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 #include <linux/if_vlan.h>
15 
16 #include "nic_reg.h"
17 #include "nic.h"
18 #include "q_struct.h"
19 #include "thunder_bgx.h"
20 
21 #define DRV_NAME	"thunder-nic"
22 #define DRV_VERSION	"1.0"
23 
24 struct hw_info {
25 	u8		bgx_cnt;
26 	u8		chans_per_lmac;
27 	u8		chans_per_bgx; /* Rx/Tx chans */
28 	u8		chans_per_rgx;
29 	u8		chans_per_lbk;
30 	u16		cpi_cnt;
31 	u16		rssi_cnt;
32 	u16		rss_ind_tbl_size;
33 	u16		tl4_cnt;
34 	u16		tl3_cnt;
35 	u8		tl2_cnt;
36 	u8		tl1_cnt;
37 	bool		tl1_per_bgx; /* TL1 per BGX or per LMAC */
38 };
39 
40 struct nicpf {
41 	struct pci_dev		*pdev;
42 	struct hw_info          *hw;
43 	u8			node;
44 	unsigned int		flags;
45 	u8			num_vf_en;      /* No of VF enabled */
46 	bool			vf_enabled[MAX_NUM_VFS_SUPPORTED];
47 	void __iomem		*reg_base;       /* Register start address */
48 	u8			num_sqs_en;	/* Secondary qsets enabled */
49 	u64			nicvf[MAX_NUM_VFS_SUPPORTED];
50 	u8			vf_sqs[MAX_NUM_VFS_SUPPORTED][MAX_SQS_PER_VF];
51 	u8			pqs_vf[MAX_NUM_VFS_SUPPORTED];
52 	bool			sqs_used[MAX_NUM_VFS_SUPPORTED];
53 	struct pkind_cfg	pkind;
54 #define	NIC_SET_VF_LMAC_MAP(bgx, lmac)	(((bgx & 0xF) << 4) | (lmac & 0xF))
55 #define	NIC_GET_BGX_FROM_VF_LMAC_MAP(map)	((map >> 4) & 0xF)
56 #define	NIC_GET_LMAC_FROM_VF_LMAC_MAP(map)	(map & 0xF)
57 	u8			*vf_lmac_map;
58 	struct delayed_work     dwork;
59 	struct workqueue_struct *check_link;
60 	u8			*link;
61 	u8			*duplex;
62 	u32			*speed;
63 	u16			cpi_base[MAX_NUM_VFS_SUPPORTED];
64 	u16			rssi_base[MAX_NUM_VFS_SUPPORTED];
65 	bool			mbx_lock[MAX_NUM_VFS_SUPPORTED];
66 
67 	/* MSI-X */
68 	bool			msix_enabled;
69 	u8			num_vec;
70 	struct msix_entry	*msix_entries;
71 	bool			irq_allocated[NIC_PF_MSIX_VECTORS];
72 	char			irq_name[NIC_PF_MSIX_VECTORS][20];
73 };
74 
75 /* Supported devices */
76 static const struct pci_device_id nic_id_table[] = {
77 	{ PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, PCI_DEVICE_ID_THUNDER_NIC_PF) },
78 	{ 0, }  /* end of table */
79 };
80 
81 MODULE_AUTHOR("Sunil Goutham");
82 MODULE_DESCRIPTION("Cavium Thunder NIC Physical Function Driver");
83 MODULE_LICENSE("GPL v2");
84 MODULE_VERSION(DRV_VERSION);
85 MODULE_DEVICE_TABLE(pci, nic_id_table);
86 
87 /* The Cavium ThunderX network controller can *only* be found in SoCs
88  * containing the ThunderX ARM64 CPU implementation.  All accesses to the device
89  * registers on this platform are implicitly strongly ordered with respect
90  * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
91  * with no memory barriers in this driver.  The readq()/writeq() functions add
92  * explicit ordering operation which in this case are redundant, and only
93  * add overhead.
94  */
95 
96 /* Register read/write APIs */
97 static void nic_reg_write(struct nicpf *nic, u64 offset, u64 val)
98 {
99 	writeq_relaxed(val, nic->reg_base + offset);
100 }
101 
102 static u64 nic_reg_read(struct nicpf *nic, u64 offset)
103 {
104 	return readq_relaxed(nic->reg_base + offset);
105 }
106 
107 /* PF -> VF mailbox communication APIs */
108 static void nic_enable_mbx_intr(struct nicpf *nic)
109 {
110 	int vf_cnt = pci_sriov_get_totalvfs(nic->pdev);
111 
112 #define INTR_MASK(vfs) ((vfs < 64) ? (BIT_ULL(vfs) - 1) : (~0ull))
113 
114 	/* Clear it, to avoid spurious interrupts (if any) */
115 	nic_reg_write(nic, NIC_PF_MAILBOX_INT, INTR_MASK(vf_cnt));
116 
117 	/* Enable mailbox interrupt for all VFs */
118 	nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S, INTR_MASK(vf_cnt));
119 	/* One mailbox intr enable reg per 64 VFs */
120 	if (vf_cnt > 64) {
121 		nic_reg_write(nic, NIC_PF_MAILBOX_INT + sizeof(u64),
122 			      INTR_MASK(vf_cnt - 64));
123 		nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S + sizeof(u64),
124 			      INTR_MASK(vf_cnt - 64));
125 	}
126 }
127 
128 static void nic_clear_mbx_intr(struct nicpf *nic, int vf, int mbx_reg)
129 {
130 	nic_reg_write(nic, NIC_PF_MAILBOX_INT + (mbx_reg << 3), BIT_ULL(vf));
131 }
132 
133 static u64 nic_get_mbx_addr(int vf)
134 {
135 	return NIC_PF_VF_0_127_MAILBOX_0_1 + (vf << NIC_VF_NUM_SHIFT);
136 }
137 
138 /* Send a mailbox message to VF
139  * @vf: vf to which this message to be sent
140  * @mbx: Message to be sent
141  */
142 static void nic_send_msg_to_vf(struct nicpf *nic, int vf, union nic_mbx *mbx)
143 {
144 	void __iomem *mbx_addr = nic->reg_base + nic_get_mbx_addr(vf);
145 	u64 *msg = (u64 *)mbx;
146 
147 	/* In first revision HW, mbox interrupt is triggerred
148 	 * when PF writes to MBOX(1), in next revisions when
149 	 * PF writes to MBOX(0)
150 	 */
151 	if (pass1_silicon(nic->pdev)) {
152 		/* see the comment for nic_reg_write()/nic_reg_read()
153 		 * functions above
154 		 */
155 		writeq_relaxed(msg[0], mbx_addr);
156 		writeq_relaxed(msg[1], mbx_addr + 8);
157 	} else {
158 		writeq_relaxed(msg[1], mbx_addr + 8);
159 		writeq_relaxed(msg[0], mbx_addr);
160 	}
161 }
162 
163 /* Responds to VF's READY message with VF's
164  * ID, node, MAC address e.t.c
165  * @vf: VF which sent READY message
166  */
167 static void nic_mbx_send_ready(struct nicpf *nic, int vf)
168 {
169 	union nic_mbx mbx = {};
170 	int bgx_idx, lmac;
171 	const char *mac;
172 
173 	mbx.nic_cfg.msg = NIC_MBOX_MSG_READY;
174 	mbx.nic_cfg.vf_id = vf;
175 
176 	mbx.nic_cfg.tns_mode = NIC_TNS_BYPASS_MODE;
177 
178 	if (vf < nic->num_vf_en) {
179 		bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
180 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
181 
182 		mac = bgx_get_lmac_mac(nic->node, bgx_idx, lmac);
183 		if (mac)
184 			ether_addr_copy((u8 *)&mbx.nic_cfg.mac_addr, mac);
185 	}
186 	mbx.nic_cfg.sqs_mode = (vf >= nic->num_vf_en) ? true : false;
