1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2015 Cavium, Inc.
4  */
5 
6 #include <linux/module.h>
7 #include <linux/interrupt.h>
8 #include <linux/pci.h>
9 #include <linux/etherdevice.h>
10 #include <linux/of.h>
11 #include <linux/if_vlan.h>
12 
13 #include "nic_reg.h"
14 #include "nic.h"
15 #include "q_struct.h"
16 #include "thunder_bgx.h"
17 
18 #define DRV_NAME	"nicpf"
19 #define DRV_VERSION	"1.0"
20 
21 #define NIC_VF_PER_MBX_REG      64
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 	u16			cpi_base[MAX_NUM_VFS_SUPPORTED];
58 	u16			rssi_base[MAX_NUM_VFS_SUPPORTED];
59 
60 	/* MSI-X */
61 	u8			num_vec;
62 	bool			irq_allocated[NIC_PF_MSIX_VECTORS];
63 	char			irq_name[NIC_PF_MSIX_VECTORS][20];
64 };
65 
66 /* Supported devices */
67 static const struct pci_device_id nic_id_table[] = {
68 	{ PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, PCI_DEVICE_ID_THUNDER_NIC_PF) },
69 	{ 0, }  /* end of table */
70 };
71 
72 MODULE_AUTHOR("Sunil Goutham");
73 MODULE_DESCRIPTION("Cavium Thunder NIC Physical Function Driver");
74 MODULE_LICENSE("GPL v2");
75 MODULE_VERSION(DRV_VERSION);
76 MODULE_DEVICE_TABLE(pci, nic_id_table);
77 
78 /* The Cavium ThunderX network controller can *only* be found in SoCs
79  * containing the ThunderX ARM64 CPU implementation.  All accesses to the device
80  * registers on this platform are implicitly strongly ordered with respect
81  * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
82  * with no memory barriers in this driver.  The readq()/writeq() functions add
83  * explicit ordering operation which in this case are redundant, and only
84  * add overhead.
85  */
86 
87 /* Register read/write APIs */
88 static void nic_reg_write(struct nicpf *nic, u64 offset, u64 val)
89 {
90 	writeq_relaxed(val, nic->reg_base + offset);
91 }
92 
93 static u64 nic_reg_read(struct nicpf *nic, u64 offset)
94 {
95 	return readq_relaxed(nic->reg_base + offset);
96 }
97 
98 /* PF -> VF mailbox communication APIs */
99 static void nic_enable_mbx_intr(struct nicpf *nic)
100 {
101 	int vf_cnt = pci_sriov_get_totalvfs(nic->pdev);
102 
103 #define INTR_MASK(vfs) ((vfs < 64) ? (BIT_ULL(vfs) - 1) : (~0ull))
104 
105 	/* Clear it, to avoid spurious interrupts (if any) */
106 	nic_reg_write(nic, NIC_PF_MAILBOX_INT, INTR_MASK(vf_cnt));
107 
108 	/* Enable mailbox interrupt for all VFs */
109 	nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S, INTR_MASK(vf_cnt));
110 	/* One mailbox intr enable reg per 64 VFs */
111 	if (vf_cnt > 64) {
112 		nic_reg_write(nic, NIC_PF_MAILBOX_INT + sizeof(u64),
113 			      INTR_MASK(vf_cnt - 64));
114 		nic_reg_write(nic, NIC_PF_MAILBOX_ENA_W1S + sizeof(u64),
115 			      INTR_MASK(vf_cnt - 64));
116 	}
117 }
118 
119 static void nic_clear_mbx_intr(struct nicpf *nic, int vf, int mbx_reg)
120 {
121 	nic_reg_write(nic, NIC_PF_MAILBOX_INT + (mbx_reg << 3), BIT_ULL(vf));
122 }
123 
124 static u64 nic_get_mbx_addr(int vf)
125 {
126 	return NIC_PF_VF_0_127_MAILBOX_0_1 + (vf << NIC_VF_NUM_SHIFT);
127 }
128 
