2 * Copyright (C) 2015 Cavium, Inc.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License
6 * as published by the Free Software Foundation.
9 #include <linux/module.h>
10 #include <linux/interrupt.h>
11 #include <linux/pci.h>
12 #include <linux/etherdevice.h>
18 #include "thunder_bgx.h"
20 #define DRV_NAME "thunder-nic"
21 #define DRV_VERSION "1.0"
26 u8 chans_per_bgx
; /* Rx/Tx chans */
36 bool tl1_per_bgx
; /* TL1 per BGX or per LMAC */
44 u8 num_vf_en
; /* No of VF enabled */
45 bool vf_enabled
[MAX_NUM_VFS_SUPPORTED
];
46 void __iomem
*reg_base
; /* Register start address */
47 u8 num_sqs_en
; /* Secondary qsets enabled */
48 u64 nicvf
[MAX_NUM_VFS_SUPPORTED
];
49 u8 vf_sqs
[MAX_NUM_VFS_SUPPORTED
][MAX_SQS_PER_VF
];
50 u8 pqs_vf
[MAX_NUM_VFS_SUPPORTED
];
51 bool sqs_used
[MAX_NUM_VFS_SUPPORTED
];
52 struct pkind_cfg pkind
;
53 #define NIC_SET_VF_LMAC_MAP(bgx, lmac) (((bgx & 0xF) << 4) | (lmac & 0xF))
54 #define NIC_GET_BGX_FROM_VF_LMAC_MAP(map) ((map >> 4) & 0xF)
55 #define NIC_GET_LMAC_FROM_VF_LMAC_MAP(map) (map & 0xF)
56 u8 vf_lmac_map
[MAX_LMAC
];
57 struct delayed_work dwork
;
58 struct workqueue_struct
*check_link
;
62 u16 cpi_base
[MAX_NUM_VFS_SUPPORTED
];
63 u16 rssi_base
[MAX_NUM_VFS_SUPPORTED
];
64 bool mbx_lock
[MAX_NUM_VFS_SUPPORTED
];
69 struct msix_entry msix_entries
[NIC_PF_MSIX_VECTORS
];
70 bool irq_allocated
[NIC_PF_MSIX_VECTORS
];
73 /* Supported devices */
74 static const struct pci_device_id nic_id_table
[] = {
75 { PCI_DEVICE(PCI_VENDOR_ID_CAVIUM
, PCI_DEVICE_ID_THUNDER_NIC_PF
) },
76 { 0, } /* end of table */
79 MODULE_AUTHOR("Sunil Goutham");
80 MODULE_DESCRIPTION("Cavium Thunder NIC Physical Function Driver");
81 MODULE_LICENSE("GPL v2");
82 MODULE_VERSION(DRV_VERSION
);
83 MODULE_DEVICE_TABLE(pci
, nic_id_table
);
85 /* The Cavium ThunderX network controller can *only* be found in SoCs
86 * containing the ThunderX ARM64 CPU implementation. All accesses to the device
87 * registers on this platform are implicitly strongly ordered with respect
88 * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
89 * with no memory barriers in this driver. The readq()/writeq() functions add
90 * explicit ordering operation which in this case are redundant, and only
94 /* Register read/write APIs */
95 static void nic_reg_write(struct nicpf
*nic
, u64 offset
, u64 val
)
97 writeq_relaxed(val
, nic
->reg_base
+ offset
);
100 static u64
nic_reg_read(struct nicpf
*nic
, u64 offset
)
102 return readq_relaxed(nic
->reg_base
+ offset
);
105 /* PF -> VF mailbox communication APIs */
106 static void nic_enable_mbx_intr(struct nicpf
*nic
)
108 /* Enable mailbox interrupt for all 128 VFs */
109 nic_reg_write(nic
, NIC_PF_MAILBOX_ENA_W1S
, ~0ull);
110 nic_reg_write(nic
, NIC_PF_MAILBOX_ENA_W1S
+ sizeof(u64
), ~0ull);
113 static void nic_clear_mbx_intr(struct nicpf
*nic
, int vf
, int mbx_reg
)
115 nic_reg_write(nic
, NIC_PF_MAILBOX_INT
+ (mbx_reg
<< 3), BIT_ULL(vf
));
118 static u64
nic_get_mbx_addr(int vf
)
120 return NIC_PF_VF_0_127_MAILBOX_0_1
+ (vf
<< NIC_VF_NUM_SHIFT
);
123 /* Send a mailbox message to VF
124 * @vf: vf to which this message to be sent
125 * @mbx: Message to be sent
127 static void nic_send_msg_to_vf(struct nicpf
*nic
, int vf
, union nic_mbx
*mbx
)
129 void __iomem
*mbx_addr
= nic
->reg_base
+ nic_get_mbx_addr(vf
);
130 u64
*msg
= (u64
*)mbx
;
132 /* In first revision HW, mbox interrupt is triggerred
133 * when PF writes to MBOX(1), in next revisions when
134 * PF writes to MBOX(0)
136 if (pass1_silicon(nic
->pdev
)) {
137 /* see the comment for nic_reg_write()/nic_reg_read()
140 writeq_relaxed(msg
[0], mbx_addr
);
