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"
28 u8 num_vf_en
; /* No of VF enabled */
29 bool vf_enabled
[MAX_NUM_VFS_SUPPORTED
];
30 void __iomem
*reg_base
; /* Register start address */
31 u8 num_sqs_en
; /* Secondary qsets enabled */
32 u64 nicvf
[MAX_NUM_VFS_SUPPORTED
];
33 u8 vf_sqs
[MAX_NUM_VFS_SUPPORTED
][MAX_SQS_PER_VF
];
34 u8 pqs_vf
[MAX_NUM_VFS_SUPPORTED
];
35 bool sqs_used
[MAX_NUM_VFS_SUPPORTED
];
36 struct pkind_cfg pkind
;
37 #define NIC_SET_VF_LMAC_MAP(bgx, lmac) (((bgx & 0xF) << 4) | (lmac & 0xF))
38 #define NIC_GET_BGX_FROM_VF_LMAC_MAP(map) ((map >> 4) & 0xF)
39 #define NIC_GET_LMAC_FROM_VF_LMAC_MAP(map) (map & 0xF)
40 u8 vf_lmac_map
[MAX_LMAC
];
41 struct delayed_work dwork
;
42 struct workqueue_struct
*check_link
;
46 u16 cpi_base
[MAX_NUM_VFS_SUPPORTED
];
48 bool mbx_lock
[MAX_NUM_VFS_SUPPORTED
];
53 struct msix_entry msix_entries
[NIC_PF_MSIX_VECTORS
];
54 bool irq_allocated
[NIC_PF_MSIX_VECTORS
];
57 /* Supported devices */
58 static const struct pci_device_id nic_id_table
[] = {
59 { PCI_DEVICE(PCI_VENDOR_ID_CAVIUM
, PCI_DEVICE_ID_THUNDER_NIC_PF
) },
60 { 0, } /* end of table */
63 MODULE_AUTHOR("Sunil Goutham");
64 MODULE_DESCRIPTION("Cavium Thunder NIC Physical Function Driver");
65 MODULE_LICENSE("GPL v2");
66 MODULE_VERSION(DRV_VERSION
);
67 MODULE_DEVICE_TABLE(pci
, nic_id_table
);
69 /* The Cavium ThunderX network controller can *only* be found in SoCs
70 * containing the ThunderX ARM64 CPU implementation. All accesses to the device
71 * registers on this platform are implicitly strongly ordered with respect
72 * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
73 * with no memory barriers in this driver. The readq()/writeq() functions add
74 * explicit ordering operation which in this case are redundant, and only
78 /* Register read/write APIs */
79 static void nic_reg_write(struct nicpf
*nic
, u64 offset
, u64 val
)
81 writeq_relaxed(val
, nic
->reg_base
+ offset
);
84 static u64
nic_reg_read(struct nicpf
*nic
, u64 offset
)
86 return readq_relaxed(nic
->reg_base
+ offset
);
89 /* PF -> VF mailbox communication APIs */
90 static void nic_enable_mbx_intr(struct nicpf
*nic
)
92 /* Enable mailbox interrupt for all 128 VFs */
93 nic_reg_write(nic
, NIC_PF_MAILBOX_ENA_W1S
, ~0ull);
94 nic_reg_write(nic
, NIC_PF_MAILBOX_ENA_W1S
+ sizeof(u64
), ~0ull);
97 static void nic_clear_mbx_intr(struct nicpf
*nic
, int vf
, int mbx_reg
)
99 nic_reg_write(nic
, NIC_PF_MAILBOX_INT
+ (mbx_reg
<< 3), BIT_ULL(vf
));
102 static u64
nic_get_mbx_addr(int vf
)
104 return NIC_PF_VF_0_127_MAILBOX_0_1
+ (vf
<< NIC_VF_NUM_SHIFT
);
107 /* Send a mailbox message to VF
108 * @vf: vf to which this message to be sent
109 * @mbx: Message to be sent
111 static void nic_send_msg_to_vf(struct nicpf
*nic
, int vf
, union nic_mbx
*mbx
)
113 void __iomem
*mbx_addr
= nic
->reg_base
+ nic_get_mbx_addr(vf
);
114 u64
*msg
= (u64
*)mbx
;
116 /* In first revision HW, mbox interrupt is triggerred
117 * when PF writes to MBOX(1), in next revisions when
118 * PF writes to MBOX(0)
120 if (nic
->rev_id
== 0) {
121 /* see the comment for nic_reg_write()/nic_reg_read()
124 writeq_relaxed(msg
[0], mbx_addr
);
125 writeq_relaxed(msg
[1], mbx_addr
+ 8);
127 writeq_relaxed(msg
[1], mbx_addr
+ 8);
128 writeq_relaxed(msg
[0], mbx_addr
);
132 /* Responds to VF's READY message with VF's
133 * ID, node, MAC address e.t.c
