1 /*
2  * Copyright (C) 2015 Cavium, Inc.
3  *
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.
7  */
8 
9 #include <linux/module.h>
10 #include <linux/interrupt.h>
11 #include <linux/pci.h>
12 #include <linux/etherdevice.h>
13 #include <linux/of.h>
14 
15 #include "nic_reg.h"
16 #include "nic.h"
17 #include "q_struct.h"
18 #include "thunder_bgx.h"
19 
20 #define DRV_NAME	"thunder-nic"
21 #define DRV_VERSION	"1.0"
22 
23 struct hw_info {
24 	u8		bgx_cnt;
25 	u8		chans_per_lmac;
26 	u8		chans_per_bgx; /* Rx/Tx chans */
27 	u8		chans_per_rgx;
28 	u8		chans_per_lbk;
29 	u16		cpi_cnt;
30 	u16		rssi_cnt;
31 	u16		rss_ind_tbl_size;
32 	u16		tl4_cnt;
33 	u16		tl3_cnt;
34 	u8		tl2_cnt;
35 	u8		tl1_cnt;
36 	bool		tl1_per_bgx; /* TL1 per BGX or per LMAC */
37 };
38 
39 struct nicpf {
40 	struct pci_dev		*pdev;
41 	struct hw_info          *hw;
42 	u8			node;
43 	unsigned int		flags;
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;
57 	struct delayed_work     dwork;
58 	struct workqueue_struct *check_link;
59 	u8			*link;
60 	u8			*duplex;
61 	u32			*speed;
62 	u16			cpi_base[MAX_NUM_VFS_SUPPORTED];
63 	u16			rssi_base[MAX_NUM_VFS_SUPPORTED];
64 	bool			mbx_lock[MAX_NUM_VFS_SUPPORTED];
65 
66 	/* MSI-X */
67 	bool			msix_enabled;
68 	u8			num_vec;
69 	struct msix_entry	*msix_entries;
70 	bool			irq_allocated[NIC_PF_MSIX_VECTORS];
71 	char			irq_name[NIC_PF_MSIX_VECTORS][20];
72 };
73 
74 /* Supported devices */
75 static const struct pci_device_id nic_id_table[] = {
76 	{ PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, PCI_DEVICE_ID_THUNDER_NIC_PF) },
77 	{ 0, }  /* end of table */
78 };
79 
80 MODULE_AUTHOR("Sunil Goutham");
81 MODULE_DESCRIPTION("Cavium Thunder NIC Physical Function Driver");
82 MODULE_LICENSE("GPL v2");
83 MODULE_VERSION(DRV_VERSION);
84 MODULE_DEVICE_TABLE(pci, nic_id_table);
85 
86 /* The Cavium ThunderX network controller can *only* be found in SoCs
87  * containing the ThunderX ARM64 CPU implementation.  All accesses to the device
88  * registers on this platform are implicitly strongly ordered with respect
89  * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
90  * with no memory barriers in this driver.  The readq()/writeq() functions add
91  * explicit ordering operation which in this case are redundant, and only
92  * add overhead.
93  */
94 
95 /* Register read/write APIs */
96 static void nic_reg_write(struct nicpf *nic, u64 offset, u64 val)
97 {
98 	writeq_relaxed(val, nic->reg_base + offset);
99 }
100 
101 static u64 nic_reg_read(struct nicpf *nic, u64 offset)
102 {
103 	return readq_relaxed(nic->reg_base + offset);
104 }
105 
106 /* PF -> VF mailbox communication APIs */
107 static void nic_enable_mbx_intr(struct nicpf *nic)
108 {
109 	int vf_cnt = pci_sriov_get_totalvfs(nic->pdev);
110 
111 #define INTR_MASK(vfs) ((vfs < 64) ? (BIT_ULL(vfs) - 1) : (~0ull))
112 
113 	/* Clear it, to avoid spurious interrupts (if any) */
114 	nic_reg_write(nic, NIC_PF_MAILBOX_INT, INTR_MASK(vf_cnt));
115 
116 	/* Enable mailbox interrupt for all VFs */
117 	nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S, INTR_MASK(vf_cnt));
118 	/* One mailbox intr enable reg per 64 VFs */
119 	if (vf_cnt > 64) {
120 		nic_reg_write(nic, NIC_PF_MAILBOX_INT + sizeof(u64),
121 			      INTR_MASK(vf_cnt - 64));
122 		nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S + sizeof(u64),
123 			      INTR_MASK(vf_cnt - 64));
124 	}
125 }
126 
127 static void nic_clear_mbx_intr(struct nicpf *nic, int vf, int mbx_reg)
128 {
129 	nic_reg_write(nic, NIC_PF_MAILBOX_INT + (mbx_reg << 3), BIT_ULL(vf));
130 }
131 
132 static u64 nic_get_mbx_addr(int vf)
133 {
134 	return NIC_PF_VF_0_127_MAILBOX_0_1 + (vf << NIC_VF_NUM_SHIFT);
135 }
136 
137 /* Send a mailbox message to VF
138  * @vf: vf to which this message to be sent
139  * @mbx: Message to be sent
140  */
141 static void nic_send_msg_to_vf(struct nicpf *nic, int vf, union nic_mbx *mbx)
142 {
143 	void __iomem *mbx_addr = nic->reg_base + nic_get_mbx_addr(vf);
144 	u64 *msg = (u64 *)mbx;
145 
146 	/* In first revision HW, mbox interrupt is triggerred
147 	 * when PF writes to MBOX(1), in next revisions when
148 	 * PF writes to MBOX(0)
149 	 */
150 	if (pass1_silicon(nic->pdev)) {
151 		/* see the comment for nic_reg_write()/nic_reg_read()
152 		 * functions above
153 		 */
154 		writeq_relaxed(msg[0], mbx_addr);
155 		writeq_relaxed(msg[1], mbx_addr + 8);
156 	} else {
157 		writeq_relaxed(msg[1], mbx_addr + 8);
158 		writeq_relaxed(msg[0], mbx_addr);
159 	}
160 }
161 
162 /* Responds to VF's READY message with VF's
163  * ID, node, MAC address e.t.c
164  * @vf: VF which sent READY message
165  */
166 static void nic_mbx_send_ready(struct nicpf *nic, int vf)
167 {
168 	union nic_mbx mbx = {};
169 	int bgx_idx, lmac;
170 	const char *mac;
171 
172 	mbx.nic_cfg.msg = NIC_MBOX_MSG_READY;
173 	mbx.nic_cfg.vf_id = vf;
174 
175 	mbx.nic_cfg.tns_mode = NIC_TNS_BYPASS_MODE;
176 
177 	if (vf < nic->num_vf_en) {
178 		bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
179 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
180 
181 		mac = bgx_get_lmac_mac(nic->node, bgx_idx, lmac);
182 		if (mac)
183 			ether_addr_copy((u8 *)&mbx.nic_cfg.mac_addr, mac);
184 	}