141 writeq_relaxed(msg
[1], mbx_addr
+ 8);
143 writeq_relaxed(msg
[1], mbx_addr
+ 8);
144 writeq_relaxed(msg
[0], mbx_addr
);
148 /* Responds to VF's READY message with VF's
149 * ID, node, MAC address e.t.c
150 * @vf: VF which sent READY message
152 static void nic_mbx_send_ready(struct nicpf
*nic
, int vf
)
154 union nic_mbx mbx
= {};
158 mbx
.nic_cfg
.msg
= NIC_MBOX_MSG_READY
;
159 mbx
.nic_cfg
.vf_id
= vf
;
161 mbx
.nic_cfg
.tns_mode
= NIC_TNS_BYPASS_MODE
;
164 bgx_idx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vf
]);
165 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vf
]);
167 mac
= bgx_get_lmac_mac(nic
->node
, bgx_idx
, lmac
);
169 ether_addr_copy((u8
*)&mbx
.nic_cfg
.mac_addr
, mac
);
171 mbx
.nic_cfg
.sqs_mode
= (vf
>= nic
->num_vf_en
) ? true : false;
172 mbx
.nic_cfg
.node_id
= nic
->node
;
174 mbx
.nic_cfg
.loopback_supported
= vf
< MAX_LMAC
;
176 nic_send_msg_to_vf(nic
, vf
, &mbx
);
179 /* ACKs VF's mailbox message
180 * @vf: VF to which ACK to be sent
182 static void nic_mbx_send_ack(struct nicpf
*nic
, int vf
)
184 union nic_mbx mbx
= {};
186 mbx
.msg
.msg
= NIC_MBOX_MSG_ACK
;
187 nic_send_msg_to_vf(nic
, vf
, &mbx
);
190 /* NACKs VF's mailbox message that PF is not able to
191 * complete the action
192 * @vf: VF to which ACK to be sent
194 static void nic_mbx_send_nack(struct nicpf
*nic
, int vf
)
196 union nic_mbx mbx
= {};
198 mbx
.msg
.msg
= NIC_MBOX_MSG_NACK
;
199 nic_send_msg_to_vf(nic
, vf
, &mbx
);
202 /* Flush all in flight receive packets to memory and
203 * bring down an active RQ
205 static int nic_rcv_queue_sw_sync(struct nicpf
*nic
)
209 nic_reg_write(nic
, NIC_PF_SW_SYNC_RX
, 0x01);
210 /* Wait till sync cycle is finished */
212 if (nic_reg_read(nic
, NIC_PF_SW_SYNC_RX_DONE
) & 0x1)
216 nic_reg_write(nic
, NIC_PF_SW_SYNC_RX
, 0x00);
218 dev_err(&nic
->pdev
->dev
, "Receive queue software sync failed");
224 /* Get BGX Rx/Tx stats and respond to VF's request */
225 static void nic_get_bgx_stats(struct nicpf
*nic
, struct bgx_stats_msg
*bgx
)
228 union nic_mbx mbx
= {};
230 bgx_idx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[bgx
->vf_id
]);
231 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[bgx
->vf_id
]);
233 mbx
.bgx_stats
.msg
= NIC_MBOX_MSG_BGX_STATS
;
234 mbx
.bgx_stats
.vf_id
= bgx
->vf_id
;
235 mbx
.bgx_stats
.rx
= bgx
->rx
;
236 mbx
.bgx_stats
.idx
= bgx
->idx
;
238 mbx
.bgx_stats
.stats
= bgx_get_rx_stats(nic
->node
, bgx_idx
,
241 mbx
.bgx_stats
.stats
= bgx_get_tx_stats(nic
->node
, bgx_idx
,
243 nic_send_msg_to_vf(nic
, bgx
->vf_id
, &mbx
);
246 /* Update hardware min/max frame size */
247 static int nic_update_hw_frs(struct nicpf
*nic
, int new_frs
, int vf
)
249 if ((new_frs
> NIC_HW_MAX_FRS
) || (new_frs
< NIC_HW_MIN_FRS
)) {
250 dev_err(&nic
->pdev
->dev
,
251 "Invalid MTU setting from VF%d rejected, should be between %d and %d\n",
252 vf
, NIC_HW_MIN_FRS
, NIC_HW_MAX_FRS
);
256 if (new_frs
<= nic
->pkind
.maxlen
)
259 nic
->pkind
.maxlen
= new_frs
;
260 nic_reg_write(nic
, NIC_PF_PKIND_0_15_CFG
, *(u64
*)&nic
->pkind
);
264 /* Set minimum transmit packet size */
265 static void nic_set_tx_pkt_pad(struct nicpf
*nic
, int size
)
270 /* Max value that can be set is 60 */
274 for (lmac
= 0; lmac
< (MAX_BGX_PER_CN88XX
* MAX_LMAC_PER_BGX
); lmac
++) {
275 lmac_cfg
= nic_reg_read(nic
, NIC_PF_LMAC_0_7_CFG
| (lmac
<< 3));
276 lmac_cfg
&= ~(0xF << 2);
277 lmac_cfg
|= ((size
/ 4) << 2);
278 nic_reg_write(nic
, NIC_PF_LMAC_0_7_CFG
| (lmac
<< 3), lmac_cfg
);
282 /* Function to check number of LMACs present and set VF::LMAC mapping.
283 * Mapping will be used while initializing channels.