129 /* Send a mailbox message to VF
130  * @vf: vf to which this message to be sent
131  * @mbx: Message to be sent
132  */
133 static void nic_send_msg_to_vf(struct nicpf *nic, int vf, union nic_mbx *mbx)
134 {
135 	void __iomem *mbx_addr = nic->reg_base + nic_get_mbx_addr(vf);
136 	u64 *msg = (u64 *)mbx;
137 
138 	/* In first revision HW, mbox interrupt is triggerred
139 	 * when PF writes to MBOX(1), in next revisions when
140 	 * PF writes to MBOX(0)
141 	 */
142 	if (pass1_silicon(nic->pdev)) {
143 		/* see the comment for nic_reg_write()/nic_reg_read()
144 		 * functions above
145 		 */
146 		writeq_relaxed(msg[0], mbx_addr);
147 		writeq_relaxed(msg[1], mbx_addr + 8);
148 	} else {
149 		writeq_relaxed(msg[1], mbx_addr + 8);
150 		writeq_relaxed(msg[0], mbx_addr);
151 	}
152 }
153 
154 /* Responds to VF's READY message with VF's
155  * ID, node, MAC address e.t.c
156  * @vf: VF which sent READY message
157  */
158 static void nic_mbx_send_ready(struct nicpf *nic, int vf)
159 {
160 	union nic_mbx mbx = {};
161 	int bgx_idx, lmac;
162 	const char *mac;
163 
164 	mbx.nic_cfg.msg = NIC_MBOX_MSG_READY;
165 	mbx.nic_cfg.vf_id = vf;
166 
167 	mbx.nic_cfg.tns_mode = NIC_TNS_BYPASS_MODE;
168 
169 	if (vf < nic->num_vf_en) {
170 		bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
171 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
172 
173 		mac = bgx_get_lmac_mac(nic->node, bgx_idx, lmac);
174 		if (mac)
175 			ether_addr_copy((u8 *)&mbx.nic_cfg.mac_addr, mac);
176 	}
177 	mbx.nic_cfg.sqs_mode = (vf >= nic->num_vf_en) ? true : false;
178 	mbx.nic_cfg.node_id = nic->node;
179 
180 	mbx.nic_cfg.loopback_supported = vf < nic->num_vf_en;
181 
182 	nic_send_msg_to_vf(nic, vf, &mbx);
183 }
184 
185 /* ACKs VF's mailbox message
186  * @vf: VF to which ACK to be sent
187  */
188 static void nic_mbx_send_ack(struct nicpf *nic, int vf)
189 {
190 	union nic_mbx mbx = {};
191 
192 	mbx.msg.msg = NIC_MBOX_MSG_ACK;
193 	nic_send_msg_to_vf(nic, vf, &mbx);
194 }
195 
196 /* NACKs VF's mailbox message that PF is not able to
197  * complete the action
198  * @vf: VF to which ACK to be sent
199  */
200 static void nic_mbx_send_nack(struct nicpf *nic, int vf)
201 {
202 	union nic_mbx mbx = {};
203 
204 	mbx.msg.msg = NIC_MBOX_MSG_NACK;
205 	nic_send_msg_to_vf(nic, vf, &mbx);
206 }
207 
208 /* Flush all in flight receive packets to memory and
209  * bring down an active RQ
210  */
211 static int nic_rcv_queue_sw_sync(struct nicpf *nic)
212 {
213 	u16 timeout = ~0x00;
214 
215 	nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x01);
216 	/* Wait till sync cycle is finished */
217 	while (timeout) {
218 		if (nic_reg_read(nic, NIC_PF_SW_SYNC_RX_DONE) & 0x1)
219 			break;
220 		timeout--;
221 	}
222 	nic_reg_write(nic, NIC_PF_SW_SYNC_RX, 0x00);
223 	if (!timeout) {
224 		dev_err(&nic->pdev->dev, "Receive queue software sync failed");
225 		return 1;
226 	}
227 	return 0;
228 }
229 
230 /* Get BGX Rx/Tx stats and respond to VF's request */
231 static void nic_get_bgx_stats(struct nicpf *nic, struct bgx_stats_msg *bgx)