185 	mbx.nic_cfg.sqs_mode = (vf >= nic->num_vf_en) ? true : false;
186 	mbx.nic_cfg.node_id = nic->node;
187 
188 	mbx.nic_cfg.loopback_supported = vf < nic->num_vf_en;
189 
190 	nic_send_msg_to_vf(nic, vf, &mbx);
191 }
192 
193 /* ACKs VF's mailbox message
194  * @vf: VF to which ACK to be sent
195  */
196 static void nic_mbx_send_ack(struct nicpf *nic, int vf)
197 {
198 	union nic_mbx mbx = {};
199 
200 	mbx.msg.msg = NIC_MBOX_MSG_ACK;
201 	nic_send_msg_to_vf(nic, vf, &mbx);
202 }
203 
204 /* NACKs VF's mailbox message that PF is not able to
205  * complete the action
206  * @vf: VF to which ACK to be sent
207  */
208 static void nic_mbx_send_nack(struct nicpf *nic, int vf)
209 {
210 	union nic_mbx mbx = {};
211 
212 	mbx.msg.msg = NIC_MBOX_MSG_NACK;
213 	nic_send_msg_to_vf(nic, vf, &mbx);
214 }
215 
216 /* Flush all in flight receive packets to memory and
217  * bring down an active RQ
218  */
219 static int nic_rcv_queue_sw_sync(struct nicpf *nic)
220 {
221 	u16 timeout = ~0x00;
222 
223 	nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x01);
224 	/* Wait till sync cycle is finished */
225 	while (timeout) {
226 		if (nic_reg_read(nic, NIC_PF_SW_SYNC_RX_DONE) & 0x1)
227 			break;
228 		timeout--;
229 	}
230 	nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x00);
231 	if (!timeout) {
232 		dev_err(&nic->pdev->dev, "Receive queue software sync failed");
233 		return 1;
234 	}
235 	return 0;
236 }
237 
238 /* Get BGX Rx/Tx stats and respond to VF's request */
239 static void nic_get_bgx_stats(struct nicpf *nic, struct bgx_stats_msg *bgx)
240 {
241 	int bgx_idx, lmac;
242 	union nic_mbx mbx = {};
243 
244 	bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
245 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
246 
247 	mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
248 	mbx.bgx_stats.vf_id = bgx->vf_id;
249 	mbx.bgx_stats.rx = bgx->rx;
250 	mbx.bgx_stats.idx = bgx->idx;
251 	if (bgx->rx)
252 		mbx.bgx_stats.stats = bgx_get_rx_stats(nic->node, bgx_idx,
253 							    lmac, bgx->idx);
254 	else
255 		mbx.bgx_stats.stats = bgx_get_tx_stats(nic->node, bgx_idx,
256 							    lmac, bgx->idx);
257 	nic_send_msg_to_vf(nic, bgx->vf_id, &mbx);
258 }
259 
260 /* Update hardware min/max frame size */
261 static int nic_update_hw_frs(struct nicpf *nic, int new_frs, int vf)
262 {
263 	if ((new_frs > NIC_HW_MAX_FRS) || (new_frs < NIC_HW_MIN_FRS)) {
264 		dev_err(&nic->pdev->dev,
265 			"Invalid MTU setting from VF%d rejected, should be between %d and %d\n",
266 			   vf, NIC_HW_MIN_FRS, NIC_HW_MAX_FRS);
267 		return 1;
268 	}
269 	new_frs += ETH_HLEN;
270 	if (new_frs <= nic->pkind.maxlen)
271 		return 0;
272 
273 	nic->pkind.maxlen = new_frs;
274 	nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG, *(u64 *)&nic->pkind);
275 	return 0;
276 }
277 
278 /* Set minimum transmit packet size */
279 static void nic_set_tx_pkt_pad(struct nicpf *nic, int size)
280 {
281 	int lmac, max_lmac;
282 	u16 sdevid;
283 	u64 lmac_cfg;
284 
285 	/* There is a issue in HW where-in while sending GSO sized
286 	 * pkts as part of TSO, if pkt len falls below this size
287 	 * NIC will zero PAD packet and also updates IP total length.
288 	 * Hence set this value to lessthan min pkt size of MAC+IP+TCP
289 	 * headers, BGX will do the padding to transmit 64 byte pkt.
290 	 */
291 	if (size > 52)
292 		size = 52;
293 
294 	pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
295 	/* 81xx's RGX has only one LMAC */
296 	if (sdevid == PCI_SUBSYS_DEVID_81XX_NIC_PF)
297 		max_lmac = ((nic->hw->bgx_cnt - 1) * MAX_LMAC_PER_BGX) + 1;
298 	else
299 		max_lmac = nic->hw->bgx_cnt * MAX_LMAC_PER_BGX;
300 
301 	for (lmac = 0; lmac < max_lmac; lmac++) {
302 		lmac_cfg = nic_reg_read(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3));
303 		lmac_cfg &= ~(0xF << 2);
304 		lmac_cfg |= ((size / 4) << 2);
305 		nic_reg_write(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3), lmac_cfg);
306 	}
307 }
308 
309 /* Function to check number of LMACs present and set VF::LMAC mapping.
310  * Mapping will be used while initializing channels.
311  */
312 static void nic_set_lmac_vf_mapping(struct nicpf *nic)
313 {
314 	unsigned bgx_map = bgx_get_map(nic->node);
315 	int bgx, next_bgx_lmac = 0;
316 	int lmac, lmac_cnt = 0;
317 	u64 lmac_credit;
318 
319 	nic->num_vf_en = 0;
320 
321 	for (bgx = 0; bgx < nic->hw->bgx_cnt; bgx++) {
322 		if (!(bgx_map & (1 << bgx)))
323 			continue;
324 		lmac_cnt = bgx_get_lmac_count(nic->node, bgx);
325 		for (lmac = 0; lmac < lmac_cnt; lmac++)
326 			nic->vf_lmac_map[next_bgx_lmac++] =
327 						NIC_SET_VF_LMAC_MAP(bgx, lmac);
328 		nic->num_vf_en += lmac_cnt;
329 
330 		/* Program LMAC credits */
331 		lmac_credit = (1ull << 1); /* channel credit enable */
332 		lmac_credit |= (0x1ff << 2); /* Max outstanding pkt count */
333 		/* 48KB BGX Tx buffer size, each unit is of size 16bytes */
334 		lmac_credit |= (((((48 * 1024) / lmac_cnt) -
335 				NIC_HW_MAX_FRS) / 16) << 12);
336 		lmac = bgx * MAX_LMAC_PER_BGX;
337 		for (; lmac < lmac_cnt + (bgx * MAX_LMAC_PER_BGX); lmac++)
338 			nic_reg_write(nic,
339 				      NIC_PF_LMAC_0_7_CREDIT + (lmac * 8),
340 				      lmac_credit);
341 
342 		/* On CN81XX there are only 8 VFs but max possible no of
343 		 * interfaces are 9.