285 static void nic_set_lmac_vf_mapping(struct nicpf
*nic
)
287 unsigned bgx_map
= bgx_get_map(nic
->node
);
288 int bgx
, next_bgx_lmac
= 0;
289 int lmac
, lmac_cnt
= 0;
294 for (bgx
= 0; bgx
< nic
->hw
->bgx_cnt
; bgx
++) {
295 if (!(bgx_map
& (1 << bgx
)))
297 lmac_cnt
= bgx_get_lmac_count(nic
->node
, bgx
);
298 for (lmac
= 0; lmac
< lmac_cnt
; lmac
++)
299 nic
->vf_lmac_map
[next_bgx_lmac
++] =
300 NIC_SET_VF_LMAC_MAP(bgx
, lmac
);
301 nic
->num_vf_en
+= lmac_cnt
;
303 /* Program LMAC credits */
304 lmac_credit
= (1ull << 1); /* channel credit enable */
305 lmac_credit
|= (0x1ff << 2); /* Max outstanding pkt count */
306 /* 48KB BGX Tx buffer size, each unit is of size 16bytes */
307 lmac_credit
|= (((((48 * 1024) / lmac_cnt
) -
308 NIC_HW_MAX_FRS
) / 16) << 12);
309 lmac
= bgx
* MAX_LMAC_PER_BGX
;
310 for (; lmac
< lmac_cnt
+ (bgx
* MAX_LMAC_PER_BGX
); lmac
++)
312 NIC_PF_LMAC_0_7_CREDIT
+ (lmac
* 8),
317 static void nic_get_hw_info(struct nicpf
*nic
)
320 struct hw_info
*hw
= nic
->hw
;
322 pci_read_config_word(nic
->pdev
, PCI_SUBSYSTEM_ID
, &sdevid
);
325 case PCI_SUBSYS_DEVID_88XX_NIC_PF
:
326 hw
->bgx_cnt
= MAX_BGX_PER_CN88XX
;
327 hw
->chans_per_lmac
= 16;
328 hw
->chans_per_bgx
= 128;
331 hw
->rss_ind_tbl_size
= NIC_MAX_RSS_IDR_TBL_SIZE
;
335 hw
->tl1_per_bgx
= true;
337 case PCI_SUBSYS_DEVID_81XX_NIC_PF
:
338 hw
->bgx_cnt
= MAX_BGX_PER_CN81XX
;
339 hw
->chans_per_lmac
= 8;
340 hw
->chans_per_bgx
= 32;
341 hw
->chans_per_rgx
= 8;
342 hw
->chans_per_lbk
= 24;
345 hw
->rss_ind_tbl_size
= 32; /* Max RSSI / Max interfaces */
349 hw
->tl1_per_bgx
= false;
351 case PCI_SUBSYS_DEVID_83XX_NIC_PF
:
352 hw
->bgx_cnt
= MAX_BGX_PER_CN83XX
;
353 hw
->chans_per_lmac
= 8;
354 hw
->chans_per_bgx
= 32;
355 hw
->chans_per_lbk
= 64;
358 hw
->rss_ind_tbl_size
= 64; /* Max RSSI / Max interfaces */
362 hw
->tl1_per_bgx
= false;
365 hw
->tl4_cnt
= MAX_QUEUES_PER_QSET
* pci_sriov_get_totalvfs(nic
->pdev
);
371 static void nic_init_hw(struct nicpf
*nic
)
376 /* Get HW capability info */
377 nic_get_hw_info(nic
);
379 /* Enable NIC HW block */
380 nic_reg_write(nic
, NIC_PF_CFG
, 0x3);
382 /* Enable backpressure */
383 nic_reg_write(nic
, NIC_PF_BP_CFG
, (1ULL << 6) | 0x03);
385 /* TNS and TNS bypass modes are present only on 88xx */
386 if (nic
->pdev
->subsystem_device
== PCI_SUBSYS_DEVID_88XX_NIC_PF
) {
387 /* Disable TNS mode on both interfaces */
388 nic_reg_write(nic
, NIC_PF_INTF_0_1_SEND_CFG
,
389 (NIC_TNS_BYPASS_MODE
<< 7) | BGX0_BLOCK
);
390 nic_reg_write(nic
, NIC_PF_INTF_0_1_SEND_CFG
| (1 << 8),
391 (NIC_TNS_BYPASS_MODE
<< 7) | BGX1_BLOCK
);
394 nic_reg_write(nic
, NIC_PF_INTF_0_1_BP_CFG
,
395 (1ULL << 63) | BGX0_BLOCK
);
396 nic_reg_write(nic
, NIC_PF_INTF_0_1_BP_CFG
+ (1 << 8),
397 (1ULL << 63) | BGX1_BLOCK
);
399 /* PKIND configuration */
400 nic
->pkind
.minlen
= 0;
401 nic
->pkind
.maxlen
= NIC_HW_MAX_FRS
+ ETH_HLEN
;
402 nic
->pkind
.lenerr_en
= 1;
403 nic
->pkind
.rx_hdr
= 0;
404 nic
->pkind
.hdr_sl
= 0;
406 for (i
= 0; i
< NIC_MAX_PKIND
; i
++)
407 nic_reg_write(nic
, NIC_PF_PKIND_0_15_CFG
| (i
<< 3),
408 *(u64
*)&nic
->pkind
);
410 nic_set_tx_pkt_pad(nic
, NIC_HW_MIN_FRS
);
413 nic_reg_write(nic
, NIC_PF_INTR_TIMER_CFG
, NICPF_CLK_PER_INT_TICK
);
415 /* Enable VLAN ethertype matching and stripping */
416 nic_reg_write(nic
, NIC_PF_RX_ETYPE_0_7
,
417 (2 << 19) | (ETYPE_ALG_VLAN_STRIP
<< 16) | ETH_P_8021Q
);
419 /* Check if HW expected value is higher (could be in future chips) */
420 cqm_cfg
= nic_reg_read(nic
, NIC_PF_CQM_CFG
);
421 if (cqm_cfg
< NICPF_CQM_MIN_DROP_LEVEL
)
422 nic_reg_write(nic
, NIC_PF_CQM_CFG
, NICPF_CQM_MIN_DROP_LEVEL
);
425 /* Channel parse index configuration */
426 static void nic_config_cpi(struct nicpf
*nic
, struct cpi_cfg_msg
*cfg
)
428 struct hw_info
*hw
= nic
->hw
;
429 u32 vnic
, bgx
, lmac
, chan
;
430 u32 padd
, cpi_count
= 0;
431 u64 cpi_base
, cpi
, rssi_base
, rssi
;
435 bgx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vnic
]);
436 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vnic
]);
438 chan
= (lmac
* hw
->chans_per_lmac
) + (bgx
* hw
->chans_per_bgx
);
439 cpi_base
= (lmac
* NIC_MAX_CPI_PER_LMAC
) +
440 (bgx
* (hw
->cpi_cnt
/ hw
->bgx_cnt
));
441 rssi_base
= (lmac
* hw