134 * @vf: VF which sent READY message
136 static void nic_mbx_send_ready(struct nicpf
*nic
, int vf
)
138 union nic_mbx mbx
= {};
142 mbx
.nic_cfg
.msg
= NIC_MBOX_MSG_READY
;
143 mbx
.nic_cfg
.vf_id
= vf
;
145 mbx
.nic_cfg
.tns_mode
= NIC_TNS_BYPASS_MODE
;
148 bgx_idx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vf
]);
149 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vf
]);
151 mac
= bgx_get_lmac_mac(nic
->node
, bgx_idx
, lmac
);
153 ether_addr_copy((u8
*)&mbx
.nic_cfg
.mac_addr
, mac
);
155 mbx
.nic_cfg
.sqs_mode
= (vf
>= nic
->num_vf_en
) ? true : false;
156 mbx
.nic_cfg
.node_id
= nic
->node
;
157 nic_send_msg_to_vf(nic
, vf
, &mbx
);
160 /* ACKs VF's mailbox message
161 * @vf: VF to which ACK to be sent
163 static void nic_mbx_send_ack(struct nicpf
*nic
, int vf
)
165 union nic_mbx mbx
= {};
167 mbx
.msg
.msg
= NIC_MBOX_MSG_ACK
;
168 nic_send_msg_to_vf(nic
, vf
, &mbx
);
171 /* NACKs VF's mailbox message that PF is not able to
172 * complete the action
173 * @vf: VF to which ACK to be sent
175 static void nic_mbx_send_nack(struct nicpf
*nic
, int vf
)
177 union nic_mbx mbx
= {};
179 mbx
.msg
.msg
= NIC_MBOX_MSG_NACK
;
180 nic_send_msg_to_vf(nic
, vf
, &mbx
);
183 /* Flush all in flight receive packets to memory and
184 * bring down an active RQ
186 static int nic_rcv_queue_sw_sync(struct nicpf
*nic
)
190 nic_reg_write(nic
, NIC_PF_SW_SYNC_RX
, 0x01);
191 /* Wait till sync cycle is finished */
193 if (nic_reg_read(nic
, NIC_PF_SW_SYNC_RX_DONE
) & 0x1)
197 nic_reg_write(nic
, NIC_PF_SW_SYNC_RX
, 0x00);
199 dev_err(&nic
->pdev
->dev
, "Receive queue software sync failed");
205 /* Get BGX Rx/Tx stats and respond to VF's request */
206 static void nic_get_bgx_stats(struct nicpf
*nic
, struct bgx_stats_msg
*bgx
)
209 union nic_mbx mbx
= {};
211 bgx_idx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[bgx
->vf_id
]);
212 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[bgx
->vf_id
]);
214 mbx
.bgx_stats
.msg
= NIC_MBOX_MSG_BGX_STATS
;
215 mbx
.bgx_stats
.vf_id
= bgx
->vf_id
;
216 mbx
.bgx_stats
.rx
= bgx
->rx
;
217 mbx
.bgx_stats
.idx
= bgx
->idx
;
219 mbx
.bgx_stats
.stats
= bgx_get_rx_stats(nic
->node
, bgx_idx
,
222 mbx
.bgx_stats
.stats
= bgx_get_tx_stats(nic
->node
, bgx_idx
,
224 nic_send_msg_to_vf(nic
, bgx
->vf_id
, &mbx
);
227 /* Update hardware min/max frame size */
228 static int nic_update_hw_frs(struct nicpf
*nic
, int new_frs
, int vf
)
230 if ((new_frs
> NIC_HW_MAX_FRS
) || (new_frs
< NIC_HW_MIN_FRS
)) {
231 dev_err(&nic
->pdev
->dev
,
232 "Invalid MTU setting from VF%d rejected, should be between %d and %d\n",
233 vf
, NIC_HW_MIN_FRS
, NIC_HW_MAX_FRS
);
237 if (new_frs
<= nic
->pkind
.maxlen
)
240 nic
->pkind
.maxlen
= new_frs
;
241 nic_reg_write(nic
, NIC_PF_PKIND_0_15_CFG
, *(u64
*)&nic
->pkind
);
245 /* Set minimum transmit packet size */
246 static void nic_set_tx_pkt_pad(struct nicpf
*nic
, int size
)
251 /* Max value that can be set is 60 */
255 for (lmac
= 0; lmac
< (MAX_BGX_PER_CN88XX
* MAX_LMAC_PER_BGX
); lmac
++) {
256 lmac_cfg
= nic_reg_read(nic
, NIC_PF_LMAC_0_7_CFG
| (lmac
<< 3));
257 lmac_cfg
&= ~(0xF << 2);
258 lmac_cfg
|= ((size
/ 4) << 2);
259 nic_reg_write(nic
, NIC_PF_LMAC_0_7_CFG
| (lmac
<< 3), lmac_cfg
);
263 /* Function to check number of LMACs present and set VF::LMAC mapping.
264 * Mapping will be used while initializing channels.