344 		 */
345 		if (nic->num_vf_en >= pci_sriov_get_totalvfs(nic->pdev)) {
346 			nic->num_vf_en = pci_sriov_get_totalvfs(nic->pdev);
347 			break;
348 		}
349 	}
350 }
351 
352 static void nic_free_lmacmem(struct nicpf *nic)
353 {
354 	kfree(nic->vf_lmac_map);
355 	kfree(nic->link);
356 	kfree(nic->duplex);
357 	kfree(nic->speed);
358 }
359 
360 static int nic_get_hw_info(struct nicpf *nic)
361 {
362 	u8 max_lmac;
363 	u16 sdevid;
364 	struct hw_info *hw = nic->hw;
365 
366 	pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
367 
368 	switch (sdevid) {
369 	case PCI_SUBSYS_DEVID_88XX_NIC_PF:
370 		hw->bgx_cnt = MAX_BGX_PER_CN88XX;
371 		hw->chans_per_lmac = 16;
372 		hw->chans_per_bgx = 128;
373 		hw->cpi_cnt = 2048;
374 		hw->rssi_cnt = 4096;
375 		hw->rss_ind_tbl_size = NIC_MAX_RSS_IDR_TBL_SIZE;
376 		hw->tl3_cnt = 256;
377 		hw->tl2_cnt = 64;
378 		hw->tl1_cnt = 2;
379 		hw->tl1_per_bgx = true;
380 		break;
381 	case PCI_SUBSYS_DEVID_81XX_NIC_PF:
382 		hw->bgx_cnt = MAX_BGX_PER_CN81XX;
383 		hw->chans_per_lmac = 8;
384 		hw->chans_per_bgx = 32;
385 		hw->chans_per_rgx = 8;
386 		hw->chans_per_lbk = 24;
387 		hw->cpi_cnt = 512;
388 		hw->rssi_cnt = 256;
389 		hw->rss_ind_tbl_size = 32; /* Max RSSI / Max interfaces */
390 		hw->tl3_cnt = 64;
391 		hw->tl2_cnt = 16;
392 		hw->tl1_cnt = 10;
393 		hw->tl1_per_bgx = false;
394 		break;
395 	case PCI_SUBSYS_DEVID_83XX_NIC_PF:
396 		hw->bgx_cnt = MAX_BGX_PER_CN83XX;
397 		hw->chans_per_lmac = 8;
398 		hw->chans_per_bgx = 32;
399 		hw->chans_per_lbk = 64;
400 		hw->cpi_cnt = 2048;
401 		hw->rssi_cnt = 1024;
402 		hw->rss_ind_tbl_size = 64; /* Max RSSI / Max interfaces */
403 		hw->tl3_cnt = 256;
404 		hw->tl2_cnt = 64;
405 		hw->tl1_cnt = 18;
406 		hw->tl1_per_bgx = false;
407 		break;
408 	}
409 	hw->tl4_cnt = MAX_QUEUES_PER_QSET * pci_sriov_get_totalvfs(nic->pdev);
410 
411 	/* Allocate memory for LMAC tracking elements */
412 	max_lmac = hw->bgx_cnt * MAX_LMAC_PER_BGX;
413 	nic->vf_lmac_map = kmalloc_array(max_lmac, sizeof(u8), GFP_KERNEL);
414 	if (!nic->vf_lmac_map)
415 		goto error;
416 	nic->link = kmalloc_array(max_lmac, sizeof(u8), GFP_KERNEL);
417 	if (!nic->link)
418 		goto error;
419 	nic->duplex = kmalloc_array(max_lmac, sizeof(u8), GFP_KERNEL);
420 	if (!nic->duplex)
421 		goto error;
422 	nic->speed = kmalloc_array(max_lmac, sizeof(u32), GFP_KERNEL);
423 	if (!nic->speed)
424 		goto error;
425 	return 0;
426 
427 error:
428 	nic_free_lmacmem(nic);
429 	return -ENOMEM;
430 }
431 
432 #define BGX0_BLOCK 8
433 #define BGX1_BLOCK 9
434 
435 static int nic_init_hw(struct nicpf *nic)
436 {
437 	int i, err;
438 	u64 cqm_cfg;
439 
440 	/* Get HW capability info */
441 	err = nic_get_hw_info(nic);
442 	if (err)
443 		return err;
444 
445 	/* Enable NIC HW block */
446 	nic_reg_write(nic, NIC_PF_CFG, 0x3);
447 
448 	/* Enable backpressure */
449 	nic_reg_write(nic, NIC_PF_BP_CFG, (1ULL << 6) | 0x03);
450 
451 	/* TNS and TNS bypass modes are present only on 88xx */
452 	if (nic->pdev->subsystem_device == PCI_SUBSYS_DEVID_88XX_NIC_PF) {
453 		/* Disable TNS mode on both interfaces */
454 		nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG,
455 			      (NIC_TNS_BYPASS_MODE << 7) | BGX0_BLOCK);
456 		nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG | (1 << 8),
457 			      (NIC_TNS_BYPASS_MODE << 7) | BGX1_BLOCK);
458 	}
459 
460 	nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG,
461 		      (1ULL << 63) | BGX0_BLOCK);
462 	nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG + (1 << 8),
463 		      (1ULL << 63) | BGX1_BLOCK);
464 
465 	/* PKIND configuration */
466 	nic->pkind.minlen = 0;
467 	nic->pkind.maxlen = NIC_HW_MAX_FRS + ETH_HLEN;
468 	nic->pkind.lenerr_en = 1;
469 	nic->pkind.rx_hdr = 0;
470 	nic->pkind.hdr_sl = 0;
471 
472 	for (i = 0; i < NIC_MAX_PKIND; i++)
473 		nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG | (i << 3),
474 			      *(u64 *)&nic->pkind);
475 
476 	nic_set_tx_pkt_pad(nic, NIC_HW_MIN_FRS);
477 
478 	/* Timer config */
479 	nic_reg_write(nic, NIC_PF_INTR_TIMER_CFG, NICPF_CLK_PER_INT_TICK);
480 
481 	/* Enable VLAN ethertype matching and stripping */
482 	nic_reg_write(nic, NIC_PF_RX_ETYPE_0_7,
483 		      (2 << 19) | (ETYPE_ALG_VLAN_STRIP << 16) | ETH_P_8021Q);
484 
485 	/* Check if HW expected value is higher (could be in future chips) */
486 	cqm_cfg = nic_reg_read(nic, NIC_PF_CQM_CFG);
487 	if (cqm_cfg < NICPF_CQM_MIN_DROP_LEVEL)
488 		nic_reg_write(nic, NIC_PF_CQM_CFG, NICPF_CQM_MIN_DROP_LEVEL);
489 
490 	return 0;
491 }
492 
493 /* Channel parse index configuration */
494 static void nic_config_cpi(struct nicpf *nic, struct cpi_cfg_msg *cfg)
495 {
496 	struct hw_info *hw = nic->hw;
497 	u32 vnic, bgx, lmac, chan;
498 	u32 padd, cpi_count = 0;
499 	u64 cpi_base, cpi, rssi_base, rssi;
500 	u8  qset, rq_idx = 0;
501 
502 	vnic = cfg->vf_id;
503 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
504 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
505 
506 	chan = (lmac * hw->chans_per_lmac) + (bgx * hw->chans_per_bgx);
507 	cpi_base = vnic * NIC_MAX_CPI_PER_LMAC;
508 	rssi_base = vnic * hw->rss_ind_tbl_size;
509 
510 	/* Rx channel configuration */
511 	nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_BP_CFG | (chan << 3),
512 		      (1ull << 63) | (vnic << 0));
513 	nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_CFG | (chan << 3),
514 		      ((u64)cfg->cpi_alg << 62) | (cpi_base << 48));
515 
516 	if (cfg->cpi_alg == CPI_ALG_NONE)
517 		cpi_count = 1;
518 	else if (cfg->cpi_alg == CPI_ALG_VLAN) /* 3 bits of PCP */
519 		cpi_count = 8;
520 	else if (cfg->cpi_alg == CPI_ALG_VLAN16) /* 3 bits PCP + DEI */
