nr_frags
== 0) ? PCIE_DESC_TX_EOP
: 0;
805 txd
->dma_len
= cpu_to_le16(skb_headlen(skb
));
806 nfp_desc_set_dma_addr(txd
, dma_addr
);
807 txd
->data_len
= cpu_to_le16(skb
->len
);
813 nfp_net_tx_tso(nn
, r_vec
, txbuf
, txd
, skb
);
815 nfp_net_tx_csum(nn
, r_vec
, txbuf
, txd
, skb
);
817 if (skb_vlan_tag_present(skb
) && nn
->ctrl
& NFP_NET_CFG_CTRL_TXVLAN
) {
818 txd
->flags
|= PCIE_DESC_TX_VLAN
;
819 txd
->vlan
= cpu_to_le16(skb_vlan_tag_get(skb
));
824 /* all descs must match except for in addr, length and eop */
827 for (f
= 0; f
< nr_frags
; f
++) {
828 frag
= &skb_shinfo(skb
)->frags
[f
];
829 fsize
= skb_frag_size(frag
);
831 dma_addr
= skb_frag_dma_map(&nn
->pdev
->dev
, frag
, 0,
832 fsize
, DMA_TO_DEVICE
);
833 if (dma_mapping_error(&nn
->pdev
->dev
, dma_addr
))
836 wr_idx
= (wr_idx
+ 1) % tx_ring
->cnt
;
837 tx_ring
->txbufs
[wr_idx
].skb
= skb
;
838 tx_ring
->txbufs
[wr_idx
].dma_addr
= dma_addr
;
839 tx_ring
->txbufs
[wr_idx
].fidx
= f
;
841 txd
= &tx_ring
->txds
[wr_idx
];
843 txd
->dma_len
= cpu_to_le16(fsize
);
844 nfp_desc_set_dma_addr(txd
, dma_addr
);
846 (f
== nr_frags
- 1) ? PCIE_DESC_TX_EOP
: 0;
849 u64_stats_update_begin(&r_vec
->tx_sync
);
851 u64_stats_update_end(&r_vec
->tx_sync
);
854 netdev_tx_sent_queue(nd_q
, txbuf
->real_len
);
856 tx_ring
->wr_p
+= nr_frags
+ 1;
857 if (nfp_net_tx_ring_should_stop(tx_ring
))
858 nfp_net_tx_ring_stop(nd_q
, tx_ring
);
860 tx_ring
->wr_ptr_add
+= nr_frags
+ 1;
861 if (!skb
->xmit_more
|| netif_xmit_stopped(nd_q
)) {
862 /* force memory write before we let HW know */
864 nfp_qcp_wr_ptr_add(tx_ring
->qcp_q
, tx_ring
->wr_ptr_add
);
865 tx_ring
->wr_ptr_add
= 0;
868 skb_tx_timestamp(skb
);
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
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 * Warning: Do *not* call if ring buffers were never put on the FW freelist
1148 * (i.e. device was not enabled)!
1150 static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring
*rx_ring
)
1152 unsigned int wr_idx
, last_idx
;
1154 /* Move the empty entry to the end of the list */
1155 wr_idx
= rx_ring
->wr_p
% rx_ring
->cnt
;
1156 last_idx
= rx_ring
->cnt
- 1;
1157 rx_ring
->rxbufs
[wr_idx
].dma_addr
= rx_ring
->rxbufs
[last_idx
].dma_addr
;
1158 rx_ring
->rxbufs
[wr_idx
].skb
= rx_ring
->rxbufs
[last_idx
].skb
;
1159 rx_ring
->rxbufs
[last_idx
].dma_addr
= 0;
1160 rx_ring
->rxbufs
[last_idx
].skb
= NULL
;
1162 memset(rx_ring
->rxds
, 0, sizeof(*rx_ring
->rxds
) * rx_ring
->cnt
);
1165 rx_ring
->wr_ptr_add
= 0;
1169 * nfp_net_rx_ring_bufs_free() - Free any buffers currently on the RX ring
1170 * @nn: NFP Net device
1171 * @rx_ring: RX ring to remove buffers from
1173 * Assumes that the device is stopped and buffers are in [0, ring->cnt - 1)
1174 * entries. After device is disabled nfp_net_rx_ring_reset() must be called
1175 * to restore required ring geometry.
1178 nfp_net_rx_ring_bufs_free(struct nfp_net
*nn
, struct nfp_net_rx_ring
*rx_ring
)
1180 struct pci_dev
*pdev
= nn
->pdev
;
1183 for (i
= 0; i
< rx_ring
->cnt
- 1; i
++) {
1184 /* NULL skb can only happen when initial filling of the ring
1185 * fails to allocate enough buffers and calls here to free
1186 * already allocated ones.
1188 if (!rx_ring
->rxbufs
[i
].skb
)
1191 dma_unmap_single(&pdev
->dev
, rx_ring
->rxbufs
[i
].dma_addr
,
1192 rx_ring
->bufsz
, DMA_FROM_DEVICE
);
1193 dev_kfree_skb_any(rx_ring
->rxbufs
[i
].skb
);
1194 rx_ring
->rxbufs
[i
].dma_addr
= 0;
1195 rx_ring
->rxbufs
[i
].skb
= NULL
;
1200 * nfp_net_rx_ring_bufs_alloc() - Fill RX ring with buffers (don't give to FW)
1201 * @nn: NFP Net device
1202 * @rx_ring: RX ring to remove buffers from
1205 nfp_net_rx_ring_bufs_alloc(struct nfp_net
*nn
, struct nfp_net_rx_ring
*rx_ring
)
1207 struct nfp_net_rx_buf
*rxbufs
;
1210 rxbufs
= rx_ring
->rxbufs
;
1212 for (i
= 0; i
< rx_ring
->cnt
- 1; i
++) {
1214 nfp_net_rx_alloc_one(rx_ring
, &rxbufs
[i
].dma_addr
,
1216 if (!rxbufs
[i
].skb
) {
1217 nfp_net_rx_ring_bufs_free(nn
, rx_ring
);
1226 * nfp_net_rx_ring_fill_freelist() - Give buffers from the ring to FW
1227 * @rx_ring: RX ring to fill
1229 static void nfp_net_rx_ring_fill_freelist(struct nfp_net_rx_ring
*rx_ring
)
1233 for (i
= 0; i
< rx_ring
->cnt
- 1; i
++)
1234 nfp_net_rx_give_one(rx_ring
, rx_ring
->rxbufs
[i
].skb
,
1235 rx_ring
->rxbufs
[i
].dma_addr
);
1239 * nfp_net_rx_csum_has_errors() - group check if rxd has any csum errors
1240 * @flags: RX descriptor flags field in CPU byte order
1242 static int nfp_net_rx_csum_has_errors(u16 flags
)
1244 u16 csum_all_checked
, csum_all_ok
;
1246 csum_all_checked
= flags
& __PCIE_DESC_RX_CSUM_ALL
;
1247 csum_all_ok
= flags
& __PCIE_DESC_RX_CSUM_ALL_OK
;
1249 return csum_all_checked
!= (csum_all_ok
<< PCIE_DESC_RX_CSUM_OK_SHIFT
);
1253 * nfp_net_rx_csum() - set SKB checksum field based on RX descriptor flags
1254 * @nn: NFP Net device
1255 * @r_vec: per-ring structure
1256 * @rxd: Pointer to RX descriptor
1257 * @skb: Pointer to SKB
1259 static void nfp_net_rx_csum(struct nfp_net
*nn
, struct nfp_net_r_vector
*r_vec
,
1260 struct nfp_net_rx_desc
*rxd
, struct sk_buff
*skb
)
1262 skb_checksum_none_assert(skb
);
1264 if (!(nn
->netdev
->features
& NETIF_F_RXCSUM
))
1267 if (nfp_net_rx_csum_has_errors(le16_to_cpu(rxd
->rxd
.flags
))) {
1268 u64_stats_update_begin(&r_vec
->rx_sync
);
1269 r_vec
->hw_csum_rx_error
++;
1270 u64_stats_update_end(&r_vec
->rx_sync
);
1274 /* Assume that the firmware will never report inner CSUM_OK unless outer
1275 * L4 headers were successfully parsed. FW will always report zero UDP
1276 * checksum as CSUM_OK.
1278 if (rxd
->rxd
.flags
& PCIE_DESC_RX_TCP_CSUM_OK
||
1279 rxd
->rxd
.flags
& PCIE_DESC_RX_UDP_CSUM_OK
) {
1280 __skb_incr_checksum_unnecessary(skb
);
1281 u64_stats_update_begin(&r_vec
->rx_sync
);
1282 r_vec
->hw_csum_rx_ok
++;
1283 u64_stats_update_end(&r_vec
->rx_sync
);
1286 if (rxd
->rxd
.flags
& PCIE_DESC_RX_I_TCP_CSUM_OK
||
1287 rxd
->rxd
.flags
& PCIE_DESC_RX_I_UDP_CSUM_OK
) {
1288 __skb_incr_checksum_unnecessary(skb
);
1289 u64_stats_update_begin(&r_vec
->rx_sync
);
1290 r_vec
->hw_csum_rx_inner_ok
++;
1291 u64_stats_update_end(&r_vec
->rx_sync
);
1296 * nfp_net_set_hash() - Set SKB hash data
1297 * @netdev: adapter's net_device structure
1298 * @skb: SKB to set the hash data on
1299 * @rxd: RX descriptor
1301 * The RSS hash and hash-type are pre-pended to the packet data.