266 static void nic_set_lmac_vf_mapping(struct nicpf
*nic
)
268 unsigned bgx_map
= bgx_get_map(nic
->node
);
269 int bgx
, next_bgx_lmac
= 0;
270 int lmac
, lmac_cnt
= 0;
275 for (bgx
= 0; bgx
< NIC_MAX_BGX
; bgx
++) {
276 if (!(bgx_map
& (1 << bgx
)))
278 lmac_cnt
= bgx_get_lmac_count(nic
->node
, bgx
);
279 for (lmac
= 0; lmac
< lmac_cnt
; lmac
++)
280 nic
->vf_lmac_map
[next_bgx_lmac
++] =
281 NIC_SET_VF_LMAC_MAP(bgx
, lmac
);
282 nic
->num_vf_en
+= lmac_cnt
;
284 /* Program LMAC credits */
285 lmac_credit
= (1ull << 1); /* channel credit enable */
286 lmac_credit
|= (0x1ff << 2); /* Max outstanding pkt count */
287 /* 48KB BGX Tx buffer size, each unit is of size 16bytes */
288 lmac_credit
|= (((((48 * 1024) / lmac_cnt
) -
289 NIC_HW_MAX_FRS
) / 16) << 12);
290 lmac
= bgx
* MAX_LMAC_PER_BGX
;
291 for (; lmac
< lmac_cnt
+ (bgx
* MAX_LMAC_PER_BGX
); lmac
++)
293 NIC_PF_LMAC_0_7_CREDIT
+ (lmac
* 8),
301 static void nic_init_hw(struct nicpf
*nic
)
305 /* Reset NIC, in case the driver is repeatedly inserted and removed */
306 nic_reg_write(nic
, NIC_PF_SOFT_RESET
, 1);
308 /* Enable NIC HW block */
309 nic_reg_write(nic
, NIC_PF_CFG
, 0x3);
311 /* Enable backpressure */
312 nic_reg_write(nic
, NIC_PF_BP_CFG
, (1ULL << 6) | 0x03);
314 /* Disable TNS mode on both interfaces */
315 nic_reg_write(nic
, NIC_PF_INTF_0_1_SEND_CFG
,
316 (NIC_TNS_BYPASS_MODE
<< 7) | BGX0_BLOCK
);
317 nic_reg_write(nic
, NIC_PF_INTF_0_1_SEND_CFG
| (1 << 8),
318 (NIC_TNS_BYPASS_MODE
<< 7) | BGX1_BLOCK
);
319 nic_reg_write(nic
, NIC_PF_INTF_0_1_BP_CFG
,
320 (1ULL << 63) | BGX0_BLOCK
);
321 nic_reg_write(nic
, NIC_PF_INTF_0_1_BP_CFG
+ (1 << 8),
322 (1ULL << 63) | BGX1_BLOCK
);
324 /* PKIND configuration */
325 nic
->pkind
.minlen
= 0;
326 nic
->pkind
.maxlen
= NIC_HW_MAX_FRS
+ ETH_HLEN
;
327 nic
->pkind
.lenerr_en
= 1;
328 nic
->pkind
.rx_hdr
= 0;
329 nic
->pkind
.hdr_sl
= 0;
331 for (i
= 0; i
< NIC_MAX_PKIND
; i
++)
332 nic_reg_write(nic
, NIC_PF_PKIND_0_15_CFG
| (i
<< 3),
333 *(u64
*)&nic
->pkind
);
335 nic_set_tx_pkt_pad(nic
, NIC_HW_MIN_FRS
);
338 nic_reg_write(nic
, NIC_PF_INTR_TIMER_CFG
, NICPF_CLK_PER_INT_TICK
);
340 /* Enable VLAN ethertype matching and stripping */
341 nic_reg_write(nic
, NIC_PF_RX_ETYPE_0_7
,
342 (2 << 19) | (ETYPE_ALG_VLAN_STRIP
<< 16) | ETH_P_8021Q
);
345 /* Channel parse index configuration */
346 static void nic_config_cpi(struct nicpf
*nic
, struct cpi_cfg_msg
*cfg
)
348 u32 vnic
, bgx
, lmac
, chan
;
349 u32 padd
, cpi_count
= 0;
350 u64 cpi_base
, cpi
, rssi_base
, rssi
;
354 bgx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vnic
]);
355 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vnic
]);
357 chan
= (lmac
* MAX_BGX_CHANS_PER_LMAC
) + (bgx
* NIC_CHANS_PER_INF
);
358 cpi_base
= (lmac
* NIC_MAX_CPI_PER_LMAC
) + (bgx
* NIC_CPI_PER_BGX
);
359 rssi_base
= (lmac
* nic
->rss_ind_tbl_size
) + (bgx
* NIC_RSSI_PER_BGX
);
361 /* Rx channel configuration */
362 nic_reg_write(nic
, NIC_PF_CHAN_0_255_RX_BP_CFG
| (chan
<< 3),
363 (1ull << 63) | (vnic
<< 0));
364 nic_reg_write(nic
, NIC_PF_CHAN_0_255_RX_CFG
| (chan
<< 3),
365 ((u64
)cfg
->cpi_alg
<< 62) | (cpi_base
<< 48));
367 if (cfg
->cpi_alg
== CPI_ALG_NONE
)
369 else if (cfg
->cpi_alg
== CPI_ALG_VLAN
) /* 3 bits of PCP */
371 else if (cfg
->cpi_alg
== CPI_ALG_VLAN16
) /* 3 bits PCP + DEI */
373 else if (cfg
->cpi_alg
== CPI_ALG_DIFF
) /* 6bits DSCP */
374 cpi_count
= NIC_MAX_CPI_PER_LMAC