521 		cpi_count = 16;
522 	else if (cfg->cpi_alg == CPI_ALG_DIFF) /* 6bits DSCP */
523 		cpi_count = NIC_MAX_CPI_PER_LMAC;
524 
525 	/* RSS Qset, Qidx mapping */
526 	qset = cfg->vf_id;
527 	rssi = rssi_base;
528 	for (; rssi < (rssi_base + cfg->rq_cnt); rssi++) {
529 		nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
530 			      (qset << 3) | rq_idx);
531 		rq_idx++;
532 	}
533 
534 	rssi = 0;
535 	cpi = cpi_base;
536 	for (; cpi < (cpi_base + cpi_count); cpi++) {
537 		/* Determine port to channel adder */
538 		if (cfg->cpi_alg != CPI_ALG_DIFF)
539 			padd = cpi % cpi_count;
540 		else
541 			padd = cpi % 8; /* 3 bits CS out of 6bits DSCP */
542 
543 		/* Leave RSS_SIZE as '0' to disable RSS */
544 		if (pass1_silicon(nic->pdev)) {
545 			nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
546 				      (vnic << 24) | (padd << 16) |
547 				      (rssi_base + rssi));
548 		} else {
549 			/* Set MPI_ALG to '0' to disable MCAM parsing */
550 			nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
551 				      (padd << 16));
552 			/* MPI index is same as CPI if MPI_ALG is not enabled */
553 			nic_reg_write(nic, NIC_PF_MPI_0_2047_CFG | (cpi << 3),
554 				      (vnic << 24) | (rssi_base + rssi));
555 		}
556 
557 		if ((rssi + 1) >= cfg->rq_cnt)
558 			continue;
559 
560 		if (cfg->cpi_alg == CPI_ALG_VLAN)
561 			rssi++;
562 		else if (cfg->cpi_alg == CPI_ALG_VLAN16)
563 			rssi = ((cpi - cpi_base) & 0xe) >> 1;
564 		else if (cfg->cpi_alg == CPI_ALG_DIFF)
565 			rssi = ((cpi - cpi_base) & 0x38) >> 3;
566 	}
567 	nic->cpi_base[cfg->vf_id] = cpi_base;
568 	nic->rssi_base[cfg->vf_id] = rssi_base;
569 }
570 
571 /* Responsds to VF with its RSS indirection table size */
572 static void nic_send_rss_size(struct nicpf *nic, int vf)
573 {
574 	union nic_mbx mbx = {};
575 	u64  *msg;
576 
577 	msg = (u64 *)&mbx;
578 
579 	mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
580 	mbx.rss_size.ind_tbl_size = nic->hw->rss_ind_tbl_size;
581 	nic_send_msg_to_vf(nic, vf, &mbx);
582 }
583 
584 /* Receive side scaling configuration
585  * configure:
586  * - RSS index
587  * - indir table i.e hash::RQ mapping
588  * - no of hash bits to consider
589  */
590 static void nic_config_rss(struct nicpf *nic, struct rss_cfg_msg *cfg)
591 {
592 	u8  qset, idx = 0;
593 	u64 cpi_cfg, cpi_base, rssi_base, rssi;
594 	u64 idx_addr;
595 
596 	rssi_base = nic->rssi_base[cfg->vf_id] + cfg->tbl_offset;
597 
598 	rssi = rssi_base;
599 	qset = cfg->vf_id;
600 
601 	for (; rssi < (rssi_base + cfg->tbl_len); rssi++) {
602 		u8 svf = cfg->ind_tbl[idx] >> 3;
603 
604 		if (svf)
605 			qset = nic->vf_sqs[cfg->vf_id][svf - 1];
606 		else
607 			qset = cfg->vf_id;
608 		nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
609 			      (qset << 3) | (cfg->ind_tbl[idx] & 0x7));
610 		idx++;
611 	}
612 
613 	cpi_base = nic->cpi_base[cfg->vf_id];
614 	if (pass1_silicon(nic->pdev))
615 		idx_addr = NIC_PF_CPI_0_2047_CFG;
616 	else
617 		idx_addr = NIC_PF_MPI_0_2047_CFG;
618 	cpi_cfg = nic_reg_read(nic, idx_addr | (cpi_base << 3));
619 	cpi_cfg &= ~(0xFULL << 20);
620 	cpi_cfg |= (cfg->hash_bits << 20);
621 	nic_reg_write(nic, idx_addr | (cpi_base << 3), cpi_cfg);
622 }
623 
624 /* 4 level transmit side scheduler configutation
625  * for TNS bypass mode
626  *
627  * Sample configuration for SQ0 on 88xx
628  * VNIC0-SQ0 -> TL4(0)   -> TL3[0]   -> TL2[0]  -> TL1[0] -> BGX0
629  * VNIC1-SQ0 -> TL4(8)   -> TL3[2]   -> TL2[0]  -> TL1[0] -> BGX0
630  * VNIC2-SQ0 -> TL4(16)  -> TL3[4]   -> TL2[1]  -> TL1[0] -> BGX0
631  * VNIC3-SQ0 -> TL4(24)  -> TL3[6]   -> TL2[1]  -> TL1[0] -> BGX0
632  * VNIC4-SQ0 -> TL4(512) -> TL3[128] -> TL2[32] -> TL1[1] -> BGX1
633  * VNIC5-SQ0 -> TL4(520) -> TL3[130] -> TL2[32] -> TL1[1] -> BGX1
634  * VNIC6-SQ0 -> TL4(528) -> TL3[132] -> TL2[33] -> TL1[1] -> BGX1
635  * VNIC7-SQ0 -> TL4(536) -> TL3[134] -> TL2[33] -> TL1[1] -> BGX1
636  */
637 static void nic_tx_channel_cfg(struct nicpf *nic, u8 vnic,
638 			       struct sq_cfg_msg *sq)
639 {
640 	struct hw_info *hw = nic->hw;
641 	u32 bgx, lmac, chan;
642 	u32 tl2, tl3, tl4;
643 	u32 rr_quantum;
644 	u8 sq_idx = sq->sq_num;
645 	u8 pqs_vnic;
646 	int svf;
647 
648 	if (sq->sqs_mode)
649 		pqs_vnic = nic->pqs_vf[vnic];
650 	else
651 		pqs_vnic = vnic;
652 
653 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
654 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
655 
656 	/* 24 bytes for FCS, IPG and preamble */
657 	rr_quantum = ((NIC_HW_MAX_FRS + 24) / 4);
658 
659 	/* For 88xx 0-511 TL4 transmits via BGX0 and
660 	 * 512-1023 TL4s transmit via BGX1.
661 	 */
662 	if (hw->tl1_per_bgx) {
663 		tl4 = bgx * (hw->tl4_cnt / hw->bgx_cnt);
664 		if (!sq->sqs_mode) {
665 			tl4 += (lmac * MAX_QUEUES_PER_QSET);
666 		} else {
667 			for (svf = 0; svf < MAX_SQS_PER_VF; svf++) {
668 				if (nic->vf_sqs[pqs_vnic][svf] == vnic)
669 					break;
670 			}
671 			tl4 += (MAX_LMAC_PER_BGX * MAX_QUEUES_PER_QSET);
672 			tl4 += (lmac * MAX_QUEUES_PER_QSET * MAX_SQS_PER_VF);
673 			tl4 += (svf * MAX_QUEUES_PER_QSET);
674 		}
675 	} else {
676 		tl4 = (vnic * MAX_QUEUES_PER_QSET);
677 	}
678 	tl4 += sq_idx;
679 
680 	tl3 = tl4 / (hw->tl4_cnt / hw->tl3_cnt);
681 	nic_reg_write(nic, NIC_PF_QSET_0_127_SQ_0_7_CFG2 |
682 		      ((u64)vnic << NIC_QS_ID_SHIFT) |
683 		      ((u32)sq_idx << NIC_Q_NUM_SHIFT), tl4);
684 	nic_reg_write(nic, NIC_PF_TL4_0_1023_CFG | (tl4 << 3),
685 		      ((u64)vnic << 27) | ((u32)sq_idx << 24) | rr_quantum);
686 
687 	nic_reg_write(nic, NIC_PF_TL3_0_255_CFG | (tl3 << 3), rr_quantum);
688 
689 	/* On 88xx 0-127 channels are for BGX0 and
690 	 * 127-255 channels for BGX1.