1302 * Extract and decode it and set the skb fields.
1304 static void nfp_net_set_hash(struct net_device
*netdev
, struct sk_buff
*skb
,
1305 struct nfp_net_rx_desc
*rxd
)
1307 struct nfp_net_rx_hash
*rx_hash
;
1309 if (!(rxd
->rxd
.flags
& PCIE_DESC_RX_RSS
) ||
1310 !(netdev
->features
& NETIF_F_RXHASH
))
1313 rx_hash
= (struct nfp_net_rx_hash
*)(skb
->data
- sizeof(*rx_hash
));
1315 switch (be32_to_cpu(rx_hash
->hash_type
)) {
1316 case NFP_NET_RSS_IPV4
:
1317 case NFP_NET_RSS_IPV6
:
1318 case NFP_NET_RSS_IPV6_EX
:
1319 skb_set_hash(skb
, be32_to_cpu(rx_hash
->hash
), PKT_HASH_TYPE_L3
);
1322 skb_set_hash(skb
, be32_to_cpu(rx_hash
->hash
), PKT_HASH_TYPE_L4
);
1328 * nfp_net_rx() - receive up to @budget packets on @rx_ring
1329 * @rx_ring: RX ring to receive from
1330 * @budget: NAPI budget
1332 * Note, this function is separated out from the napi poll function to
1333 * more cleanly separate packet receive code from other bookkeeping
1334 * functions performed in the napi poll function.
1336 * There are differences between the NFP-3200 firmware and the
1337 * NFP-6000 firmware. The NFP-3200 firmware uses a dedicated RX queue
1338 * to indicate that new packets have arrived. The NFP-6000 does not
1339 * have this queue and uses the DD bit in the RX descriptor. This
1340 * method cannot be used on the NFP-3200 as it causes a race
1341 * condition: The RX ring write pointer on the NFP-3200 is updated
1342 * after packets (and descriptors) have been DMAed. If the DD bit is
1343 * used and subsequently the read pointer is updated this may lead to
1344 * the RX queue to underflow (if the firmware has not yet update the
1345 * write pointer). Therefore we use slightly ugly conditional code
1346 * below to handle the differences. We may, in the future update the
1347 * NFP-3200 firmware to behave the same as the firmware on the
1350 * Return: Number of packets received.
1352 static int nfp_net_rx(struct nfp_net_rx_ring
*rx_ring
, int budget
)
1354 struct nfp_net_r_vector
*r_vec
= rx_ring
->r_vec
;
1355 struct nfp_net
*nn
= r_vec
->nfp_net
;
1356 unsigned int data_len
, meta_len
;
1357 int avail
= 0, pkts_polled
= 0;
1358 struct sk_buff
*skb
, *new_skb
;
1359 struct nfp_net_rx_desc
*rxd
;
1360 dma_addr_t new_dma_addr
;
1364 if (nn
->is_nfp3200
) {
1365 /* Work out how many packets arrived */
1366 qcp_wr_p
= nfp_qcp_wr_ptr_read(rx_ring
->qcp_rx
);
1367 idx
= rx_ring
->rd_p
% rx_ring
->cnt
;
1369 if (qcp_wr_p
== idx
)
1370 /* No new packets */
1374 avail
= qcp_wr_p
- idx
;
1376 avail
= qcp_wr_p
+ rx_ring
->cnt
- idx
;
1381 while (avail
> 0 && pkts_polled
< budget
) {
1382 idx
= rx_ring
->rd_p
% rx_ring
->cnt
;
1384 rxd
= &rx_ring
->rxds
[idx
];
1385 if (!(rxd
->rxd
.meta_len_dd
& PCIE_DESC_RX_DD
)) {
1387 nn_dbg(nn
, "RX descriptor not valid (DD)%d:%u rxd[0]=%#x rxd[1]=%#x\n",
1389 rxd
->vals
[0], rxd
->vals
[1]);
1392 /* Memory barrier to ensure that we won't do other reads
1393 * before the DD bit.
1401 skb
= rx_ring
->rxbufs
[idx
].skb
;
1403 new_skb
= nfp_net_rx_alloc_one(rx_ring
, &new_dma_addr
,
1406 nfp_net_rx_give_one(rx_ring
, rx_ring
->rxbufs
[idx
].skb
,
1407 rx_ring
->rxbufs
[idx
].dma_addr
);
1408 u64_stats_update_begin(&r_vec
->rx_sync
);
1410 u64_stats_update_end(&r_vec
->rx_sync
);
1414 dma_unmap_single(&nn
->pdev
->dev
,
1415 rx_ring
->rxbufs
[idx
].dma_addr
,
1416 nn
->fl_bufsz
, DMA_FROM_DEVICE
);
1418 nfp_net_rx_give_one(rx_ring
, new_skb
, new_dma_addr
);
1421 * <-- [rx_offset] -->
1422 * ---------------------------------------------------------
1423 * | [XX] | metadata | packet | XXXX |
1424 * ---------------------------------------------------------
1425 * <---------------- data_len --------------->
1427 * The rx_offset is fixed for all packets, the meta_len can vary
1428 * on a packet by packet basis. If rx_offset is set to zero
1429 * (_RX_OFFSET_DYNAMIC) metadata starts at the beginning of the
1430 * buffer and is immediately followed by the packet (no [XX]).
1432 meta_len
= rxd
->rxd
.meta_len_dd
& PCIE_DESC_RX_META_LEN_MASK
;
1433 data_len
= le16_to_cpu(rxd
->rxd
.data_len
);
1435 if (nn
->rx_offset
== NFP_NET_CFG_RX_OFFSET_DYNAMIC
)
1436 skb_reserve(skb
, meta_len
);
1438 skb_reserve(skb
, nn
->rx_offset
);
1439 skb_put(skb
, data_len
- meta_len
);
1441 nfp_net_set_hash(nn
->netdev
, skb
, rxd
);
1444 u64_stats_update_begin(&r_vec
->rx_sync
);
1446 r_vec
->rx_bytes
+= skb
->len
;
1447 u64_stats_update_end(&r_vec
->rx_sync
);
1449 skb_record_rx_queue(skb
, rx_ring
->idx
);
1450 skb
->protocol
= eth_type_trans(skb
, nn
->netdev
);
1452 nfp_net_rx_csum(nn
, r_vec
, rxd
, skb
);
1454 if (rxd
->rxd
.flags
& PCIE_DESC_RX_VLAN
)
1455 __vlan_hwaccel_put_tag(skb
, htons(ETH_P_8021Q
),
1456 le16_to_cpu(rxd
->rxd
.vlan
));
1458 napi_gro_receive(&rx_ring
->r_vec
->napi
, skb
);
1462 nfp_qcp_rd_ptr_add(rx_ring
->qcp_rx
, pkts_polled
);
1468 * nfp_net_poll() - napi poll function
1469 * @napi: NAPI structure
1470 * @budget: NAPI budget
1472 * Return: number of packets polled.
1474 static int nfp_net_poll(struct napi_struct
*napi
, int budget
)
1476 struct nfp_net_r_vector
*r_vec
=
1477 container_of(napi
, struct nfp_net_r_vector
, napi
);
1478 struct nfp_net_rx_ring
*rx_ring
= r_vec
->rx_ring
;
1479 struct nfp_net_tx_ring
*tx_ring
= r_vec
->tx_ring
;
1480 struct nfp_net
*nn
= r_vec
->nfp_net
;
1481 struct netdev_queue
*txq
;
1482 unsigned int pkts_polled
;
1484 tx_ring
= &nn
->tx_rings
[rx_ring
->idx
];
1485 txq
= netdev_get_tx_queue(nn
->netdev
, tx_ring
->idx
);
1486 nfp_net_tx_complete(tx_ring
);
1488 pkts_polled
= nfp_net_rx(rx_ring
, budget
);
1490 if (pkts_polled
< budget
) {
1491 napi_complete_done(napi
, pkts_polled
);
1492 nfp_net_irq_unmask(nn
, r_vec
->irq_idx
);
1498 /* Setup and Configuration
1502 * nfp_net_tx_ring_free() - Free resources allocated to a TX ring
1503 * @tx_ring: TX ring to free
1505 static void nfp_net_tx_ring_free(struct nfp_net_tx_ring
*tx_ring
)
1507 struct nfp_net_r_vector
*r_vec
= tx_ring
->r_vec
;
1508 struct nfp_net
*nn
= r_vec
->nfp_net
;
1509 struct pci_dev
*pdev
= nn
->pdev
;
1511 kfree(tx_ring
->txbufs
);
1514 dma_free_coherent(&pdev
->dev
, tx_ring
->size
,
1515 tx_ring
->txds
, tx_ring
->dma
);
1518 tx_ring
->txbufs
= NULL
;
1519 tx_ring
->txds
= NULL
;
1525 * nfp_net_tx_ring_alloc() - Allocate resource for a TX ring
1526 * @tx_ring: TX Ring structure to allocate
1527 * @cnt: Ring buffer count
1529 * Return: 0 on success, negative errno otherwise.
