1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2015 Cavium, Inc.
6 #include <linux/module.h>
7 #include <linux/interrupt.h>
9 #include <linux/netdevice.h>
10 #include <linux/if_vlan.h>
11 #include <linux/etherdevice.h>
12 #include <linux/ethtool.h>
13 #include <linux/log2.h>
14 #include <linux/prefetch.h>
15 #include <linux/irq.h>
16 #include <linux/iommu.h>
17 #include <linux/bpf.h>
18 #include <linux/bpf_trace.h>
19 #include <linux/filter.h>
20 #include <linux/net_tstamp.h>
21 #include <linux/workqueue.h>
25 #include "nicvf_queues.h"
26 #include "thunder_bgx.h"
27 #include "../common/cavium_ptp.h"
29 #define DRV_NAME "nicvf"
30 #define DRV_VERSION "1.0"
32 /* NOTE: Packets bigger than 1530 are split across multiple pages and XDP needs
33 * the buffer to be contiguous. Allow XDP to be set up only if we don't exceed
34 * this value, keeping headroom for the 14 byte Ethernet header and two
35 * VLAN tags (for QinQ)
37 #define MAX_XDP_MTU (1530 - ETH_HLEN - VLAN_HLEN * 2)
39 /* Supported devices */
40 static const struct pci_device_id nicvf_id_table
[] = {
41 { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM
,
42 PCI_DEVICE_ID_THUNDER_NIC_VF
,
44 PCI_SUBSYS_DEVID_88XX_NIC_VF
) },
45 { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM
,
46 PCI_DEVICE_ID_THUNDER_PASS1_NIC_VF
,
48 PCI_SUBSYS_DEVID_88XX_PASS1_NIC_VF
) },
49 { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM
,
50 PCI_DEVICE_ID_THUNDER_NIC_VF
,
52 PCI_SUBSYS_DEVID_81XX_NIC_VF
) },
53 { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM
,
54 PCI_DEVICE_ID_THUNDER_NIC_VF
,
56 PCI_SUBSYS_DEVID_83XX_NIC_VF
) },
57 { 0, } /* end of table */
60 MODULE_AUTHOR("Sunil Goutham");
61 MODULE_DESCRIPTION("Cavium Thunder NIC Virtual Function Driver");
62 MODULE_LICENSE("GPL v2");
63 MODULE_VERSION(DRV_VERSION
);
64 MODULE_DEVICE_TABLE(pci
, nicvf_id_table
);
66 static int debug
= 0x00;
67 module_param(debug
, int, 0644);
68 MODULE_PARM_DESC(debug
, "Debug message level bitmap");
70 static int cpi_alg
= CPI_ALG_NONE
;
71 module_param(cpi_alg
, int, 0444);
72 MODULE_PARM_DESC(cpi_alg
,
73 "PFC algorithm (0=none, 1=VLAN, 2=VLAN16, 3=IP Diffserv)");
75 static inline u8
nicvf_netdev_qidx(struct nicvf
*nic
, u8 qidx
)
78 return qidx
+ ((nic
->sqs_id
+ 1) * MAX_CMP_QUEUES_PER_QS
);
83 /* The Cavium ThunderX network controller can *only* be found in SoCs
84 * containing the ThunderX ARM64 CPU implementation. All accesses to the device
85 * registers on this platform are implicitly strongly ordered with respect
86 * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
87 * with no memory barriers in this driver. The readq()/writeq() functions add
88 * explicit ordering operation which in this case are redundant, and only
92 /* Register read/write APIs */
93 void nicvf_reg_write(struct nicvf
*nic
, u64 offset
, u64 val
)
95 writeq_relaxed(val
, nic
->reg_base
+ offset
);
98 u64
nicvf_reg_read(struct nicvf
*nic
, u64 offset
)
100 return readq_relaxed(nic
->reg_base
+ offset
);
103 void nicvf_queue_reg_write(struct nicvf
*nic
, u64 offset
,
106 void __iomem
*addr
= nic
->reg_base
+ offset
;
108 writeq_relaxed(val
, addr
+ (qidx
<< NIC_Q_NUM_SHIFT
));
111 u64
nicvf_queue_reg_read(struct nicvf
*nic
, u64 offset
, u64 qidx
)
113 void __iomem
*addr
= nic
->reg_base
+ offset
;
115 return readq_relaxed(addr
+ (qidx
<< NIC_Q_NUM_SHIFT
));
118 /* VF -> PF mailbox communication */
119 static void nicvf_write_to_mbx(struct nicvf
*nic
, union nic_mbx
*mbx
)
121 u64
*msg
= (u64
*)mbx
;
123 nicvf_reg_write(nic
, NIC_VF_PF_MAILBOX_0_1
+ 0, msg
[0]);
124 nicvf_reg_write(nic
, NIC_VF_PF_MAILBOX_0_1
+ 8, msg
[1]);
127 int nicvf_send_msg_to_pf(struct nicvf
*nic
, union nic_mbx
*mbx
)
129 unsigned long timeout
;
132 mutex_lock(&nic
->rx_mode_mtx
);
134 nic
->pf_acked
= false;
135 nic
->pf_nacked
= false;
137 nicvf_write_to_mbx(nic
, mbx
);
139 timeout
= jiffies
+ msecs_to_jiffies(NIC_MBOX_MSG_TIMEOUT
);
140 /* Wait for previous message to be acked, timeout 2sec */
141 while (!nic
->pf_acked
) {
142 if (nic
->pf_nacked
) {
143 netdev_err(nic
->netdev
,
144 "PF NACK to mbox msg 0x%02x from VF%d\n",
145 (mbx
->msg
.msg
& 0xFF), nic
->vf_id
);
149 usleep_range(8000, 10000);
152 if (time_after(jiffies
, timeout
)) {
153 netdev_err(nic
->netdev
,
154 "PF didn't ACK to mbox msg 0x%02x from VF%d\n",
155 (mbx
->msg
.msg
& 0xFF), nic
->vf_id
);
160 mutex_unlock(&nic
->rx_mode_mtx
);
164 /* Checks if VF is able to comminicate with PF
165 * and also gets the VNIC number this VF is associated to.
167 static int nicvf_check_pf_ready(struct nicvf
*nic
)
169 union nic_mbx mbx
= {};
171 mbx
.msg
.msg
= NIC_MBOX_MSG_READY
;
172 if (nicvf_send_msg_to_pf(nic
, &mbx
)) {
173 netdev_err(nic
->netdev
,
174 "PF didn't respond to READY msg\n");
181 static void nicvf_send_cfg_done(struct nicvf
*nic
)
183 union nic_mbx mbx
= {};
185 mbx
.msg
.msg
= NIC_MBOX_MSG_CFG_DONE
;
186 if (nicvf_send_msg_to_pf(nic
, &mbx
)) {
187 netdev_err(nic
->netdev
,
188 "PF didn't respond to CFG DONE msg\n");
192 static void nicvf_read_bgx_stats(struct nicvf
*nic
, struct bgx_stats_msg
*bgx
)
195 nic
->bgx_stats
.rx_stats
[bgx
->idx
] = bgx
->stats
;
197 nic
->bgx_stats
.tx_stats
[bgx
->idx
] = bgx
->stats
;
200 static void nicvf_handle_mbx_intr(struct nicvf
*nic
)
202 union nic_mbx mbx
= {};
207 mbx_addr
= NIC_VF_PF_MAILBOX_0_1
;
208 mbx_data
= (u64
*)&mbx
;
210 for (i
= 0; i
< NIC_PF_VF_MAILBOX_SIZE
; i
++) {
211 *mbx_data
= nicvf_reg_read(nic
, mbx_addr
);
213 mbx_addr
+= sizeof(u64
);
216 netdev_dbg(nic
->netdev
, "Mbox message: msg: 0x%x\n", mbx
.msg
.msg
);
217 switch (mbx
.msg
.msg
) {
218 case NIC_MBOX_MSG_READY
:
219 nic
->pf_acked
= true;
220 nic
->vf_id
= mbx
.nic_cfg
.vf_id
& 0x7F;
221 nic
->tns_mode
= mbx
.nic_cfg
.tns_mode
& 0x7F;
222 nic
->node
= mbx
.nic_cfg
.node_id
;
223 if (!nic
->set_mac_pending
)
224 ether_addr_copy(nic
->netdev
->dev_addr
,
225 mbx
.nic_cfg
.mac_addr
);
226 nic
->sqs_mode
= mbx
.nic_cfg
.sqs_mode
;
227 nic
->loopback_supported
= mbx
.nic_cfg
.loopback_supported
;
228 nic
->link_up
= false;
232 case NIC_MBOX_MSG_ACK
:
233 nic
->pf_acked
= true;
235 case NIC_MBOX_MSG_NACK
:
236 nic
->pf_nacked
= true;
238 case NIC_MBOX_MSG_RSS_SIZE
:
239 nic
->rss_info
.rss_size
= mbx
.rss_size
.ind_tbl_size
;
240 nic
->pf_acked
= true;
242 case NIC_MBOX_MSG_BGX_STATS
:
243 nicvf_read_bgx_stats(nic
, &mbx
.bgx_stats
);
244 nic
->pf_acked
= true;
246 case NIC_MBOX_MSG_BGX_LINK_CHANGE
:
247 nic
->pf_acked
= true;
248 if (nic
->link_up
!= mbx
.link_status
.link_up
) {
249 nic
->link_up
= mbx
.link_status
.link_up
;
250 nic
->duplex
= mbx
.link_status
.duplex
;
251 nic
->speed
= mbx
.link_status
.speed
;
252 nic
->mac_type
= mbx
.link_status
.mac_type
;
254 netdev_info(nic
->netdev
,
255 "Link is Up %d Mbps %s duplex\n",
257 nic
->duplex
== DUPLEX_FULL
?
259 netif_carrier_on(nic
->netdev
);
260 netif_tx_start_all_queues(nic
->netdev
);
262 netdev_info(nic
->netdev
, "Link is Down\n");
263 netif_carrier_off(nic
->netdev
);
264 netif_tx_stop_all_queues(nic
->netdev
);
268 case NIC_MBOX_MSG_ALLOC_SQS
:
269 nic
->sqs_count
= mbx
.sqs_alloc
.qs_count
;
270 nic
->pf_acked
= true;
272 case NIC_MBOX_MSG_SNICVF_PTR
:
273 /* Primary VF: make note of secondary VF's pointer
274 * to be used while packet transmission.
276 nic
->snicvf
[mbx
.nicvf
.sqs_id
] =
277 (struct nicvf
*)mbx
.nicvf
.nicvf
;
278 nic
->pf_acked
= true;
280 case NIC_MBOX_MSG_PNICVF_PTR
:
281 /* Secondary VF/Qset: make note of primary VF's pointer
282 * to be used while packet reception, to handover packet
283 * to primary VF's netdev.