691 	 *
692 	 * On 81xx/83xx TL3_CHAN reg should be configured with channel
693 	 * within LMAC i.e 0-7 and not the actual channel number like on 88xx
694 	 */
695 	chan = (lmac * hw->chans_per_lmac) + (bgx * hw->chans_per_bgx);
696 	if (hw->tl1_per_bgx)
697 		nic_reg_write(nic, NIC_PF_TL3_0_255_CHAN | (tl3 << 3), chan);
698 	else
699 		nic_reg_write(nic, NIC_PF_TL3_0_255_CHAN | (tl3 << 3), 0);
700 
701 	/* Enable backpressure on the channel */
702 	nic_reg_write(nic, NIC_PF_CHAN_0_255_TX_CFG | (chan << 3), 1);
703 
704 	tl2 = tl3 >> 2;
705 	nic_reg_write(nic, NIC_PF_TL3A_0_63_CFG | (tl2 << 3), tl2);
706 	nic_reg_write(nic, NIC_PF_TL2_0_63_CFG | (tl2 << 3), rr_quantum);
707 	/* No priorities as of now */
708 	nic_reg_write(nic, NIC_PF_TL2_0_63_PRI | (tl2 << 3), 0x00);
709 
710 	/* Unlike 88xx where TL2s 0-31 transmits to TL1 '0' and rest to TL1 '1'
711 	 * on 81xx/83xx TL2 needs to be configured to transmit to one of the
712 	 * possible LMACs.
713 	 *
714 	 * This register doesn't exist on 88xx.
715 	 */
716 	if (!hw->tl1_per_bgx)
717 		nic_reg_write(nic, NIC_PF_TL2_LMAC | (tl2 << 3),
718 			      lmac + (bgx * MAX_LMAC_PER_BGX));
719 }
720 
721 /* Send primary nicvf pointer to secondary QS's VF */
722 static void nic_send_pnicvf(struct nicpf *nic, int sqs)
723 {
724 	union nic_mbx mbx = {};
725 
726 	mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
727 	mbx.nicvf.nicvf = nic->nicvf[nic->pqs_vf[sqs]];
728 	nic_send_msg_to_vf(nic, sqs, &mbx);
729 }
730 
731 /* Send SQS's nicvf pointer to primary QS's VF */
732 static void nic_send_snicvf(struct nicpf *nic, struct nicvf_ptr *nicvf)
733 {
734 	union nic_mbx mbx = {};
735 	int sqs_id = nic->vf_sqs[nicvf->vf_id][nicvf->sqs_id];
736 
737 	mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
738 	mbx.nicvf.sqs_id = nicvf->sqs_id;
739 	mbx.nicvf.nicvf = nic->nicvf[sqs_id];
740 	nic_send_msg_to_vf(nic, nicvf->vf_id, &mbx);
741 }
742 
743 /* Find next available Qset that can be assigned as a
744  * secondary Qset to a VF.
745  */
746 static int nic_nxt_avail_sqs(struct nicpf *nic)
747 {
748 	int sqs;
749 
750 	for (sqs = 0; sqs < nic->num_sqs_en; sqs++) {
751 		if (!nic->sqs_used[sqs])
752 			nic->sqs_used[sqs] = true;
753 		else
754 			continue;
755 		return sqs + nic->num_vf_en;
756 	}
757 	return -1;
758 }
759 
760 /* Allocate additional Qsets for requested VF */
761 static void nic_alloc_sqs(struct nicpf *nic, struct sqs_alloc *sqs)
762 {
763 	union nic_mbx mbx = {};
764 	int idx, alloc_qs = 0;
765 	int sqs_id;
766 
767 	if (!nic->num_sqs_en)
768 		goto send_mbox;
769 
770 	for (idx = 0; idx < sqs->qs_count; idx++) {
771 		sqs_id = nic_nxt_avail_sqs(nic);
772 		if (sqs_id < 0)
773 			break;
774 		nic->vf_sqs[sqs->vf_id][idx] = sqs_id;
775 		nic->pqs_vf[sqs_id] = sqs->vf_id;
776 		alloc_qs++;
777 	}
778 
779 send_mbox:
780 	mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
781 	mbx.sqs_alloc.vf_id = sqs->vf_id;
782 	mbx.sqs_alloc.qs_count = alloc_qs;
783 	nic_send_msg_to_vf(nic, sqs->vf_id, &mbx);
784 }
785 
786 static int nic_config_loopback(struct nicpf *nic, struct set_loopback *lbk)
787 {
788 	int bgx_idx, lmac_idx;
789 
790 	if (lbk->vf_id >= nic->num_vf_en)
791 		return -1;
792 
793 	bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
794 	lmac_idx = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
795 
796 	bgx_lmac_internal_loopback(nic->node, bgx_idx, lmac_idx, lbk->enable);
797 
798 	return 0;
799 }
800 
801 /* Reset statistics counters */
802 static int nic_reset_stat_counters(struct nicpf *nic,
803 				   int vf, struct reset_stat_cfg *cfg)
804 {
805 	int i, stat, qnum;
806 	u64 reg_addr;
807 
808 	for (i = 0; i < RX_STATS_ENUM_LAST; i++) {
809 		if (cfg->rx_stat_mask & BIT(i)) {
810 			reg_addr = NIC_PF_VNIC_0_127_RX_STAT_0_13 |
811 				   (vf << NIC_QS_ID_SHIFT) |
812 				   (i << 3);
813 			nic_reg_write(nic, reg_addr, 0);
814 		}
815 	}
816 
817 	for (i = 0; i < TX_STATS_ENUM_LAST; i++) {
818 		if (cfg->tx_stat_mask & BIT(i)) {
819 			reg_addr = NIC_PF_VNIC_0_127_TX_STAT_0_4 |
820 				   (vf << NIC_QS_ID_SHIFT) |
821 				   (i << 3);
822 			nic_reg_write(nic, reg_addr, 0);
823 		}
824 	}
825 
826 	for (i = 0; i <= 15; i++) {
827 		qnum = i >> 1;
828 		stat = i & 1 ? 1 : 0;
829 		reg_addr = (vf << NIC_QS_ID_SHIFT) |
830 			   (qnum << NIC_Q_NUM_SHIFT) | (stat << 3);
831 		if (cfg->rq_stat_mask & BIT(i)) {
832 			reg_addr |= NIC_PF_QSET_0_127_RQ_0_7_STAT_0_1;
833 			nic_reg_write(nic, reg_addr, 0);
834 		}
835 		if (cfg->sq_stat_mask & BIT(i)) {
836 			reg_addr |= NIC_PF_QSET_0_127_SQ_0_7_STAT_0_1;
837 			nic_reg_write(nic, reg_addr, 0);
838 		}
839 	}
840 	return 0;
841 }
842 
843 static void nic_enable_tunnel_parsing(struct nicpf *nic, int vf)
844 {
845 	u64 prot_def = (IPV6_PROT << 32) | (IPV4_PROT << 16) | ET_PROT;