1531 static int nfp_net_tx_ring_alloc(struct nfp_net_tx_ring
*tx_ring
, u32 cnt
)
1533 struct nfp_net_r_vector
*r_vec
= tx_ring
->r_vec
;
1534 struct nfp_net
*nn
= r_vec
->nfp_net
;
1535 struct pci_dev
*pdev
= nn
->pdev
;
1540 tx_ring
->size
= sizeof(*tx_ring
->txds
) * tx_ring
->cnt
;
1541 tx_ring
->txds
= dma_zalloc_coherent(&pdev
->dev
, tx_ring
->size
,
1542 &tx_ring
->dma
, GFP_KERNEL
);
1546 sz
= sizeof(*tx_ring
->txbufs
) * tx_ring
->cnt
;
1547 tx_ring
->txbufs
= kzalloc(sz
, GFP_KERNEL
);
1548 if (!tx_ring
->txbufs
)
1551 netif_set_xps_queue(nn
->netdev
, &r_vec
->affinity_mask
, tx_ring
->idx
);
1553 nn_dbg(nn
, "TxQ%02d: QCidx=%02d cnt=%d dma=%#llx host=%p\n",
1554 tx_ring
->idx
, tx_ring
->qcidx
,
1555 tx_ring
->cnt
, (unsigned long long)tx_ring
->dma
, tx_ring
->txds
);
1560 nfp_net_tx_ring_free(tx_ring
);
1564 static struct nfp_net_tx_ring
*
1565 nfp_net_shadow_tx_rings_prepare(struct nfp_net
*nn
, u32 buf_cnt
)
1567 struct nfp_net_tx_ring
*rings
;
1570 rings
= kcalloc(nn
->num_tx_rings
, sizeof(*rings
), GFP_KERNEL
);
1574 for (r
= 0; r
< nn
->num_tx_rings
; r
++) {
1575 nfp_net_tx_ring_init(&rings
[r
], nn
->tx_rings
[r
].r_vec
, r
);
1577 if (nfp_net_tx_ring_alloc(&rings
[r
], buf_cnt
))
1585 nfp_net_tx_ring_free(&rings
[r
]);
1590 static struct nfp_net_tx_ring
*
1591 nfp_net_shadow_tx_rings_swap(struct nfp_net
*nn
, struct nfp_net_tx_ring
*rings
)
1593 struct nfp_net_tx_ring
*old
= nn
->tx_rings
;
1596 for (r
= 0; r
< nn
->num_tx_rings
; r
++)
1597 old
[r
].r_vec
->tx_ring
= &rings
[r
];
1599 nn
->tx_rings
= rings
;
1604 nfp_net_shadow_tx_rings_free(struct nfp_net
*nn
, struct nfp_net_tx_ring
*rings
)
1611 for (r
= 0; r
< nn
->num_tx_rings
; r
++)
1612 nfp_net_tx_ring_free(&rings
[r
]);
1618 * nfp_net_rx_ring_free() - Free resources allocated to a RX ring
1619 * @rx_ring: RX ring to free
1621 static void nfp_net_rx_ring_free(struct nfp_net_rx_ring
*rx_ring
)
1623 struct nfp_net_r_vector
*r_vec
= rx_ring
->r_vec
;
1624 struct nfp_net
*nn
= r_vec
->nfp_net
;
1625 struct pci_dev
*pdev
= nn
->pdev
;
1627 kfree(rx_ring
->rxbufs
);
1630 dma_free_coherent(&pdev
->dev
, rx_ring
->size
,
1631 rx_ring
->rxds
, rx_ring
->dma
);
1634 rx_ring
->rxbufs
= NULL
;
1635 rx_ring
->rxds
= NULL
;
1641 * nfp_net_rx_ring_alloc() - Allocate resource for a RX ring
1642 * @rx_ring: RX ring to allocate
1643 * @fl_bufsz: Size of buffers to allocate
1644 * @cnt: Ring buffer count
1646 * Return: 0 on success, negative errno otherwise.
1649 nfp_net_rx_ring_alloc(struct nfp_net_rx_ring
*rx_ring
, unsigned int fl_bufsz
,
1652 struct nfp_net_r_vector
*r_vec
= rx_ring
->r_vec
;
1653 struct nfp_net
*nn
= r_vec
->nfp_net
;
1654 struct pci_dev
*pdev
= nn
->pdev
;
1658 rx_ring
->bufsz
= fl_bufsz
;
1660 rx_ring
->size
= sizeof(*rx_ring
->rxds
) * rx_ring
->cnt
;
1661 rx_ring
->rxds
= dma_zalloc_coherent(&pdev
->dev
, rx_ring
->size
,
1662 &rx_ring
->dma
, GFP_KERNEL
);
1666 sz
= sizeof(*rx_ring
->rxbufs
) * rx_ring
->cnt
;
1667 rx_ring
->rxbufs
= kzalloc(sz
, GFP_KERNEL
);
1668 if (!rx_ring
->rxbufs
)
1671 nn_dbg(nn
, "RxQ%02d: FlQCidx=%02d RxQCidx=%02d cnt=%d dma=%#llx host=%p\n",
1672 rx_ring
->idx
, rx_ring
->fl_qcidx
, rx_ring
->rx_qcidx
,
1673 rx_ring
->cnt
, (unsigned long long)rx_ring
->dma
, rx_ring
->rxds
);
1678 nfp_net_rx_ring_free(rx_ring
);
1682 static struct nfp_net_rx_ring
*
1683 nfp_net_shadow_rx_rings_prepare(struct nfp_net
*nn
, unsigned int fl_bufsz
,
1686 struct nfp_net_rx_ring
*rings
;
1689 rings
= kcalloc(nn
->num_rx_rings
, sizeof(*rings
), GFP_KERNEL
);
1693 for (r
= 0; r
< nn
->num_rx_rings
; r
++) {
1694 nfp_net_rx_ring_init(&rings
[r
], nn
->rx_rings
[r
].r_vec
, r
);
1696 if (nfp_net_rx_ring_alloc(&rings
[r
], fl_bufsz
, buf_cnt
))
1699 if (nfp_net_rx_ring_bufs_alloc(nn
, &rings
[r
]))
1707 nfp_net_rx_ring_bufs_free(nn
, &rings
[r
]);
1709 nfp_net_rx_ring_free(&rings
[r
]);
1715 static struct nfp_net_rx_ring
*
1716 nfp_net_shadow_rx_rings_swap(struct nfp_net
*nn
, struct nfp_net_rx_ring
*rings
)
1718 struct nfp_net_rx_ring
*old
= nn
->rx_rings
;
1721 for (r
= 0; r
< nn
->num_rx_rings
; r
++)
1722 old
[r
].r_vec
->rx_ring
= &rings
[r
];
1724 nn
->rx_rings
= rings
;
1729 nfp_net_shadow_rx_rings_free(struct nfp_net
*nn
, struct nfp_net_rx_ring
*rings
)
1736 for (r
= 0; r
< nn
->num_r_vecs
; r
++) {
1737 nfp_net_rx_ring_bufs_free(nn
, &rings
[r
]);
1738 nfp_net_rx_ring_free(&rings
[r
]);
1745 nfp_net_prepare_vector(struct nfp_net
*nn
, struct nfp_net_r_vector
*r_vec
,
1748 struct msix_entry
*entry
= &nn
->irq_entries
[r_vec
->irq_idx
];
1751 r_vec
->tx_ring
= &nn
->tx_rings
[idx
];
1752 nfp_net_tx_ring_init(r_vec
->tx_ring
, r_vec
, idx
);
1754 r_vec
->rx_ring
= &nn
->rx_rings
[idx
];
1755 nfp_net_rx_ring_init(r_vec
->rx_ring
, r_vec
, idx
);
1757 snprintf(r_vec
->name
, sizeof(r_vec
->name
),
1758 "%s-rxtx-%d", nn
->netdev
->name
, idx
);
1759 err
= request_irq(entry
->vector
, r_vec
->handler
, 0, r_vec
->name
, r_vec
);
1761 nn_err(nn
, "Error requesting IRQ %d\n", entry
->vector
);
1764 disable_irq(entry
->vector
);
1767 netif_napi_add(nn
->netdev
, &r_vec
->napi
,
1768 nfp_net_poll
, NAPI_POLL_WEIGHT
);
1770 irq_set_affinity_hint(entry
->vector
, &r_vec
->affinity_mask
);
1772 nn_dbg(nn
, "RV%02d: irq=%03d/%03d\n", idx
, entry
->vector
, entry
->entry
);
1778 nfp_net_cleanup_vector(struct nfp_net
*nn
, struct nfp_net_r_vector
*r_vec
)
1780 struct msix_entry
*entry
= &nn
->irq_entries
[r_vec
->irq_idx
];
1782 irq_set_affinity_hint(entry
->vector
, NULL
);
1783 netif_napi_del(&r_vec
->napi
);
1784 free_irq(entry
->vector
, r_vec
);
1788 * nfp_net_rss_write_itbl() - Write RSS indirection table to device