->rss_ind_tbl_size
) +
442 (bgx
* (hw
->rssi_cnt
/ hw
->bgx_cnt
));
444 /* Rx channel configuration */
445 nic_reg_write(nic
, NIC_PF_CHAN_0_255_RX_BP_CFG
| (chan
<< 3),
446 (1ull << 63) | (vnic
<< 0));
447 nic_reg_write(nic
, NIC_PF_CHAN_0_255_RX_CFG
| (chan
<< 3),
448 ((u64
)cfg
->cpi_alg
<< 62) | (cpi_base
<< 48));
450 if (cfg
->cpi_alg
== CPI_ALG_NONE
)
452 else if (cfg
->cpi_alg
== CPI_ALG_VLAN
) /* 3 bits of PCP */
454 else if (cfg
->cpi_alg
== CPI_ALG_VLAN16
) /* 3 bits PCP + DEI */
456 else if (cfg
->cpi_alg
== CPI_ALG_DIFF
) /* 6bits DSCP */
457 cpi_count
= NIC_MAX_CPI_PER_LMAC
;
459 /* RSS Qset, Qidx mapping */
462 for (; rssi
< (rssi_base
+ cfg
->rq_cnt
); rssi
++) {
463 nic_reg_write(nic
, NIC_PF_RSSI_0_4097_RQ
| (rssi
<< 3),
464 (qset
<< 3) | rq_idx
);
470 for (; cpi
< (cpi_base
+ cpi_count
); cpi
++) {
471 /* Determine port to channel adder */
472 if (cfg
->cpi_alg
!= CPI_ALG_DIFF
)
473 padd
= cpi
% cpi_count
;
475 padd
= cpi
% 8; /* 3 bits CS out of 6bits DSCP */
477 /* Leave RSS_SIZE as '0' to disable RSS */
478 if (pass1_silicon(nic
->pdev
)) {
479 nic_reg_write(nic
, NIC_PF_CPI_0_2047_CFG
| (cpi
<< 3),
480 (vnic
<< 24) | (padd
<< 16) |
483 /* Set MPI_ALG to '0' to disable MCAM parsing */
484 nic_reg_write(nic
, NIC_PF_CPI_0_2047_CFG
| (cpi
<< 3),
486 /* MPI index is same as CPI if MPI_ALG is not enabled */
487 nic_reg_write(nic
, NIC_PF_MPI_0_2047_CFG
| (cpi
<< 3),
488 (vnic
<< 24) | (rssi_base
+ rssi
));
491 if ((rssi
+ 1) >= cfg
->rq_cnt
)
494 if (cfg
->cpi_alg
== CPI_ALG_VLAN
)
496 else if (cfg
->cpi_alg
== CPI_ALG_VLAN16
)
497 rssi
= ((cpi
- cpi_base
) & 0xe) >> 1;
498 else if (cfg
->cpi_alg
== CPI_ALG_DIFF
)
499 rssi
= ((cpi
- cpi_base
) & 0x38) >> 3;
501 nic
->cpi_base
[cfg
->vf_id
] = cpi_base
;
502 nic
->rssi_base
[cfg
->vf_id
] = rssi_base
;
505 /* Responsds to VF with its RSS indirection table size */
506 static void nic_send_rss_size(struct nicpf
*nic
, int vf
)
508 union nic_mbx mbx
= {};
513 mbx
.rss_size
.msg
= NIC_MBOX_MSG_RSS_SIZE
;
514 mbx
.rss_size
.ind_tbl_size
= nic
->hw
->rss_ind_tbl_size
;
515 nic_send_msg_to_vf(nic
, vf
, &mbx
);
518 /* Receive side scaling configuration
521 * - indir table i.e hash::RQ mapping
522 * - no of hash bits to consider
524 static void nic_config_rss(struct nicpf
*nic
, struct rss_cfg_msg
*cfg
)
527 u64 cpi_cfg
, cpi_base
, rssi_base
, rssi
;
530 rssi_base
= nic
->rssi_base
[cfg
->vf_id
] + cfg
->tbl_offset
;
535 for (; rssi
< (rssi_base
+ cfg
->tbl_len
); rssi
++) {
536 u8 svf
= cfg
->ind_tbl
[idx
] >> 3;
539 qset
= nic
->vf_sqs
[cfg
->vf_id
][svf
- 1];
542 nic_reg_write(nic
, NIC_PF_RSSI_0_4097_RQ
| (rssi
<< 3),
543 (qset
<< 3) | (cfg
->ind_tbl
[idx
] & 0x7));
547 cpi_base
= nic
->cpi_base
[cfg
->vf_id
];
548 if (pass1_silicon(nic
->pdev
))
549 idx_addr
= NIC_PF_CPI_0_2047_CFG
;
551 idx_addr
= NIC_PF_MPI_0_2047_CFG
;
552 cpi_cfg
= nic_reg_read(nic
, idx_addr
| (cpi_base
<< 3));
553 cpi_cfg
&= ~(0xFULL
<< 20);
554 cpi_cfg
|= (cfg
->hash_bits
<< 20);
555 nic_reg_write(nic
, idx_addr
| (cpi_base
<< 3), cpi_cfg
);
558 /* 4 level transmit side scheduler configutation
559 * for TNS bypass mode
561 * Sample configuration for SQ0 on 88xx
562 * VNIC0-SQ0 -> TL4(0) -> TL3[0] -> TL2[0] -> TL1[0] -> BGX0
563 * VNIC1-SQ0 -> TL4(8) -> TL3[2] -> TL2[0] -> TL1[0] -> BGX0
564 * VNIC2-SQ0 -> TL4(16) -> TL3[4] -> TL2[1] -> TL1[0] -> BGX0
565 * VNIC3-SQ0 -> TL4(24) -> TL3[6] -> TL2[1] -> TL1[0] -> BGX0
566 * VNIC4-SQ0 -> TL4(512) -> TL3[128] -> TL2[32] -> TL1[1] -> BGX1
567 * VNIC5-SQ0 -> TL4(520) -> TL3[130] -> TL2[32] -> TL1[1] -> BGX1
568 * VNIC6-SQ0 -> TL4(528) -> TL3[132] -> TL2[33] -> TL1[1] -> BGX1
569 * VNIC7-SQ0 -> TL4(536) -> TL3[134] -> TL2[33] -> TL1[1] -> BGX1
571 static void nic_tx_channel_cfg(struct nicpf
*nic
, u8 vnic
,
572 struct sq_cfg_msg
*sq
)
574 struct hw_info
*hw
= nic
->hw
;
578 u8 sq_idx
= sq
->sq_num
;
583 pqs_vnic
= nic
->pqs_vf
[vnic
];
587 bgx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[pqs_vnic
]);
588 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[pqs_vnic
]);
590 /* 24 bytes for FCS, IPG and preamble */
591 rr_quantum
= ((NIC_HW_MAX_FRS
+ 24) / 4);
593 /* For 88xx 0-511 TL4 transmits via BGX0 and
594 * 512-1023 TL4s transmit via BGX1.