846 	u64 vxlan_prot_def = (IPV6_PROT_DEF << 32) |
847 			      (IPV4_PROT_DEF) << 16 | ET_PROT_DEF;
848 
849 	/* Configure tunnel parsing parameters */
850 	nic_reg_write(nic, NIC_PF_RX_GENEVE_DEF,
851 		      (1ULL << 63 | UDP_GENEVE_PORT_NUM));
852 	nic_reg_write(nic, NIC_PF_RX_GENEVE_PROT_DEF,
853 		      ((7ULL << 61) | prot_def));
854 	nic_reg_write(nic, NIC_PF_RX_NVGRE_PROT_DEF,
855 		      ((7ULL << 61) | prot_def));
856 	nic_reg_write(nic, NIC_PF_RX_VXLAN_DEF_0_1,
857 		      ((1ULL << 63) | UDP_VXLAN_PORT_NUM));
858 	nic_reg_write(nic, NIC_PF_RX_VXLAN_PROT_DEF,
859 		      ((0xfULL << 60) | vxlan_prot_def));
860 }
861 
862 static void nic_enable_vf(struct nicpf *nic, int vf, bool enable)
863 {
864 	int bgx, lmac;
865 
866 	nic->vf_enabled[vf] = enable;
867 
868 	if (vf >= nic->num_vf_en)
869 		return;
870 
871 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
872 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
873 
874 	bgx_lmac_rx_tx_enable(nic->node, bgx, lmac, enable);
875 }
876 
877 /* Interrupt handler to handle mailbox messages from VFs */
878 static void nic_handle_mbx_intr(struct nicpf *nic, int vf)
879 {
880 	union nic_mbx mbx = {};
881 	u64 *mbx_data;
882 	u64 mbx_addr;
883 	u64 reg_addr;
884 	u64 cfg;
885 	int bgx, lmac;
886 	int i;
887 	int ret = 0;
888 
889 	nic->mbx_lock[vf] = true;
890 
891 	mbx_addr = nic_get_mbx_addr(vf);
892 	mbx_data = (u64 *)&mbx;
893 
894 	for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
895 		*mbx_data = nic_reg_read(nic, mbx_addr);
896 		mbx_data++;
897 		mbx_addr += sizeof(u64);
898 	}
899 
900 	dev_dbg(&nic->pdev->dev, "%s: Mailbox msg 0x%02x from VF%d\n",
901 		__func__, mbx.msg.msg, vf);
902 	switch (mbx.msg.msg) {
903 	case NIC_MBOX_MSG_READY:
904 		nic_mbx_send_ready(nic, vf);
905 		if (vf < nic->num_vf_en) {
906 			nic->link[vf] = 0;
907 			nic->duplex[vf] = 0;
908 			nic->speed[vf] = 0;
909 		}
910 		goto unlock;
911 	case NIC_MBOX_MSG_QS_CFG:
912 		reg_addr = NIC_PF_QSET_0_127_CFG |
913 			   (mbx.qs.num << NIC_QS_ID_SHIFT);
914 		cfg = mbx.qs.cfg;
915 		/* Check if its a secondary Qset */
916 		if (vf >= nic->num_vf_en) {
917 			cfg = cfg & (~0x7FULL);
918 			/* Assign this Qset to primary Qset's VF */
919 			cfg |= nic->pqs_vf[vf];
920 		}
921 		nic_reg_write(nic, reg_addr, cfg);
922 		break;
923 	case NIC_MBOX_MSG_RQ_CFG:
924 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_CFG |
925 			   (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
926 			   (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
927 		nic_reg_write(nic, reg_addr, mbx.rq.cfg);
928 		/* Enable CQE_RX2_S extension in CQE_RX descriptor.
929 		 * This gets appended by default on 81xx/83xx chips,
930 		 * for consistency enabling the same on 88xx pass2
931 		 * where this is introduced.
932 		 */
933 		if (pass2_silicon(nic->pdev))
934 			nic_reg_write(nic, NIC_PF_RX_CFG, 0x01);
935 		if (!pass1_silicon(nic->pdev))
936 			nic_enable_tunnel_parsing(nic, vf);
937 		break;
938 	case NIC_MBOX_MSG_RQ_BP_CFG:
939 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_BP_CFG |
940 			   (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
941 			   (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
942 		nic_reg_write(nic, reg_addr, mbx.rq.cfg);
943 		break;
944 	case NIC_MBOX_MSG_RQ_SW_SYNC:
945 		ret = nic_rcv_queue_sw_sync(nic);
946 		break;
947 	case NIC_MBOX_MSG_RQ_DROP_CFG:
948 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_DROP_CFG |
949 			   (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
950 			   (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
951 		nic_reg_write(nic, reg_addr, mbx.rq.cfg);
952 		break;
953 	case NIC_MBOX_MSG_SQ_CFG:
954 		reg_addr = NIC_PF_QSET_0_127_SQ_0_7_CFG |
955 			   (mbx.sq.qs_num << NIC_QS_ID_SHIFT) |
956 			   (mbx.sq.sq_num << NIC_Q_NUM_SHIFT);
957 		nic_reg_write(nic, reg_addr, mbx.sq.cfg);
958 		nic_tx_channel_cfg(nic, mbx.qs.num, &mbx.sq);
959 		break;
960 	case NIC_MBOX_MSG_SET_MAC:
961 		if (vf >= nic->num_vf_en) {
962 			ret = -1; /* NACK */
963 			break;
964 		}
965 		lmac = mbx.mac.vf_id;
966 		bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
967 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
968 		bgx_set_lmac_mac(nic->node, bgx, lmac, mbx.mac.mac_addr);
969 		break;
970 	case NIC_MBOX_MSG_SET_MAX_FRS:
971 		ret = nic_update_hw_frs(nic, mbx.frs.max_frs,
972 					mbx.frs.vf_id);
973 		break;
974 	case NIC_MBOX_MSG_CPI_CFG:
975 		nic_config_cpi(nic, &mbx.cpi_cfg);
976 		break;
977 	case NIC_MBOX_MSG_RSS_SIZE:
978 		nic_send_rss_size(nic, vf);
979 		goto unlock;
980 	case NIC_MBOX_MSG_RSS_CFG:
981 	case NIC_MBOX_MSG_RSS_CFG_CONT:
982 		nic_config_rss(nic, &mbx.rss_cfg);
983 		break;
984 	case NIC_MBOX_MSG_CFG_DONE:
985 		/* Last message of VF config msg sequence */
986 		nic_enable_vf(nic, vf, true);
987 		goto unlock;
988 	case NIC_MBOX_MSG_SHUTDOWN:
989 		/* First msg in VF teardown sequence */
990 		if (vf >= nic->num_vf_en)
991 			nic->sqs_used[vf - nic->num_vf_en] = false;
992 		nic->pqs_vf[vf] = 0;
993 		nic_enable_vf(nic, vf, false);
994 		break;
995 	case NIC_MBOX_MSG_ALLOC_SQS:
996 		nic_alloc_sqs(nic, &mbx.sqs_alloc);
997 		goto unlock;
998 	case NIC_MBOX_MSG_NICVF_PTR:
999 		nic->nicvf[vf] = mbx.nicvf.nicvf;
1000 		break;
1001 	case NIC_MBOX_MSG_PNICVF_PTR:
1002 		nic_send_pnicvf(nic, vf);
1003 		goto unlock;
1004 	case NIC_MBOX_MSG_SNICVF_PTR:
1005 		nic_send_snicvf(nic, &mbx.nicvf);
1006 		goto unlock;
1007 	case NIC_MBOX_MSG_BGX_STATS:
1008 		nic_get_bgx_stats(nic, &mbx.bgx_stats);
1009 		goto unlock;
1010 	case NIC_MBOX_MSG_LOOPBACK:
1011 		ret = nic_config_loopback(nic, &mbx.lbk);
1012 		break;
1013 	case NIC_MBOX_MSG_RESET_STAT_COUNTER:
1014 		ret = nic_reset_stat_counters(nic, vf, &mbx.reset_stat);
1015 		break;
1016 	default:
1017 		dev_err(&nic->pdev->dev,
1018 			"Invalid msg from VF%d, msg 0x%x\n", vf, mbx.msg.msg);
1019 		break;
1020 	}
1021 
1022 	if (!ret) {
1023 		nic_mbx_send_ack(nic, vf);
1024 	} else if (mbx.msg.msg != NIC_MBOX_MSG_READY) {
1025 		dev_err(&nic->pdev->dev, "NACK for MBOX 0x%02x from VF %d\n",
1026 			mbx.msg.msg, vf);
1027 		nic_mbx_send_nack(nic, vf);
1028 	}
1029 unlock:
1030 	nic->mbx_lock[vf] = false;
1031 }
1032 
1033 static irqreturn_t nic_mbx_intr_handler(int irq, void *nic_irq)
1034 {
1035 	struct nicpf *nic = (struct nicpf *)nic_irq;
1036 	int mbx;
1037 	u64 intr;
1038 	u8  vf, vf_per_mbx_reg = 64;
1039 
1040 	if (irq == nic->msix_entries[NIC_PF_INTR_ID_MBOX0].vector)
1041 		mbx = 0;
1042 	else
1043 		mbx = 1;
1044 
1045 	intr = nic_reg_read(nic, NIC_PF_MAILBOX_INT + (mbx << 3));
1046 	dev_dbg(&nic->pdev->dev, "PF interrupt Mbox%d 0x%llx\n", mbx, intr);
1047 	for (vf = 0; vf < vf_per_mbx_reg; vf++) {
1048 		if (intr & (1ULL << vf)) {
1049 			dev_dbg(&nic->pdev->dev, "Intr from VF %d\n",
1050 				vf + (mbx * vf_per_mbx_reg));
1051 
1052 			nic_handle_mbx_intr(nic, vf + (mbx * vf_per_mbx_reg));
1053 			nic_clear_mbx_intr(nic, vf, mbx);
1054 		}
1055 	}
1056 	return IRQ_HANDLED;
1057 }
1058 
1059 static int nic_enable_msix(struct nicpf *nic)
1060 {
1061 	int i, ret;
1062 
1063 	nic->num_vec = pci_msix_vec_count(nic->pdev);
1064 
1065 	nic->msix_entries = kmalloc_array(nic->num_vec,
1066 					  sizeof(struct msix_entry),
1067 					  GFP_KERNEL);
1068 	if (!nic->msix_entries)
1069 		return -ENOMEM;
1070 
1071 	for (i = 0; i < nic->num_vec; i++)
1072 		nic->msix_entries[i].entry = i;
1073 
1074 	ret = pci_enable_msix(nic->pdev, nic->msix_entries, nic->num_vec);
1075 	if (ret) {
1076 		dev_err(&nic->pdev->dev,
1077 			"Request for #%d msix vectors failed, returned %d\n",
1078 			   nic->num_vec, ret);
1079 		kfree(nic->msix_entries);
1080 		return ret;
1081 	}
1082 
1083 	nic->msix_enabled = 1;
1084 	return 0;
1085 }
1086 
1087 static void nic_disable_msix(struct nicpf *nic)
1088 {
1089 	if (nic->msix_enabled) {
1090 		pci_disable_msix(nic->pdev);
1091 		kfree(nic->msix_entries);
1092 		nic->msix_enabled = 0;
1093 		nic->num_vec = 0;
1094 	}
1095 }
1096 
1097 static void nic_free_all_interrupts(struct nicpf *nic)
1098 {
1099 	int irq;
1100 
1101 	for (irq = 0; irq < nic->num_vec; irq++) {
1102 		if (nic->irq_allocated[irq])
1103 			free_irq(nic->msix_entries[irq].vector, nic);
1104 		nic->irq_allocated[irq] = false;
1105 	}
1106 }
1107 
1108 static int nic_register_interrupts(struct nicpf *nic)
1109 {
1110 	int i, ret;
1111 
1112 	/* Enable MSI-X */
1113 	ret = nic_enable_msix(nic);
1114 	if (ret)
1115 		return ret;
1116 
1117 	/* Register mailbox interrupt handler */
1118 	for (i = NIC_PF_INTR_ID_MBOX0; i < nic->num_vec; i++) {
1119 		sprintf(nic->irq_name[i],
1120 			"NICPF Mbox%d", (i - NIC_PF_INTR_ID_MBOX0));
1121 
1122 		ret = request_irq(nic->msix_entries[i].vector,
1123 				  nic_mbx_intr_handler, 0,
1124 				  nic->irq_name[i], nic);
1125 		if (ret)
1126 			goto fail;
1127 
1128 		nic->irq_allocated[i] = true;
1129 	}
1130 
1131 	/* Enable mailbox interrupt */
1132 	nic_enable_mbx_intr(nic);
1133 	return 0;
1134 
1135 fail:
1136 	dev_err(&nic->pdev->dev, "Request irq failed\n");
1137 	nic_free_all_interrupts(nic);
1138 	nic_disable_msix(nic);
1139 	return ret;
1140 }
1141 
1142 static void nic_unregister_interrupts(struct nicpf *nic)
1143 {
1144 	nic_free_all_interrupts(nic);
1145 	nic_disable_msix(nic);
1146 }
1147 
1148 static int nic_num_sqs_en(struct nicpf *nic, int vf_en)
1149 {
1150 	int pos, sqs_per_vf = MAX_SQS_PER_VF_SINGLE_NODE;
1151 	u16 total_vf;
1152 
1153 	/* Secondary Qsets are needed only if CPU count is
1154 	 * morethan MAX_QUEUES_PER_QSET.