232 {
233 	int bgx_idx, lmac;
234 	union nic_mbx mbx = {};
235 
236 	bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
237 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[bgx->vf_id]);
238 
239 	mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
240 	mbx.bgx_stats.vf_id = bgx->vf_id;
241 	mbx.bgx_stats.rx = bgx->rx;
242 	mbx.bgx_stats.idx = bgx->idx;
243 	if (bgx->rx)
244 		mbx.bgx_stats.stats = bgx_get_rx_stats(nic->node, bgx_idx,
245 							    lmac, bgx->idx);
246 	else
247 		mbx.bgx_stats.stats = bgx_get_tx_stats(nic->node, bgx_idx,
248 							    lmac, bgx->idx);
249 	nic_send_msg_to_vf(nic, bgx->vf_id, &mbx);
250 }
251 
252 /* Update hardware min/max frame size */
253 static int nic_update_hw_frs(struct nicpf *nic, int new_frs, int vf)
254 {
255 	int bgx, lmac, lmac_cnt;
256 	u64 lmac_credits;
257 
258 	if ((new_frs > NIC_HW_MAX_FRS) || (new_frs < NIC_HW_MIN_FRS))
259 		return 1;
260 
261 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
262 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
263 	lmac += bgx * MAX_LMAC_PER_BGX;
264 
265 	new_frs += VLAN_ETH_HLEN + ETH_FCS_LEN + 4;
266 
267 	/* Update corresponding LMAC credits */
268 	lmac_cnt = bgx_get_lmac_count(nic->node, bgx);
269 	lmac_credits = nic_reg_read(nic, NIC_PF_LMAC_0_7_CREDIT + (lmac * 8));
270 	lmac_credits &= ~(0xFFFFFULL << 12);
271 	lmac_credits |= (((((48 * 1024) / lmac_cnt) - new_frs) / 16) << 12);
272 	nic_reg_write(nic, NIC_PF_LMAC_0_7_CREDIT + (lmac * 8), lmac_credits);
273 
274 	/* Enforce MTU in HW
275 	 * This config is supported only from 88xx pass 2.0 onwards.
276 	 */
277 	if (!pass1_silicon(nic->pdev))
278 		nic_reg_write(nic,
279 			      NIC_PF_LMAC_0_7_CFG2 + (lmac * 8), new_frs);
280 	return 0;
281 }
282 
283 /* Set minimum transmit packet size */
284 static void nic_set_tx_pkt_pad(struct nicpf *nic, int size)
285 {
286 	int lmac, max_lmac;
287 	u16 sdevid;
288 	u64 lmac_cfg;
289 
290 	/* There is a issue in HW where-in while sending GSO sized
291 	 * pkts as part of TSO, if pkt len falls below this size
292 	 * NIC will zero PAD packet and also updates IP total length.
293 	 * Hence set this value to lessthan min pkt size of MAC+IP+TCP
294 	 * headers, BGX will do the padding to transmit 64 byte pkt.
295 	 */
296 	if (size > 52)
297 		size = 52;
298 
299 	pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
300 	/* 81xx's RGX has only one LMAC */
301 	if (sdevid == PCI_SUBSYS_DEVID_81XX_NIC_PF)
302 		max_lmac = ((nic->hw->bgx_cnt - 1) * MAX_LMAC_PER_BGX) + 1;
303 	else
304 		max_lmac = nic->hw->bgx_cnt * MAX_LMAC_PER_BGX;
305 
306 	for (lmac = 0; lmac < max_lmac; lmac++) {
307 		lmac_cfg = nic_reg_read(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3));
308 		lmac_cfg &= ~(0xF << 2);
309 		lmac_cfg |= ((size / 4) << 2);
310 		nic_reg_write(nic, NIC_PF_LMAC_0_7_CFG | (lmac << 3), lmac_cfg);
311 	}
312 }
313 
314 /* Function to check number of LMACs present and set VF::LMAC mapping.
315  * Mapping will be used while initializing channels.