285 nic
->pnicvf
= (struct nicvf
*)mbx
.nicvf
.nicvf
;
286 nic
->pf_acked
= true;
288 case NIC_MBOX_MSG_PFC
:
289 nic
->pfc
.autoneg
= mbx
.pfc
.autoneg
;
290 nic
->pfc
.fc_rx
= mbx
.pfc
.fc_rx
;
291 nic
->pfc
.fc_tx
= mbx
.pfc
.fc_tx
;
292 nic
->pf_acked
= true;
295 netdev_err(nic
->netdev
,
296 "Invalid message from PF, msg 0x%x\n", mbx
.msg
.msg
);
299 nicvf_clear_intr(nic
, NICVF_INTR_MBOX
, 0);
302 static int nicvf_hw_set_mac_addr(struct nicvf
*nic
, struct net_device
*netdev
)
304 union nic_mbx mbx
= {};
306 mbx
.mac
.msg
= NIC_MBOX_MSG_SET_MAC
;
307 mbx
.mac
.vf_id
= nic
->vf_id
;
308 ether_addr_copy(mbx
.mac
.mac_addr
, netdev
->dev_addr
);
310 return nicvf_send_msg_to_pf(nic
, &mbx
);
313 static void nicvf_config_cpi(struct nicvf
*nic
)
315 union nic_mbx mbx
= {};
317 mbx
.cpi_cfg
.msg
= NIC_MBOX_MSG_CPI_CFG
;
318 mbx
.cpi_cfg
.vf_id
= nic
->vf_id
;
319 mbx
.cpi_cfg
.cpi_alg
= nic
->cpi_alg
;
320 mbx
.cpi_cfg
.rq_cnt
= nic
->qs
->rq_cnt
;
322 nicvf_send_msg_to_pf(nic
, &mbx
);
325 static void nicvf_get_rss_size(struct nicvf
*nic
)
327 union nic_mbx mbx
= {};
329 mbx
.rss_size
.msg
= NIC_MBOX_MSG_RSS_SIZE
;
330 mbx
.rss_size
.vf_id
= nic
->vf_id
;
331 nicvf_send_msg_to_pf(nic
, &mbx
);
334 void nicvf_config_rss(struct nicvf
*nic
)
336 union nic_mbx mbx
= {};
337 struct nicvf_rss_info
*rss
= &nic
->rss_info
;
338 int ind_tbl_len
= rss
->rss_size
;
341 mbx
.rss_cfg
.vf_id
= nic
->vf_id
;
342 mbx
.rss_cfg
.hash_bits
= rss
->hash_bits
;
343 while (ind_tbl_len
) {
344 mbx
.rss_cfg
.tbl_offset
= nextq
;
345 mbx
.rss_cfg
.tbl_len
= min(ind_tbl_len
,
346 RSS_IND_TBL_LEN_PER_MBX_MSG
);
347 mbx
.rss_cfg
.msg
= mbx
.rss_cfg
.tbl_offset
?
348 NIC_MBOX_MSG_RSS_CFG_CONT
: NIC_MBOX_MSG_RSS_CFG
;
350 for (i
= 0; i
< mbx
.rss_cfg
.tbl_len
; i
++)
351 mbx
.rss_cfg
.ind_tbl
[i
] = rss
->ind_tbl
[nextq
++];
353 nicvf_send_msg_to_pf(nic
, &mbx
);
355 ind_tbl_len
-= mbx
.rss_cfg
.tbl_len
;
359 void nicvf_set_rss_key(struct nicvf
*nic
)
361 struct nicvf_rss_info
*rss
= &nic
->rss_info
;
362 u64 key_addr
= NIC_VNIC_RSS_KEY_0_4
;
365 for (idx
= 0; idx
< RSS_HASH_KEY_SIZE
; idx
++) {
366 nicvf_reg_write(nic
, key_addr
, rss
->key
[idx
]);
367 key_addr
+= sizeof(u64
);
371 static int nicvf_rss_init(struct nicvf
*nic
)
373 struct nicvf_rss_info
*rss
= &nic
->rss_info
;
376 nicvf_get_rss_size(nic
);
378 if (cpi_alg
!= CPI_ALG_NONE
) {
386 netdev_rss_key_fill(rss
->key
, RSS_HASH_KEY_SIZE
* sizeof(u64
));
387 nicvf_set_rss_key(nic
);
389 rss
->cfg
= RSS_IP_HASH_ENA
| RSS_TCP_HASH_ENA
| RSS_UDP_HASH_ENA
;
390 nicvf_reg_write(nic
, NIC_VNIC_RSS_CFG
, rss
->cfg
);
392 rss
->hash_bits
= ilog2(rounddown_pow_of_two(rss
->rss_size
));
394 for (idx
= 0; idx
< rss
->rss_size
; idx
++)
395 rss
->ind_tbl
[idx
] = ethtool_rxfh_indir_default(idx
,
397 nicvf_config_rss(nic
);
401 /* Request PF to allocate additional Qsets */
402 static void nicvf_request_sqs(struct nicvf
*nic
)
404 union nic_mbx mbx
= {};
406 int sqs_count
= nic
->sqs_count
;
407 int rx_queues
= 0, tx_queues
= 0;
409 /* Only primary VF should request */
410 if (nic
->sqs_mode
|| !nic
->sqs_count
)
413 mbx
.sqs_alloc
.msg
= NIC_MBOX_MSG_ALLOC_SQS
;
414 mbx
.sqs_alloc
.vf_id
= nic
->vf_id
;
415 mbx
.sqs_alloc
.qs_count
= nic
->sqs_count
;
416 if (nicvf_send_msg_to_pf(nic
, &mbx
)) {
417 /* No response from PF */
422 /* Return if no Secondary Qsets available */
426 if (nic
->rx_queues
> MAX_RCV_QUEUES_PER_QS
)
427 rx_queues
= nic
->rx_queues
- MAX_RCV_QUEUES_PER_QS
;
429 tx_queues
= nic
->tx_queues
+ nic
->xdp_tx_queues
;
430 if (tx_queues
> MAX_SND_QUEUES_PER_QS
)
431 tx_queues
= tx_queues
- MAX_SND_QUEUES_PER_QS
;
433 /* Set no of Rx/Tx queues in each of the SQsets */
434 for (sqs
= 0; sqs
< nic
->sqs_count
; sqs
++) {
435 mbx
.nicvf
.msg
= NIC_MBOX_MSG_SNICVF_PTR
;
436 mbx
.nicvf
.vf_id
= nic
->vf_id
;
437 mbx
.nicvf
.sqs_id
= sqs
;
438 nicvf_send_msg_to_pf(nic
, &mbx
);
440 nic
->snicvf
[sqs
]->sqs_id
= sqs
;
441 if (rx_queues
> MAX_RCV_QUEUES_PER_QS
) {
442 nic
->snicvf
[sqs
]->qs
->rq_cnt
= MAX_RCV_QUEUES_PER_QS
;
443 rx_queues
-= MAX_RCV_QUEUES_PER_QS
;
445 nic
->snicvf
[sqs
]->qs
->rq_cnt
= rx_queues
;
449 if (tx_queues
> MAX_SND_QUEUES_PER_QS
) {
450 nic
->snicvf
[sqs
]->qs
->sq_cnt
= MAX_SND_QUEUES_PER_QS
;
451 tx_queues
-= MAX_SND_QUEUES_PER_QS
;
453 nic
->snicvf
[sqs
]->qs
->sq_cnt
= tx_queues
;
457 nic
->snicvf
[sqs
]->qs
->cq_cnt
=
458 max(nic
->snicvf
[sqs
]->qs
->rq_cnt
, nic
->snicvf
[sqs
]->qs
->sq_cnt
);
460 /* Initialize secondary Qset's queues and its interrupts */
461 nicvf_open(nic
->snicvf
[sqs
]->netdev
);
464 /* Update stack with actual Rx/Tx queue count allocated */
465 if (sqs_count
!= nic
->sqs_count
)
466 nicvf_set_real_num_queues(nic
->netdev
,
467 nic
->tx_queues
, nic
->rx_queues
);
470 /* Send this Qset's nicvf pointer to PF.
471 * PF inturn sends primary VF's nicvf struct to secondary Qsets/VFs
472 * so that packets received by these Qsets can use primary VF's netdev
474 static void nicvf_send_vf_struct(struct nicvf
*nic
)
476 union nic_mbx mbx
= {};
478 mbx
.nicvf
.msg
= NIC_MBOX_MSG_NICVF_PTR
;
479 mbx
.nicvf
.sqs_mode
= nic
->sqs_mode
;
480 mbx
.nicvf
.nicvf
= (u64
)nic
;
481 nicvf_send_msg_to_pf(nic
, &mbx
);
484 static void nicvf_get_primary_vf_struct(struct nicvf
*nic
)
486 union nic_mbx mbx
= {};
488 mbx
.nicvf
.msg
= NIC_MBOX_MSG_PNICVF_PTR
;
489 nicvf_send_msg_to_pf(nic
, &mbx
);
492 int nicvf_set_real_num_queues(struct net_device
*netdev
,
493 int tx_queues
, int rx_queues
)
497 err
= netif_set_real_num_tx_queues(netdev
, tx_queues
);
500 "Failed to set no of Tx queues: %d\n", tx_queues
);
504 err
= netif_set_real_num_rx_queues(netdev
, rx_queues
);
507 "Failed to set no of Rx queues: %d\n", rx_queues
);
511 static int nicvf_init_resources(struct nicvf
*nic
)
516 nicvf_qset_config(nic
, true);
518 /* Initialize queues and HW for data transfer */
519 err
= nicvf_config_data_transfer(nic
, true);
521 netdev_err(nic
->netdev
,
522 "Failed to alloc/config VF's QSet resources\n");
529 static inline bool nicvf_xdp_rx(struct nicvf
*nic
, struct bpf_prog
*prog
,
530 struct cqe_rx_t
*cqe_rx
, struct snd_queue
*sq
,
531 struct rcv_queue
*rq
, struct sk_buff
**skb
)
533 unsigned char *hard_start
, *data
;
538 u64 dma_addr
, cpu_addr
;
541 /* Retrieve packet buffer's DMA address and length */
542 len
= *((u16
*)((void *)cqe_rx
+ (3 * sizeof(u64
))));
543 dma_addr
= *((u64
*)((void *)cqe_rx
+ (7 * sizeof(u64
))));
545 cpu_addr
= nicvf_iova_to_phys(nic
, dma_addr
);
548 cpu_addr
= (u64
)phys_to_virt(cpu_addr
);
549 page
= virt_to_page((void *)cpu_addr
);
551 xdp_init_buff(&xdp
, RCV_FRAG_LEN
+ XDP_PACKET_HEADROOM
,
553 hard_start
= page_address(page
);
554 data
= (unsigned char *)cpu_addr
;
555 xdp_prepare_buff(&xdp
, hard_start
, data
- hard_start
, len
, false);
556 orig_data
= xdp
.data
;
558 action
= bpf_prog_run_xdp(prog
, &xdp
);
560 len
= xdp
.data_end
- xdp
.data
;
561 /* Check if XDP program has changed headers */
562 if (orig_data
!= xdp
.data
) {
563 offset
= orig_data
- xdp
.data
;
569 /* Check if it's a recycled page, if not
570 * unmap the DMA mapping.