187 	mbx.nic_cfg.node_id = nic->node;
188 
189 	mbx.nic_cfg.loopback_supported = vf < nic->num_vf_en;
190 
191 	nic_send_msg_to_vf(nic, vf, &mbx);
192 }
193 
194 /* ACKs VF's mailbox message
195  * @vf: VF to which ACK to be sent
196  */
197 static void nic_mbx_send_ack(struct nicpf *nic, int vf)
198 {
199 	union nic_mbx mbx = {};
200 
201 	mbx.msg.msg = NIC_MBOX_MSG_ACK;
202 	nic_send_msg_to_vf(nic, vf, &mbx);
203 }
204 
205 /* NACKs VF's mailbox message that PF is not able to
206  * complete the action
207  * @vf: VF to which ACK to be sent
208  */
209 static void nic_mbx_send_nack(struct nicpf *nic, int vf)
210 {
211 	union nic_mbx mbx = {};
212 
213 	mbx.msg.msg = NIC_MBOX_MSG_NACK;
214 	nic_send_msg_to_vf(nic, vf, &mbx);
215 }
216 
217 /* Flush all in flight receive packets to memory and
218  * bring down an active RQ
219  */
220 static int nic_rcv_queue_sw_sync(struct nicpf *nic)
221 {
222 	u16 timeout = ~0x00;
223 
224 	nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x01);
225 	/* Wait till sync cycle is finished */
226 	while (timeout) {
227 		if (nic_reg_read(nic, NIC_PF_SW_SYNC_RX_DONE) & 0x1)
228 			break;
229 		timeout--;
230 	}
231 	nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x00);
232 	if (!timeout) {
233 		dev_err(&nic->pdev->dev, "Receive queue software sync failed");
234 		return 1;
235 	}
236 	return 0;
237 }
238 
239 /* Get BGX Rx/Tx stats and respond to VF's request */
240 static void nic_get_bgx_stats(struct nicpf *nic, struct bgx_stats_msg *bgx)
241 {
242 	int bgx_idx, lmac;
243 	union nic_mbx mbx = {};
244 
245 	bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
246 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
247 
248 	mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
249 	mbx.bgx_stats.vf_id = bgx->vf_id;
250 	mbx.bgx_stats.rx = bgx->rx;
251 	mbx.bgx_stats.idx = bgx->idx;
252 	if (bgx->rx)
253 		mbx.bgx_stats.stats = bgx_get_rx_stats(nic->node, bgx_idx,
254 							    lmac, bgx->idx);
255 	else
256 		mbx.bgx_stats.stats = bgx_get_tx_stats(nic->node, bgx_idx,
257 							    lmac, bgx->idx);
258 	nic_send_msg_to_vf(nic, bgx->vf_id, &mbx);
259 }
260 
261 /* Update hardware min/max frame size */
262 static int nic_update_hw_frs(struct nicpf *nic, int new_frs, int vf)
263 {
264 	int bgx, lmac, lmac_cnt;
265 	u64 lmac_credits;
266 
267 	if ((new_frs > NIC_HW_MAX_FRS) || (new_frs < NIC_HW_MIN_FRS))
268 		return 1;
269 
270 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
271 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
272 	lmac += bgx * MAX_LMAC_PER_BGX;
273 
274 	new_frs += VLAN_ETH_HLEN + ETH_FCS_LEN + 4;
275 
276 	/* Update corresponding LMAC credits */
277 	lmac_cnt = bgx_get_lmac_count(nic->node, bgx);
278 	lmac_credits = nic_reg_read(nic, NIC_PF_LMAC_0_7_CREDIT + (lmac * 8));
279 	lmac_credits &= ~(0xFFFFFULL << 12);
280 	lmac_credits |= (((((48 * 1024) / lmac_cnt) - new_frs) / 16) << 12);
281 	nic_reg_write(nic, NIC_PF_LMAC_0_7_CREDIT + (lmac * 8), lmac_credits);
282 
283 	/* Enforce MTU in HW
284 	 * This config is supported only from 88xx pass 2.0 onwards.
285 	 */
286 	if (!pass1_silicon(nic->pdev))
287 		nic_reg_write(nic,
288 			      NIC_PF_LMAC_0_7_CFG2 + (lmac * 8), new_frs);
289 	return 0;
290 }
291 
292 /* Set minimum transmit packet size */
293 static void nic_set_tx_pkt_pad(struct nicpf *nic, int size)
294 {
295 	int lmac, max_lmac;
296 	u16 sdevid;
297 	u64 lmac_cfg;
298 
299 	/* There is a issue in HW where-in while sending GSO sized
300 	 * pkts as part of TSO, if pkt len falls below this size
301 	 * NIC will zero PAD packet and also updates IP total length.
302 	 * Hence set this value to lessthan min pkt size of MAC+IP+TCP
303 	 * headers, BGX will do the padding to transmit 64 byte pkt.
304 	 */
305 	if (size > 52)
306 		size = 52;
307 
308 	pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
309 	/* 81xx's RGX has only one LMAC */
310 	if (sdevid == PCI_SUBSYS_DEVID_81XX_NIC_PF)
311 		max_lmac = ((nic->hw->bgx_cnt - 1) * MAX_LMAC_PER_BGX) + 1;
312 	else
313 		max_lmac = nic->hw->bgx_cnt * MAX_LMAC_PER_BGX;
314 
315 	for (lmac = 0; lmac < max_lmac; lmac++) {
316 		lmac_cfg = nic_reg_read(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3));
317 		lmac_cfg &= ~(0xF << 2);
318 		lmac_cfg |= ((size / 4) << 2);
319 		nic_reg_write(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3), lmac_cfg);
320 	}
321 }
322 
323 /* Function to check number of LMACs present and set VF::LMAC mapping.
324  * Mapping will be used while initializing channels.
325  */
326 static void nic_set_lmac_vf_mapping(struct nicpf *nic)
327 {
328 	unsigned bgx_map = bgx_get_map(nic->node);
329 	int bgx, next_bgx_lmac = 0;
330 	int lmac, lmac_cnt = 0;
331 	u64 lmac_credit;
332 
333 	nic->num_vf_en = 0;
334 
335 	for (bgx = 0; bgx < nic->hw->bgx_cnt; bgx++) {
336 		if (!(bgx_map & (1 << bgx)))
337 			continue;
338 		lmac_cnt = bgx_get_lmac_count(nic->node, bgx);
339 		for (lmac = 0; lmac < lmac_cnt; lmac++)
340 			nic->vf_lmac_map[next_bgx_lmac++] =
341 						NIC_SET_VF_LMAC_MAP(bgx, lmac);
342 		nic->num_vf_en += lmac_cnt;
343 
344 		/* Program LMAC credits */
345 		lmac_credit = (1ull << 1); /* channel credit enable */
346 		lmac_credit |= (0x1ff << 2); /* Max outstanding pkt count */
347 		/* 48KB BGX Tx buffer size, each unit is of size 16bytes */
348 		lmac_credit |= (((((48 * 1024) / lmac_cnt) -
349 				NIC_HW_MAX_FRS) / 16) << 12);
350 		lmac = bgx * MAX_LMAC_PER_BGX;
351 		for (; lmac < lmac_cnt + (bgx * MAX_LMAC_PER_BGX); lmac++)
352 			nic_reg_write(nic,
353 				      NIC_PF_LMAC_0_7_CREDIT + (lmac * 8),
354 				      lmac_credit);
355 
356 		/* On CN81XX there are only 8 VFs but max possible no of
357 		 * interfaces are 9.