316  */
317 static void nic_set_lmac_vf_mapping(struct nicpf *nic)
318 {
319 	unsigned bgx_map = bgx_get_map(nic->node);
320 	int bgx, next_bgx_lmac = 0;
321 	int lmac, lmac_cnt = 0;
322 	u64 lmac_credit;
323 
324 	nic->num_vf_en = 0;
325 
326 	for (bgx = 0; bgx < nic->hw->bgx_cnt; bgx++) {
327 		if (!(bgx_map & (1 << bgx)))
328 			continue;
329 		lmac_cnt = bgx_get_lmac_count(nic->node, bgx);
330 		for (lmac = 0; lmac < lmac_cnt; lmac++)
331 			nic->vf_lmac_map[next_bgx_lmac++] =
332 						NIC_SET_VF_LMAC_MAP(bgx, lmac);
333 		nic->num_vf_en += lmac_cnt;
334 
335 		/* Program LMAC credits */
336 		lmac_credit = (1ull << 1); /* channel credit enable */
337 		lmac_credit |= (0x1ff << 2); /* Max outstanding pkt count */
338 		/* 48KB BGX Tx buffer size, each unit is of size 16bytes */
339 		lmac_credit |= (((((48 * 1024) / lmac_cnt) -
340 				NIC_HW_MAX_FRS) / 16) << 12);
341 		lmac = bgx * MAX_LMAC_PER_BGX;
342 		for (; lmac < lmac_cnt + (bgx * MAX_LMAC_PER_BGX); lmac++)
343 			nic_reg_write(nic,
344 				      NIC_PF_LMAC_0_7_CREDIT + (lmac * 8),
345 				      lmac_credit);
346 
347 		/* On CN81XX there are only 8 VFs but max possible no of
348 		 * interfaces are 9.
349 		 */
350 		if (nic->num_vf_en >= pci_sriov_get_totalvfs(nic->pdev)) {
351 			nic->num_vf_en = pci_sriov_get_totalvfs(nic->pdev);
352 			break;
353 		}
354 	}
355 }
356 
357 static void nic_get_hw_info(struct nicpf *nic)
358 {
359 	u16 sdevid;
360 	struct hw_info *hw = nic->hw;
361 
362 	pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
363 
364 	switch (sdevid) {
365 	case PCI_SUBSYS_DEVID_88XX_NIC_PF:
366 		hw->bgx_cnt = MAX_BGX_PER_CN88XX;
367 		hw->chans_per_lmac = 16;
368 		hw->chans_per_bgx = 128;
369 		hw->cpi_cnt = 2048;
370 		hw->rssi_cnt = 4096;
371 		hw->rss_ind_tbl_size = NIC_MAX_RSS_IDR_TBL_SIZE;
372 		hw->tl3_cnt = 256;
373 		hw->tl2_cnt = 64;
374 		hw->tl1_cnt = 2;
375 		hw->tl1_per_bgx = true;
376 		break;
377 	case PCI_SUBSYS_DEVID_81XX_NIC_PF:
378 		hw->bgx_cnt = MAX_BGX_PER_CN81XX;
379 		hw->chans_per_lmac = 8;
380 		hw->chans_per_bgx = 32;
381 		hw->chans_per_rgx = 8;
382 		hw->chans_per_lbk = 24;
383 		hw->cpi_cnt = 512;
384 		hw->rssi_cnt = 256;
385 		hw->rss_ind_tbl_size = 32; /* Max RSSI / Max interfaces */
386 		hw->tl3_cnt = 64;
387 		hw->tl2_cnt = 16;
388 		hw->tl1_cnt = 10;
389 		hw->tl1_per_bgx = false;
390 		break;
391 	case PCI_SUBSYS_DEVID_83XX_NIC_PF:
392 		hw->bgx_cnt = MAX_BGX_PER_CN83XX;
393 		hw->chans_per_lmac = 8;
394 		hw->chans_per_bgx = 32;
395 		hw->chans_per_lbk = 64;
396 		hw->cpi_cnt = 2048;
397 		hw->rssi_cnt = 1024;
398 		hw->rss_ind_tbl_size = 64; /* Max RSSI / Max interfaces */
399 		hw->tl3_cnt = 256;
400 		hw->tl2_cnt = 64;
401 		hw->tl1_cnt = 18;
402 		hw->tl1_per_bgx = false;
403 		break;
404 	}
405 	hw->tl4_cnt = MAX_QUEUES_PER_QSET * pci_sriov_get_totalvfs(nic->pdev);
406 }
407 
408 #define BGX0_BLOCK 8
409 #define BGX1_BLOCK 9
410 
411 static void nic_init_hw(struct nicpf *nic)
412 {
413 	int i;
414 	u64 cqm_cfg;
415 
416 	/* Enable NIC HW block */
417 	nic_reg_write(nic, NIC_PF_CFG, 0x3);
418 
419 	/* Enable backpressure */
420 	nic_reg_write(nic, NIC_PF_BP_CFG, (1ULL << 6) | 0x03);
421 
422 	/* TNS and TNS bypass modes are present only on 88xx
423 	 * Also offset of this CSR has changed in 81xx and 83xx.