572 * Recycled page holds an extra reference.
574 if (page_ref_count(page
) == 1) {
575 dma_addr
&= PAGE_MASK
;
576 dma_unmap_page_attrs(&nic
->pdev
->dev
, dma_addr
,
577 RCV_FRAG_LEN
+ XDP_PACKET_HEADROOM
,
579 DMA_ATTR_SKIP_CPU_SYNC
);
582 /* Build SKB and pass on packet to network stack */
583 *skb
= build_skb(xdp
.data
,
584 RCV_FRAG_LEN
- cqe_rx
->align_pad
+ offset
);
591 nicvf_xdp_sq_append_pkt(nic
, sq
, (u64
)xdp
.data
, dma_addr
, len
);
594 bpf_warn_invalid_xdp_action(action
);
597 trace_xdp_exception(nic
->netdev
, prog
, action
);
600 /* Check if it's a recycled page, if not
601 * unmap the DMA mapping.
603 * Recycled page holds an extra reference.
605 if (page_ref_count(page
) == 1) {
606 dma_addr
&= PAGE_MASK
;
607 dma_unmap_page_attrs(&nic
->pdev
->dev
, dma_addr
,
608 RCV_FRAG_LEN
+ XDP_PACKET_HEADROOM
,
610 DMA_ATTR_SKIP_CPU_SYNC
);
618 static void nicvf_snd_ptp_handler(struct net_device
*netdev
,
619 struct cqe_send_t
*cqe_tx
)
621 struct nicvf
*nic
= netdev_priv(netdev
);
622 struct skb_shared_hwtstamps ts
;
627 /* Sync for 'ptp_skb' */
630 /* New timestamp request can be queued now */
631 atomic_set(&nic
->tx_ptp_skbs
, 0);
633 /* Check for timestamp requested skb */
637 /* Check if timestamping is timedout, which is set to 10us */
638 if (cqe_tx
->send_status
== CQ_TX_ERROP_TSTMP_TIMEOUT
||
639 cqe_tx
->send_status
== CQ_TX_ERROP_TSTMP_CONFLICT
)
642 /* Get the timestamp */
643 memset(&ts
, 0, sizeof(ts
));
644 ns
= cavium_ptp_tstamp2time(nic
->ptp_clock
, cqe_tx
->ptp_timestamp
);
645 ts
.hwtstamp
= ns_to_ktime(ns
);
646 skb_tstamp_tx(nic
->ptp_skb
, &ts
);
649 /* Free the original skb */
650 dev_kfree_skb_any(nic
->ptp_skb
);
656 static void nicvf_snd_pkt_handler(struct net_device
*netdev
,
657 struct cqe_send_t
*cqe_tx
,
658 int budget
, int *subdesc_cnt
,
659 unsigned int *tx_pkts
, unsigned int *tx_bytes
)
661 struct sk_buff
*skb
= NULL
;
663 struct nicvf
*nic
= netdev_priv(netdev
);
664 struct snd_queue
*sq
;
665 struct sq_hdr_subdesc
*hdr
;
666 struct sq_hdr_subdesc
*tso_sqe
;
668 sq
= &nic
->qs
->sq
[cqe_tx
->sq_idx
];
670 hdr
= (struct sq_hdr_subdesc
*)GET_SQ_DESC(sq
, cqe_tx
->sqe_ptr
);
671 if (hdr
->subdesc_type
!= SQ_DESC_TYPE_HEADER
)
674 /* Check for errors */
675 if (cqe_tx
->send_status
)
676 nicvf_check_cqe_tx_errs(nic
->pnicvf
, cqe_tx
);
678 /* Is this a XDP designated Tx queue */
680 page
= (struct page
*)sq
->xdp_page
[cqe_tx
->sqe_ptr
];
681 /* Check if it's recycled page or else unmap DMA mapping */
682 if (page
&& (page_ref_count(page
) == 1))
683 nicvf_unmap_sndq_buffers(nic
, sq
, cqe_tx
->sqe_ptr
,
686 /* Release page reference for recycling */
689 sq
->xdp_page
[cqe_tx
->sqe_ptr
] = (u64
)NULL
;
690 *subdesc_cnt
+= hdr
->subdesc_cnt
+ 1;
694 skb
= (struct sk_buff
*)sq
->skbuff
[cqe_tx
->sqe_ptr
];
696 /* Check for dummy descriptor used for HW TSO offload on 88xx */
697 if (hdr
->dont_send
) {
698 /* Get actual TSO descriptors and free them */
700 (struct sq_hdr_subdesc
*)GET_SQ_DESC(sq
, hdr
->rsvd2
);
701 nicvf_unmap_sndq_buffers(nic
, sq
, hdr
->rsvd2
,
702 tso_sqe
->subdesc_cnt
);
703 *subdesc_cnt
+= tso_sqe
->subdesc_cnt
+ 1;
705 nicvf_unmap_sndq_buffers(nic
, sq
, cqe_tx
->sqe_ptr
,
708 *subdesc_cnt
+= hdr
->subdesc_cnt
+ 1;
711 *tx_bytes
+= skb
->len
;
712 /* If timestamp is requested for this skb, don't free it */
713 if (skb_shinfo(skb
)->tx_flags
& SKBTX_IN_PROGRESS
&&
714 !nic
->pnicvf
->ptp_skb
)
715 nic
->pnicvf
->ptp_skb
= skb
;
717 napi_consume_skb(skb
, budget
);
718 sq
->skbuff
[cqe_tx
->sqe_ptr
] = (u64
)NULL
;
720 /* In case of SW TSO on 88xx, only last segment will have
721 * a SKB attached, so just free SQEs here.
724 *subdesc_cnt
+= hdr
->subdesc_cnt
+ 1;
728 static inline void nicvf_set_rxhash(struct net_device
*netdev
,
729 struct cqe_rx_t
*cqe_rx
,
735 if (!(netdev
->features
& NETIF_F_RXHASH
))
738 switch (cqe_rx
->rss_alg
) {
741 hash_type
= PKT_HASH_TYPE_L4
;
742 hash
= cqe_rx
->rss_tag
;
745 hash_type
= PKT_HASH_TYPE_L3
;
746 hash
= cqe_rx
->rss_tag
;
749 hash_type
= PKT_HASH_TYPE_NONE
;
753 skb_set_hash(skb
, hash
, hash_type
);
756 static inline void nicvf_set_rxtstamp(struct nicvf
*nic
, struct sk_buff
*skb
)
760 if (!nic
->ptp_clock
|| !nic
->hw_rx_tstamp
)
763 /* The first 8 bytes is the timestamp */
764 ns
= cavium_ptp_tstamp2time(nic
->ptp_clock
,
765 be64_to_cpu(*(__be64
*)skb
->data
));
766 skb_hwtstamps(skb
)->hwtstamp
= ns_to_ktime(ns
);
771 static void nicvf_rcv_pkt_handler(struct net_device
*netdev
,
772 struct napi_struct
*napi
,
773 struct cqe_rx_t
*cqe_rx
,
774 struct snd_queue
*sq
, struct rcv_queue
*rq
)
776 struct sk_buff
*skb
= NULL
;
777 struct nicvf
*nic
= netdev_priv(netdev
);
778 struct nicvf
*snic
= nic
;
782 rq_idx
= nicvf_netdev_qidx(nic
, cqe_rx
->rq_idx
);
785 /* Use primary VF's 'nicvf' struct */
787 netdev
= nic
->netdev
;
790 /* Check for errors */
791 if (cqe_rx
->err_level
|| cqe_rx
->err_opcode
) {
792 err
= nicvf_check_cqe_rx_errs(nic
, cqe_rx
);
793 if (err
&& !cqe_rx
->rb_cnt
)
797 /* For XDP, ignore pkts spanning multiple pages */
798 if (nic
->xdp_prog
&& (cqe_rx
->rb_cnt
== 1)) {
799 /* Packet consumed by XDP */
800 if (nicvf_xdp_rx(snic
, nic
->xdp_prog
, cqe_rx
, sq
, rq
, &skb
))
803 skb
= nicvf_get_rcv_skb(snic
, cqe_rx
,
804 nic
->xdp_prog
? true : false);
810 if (netif_msg_pktdata(nic
)) {
811 netdev_info(nic
->netdev
, "skb 0x%p, len=%d\n", skb
, skb
->len
);
812 print_hex_dump(KERN_INFO
, "", DUMP_PREFIX_OFFSET
, 16, 1,
813 skb
->data
, skb
->len
, true);
816 /* If error packet, drop it here */
818 dev_kfree_skb_any(skb
);
822 nicvf_set_rxtstamp(nic
, skb
);
823 nicvf_set_rxhash(netdev
, cqe_rx
, skb
);
825 skb_record_rx_queue(skb
, rq_idx
);
826 if (netdev
->hw_features
& NETIF_F_RXCSUM
) {
827 /* HW by default verifies TCP/UDP/SCTP checksums */
828 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
830 skb_checksum_none_assert(skb
);
833 skb
->protocol
= eth_type_trans(skb
, netdev
);
835 /* Check for stripped VLAN */
836 if (cqe_rx
->vlan_found
&& cqe_rx
->vlan_stripped
)
837 __vlan_hwaccel_put_tag(skb
, htons(ETH_P_8021Q
),
838 ntohs((__force __be16
)cqe_rx
->vlan_tci
));
840 if (napi
&& (netdev
->features
& NETIF_F_GRO
))
841 napi_gro_receive(napi
, skb
);
843 netif_receive_skb(skb
);
846 static int nicvf_cq_intr_handler(struct net_device
*netdev
, u8 cq_idx
,
847 struct napi_struct
*napi
, int budget
)
849 int processed_cqe
, work_done
= 0, tx_done
= 0;
850 int cqe_count
, cqe_head
;
852 struct nicvf
*nic
= netdev_priv(netdev
);
853 struct queue_set
*qs
= nic
->qs
;
854 struct cmp_queue
*cq
= &qs
->cq
[cq_idx
];
855 struct cqe_rx_t
*cq_desc
;
856 struct netdev_queue
*txq
;
857 struct snd_queue
*sq
= &qs
->sq
[cq_idx
];
858 struct rcv_queue
*rq
= &qs
->rq
[cq_idx
];
859 unsigned int tx_pkts
= 0, tx_bytes
= 0, txq_idx
;
861 spin_lock_bh(&cq
->lock
);
864 /* Get no of valid CQ entries to process */
865 cqe_count
= nicvf_queue_reg_read(nic
, NIC_QSET_CQ_0_7_STATUS
, cq_idx
);
866 cqe_count
&= CQ_CQE_COUNT
;
870 /* Get head of the valid CQ entries */
871 cqe_head
= nicvf_queue_reg_read(nic
, NIC_QSET_CQ_0_7_HEAD
, cq_idx
) >> 9;
874 while (processed_cqe
< cqe_count
) {
875 /* Get the CQ descriptor */
876 cq_desc
= (struct cqe_rx_t
*)GET_CQ_DESC(cq
, cqe_head
);
878 cqe_head
&= (cq
->dmem
.q_len
- 1);
879 /* Initiate prefetch for next descriptor */
880 prefetch((struct cqe_rx_t
*)GET_CQ_DESC(cq
, cqe_head
));
882 if ((work_done
>= budget
) && napi
&&
883 (cq_desc
->cqe_type
!= CQE_TYPE_SEND
)) {
887 switch (cq_desc
->cqe_type
) {
889 nicvf_rcv_pkt_handler(netdev
, napi
, cq_desc
, sq
, rq
);
893 nicvf_snd_pkt_handler(netdev
, (void *)cq_desc
,
894 budget
, &subdesc_cnt
,
895 &tx_pkts
, &tx_bytes
);
898 case CQE_TYPE_SEND_PTP