358 		 */
359 		if (nic->num_vf_en >= pci_sriov_get_totalvfs(nic->pdev)) {
360 			nic->num_vf_en = pci_sriov_get_totalvfs(nic->pdev);
361 			break;
362 		}
363 	}
364 }
365 
366 static void nic_free_lmacmem(struct nicpf *nic)
367 {
368 	kfree(nic->vf_lmac_map);
369 	kfree(nic->link);
370 	kfree(nic->duplex);
371 	kfree(nic->speed);
372 }
373 
374 static int nic_get_hw_info(struct nicpf *nic)
375 {
376 	u8 max_lmac;
377 	u16 sdevid;
378 	struct hw_info *hw = nic->hw;
379 
380 	pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
381 
382 	switch (sdevid) {
383 	case PCI_SUBSYS_DEVID_88XX_NIC_PF:
384 		hw->bgx_cnt = MAX_BGX_PER_CN88XX;
385 		hw->chans_per_lmac = 16;
386 		hw->chans_per_bgx = 128;
387 		hw->cpi_cnt = 2048;
388 		hw->rssi_cnt = 4096;
389 		hw->rss_ind_tbl_size = NIC_MAX_RSS_IDR_TBL_SIZE;
390 		hw->tl3_cnt = 256;
391 		hw->tl2_cnt = 64;
392 		hw->tl1_cnt = 2;
393 		hw->tl1_per_bgx = true;
394 		break;
395 	case PCI_SUBSYS_DEVID_81XX_NIC_PF:
396 		hw->bgx_cnt = MAX_BGX_PER_CN81XX;
397 		hw->chans_per_lmac = 8;
398 		hw->chans_per_bgx = 32;
399 		hw->chans_per_rgx = 8;
400 		hw->chans_per_lbk = 24;
401 		hw->cpi_cnt = 512;
402 		hw->rssi_cnt = 256;
403 		hw->rss_ind_tbl_size = 32; /* Max RSSI / Max interfaces */
404 		hw->tl3_cnt = 64;
405 		hw->tl2_cnt = 16;
406 		hw->tl1_cnt = 10;
407 		hw->tl1_per_bgx = false;
408 		break;
409 	case PCI_SUBSYS_DEVID_83XX_NIC_PF:
410 		hw->bgx_cnt = MAX_BGX_PER_CN83XX;
411 		hw->chans_per_lmac = 8;
412 		hw->chans_per_bgx = 32;
413 		hw->chans_per_lbk = 64;
414 		hw->cpi_cnt = 2048;
415 		hw->rssi_cnt = 1024;
416 		hw->rss_ind_tbl_size = 64; /* Max RSSI / Max interfaces */
417 		hw->tl3_cnt = 256;
418 		hw->tl2_cnt = 64;
419 		hw->tl1_cnt = 18;
420 		hw->tl1_per_bgx = false;
421 		break;
422 	}
423 	hw->tl4_cnt = MAX_QUEUES_PER_QSET * pci_sriov_get_totalvfs(nic->pdev);
424 
425 	/* Allocate memory for LMAC tracking elements */
426 	max_lmac = hw->bgx_cnt * MAX_LMAC_PER_BGX;
427 	nic->vf_lmac_map = kmalloc_array(max_lmac, sizeof(u8), GFP_KERNEL);
428 	if (!nic->vf_lmac_map)
429 		goto error;
430 	nic->link = kmalloc_array(max_lmac, sizeof(u8), GFP_KERNEL);
431 	if (!nic->link)
432 		goto error;
433 	nic->duplex = kmalloc_array(max_lmac, sizeof(u8), GFP_KERNEL);
434 	if (!nic->duplex)
435 		goto error;
436 	nic->speed = kmalloc_array(max_lmac, sizeof(u32), GFP_KERNEL);
437 	if (!nic->speed)
438 		goto error;
439 	return 0;
440 
441 error:
442 	nic_free_lmacmem(nic);
443 	return -ENOMEM;
444 }
445 
446 #define BGX0_BLOCK 8
447 #define BGX1_BLOCK 9
448 
449 static int nic_init_hw(struct nicpf *nic)
450 {
451 	int i, err;
452 	u64 cqm_cfg;
453 
454 	/* Get HW capability info */
455 	err = nic_get_hw_info(nic);
456 	if (err)
457 		return err;
458 
459 	/* Enable NIC HW block */
460 	nic_reg_write(nic, NIC_PF_CFG, 0x3);
461 
462 	/* Enable backpressure */
463 	nic_reg_write(nic, NIC_PF_BP_CFG, (1ULL << 6) | 0x03);
464 
465 	/* TNS and TNS bypass modes are present only on 88xx */
466 	if (nic->pdev->subsystem_device == PCI_SUBSYS_DEVID_88XX_NIC_PF) {
467 		/* Disable TNS mode on both interfaces */
468 		nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG,
469 			      (NIC_TNS_BYPASS_MODE << 7) | BGX0_BLOCK);
470 		nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG | (1 << 8),
471 			      (NIC_TNS_BYPASS_MODE << 7) | BGX1_BLOCK);
472 	}
473 
474 	nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG,
475 		      (1ULL << 63) | BGX0_BLOCK);
476 	nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG + (1 << 8),
477 		      (1ULL << 63) | BGX1_BLOCK);
478 
479 	/* PKIND configuration */
480 	nic->pkind.minlen = 0;
481 	nic->pkind.maxlen = NIC_HW_MAX_FRS + VLAN_ETH_HLEN + ETH_FCS_LEN + 4;
482 	nic->pkind.lenerr_en = 1;
483 	nic->pkind.rx_hdr = 0;
484 	nic->pkind.hdr_sl = 0;
485 
486 	for (i = 0; i < NIC_MAX_PKIND; i++)
487 		nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG | (i << 3),
488 			      *(u64 *)&nic->pkind);
489 
490 	nic_set_tx_pkt_pad(nic, NIC_HW_MIN_FRS);
491 
492 	/* Timer config */
493 	nic_reg_write(nic, NIC_PF_INTR_TIMER_CFG, NICPF_CLK_PER_INT_TICK);
494 
495 	/* Enable VLAN ethertype matching and stripping */
496 	nic_reg_write(nic, NIC_PF_RX_ETYPE_0_7,
497 		      (2 << 19) | (ETYPE_ALG_VLAN_STRIP << 16) | ETH_P_8021Q);
498 
499 	/* Check if HW expected value is higher (could be in future chips) */
500 	cqm_cfg = nic_reg_read(nic, NIC_PF_CQM_CFG);
501 	if (cqm_cfg < NICPF_CQM_MIN_DROP_LEVEL)
502 		nic_reg_write(nic, NIC_PF_CQM_CFG, NICPF_CQM_MIN_DROP_LEVEL);
503 
504 	return 0;
505 }
506 
507 /* Channel parse index configuration */
508 static void nic_config_cpi(struct nicpf *nic, struct cpi_cfg_msg *cfg)
509 {
510 	struct hw_info *hw = nic->hw;
511 	u32 vnic, bgx, lmac, chan;
512 	u32 padd, cpi_count = 0;
513 	u64 cpi_base, cpi, rssi_base, rssi;
514 	u8  qset, rq_idx = 0;
515 
516 	vnic = cfg->vf_id;
517 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
518 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
519 
520 	chan = (lmac * hw->chans_per_lmac) + (bgx * hw->chans_per_bgx);
521 	cpi_base = vnic * NIC_MAX_CPI_PER_LMAC;
522 	rssi_base = vnic * hw->rss_ind_tbl_size;
523 
524 	/* Rx channel configuration */
525 	nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_BP_CFG | (chan << 3),
526 		      (1ull << 63) | (vnic << 0));
527 	nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_CFG | (chan << 3),
528 		      ((u64)cfg->cpi_alg << 62) | (cpi_base << 48));
529 
530 	if (cfg->cpi_alg == CPI_ALG_NONE)
531 		cpi_count = 1;
532 	else if (cfg->cpi_alg == CPI_ALG_VLAN) /* 3 bits of PCP */
533 		cpi_count = 8;
534 	else if (cfg->cpi_alg == CPI_ALG_VLAN16) /* 3 bits PCP + DEI */