1789 * @nn: NFP Net device to reconfigure
1791 void nfp_net_rss_write_itbl(struct nfp_net
*nn
)
1795 for (i
= 0; i
< NFP_NET_CFG_RSS_ITBL_SZ
; i
+= 4)
1796 nn_writel(nn
, NFP_NET_CFG_RSS_ITBL
+ i
,
1797 get_unaligned_le32(nn
->rss_itbl
+ i
));
1801 * nfp_net_rss_write_key() - Write RSS hash key to device
1802 * @nn: NFP Net device to reconfigure
1804 void nfp_net_rss_write_key(struct nfp_net
*nn
)
1808 for (i
= 0; i
< NFP_NET_CFG_RSS_KEY_SZ
; i
+= 4)
1809 nn_writel(nn
, NFP_NET_CFG_RSS_KEY
+ i
,
1810 get_unaligned_le32(nn
->rss_key
+ i
));
1814 * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW
1815 * @nn: NFP Net device to reconfigure
1817 void nfp_net_coalesce_write_cfg(struct nfp_net
*nn
)
1823 /* Compute factor used to convert coalesce '_usecs' parameters to
1824 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp
1827 factor
= nn
->me_freq_mhz
/ 16;
1829 /* copy RX interrupt coalesce parameters */
1830 value
= (nn
->rx_coalesce_max_frames
<< 16) |
1831 (factor
* nn
->rx_coalesce_usecs
);
1832 for (i
= 0; i
< nn
->num_r_vecs
; i
++)
1833 nn_writel(nn
, NFP_NET_CFG_RXR_IRQ_MOD(i
), value
);
1835 /* copy TX interrupt coalesce parameters */
1836 value
= (nn
->tx_coalesce_max_frames
<< 16) |
1837 (factor
* nn
->tx_coalesce_usecs
);
1838 for (i
= 0; i
< nn
->num_r_vecs
; i
++)
1839 nn_writel(nn
, NFP_NET_CFG_TXR_IRQ_MOD(i
), value
);
1843 * nfp_net_write_mac_addr() - Write mac address to the device control BAR
1844 * @nn: NFP Net device to reconfigure
1846 * Writes the MAC address from the netdev to the device control BAR. Does not
1847 * perform the required reconfig. We do a bit of byte swapping dance because
1850 static void nfp_net_write_mac_addr(struct nfp_net
*nn
)
1852 nn_writel(nn
, NFP_NET_CFG_MACADDR
+ 0,
1853 get_unaligned_be32(nn
->netdev
->dev_addr
));
1854 /* We can't do writew for NFP-3200 compatibility */
1855 nn_writel(nn
, NFP_NET_CFG_MACADDR
+ 4,
1856 get_unaligned_be16(nn
->netdev
->dev_addr
+ 4) << 16);
1859 static void nfp_net_vec_clear_ring_data(struct nfp_net
*nn
, unsigned int idx
)
1861 nn_writeq(nn
, NFP_NET_CFG_RXR_ADDR(idx
), 0);
1862 nn_writeb(nn
, NFP_NET_CFG_RXR_SZ(idx
), 0);
1863 nn_writeb(nn
, NFP_NET_CFG_RXR_VEC(idx
), 0);
1865 nn_writeq(nn
, NFP_NET_CFG_TXR_ADDR(idx
), 0);
1866 nn_writeb(nn
, NFP_NET_CFG_TXR_SZ(idx
), 0);
1867 nn_writeb(nn
, NFP_NET_CFG_TXR_VEC(idx
), 0);
1871 * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP
1872 * @nn: NFP Net device to reconfigure
1874 static void nfp_net_clear_config_and_disable(struct nfp_net
*nn
)
1876 u32 new_ctrl
, update
;
1880 new_ctrl
= nn
->ctrl
;
1881 new_ctrl
&= ~NFP_NET_CFG_CTRL_ENABLE
;
1882 update
= NFP_NET_CFG_UPDATE_GEN
;
1883 update
|= NFP_NET_CFG_UPDATE_MSIX
;
1884 update
|= NFP_NET_CFG_UPDATE_RING
;
1886 if (nn
->cap
& NFP_NET_CFG_CTRL_RINGCFG
)
1887 new_ctrl
&= ~NFP_NET_CFG_CTRL_RINGCFG
;
1889 nn_writeq(nn
, NFP_NET_CFG_TXRS_ENABLE
, 0);
1890 nn_writeq(nn
, NFP_NET_CFG_RXRS_ENABLE
, 0);
1892 nn_writel(nn
, NFP_NET_CFG_CTRL
, new_ctrl
);
1893 err
= nfp_net_reconfig(nn
, update
);
1895 nn_err(nn
, "Could not disable device: %d\n", err
);
1897 for (r
= 0; r
< nn
->num_r_vecs
; r
++) {
1898 nfp_net_rx_ring_reset(nn
->r_vecs
[r
].rx_ring
);
1899 nfp_net_tx_ring_reset(nn
, nn
->r_vecs
[r
].tx_ring
);
1900 nfp_net_vec_clear_ring_data(nn
, r
);
1903 nn
->ctrl
= new_ctrl
;
1907 nfp_net_vec_write_ring_data(struct nfp_net
*nn
, struct nfp_net_r_vector
*r_vec
,
1910 /* Write the DMA address, size and MSI-X info to the device */
1911 nn_writeq(nn
, NFP_NET_CFG_RXR_ADDR(idx
), r_vec
->rx_ring
->dma
);
1912 nn_writeb(nn
, NFP_NET_CFG_RXR_SZ(idx
), ilog2(r_vec
->rx_ring
->cnt
));
1913 nn_writeb(nn
, NFP_NET_CFG_RXR_VEC(idx
), r_vec
->irq_idx
);
1915 nn_writeq(nn
, NFP_NET_CFG_TXR_ADDR(idx
), r_vec
->tx_ring
->dma
);
1916 nn_writeb(nn
, NFP_NET_CFG_TXR_SZ(idx
), ilog2(r_vec
->tx_ring
->cnt
));
1917 nn_writeb(nn
, NFP_NET_CFG_TXR_VEC(idx
), r_vec
->irq_idx
);
1920 static int __nfp_net_set_config_and_enable(struct nfp_net
*nn
)
1922 u32 new_ctrl
, update
= 0;
1926 new_ctrl
= nn
->ctrl
;
1928 if (nn
->cap
& NFP_NET_CFG_CTRL_RSS
) {
1929 nfp_net_rss_write_key(nn
);
1930 nfp_net_rss_write_itbl(nn
);
1931 nn_writel(nn
, NFP_NET_CFG_RSS_CTRL
, nn
->rss_cfg
);
1932 update
|= NFP_NET_CFG_UPDATE_RSS
;
1935 if (nn
->cap
& NFP_NET_CFG_CTRL_IRQMOD
) {
1936 nfp_net_coalesce_write_cfg(nn
);
1938 new_ctrl
|= NFP_NET_CFG_CTRL_IRQMOD
;
1939 update
|= NFP_NET_CFG_UPDATE_IRQMOD
;
1942 for (r
= 0; r
< nn
->num_r_vecs
; r
++)
1943 nfp_net_vec_write_ring_data(nn
, &nn
->r_vecs
[r
], r
);
1945 nn_writeq(nn
, NFP_NET_CFG_TXRS_ENABLE
, nn
->num_tx_rings
== 64 ?
1946 0xffffffffffffffffULL
: ((u64
)1 << nn
->num_tx_rings
) - 1);
1948 nn_writeq(nn
, NFP_NET_CFG_RXRS_ENABLE
, nn
->num_rx_rings
== 64 ?
1949 0xffffffffffffffffULL
: ((u64
)1 << nn
->num_rx_rings
) - 1);
1951 nfp_net_write_mac_addr(nn
);
1953 nn_writel(nn
, NFP_NET_CFG_MTU
, nn
->netdev
->mtu
);
1954 nn_writel(nn
, NFP_NET_CFG_FLBUFSZ
, nn
->fl_bufsz
);
1957 new_ctrl
|= NFP_NET_CFG_CTRL_ENABLE
;
1958 update
|= NFP_NET_CFG_UPDATE_GEN
;
1959 update
|= NFP_NET_CFG_UPDATE_MSIX
;
1960 update
|= NFP_NET_CFG_UPDATE_RING
;
1961 if (nn
->cap
& NFP_NET_CFG_CTRL_RINGCFG
)
1962 new_ctrl
|= NFP_NET_CFG_CTRL_RINGCFG
;
1964 nn_writel(nn
, NFP_NET_CFG_CTRL
, new_ctrl
);
1965 err
= nfp_net_reconfig(nn
, update
);
1967 nn
->ctrl
= new_ctrl
;
1969 for (r
= 0; r
< nn
->num_r_vecs
; r
++)
1970 nfp_net_rx_ring_fill_freelist(nn
->r_vecs
[r
].rx_ring
);
1972 /* Since reconfiguration requests while NFP is down are ignored we
1973 * have to wipe the entire VXLAN configuration and reinitialize it.