596 if (hw
->tl1_per_bgx
) {
597 tl4
= bgx
* (hw
->tl4_cnt
/ hw
->bgx_cnt
);
599 tl4
+= (lmac
* MAX_QUEUES_PER_QSET
);
601 for (svf
= 0; svf
< MAX_SQS_PER_VF
; svf
++) {
602 if (nic
->vf_sqs
[pqs_vnic
][svf
] == vnic
)
605 tl4
+= (MAX_LMAC_PER_BGX
* MAX_QUEUES_PER_QSET
);
606 tl4
+= (lmac
* MAX_QUEUES_PER_QSET
* MAX_SQS_PER_VF
);
607 tl4
+= (svf
* MAX_QUEUES_PER_QSET
);
610 tl4
= (vnic
* MAX_QUEUES_PER_QSET
);
614 tl3
= tl4
/ (hw
->tl4_cnt
/ hw
->tl3_cnt
);
615 nic_reg_write(nic
, NIC_PF_QSET_0_127_SQ_0_7_CFG2
|
616 ((u64
)vnic
<< NIC_QS_ID_SHIFT
) |
617 ((u32
)sq_idx
<< NIC_Q_NUM_SHIFT
), tl4
);
618 nic_reg_write(nic
, NIC_PF_TL4_0_1023_CFG
| (tl4
<< 3),
619 ((u64
)vnic
<< 27) | ((u32
)sq_idx
<< 24) | rr_quantum
);
621 nic_reg_write(nic
, NIC_PF_TL3_0_255_CFG
| (tl3
<< 3), rr_quantum
);
623 /* On 88xx 0-127 channels are for BGX0 and
624 * 127-255 channels for BGX1.
626 * On 81xx/83xx TL3_CHAN reg should be configured with channel
627 * within LMAC i.e 0-7 and not the actual channel number like on 88xx
629 chan
= (lmac
* hw
->chans_per_lmac
) + (bgx
* hw
->chans_per_bgx
);
631 nic_reg_write(nic
, NIC_PF_TL3_0_255_CHAN
| (tl3
<< 3), chan
);
633 nic_reg_write(nic
, NIC_PF_TL3_0_255_CHAN
| (tl3
<< 3), 0);
635 /* Enable backpressure on the channel */
636 nic_reg_write(nic
, NIC_PF_CHAN_0_255_TX_CFG
| (chan
<< 3), 1);
639 nic_reg_write(nic
, NIC_PF_TL3A_0_63_CFG
| (tl2
<< 3), tl2
);
640 nic_reg_write(nic
, NIC_PF_TL2_0_63_CFG
| (tl2
<< 3), rr_quantum
);
641 /* No priorities as of now */
642 nic_reg_write(nic
, NIC_PF_TL2_0_63_PRI
| (tl2
<< 3), 0x00);
644 /* Unlike 88xx where TL2s 0-31 transmits to TL1 '0' and rest to TL1 '1'
645 * on 81xx/83xx TL2 needs to be configured to transmit to one of the
648 * This register doesn't exist on 88xx.
650 if (!hw
->tl1_per_bgx
)
651 nic_reg_write(nic
, NIC_PF_TL2_LMAC
| (tl2
<< 3),
652 lmac
+ (bgx
* MAX_LMAC_PER_BGX
));
655 /* Send primary nicvf pointer to secondary QS's VF */
656 static void nic_send_pnicvf(struct nicpf
*nic
, int sqs
)
658 union nic_mbx mbx
= {};
660 mbx
.nicvf
.msg
= NIC_MBOX_MSG_PNICVF_PTR
;
661 mbx
.nicvf
.nicvf
= nic
->nicvf
[nic
->pqs_vf
[sqs
]];
662 nic_send_msg_to_vf(nic
, sqs
, &mbx
);
665 /* Send SQS's nicvf pointer to primary QS's VF */
666 static void nic_send_snicvf(struct nicpf
*nic
, struct nicvf_ptr
*nicvf
)
668 union nic_mbx mbx
= {};
669 int sqs_id
= nic
->vf_sqs
[nicvf
->vf_id
][nicvf
->sqs_id
];
671 mbx
.nicvf
.msg
= NIC_MBOX_MSG_SNICVF_PTR
;
672 mbx
.nicvf
.sqs_id
= nicvf
->sqs_id
;
673 mbx
.nicvf
.nicvf
= nic
->nicvf
[sqs_id
];
674 nic_send_msg_to_vf(nic
, nicvf
->vf_id
, &mbx
);
677 /* Find next available Qset that can be assigned as a
678 * secondary Qset to a VF.