;
376 /* RSS Qset, Qidx mapping */
379 for (; rssi
< (rssi_base
+ cfg
->rq_cnt
); rssi
++) {
380 nic_reg_write(nic
, NIC_PF_RSSI_0_4097_RQ
| (rssi
<< 3),
381 (qset
<< 3) | rq_idx
);
387 for (; cpi
< (cpi_base
+ cpi_count
); cpi
++) {
388 /* Determine port to channel adder */
389 if (cfg
->cpi_alg
!= CPI_ALG_DIFF
)
390 padd
= cpi
% cpi_count
;
392 padd
= cpi
% 8; /* 3 bits CS out of 6bits DSCP */
394 /* Leave RSS_SIZE as '0' to disable RSS */
395 nic_reg_write(nic
, NIC_PF_CPI_0_2047_CFG
| (cpi
<< 3),
396 (vnic
<< 24) | (padd
<< 16) | (rssi_base
+ rssi
));
398 if ((rssi
+ 1) >= cfg
->rq_cnt
)
401 if (cfg
->cpi_alg
== CPI_ALG_VLAN
)
403 else if (cfg
->cpi_alg
== CPI_ALG_VLAN16
)
404 rssi
= ((cpi
- cpi_base
) & 0xe) >> 1;
405 else if (cfg
->cpi_alg
== CPI_ALG_DIFF
)
406 rssi
= ((cpi
- cpi_base
) & 0x38) >> 3;
408 nic
->cpi_base
[cfg
->vf_id
] = cpi_base
;
411 /* Responsds to VF with its RSS indirection table size */
412 static void nic_send_rss_size(struct nicpf
*nic
, int vf
)
414 union nic_mbx mbx
= {};
419 mbx
.rss_size
.msg
= NIC_MBOX_MSG_RSS_SIZE
;
420 mbx
.rss_size
.ind_tbl_size
= nic
->rss_ind_tbl_size
;
421 nic_send_msg_to_vf(nic
, vf
, &mbx
);
424 /* Receive side scaling configuration
427 * - indir table i.e hash::RQ mapping
428 * - no of hash bits to consider
430 static void nic_config_rss(struct nicpf
*nic
, struct rss_cfg_msg
*cfg
)
433 u64 cpi_cfg
, cpi_base
, rssi_base
, rssi
;
435 cpi_base
= nic
->cpi_base
[cfg
->vf_id
];
436 cpi_cfg
= nic_reg_read(nic
, NIC_PF_CPI_0_2047_CFG
| (cpi_base
<< 3));
437 rssi_base
= (cpi_cfg
& 0x0FFF) + cfg
->tbl_offset
;
442 for (; rssi
< (rssi_base
+ cfg
->tbl_len
); rssi
++) {
443 u8 svf
= cfg
->ind_tbl
[idx
] >> 3;
446 qset
= nic
->vf_sqs
[cfg
->vf_id
][svf
- 1];
449 nic_reg_write(nic
, NIC_PF_RSSI_0_4097_RQ
| (rssi
<< 3),
450 (qset
<< 3) | (cfg
->ind_tbl
[idx
] & 0x7));
454 cpi_cfg
&= ~(0xFULL
<< 20);
455 cpi_cfg
|= (cfg
->hash_bits
<< 20);
456 nic_reg_write(nic
, NIC_PF_CPI_0_2047_CFG
| (cpi_base
<< 3), cpi_cfg
);
459 /* 4 level transmit side scheduler configutation
460 * for TNS bypass mode
462 * Sample configuration for SQ0
463 * VNIC0-SQ0 -> TL4(0) -> TL3[0] -> TL2[0] -> TL1[0] -> BGX0
464 * VNIC1-SQ0 -> TL4(8) -> TL3[2] -> TL2[0] -> TL1[0] -> BGX0
465 * VNIC2-SQ0 -> TL4(16) -> TL3[4] -> TL2[1] -> TL1[0] -> BGX0
466 * VNIC3-SQ0 -> TL4(24) -> TL3[6] -> TL2[1] -> TL1[0] -> BGX0
467 * VNIC4-SQ0 -> TL4(512) -> TL3[128] -> TL2[32] -> TL1[1] -> BGX1
468 * VNIC5-SQ0 -> TL4(520) -> TL3[130] -> TL2[32] -> TL1[1] -> BGX1
469 * VNIC6-SQ0 -> TL4(528) -> TL3[132] -> TL2[33] -> TL1[1] -> BGX1
470 * VNIC7-SQ0 -> TL4(536) -> TL3[134] -> TL2[33] -> TL1[1] -> BGX1
472 static void nic_tx_channel_cfg(struct nicpf
*nic
, u8 vnic
,
473 struct sq_cfg_msg
*sq
)
478 u8 sq_idx
= sq
->sq_num
;
482 pqs_vnic
= nic
->pqs_vf
[vnic
];
486 bgx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[pqs_vnic
]);
487 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[pqs_vnic
]);
489 /* 24 bytes for FCS, IPG and preamble */
490 rr_quantum
= ((NIC_HW_MAX_FRS
+ 24) / 4);
492 tl4
= (lmac
* NIC_TL4_PER_LMAC
) + (bgx
* NIC_TL4_PER_BGX
);
497 tl3
= tl4
/ (NIC_MAX_TL4
/ NIC_MAX_TL3
);
498 nic_reg_write(nic
, NIC_PF_QSET_0_127_SQ_0_7_CFG2
|
499 ((u64
)vnic
<< NIC_QS_ID_SHIFT
) |
500 ((u32
)sq_idx
<< NIC_Q_NUM_SHIFT
), tl4
);
501 nic_reg_write(nic
, NIC_PF_TL4_0_1023_CFG
| (tl4
<< 3),
502 ((u64
)vnic
<< 27) | ((u32
)sq_idx
<< 24) | rr_quantum