424 	 */
425 	if (nic->pdev->subsystem_device == PCI_SUBSYS_DEVID_88XX_NIC_PF) {
426 		/* Disable TNS mode on both interfaces */
427 		nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG,
428 			      (NIC_TNS_BYPASS_MODE << 7) |
429 			      BGX0_BLOCK | (1ULL << 16));
430 		nic_reg_write(nic, NIC_PF_INTF_0_1_SEND_CFG | (1 << 8),
431 			      (NIC_TNS_BYPASS_MODE << 7) |
432 			      BGX1_BLOCK | (1ULL << 16));
433 	} else {
434 		/* Configure timestamp generation timeout to 10us */
435 		for (i = 0; i < nic->hw->bgx_cnt; i++)
436 			nic_reg_write(nic, NIC_PF_INTFX_SEND_CFG | (i << 3),
437 				      (1ULL << 16));
438 	}
439 
440 	nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG,
441 		      (1ULL << 63) | BGX0_BLOCK);
442 	nic_reg_write(nic, NIC_PF_INTF_0_1_BP_CFG + (1 << 8),
443 		      (1ULL << 63) | BGX1_BLOCK);
444 
445 	/* PKIND configuration */
446 	nic->pkind.minlen = 0;
447 	nic->pkind.maxlen = NIC_HW_MAX_FRS + VLAN_ETH_HLEN + ETH_FCS_LEN + 4;
448 	nic->pkind.lenerr_en = 1;
449 	nic->pkind.rx_hdr = 0;
450 	nic->pkind.hdr_sl = 0;
451 
452 	for (i = 0; i < NIC_MAX_PKIND; i++)
453 		nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG | (i << 3),
454 			      *(u64 *)&nic->pkind);
455 
456 	nic_set_tx_pkt_pad(nic, NIC_HW_MIN_FRS);
457 
458 	/* Timer config */
459 	nic_reg_write(nic, NIC_PF_INTR_TIMER_CFG, NICPF_CLK_PER_INT_TICK);
460 
461 	/* Enable VLAN ethertype matching and stripping */
462 	nic_reg_write(nic, NIC_PF_RX_ETYPE_0_7,
463 		      (2 << 19) | (ETYPE_ALG_VLAN_STRIP << 16) | ETH_P_8021Q);
464 
465 	/* Check if HW expected value is higher (could be in future chips) */
466 	cqm_cfg = nic_reg_read(nic, NIC_PF_CQM_CFG);
467 	if (cqm_cfg < NICPF_CQM_MIN_DROP_LEVEL)
468 		nic_reg_write(nic, NIC_PF_CQM_CFG, NICPF_CQM_MIN_DROP_LEVEL);
469 }
470 
471 /* Channel parse index configuration */
472 static void nic_config_cpi(struct nicpf *nic, struct cpi_cfg_msg *cfg)
473 {
474 	struct hw_info *hw = nic->hw;
475 	u32 vnic, bgx, lmac, chan;
476 	u32 padd, cpi_count = 0;
477 	u64 cpi_base, cpi, rssi_base, rssi;
478 	u8  qset, rq_idx = 0;
479 
480 	vnic = cfg->vf_id;
481 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
482 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
483 
484 	chan = (lmac * hw->chans_per_lmac) + (bgx * hw->chans_per_bgx);
485 	cpi_base = vnic * NIC_MAX_CPI_PER_LMAC;
486 	rssi_base = vnic * hw->rss_ind_tbl_size;
487 
488 	/* Rx channel configuration */
489 	nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_BP_CFG | (chan << 3),
490 		      (1ull << 63) | (vnic << 0));
491 	nic_reg_write(nic, NIC_PF_CHAN_0_255_RX_CFG | (chan << 3),
492 		      ((u64)cfg->cpi_alg << 62) | (cpi_base << 48));
493 
494 	if (cfg->cpi_alg == CPI_ALG_NONE)
495 		cpi_count = 1;
496 	else if (cfg->cpi_alg == CPI_ALG_VLAN) /* 3 bits of PCP */
497 		cpi_count = 8;