:
899 nicvf_snd_ptp_handler(netdev
, (void *)cq_desc
);
901 case CQE_TYPE_INVALID
:
902 case CQE_TYPE_RX_SPLIT
:
903 case CQE_TYPE_RX_TCP
:
910 /* Ring doorbell to inform H/W to reuse processed CQEs */
911 nicvf_queue_reg_write(nic
, NIC_QSET_CQ_0_7_DOOR
,
912 cq_idx
, processed_cqe
);
914 if ((work_done
< budget
) && napi
)
918 /* Update SQ's descriptor free count */
920 nicvf_put_sq_desc(sq
, subdesc_cnt
);
922 txq_idx
= nicvf_netdev_qidx(nic
, cq_idx
);
923 /* Handle XDP TX queues */
924 if (nic
->pnicvf
->xdp_prog
) {
925 if (txq_idx
< nic
->pnicvf
->xdp_tx_queues
) {
926 nicvf_xdp_sq_doorbell(nic
, sq
, cq_idx
);
930 txq_idx
-= nic
->pnicvf
->xdp_tx_queues
;
933 /* Wakeup TXQ if its stopped earlier due to SQ full */
935 (atomic_read(&sq
->free_cnt
) >= MIN_SQ_DESC_PER_PKT_XMIT
)) {
936 netdev
= nic
->pnicvf
->netdev
;
937 txq
= netdev_get_tx_queue(netdev
, txq_idx
);
939 netdev_tx_completed_queue(txq
, tx_pkts
, tx_bytes
);
941 /* To read updated queue and carrier status */
943 if (netif_tx_queue_stopped(txq
) && netif_carrier_ok(netdev
)) {
944 netif_tx_wake_queue(txq
);
946 this_cpu_inc(nic
->drv_stats
->txq_wake
);
947 netif_warn(nic
, tx_err
, netdev
,
948 "Transmit queue wakeup SQ%d\n", txq_idx
);
953 spin_unlock_bh(&cq
->lock
);
957 static int nicvf_poll(struct napi_struct
*napi
, int budget
)
961 struct net_device
*netdev
= napi
->dev
;
962 struct nicvf
*nic
= netdev_priv(netdev
);
963 struct nicvf_cq_poll
*cq
;
965 cq
= container_of(napi
, struct nicvf_cq_poll
, napi
);
966 work_done
= nicvf_cq_intr_handler(netdev
, cq
->cq_idx
, napi
, budget
);
968 if (work_done
< budget
) {
969 /* Slow packet rate, exit polling */
970 napi_complete_done(napi
, work_done
);
971 /* Re-enable interrupts */
972 cq_head
= nicvf_queue_reg_read(nic
, NIC_QSET_CQ_0_7_HEAD
,
974 nicvf_clear_intr(nic
, NICVF_INTR_CQ
, cq
->cq_idx
);
975 nicvf_queue_reg_write(nic
, NIC_QSET_CQ_0_7_HEAD
,
976 cq
->cq_idx
, cq_head
);
977 nicvf_enable_intr(nic
, NICVF_INTR_CQ
, cq
->cq_idx
);
982 /* Qset error interrupt handler
984 * As of now only CQ errors are handled
986 static void nicvf_handle_qs_err(struct tasklet_struct
*t
)
988 struct nicvf
*nic
= from_tasklet(nic
, t
, qs_err_task
);
989 struct queue_set
*qs
= nic
->qs
;
993 netif_tx_disable(nic
->netdev
);
995 /* Check if it is CQ err */
996 for (qidx
= 0; qidx
< qs
->cq_cnt
; qidx
++) {
997 status
= nicvf_queue_reg_read(nic
, NIC_QSET_CQ_0_7_STATUS
,
999 if (!(status
& CQ_ERR_MASK
))
1001 /* Process already queued CQEs and reconfig CQ */
1002 nicvf_disable_intr(nic
, NICVF_INTR_CQ
, qidx
);
1003 nicvf_sq_disable(nic
, qidx
);
1004 nicvf_cq_intr_handler(nic
->netdev
, qidx
, NULL
, 0);
1005 nicvf_cmp_queue_config(nic
, qs
, qidx
, true);
1006 nicvf_sq_free_used_descs(nic
->netdev
, &qs
->sq
[qidx
], qidx
);
1007 nicvf_sq_enable(nic
, &qs
->sq
[qidx
], qidx
);
1009 nicvf_enable_intr(nic
, NICVF_INTR_CQ
, qidx
);
1012 netif_tx_start_all_queues(nic
->netdev
);
1013 /* Re-enable Qset error interrupt */
1014 nicvf_enable_intr(nic
, NICVF_INTR_QS_ERR
, 0);
1017 static void nicvf_dump_intr_status(struct nicvf
*nic
)
1019 netif_info(nic
, intr
, nic
->netdev
, "interrupt status 0x%llx\n",
1020 nicvf_reg_read(nic
, NIC_VF_INT
));
1023 static irqreturn_t
nicvf_misc_intr_handler(int irq
, void *nicvf_irq
)
1025 struct nicvf
*nic
= (struct nicvf
*)nicvf_irq
;
1028 nicvf_dump_intr_status(nic
);
1030 intr
= nicvf_reg_read(nic
, NIC_VF_INT
);
1031 /* Check for spurious interrupt */
1032 if (!(intr
& NICVF_INTR_MBOX_MASK
))
1035 nicvf_handle_mbx_intr(nic
);
1040 static irqreturn_t
nicvf_intr_handler(int irq
, void *cq_irq
)
1042 struct nicvf_cq_poll
*cq_poll
= (struct nicvf_cq_poll
*)cq_irq
;
1043 struct nicvf
*nic
= cq_poll
->nicvf
;
1044 int qidx
= cq_poll
->cq_idx
;
1046 nicvf_dump_intr_status(nic
);
1048 /* Disable interrupts */
1049 nicvf_disable_intr(nic
, NICVF_INTR_CQ
, qidx
);
1052 napi_schedule_irqoff(&cq_poll
->napi
);
1054 /* Clear interrupt */
1055 nicvf_clear_intr(nic
, NICVF_INTR_CQ
, qidx
);
1060 static irqreturn_t
nicvf_rbdr_intr_handler(int irq
, void *nicvf_irq
)
1062 struct nicvf
*nic
= (struct nicvf
*)nicvf_irq
;
1066 nicvf_dump_intr_status(nic
);
1068 /* Disable RBDR interrupt and schedule softirq */
1069 for (qidx
= 0; qidx
< nic
->qs
->rbdr_cnt
; qidx
++) {
1070 if (!nicvf_is_intr_enabled(nic
, NICVF_INTR_RBDR
, qidx
))
1072 nicvf_disable_intr(nic
, NICVF_INTR_RBDR
, qidx
);
1073 tasklet_hi_schedule(&nic
->rbdr_task
);
1074 /* Clear interrupt */
1075 nicvf_clear_intr(nic
, NICVF_INTR_RBDR
, qidx
);
1081 static irqreturn_t
nicvf_qs_err_intr_handler(int irq
, void *nicvf_irq
)
1083 struct nicvf
*nic
= (struct nicvf
*)nicvf_irq
;
1085 nicvf_dump_intr_status(nic
);
1087 /* Disable Qset err interrupt and schedule softirq */
1088 nicvf_disable_intr(nic
, NICVF_INTR_QS_ERR
, 0);
1089 tasklet_hi_schedule(&nic
->qs_err_task
);
1090 nicvf_clear_intr(nic
, NICVF_INTR_QS_ERR
, 0);
1095 static void nicvf_set_irq_affinity(struct nicvf
*nic
)
1099 for (vec
= 0; vec
< nic
->num_vec
; vec
++) {
1100 if (!nic
->irq_allocated
[vec
])
1103 if (!zalloc_cpumask_var(&nic
->affinity_mask
[vec
], GFP_KERNEL
))
1106 if (vec
< NICVF_INTR_ID_SQ
)
1107 /* Leave CPU0 for RBDR and other interrupts */
1108 cpu
= nicvf_netdev_qidx(nic
, vec
) + 1;
1112 cpumask_set_cpu(cpumask_local_spread(cpu
, nic
->node
),
1113 nic
->affinity_mask
[vec
]);
1114 irq_set_affinity_hint(pci_irq_vector(nic
->pdev
, vec
),
1115 nic
->affinity_mask
[vec
]);
1119 static int nicvf_register_interrupts(struct nicvf
*nic
)
1123 for_each_cq_irq(irq
)
1124 sprintf(nic
->irq_name
[irq
], "%s-rxtx-%d",
1125 nic
->pnicvf
->netdev
->name
,
1126 nicvf_netdev_qidx(nic
, irq
));
1128 for_each_sq_irq(irq
)
1129 sprintf(nic
->irq_name
[irq
], "%s-sq-%d",
1130 nic
->pnicvf
->netdev
->name
,
1131 nicvf_netdev_qidx(nic
, irq
- NICVF_INTR_ID_SQ
));
1133 for_each_rbdr_irq(irq
)
1134 sprintf(nic
->irq_name
[irq
], "%s-rbdr-%d",
1135 nic
->pnicvf
->netdev
->name
,
1136 nic
->sqs_mode
? (nic
->sqs_id
+ 1) : 0);
1138 /* Register CQ interrupts */
1139 for (irq
= 0; irq
< nic
->qs
->cq_cnt
; irq
++) {
1140 ret
= request_irq(pci_irq_vector(nic
->pdev
, irq
),
1142 0, nic
->irq_name
[irq
], nic
->napi
[irq
]);
1145 nic
->irq_allocated
[irq
] = true;
1148 /* Register RBDR interrupt */
1149 for (irq
= NICVF_INTR_ID_RBDR
;
1150 irq
< (NICVF_INTR_ID_RBDR
+ nic
->qs
->rbdr_cnt
); irq
++) {
1151 ret
= request_irq(pci_irq_vector(nic
->pdev
, irq
),
1152 nicvf_rbdr_intr_handler
,
1153 0, nic
->irq_name
[irq
], nic
);
1156 nic
->irq_allocated
[irq
] = true;
1159 /* Register QS error interrupt */
1160 sprintf(nic
->irq_name
[NICVF_INTR_ID_QS_ERR
], "%s-qset-err-%d",
1161 nic
->pnicvf
->netdev
->name
,
1162 nic
->sqs_mode
? (nic
->sqs_id
+ 1) : 0);
1163 irq
= NICVF_INTR_ID_QS_ERR
;
1164 ret
= request_irq(pci_irq_vector(nic
->pdev
, irq
),
1165 nicvf_qs_err_intr_handler
,
1166 0, nic
->irq_name
[irq
], nic
);
1170 nic
->irq_allocated
[irq
] = true;
1172 /* Set IRQ affinities */
1173 nicvf_set_irq_affinity(nic
);
1177 netdev_err(nic
->netdev
, "request_irq failed, vector %d\n", irq
);
1182 static void nicvf_unregister_interrupts(struct nicvf
*nic
)
1184 struct pci_dev
*pdev
= nic
->pdev
;
1187 /* Free registered interrupts */
1188 for (irq
= 0; irq
< nic
->num_vec
; irq
++) {
1189 if (!nic
->irq_allocated
[irq
])
1192 irq_set_affinity_hint(pci_irq_vector(pdev
, irq
), NULL
);
1193 free_cpumask_var(nic
->affinity_mask
[irq
]);
1195 if (irq
< NICVF_INTR_ID_SQ
)
1196 free_irq(pci_irq_vector(pdev
, irq
), nic
->napi
[irq
]);
1198 free_irq(pci_irq_vector(pdev
, irq
), nic
);
1200 nic
->irq_allocated
[irq
] = false;
1204 pci_free_irq_vectors(pdev
);
1208 /* Initialize MSIX vectors and register MISC interrupt.