);
504 nic_reg_write(nic
, NIC_PF_TL3_0_255_CFG
| (tl3
<< 3), rr_quantum
);
505 chan
= (lmac
* MAX_BGX_CHANS_PER_LMAC
) + (bgx
* NIC_CHANS_PER_INF
);
506 nic_reg_write(nic
, NIC_PF_TL3_0_255_CHAN
| (tl3
<< 3), chan
);
507 /* Enable backpressure on the channel */
508 nic_reg_write(nic
, NIC_PF_CHAN_0_255_TX_CFG
| (chan
<< 3), 1);
511 nic_reg_write(nic
, NIC_PF_TL3A_0_63_CFG
| (tl2
<< 3), tl2
);
512 nic_reg_write(nic
, NIC_PF_TL2_0_63_CFG
| (tl2
<< 3), rr_quantum
);
513 /* No priorities as of now */
514 nic_reg_write(nic
, NIC_PF_TL2_0_63_PRI
| (tl2
<< 3), 0x00);
517 /* Send primary nicvf pointer to secondary QS's VF */
518 static void nic_send_pnicvf(struct nicpf
*nic
, int sqs
)
520 union nic_mbx mbx
= {};
522 mbx
.nicvf
.msg
= NIC_MBOX_MSG_PNICVF_PTR
;
523 mbx
.nicvf
.nicvf
= nic
->nicvf
[nic
->pqs_vf
[sqs
]];
524 nic_send_msg_to_vf(nic
, sqs
, &mbx
);
527 /* Send SQS's nicvf pointer to primary QS's VF */
528 static void nic_send_snicvf(struct nicpf
*nic
, struct nicvf_ptr
*nicvf
)
530 union nic_mbx mbx
= {};
531 int sqs_id
= nic
->vf_sqs
[nicvf
->vf_id
][nicvf
->sqs_id
];
533 mbx
.nicvf
.msg
= NIC_MBOX_MSG_SNICVF_PTR
;
534 mbx
.nicvf
.sqs_id
= nicvf
->sqs_id
;
535 mbx
.nicvf
.nicvf
= nic
->nicvf
[sqs_id
];
536 nic_send_msg_to_vf(nic
, nicvf
->vf_id
, &mbx
);
539 /* Find next available Qset that can be assigned as a
540 * secondary Qset to a VF.
542 static int nic_nxt_avail_sqs(struct nicpf
*nic
)
546 for (sqs
= 0; sqs
< nic
->num_sqs_en
; sqs
++) {
547 if (!nic
->sqs_used
[sqs
])
548 nic
->sqs_used
[sqs
] = true;
551 return sqs
+ nic
->num_vf_en
;
556 /* Allocate additional Qsets for requested VF */
557 static void nic_alloc_sqs(struct nicpf
*nic
, struct sqs_alloc
*sqs
)
559 union nic_mbx mbx
= {};
560 int idx
, alloc_qs
= 0;
563 if (!nic
->num_sqs_en
)
566 for (idx
= 0; idx
< sqs
->qs_count
; idx
++) {
567 sqs_id
= nic_nxt_avail_sqs(nic
);
570 nic
->vf_sqs
[sqs
->vf_id
][idx
] = sqs_id
;
571 nic
->pqs_vf
[sqs_id
] = sqs
->vf_id
;
576 mbx
.sqs_alloc
.msg
= NIC_MBOX_MSG_ALLOC_SQS
;
577 mbx
.sqs_alloc
.vf_id
= sqs
->vf_id
;
578 mbx
.sqs_alloc
.qs_count
= alloc_qs
;
579 nic_send_msg_to_vf(nic
, sqs
->vf_id
, &mbx
);
582 /* Interrupt handler to handle mailbox messages from VFs */
583 static void nic_handle_mbx_intr(struct nicpf
*nic
, int vf
)
585 union nic_mbx mbx
= {};
594 nic
->mbx_lock
[vf
] = true;
596 mbx_addr
= nic_get_mbx_addr(vf
);
597 mbx_data
= (u64
*)&mbx
;
599 for (i
= 0; i
< NIC_PF_VF_MAILBOX_SIZE
; i
++) {
600 *mbx_data
= nic_reg_read(nic
, mbx_addr
);
602 mbx_addr
+= sizeof(u64
);
605 dev_dbg(&nic
->pdev
->dev
, "%s: Mailbox msg %d from VF%d\n",
606 __func__
, mbx
.msg
.msg
, vf
);
607 switch (mbx
.msg
.msg
) {
608 case NIC_MBOX_MSG_READY
:
609 nic_mbx_send_ready(nic
, vf
);
617 case NIC_MBOX_MSG_QS_CFG
:
618 reg_addr
= NIC_PF_QSET_0_127_CFG
|
619 (mbx
.qs
.num
<< NIC_QS_ID_SHIFT
);
621 /* Check if its a secondary Qset */
622 if (vf
>= nic
->num_vf_en
) {
623 cfg
= cfg
& (~0x7FULL
);
624 /* Assign this Qset to primary Qset's VF */
625 cfg
|= nic
->pqs_vf
[vf
];
627 nic_reg_write(nic
, reg_addr
, cfg
);
629 case NIC_MBOX_MSG_RQ_CFG
:
630 reg_addr
= NIC_PF_QSET_0_127_RQ_0_7_CFG
|
631 (mbx
.rq
.qs_num
<< NIC_QS_ID_SHIFT
) |
632 (mbx
.rq
.rq_num
<< NIC_Q_NUM_SHIFT
);
633 nic_reg_write(nic
, reg_addr
, mbx
.rq
.cfg
);
635 case NIC_MBOX_MSG_RQ_BP_CFG
:
636 reg_addr
= NIC_PF_QSET_0_127_RQ_0_7_BP_CFG
|
637 (mbx
.rq
.qs_num
<< NIC_QS_ID_SHIFT