535 		cpi_count = 16;
536 	else if (cfg->cpi_alg == CPI_ALG_DIFF) /* 6bits DSCP */
537 		cpi_count = NIC_MAX_CPI_PER_LMAC;
538 
539 	/* RSS Qset, Qidx mapping */
540 	qset = cfg->vf_id;
541 	rssi = rssi_base;
542 	for (; rssi < (rssi_base + cfg->rq_cnt); rssi++) {
543 		nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
544 			      (qset << 3) | rq_idx);
545 		rq_idx++;
546 	}
547 
548 	rssi = 0;
549 	cpi = cpi_base;
550 	for (; cpi < (cpi_base + cpi_count); cpi++) {
551 		/* Determine port to channel adder */
552 		if (cfg->cpi_alg != CPI_ALG_DIFF)
553 			padd = cpi % cpi_count;
554 		else
555 			padd = cpi % 8; /* 3 bits CS out of 6bits DSCP */
556 
557 		/* Leave RSS_SIZE as '0' to disable RSS */
558 		if (pass1_silicon(nic->pdev)) {
559 			nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
560 				      (vnic << 24) | (padd << 16) |
561 				      (rssi_base + rssi));
562 		} else {
563 			/* Set MPI_ALG to '0' to disable MCAM parsing */
564 			nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
565 				      (padd << 16));
566 			/* MPI index is same as CPI if MPI_ALG is not enabled */
567 			nic_reg_write(nic, NIC_PF_MPI_0_2047_CFG | (cpi << 3),
568 				      (vnic << 24) | (rssi_base + rssi));
569 		}
570 
571 		if ((rssi + 1) >= cfg->rq_cnt)
572 			continue;
573 
574 		if (cfg->cpi_alg == CPI_ALG_VLAN)
575 			rssi++;
576 		else if (cfg->cpi_alg == CPI_ALG_VLAN16)
577 			rssi = ((cpi - cpi_base) & 0xe) >> 1;
578 		else if (cfg->cpi_alg == CPI_ALG_DIFF)
579 			rssi = ((cpi - cpi_base) & 0x38) >> 3;
580 	}
581 	nic->cpi_base[cfg->vf_id] = cpi_base;
582 	nic->rssi_base[cfg->vf_id] = rssi_base;
583 }
584 
585 /* Responsds to VF with its RSS indirection table size */
586 static void nic_send_rss_size(struct nicpf *nic, int vf)
587 {
588 	union nic_mbx mbx = {};
589 	u64  *msg;
590 
591 	msg = (u64 *)&mbx;
592 
593 	mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
594 	mbx.rss_size.ind_tbl_size = nic->hw->rss_ind_tbl_size;
595 	nic_send_msg_to_vf(nic, vf, &mbx);
596 }
597 
598 /* Receive side scaling configuration
599  * configure:
600  * - RSS index
601  * - indir table i.e hash::RQ mapping
602  * - no of hash bits to consider
603  */
604 static void nic_config_rss(struct nicpf *nic, struct rss_cfg_msg *cfg)
605 {
606 	u8  qset, idx = 0;
607 	u64 cpi_cfg, cpi_base, rssi_base, rssi;
608 	u64 idx_addr;
609 
610 	rssi_base = nic->rssi_base[cfg->vf_id] + cfg->tbl_offset;
611 
612 	rssi = rssi_base;
613 	qset = cfg->vf_id;
614 
615 	for (; rssi < (rssi_base + cfg->tbl_len); rssi++) {
616 		u8 svf = cfg->ind_tbl[idx] >> 3;
617 
618 		if (svf)
619 			qset = nic->vf_sqs[cfg->vf_id][svf - 1];
620 		else
621 			qset = cfg->vf_id;
622 		nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
623 			      (qset << 3) | (cfg->ind_tbl[idx] & 0x7));
624 		idx++;
625 	}
626 
627 	cpi_base = nic->cpi_base[cfg->vf_id];
628 	if (pass1_silicon(nic->pdev))
629 		idx_addr = NIC_PF_CPI_0_2047_CFG;
630 	else
631 		idx_addr = NIC_PF_MPI_0_2047_CFG;
632 	cpi_cfg = nic_reg_read(nic, idx_addr | (cpi_base << 3));
633 	cpi_cfg &= ~(0xFULL << 20);
634 	cpi_cfg |= (cfg->hash_bits << 20);
635 	nic_reg_write(nic, idx_addr | (cpi_base << 3), cpi_cfg);
636 }
637 
638 /* 4 level transmit side scheduler configutation
639  * for TNS bypass mode
640  *
641  * Sample configuration for SQ0 on 88xx
642  * VNIC0-SQ0 -> TL4(0)   -> TL3[0]   -> TL2[0]  -> TL1[0] -> BGX0
643  * VNIC1-SQ0 -> TL4(8)   -> TL3[2]   -> TL2[0]  -> TL1[0] -> BGX0
644  * VNIC2-SQ0 -> TL4(16)  -> TL3[4]   -> TL2[1]  -> TL1[0] -> BGX0
645  * VNIC3-SQ0 -> TL4(24)  -> TL3[6]   -> TL2[1]  -> TL1[0] -> BGX0
646  * VNIC4-SQ0 -> TL4(512) -> TL3[128] -> TL2[32] -> TL1[1] -> BGX1
647  * VNIC5-SQ0 -> TL4(520) -> TL3[130] -> TL2[32] -> TL1[1] -> BGX1
648  * VNIC6-SQ0 -> TL4(528) -> TL3[132] -> TL2[33] -> TL1[1] -> BGX1
649  * VNIC7-SQ0 -> TL4(536) -> TL3[134] -> TL2[33] -> TL1[1] -> BGX1
650  */
651 static void nic_tx_channel_cfg(struct nicpf *nic, u8 vnic,
652 			       struct sq_cfg_msg *sq)
653 {
654 	struct hw_info *hw = nic->hw;
655 	u32 bgx, lmac, chan;
656 	u32 tl2, tl3, tl4;
657 	u32 rr_quantum;
658 	u8 sq_idx = sq->sq_num;
659 	u8 pqs_vnic;
660 	int svf;
661 
662 	if (sq->sqs_mode)
663 		pqs_vnic = nic->pqs_vf[vnic];
664 	else
665 		pqs_vnic = vnic;
666 
667 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
668 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
669 
670 	/* 24 bytes for FCS, IPG and preamble */
671 	rr_quantum = ((NIC_HW_MAX_FRS + 24) / 4);
672 
673 	/* For 88xx 0-511 TL4 transmits via BGX0 and
674 	 * 512-1023 TL4s transmit via BGX1.
675 	 */
676 	if (hw->tl1_per_bgx) {
677 		tl4 = bgx * (hw->tl4_cnt / hw->bgx_cnt);
678 		if (!sq->sqs_mode) {
679 			tl4 += (lmac * MAX_QUEUES_PER_QSET);
680 		} else {
681 			for (svf = 0; svf < MAX_SQS_PER_VF; svf++) {
682 				if (nic->vf_sqs[pqs_vnic][svf] == vnic)
683 					break;
684 			}
685 			tl4 += (MAX_LMAC_PER_BGX * MAX_QUEUES_PER_QSET);
686 			tl4 += (lmac * MAX_QUEUES_PER_QSET * MAX_SQS_PER_VF);
687 			tl4 += (svf * MAX_QUEUES_PER_QSET);
688 		}
689 	} else {
690 		tl4 = (vnic * MAX_QUEUES_PER_QSET);
691 	}
692 	tl4 += sq_idx;
693 
694 	tl3 = tl4 / (hw->tl4_cnt / hw->tl3_cnt);
695 	nic_reg_write(nic, NIC_PF_QSET_0_127_SQ_0_7_CFG2 |
696 		      ((u64)vnic << NIC_QS_ID_SHIFT) |
697 		      ((u32)sq_idx << NIC_Q_NUM_SHIFT), tl4);
698 	nic_reg_write(nic, NIC_PF_TL4_0_1023_CFG | (tl4 << 3),
699 		      ((u64)vnic << 27) | ((u32)sq_idx << 24) | rr_quantum);
700 
701 	nic_reg_write(nic, NIC_PF_TL3_0_255_CFG | (tl3 << 3), rr_quantum);
702 
703 	/* On 88xx 0-127 channels are for BGX0 and
704 	 * 127-255 channels for BGX1.