1155 	 */
1156 	if (num_online_cpus() <= MAX_QUEUES_PER_QSET)
1157 		return 0;
1158 
1159 	/* Check if its a multi-node environment */
1160 	if (nr_node_ids > 1)
1161 		sqs_per_vf = MAX_SQS_PER_VF;
1162 
1163 	pos = pci_find_ext_capability(nic->pdev, PCI_EXT_CAP_ID_SRIOV);
1164 	pci_read_config_word(nic->pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf);
1165 	return min(total_vf - vf_en, vf_en * sqs_per_vf);
1166 }
1167 
1168 static int nic_sriov_init(struct pci_dev *pdev, struct nicpf *nic)
1169 {
1170 	int pos = 0;
1171 	int vf_en;
1172 	int err;
1173 	u16 total_vf_cnt;
1174 
1175 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
1176 	if (!pos) {
1177 		dev_err(&pdev->dev, "SRIOV capability is not found in PCIe config space\n");
1178 		return -ENODEV;
1179 	}
1180 
1181 	pci_read_config_word(pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf_cnt);
1182 	if (total_vf_cnt < nic->num_vf_en)
1183 		nic->num_vf_en = total_vf_cnt;
1184 
1185 	if (!total_vf_cnt)
1186 		return 0;
1187 
1188 	vf_en = nic->num_vf_en;
1189 	nic->num_sqs_en = nic_num_sqs_en(nic, nic->num_vf_en);
1190 	vf_en += nic->num_sqs_en;
1191 
1192 	err = pci_enable_sriov(pdev, vf_en);
1193 	if (err) {
1194 		dev_err(&pdev->dev, "SRIOV enable failed, num VF is %d\n",
1195 			vf_en);
1196 		nic->num_vf_en = 0;
1197 		return err;
1198 	}
1199 
1200 	dev_info(&pdev->dev, "SRIOV enabled, number of VF available %d\n",
1201 		 vf_en);
1202 
1203 	nic->flags |= NIC_SRIOV_ENABLED;
1204 	return 0;
1205 }
1206 
1207 /* Poll for BGX LMAC link status and update corresponding VF
1208  * if there is a change, valid only if internal L2 switch
1209  * is not present otherwise VF link is always treated as up
1210  */
1211 static void nic_poll_for_link(struct work_struct *work)
1212 {
1213 	union nic_mbx mbx = {};
1214 	struct nicpf *nic;
1215 	struct bgx_link_status link;
1216 	u8 vf, bgx, lmac;
1217 
1218 	nic = container_of(work, struct nicpf, dwork.work);
1219 
1220 	mbx.link_status.msg = NIC_MBOX_MSG_BGX_LINK_CHANGE;
1221 
1222 	for (vf = 0; vf < nic->num_vf_en; vf++) {
1223 		/* Poll only if VF is UP */
1224 		if (!nic->vf_enabled[vf])
1225 			continue;
1226 
1227 		/* Get BGX, LMAC indices for the VF */
1228 		bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1229 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1230 		/* Get interface link status */
1231 		bgx_get_lmac_link_state(nic->node, bgx, lmac, &link);
1232 
1233 		/* Inform VF only if link status changed */
1234 		if (nic->link[vf] == link.link_up)
1235 			continue;
1236 
1237 		if (!nic->mbx_lock[vf]) {
1238 			nic->link[vf] = link.link_up;
1239 			nic->duplex[vf] = link.duplex;
1240 			nic->speed[vf] = link.speed;
1241 
1242 			/* Send a mbox message to VF with current link status */
1243 			mbx.link_status.link_up = link.link_up;
1244 			mbx.link_status.duplex = link.duplex;
1245 			mbx.link_status.speed = link.speed;
1246 			nic_send_msg_to_vf(nic, vf, &mbx);
1247 		}
1248 	}
1249 	queue_delayed_work(nic->check_link, &nic->dwork, HZ * 2);
1250 }
1251 
1252 static int nic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1253 {
1254 	struct device *dev = &pdev->dev;
1255 	struct nicpf *nic;
1256 	int    err;
1257 
1258 	BUILD_BUG_ON(sizeof(union nic_mbx) > 16);
1259 
1260 	nic = devm_kzalloc(dev, sizeof(*nic), GFP_KERNEL);
1261 	if (!nic)
1262 		return -ENOMEM;
1263 
1264 	nic->hw = devm_kzalloc(dev, sizeof(struct hw_info), GFP_KERNEL);
1265 	if (!nic->hw) {
1266 		devm_kfree(dev, nic);
1267 		return -ENOMEM;
1268 	}
1269 
1270 	pci_set_drvdata(pdev, nic);
1271 
1272 	nic->pdev = pdev;
1273 
1274 	err = pci_enable_device(pdev);
1275 	if (err) {
1276 		dev_err(dev, "Failed to enable PCI device\n");
1277 		pci_set_drvdata(pdev, NULL);
1278 		return err;
1279 	}
1280 
1281 	err = pci_request_regions(pdev, DRV_NAME);
1282 	if (err) {
1283 		dev_err(dev, "PCI request regions failed 0x%x\n", err);
1284 		goto err_disable_device;
1285 	}
1286 
1287 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
1288 	if (err) {
1289 		dev_err(dev, "Unable to get usable DMA configuration\n");
1290 		goto err_release_regions;
1291 	}
1292 
1293 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
1294 	if (err) {
1295 		dev_err(dev, "Unable to get 48-bit DMA for consistent allocations\n");
1296 		goto err_release_regions;
1297 	}
1298 
1299 	/* MAP PF's configuration registers */
1300 	nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
1301 	if (!nic->reg_base) {
1302 		dev_err(dev, "Cannot map config register space, aborting\n");
1303 		err = -ENOMEM;
1304 		goto err_release_regions;
1305 	}
1306 
1307 	nic->node = nic_get_node_id(pdev);
1308 
1309 	/* Initialize hardware */
1310 	err = nic_init_hw(nic);
1311 	if (err)
1312 		goto err_release_regions;
1313 
1314 	nic_set_lmac_vf_mapping(nic);
1315 
1316 	/* Register interrupts */
1317 	err = nic_register_interrupts(nic);
1318 	if (err)
1319 		goto err_release_regions;
1320 
1321 	/* Configure SRIOV */
1322 	err = nic_sriov_init(pdev, nic);
1323 	if (err)
1324 		goto err_unregister_interrupts;
1325 
1326 	/* Register a physical link status poll fn() */
1327 	nic->check_link = alloc_workqueue("check_link_status",
1328 					  WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1329 	if (!nic->check_link) {
1330 		err = -ENOMEM;
1331 		goto err_disable_sriov;
1332 	}
1333 
1334 	INIT_DELAYED_WORK(&nic->dwork, nic_poll_for_link);
1335 	queue_delayed_work(nic->check_link, &nic->dwork, 0);
1336 
1337 	return 0;
1338 
1339 err_disable_sriov:
1340 	if (nic->flags & NIC_SRIOV_ENABLED)
1341 		pci_disable_sriov(pdev);
1342 err_unregister_interrupts:
1343 	nic_unregister_interrupts(nic);
1344 err_release_regions:
1345 	pci_release_regions(pdev);
1346 err_disable_device:
1347 	nic_free_lmacmem(nic);
1348 	devm_kfree(dev, nic->hw);
1349 	devm_kfree(dev, nic);
1350 	pci_disable_device(pdev);
1351 	pci_set_drvdata(pdev, NULL);
1352 	return err;
1353 }
1354 
1355 static void nic_remove(struct pci_dev *pdev)
1356 {
1357 	struct nicpf *nic = pci_get_drvdata(pdev);
1358 
1359 	if (nic->flags & NIC_SRIOV_ENABLED)
1360 		pci_disable_sriov(pdev);
1361 
1362 	if (nic->check_link) {
1363 		/* Destroy work Queue */
1364 		cancel_delayed_work_sync(&nic->dwork);
1365 		destroy_workqueue(nic->check_link);
1366 	}
1367 
1368 	nic_unregister_interrupts(nic);
1369 	pci_release_regions(pdev);
1370 
1371 	nic_free_lmacmem(nic);
1372 	devm_kfree(&pdev->dev, nic->hw);
1373 	devm_kfree(&pdev->dev, nic);
1374 
1375 	pci_disable_device(pdev);
1376 	pci_set_drvdata(pdev, NULL);
1377 }
1378 
1379 static struct pci_driver nic_driver = {
1380 	.name = DRV_NAME,
1381 	.id_table = nic_id_table,
1382 	.probe = nic_probe,
1383 	.remove = nic_remove,
1384 };
1385 
1386 static int __init nic_init_module(void)
1387 {
1388 	pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
1389 
1390 	return pci_register_driver(&nic_driver);
1391 }
1392 
1393 static void __exit nic_cleanup_module(void)
1394 {
1395 	pci_unregister_driver(&nic_driver);
1396 }
1397 
1398 module_init(nic_init_module);
1399 module_exit(nic_cleanup_module);
1400