498 	else if (cfg->cpi_alg == CPI_ALG_VLAN16) /* 3 bits PCP + DEI */
499 		cpi_count = 16;
500 	else if (cfg->cpi_alg == CPI_ALG_DIFF) /* 6bits DSCP */
501 		cpi_count = NIC_MAX_CPI_PER_LMAC;
502 
503 	/* RSS Qset, Qidx mapping */
504 	qset = cfg->vf_id;
505 	rssi = rssi_base;
506 	for (; rssi < (rssi_base + cfg->rq_cnt); rssi++) {
507 		nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
508 			      (qset << 3) | rq_idx);
509 		rq_idx++;
510 	}
511 
512 	rssi = 0;
513 	cpi = cpi_base;
514 	for (; cpi < (cpi_base + cpi_count); cpi++) {
515 		/* Determine port to channel adder */
516 		if (cfg->cpi_alg != CPI_ALG_DIFF)
517 			padd = cpi % cpi_count;
518 		else
519 			padd = cpi % 8; /* 3 bits CS out of 6bits DSCP */
520 
521 		/* Leave RSS_SIZE as '0' to disable RSS */
522 		if (pass1_silicon(nic->pdev)) {
523 			nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
524 				      (vnic << 24) | (padd << 16) |
525 				      (rssi_base + rssi));
526 		} else {
527 			/* Set MPI_ALG to '0' to disable MCAM parsing */
528 			nic_reg_write(nic, NIC_PF_CPI_0_2047_CFG | (cpi << 3),
529 				      (padd << 16));
530 			/* MPI index is same as CPI if MPI_ALG is not enabled */
531 			nic_reg_write(nic, NIC_PF_MPI_0_2047_CFG | (cpi << 3),
532 				      (vnic << 24) | (rssi_base + rssi));
533 		}
534 
535 		if ((rssi + 1) >= cfg->rq_cnt)
536 			continue;
537 
538 		if (cfg->cpi_alg == CPI_ALG_VLAN)
539 			rssi++;
540 		else if (cfg->cpi_alg == CPI_ALG_VLAN16)
541 			rssi = ((cpi - cpi_base) & 0xe) >> 1;
542 		else if (cfg->cpi_alg == CPI_ALG_DIFF)
543 			rssi = ((cpi - cpi_base) & 0x38) >> 3;
544 	}
545 	nic->cpi_base[cfg->vf_id] = cpi_base;
546 	nic->rssi_base[cfg->vf_id] = rssi_base;
547 }
548 
549 /* Responsds to VF with its RSS indirection table size */
550 static void nic_send_rss_size(struct nicpf *nic, int vf)
551 {
552 	union nic_mbx mbx = {};
553 
554 	mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
555 	mbx.rss_size.ind_tbl_size = nic->hw->rss_ind_tbl_size;
556 	nic_send_msg_to_vf(nic, vf, &mbx);
557 }
558 
559 /* Receive side scaling configuration
560  * configure:
561  * - RSS index
562  * - indir table i.e hash::RQ mapping
563  * - no of hash bits to consider
564  */
565 static void nic_config_rss(struct nicpf *nic, struct rss_cfg_msg *cfg)
566 {
567 	u8  qset, idx = 0;
568 	u64 cpi_cfg, cpi_base, rssi_base, rssi;
569 	u64 idx_addr;
570 
571 	rssi_base = nic->rssi_base[cfg->vf_id] + cfg->tbl_offset;
572 
573 	rssi = rssi_base;
574 
575 	for (; rssi < (rssi_base + cfg->tbl_len); rssi++) {
576 		u8 svf = cfg->ind_tbl[idx] >> 3;
577 
578 		if (svf)
579 			qset = nic->vf_sqs[cfg->vf_id][svf - 1];
580 		else
581 			qset = cfg->vf_id;
582 		nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
583 			      (qset << 3) | (cfg->ind_tbl[idx] & 0x7));
584 		idx++;
585 	}
586 