1209 * Send READY message to PF to check if its alive
1211 static int nicvf_register_misc_interrupt(struct nicvf
*nic
)
1214 int irq
= NICVF_INTR_ID_MISC
;
1216 /* Return if mailbox interrupt is already registered */
1217 if (nic
->pdev
->msix_enabled
)
1221 nic
->num_vec
= pci_msix_vec_count(nic
->pdev
);
1222 ret
= pci_alloc_irq_vectors(nic
->pdev
, nic
->num_vec
, nic
->num_vec
,
1225 netdev_err(nic
->netdev
,
1226 "Req for #%d msix vectors failed\n", nic
->num_vec
);
1230 sprintf(nic
->irq_name
[irq
], "%s Mbox", "NICVF");
1231 /* Register Misc interrupt */
1232 ret
= request_irq(pci_irq_vector(nic
->pdev
, irq
),
1233 nicvf_misc_intr_handler
, 0, nic
->irq_name
[irq
], nic
);
1237 nic
->irq_allocated
[irq
] = true;
1239 /* Enable mailbox interrupt */
1240 nicvf_enable_intr(nic
, NICVF_INTR_MBOX
, 0);
1242 /* Check if VF is able to communicate with PF */
1243 if (!nicvf_check_pf_ready(nic
)) {
1244 nicvf_disable_intr(nic
, NICVF_INTR_MBOX
, 0);
1245 nicvf_unregister_interrupts(nic
);
1252 static netdev_tx_t
nicvf_xmit(struct sk_buff
*skb
, struct net_device
*netdev
)
1254 struct nicvf
*nic
= netdev_priv(netdev
);
1255 int qid
= skb_get_queue_mapping(skb
);
1256 struct netdev_queue
*txq
= netdev_get_tx_queue(netdev
, qid
);
1258 struct snd_queue
*sq
;
1261 /* Check for minimum packet length */
1262 if (skb
->len
<= ETH_HLEN
) {
1264 return NETDEV_TX_OK
;
1267 /* In XDP case, initial HW tx queues are used for XDP,
1268 * but stack's queue mapping starts at '0', so skip the
1269 * Tx queues attached to Rx queues for XDP.
1272 qid
+= nic
->xdp_tx_queues
;
1275 /* Get secondary Qset's SQ structure */
1276 if (qid
>= MAX_SND_QUEUES_PER_QS
) {
1277 tmp
= qid
/ MAX_SND_QUEUES_PER_QS
;
1278 snic
= (struct nicvf
*)nic
->snicvf
[tmp
- 1];
1280 netdev_warn(nic
->netdev
,
1281 "Secondary Qset#%d's ptr not initialized\n",
1284 return NETDEV_TX_OK
;
1286 qid
= qid
% MAX_SND_QUEUES_PER_QS
;
1289 sq
= &snic
->qs
->sq
[qid
];
1290 if (!netif_tx_queue_stopped(txq
) &&
1291 !nicvf_sq_append_skb(snic
, sq
, skb
, qid
)) {
1292 netif_tx_stop_queue(txq
);
1294 /* Barrier, so that stop_queue visible to other cpus */
1297 /* Check again, incase another cpu freed descriptors */
1298 if (atomic_read(&sq
->free_cnt
) > MIN_SQ_DESC_PER_PKT_XMIT
) {
1299 netif_tx_wake_queue(txq
);
1301 this_cpu_inc(nic
->drv_stats
->txq_stop
);
1302 netif_warn(nic
, tx_err
, netdev
,
1303 "Transmit ring full, stopping SQ%d\n", qid
);
1305 return NETDEV_TX_BUSY
;
1308 return NETDEV_TX_OK
;
1311 static inline void nicvf_free_cq_poll(struct nicvf
*nic
)
1313 struct nicvf_cq_poll
*cq_poll
;
1316 for (qidx
= 0; qidx
< nic
->qs
->cq_cnt
; qidx
++) {
1317 cq_poll
= nic
->napi
[qidx
];
1320 nic
->napi
[qidx
] = NULL
;
1325 int nicvf_stop(struct net_device
*netdev
)
1328 struct nicvf
*nic
= netdev_priv(netdev
);
1329 struct queue_set
*qs
= nic
->qs
;
1330 struct nicvf_cq_poll
*cq_poll
= NULL
;
1331 union nic_mbx mbx
= {};
1333 /* wait till all queued set_rx_mode tasks completes */
1334 if (nic
->nicvf_rx_mode_wq
) {
1335 cancel_delayed_work_sync(&nic
->link_change_work
);
1336 drain_workqueue(nic
->nicvf_rx_mode_wq
);
1339 mbx
.msg
.msg
= NIC_MBOX_MSG_SHUTDOWN
;
1340 nicvf_send_msg_to_pf(nic
, &mbx
);
1342 netif_carrier_off(netdev
);
1343 netif_tx_stop_all_queues(nic
->netdev
);
1344 nic
->link_up
= false;
1346 /* Teardown secondary qsets first */
1347 if (!nic
->sqs_mode
) {
1348 for (qidx
= 0; qidx
< nic
->sqs_count
; qidx
++) {
1349 if (!nic
->snicvf
[qidx
])
1351 nicvf_stop(nic
->snicvf
[qidx
]->netdev
);
1352 nic
->snicvf
[qidx
] = NULL
;
1356 /* Disable RBDR & QS error interrupts */
1357 for (qidx
= 0; qidx
< qs
->rbdr_cnt
; qidx
++) {
1358 nicvf_disable_intr(nic
, NICVF_INTR_RBDR
, qidx
);
1359 nicvf_clear_intr(nic
, NICVF_INTR_RBDR
, qidx
);
1361 nicvf_disable_intr(nic
, NICVF_INTR_QS_ERR
, 0);
1362 nicvf_clear_intr(nic
, NICVF_INTR_QS_ERR
, 0);
1364 /* Wait for pending IRQ handlers to finish */
1365 for (irq
= 0; irq
< nic
->num_vec
; irq
++)
1366 synchronize_irq(pci_irq_vector(nic
->pdev
, irq
));
1368 tasklet_kill(&nic
->rbdr_task
);
1369 tasklet_kill(&nic
->qs_err_task
);
1370 if (nic
->rb_work_scheduled
)
1371 cancel_delayed_work_sync(&nic
->rbdr_work
);
1373 for (qidx
= 0; qidx
< nic
->qs
->cq_cnt
; qidx
++) {
1374 cq_poll
= nic
->napi
[qidx
];
1377 napi_synchronize(&cq_poll
->napi
);
1378 /* CQ intr is enabled while napi_complete,
1381 nicvf_disable_intr(nic
, NICVF_INTR_CQ
, qidx
);
1382 nicvf_clear_intr(nic
, NICVF_INTR_CQ
, qidx
);
1383 napi_disable(&cq_poll
->napi
);
1384 netif_napi_del(&cq_poll
->napi
);
1387 netif_tx_disable(netdev
);
1389 for (qidx
= 0; qidx
< netdev
->num_tx_queues
; qidx
++)
1390 netdev_tx_reset_queue(netdev_get_tx_queue(netdev
, qidx
));
1392 /* Free resources */
1393 nicvf_config_data_transfer(nic
, false);
1395 /* Disable HW Qset */
1396 nicvf_qset_config(nic
, false);
1398 /* disable mailbox interrupt */
1399 nicvf_disable_intr(nic
, NICVF_INTR_MBOX
, 0);
1401 nicvf_unregister_interrupts(nic
);
1403 nicvf_free_cq_poll(nic
);
1405 /* Free any pending SKB saved to receive timestamp */
1407 dev_kfree_skb_any(nic
->ptp_skb
);
1408 nic
->ptp_skb
= NULL
;
1411 /* Clear multiqset info */
1417 static int nicvf_config_hw_rx_tstamp(struct nicvf
*nic
, bool enable
)
1419 union nic_mbx mbx
= {};
1421 mbx
.ptp
.msg
= NIC_MBOX_MSG_PTP_CFG
;
1422 mbx
.ptp
.enable
= enable
;
1424 return nicvf_send_msg_to_pf(nic
, &mbx
);
1427 static int nicvf_update_hw_max_frs(struct nicvf
*nic
, int mtu
)
1429 union nic_mbx mbx
= {};
1431 mbx
.frs
.msg
= NIC_MBOX_MSG_SET_MAX_FRS
;
1432 mbx
.frs
.max_frs
= mtu
;
1433 mbx
.frs
.vf_id
= nic
->vf_id
;
1435 return nicvf_send_msg_to_pf(nic
, &mbx
);
1438 static void nicvf_link_status_check_task(struct work_struct
*work_arg
)
1440 struct nicvf
*nic
= container_of(work_arg
,
1442 link_change_work
.work
);
1443 union nic_mbx mbx
= {};
1444 mbx
.msg
.msg
= NIC_MBOX_MSG_BGX_LINK_CHANGE
;
1445 nicvf_send_msg_to_pf(nic
, &mbx
);
1446 queue_delayed_work(nic
->nicvf_rx_mode_wq
,
1447 &nic
->link_change_work
, 2 * HZ
);
1450 int nicvf_open(struct net_device
*netdev
)
1453 struct nicvf
*nic
= netdev_priv(netdev
);
1454 struct queue_set
*qs
= nic
->qs
;
1455 struct nicvf_cq_poll
*cq_poll
= NULL
;
1457 /* wait till all queued set_rx_mode tasks completes if any */