680 static int nic_nxt_avail_sqs(struct nicpf
*nic
)
684 for (sqs
= 0; sqs
< nic
->num_sqs_en
; sqs
++) {
685 if (!nic
->sqs_used
[sqs
])
686 nic
->sqs_used
[sqs
] = true;
689 return sqs
+ nic
->num_vf_en
;
694 /* Allocate additional Qsets for requested VF */
695 static void nic_alloc_sqs(struct nicpf
*nic
, struct sqs_alloc
*sqs
)
697 union nic_mbx mbx
= {};
698 int idx
, alloc_qs
= 0;
701 if (!nic
->num_sqs_en
)
704 for (idx
= 0; idx
< sqs
->qs_count
; idx
++) {
705 sqs_id
= nic_nxt_avail_sqs(nic
);
708 nic
->vf_sqs
[sqs
->vf_id
][idx
] = sqs_id
;
709 nic
->pqs_vf
[sqs_id
] = sqs
->vf_id
;
714 mbx
.sqs_alloc
.msg
= NIC_MBOX_MSG_ALLOC_SQS
;
715 mbx
.sqs_alloc
.vf_id
= sqs
->vf_id
;
716 mbx
.sqs_alloc
.qs_count
= alloc_qs
;
717 nic_send_msg_to_vf(nic
, sqs
->vf_id
, &mbx
);
720 static int nic_config_loopback(struct nicpf
*nic
, struct set_loopback
*lbk
)
722 int bgx_idx
, lmac_idx
;
724 if (lbk
->vf_id
> MAX_LMAC
)
727 bgx_idx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[lbk
->vf_id
]);
728 lmac_idx
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[lbk
->vf_id
]);
730 bgx_lmac_internal_loopback(nic
->node
, bgx_idx
, lmac_idx
, lbk
->enable
);
735 static void nic_enable_vf(struct nicpf
*nic
, int vf
, bool enable
)
739 nic
->vf_enabled
[vf
] = enable
;
741 if (vf
>= nic
->num_vf_en
)
744 bgx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vf
]);
745 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vf
]);
747 bgx_lmac_rx_tx_enable(nic
->node
, bgx
, lmac
, enable
);
750 /* Interrupt handler to handle mailbox messages from VFs */
751 static void nic_handle_mbx_intr(struct nicpf
*nic
, int vf
)
753 union nic_mbx mbx
= {};
762 nic
->mbx_lock
[vf
] = true;
764 mbx_addr
= nic_get_mbx_addr(vf
);
765 mbx_data
= (u64
*)&mbx
;
767 for (i
= 0; i
< NIC_PF_VF_MAILBOX_SIZE
; i
++) {
768 *mbx_data
= nic_reg_read(nic
, mbx_addr
);
770 mbx_addr
+= sizeof(u64
);
773 dev_dbg(&nic
->pdev
->dev
, "%s: Mailbox msg %d from VF%d\n",
774 __func__
, mbx
.msg
.msg
, vf
);
775 switch (mbx
.msg
.msg
) {
776 case NIC_MBOX_MSG_READY
:
777 nic_mbx_send_ready(nic
, vf
);
785 case NIC_MBOX_MSG_QS_CFG
:
786 reg_addr
= NIC_PF_QSET_0_127_CFG
|
787 (mbx
.qs
.num
<< NIC_QS_ID_SHIFT
);
789 /* Check if its a secondary Qset */
790 if (vf
>= nic
->num_vf_en
) {
791 cfg
= cfg
& (~0x7FULL
);
792 /* Assign this Qset to primary Qset's VF */
793 cfg
|= nic
->pqs_vf
[vf
];
795 nic_reg_write(nic
, reg_addr
, cfg
);
797 case NIC_MBOX_MSG_RQ_CFG
:
798 reg_addr
= NIC_PF_QSET_0_127_RQ_0_7_CFG
|
799 (mbx
.rq
.qs_num
<< NIC_QS_ID_SHIFT
) |
800 (mbx
.rq
.rq_num
<< NIC_Q_NUM_SHIFT
);
801 nic_reg_write(nic
, reg_addr
, mbx
.rq
.cfg
);
803 case NIC_MBOX_MSG_RQ_BP_CFG
:
804 reg_addr
= NIC_PF_QSET_0_127_RQ_0_7_BP_CFG
|
805 (mbx
.rq
.qs_num
<< NIC_QS_ID_SHIFT
) |
806 (mbx
.rq
.rq_num
<< NIC_Q_NUM_SHIFT
);
807 nic_reg_write(nic
, reg_addr
, mbx
.rq
.cfg
);
809 case NIC_MBOX_MSG_RQ_SW_SYNC
:
810 ret
= nic_rcv_queue_sw_sync(nic
);
812 case NIC_MBOX_MSG_RQ_DROP_CFG
:
813 reg_addr
= NIC_PF_QSET_0_127_RQ_0_7_DROP_CFG
|
814 (mbx
.rq
.qs_num
<< NIC_QS_ID_SHIFT
) |
815 (mbx
.rq
.rq_num
<< NIC_Q_NUM_SHIFT
);
816 nic_reg_write(nic
, reg_addr
, mbx
.rq
.cfg
);
818 case NIC_MBOX_MSG_SQ_CFG
:
819 reg_addr
= NIC_PF_QSET_0_127_SQ_0_7_CFG
|
820 (mbx
.sq
.qs_num
<< NIC_QS_ID_SHIFT
) |
821 (mbx
.sq
.sq_num
<< NIC_Q_NUM_SHIFT
);
822 nic_reg_write(nic
, reg_addr
, mbx
.sq
.cfg
);
823 nic_tx_channel_cfg(nic
, mbx
.qs
.num
, &mbx
.sq
);
825 case NIC_MBOX_MSG_SET_MAC
:
826 if (vf
>= nic
->num_vf_en
)
828 lmac
= mbx
.mac
.vf_id
;
829 bgx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[lmac
]);
830 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[lmac
]);
831 bgx_set_lmac_mac(nic
->node
, bgx
, lmac
, mbx
.mac
.mac_addr
);
833 case NIC_MBOX_MSG_SET_MAX_FRS
:
834 ret
= nic_update_hw_frs(nic
, mbx
.frs
.max_frs
,
837 case NIC_MBOX_MSG_CPI_CFG
:
838 nic_config_cpi(nic
, &mbx
.cpi_cfg
);
840 case NIC_MBOX_MSG_RSS_SIZE
:
841 nic_send_rss_size(nic
, vf
);
843 case NIC_MBOX_MSG_RSS_CFG
:
844 case NIC_MBOX_MSG_RSS_CFG_CONT
:
845 nic_config_rss(nic
, &mbx
.rss_cfg
);
847 case NIC_MBOX_MSG_CFG_DONE
:
848 /* Last message of VF config msg sequence */
849 nic_enable_vf(nic
, vf
, true);
851 case NIC_MBOX_MSG_SHUTDOWN
:
852 /* First msg in VF teardown sequence */
853 if (vf
>= nic
->num_vf_en
)
854 nic
->sqs_used
[vf
- nic
->num_vf_en
] = false;
856 nic_enable_vf(nic
, vf
, false);