1975 if (nn
->ctrl
& NFP_NET_CFG_CTRL_VXLAN
) {
1976 memset(&nn
->vxlan_ports
, 0, sizeof(nn
->vxlan_ports
));
1977 memset(&nn
->vxlan_usecnt
, 0, sizeof(nn
->vxlan_usecnt
));
1978 udp_tunnel_get_rx_info(nn
->netdev
);
1985 * nfp_net_set_config_and_enable() - Write control BAR and enable NFP
1986 * @nn: NFP Net device to reconfigure
1988 static int nfp_net_set_config_and_enable(struct nfp_net
*nn
)
1992 err
= __nfp_net_set_config_and_enable(nn
);
1994 nfp_net_clear_config_and_disable(nn
);
2000 * nfp_net_open_stack() - Start the device from stack's perspective
2001 * @nn: NFP Net device to reconfigure
2003 static void nfp_net_open_stack(struct nfp_net
*nn
)
2007 for (r
= 0; r
< nn
->num_r_vecs
; r
++) {
2008 napi_enable(&nn
->r_vecs
[r
].napi
);
2009 enable_irq(nn
->irq_entries
[nn
->r_vecs
[r
].irq_idx
].vector
);
2012 netif_tx_wake_all_queues(nn
->netdev
);
2014 enable_irq(nn
->irq_entries
[NFP_NET_IRQ_LSC_IDX
].vector
);
2015 nfp_net_read_link_status(nn
);
2018 static int nfp_net_netdev_open(struct net_device
*netdev
)
2020 struct nfp_net
*nn
= netdev_priv(netdev
);
2023 if (nn
->ctrl
& NFP_NET_CFG_CTRL_ENABLE
) {
2024 nn_err(nn
, "Dev is already enabled: 0x%08x\n", nn
->ctrl
);
2028 /* Step 1: Allocate resources for rings and the like
2029 * - Request interrupts
2030 * - Allocate RX and TX ring resources
2031 * - Setup initial RSS table
2033 err
= nfp_net_aux_irq_request(nn
, NFP_NET_CFG_EXN
, "%s-exn",
2034 nn
->exn_name
, sizeof(nn
->exn_name
),
2035 NFP_NET_IRQ_EXN_IDX
, nn
->exn_handler
);
2038 err
= nfp_net_aux_irq_request(nn
, NFP_NET_CFG_LSC
, "%s-lsc",
2039 nn
->lsc_name
, sizeof(nn
->lsc_name
),
2040 NFP_NET_IRQ_LSC_IDX
, nn
->lsc_handler
);
2043 disable_irq(nn
->irq_entries
[NFP_NET_IRQ_LSC_IDX
].vector
);
2045 nn
->rx_rings
= kcalloc(nn
->num_rx_rings
, sizeof(*nn
->rx_rings
),
2049 nn
->tx_rings
= kcalloc(nn
->num_tx_rings
, sizeof(*nn
->tx_rings
),
2052 goto err_free_rx_rings
;
2054 for (r
= 0; r
< nn
->num_r_vecs
; r
++) {
2055 err
= nfp_net_prepare_vector(nn
, &nn
->r_vecs
[r
], r
);
2057 goto err_free_prev_vecs
;
2059 err
= nfp_net_tx_ring_alloc(nn
->r_vecs
[r
].tx_ring
, nn
->txd_cnt
);
2061 goto err_cleanup_vec_p
;
2063 err
= nfp_net_rx_ring_alloc(nn
->r_vecs
[r
].rx_ring
,
2064 nn
->fl_bufsz
, nn
->rxd_cnt
);
2066 goto err_free_tx_ring_p
;
2068 err
= nfp_net_rx_ring_bufs_alloc(nn
, nn
->r_vecs
[r
].rx_ring
);
2070 goto err_flush_rx_ring_p
;
2073 err
= netif_set_real_num_tx_queues(netdev
, nn
->num_tx_rings
);
2075 goto err_free_rings
;
2077 err
= netif_set_real_num_rx_queues(netdev
, nn
->num_rx_rings
);
2079 goto err_free_rings
;
2081 /* Step 2: Configure the NFP
2082 * - Enable rings from 0 to tx_rings/rx_rings - 1.
2083 * - Write MAC address (in case it changed)
2085 * - Set the Freelist buffer size
2088 err
= nfp_net_set_config_and_enable(nn
);
2090 goto err_free_rings
;
2092 /* Step 3: Enable for kernel
2093 * - put some freelist descriptors on each RX ring
2094 * - enable NAPI on each ring
2095 * - enable all TX queues
2098 nfp_net_open_stack(nn
);
2106 nfp_net_rx_ring_bufs_free(nn
, nn
->r_vecs
[r
].rx_ring
);
2107 err_flush_rx_ring_p
:
2108 nfp_net_rx_ring_free(nn
->r_vecs
[r
].rx_ring
);
2110 nfp_net_tx_ring_free(nn
->r_vecs
[r
].tx_ring
);
2112 nfp_net_cleanup_vector(nn
, &nn
->r_vecs
[r
]);
2114 kfree(nn
->tx_rings
);
2116 kfree(nn
->rx_rings
);
2118 nfp_net_aux_irq_free(nn
, NFP_NET_CFG_LSC
, NFP_NET_IRQ_LSC_IDX
);
2120 nfp_net_aux_irq_free(nn
, NFP_NET_CFG_EXN
, NFP_NET_IRQ_EXN_IDX
);
2125 * nfp_net_close_stack() - Quiescent the stack (part of close)
2126 * @nn: NFP Net device to reconfigure
2128 static void nfp_net_close_stack(struct nfp_net
*nn
)
2132 disable_irq(nn
->irq_entries
[NFP_NET_IRQ_LSC_IDX
].vector
);
2133 netif_carrier_off(nn
->netdev
);
2134 nn
->link_up
= false;
2136 for (r
= 0; r
< nn
->num_r_vecs
; r
++) {
2137 disable_irq(nn
->irq_entries
[nn
->r_vecs
[r
].irq_idx
].vector
);
2138 napi_disable(&nn
->r_vecs
[r
].napi
);
2141 netif_tx_disable(nn
->netdev
);
2145 * nfp_net_close_free_all() - Free all runtime resources
2146 * @nn: NFP Net device to reconfigure
2148 static void nfp_net_close_free_all(struct nfp_net
*nn
)
2152 for (r
= 0; r
< nn
->num_r_vecs
; r
++) {
2153 nfp_net_rx_ring_bufs_free(nn
, nn
->r_vecs
[r
].rx_ring
);
2154 nfp_net_rx_ring_free(nn
->r_vecs
[r
].rx_ring
);
2155 nfp_net_tx_ring_free(nn
->r_vecs
[r
].tx_ring
);
2156 nfp_net_cleanup_vector(nn
, &nn
->r_vecs
[r
]);
2159 kfree(nn
->rx_rings
);
2160 kfree(nn
->tx_rings
);
2162 nfp_net_aux_irq_free(nn
, NFP_NET_CFG_LSC
, NFP_NET_IRQ_LSC_IDX
);
2163 nfp_net_aux_irq_free(nn
, NFP_NET_CFG_EXN
, NFP_NET_IRQ_EXN_IDX
);
2167 * nfp_net_netdev_close() - Called when the device is downed
2168 * @netdev: netdev structure
2170 static int nfp_net_netdev_close(struct net_device
*netdev
)
2172 struct nfp_net
*nn
= netdev_priv(netdev
);
2174 if (!(nn
->ctrl
& NFP_NET_CFG_CTRL_ENABLE
)) {
2175 nn_err(nn
, "Dev is not up: 0x%08x\n", nn
->ctrl
);
2179 /* Step 1: Disable RX and TX rings from the Linux kernel perspective
2181 nfp_net_close_stack(nn
);
2185 nfp_net_clear_config_and_disable(nn
);
2187 /* Step 3: Free resources
2189 nfp_net_close_free_all(nn
);
2191 nn_dbg(nn
, "%s down", netdev
->name
);
2195 static void nfp_net_set_rx_mode(struct net_device
*netdev
)
2197 struct nfp_net
*nn
= netdev_priv(netdev
);
2200 new_ctrl
= nn
->ctrl
;
2202 if (netdev
->flags
& IFF_PROMISC
) {
2203 if (nn
->cap
& NFP_NET_CFG_CTRL_PROMISC
)
2204 new_ctrl
|= NFP_NET_CFG_CTRL_PROMISC
;
2206 nn_warn(nn
, "FW does not support promiscuous mode\n");
2208 new_ctrl
&= ~NFP_NET_CFG_CTRL_PROMISC
;
2211 if (new_ctrl
== nn
->ctrl
)
2214 nn_writel(nn
, NFP_NET_CFG_CTRL
, new_ctrl
);
2215 nfp_net_reconfig_post(nn
, NFP_NET_CFG_UPDATE_GEN
);
2217 nn
->ctrl
= new_ctrl
;