705 	 *
706 	 * On 81xx/83xx TL3_CHAN reg should be configured with channel
707 	 * within LMAC i.e 0-7 and not the actual channel number like on 88xx
708 	 */
709 	chan = (lmac * hw->chans_per_lmac) + (bgx * hw->chans_per_bgx);
710 	if (hw->tl1_per_bgx)
711 		nic_reg_write(nic, NIC_PF_TL3_0_255_CHAN | (tl3 << 3), chan);
712 	else
713 		nic_reg_write(nic, NIC_PF_TL3_0_255_CHAN | (tl3 << 3), 0);
714 
715 	/* Enable backpressure on the channel */
716 	nic_reg_write(nic, NIC_PF_CHAN_0_255_TX_CFG | (chan << 3), 1);
717 
718 	tl2 = tl3 >> 2;
719 	nic_reg_write(nic, NIC_PF_TL3A_0_63_CFG | (tl2 << 3), tl2);
720 	nic_reg_write(nic, NIC_PF_TL2_0_63_CFG | (tl2 << 3), rr_quantum);
721 	/* No priorities as of now */
722 	nic_reg_write(nic, NIC_PF_TL2_0_63_PRI | (tl2 << 3), 0x00);
723 
724 	/* Unlike 88xx where TL2s 0-31 transmits to TL1 '0' and rest to TL1 '1'
725 	 * on 81xx/83xx TL2 needs to be configured to transmit to one of the
726 	 * possible LMACs.
727 	 *
728 	 * This register doesn't exist on 88xx.
729 	 */
730 	if (!hw->tl1_per_bgx)
731 		nic_reg_write(nic, NIC_PF_TL2_LMAC | (tl2 << 3),
732 			      lmac + (bgx * MAX_LMAC_PER_BGX));
733 }
734 
735 /* Send primary nicvf pointer to secondary QS's VF */
736 static void nic_send_pnicvf(struct nicpf *nic, int sqs)
737 {
738 	union nic_mbx mbx = {};
739 
740 	mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
741 	mbx.nicvf.nicvf = nic->nicvf[nic->pqs_vf[sqs]];
742 	nic_send_msg_to_vf(nic, sqs, &mbx);
743 }
744 
745 /* Send SQS's nicvf pointer to primary QS's VF */
746 static void nic_send_snicvf(struct nicpf *nic, struct nicvf_ptr *nicvf)
747 {
748 	union nic_mbx mbx = {};
749 	int sqs_id = nic->vf_sqs[nicvf->vf_id][nicvf->sqs_id];
750 
751 	mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
752 	mbx.nicvf.sqs_id = nicvf->sqs_id;
753 	mbx.nicvf.nicvf = nic->nicvf[sqs_id];
754 	nic_send_msg_to_vf(nic, nicvf->vf_id, &mbx);
755 }
756 
757 /* Find next available Qset that can be assigned as a
758  * secondary Qset to a VF.
759  */
760 static int nic_nxt_avail_sqs(struct nicpf *nic)
761 {
762 	int sqs;
763 
764 	for (sqs = 0; sqs < nic->num_sqs_en; sqs++) {
765 		if (!nic->sqs_used[sqs])
766 			nic->sqs_used[sqs] = true;
767 		else
768 			continue;
769 		return sqs + nic->num_vf_en;
770 	}
771 	return -1;
772 }
773 
774 /* Allocate additional Qsets for requested VF */
775 static void nic_alloc_sqs(struct nicpf *nic, struct sqs_alloc *sqs)
776 {
777 	union nic_mbx mbx = {};
778 	int idx, alloc_qs = 0;
779 	int sqs_id;
780 
781 	if (!nic->num_sqs_en)
782 		goto send_mbox;
783 
784 	for (idx = 0; idx < sqs->qs_count; idx++) {
785 		sqs_id = nic_nxt_avail_sqs(nic);
786 		if (sqs_id < 0)
787 			break;
788 		nic->vf_sqs[sqs->vf_id][idx] = sqs_id;
789 		nic->pqs_vf[sqs_id] = sqs->vf_id;
790 		alloc_qs++;
791 	}
792 
793 send_mbox:
794 	mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
795 	mbx.sqs_alloc.vf_id = sqs->vf_id;
796 	mbx.sqs_alloc.qs_count = alloc_qs;
797 	nic_send_msg_to_vf(nic, sqs->vf_id, &mbx);
798 }
799 
800 static int nic_config_loopback(struct nicpf *nic, struct set_loopback *lbk)
801 {
802 	int bgx_idx, lmac_idx;
803 
804 	if (lbk->vf_id >= nic->num_vf_en)
805 		return -1;
806 
807 	bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
808 	lmac_idx = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
809 
810 	bgx_lmac_internal_loopback(nic->node, bgx_idx, lmac_idx, lbk->enable);
811 
812 	/* Enable moving average calculation.
813 	 * Keep the LVL/AVG delay to HW enforced minimum so that, not too many
814 	 * packets sneek in between average calculations.
815 	 */
816 	nic_reg_write(nic, NIC_PF_CQ_AVG_CFG,
817 		      (BIT_ULL(20) | 0x2ull << 14 | 0x1));
818 	nic_reg_write(nic, NIC_PF_RRM_AVG_CFG,
819 		      (BIT_ULL(20) | 0x3ull << 14 | 0x1));
820 
821 	return 0;
822 }
823 
824 /* Reset statistics counters */
825 static int nic_reset_stat_counters(struct nicpf *nic,
826 				   int vf, struct reset_stat_cfg *cfg)
827 {
828 	int i, stat, qnum;
829 	u64 reg_addr;
830 
831 	for (i = 0; i < RX_STATS_ENUM_LAST; i++) {
832 		if (cfg->rx_stat_mask & BIT(i)) {
833 			reg_addr = NIC_PF_VNIC_0_127_RX_STAT_0_13 |
834 				   (vf << NIC_QS_ID_SHIFT) |
835 				   (i << 3);
836 			nic_reg_write(nic, reg_addr, 0);
837 		}
838 	}
839 
840 	for (i = 0; i < TX_STATS_ENUM_LAST; i++) {
841 		if (cfg->tx_stat_mask & BIT(i)) {
842 			reg_addr = NIC_PF_VNIC_0_127_TX_STAT_0_4 |
843 				   (vf << NIC_QS_ID_SHIFT) |
844 				   (i << 3);
845 			nic_reg_write(nic, reg_addr, 0);
846 		}
847 	}
848 
849 	for (i = 0; i <= 15; i++) {
850 		qnum = i >> 1;
851 		stat = i & 1 ? 1 : 0;
852 		reg_addr = (vf << NIC_QS_ID_SHIFT) |
853 			   (qnum << NIC_Q_NUM_SHIFT) | (stat << 3);
854 		if (cfg->rq_stat_mask & BIT(i)) {
855 			reg_addr |= NIC_PF_QSET_0_127_RQ_0_7_STAT_0_1;
856 			nic_reg_write(nic, reg_addr, 0);
857 		}
858 		if (cfg->sq_stat_mask & BIT(i)) {
859 			reg_addr |= NIC_PF_QSET_0_127_SQ_0_7_STAT_0_1;
860 			nic_reg_write(nic, reg_addr, 0);
861 		}
862 	}
863 
864 	return 0;
865 }
866 
867 static void nic_enable_tunnel_parsing(struct nicpf *nic, int vf)
868 {
869 	u64 prot_def = (IPV6_PROT << 32) | (IPV4_PROT << 16) | ET_PROT;