858 case NIC_MBOX_MSG_ALLOC_SQS
:
859 nic_alloc_sqs(nic
, &mbx
.sqs_alloc
);
861 case NIC_MBOX_MSG_NICVF_PTR
:
862 nic
->nicvf
[vf
] = mbx
.nicvf
.nicvf
;
864 case NIC_MBOX_MSG_PNICVF_PTR
:
865 nic_send_pnicvf(nic
, vf
);
867 case NIC_MBOX_MSG_SNICVF_PTR
:
868 nic_send_snicvf(nic
, &mbx
.nicvf
);
870 case NIC_MBOX_MSG_BGX_STATS
:
871 nic_get_bgx_stats(nic
, &mbx
.bgx_stats
);
873 case NIC_MBOX_MSG_LOOPBACK
:
874 ret
= nic_config_loopback(nic
, &mbx
.lbk
);
877 dev_err(&nic
->pdev
->dev
,
878 "Invalid msg from VF%d, msg 0x%x\n", vf
, mbx
.msg
.msg
);
883 nic_mbx_send_ack(nic
, vf
);
884 else if (mbx
.msg
.msg
!= NIC_MBOX_MSG_READY
)
885 nic_mbx_send_nack(nic
, vf
);
887 nic
->mbx_lock
[vf
] = false;
890 static void nic_mbx_intr_handler (struct nicpf
*nic
, int mbx
)
893 u8 vf
, vf_per_mbx_reg
= 64;
895 intr
= nic_reg_read(nic
, NIC_PF_MAILBOX_INT
+ (mbx
<< 3));
896 dev_dbg(&nic
->pdev
->dev
, "PF interrupt Mbox%d 0x%llx\n", mbx
, intr
);
897 for (vf
= 0; vf
< vf_per_mbx_reg
; vf
++) {
898 if (intr
& (1ULL << vf
)) {
899 dev_dbg(&nic
->pdev
->dev
, "Intr from VF %d\n",
900 vf
+ (mbx
* vf_per_mbx_reg
));
902 nic_handle_mbx_intr(nic
, vf
+ (mbx
* vf_per_mbx_reg
));
903 nic_clear_mbx_intr(nic
, vf
, mbx
);
908 static irqreturn_t
nic_mbx0_intr_handler (int irq
, void *nic_irq
)
910 struct nicpf
*nic
= (struct nicpf
*)nic_irq
;
912 nic_mbx_intr_handler(nic
, 0);
917 static irqreturn_t
nic_mbx1_intr_handler (int irq
, void *nic_irq
)
919 struct nicpf
*nic
= (struct nicpf
*)nic_irq
;
921 nic_mbx_intr_handler(nic
, 1);
926 static int nic_enable_msix(struct nicpf
*nic
)
930 nic
->num_vec
= NIC_PF_MSIX_VECTORS
;
932 for (i
= 0; i
< nic
->num_vec
; i
++)
933 nic
->msix_entries
[i
].entry
= i
;
935 ret
= pci_enable_msix(nic
->pdev
, nic
->msix_entries
, nic
->num_vec
);
937 dev_err(&nic
->pdev
->dev
,
938 "Request for #%d msix vectors failed\n",
943 nic
->msix_enabled
= 1;
947 static void nic_disable_msix(struct nicpf
*nic
)
949 if (nic
->msix_enabled
) {
950 pci_disable_msix(nic
->pdev
);
951 nic
->msix_enabled
= 0;
956 static void nic_free_all_interrupts(struct nicpf
*nic
)
960 for (irq
= 0; irq
< nic
->num_vec
; irq
++) {
961 if (nic
->irq_allocated
[irq
])
962 free_irq(nic
->msix_entries
[irq
].vector
, nic
);
963 nic
->irq_allocated
[irq
] = false;
967 static int nic_register_interrupts(struct nicpf
*nic
)
972 ret
= nic_enable_msix(nic
);
976 /* Register mailbox interrupt handlers */
977 ret
= request_irq(nic
->msix_entries
[NIC_PF_INTR_ID_MBOX0
].vector
,
978 nic_mbx0_intr_handler
, 0, "NIC Mbox0", nic
);
982 nic
->irq_allocated
[NIC_PF_INTR_ID_MBOX0
] = true;
984 ret
= request_irq(nic
->msix_entries
[NIC_PF_INTR_ID_MBOX1
].vector
,
985 nic_mbx1_intr_handler
, 0, "NIC Mbox1", nic
);
989 nic
->irq_allocated
[NIC_PF_INTR_ID_MBOX1
] = true;
991 /* Enable mailbox interrupt */
992 nic_enable_mbx_intr(nic
);
996 dev_err(&nic
->pdev
->dev
, "Request irq failed\n");
997 nic_free_all_interrupts(nic
);
1001 static void nic_unregister_interrupts(struct nicpf
*nic
)
1003 nic_free_all_interrupts(nic
);
1004 nic_disable_msix(nic
);
1007 static int nic_num_sqs_en(struct nicpf
*nic
, int vf_en
)
1009 int pos
, sqs_per_vf
= MAX_SQS_PER_VF_SINGLE_NODE
;
1012 /* Check if its a multi-node environment */
1013 if (nr_node_ids
> 1)
1014 sqs_per_vf
= MAX_SQS_PER_VF
;
1016 pos
= pci_find_ext_capability(nic
->pdev
, PCI_EXT_CAP_ID_SRIOV
);
1017 pci_read_config_word(nic
->pdev
, (pos
+ PCI_SRIOV_TOTAL_VF
), &total_vf
);
1018 return min(total_vf
- vf_en
, vf_en
* sqs_per_vf
);
1021 static int nic_sriov_init(struct pci_dev
*pdev
, struct nicpf
*nic
)
1028 pos
= pci_find_ext_capability(pdev
, PCI_EXT_CAP_ID_SRIOV
);
1030 dev_err(&pdev
->dev
, "SRIOV capability is not found in PCIe config space\n");
1034 pci_read_config_word(pdev
, (pos
+ PCI_SRIOV_TOTAL_VF
), &total_vf_cnt
);
1035 if (total_vf_cnt
< nic
->num_vf_en
)
1036 nic
->num_vf_en
= total_vf_cnt
;
1041 vf_en
= nic
->num_vf_en
;
1042 nic
->num_sqs_en
= nic_num_sqs_en(nic
, nic
->num_vf_en
);
1043 vf_en
+= nic
->num_sqs_en
;
1045 err
= pci_enable_sriov(pdev
, vf_en
);
1047 dev_err(&pdev
->dev
, "SRIOV enable failed, num VF is %d\n",
1053 dev_info(&pdev
->dev
, "SRIOV enabled, number of VF available %d\n",
1056 nic
->flags
|= NIC_SRIOV_ENABLED
;
1060 /* Poll for BGX LMAC link status and update corresponding VF
1061 * if there is a change, valid only if internal L2 switch