2220 static int nfp_net_change_mtu(struct net_device
*netdev
, int new_mtu
)
2222 unsigned int old_mtu
, old_fl_bufsz
, new_fl_bufsz
;
2223 struct nfp_net
*nn
= netdev_priv(netdev
);
2224 struct nfp_net_rx_ring
*tmp_rings
;
2227 if (new_mtu
< 68 || new_mtu
> nn
->max_mtu
) {
2228 nn_err(nn
, "New MTU (%d) is not valid\n", new_mtu
);
2232 old_mtu
= netdev
->mtu
;
2233 old_fl_bufsz
= nn
->fl_bufsz
;
2234 new_fl_bufsz
= NFP_NET_MAX_PREPEND
+ ETH_HLEN
+ VLAN_HLEN
* 2 + new_mtu
;
2236 if (!netif_running(netdev
)) {
2237 netdev
->mtu
= new_mtu
;
2238 nn
->fl_bufsz
= new_fl_bufsz
;
2242 /* Prepare new rings */
2243 tmp_rings
= nfp_net_shadow_rx_rings_prepare(nn
, new_fl_bufsz
,
2248 /* Stop device, swap in new rings, try to start the firmware */
2249 nfp_net_close_stack(nn
);
2250 nfp_net_clear_config_and_disable(nn
);
2252 tmp_rings
= nfp_net_shadow_rx_rings_swap(nn
, tmp_rings
);
2254 netdev
->mtu
= new_mtu
;
2255 nn
->fl_bufsz
= new_fl_bufsz
;
2257 err
= nfp_net_set_config_and_enable(nn
);
2259 const int err_new
= err
;
2261 /* Try with old configuration and old rings */
2262 tmp_rings
= nfp_net_shadow_rx_rings_swap(nn
, tmp_rings
);
2264 netdev
->mtu
= old_mtu
;
2265 nn
->fl_bufsz
= old_fl_bufsz
;
2267 err
= __nfp_net_set_config_and_enable(nn
);
2269 nn_err(nn
, "Can't restore MTU - FW communication failed (%d,%d)\n",
2273 nfp_net_shadow_rx_rings_free(nn
, tmp_rings
);
2275 nfp_net_open_stack(nn
);
2280 int nfp_net_set_ring_size(struct nfp_net
*nn
, u32 rxd_cnt
, u32 txd_cnt
)
2282 struct nfp_net_tx_ring
*tx_rings
= NULL
;
2283 struct nfp_net_rx_ring
*rx_rings
= NULL
;
2284 u32 old_rxd_cnt
, old_txd_cnt
;
2287 if (!netif_running(nn
->netdev
)) {
2288 nn
->rxd_cnt
= rxd_cnt
;
2289 nn
->txd_cnt
= txd_cnt
;
2293 old_rxd_cnt
= nn
->rxd_cnt
;
2294 old_txd_cnt
= nn
->txd_cnt
;
2296 /* Prepare new rings */
2297 if (nn
->rxd_cnt
!= rxd_cnt
) {
2298 rx_rings
= nfp_net_shadow_rx_rings_prepare(nn
, nn
->fl_bufsz
,
2303 if (nn
->txd_cnt
!= txd_cnt
) {
2304 tx_rings
= nfp_net_shadow_tx_rings_prepare(nn
, txd_cnt
);
2306 nfp_net_shadow_rx_rings_free(nn
, rx_rings
);
2311 /* Stop device, swap in new rings, try to start the firmware */
2312 nfp_net_close_stack(nn
);
2313 nfp_net_clear_config_and_disable(nn
);
2316 rx_rings
= nfp_net_shadow_rx_rings_swap(nn
, rx_rings
);
2318 tx_rings
= nfp_net_shadow_tx_rings_swap(nn
, tx_rings
);
2320 nn
->rxd_cnt
= rxd_cnt
;
2321 nn
->txd_cnt
= txd_cnt
;
2323 err
= nfp_net_set_config_and_enable(nn
);
2325 const int err_new
= err
;
2327 /* Try with old configuration and old rings */
2329 rx_rings
= nfp_net_shadow_rx_rings_swap(nn
, rx_rings
);
2331 tx_rings
= nfp_net_shadow_tx_rings_swap(nn
, tx_rings
);
2333 nn
->rxd_cnt
= old_rxd_cnt
;
2334 nn
->txd_cnt
= old_txd_cnt
;
2336 err
= __nfp_net_set_config_and_enable(nn
);
2338 nn_err(nn
, "Can't restore ring config - FW communication failed (%d,%d)\n",
2342 nfp_net_shadow_rx_rings_free(nn
, rx_rings
);
2343 nfp_net_shadow_tx_rings_free(nn
, tx_rings
);
2345 nfp_net_open_stack(nn
);
2350 static struct rtnl_link_stats64
*nfp_net_stat64(struct net_device
*netdev
,
2351 struct rtnl_link_stats64
*stats
)
2353 struct nfp_net
*nn
= netdev_priv(netdev
);
2356 for (r
= 0; r
< nn
->num_r_vecs
; r
++) {
2357 struct nfp_net_r_vector
*r_vec
= &nn
->r_vecs
[r
];
2362 start
= u64_stats_fetch_begin(&r_vec
->rx_sync
);
2363 data
[0] = r_vec
->rx_pkts
;
2364 data
[1] = r_vec
->rx_bytes
;
2365 data
[2] = r_vec
->rx_drops
;
2366 } while (u64_stats_fetch_retry(&r_vec
->rx_sync
, start
));
2367 stats
->rx_packets
+= data
[0];
2368 stats
->rx_bytes
+= data
[1];
2369 stats
->rx_dropped
+= data
[2];
2372 start
= u64_stats_fetch_begin(&r_vec
->tx_sync
);
2373 data
[0] = r_vec
->tx_pkts
;
2374 data
[1] = r_vec
->tx_bytes
;
2375 data
[2] = r_vec
->tx_errors
;
2376 } while (u64_stats_fetch_retry(&r_vec
->tx_sync
, start
));
2377 stats
->tx_packets
+= data
[0];
2378 stats
->tx_bytes
+= data
[1];
2379 stats
->tx_errors
+= data
[2];
2385 static int nfp_net_set_features(struct net_device
*netdev
,
2386 netdev_features_t features
)
2388 netdev_features_t changed
= netdev
->features
^ features
;
2389 struct nfp_net
*nn
= netdev_priv(netdev
);
2393 /* Assume this is not called with features we have not advertised */
2395 new_ctrl
= nn
->ctrl
;
2397 if (changed
& NETIF_F_RXCSUM
) {
2398 if (features
& NETIF_F_RXCSUM
)
2399 new_ctrl
|= NFP_NET_CFG_CTRL_RXCSUM
;
2401 new_ctrl
&= ~NFP_NET_CFG_CTRL_RXCSUM
;
2404 if (changed
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
)) {
2405 if (features
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
))
2406 new_ctrl
|= NFP_NET_CFG_CTRL_TXCSUM
;
2408 new_ctrl
&= ~NFP_NET_CFG_CTRL_TXCSUM
;
2411 if (changed
& (NETIF_F_TSO
| NETIF_F_TSO6
)) {
2412 if (features
& (NETIF_F_TSO
| NETIF_F_TSO6
))
2413 new_ctrl
|= NFP_NET_CFG_CTRL_LSO
;
2415 new_ctrl
&= ~NFP_NET_CFG_CTRL_LSO
;
2418 if (changed
& NETIF_F_HW_VLAN_CTAG_RX
) {
2419 if (features
& NETIF_F_HW_VLAN_CTAG_RX
)
2420 new_ctrl
|= NFP_NET_CFG_CTRL_RXVLAN
;
2422 new_ctrl
&= ~NFP_NET_CFG_CTRL_RXVLAN
;
2425 if (changed
& NETIF_F_HW_VLAN_CTAG_TX
) {
2426 if (features
& NETIF_F_HW_VLAN_CTAG_TX
)
2427 new_ctrl
|= NFP_NET_CFG_CTRL_TXVLAN
;
2429 new_ctrl
&= ~NFP_NET_CFG_CTRL_TXVLAN
;
2432 if (changed
& NETIF_F_SG
) {
2433 if (features
& NETIF_F_SG
)
2434 new_ctrl
|= NFP_NET_CFG_CTRL_GATHER
;
2436 new_ctrl
&= ~NFP_NET_CFG_CTRL_GATHER
;
2439 nn_dbg(nn
, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n",
2440 netdev
->features
, features
, changed
);
2442 if (new_ctrl
== nn
->ctrl
)
2445 nn_dbg(nn
, "NIC ctrl: 0x%x -> 0x%x\n", nn
->ctrl
, new_ctrl
);
2446 nn_writel(nn
, NFP_NET_CFG_CTRL
, new_ctrl
);
2447 err
= nfp_net_reconfig(nn
, NFP_NET_CFG_UPDATE_GEN
);
2451 nn
->ctrl
= new_ctrl
;
2456 static netdev_features_t
2457 nfp_net_features_check(struct sk_buff