1458 if (nic
->nicvf_rx_mode_wq
)
1459 drain_workqueue(nic
->nicvf_rx_mode_wq
);
1461 netif_carrier_off(netdev
);
1463 err
= nicvf_register_misc_interrupt(nic
);
1467 /* Register NAPI handler for processing CQEs */
1468 for (qidx
= 0; qidx
< qs
->cq_cnt
; qidx
++) {
1469 cq_poll
= kzalloc(sizeof(*cq_poll
), GFP_KERNEL
);
1474 cq_poll
->cq_idx
= qidx
;
1475 cq_poll
->nicvf
= nic
;
1476 netif_napi_add(netdev
, &cq_poll
->napi
, nicvf_poll
,
1478 napi_enable(&cq_poll
->napi
);
1479 nic
->napi
[qidx
] = cq_poll
;
1482 /* Check if we got MAC address from PF or else generate a radom MAC */
1483 if (!nic
->sqs_mode
&& is_zero_ether_addr(netdev
->dev_addr
)) {
1484 eth_hw_addr_random(netdev
);
1485 nicvf_hw_set_mac_addr(nic
, netdev
);
1488 if (nic
->set_mac_pending
) {
1489 nic
->set_mac_pending
= false;
1490 nicvf_hw_set_mac_addr(nic
, netdev
);
1493 /* Init tasklet for handling Qset err interrupt */
1494 tasklet_setup(&nic
->qs_err_task
, nicvf_handle_qs_err
);
1496 /* Init RBDR tasklet which will refill RBDR */
1497 tasklet_setup(&nic
->rbdr_task
, nicvf_rbdr_task
);
1498 INIT_DELAYED_WORK(&nic
->rbdr_work
, nicvf_rbdr_work
);
1500 /* Configure CPI alorithm */
1501 nic
->cpi_alg
= cpi_alg
;
1503 nicvf_config_cpi(nic
);
1505 nicvf_request_sqs(nic
);
1507 nicvf_get_primary_vf_struct(nic
);
1509 /* Configure PTP timestamp */
1511 nicvf_config_hw_rx_tstamp(nic
, nic
->hw_rx_tstamp
);
1512 atomic_set(&nic
->tx_ptp_skbs
, 0);
1513 nic
->ptp_skb
= NULL
;
1515 /* Configure receive side scaling and MTU */
1516 if (!nic
->sqs_mode
) {
1517 nicvf_rss_init(nic
);
1518 err
= nicvf_update_hw_max_frs(nic
, netdev
->mtu
);
1522 /* Clear percpu stats */
1523 for_each_possible_cpu(cpu
)
1524 memset(per_cpu_ptr(nic
->drv_stats
, cpu
), 0,
1525 sizeof(struct nicvf_drv_stats
));
1528 err
= nicvf_register_interrupts(nic
);
1532 /* Initialize the queues */
1533 err
= nicvf_init_resources(nic
);
1537 /* Make sure queue initialization is written */
1540 nicvf_reg_write(nic
, NIC_VF_INT
, -1);
1541 /* Enable Qset err interrupt */
1542 nicvf_enable_intr(nic
, NICVF_INTR_QS_ERR
, 0);
1544 /* Enable completion queue interrupt */
1545 for (qidx
= 0; qidx
< qs
->cq_cnt
; qidx
++)
1546 nicvf_enable_intr(nic
, NICVF_INTR_CQ
, qidx
);
1548 /* Enable RBDR threshold interrupt */
1549 for (qidx
= 0; qidx
< qs
->rbdr_cnt
; qidx
++)
1550 nicvf_enable_intr(nic
, NICVF_INTR_RBDR
, qidx
);
1552 /* Send VF config done msg to PF */
1553 nicvf_send_cfg_done(nic
);
1555 if (nic
->nicvf_rx_mode_wq
) {
1556 INIT_DELAYED_WORK(&nic
->link_change_work
,
1557 nicvf_link_status_check_task
);
1558 queue_delayed_work(nic
->nicvf_rx_mode_wq
,
1559 &nic
->link_change_work
, 0);
1564 nicvf_disable_intr(nic
, NICVF_INTR_MBOX
, 0);
1565 nicvf_unregister_interrupts(nic
);
1566 tasklet_kill(&nic
->qs_err_task
);
1567 tasklet_kill(&nic
->rbdr_task
);
1569 for (qidx
= 0; qidx
< qs
->cq_cnt
; qidx
++) {
1570 cq_poll
= nic
->napi
[qidx
];
1573 napi_disable(&cq_poll
->napi
);
1574 netif_napi_del(&cq_poll
->napi
);
1576 nicvf_free_cq_poll(nic
);
1580 static int nicvf_change_mtu(struct net_device
*netdev
, int new_mtu
)
1582 struct nicvf
*nic
= netdev_priv(netdev
);
1583 int orig_mtu
= netdev
->mtu
;
1585 /* For now just support only the usual MTU sized frames,
1586 * plus some headroom for VLAN, QinQ.
1588 if (nic
->xdp_prog
&& new_mtu
> MAX_XDP_MTU
) {
1589 netdev_warn(netdev
, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
1594 netdev
->mtu
= new_mtu
;
1596 if (!netif_running(netdev
))
1599 if (nicvf_update_hw_max_frs(nic
, new_mtu
)) {
1600 netdev
->mtu
= orig_mtu
;
1607 static int nicvf_set_mac_address(struct net_device
*netdev
, void *p
)
1609 struct sockaddr
*addr
= p
;
1610 struct nicvf
*nic
= netdev_priv(netdev
);
1612 if (!is_valid_ether_addr(addr
->sa_data
))
1613 return -EADDRNOTAVAIL
;
1615 memcpy(netdev
->dev_addr
, addr
->sa_data
, netdev
->addr_len
);
1617 if (nic
->pdev
->msix_enabled
) {
1618 if (nicvf_hw_set_mac_addr(nic
, netdev
))
1621 nic
->set_mac_pending
= true;
1627 void nicvf_update_lmac_stats(struct nicvf
*nic
)
1630 union nic_mbx mbx
= {};
1632 if (!netif_running(nic
->netdev
))
1635 mbx
.bgx_stats
.msg
= NIC_MBOX_MSG_BGX_STATS
;
1636 mbx
.bgx_stats
.vf_id
= nic
->vf_id
;
1638 mbx
.bgx_stats
.rx
= 1;
1639 while (stat
< BGX_RX_STATS_COUNT
) {
1640 mbx
.bgx_stats
.idx
= stat
;
1641 if (nicvf_send_msg_to_pf(nic
, &mbx
))
1649 mbx
.bgx_stats
.rx
= 0;
1650 while (stat
< BGX_TX_STATS_COUNT
) {
1651 mbx
.bgx_stats
.idx
= stat
;
1652 if (nicvf_send_msg_to_pf(nic
, &mbx
))
1658 void nicvf_update_stats(struct nicvf
*nic
)
1662 struct nicvf_hw_stats
*stats
= &nic
->hw_stats
;
1663 struct nicvf_drv_stats
*drv_stats
;
1664 struct queue_set
*qs
= nic
->qs
;
1666 #define GET_RX_STATS(reg) \
1667 nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | (reg << 3))
1668 #define GET_TX_STATS(reg) \
1669 nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | (reg << 3))
1671 stats
->rx_bytes
= GET_RX_STATS(RX_OCTS
);
1672 stats
->rx_ucast_frames
= GET_RX_STATS(RX_UCAST
);
1673 stats
->rx_bcast_frames
= GET_RX_STATS(RX_BCAST
);
1674 stats
->rx_mcast_frames
= GET_RX_STATS(RX_MCAST
);
1675 stats
->rx_fcs_errors
= GET_RX_STATS(RX_FCS
);
1676 stats
->rx_l2_errors
= GET_RX_STATS(RX_L2ERR
);
1677 stats
->rx_drop_red
= GET_RX_STATS(RX_RED
);
1678 stats
->rx_drop_red_bytes
= GET_RX_STATS(RX_RED_OCTS
);
1679 stats
->rx_drop_overrun
= GET_RX_STATS(RX_ORUN
);
1680 stats
->rx_drop_overrun_bytes
= GET_RX_STATS(RX_ORUN_OCTS
);
1681 stats
->rx_drop_bcast
= GET_RX_STATS(RX_DRP_BCAST
);
1682 stats
->rx_drop_mcast
= GET_RX_STATS(RX_DRP_MCAST
);
1683 stats
->rx_drop_l3_bcast
= GET_RX_STATS(RX_DRP_L3BCAST
);
1684 stats
->rx_drop_l3_mcast
= GET_RX_STATS(RX_DRP_L3MCAST
);
1686 stats
->tx_bytes
= GET_TX_STATS(TX_OCTS
);
1687 stats
->tx_ucast_frames
= GET_TX_STATS(TX_UCAST
);
1688 stats
->tx_bcast_frames
= GET_TX_STATS(TX_BCAST
);
1689 stats
->tx_mcast_frames
= GET_TX_STATS(TX_MCAST
);
1690 stats
->tx_drops
= GET_TX_STATS(TX_DROP
);
1692 /* On T88 pass 2.0, the dummy SQE added for TSO notification
1693 * via CQE has 'dont_send' set. Hence HW drops the pkt pointed
1694 * pointed by dummy SQE and results in tx_drops counter being
1695 * incremented. Subtracting it from tx_tso counter will give
1696 * exact tx_drops counter.