1062 * is not present otherwise VF link is always treated as up
1064 static void nic_poll_for_link(struct work_struct
*work
)
1066 union nic_mbx mbx
= {};
1068 struct bgx_link_status link
;
1071 nic
= container_of(work
, struct nicpf
, dwork
.work
);
1073 mbx
.link_status
.msg
= NIC_MBOX_MSG_BGX_LINK_CHANGE
;
1075 for (vf
= 0; vf
< nic
->num_vf_en
; vf
++) {
1076 /* Poll only if VF is UP */
1077 if (!nic
->vf_enabled
[vf
])
1080 /* Get BGX, LMAC indices for the VF */
1081 bgx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vf
]);
1082 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vf
]);
1083 /* Get interface link status */
1084 bgx_get_lmac_link_state(nic
->node
, bgx
, lmac
, &link
);
1086 /* Inform VF only if link status changed */
1087 if (nic
->link
[vf
] == link
.link_up
)
1090 if (!nic
->mbx_lock
[vf
]) {
1091 nic
->link
[vf
] = link
.link_up
;
1092 nic
->duplex
[vf
] = link
.duplex
;
1093 nic
->speed
[vf
] = link
.speed
;
1095 /* Send a mbox message to VF with current link status */
1096 mbx
.link_status
.link_up
= link
.link_up
;
1097 mbx
.link_status
.duplex
= link
.duplex
;
1098 mbx
.link_status
.speed
= link
.speed
;
1099 nic_send_msg_to_vf(nic
, vf
, &mbx
);
1102 queue_delayed_work(nic
->check_link
, &nic
->dwork
, HZ
* 2);
1105 static int nic_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
1107 struct device
*dev
= &pdev
->dev
;
1111 BUILD_BUG_ON(sizeof(union nic_mbx
) > 16);
1113 nic
= devm_kzalloc(dev
, sizeof(*nic
), GFP_KERNEL
);
1117 nic
->hw
= devm_kzalloc(dev
, sizeof(struct hw_info
), GFP_KERNEL
);
1119 devm_kfree(dev
, nic
);
1123 pci_set_drvdata(pdev
, nic
);
1127 err
= pci_enable_device(pdev
);
1129 dev_err(dev
, "Failed to enable PCI device\n");
1130 pci_set_drvdata(pdev
, NULL
);
1134 err
= pci_request_regions(pdev
, DRV_NAME
);
1136 dev_err(dev
, "PCI request regions failed 0x%x\n", err
);
1137 goto err_disable_device
;
1140 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(48));
1142 dev_err(dev
, "Unable to get usable DMA configuration\n");
1143 goto err_release_regions
;
1146 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(48));
1148 dev_err(dev
, "Unable to get 48-bit DMA for consistent allocations\n");
1149 goto err_release_regions
;
1152 /* MAP PF's configuration registers */
1153 nic
->reg_base
= pcim_iomap(pdev
, PCI_CFG_REG_BAR_NUM
, 0);
1154 if (!nic
->reg_base
) {
1155 dev_err(dev
, "Cannot map config register space, aborting\n");
1157 goto err_release_regions
;
1160 nic
->node
= nic_get_node_id(pdev
);
1162 /* Initialize hardware */
1165 nic_set_lmac_vf_mapping(nic
);
1167 /* Register interrupts */
1168 err
= nic_register_interrupts(nic
);
1170 goto err_release_regions
;
1172 /* Configure SRIOV */
1173 err
= nic_sriov_init(pdev
, nic
);
1175 goto err_unregister_interrupts
;
1177 /* Register a physical link status poll fn() */
1178 nic
->check_link
= alloc_workqueue("check_link_status",
1179 WQ_UNBOUND
| WQ_MEM_RECLAIM
, 1);
1180 if (!nic
->check_link
) {
1182 goto err_disable_sriov
;
1185 INIT_DELAYED_WORK(&nic
->dwork
, nic_poll_for_link
);
1186 queue_delayed_work(nic
->check_link
, &nic
->dwork
, 0);
1191 if (nic
->flags
& NIC_SRIOV_ENABLED
)
1192 pci_disable_sriov(pdev
);
1193 err_unregister_interrupts
:
1194 nic_unregister_interrupts(nic
);
1195 err_release_regions
:
1196 pci_release_regions(pdev
);
1198 devm_kfree(dev
, nic
->hw
);
1199 devm_kfree(dev
, nic
);
1200 pci_disable_device(pdev
);
1201 pci_set_drvdata(pdev
, NULL
);
1205 static void nic_remove(struct pci_dev
*pdev
)
1207 struct nicpf
*nic
= pci_get_drvdata(pdev
);
1209 if (nic
->flags
& NIC_SRIOV_ENABLED
)
1210 pci_disable_sriov(pdev
);
1212 if (nic
->check_link
) {
1213 /* Destroy work Queue */
1214 cancel_delayed_work_sync(&nic
->dwork
);
1215 destroy_workqueue(nic
->check_link
);
1218 nic_unregister_interrupts(nic
);
1219 pci_release_regions(pdev
);
1221 devm_kfree(&pdev
->dev
, nic
->hw
);
1222 devm_kfree(&pdev
->dev
, nic
);
1224 pci_disable_device(pdev
);
1225 pci_set_drvdata(pdev
, NULL
);
1228 static struct pci_driver nic_driver
= {
1230 .id_table
= nic_id_table
,
1232 .remove
= nic_remove
,
1235 static int __init
nic_init_module(void)
1237 pr_info("%s, ver %s\n", DRV_NAME
, DRV_VERSION
);
1239 return pci_register_driver(&nic_driver
);
1242 static void __exit
nic_cleanup_module(void)
1244 pci_unregister_driver(&nic_driver
);
1247 module_init(nic_init_module
);
1248 module_exit(nic_cleanup_module
);