*skb
, struct net_device
*dev
,
2458 netdev_features_t features
)
2462 /* We can't do TSO over double tagged packets (802.1AD) */
2463 features
&= vlan_features_check(skb
, features
);
2465 if (!skb
->encapsulation
)
2468 /* Ensure that inner L4 header offset fits into TX descriptor field */
2469 if (skb_is_gso(skb
)) {
2472 hdrlen
= skb_inner_transport_header(skb
) - skb
->data
+
2473 inner_tcp_hdrlen(skb
);
2475 if (unlikely(hdrlen
> NFP_NET_LSO_MAX_HDR_SZ
))
2476 features
&= ~NETIF_F_GSO_MASK
;
2479 /* VXLAN/GRE check */
2480 switch (vlan_get_protocol(skb
)) {
2481 case htons(ETH_P_IP
):
2482 l4_hdr
= ip_hdr(skb
)->protocol
;
2484 case htons(ETH_P_IPV6
):
2485 l4_hdr
= ipv6_hdr(skb
)->nexthdr
;
2488 return features
& ~(NETIF_F_CSUM_MASK
| NETIF_F_GSO_MASK
);
2491 if (skb
->inner_protocol_type
!= ENCAP_TYPE_ETHER
||
2492 skb
->inner_protocol
!= htons(ETH_P_TEB
) ||
2493 (l4_hdr
!= IPPROTO_UDP
&& l4_hdr
!= IPPROTO_GRE
) ||
2494 (l4_hdr
== IPPROTO_UDP
&&
2495 (skb_inner_mac_header(skb
) - skb_transport_header(skb
) !=
2496 sizeof(struct udphdr
) + sizeof(struct vxlanhdr
))))
2497 return features
& ~(NETIF_F_CSUM_MASK
| NETIF_F_GSO_MASK
);
2503 * nfp_net_set_vxlan_port() - set vxlan port in SW and reconfigure HW
2504 * @nn: NFP Net device to reconfigure
2505 * @idx: Index into the port table where new port should be written
2506 * @port: UDP port to configure (pass zero to remove VXLAN port)
2508 static void nfp_net_set_vxlan_port(struct nfp_net
*nn
, int idx
, __be16 port
)
2512 nn
->vxlan_ports
[idx
] = port
;
2514 if (!(nn
->ctrl
& NFP_NET_CFG_CTRL_VXLAN
))
2517 BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS
& 1);
2518 for (i
= 0; i
< NFP_NET_N_VXLAN_PORTS
; i
+= 2)
2519 nn_writel(nn
, NFP_NET_CFG_VXLAN_PORT
+ i
* sizeof(port
),
2520 be16_to_cpu(nn
->vxlan_ports
[i
+ 1]) << 16 |
2521 be16_to_cpu(nn
->vxlan_ports
[i
]));
2523 nfp_net_reconfig_post(nn
, NFP_NET_CFG_UPDATE_VXLAN
);
2527 * nfp_net_find_vxlan_idx() - find table entry of the port or a free one
2528 * @nn: NFP Network structure
2529 * @port: UDP port to look for
2531 * Return: if the port is already in the table -- it's position;
2532 * if the port is not in the table -- free position to use;
2533 * if the table is full -- -ENOSPC.
2535 static int nfp_net_find_vxlan_idx(struct nfp_net
*nn
, __be16 port
)
2537 int i
, free_idx
= -ENOSPC
;
2539 for (i
= 0; i
< NFP_NET_N_VXLAN_PORTS
; i
++) {
2540 if (nn
->vxlan_ports
[i
] == port
)
2542 if (!nn
->vxlan_usecnt
[i
])
2549 static void nfp_net_add_vxlan_port(struct net_device
*netdev
,
2550 struct udp_tunnel_info
*ti
)
2552 struct nfp_net
*nn
= netdev_priv(netdev
);
2555 if (ti
->type
!= UDP_TUNNEL_TYPE_VXLAN
)
2558 idx
= nfp_net_find_vxlan_idx(nn
, ti
->port
);
2562 if (!nn
->vxlan_usecnt
[idx
]++)
2563 nfp_net_set_vxlan_port(nn
, idx
, ti
->port
);
2566 static void nfp_net_del_vxlan_port(struct net_device
*netdev
,
2567 struct udp_tunnel_info
*ti
)
2569 struct nfp_net
*nn
= netdev_priv(netdev
);
2572 if (ti
->type
!= UDP_TUNNEL_TYPE_VXLAN
)
2575 idx
= nfp_net_find_vxlan_idx(nn
, ti
->port
);
2576 if (idx
== -ENOSPC
|| !nn
->vxlan_usecnt
[idx
])
2579 if (!--nn
->vxlan_usecnt
[idx
])
2580 nfp_net_set_vxlan_port(nn
, idx
, 0);
2583 static const struct net_device_ops nfp_net_netdev_ops
= {
2584 .ndo_open
= nfp_net_netdev_open
,
2585 .ndo_stop
= nfp_net_netdev_close
,
2586 .ndo_start_xmit
= nfp_net_tx
,
2587 .ndo_get_stats64
= nfp_net_stat64
,
2588 .ndo_tx_timeout
= nfp_net_tx_timeout
,
2589 .ndo_set_rx_mode
= nfp_net_set_rx_mode
,
2590 .ndo_change_mtu
= nfp_net_change_mtu
,
2591 .ndo_set_mac_address
= eth_mac_addr
,
2592 .ndo_set_features
= nfp_net_set_features
,
2593 .ndo_features_check
= nfp_net_features_check
,
2594 .ndo_udp_tunnel_add
= nfp_net_add_vxlan_port
,
2595 .ndo_udp_tunnel_del
= nfp_net_del_vxlan_port
,
2599 * nfp_net_info() - Print general info about the NIC
2600 * @nn: NFP Net device to reconfigure
2602 void nfp_net_info(struct nfp_net
*nn
)
2604 nn_info(nn
, "Netronome %s %sNetdev: TxQs=%d/%d RxQs=%d/%d\n",
2605 nn
->is_nfp3200
? "NFP-32xx" : "NFP-6xxx",
2606 nn
->is_vf
? "VF " : "",
2607 nn
->num_tx_rings
, nn
->max_tx_rings
,
2608 nn
->num_rx_rings
, nn
->max_rx_rings
);
2609 nn_info(nn
, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n",
2610 nn
->fw_ver
.resv
, nn
->fw_ver
.class,
2611 nn
->fw_ver
.major
, nn
->fw_ver
.minor
,
2613 nn_info(nn
, "CAP: %#x %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2615 nn
->cap
& NFP_NET_CFG_CTRL_PROMISC
? "PROMISC " : "",
2616 nn
->cap
& NFP_NET_CFG_CTRL_L2BC
? "L2BCFILT " : "",
2617 nn
->cap
& NFP_NET_CFG_CTRL_L2MC
? "L2MCFILT " : "",
2618 nn
->cap
& NFP_NET_CFG_CTRL_RXCSUM
? "RXCSUM " : "",
2619 nn
->cap
& NFP_NET_CFG_CTRL_TXCSUM
? "TXCSUM " : "",
2620 nn
->cap
& NFP_NET_CFG_CTRL_RXVLAN
? "RXVLAN " : "",
2621 nn
->cap
& NFP_NET_CFG_CTRL_TXVLAN
? "TXVLAN " : "",
2622 nn
->cap
& NFP_NET_CFG_CTRL_SCATTER
? "SCATTER " : "",
2623 nn
->cap
& NFP_NET_CFG_CTRL_GATHER
? "GATHER " : "",
2624 nn
->cap
& NFP_NET_CFG_CTRL_LSO
? "TSO " : "",
2625 nn
->cap
& NFP_NET_CFG_CTRL_RSS
? "RSS " : "",
2626 nn
->cap
& NFP_NET_CFG_CTRL_L2SWITCH
? "L2SWITCH " : "",
2627 nn
->cap
& NFP_NET_CFG_CTRL_MSIXAUTO
? "AUTOMASK " : "",
2628 nn
->cap
& NFP_NET_CFG_CTRL_IRQMOD
? "IRQMOD " : "",
2629 nn
->cap
& NFP_NET_CFG_CTRL_VXLAN
? "VXLAN " : "",
2630 nn
->cap
& NFP_NET_CFG_CTRL_NVGRE
? "NVGRE " : "");
2634 * nfp_net_netdev_alloc() - Allocate netdev and related structure
2636 * @max_tx_rings: Maximum number of TX rings supported by device
2637 * @max_rx_rings: Maximum number of RX rings supported by device
2639 * This function allocates a netdev device and fills in the initial
2640 * part of the @struct nfp_net structure.
2642 * Return: NFP Net device structure, or ERR_PTR on error.