1698 if (nic
->t88
&& nic
->hw_tso
) {
1699 for_each_possible_cpu(cpu
) {
1700 drv_stats
= per_cpu_ptr(nic
->drv_stats
, cpu
);
1701 tmp_stats
+= drv_stats
->tx_tso
;
1703 stats
->tx_drops
= tmp_stats
- stats
->tx_drops
;
1705 stats
->tx_frames
= stats
->tx_ucast_frames
+
1706 stats
->tx_bcast_frames
+
1707 stats
->tx_mcast_frames
;
1708 stats
->rx_frames
= stats
->rx_ucast_frames
+
1709 stats
->rx_bcast_frames
+
1710 stats
->rx_mcast_frames
;
1711 stats
->rx_drops
= stats
->rx_drop_red
+
1712 stats
->rx_drop_overrun
;
1714 /* Update RQ and SQ stats */
1715 for (qidx
= 0; qidx
< qs
->rq_cnt
; qidx
++)
1716 nicvf_update_rq_stats(nic
, qidx
);
1717 for (qidx
= 0; qidx
< qs
->sq_cnt
; qidx
++)
1718 nicvf_update_sq_stats(nic
, qidx
);
1721 static void nicvf_get_stats64(struct net_device
*netdev
,
1722 struct rtnl_link_stats64
*stats
)
1724 struct nicvf
*nic
= netdev_priv(netdev
);
1725 struct nicvf_hw_stats
*hw_stats
= &nic
->hw_stats
;
1727 nicvf_update_stats(nic
);
1729 stats
->rx_bytes
= hw_stats
->rx_bytes
;
1730 stats
->rx_packets
= hw_stats
->rx_frames
;
1731 stats
->rx_dropped
= hw_stats
->rx_drops
;
1732 stats
->multicast
= hw_stats
->rx_mcast_frames
;
1734 stats
->tx_bytes
= hw_stats
->tx_bytes
;
1735 stats
->tx_packets
= hw_stats
->tx_frames
;
1736 stats
->tx_dropped
= hw_stats
->tx_drops
;
1740 static void nicvf_tx_timeout(struct net_device
*dev
, unsigned int txqueue
)
1742 struct nicvf
*nic
= netdev_priv(dev
);
1744 netif_warn(nic
, tx_err
, dev
, "Transmit timed out, resetting\n");
1746 this_cpu_inc(nic
->drv_stats
->tx_timeout
);
1747 schedule_work(&nic
->reset_task
);
1750 static void nicvf_reset_task(struct work_struct
*work
)
1754 nic
= container_of(work
, struct nicvf
, reset_task
);
1756 if (!netif_running(nic
->netdev
))
1759 nicvf_stop(nic
->netdev
);
1760 nicvf_open(nic
->netdev
);
1761 netif_trans_update(nic
->netdev
);
1764 static int nicvf_config_loopback(struct nicvf
*nic
,
1765 netdev_features_t features
)
1767 union nic_mbx mbx
= {};
1769 mbx
.lbk
.msg
= NIC_MBOX_MSG_LOOPBACK
;
1770 mbx
.lbk
.vf_id
= nic
->vf_id
;
1771 mbx
.lbk
.enable
= (features
& NETIF_F_LOOPBACK
) != 0;
1773 return nicvf_send_msg_to_pf(nic
, &mbx
);
1776 static netdev_features_t
nicvf_fix_features(struct net_device
*netdev
,
1777 netdev_features_t features
)
1779 struct nicvf
*nic
= netdev_priv(netdev
);
1781 if ((features
& NETIF_F_LOOPBACK
) &&
1782 netif_running(netdev
) && !nic
->loopback_supported
)
1783 features
&= ~NETIF_F_LOOPBACK
;
1788 static int nicvf_set_features(struct net_device
*netdev
,
1789 netdev_features_t features
)
1791 struct nicvf
*nic
= netdev_priv(netdev
);
1792 netdev_features_t changed
= features
^ netdev
->features
;
1794 if (changed
& NETIF_F_HW_VLAN_CTAG_RX
)
1795 nicvf_config_vlan_stripping(nic
, features
);
1797 if ((changed
& NETIF_F_LOOPBACK
) && netif_running(netdev
))
1798 return nicvf_config_loopback(nic
, features
);
1803 static void nicvf_set_xdp_queues(struct nicvf
*nic
, bool bpf_attached
)
1805 u8 cq_count
, txq_count
;
1807 /* Set XDP Tx queue count same as Rx queue count */
1809 nic
->xdp_tx_queues
= 0;
1811 nic
->xdp_tx_queues
= nic
->rx_queues
;
1813 /* If queue count > MAX_CMP_QUEUES_PER_QS, then additional qsets
1814 * needs to be allocated, check how many.
1816 txq_count
= nic
->xdp_tx_queues
+ nic
->tx_queues
;
1817 cq_count
= max(nic
->rx_queues
, txq_count
);
1818 if (cq_count
> MAX_CMP_QUEUES_PER_QS
) {
1819 nic
->sqs_count
= roundup(cq_count
, MAX_CMP_QUEUES_PER_QS
);
1820 nic
->sqs_count
= (nic
->sqs_count
/ MAX_CMP_QUEUES_PER_QS
) - 1;
1825 /* Set primary Qset's resources */
1826 nic
->qs
->rq_cnt
= min_t(u8
, nic
->rx_queues
, MAX_RCV_QUEUES_PER_QS
);
1827 nic
->qs
->sq_cnt
= min_t(u8
, txq_count
, MAX_SND_QUEUES_PER_QS
);
1828 nic
->qs
->cq_cnt
= max_t(u8
, nic
->qs
->rq_cnt
, nic
->qs
->sq_cnt
);
1831 nicvf_set_real_num_queues(nic
->netdev
, nic
->tx_queues
, nic
->rx_queues
);
1834 static int nicvf_xdp_setup(struct nicvf
*nic
, struct bpf_prog
*prog
)
1836 struct net_device
*dev
= nic
->netdev
;
1837 bool if_up
= netif_running(nic
->netdev
);
1838 struct bpf_prog
*old_prog
;
1839 bool bpf_attached
= false;
1842 /* For now just support only the usual MTU sized frames,
1843 * plus some headroom for VLAN, QinQ.
1845 if (prog
&& dev
->mtu
> MAX_XDP_MTU
) {
1846 netdev_warn(dev
, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
1851 /* ALL SQs attached to CQs i.e same as RQs, are treated as
1852 * XDP Tx queues and more Tx queues are allocated for
1853 * network stack to send pkts out.
1855 * No of Tx queues are either same as Rx queues or whatever
1856 * is left in max no of queues possible.
1858 if ((nic
->rx_queues
+ nic
->tx_queues
) > nic
->max_queues
) {
1860 "Failed to attach BPF prog, RXQs + TXQs > Max %d\n",
1866 nicvf_stop(nic
->netdev
);
1868 old_prog
= xchg(&nic
->xdp_prog
, prog
);
1869 /* Detach old prog, if any */
1871 bpf_prog_put(old_prog
);
1873 if (nic
->xdp_prog
) {
1874 /* Attach BPF program */
1875 bpf_prog_add(nic
->xdp_prog
, nic
->rx_queues
- 1);
1876 bpf_attached
= true;
1879 /* Calculate Tx queues needed for XDP and network stack */
1880 nicvf_set_xdp_queues(nic
, bpf_attached
);
1883 /* Reinitialize interface, clean slate */
1884 nicvf_open(nic
->netdev
);
1885 netif_trans_update(nic
->netdev
);
1891 static int nicvf_xdp(struct net_device
*netdev
, struct netdev_bpf
*xdp
)
1893 struct nicvf
*nic
= netdev_priv(netdev
);
1895 /* To avoid checks while retrieving buffer address from CQE_RX,
1896 * do not support XDP for T88 pass1.x silicons which are anyway
1897 * not in use widely.
1899 if (pass1_silicon(nic
->pdev
))
1902 switch (xdp
->command
) {
1903 case XDP_SETUP_PROG
:
1904 return nicvf_xdp_setup(nic
, xdp
->prog
);
1910 static int nicvf_config_hwtstamp(struct net_device
*netdev
, struct ifreq
*ifr
)
1912 struct hwtstamp_config config
;
1913 struct nicvf
*nic
= netdev_priv(netdev
);
1915 if (!nic
->ptp_clock
)
1918 if (copy_from_user(&config
, ifr
->ifr_data
, sizeof(config
)))
1921 /* reserved for future extensions */
1925 switch (config
.tx_type
) {
1926 case HWTSTAMP_TX_OFF
:
1927 case HWTSTAMP_TX_ON
:
1933 switch (config
.rx_filter
) {
1934 case HWTSTAMP_FILTER_NONE
:
1935 nic
->hw_rx_tstamp
= false;
1937 case HWTSTAMP_FILTER_ALL
:
1938 case HWTSTAMP_FILTER_SOME
:
1939 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT
:
1940 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC
:
1941 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ
:
1942 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT
:
1943 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC
:
1944 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ
:
1945 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT
:
1946 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC
:
1947 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ
:
1948 case HWTSTAMP_FILTER_PTP_V2_EVENT
:
1949 case HWTSTAMP_FILTER_PTP_V2_SYNC
:
1950 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ
:
1951 nic
->hw_rx_tstamp
= true;
1952 config
.rx_filter
= HWTSTAMP_FILTER_ALL
;
1958 if (netif_running(netdev
))
1959 nicvf_config_hw_rx_tstamp(nic
, nic
->hw_rx_tstamp
);
1961 if (copy_to_user(ifr
->ifr_data
, &config
, sizeof(config
)))
1967 static int nicvf_ioctl(struct net_device
*netdev
, struct ifreq
*req
, int cmd
)
1971 return nicvf_config_hwtstamp(netdev
, req
);
1977 static void __nicvf_set_rx_mode_task(u8 mode
, struct xcast_addr_list
*mc_addrs
,
1980 union nic_mbx mbx
= {};
1983 /* From the inside of VM code flow we have only 128 bits memory
1984 * available to send message to host's PF, so send all mc addrs
1985 * one by one, starting from flush command in case if kernel
1986 * requests to configure specific MAC filtering
1989 /* flush DMAC filters and reset RX mode */
1990 mbx
.xcast
.msg
= NIC_MBOX_MSG_RESET_XCAST
;
1991 if (nicvf_send_msg_to_pf(nic
, &mbx
) < 0)
1994 if (mode
& BGX_XCAST_MCAST_FILTER
) {
1995 /* once enabling filtering, we need to signal to PF to add
1996 * its' own LMAC to the filter to accept packets for it.
1998 mbx
.xcast
.msg
= NIC_MBOX_MSG_ADD_MCAST
;
2000 if (nicvf_send_msg_to_pf(nic
, &mbx
) < 0)
2004 /* check if we have any specific MACs to be added to PF DMAC filter */
2006 /* now go through kernel list of MACs and add them one by one */
2007 for (idx
= 0; idx
< mc_addrs
->count
; idx
++) {
2008 mbx
.xcast
.msg
= NIC_MBOX_MSG_ADD_MCAST
;
2009 mbx
.xcast
.mac
= mc_addrs
->mc
[idx
];
2010 if (nicvf_send_msg_to_pf(nic
, &mbx
) < 0)
2015 /* and finally set rx mode for PF accordingly */
2016 mbx
.xcast
.msg
= NIC_MBOX_MSG_SET_XCAST
;
2017 mbx
.xcast
.mode
= mode
;
2019 nicvf_send_msg_to_pf(nic
, &mbx
);
2024 static void nicvf_set_rx_mode_task(struct work_struct
*work_arg
)
2026 struct nicvf_work
*vf_work
= container_of(work_arg
, struct nicvf_work
,
2028 struct nicvf
*nic
= container_of(vf_work
, struct nicvf
, rx_mode_work
);
2030 struct xcast_addr_list
*mc
;
2035 /* Save message data locally to prevent them from
2036 * being overwritten by next ndo_set_rx_mode call().