) |
638 (mbx
.rq
.rq_num
<< NIC_Q_NUM_SHIFT
);
639 nic_reg_write(nic
, reg_addr
, mbx
.rq
.cfg
);
641 case NIC_MBOX_MSG_RQ_SW_SYNC
:
642 ret
= nic_rcv_queue_sw_sync(nic
);
644 case NIC_MBOX_MSG_RQ_DROP_CFG
:
645 reg_addr
= NIC_PF_QSET_0_127_RQ_0_7_DROP_CFG
|
646 (mbx
.rq
.qs_num
<< NIC_QS_ID_SHIFT
) |
647 (mbx
.rq
.rq_num
<< NIC_Q_NUM_SHIFT
);
648 nic_reg_write(nic
, reg_addr
, mbx
.rq
.cfg
);
650 case NIC_MBOX_MSG_SQ_CFG
:
651 reg_addr
= NIC_PF_QSET_0_127_SQ_0_7_CFG
|
652 (mbx
.sq
.qs_num
<< NIC_QS_ID_SHIFT
) |
653 (mbx
.sq
.sq_num
<< NIC_Q_NUM_SHIFT
);
654 nic_reg_write(nic
, reg_addr
, mbx
.sq
.cfg
);
655 nic_tx_channel_cfg(nic
, mbx
.qs
.num
, &mbx
.sq
);
657 case NIC_MBOX_MSG_SET_MAC
:
658 if (vf
>= nic
->num_vf_en
)
660 lmac
= mbx
.mac
.vf_id
;
661 bgx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[lmac
]);
662 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[lmac
]);
663 bgx_set_lmac_mac(nic
->node
, bgx
, lmac
, mbx
.mac
.mac_addr
);
665 case NIC_MBOX_MSG_SET_MAX_FRS
:
666 ret
= nic_update_hw_frs(nic
, mbx
.frs
.max_frs
,
669 case NIC_MBOX_MSG_CPI_CFG
:
670 nic_config_cpi(nic
, &mbx
.cpi_cfg
);
672 case NIC_MBOX_MSG_RSS_SIZE
:
673 nic_send_rss_size(nic
, vf
);
675 case NIC_MBOX_MSG_RSS_CFG
:
676 case NIC_MBOX_MSG_RSS_CFG_CONT
:
677 nic_config_rss(nic
, &mbx
.rss_cfg
);
679 case NIC_MBOX_MSG_CFG_DONE
:
680 /* Last message of VF config msg sequence */
681 nic
->vf_enabled
[vf
] = true;
683 case NIC_MBOX_MSG_SHUTDOWN
:
684 /* First msg in VF teardown sequence */
685 nic
->vf_enabled
[vf
] = false;
686 if (vf
>= nic
->num_vf_en
)
687 nic
->sqs_used
[vf
- nic
->num_vf_en
] = false;
690 case NIC_MBOX_MSG_ALLOC_SQS
:
691 nic_alloc_sqs(nic
, &mbx
.sqs_alloc
);
693 case NIC_MBOX_MSG_NICVF_PTR
:
694 nic
->nicvf
[vf
] = mbx
.nicvf
.nicvf
;
696 case NIC_MBOX_MSG_PNICVF_PTR
:
697 nic_send_pnicvf(nic
, vf
);
699 case NIC_MBOX_MSG_SNICVF_PTR
:
700 nic_send_snicvf(nic
, &mbx
.nicvf
);
702 case NIC_MBOX_MSG_BGX_STATS
:
703 nic_get_bgx_stats(nic
, &mbx
.bgx_stats
);
706 dev_err(&nic
->pdev
->dev
,
707 "Invalid msg from VF%d, msg 0x%x\n", vf
, mbx
.msg
.msg
);
712 nic_mbx_send_ack(nic
, vf
);
713 else if (mbx
.msg
.msg
!= NIC_MBOX_MSG_READY
)
714 nic_mbx_send_nack(nic
, vf
);
716 nic
->mbx_lock
[vf
] = false;
719 static void nic_mbx_intr_handler (struct nicpf
*nic
, int mbx
)
722 u8 vf
, vf_per_mbx_reg
= 64;
724 intr
= nic_reg_read(nic
, NIC_PF_MAILBOX_INT
+ (mbx
<< 3));
725 dev_dbg(&nic
->pdev
->dev
, "PF interrupt Mbox%d 0x%llx\n", mbx
, intr
);
726 for (vf
= 0; vf
< vf_per_mbx_reg
; vf
++) {
727 if (intr
& (1ULL << vf
)) {
728 dev_dbg(&nic
->pdev
->dev
, "Intr from VF %d\n",
729 vf
+ (mbx
* vf_per_mbx_reg
));
731 nic_handle_mbx_intr(nic
, vf
+ (mbx
* vf_per_mbx_reg
));
732 nic_clear_mbx_intr(nic
, vf
, mbx
);
737 static irqreturn_t
nic_mbx0_intr_handler (int irq
, void *nic_irq
)
739 struct nicpf
*nic
= (struct nicpf
*)nic_irq
;
741 nic_mbx_intr_handler(nic
, 0);
746 static irqreturn_t
nic_mbx1_intr_handler (int irq
, void *nic_irq
)
748 struct nicpf
*nic
= (struct nicpf
*)nic_irq
;
750 nic_mbx_intr_handler(nic
, 1);
755 static int nic_enable_msix(struct nicpf
*nic
)
759 nic
->num_vec
= NIC_PF_MSIX_VECTORS
;
761 for (i
= 0; i
< nic
->num_vec
; i
++)
762 nic
->msix_entries
[i
].entry
= i
;
764 ret
= pci_enable_msix(nic
->pdev
, nic
->msix_entries
, nic
->num_vec
);
766 dev_err(&nic
->pdev
->dev
,
767 "Request for #%d msix vectors failed\n",