870 	u64 vxlan_prot_def = (IPV6_PROT_DEF << 32) |
871 			      (IPV4_PROT_DEF) << 16 | ET_PROT_DEF;
872 
873 	/* Configure tunnel parsing parameters */
874 	nic_reg_write(nic, NIC_PF_RX_GENEVE_DEF,
875 		      (1ULL << 63 | UDP_GENEVE_PORT_NUM));
876 	nic_reg_write(nic, NIC_PF_RX_GENEVE_PROT_DEF,
877 		      ((7ULL << 61) | prot_def));
878 	nic_reg_write(nic, NIC_PF_RX_NVGRE_PROT_DEF,
879 		      ((7ULL << 61) | prot_def));
880 	nic_reg_write(nic, NIC_PF_RX_VXLAN_DEF_0_1,
881 		      ((1ULL << 63) | UDP_VXLAN_PORT_NUM));
882 	nic_reg_write(nic, NIC_PF_RX_VXLAN_PROT_DEF,
883 		      ((0xfULL << 60) | vxlan_prot_def));
884 }
885 
886 static void nic_enable_vf(struct nicpf *nic, int vf, bool enable)
887 {
888 	int bgx, lmac;
889 
890 	nic->vf_enabled[vf] = enable;
891 
892 	if (vf >= nic->num_vf_en)
893 		return;
894 
895 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
896 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
897 
898 	bgx_lmac_rx_tx_enable(nic->node, bgx, lmac, enable);
899 }
900 
901 static void nic_pause_frame(struct nicpf *nic, int vf, struct pfc *cfg)
902 {
903 	int bgx, lmac;
904 	struct pfc pfc;
905 	union nic_mbx mbx = {};
906 
907 	if (vf >= nic->num_vf_en)
908 		return;
909 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
910 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
911 
912 	if (cfg->get) {
913 		bgx_lmac_get_pfc(nic->node, bgx, lmac, &pfc);
914 		mbx.pfc.msg = NIC_MBOX_MSG_PFC;
915 		mbx.pfc.autoneg = pfc.autoneg;
916 		mbx.pfc.fc_rx = pfc.fc_rx;
917 		mbx.pfc.fc_tx = pfc.fc_tx;
918 		nic_send_msg_to_vf(nic, vf, &mbx);
919 	} else {
920 		bgx_lmac_set_pfc(nic->node, bgx, lmac, cfg);
921 		nic_mbx_send_ack(nic, vf);
922 	}
923 }
924 
925 /* Interrupt handler to handle mailbox messages from VFs */
926 static void nic_handle_mbx_intr(struct nicpf *nic, int vf)
927 {
928 	union nic_mbx mbx = {};
929 	u64 *mbx_data;
930 	u64 mbx_addr;
931 	u64 reg_addr;
932 	u64 cfg;
933 	int bgx, lmac;
934 	int i;
935 	int ret = 0;
936 
937 	nic->mbx_lock[vf] = true;
938 
939 	mbx_addr = nic_get_mbx_addr(vf);
940 	mbx_data = (u64 *)&mbx;
941 
942 	for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
943 		*mbx_data = nic_reg_read(nic, mbx_addr);
944 		mbx_data++;
945 		mbx_addr += sizeof(u64);
946 	}
947 
948 	dev_dbg(&nic->pdev->dev, "%s: Mailbox msg 0x%02x from VF%d\n",
949 		__func__, mbx.msg.msg, vf);
950 	switch (mbx.msg.msg) {
951 	case NIC_MBOX_MSG_READY:
952 		nic_mbx_send_ready(nic, vf);
953 		if (vf < nic->num_vf_en) {
954 			nic->link[vf] = 0;
955 			nic->duplex[vf] = 0;
956 			nic->speed[vf] = 0;
957 		}
958 		goto unlock;
959 	case NIC_MBOX_MSG_QS_CFG:
960 		reg_addr = NIC_PF_QSET_0_127_CFG |
961 			   (mbx.qs.num << NIC_QS_ID_SHIFT);
962 		cfg = mbx.qs.cfg;
963 		/* Check if its a secondary Qset */
964 		if (vf >= nic->num_vf_en) {
965 			cfg = cfg & (~0x7FULL);
966 			/* Assign this Qset to primary Qset's VF */
967 			cfg |= nic->pqs_vf[vf];
968 		}
969 		nic_reg_write(nic, reg_addr, cfg);
970 		break;
971 	case NIC_MBOX_MSG_RQ_CFG:
972 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_CFG |
973 			   (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
974 			   (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
975 		nic_reg_write(nic, reg_addr, mbx.rq.cfg);
976 		/* Enable CQE_RX2_S extension in CQE_RX descriptor.
977 		 * This gets appended by default on 81xx/83xx chips,
978 		 * for consistency enabling the same on 88xx pass2
979 		 * where this is introduced.
980 		 */
981 		if (pass2_silicon(nic->pdev))
982 			nic_reg_write(nic, NIC_PF_RX_CFG, 0x01);
983 		if (!pass1_silicon(nic->pdev))
984 			nic_enable_tunnel_parsing(nic, vf);
985 		break;
986 	case NIC_MBOX_MSG_RQ_BP_CFG:
987 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_BP_CFG |
988 			   (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
989 			   (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
990 		nic_reg_write(nic, reg_addr, mbx.rq.cfg);
991 		break;
992 	case NIC_MBOX_MSG_RQ_SW_SYNC:
993 		ret = nic_rcv_queue_sw_sync(nic);
994 		break;
995 	case NIC_MBOX_MSG_RQ_DROP_CFG:
996 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_DROP_CFG |
997 			   (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
998 			   (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
999 		nic_reg_write(nic, reg_addr, mbx.rq.cfg);
1000 		break;
1001 	case NIC_MBOX_MSG_SQ_CFG:
1002 		reg_addr = NIC_PF_QSET_0_127_SQ_0_7_CFG |
1003 			   (mbx.sq.qs_num << NIC_QS_ID_SHIFT) |
1004 			   (mbx.sq.sq_num << NIC_Q_NUM_SHIFT);
1005 		nic_reg_write(nic, reg_addr, mbx.sq.cfg);
1006 		nic_tx_channel_cfg(nic, mbx.qs.num, &mbx.sq);
1007 		break;
1008 	case NIC_MBOX_MSG_SET_MAC:
1009 		if (vf >= nic->num_vf_en) {
1010 			ret = -1; /* NACK */
1011 			break;
1012 		}
1013 		lmac = mbx.mac.vf_id;
1014 		bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
1015 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
1016 		bgx_set_lmac_mac(nic->node, bgx, lmac, mbx.mac.mac_addr);
1017 		break;
1018 	case NIC_MBOX_MSG_SET_MAX_FRS:
1019 		ret = nic_update_hw_frs(nic, mbx.frs.max_frs,
1020 					mbx.frs.vf_id);
1021 		break;
1022 	case NIC_MBOX_MSG_CPI_CFG:
1023 		nic_config_cpi(nic, &mbx.cpi_cfg);