2644 struct nfp_net
*nfp_net_netdev_alloc(struct pci_dev
*pdev
,
2645 int max_tx_rings
, int max_rx_rings
)
2647 struct net_device
*netdev
;
2651 netdev
= alloc_etherdev_mqs(sizeof(struct nfp_net
),
2652 max_tx_rings
, max_rx_rings
);
2654 return ERR_PTR(-ENOMEM
);
2656 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
2657 nn
= netdev_priv(netdev
);
2659 nn
->netdev
= netdev
;
2662 nn
->max_tx_rings
= max_tx_rings
;
2663 nn
->max_rx_rings
= max_rx_rings
;
2665 nqs
= netif_get_num_default_rss_queues();
2666 nn
->num_tx_rings
= min_t(int, nqs
, max_tx_rings
);
2667 nn
->num_rx_rings
= min_t(int, nqs
, max_rx_rings
);
2669 nn
->txd_cnt
= NFP_NET_TX_DESCS_DEFAULT
;
2670 nn
->rxd_cnt
= NFP_NET_RX_DESCS_DEFAULT
;
2672 spin_lock_init(&nn
->reconfig_lock
);
2673 spin_lock_init(&nn
->link_status_lock
);
2675 setup_timer(&nn
->reconfig_timer
,
2676 nfp_net_reconfig_timer
, (unsigned long)nn
);
2682 * nfp_net_netdev_free() - Undo what @nfp_net_netdev_alloc() did
2683 * @nn: NFP Net device to reconfigure
2685 void nfp_net_netdev_free(struct nfp_net
*nn
)
2687 free_netdev(nn
->netdev
);
2691 * nfp_net_rss_init() - Set the initial RSS parameters
2692 * @nn: NFP Net device to reconfigure
2694 static void nfp_net_rss_init(struct nfp_net
*nn
)
2698 netdev_rss_key_fill(nn
->rss_key
, NFP_NET_CFG_RSS_KEY_SZ
);
2700 for (i
= 0; i
< sizeof(nn
->rss_itbl
); i
++)
2702 ethtool_rxfh_indir_default(i
, nn
->num_rx_rings
);
2704 /* Enable IPv4/IPv6 TCP by default */
2705 nn
->rss_cfg
= NFP_NET_CFG_RSS_IPV4_TCP
|
2706 NFP_NET_CFG_RSS_IPV6_TCP
|
2707 NFP_NET_CFG_RSS_TOEPLITZ
|
2708 NFP_NET_CFG_RSS_MASK
;
2712 * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters
2713 * @nn: NFP Net device to reconfigure
2715 static void nfp_net_irqmod_init(struct nfp_net
*nn
)
2717 nn
->rx_coalesce_usecs
= 50;
2718 nn
->rx_coalesce_max_frames
= 64;
2719 nn
->tx_coalesce_usecs
= 50;
2720 nn
->tx_coalesce_max_frames
= 64;
2724 * nfp_net_netdev_init() - Initialise/finalise the netdev structure
2725 * @netdev: netdev structure
2727 * Return: 0 on success or negative errno on error.
2729 int nfp_net_netdev_init(struct net_device
*netdev
)
2731 struct nfp_net
*nn
= netdev_priv(netdev
);
2734 /* Get some of the read-only fields from the BAR */
2735 nn
->cap
= nn_readl(nn
, NFP_NET_CFG_CAP
);
2736 nn
->max_mtu
= nn_readl(nn
, NFP_NET_CFG_MAX_MTU
);
2738 nfp_net_write_mac_addr(nn
);
2740 /* Set default MTU and Freelist buffer size */
2741 if (nn
->max_mtu
< NFP_NET_DEFAULT_MTU
)
2742 netdev
->mtu
= nn
->max_mtu
;
2744 netdev
->mtu
= NFP_NET_DEFAULT_MTU
;
2745 nn
->fl_bufsz
= NFP_NET_DEFAULT_RX_BUFSZ
;
2747 /* Advertise/enable offloads based on capabilities
2749 * Note: netdev->features show the currently enabled features
2750 * and netdev->hw_features advertises which features are
2751 * supported. By default we enable most features.
2753 netdev
->hw_features
= NETIF_F_HIGHDMA
;
2754 if (nn
->cap
& NFP_NET_CFG_CTRL_RXCSUM
) {
2755 netdev
->hw_features
|= NETIF_F_RXCSUM
;
2756 nn
->ctrl
|= NFP_NET_CFG_CTRL_RXCSUM
;
2758 if (nn
->cap
& NFP_NET_CFG_CTRL_TXCSUM
) {
2759 netdev
->hw_features
|= NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
;
2760 nn
->ctrl
|= NFP_NET_CFG_CTRL_TXCSUM
;
2762 if (nn
->cap
& NFP_NET_CFG_CTRL_GATHER
) {
2763 netdev
->hw_features
|= NETIF_F_SG
;
2764 nn
->ctrl
|= NFP_NET_CFG_CTRL_GATHER
;
2766 if ((nn
->cap
& NFP_NET_CFG_CTRL_LSO
) && nn
->fw_ver
.major
> 2) {
2767 netdev
->hw_features
|= NETIF_F_TSO
| NETIF_F_TSO6
;
2768 nn
->ctrl
|= NFP_NET_CFG_CTRL_LSO
;
2770 if (nn
->cap
& NFP_NET_CFG_CTRL_RSS
) {
2771 netdev
->hw_features
|= NETIF_F_RXHASH
;
2772 nfp_net_rss_init(nn
);
2773 nn
->ctrl
|= NFP_NET_CFG_CTRL_RSS
;
2775 if (nn
->cap
& NFP_NET_CFG_CTRL_VXLAN
&&
2776 nn
->cap
& NFP_NET_CFG_CTRL_NVGRE
) {
2777 if (nn
->cap
& NFP_NET_CFG_CTRL_LSO
)
2778 netdev
->hw_features
|= NETIF_F_GSO_GRE
|
2779 NETIF_F_GSO_UDP_TUNNEL
;
2780 nn
->ctrl
|= NFP_NET_CFG_CTRL_VXLAN
| NFP_NET_CFG_CTRL_NVGRE
;
2782 netdev
->hw_enc_features
= netdev
->hw_features
;
2785 netdev
->vlan_features
= netdev
->hw_features
;
2787 if (nn
->cap
& NFP_NET_CFG_CTRL_RXVLAN
) {
2788 netdev
->hw_features
|= NETIF_F_HW_VLAN_CTAG_RX
;
2789 nn
->ctrl
|= NFP_NET_CFG_CTRL_RXVLAN
;
2791 if (nn
->cap
& NFP_NET_CFG_CTRL_TXVLAN
) {
2792 netdev
->hw_features
|= NETIF_F_HW_VLAN_CTAG_TX
;
2793 nn
->ctrl
|= NFP_NET_CFG_CTRL_TXVLAN
;
2796 netdev
->features
= netdev
->hw_features
;
2798 /* Advertise but disable TSO by default. */
2799 netdev
->features
&= ~(NETIF_F_TSO
| NETIF_F_TSO6
);
2801 /* Allow L2 Broadcast and Multicast through by default, if supported */
2802 if (nn
->cap
& NFP_NET_CFG_CTRL_L2BC
)
2803 nn
->ctrl
|= NFP_NET_CFG_CTRL_L2BC
;
2804 if (nn
->cap
& NFP_NET_CFG_CTRL_L2MC
)
2805 nn
->ctrl
|= NFP_NET_CFG_CTRL_L2MC
;
2807 /* Allow IRQ moderation, if supported */
2808 if (nn
->cap
& NFP_NET_CFG_CTRL_IRQMOD
) {
2809 nfp_net_irqmod_init(nn
);
2810 nn
->ctrl
|= NFP_NET_CFG_CTRL_IRQMOD
;
2813 /* On NFP-3200 enable MSI-X auto-masking, if supported and the
2814 * interrupts are not shared.
2816 if (nn
->is_nfp3200
&& nn
->cap
& NFP_NET_CFG_CTRL_MSIXAUTO
)
2817 nn
->ctrl
|= NFP_NET_CFG_CTRL_MSIXAUTO
;
2819 /* On NFP4000/NFP6000, determine RX packet/metadata boundary offset */
2820 if (nn
->fw_ver
.major
>= 2)
2821 nn
->rx_offset
= nn_readl(nn
, NFP_NET_CFG_RX_OFFSET
);
2823 nn
->rx_offset
= NFP_NET_RX_OFFSET
;
2825 /* Stash the re-configuration queue away. First odd queue in TX Bar */
2826 nn
->qcp_cfg
= nn
->tx_bar
+ NFP_QCP_QUEUE_ADDR_SZ
;
2828 /* Make sure the FW knows the netdev is supposed to be disabled here */
2829 nn_writel(nn
, NFP_NET_CFG_CTRL
, 0);
2830 nn_writeq(nn
, NFP_NET_CFG_TXRS_ENABLE
, 0);
2831 nn_writeq(nn
, NFP_NET_CFG_RXRS_ENABLE
, 0);
2832 err
= nfp_net_reconfig(nn
, NFP_NET_CFG_UPDATE_RING
|
2833 NFP_NET_CFG_UPDATE_GEN
);
2837 /* Finalise the netdev setup */
2838 ether_setup(netdev
);
2839 netdev
->netdev_ops
= &nfp_net_netdev_ops
;
2840 netdev
->watchdog_timeo
= msecs_to_jiffies(5 * 1000);
2841 netif_carrier_off(netdev
);
2843 nfp_net_set_ethtool_ops(netdev
);
2844 nfp_net_irqs_assign(netdev
);
2846 return register_netdev(netdev
);
2850 * nfp_net_netdev_clean() - Undo what nfp_net_netdev_init() did.
2851 * @netdev: netdev structure
2853 void nfp_net_netdev_clean(struct net_device
*netdev
)
2855 unregister_netdev(netdev
);