772 nic
->msix_enabled
= 1;
776 static void nic_disable_msix(struct nicpf
*nic
)
778 if (nic
->msix_enabled
) {
779 pci_disable_msix(nic
->pdev
);
780 nic
->msix_enabled
= 0;
785 static void nic_free_all_interrupts(struct nicpf
*nic
)
789 for (irq
= 0; irq
< nic
->num_vec
; irq
++) {
790 if (nic
->irq_allocated
[irq
])
791 free_irq(nic
->msix_entries
[irq
].vector
, nic
);
792 nic
->irq_allocated
[irq
] = false;
796 static int nic_register_interrupts(struct nicpf
*nic
)
801 ret
= nic_enable_msix(nic
);
805 /* Register mailbox interrupt handlers */
806 ret
= request_irq(nic
->msix_entries
[NIC_PF_INTR_ID_MBOX0
].vector
,
807 nic_mbx0_intr_handler
, 0, "NIC Mbox0", nic
);
811 nic
->irq_allocated
[NIC_PF_INTR_ID_MBOX0
] = true;
813 ret
= request_irq(nic
->msix_entries
[NIC_PF_INTR_ID_MBOX1
].vector
,
814 nic_mbx1_intr_handler
, 0, "NIC Mbox1", nic
);
818 nic
->irq_allocated
[NIC_PF_INTR_ID_MBOX1
] = true;
820 /* Enable mailbox interrupt */
821 nic_enable_mbx_intr(nic
);
825 dev_err(&nic
->pdev
->dev
, "Request irq failed\n");
826 nic_free_all_interrupts(nic
);
830 static void nic_unregister_interrupts(struct nicpf
*nic
)
832 nic_free_all_interrupts(nic
);
833 nic_disable_msix(nic
);
836 static int nic_num_sqs_en(struct nicpf
*nic
, int vf_en
)
838 int pos
, sqs_per_vf
= MAX_SQS_PER_VF_SINGLE_NODE
;
841 /* Check if its a multi-node environment */
843 sqs_per_vf
= MAX_SQS_PER_VF
;
845 pos
= pci_find_ext_capability(nic
->pdev
, PCI_EXT_CAP_ID_SRIOV
);
846 pci_read_config_word(nic
->pdev
, (pos
+ PCI_SRIOV_TOTAL_VF
), &total_vf
);
847 return min(total_vf
- vf_en
, vf_en
* sqs_per_vf
);
850 static int nic_sriov_init(struct pci_dev
*pdev
, struct nicpf
*nic
)
857 pos
= pci_find_ext_capability(pdev
, PCI_EXT_CAP_ID_SRIOV
);
859 dev_err(&pdev
->dev
, "SRIOV capability is not found in PCIe config space\n");
863 pci_read_config_word(pdev
, (pos
+ PCI_SRIOV_TOTAL_VF
), &total_vf_cnt
);
864 if (total_vf_cnt
< nic
->num_vf_en
)
865 nic
->num_vf_en
= total_vf_cnt
;
870 vf_en
= nic
->num_vf_en
;
871 nic
->num_sqs_en
= nic_num_sqs_en(nic
, nic
->num_vf_en
);
872 vf_en
+= nic
->num_sqs_en
;
874 err
= pci_enable_sriov(pdev
, vf_en
);
876 dev_err(&pdev
->dev
, "SRIOV enable failed, num VF is %d\n",
882 dev_info(&pdev
->dev
, "SRIOV enabled, number of VF available %d\n",
885 nic
->flags
|= NIC_SRIOV_ENABLED
;
889 /* Poll for BGX LMAC link status and update corresponding VF
890 * if there is a change, valid only if internal L2 switch
891 * is not present otherwise VF link is always treated as up
893 static void nic_poll_for_link(struct work_struct
*work
)
895 union nic_mbx mbx
= {};
897 struct bgx_link_status link
;
900 nic
= container_of(work
, struct nicpf
, dwork
.work
);
902 mbx
.link_status
.msg
= NIC_MBOX_MSG_BGX_LINK_CHANGE
;
904 for (vf
= 0; vf
< nic
->num_vf_en
; vf
++) {
905 /* Poll only if VF is UP */
906 if (!nic
->vf_enabled
[vf
])
909 /* Get BGX, LMAC indices for the VF */
910 bgx
= NIC_GET_BGX_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vf
]);
911 lmac
= NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic
->vf_lmac_map
[vf
]);
912 /* Get interface link status */
913 bgx_get_lmac_link_state(nic
->node
, bgx
, lmac
, &link
);
915 /* Inform VF only if link status changed */
916 if (nic
->link
[vf
] == link
.link_up
)
919 if (!nic
->mbx_lock
[vf
]) {
920 nic
->link
[vf
] = link
.link_up
;
921 nic
->duplex
[vf
] = link
.duplex
;
922 nic
->speed