1024 		break;
1025 	case NIC_MBOX_MSG_RSS_SIZE:
1026 		nic_send_rss_size(nic, vf);
1027 		goto unlock;
1028 	case NIC_MBOX_MSG_RSS_CFG:
1029 	case NIC_MBOX_MSG_RSS_CFG_CONT:
1030 		nic_config_rss(nic, &mbx.rss_cfg);
1031 		break;
1032 	case NIC_MBOX_MSG_CFG_DONE:
1033 		/* Last message of VF config msg sequence */
1034 		nic_enable_vf(nic, vf, true);
1035 		goto unlock;
1036 	case NIC_MBOX_MSG_SHUTDOWN:
1037 		/* First msg in VF teardown sequence */
1038 		if (vf >= nic->num_vf_en)
1039 			nic->sqs_used[vf - nic->num_vf_en] = false;
1040 		nic->pqs_vf[vf] = 0;
1041 		nic_enable_vf(nic, vf, false);
1042 		break;
1043 	case NIC_MBOX_MSG_ALLOC_SQS:
1044 		nic_alloc_sqs(nic, &mbx.sqs_alloc);
1045 		goto unlock;
1046 	case NIC_MBOX_MSG_NICVF_PTR:
1047 		nic->nicvf[vf] = mbx.nicvf.nicvf;
1048 		break;
1049 	case NIC_MBOX_MSG_PNICVF_PTR:
1050 		nic_send_pnicvf(nic, vf);
1051 		goto unlock;
1052 	case NIC_MBOX_MSG_SNICVF_PTR:
1053 		nic_send_snicvf(nic, &mbx.nicvf);
1054 		goto unlock;
1055 	case NIC_MBOX_MSG_BGX_STATS:
1056 		nic_get_bgx_stats(nic, &mbx.bgx_stats);
1057 		goto unlock;
1058 	case NIC_MBOX_MSG_LOOPBACK:
1059 		ret = nic_config_loopback(nic, &mbx.lbk);
1060 		break;
1061 	case NIC_MBOX_MSG_RESET_STAT_COUNTER:
1062 		ret = nic_reset_stat_counters(nic, vf, &mbx.reset_stat);
1063 		break;
1064 	case NIC_MBOX_MSG_PFC:
1065 		nic_pause_frame(nic, vf, &mbx.pfc);
1066 		goto unlock;
1067 	default:
1068 		dev_err(&nic->pdev->dev,
1069 			"Invalid msg from VF%d, msg 0x%x\n", vf, mbx.msg.msg);
1070 		break;
1071 	}
1072 
1073 	if (!ret) {
1074 		nic_mbx_send_ack(nic, vf);
1075 	} else if (mbx.msg.msg != NIC_MBOX_MSG_READY) {
1076 		dev_err(&nic->pdev->dev, "NACK for MBOX 0x%02x from VF %d\n",
1077 			mbx.msg.msg, vf);
1078 		nic_mbx_send_nack(nic, vf);
1079 	}
1080 unlock:
1081 	nic->mbx_lock[vf] = false;
1082 }
1083 
1084 static irqreturn_t nic_mbx_intr_handler(int irq, void *nic_irq)
1085 {
1086 	struct nicpf *nic = (struct nicpf *)nic_irq;
1087 	int mbx;
1088 	u64 intr;
1089 	u8  vf, vf_per_mbx_reg = 64;
1090 
1091 	if (irq == nic->msix_entries[NIC_PF_INTR_ID_MBOX0].vector)
1092 		mbx = 0;
1093 	else
1094 		mbx = 1;
1095 
1096 	intr = nic_reg_read(nic, NIC_PF_MAILBOX_INT + (mbx << 3));
1097 	dev_dbg(&nic->pdev->dev, "PF interrupt Mbox%d 0x%llx\n", mbx, intr);
1098 	for (vf = 0; vf < vf_per_mbx_reg; vf++) {
1099 		if (intr & (1ULL << vf)) {
1100 			dev_dbg(&nic->pdev->dev, "Intr from VF %d\n",
1101 				vf + (mbx * vf_per_mbx_reg));
1102 
1103 			nic_handle_mbx_intr(nic, vf + (mbx * vf_per_mbx_reg));
1104 			nic_clear_mbx_intr(nic, vf, mbx);
1105 		}
1106 	}
1107 	return IRQ_HANDLED;
1108 }
1109 
1110 static int nic_enable_msix(struct nicpf *nic)
1111 {
1112 	int i, ret;
1113 
1114 	nic->num_vec = pci_msix_vec_count(nic->pdev);
1115 
1116 	nic->msix_entries = kmalloc_array(nic->num_vec,
1117 					  sizeof(struct msix_entry),
1118 					  GFP_KERNEL);
1119 	if (!nic->msix_entries)
1120 		return -ENOMEM;
1121 
1122 	for (i = 0; i < nic->num_vec; i++)
1123 		nic->msix_entries[i].entry = i;
1124 
1125 	ret = pci_enable_msix(nic->pdev, nic->msix_entries, nic->num_vec);
1126 	if (ret) {
1127 		dev_err(&nic->pdev->dev,
1128 			"Request for #%d msix vectors failed, returned %d\n",
1129 			   nic->num_vec, ret);
1130 		kfree(nic->msix_entries);
1131 		return ret;
1132 	}
1133 
1134 	nic->msix_enabled = 1;
1135 	return 0;
1136 }
1137 
1138 static void nic_disable_msix(struct nicpf *nic)
1139 {
1140 	if (nic->msix_enabled) {
1141 		pci_disable_msix(nic->pdev);
1142 		kfree(nic->msix_entries);
1143 		nic->msix_enabled = 0;
1144 		nic->num_vec = 0;
1145 	}
1146 }
1147 
1148 static void nic_free_all_interrupts(struct nicpf *nic)
1149 {
1150 	int irq;
1151 
1152 	for (irq = 0; irq < nic->num_vec; irq++) {
1153 		if (nic->irq_allocated[irq])
1154 			free_irq(nic->msix_entries[irq].vector, nic);
1155 		nic->irq_allocated[irq] = false;
1156 	}
1157 }
1158 
1159 static int nic_register_interrupts(struct nicpf *nic)
1160 {
1161 	int i, ret;
1162 
1163 	/* Enable MSI-X */
1164 	ret = nic_enable_msix(nic);
1165 	if (ret)
1166 		return ret;
1167 
1168 	/* Register mailbox interrupt handler */
1169 	for (i = NIC_PF_INTR_ID_MBOX0; i < nic->num_vec; i++) {
1170 		sprintf(nic->irq_name[i],
1171 			"NICPF Mbox%d", (i - NIC_PF_INTR_ID_MBOX0));
1172 
1173 		ret = request_irq(nic->msix_entries[i].vector,
1174 				  nic_mbx_intr_handler, 0,
1175 				  nic->irq_name[i], nic);
1176 		if (ret)
1177 			goto fail;
1178 
1179 		nic->irq_allocated[i] = true;
1180 	}
1181 
1182 	/* Enable mailbox interrupt */
1183 	nic_enable_mbx_intr(nic);
1184 	return 0;
1185 
1186 fail:
1187 	dev_err(&nic->pdev->dev, "Request irq failed\n");
1188 	nic_free_all_interrupts(nic);
1189 	nic_disable_msix(nic);
1190 	return ret;
1191 }
1192 
1193 static void nic_unregister_interrupts(struct nicpf *nic)
1194 {
1195 	nic_free_all_interrupts(nic);
1196 	nic_disable_msix(nic);
1197 }
1198 
1199 static int nic_num_sqs_en(struct nicpf *nic, int vf_en)
1200 {
1201 	int pos, sqs_per_vf = MAX_SQS_PER_VF_SINGLE_NODE;
1202 	u16 total_vf;
1203 
1204 	/* Secondary Qsets are needed only if CPU count is
1205 	 * morethan MAX_QUEUES_PER_QSET.