2038 spin_lock_bh(&nic
->rx_mode_wq_lock
);
2039 mode
= vf_work
->mode
;
2042 spin_unlock_bh(&nic
->rx_mode_wq_lock
);
2044 __nicvf_set_rx_mode_task(mode
, mc
, nic
);
2047 static void nicvf_set_rx_mode(struct net_device
*netdev
)
2049 struct nicvf
*nic
= netdev_priv(netdev
);
2050 struct netdev_hw_addr
*ha
;
2051 struct xcast_addr_list
*mc_list
= NULL
;
2054 if (netdev
->flags
& IFF_PROMISC
) {
2055 mode
= BGX_XCAST_BCAST_ACCEPT
| BGX_XCAST_MCAST_ACCEPT
;
2057 if (netdev
->flags
& IFF_BROADCAST
)
2058 mode
|= BGX_XCAST_BCAST_ACCEPT
;
2060 if (netdev
->flags
& IFF_ALLMULTI
) {
2061 mode
|= BGX_XCAST_MCAST_ACCEPT
;
2062 } else if (netdev
->flags
& IFF_MULTICAST
) {
2063 mode
|= BGX_XCAST_MCAST_FILTER
;
2064 /* here we need to copy mc addrs */
2065 if (netdev_mc_count(netdev
)) {
2066 mc_list
= kmalloc(struct_size(mc_list
, mc
,
2067 netdev_mc_count(netdev
)),
2069 if (unlikely(!mc_list
))
2072 netdev_hw_addr_list_for_each(ha
, &netdev
->mc
) {
2073 mc_list
->mc
[mc_list
->count
] =
2074 ether_addr_to_u64(ha
->addr
);
2080 spin_lock(&nic
->rx_mode_wq_lock
);
2081 kfree(nic
->rx_mode_work
.mc
);
2082 nic
->rx_mode_work
.mc
= mc_list
;
2083 nic
->rx_mode_work
.mode
= mode
;
2084 queue_work(nic
->nicvf_rx_mode_wq
, &nic
->rx_mode_work
.work
);
2085 spin_unlock(&nic
->rx_mode_wq_lock
);
2088 static const struct net_device_ops nicvf_netdev_ops
= {
2089 .ndo_open
= nicvf_open
,
2090 .ndo_stop
= nicvf_stop
,
2091 .ndo_start_xmit
= nicvf_xmit
,
2092 .ndo_change_mtu
= nicvf_change_mtu
,
2093 .ndo_set_mac_address
= nicvf_set_mac_address
,
2094 .ndo_get_stats64
= nicvf_get_stats64
,
2095 .ndo_tx_timeout
= nicvf_tx_timeout
,
2096 .ndo_fix_features
= nicvf_fix_features
,
2097 .ndo_set_features
= nicvf_set_features
,
2098 .ndo_bpf
= nicvf_xdp
,
2099 .ndo_eth_ioctl
= nicvf_ioctl
,
2100 .ndo_set_rx_mode
= nicvf_set_rx_mode
,
2103 static int nicvf_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
2105 struct device
*dev
= &pdev
->dev
;
2106 struct net_device
*netdev
;
2110 struct cavium_ptp
*ptp_clock
;
2112 ptp_clock
= cavium_ptp_get();
2113 if (IS_ERR(ptp_clock
)) {
2114 if (PTR_ERR(ptp_clock
) == -ENODEV
)
2115 /* In virtualized environment we proceed without ptp */
2118 return PTR_ERR(ptp_clock
);
2121 err
= pci_enable_device(pdev
);
2123 dev_err(dev
, "Failed to enable PCI device\n");
2127 err
= pci_request_regions(pdev
, DRV_NAME
);
2129 dev_err(dev
, "PCI request regions failed 0x%x\n", err
);
2130 goto err_disable_device
;
2133 err
= dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(48));
2135 dev_err(dev
, "Unable to get usable DMA configuration\n");
2136 goto err_release_regions
;
2139 qcount
= netif_get_num_default_rss_queues();
2141 /* Restrict multiqset support only for host bound VFs */
2142 if (pdev
->is_virtfn
) {
2143 /* Set max number of queues per VF */
2144 qcount
= min_t(int, num_online_cpus(),
2145 (MAX_SQS_PER_VF
+ 1) * MAX_CMP_QUEUES_PER_QS
);
2148 netdev
= alloc_etherdev_mqs(sizeof(struct nicvf
), qcount
, qcount
);
2151 goto err_release_regions
;
2154 pci_set_drvdata(pdev
, netdev
);
2156 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
2158 nic
= netdev_priv(netdev
);
2159 nic
->netdev
= netdev
;
2162 nic
->max_queues
= qcount
;
2163 /* If no of CPUs are too low, there won't be any queues left
2164 * for XDP_TX, hence double it.
2167 nic
->max_queues
*= 2;
2168 nic
->ptp_clock
= ptp_clock
;
2170 /* Initialize mutex that serializes usage of VF's mailbox */
2171 mutex_init(&nic
->rx_mode_mtx
);
2173 /* MAP VF's configuration registers */
2174 nic
->reg_base
= pcim_iomap(pdev
, PCI_CFG_REG_BAR_NUM
, 0);
2175 if (!nic
->reg_base
) {
2176 dev_err(dev
, "Cannot map config register space, aborting\n");
2178 goto err_free_netdev
;
2181 nic
->drv_stats
= netdev_alloc_pcpu_stats(struct nicvf_drv_stats
);
2182 if (!nic
->drv_stats
) {
2184 goto err_free_netdev
;
2187 err
= nicvf_set_qset_resources(nic
);
2189 goto err_free_netdev
;
2191 /* Check if PF is alive and get MAC address for this VF */
2192 err
= nicvf_register_misc_interrupt(nic
);
2194 goto err_free_netdev
;
2196 nicvf_send_vf_struct(nic
);
2198 if (!pass1_silicon(nic
->pdev
))
2201 /* Get iommu domain for iova to physical addr conversion */
2202 nic
->iommu_domain
= iommu_get_domain_for_dev(dev
);
2204 pci_read_config_word(nic
->pdev
, PCI_SUBSYSTEM_ID
, &sdevid
);
2205 if (sdevid
== 0xA134)
2208 /* Check if this VF is in QS only mode */
2212 err
= nicvf_set_real_num_queues(netdev
, nic
->tx_queues
, nic
->rx_queues
);
2214 goto err_unregister_interrupts
;
2216 netdev
->hw_features
= (NETIF_F_RXCSUM
| NETIF_F_SG
|
2217 NETIF_F_TSO
| NETIF_F_GRO
| NETIF_F_TSO6
|
2218 NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
|
2219 NETIF_F_HW_VLAN_CTAG_RX
);
2221 netdev
->hw_features
|= NETIF_F_RXHASH
;
2223 netdev
->features
|= netdev
->hw_features
;
2224 netdev
->hw_features
|= NETIF_F_LOOPBACK
;
2226 netdev
->vlan_features
= NETIF_F_SG
| NETIF_F_IP_CSUM
|
2227 NETIF_F_IPV6_CSUM
| NETIF_F_TSO
| NETIF_F_TSO6
;
2229 netdev
->netdev_ops
= &nicvf_netdev_ops
;
2230 netdev
->watchdog_timeo
= NICVF_TX_TIMEOUT
;
2232 /* MTU range: 64 - 9200 */
2233 netdev
->min_mtu
= NIC_HW_MIN_FRS
;
2234 netdev
->max_mtu
= NIC_HW_MAX_FRS
;
2236 INIT_WORK(&nic
->reset_task
, nicvf_reset_task
);
2238 nic
->nicvf_rx_mode_wq
= alloc_ordered_workqueue("nicvf_rx_mode_wq_VF%d",
2241 if (!nic
->nicvf_rx_mode_wq
) {
2243 dev_err(dev
, "Failed to allocate work queue\n");
2244 goto err_unregister_interrupts
;
2247 INIT_WORK(&nic
->rx_mode_work
.work
, nicvf_set_rx_mode_task
);
2248 spin_lock_init(&nic
->rx_mode_wq_lock
);
2250 err
= register_netdev(netdev
);
2252 dev_err(dev
, "Failed to register netdevice\n");
2253 goto err_unregister_interrupts
;
2256 nic
->msg_enable
= debug
;
2258 nicvf_set_ethtool_ops(netdev
);
2262 err_unregister_interrupts
:
2263 nicvf_unregister_interrupts(nic
);
2265 pci_set_drvdata(pdev
, NULL
);
2267 free_percpu(nic
->drv_stats
);
2268 free_netdev(netdev
);
2269 err_release_regions
:
2270 pci_release_regions(pdev
);
2272 pci_disable_device(pdev
);
2276 static void nicvf_remove(struct pci_dev
*pdev
)
2278 struct net_device
*netdev
= pci_get_drvdata(pdev
);
2280 struct net_device
*pnetdev
;
2285 nic
= netdev_priv(netdev
);
2286 pnetdev
= nic
->pnicvf
->netdev
;
2288 /* Check if this Qset is assigned to different VF.
2289 * If yes, clean primary and all secondary Qsets.
2291 if (pnetdev
&& (pnetdev
->reg_state
== NETREG_REGISTERED
))
2292 unregister_netdev(pnetdev
);
2293 if (nic
->nicvf_rx_mode_wq
) {
2294 destroy_workqueue(nic
->nicvf_rx_mode_wq
);
2295 nic
->nicvf_rx_mode_wq
= NULL
;
2297 nicvf_unregister_interrupts(nic
);
2298 pci_set_drvdata(pdev
, NULL
);
2300 free_percpu(nic
->drv_stats
);
2301 cavium_ptp_put(nic
->ptp_clock
);
2302 free_netdev(netdev
);
2303 pci_release_regions(pdev
);
2304 pci_disable_device(pdev
);
2307 static void nicvf_shutdown(struct pci_dev
*pdev
)
2312 static struct pci_driver nicvf_driver
= {
2314 .id_table
= nicvf_id_table
,
2315 .probe
= nicvf_probe
,
2316 .remove
= nicvf_remove
,
2317 .shutdown
= nicvf_shutdown
,
2320 static int __init
nicvf_init_module(void)
2322 pr_info("%s, ver %s\n", DRV_NAME
, DRV_VERSION
);
2323 return pci_register_driver(&nicvf_driver
);
2326 static void __exit
nicvf_cleanup_module(void)
2328 pci_unregister_driver(&nicvf_driver
);
2331 module_init(nicvf_init_module
);
2332 module_exit(nicvf_cleanup_module
);