[vf
] = link
.speed
;
924 /* Send a mbox message to VF with current link status */
925 mbx
.link_status
.link_up
= link
.link_up
;
926 mbx
.link_status
.duplex
= link
.duplex
;
927 mbx
.link_status
.speed
= link
.speed
;
928 nic_send_msg_to_vf(nic
, vf
, &mbx
);
931 queue_delayed_work(nic
->check_link
, &nic
->dwork
, HZ
* 2);
934 static int nic_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
936 struct device
*dev
= &pdev
->dev
;
940 BUILD_BUG_ON(sizeof(union nic_mbx
) > 16);
942 nic
= devm_kzalloc(dev
, sizeof(*nic
), GFP_KERNEL
);
946 pci_set_drvdata(pdev
, nic
);
950 err
= pci_enable_device(pdev
);
952 dev_err(dev
, "Failed to enable PCI device\n");
953 pci_set_drvdata(pdev
, NULL
);
957 err
= pci_request_regions(pdev
, DRV_NAME
);
959 dev_err(dev
, "PCI request regions failed 0x%x\n", err
);
960 goto err_disable_device
;
963 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(48));
965 dev_err(dev
, "Unable to get usable DMA configuration\n");
966 goto err_release_regions
;
969 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(48));
971 dev_err(dev
, "Unable to get 48-bit DMA for consistent allocations\n");
972 goto err_release_regions
;
975 /* MAP PF's configuration registers */
976 nic
->reg_base
= pcim_iomap(pdev
, PCI_CFG_REG_BAR_NUM
, 0);
977 if (!nic
->reg_base
) {
978 dev_err(dev
, "Cannot map config register space, aborting\n");
980 goto err_release_regions
;
983 pci_read_config_byte(pdev
, PCI_REVISION_ID
, &nic
->rev_id
);
985 nic
->node
= nic_get_node_id(pdev
);
987 nic_set_lmac_vf_mapping(nic
);
989 /* Initialize hardware */
992 /* Set RSS TBL size for each VF */
993 nic
->rss_ind_tbl_size
= NIC_MAX_RSS_IDR_TBL_SIZE
;
995 /* Register interrupts */
996 err
= nic_register_interrupts(nic
);
998 goto err_release_regions
;
1000 /* Configure SRIOV */
1001 err
= nic_sriov_init(pdev
, nic
);
1003 goto err_unregister_interrupts
;
1005 /* Register a physical link status poll fn() */
1006 nic
->check_link
= alloc_workqueue("check_link_status",
1007 WQ_UNBOUND
| WQ_MEM_RECLAIM
, 1);
1008 if (!nic
->check_link
) {
1010 goto err_disable_sriov
;
1013 INIT_DELAYED_WORK(&nic
->dwork
, nic_poll_for_link
);
1014 queue_delayed_work(nic
->check_link
, &nic
->dwork
, 0);
1019 if (nic
->flags
& NIC_SRIOV_ENABLED
)
1020 pci_disable_sriov(pdev
);
1021 err_unregister_interrupts
:
1022 nic_unregister_interrupts(nic
);
1023 err_release_regions
:
1024 pci_release_regions(pdev
);
1026 pci_disable_device(pdev
);
1027 pci_set_drvdata(pdev
, NULL
);
1031 static void nic_remove(struct pci_dev
*pdev
)
1033 struct nicpf
*nic
= pci_get_drvdata(pdev
);
1035 if (nic
->flags
& NIC_SRIOV_ENABLED
)
1036 pci_disable_sriov(pdev
);
1038 if (nic
->check_link
) {
1039 /* Destroy work Queue */
1040 cancel_delayed_work(&nic
->dwork
);
1041 flush_workqueue(nic
->check_link
);
1042 destroy_workqueue(nic
->check_link
);
1045 nic_unregister_interrupts(nic
);
1046 pci_release_regions(pdev
);
1047 pci_disable_device(pdev
);
1048 pci_set_drvdata(pdev
, NULL
);
1051 static struct pci_driver nic_driver
= {
1053 .id_table
= nic_id_table
,
1055 .remove
= nic_remove
,
1058 static int __init
nic_init_module(void)
1060 pr_info("%s, ver %s\n", DRV_NAME
, DRV_VERSION
);
1062 return pci_register_driver(&nic_driver
);
1065 static void __exit
nic_cleanup_module(void)
1067 pci_unregister_driver(&nic_driver
);
1070 module_init(nic_init_module
);
1071 module_exit(nic_cleanup_module
);