1206 	 */
1207 	if (num_online_cpus() <= MAX_QUEUES_PER_QSET)
1208 		return 0;
1209 
1210 	/* Check if its a multi-node environment */
1211 	if (nr_node_ids > 1)
1212 		sqs_per_vf = MAX_SQS_PER_VF;
1213 
1214 	pos = pci_find_ext_capability(nic->pdev, PCI_EXT_CAP_ID_SRIOV);
1215 	pci_read_config_word(nic->pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf);
1216 	return min(total_vf - vf_en, vf_en * sqs_per_vf);
1217 }
1218 
1219 static int nic_sriov_init(struct pci_dev *pdev, struct nicpf *nic)
1220 {
1221 	int pos = 0;
1222 	int vf_en;
1223 	int err;
1224 	u16 total_vf_cnt;
1225 
1226 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
1227 	if (!pos) {
1228 		dev_err(&pdev->dev, "SRIOV capability is not found in PCIe config space\n");
1229 		return -ENODEV;
1230 	}
1231 
1232 	pci_read_config_word(pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf_cnt);
1233 	if (total_vf_cnt < nic->num_vf_en)
1234 		nic->num_vf_en = total_vf_cnt;
1235 
1236 	if (!total_vf_cnt)
1237 		return 0;
1238 
1239 	vf_en = nic->num_vf_en;
1240 	nic->num_sqs_en = nic_num_sqs_en(nic, nic->num_vf_en);
1241 	vf_en += nic->num_sqs_en;
1242 
1243 	err = pci_enable_sriov(pdev, vf_en);
1244 	if (err) {
1245 		dev_err(&pdev->dev, "SRIOV enable failed, num VF is %d\n",
1246 			vf_en);
1247 		nic->num_vf_en = 0;
1248 		return err;
1249 	}
1250 
1251 	dev_info(&pdev->dev, "SRIOV enabled, number of VF available %d\n",
1252 		 vf_en);
1253 
1254 	nic->flags |= NIC_SRIOV_ENABLED;
1255 	return 0;
1256 }
1257 
1258 /* Poll for BGX LMAC link status and update corresponding VF
1259  * if there is a change, valid only if internal L2 switch
1260  * is not present otherwise VF link is always treated as up
1261  */
1262 static void nic_poll_for_link(struct work_struct *work)
1263 {
1264 	union nic_mbx mbx = {};
1265 	struct nicpf *nic;
1266 	struct bgx_link_status link;
1267 	u8 vf, bgx, lmac;
1268 
1269 	nic = container_of(work, struct nicpf, dwork.work);
1270 
1271 	mbx.link_status.msg = NIC_MBOX_MSG_BGX_LINK_CHANGE;
1272 
1273 	for (vf = 0; vf < nic->num_vf_en; vf++) {
1274 		/* Poll only if VF is UP */
1275 		if (!nic->vf_enabled[vf])
1276 			continue;
1277 
1278 		/* Get BGX, LMAC indices for the VF */
1279 		bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1280 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1281 		/* Get interface link status */
1282 		bgx_get_lmac_link_state(nic->node, bgx, lmac, &link);
1283 
1284 		/* Inform VF only if link status changed */
1285 		if (nic->link[vf] == link.link_up)
1286 			continue;
1287 
1288 		if (!nic->mbx_lock[vf]) {
1289 			nic->link[vf] = link.link_up;
1290 			nic->duplex[vf] = link.duplex;
1291 			nic->speed[vf] = link.speed;
1292 
1293 			/* Send a mbox message to VF with current link status */
1294 			mbx.link_status.link_up = link.link_up;
1295 			mbx.link_status.duplex = link.duplex;
1296 			mbx.link_status.speed = link.speed;
1297 			mbx.link_status.mac_type = link.mac_type;
1298 			nic_send_msg_to_vf(nic, vf, &mbx);
1299 		}
1300 	}
1301 	queue_delayed_work(nic->check_link, &nic->dwork, HZ * 2);
1302 }
1303 
1304 static int nic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1305 {
1306 	struct device *dev = &pdev->dev;
1307 	struct nicpf *nic;
1308 	int    err;
1309 
1310 	BUILD_BUG_ON(sizeof(union nic_mbx) > 16);
1311 
1312 	nic = devm_kzalloc(dev, sizeof(*nic), GFP_KERNEL);
1313 	if (!nic)
1314 		return -ENOMEM;
1315 
1316 	nic->hw = devm_kzalloc(dev, sizeof(struct hw_info), GFP_KERNEL);
1317 	if (!nic->hw) {
1318 		devm_kfree(dev, nic);
1319 		return -ENOMEM;
1320 	}
1321 
1322 	pci_set_drvdata(pdev, nic);
1323 
1324 	nic->pdev = pdev;
1325 
1326 	err = pci_enable_device(pdev);
1327 	if (err) {
1328 		dev_err(dev, "Failed to enable PCI device\n");
1329 		pci_set_drvdata(pdev, NULL);
1330 		return err;
1331 	}
1332 
1333 	err = pci_request_regions(pdev, DRV_NAME);
1334 	if (err) {
1335 		dev_err(dev, "PCI request regions failed 0x%x\n", err);
1336 		goto err_disable_device;
1337 	}
1338 
1339 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
1340 	if (err) {
1341 		dev_err(dev, "Unable to get usable DMA configuration\n");
1342 		goto err_release_regions;
1343 	}
1344 
1345 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
1346 	if (err) {
1347 		dev_err(dev, "Unable to get 48-bit DMA for consistent allocations\n");
1348 		goto err_release_regions;
1349 	}
1350 
1351 	/* MAP PF's configuration registers */
1352 	nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
1353 	if (!nic->reg_base) {
1354 		dev_err(dev, "Cannot map config register space, aborting\n");
1355 		err = -ENOMEM;
1356 		goto err_release_regions;
1357 	}
1358 
1359 	nic->node = nic_get_node_id(pdev);
1360 
1361 	/* Initialize hardware */
1362 	err = nic_init_hw(nic);
1363 	if (err)
1364 		goto err_release_regions;
1365 
1366 	nic_set_lmac_vf_mapping(nic);
1367 
1368 	/* Register interrupts */
1369 	err = nic_register_interrupts(nic);
1370 	if (err)
1371 		goto err_release_regions;
1372 
1373 	/* Configure SRIOV */
1374 	err = nic_sriov_init(pdev, nic);
1375 	if (err)
1376 		goto err_unregister_interrupts;
1377 
1378 	/* Register a physical link status poll fn() */
1379 	nic->check_link = alloc_workqueue("check_link_status",
1380 					  WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1381 	if (!nic->check_link) {
1382 		err = -ENOMEM;
1383 		goto err_disable_sriov;
1384 	}
1385 
1386 	INIT_DELAYED_WORK(&nic->dwork, nic_poll_for_link);
1387 	queue_delayed_work(nic->check_link, &nic->dwork, 0);
1388 
1389 	return 0;
1390 
1391 err_disable_sriov:
1392 	if (nic->flags & NIC_SRIOV_ENABLED)
1393 		pci_disable_sriov(pdev);
1394 err_unregister_interrupts:
1395 	nic_unregister_interrupts(nic);
1396 err_release_regions:
1397 	pci_release_regions(pdev);
1398 err_disable_device:
1399 	nic_free_lmacmem(nic);
1400 	devm_kfree(dev, nic->hw);
1401 	devm_kfree(dev, nic);
1402 	pci_disable_device(pdev);
1403 	pci_set_drvdata(pdev, NULL);
1404 	return err;
1405 }
1406 
1407 static void nic_remove(struct pci_dev *pdev)
1408 {
1409 	struct nicpf *nic = pci_get_drvdata(pdev);
1410 
1411 	if (nic->flags & NIC_SRIOV_ENABLED)
1412 		pci_disable_sriov(pdev);
1413 
1414 	if (nic->check_link) {
1415 		/* Destroy work Queue */
1416 		cancel_delayed_work_sync(&nic->dwork);
1417 		destroy_workqueue(nic->check_link);
1418 	}
1419 
1420 	nic_unregister_interrupts(nic);
1421 	pci_release_regions(pdev);
1422 
1423 	nic_free_lmacmem(nic);
1424 	devm_kfree(&pdev->dev, nic->hw);
1425 	devm_kfree(&pdev->dev, nic);
1426 
1427 	pci_disable_device(pdev);
1428 	pci_set_drvdata(pdev, NULL);
1429 }
1430 
1431 static struct pci_driver nic_driver = {
1432 	.name = DRV_NAME,
1433 	.id_table = nic_id_table,
1434 	.probe = nic_probe,
1435 	.remove = nic_remove,
1436 };
1437 
1438 static int __init nic_init_module(void)
1439 {
1440 	pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
1441 
1442 	return pci_register_driver(&nic_driver);
1443 }
1444 
1445 static void __exit nic_cleanup_module(void)
1446 {
1447 	pci_unregister_driver(&nic_driver);
1448 }
1449 
1450 module_init(nic_init_module);
1451 module_exit(nic_cleanup_module);
1452