587 	cpi_base = nic->cpi_base[cfg->vf_id];
588 	if (pass1_silicon(nic->pdev))
589 		idx_addr = NIC_PF_CPI_0_2047_CFG;
590 	else
591 		idx_addr = NIC_PF_MPI_0_2047_CFG;
592 	cpi_cfg = nic_reg_read(nic, idx_addr | (cpi_base << 3));
593 	cpi_cfg &= ~(0xFULL << 20);
594 	cpi_cfg |= (cfg->hash_bits << 20);
595 	nic_reg_write(nic, idx_addr | (cpi_base << 3), cpi_cfg);
596 }
597 
598 /* 4 level transmit side scheduler configutation
599  * for TNS bypass mode
600  *
601  * Sample configuration for SQ0 on 88xx
602  * VNIC0-SQ0 -> TL4(0)   -> TL3[0]   -> TL2[0]  -> TL1[0] -> BGX0
603  * VNIC1-SQ0 -> TL4(8)   -> TL3[2]   -> TL2[0]  -> TL1[0] -> BGX0
604  * VNIC2-SQ0 -> TL4(16)  -> TL3[4]   -> TL2[1]  -> TL1[0] -> BGX0
605  * VNIC3-SQ0 -> TL4(24)  -> TL3[6]   -> TL2[1]  -> TL1[0] -> BGX0
606  * VNIC4-SQ0 -> TL4(512) -> TL3[128] -> TL2[32] -> TL1[1] -> BGX1
607  * VNIC5-SQ0 -> TL4(520) -> TL3[130] -> TL2[32] -> TL1[1] -> BGX1
608  * VNIC6-SQ0 -> TL4(528) -> TL3[132] -> TL2[33] -> TL1[1] -> BGX1
609  * VNIC7-SQ0 -> TL4(536) -> TL3[134] -> TL2[33] -> TL1[1] -> BGX1
610  */
611 static void nic_tx_channel_cfg(struct nicpf *nic, u8 vnic,
612 			       struct sq_cfg_msg *sq)
613 {
614 	struct hw_info *hw = nic->hw;
615 	u32 bgx, lmac, chan;
616 	u32 tl2, tl3, tl4;
617 	u32 rr_quantum;
618 	u8 sq_idx = sq->sq_num;
619 	u8 pqs_vnic;
620 	int svf;
621 
622 	if (sq->sqs_mode)
623 		pqs_vnic = nic->pqs_vf[vnic];
624 	else
625 		pqs_vnic = vnic;
626 
627 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
628 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
629 
630 	/* 24 bytes for FCS, IPG and preamble */
631 	rr_quantum = ((NIC_HW_MAX_FRS + 24) / 4);
632 
633 	/* For 88xx 0-511 TL4 transmits via BGX0 and
634 	 * 512-1023 TL4s transmit via BGX1.
635 	 */
636 	if (hw->tl1_per_bgx) {
637 		tl4 = bgx * (hw->tl4_cnt / hw->bgx_cnt);
638 		if (!sq->sqs_mode) {
639 			tl4 += (lmac * MAX_QUEUES_PER_QSET);
640 		} else {
641 			for (svf = 0; svf < MAX_SQS_PER_VF; svf++) {
642 				if (nic->vf_sqs[pqs_vnic][svf] == vnic)
643 					break;
644 			}
645 			tl4 += (MAX_LMAC_PER_BGX * MAX_QUEUES_PER_QSET);
646 			tl4 += (lmac * MAX_QUEUES_PER_QSET * MAX_SQS_PER_VF);
647 			tl4 += (svf * MAX_QUEUES_PER_QSET);
648 		}
649 	} else {
650 		tl4 = (vnic * MAX_QUEUES_PER_QSET);
651 	}
652 	tl4 += sq_idx;
653 
654 	tl3 = tl4 / (hw->tl4_cnt / hw->tl3_cnt);
655 	nic_reg_write(nic, NIC_PF_QSET_0_127_SQ_0_7_CFG2 |
656 		      ((u64)vnic << NIC_QS_ID_SHIFT) |
657 		      ((u32)sq_idx << NIC_Q_NUM_SHIFT), tl4);
658 	nic_reg_write(nic, NIC_PF_TL4_0_1023_CFG | (tl4 << 3),
659 		      ((u64)vnic << 27) | ((u32)sq_idx << 24) | rr_quantum);
660 
661 	nic_reg_write(nic, NIC_PF_TL3_0_255_CFG | (tl3 << 3), rr_quantum);
662 
663 	/* On 88xx 0-127 channels are for BGX0 and
664 	 * 127-255 channels for BGX1.
665 	 *
666 	 * On 81xx/83xx TL3_CHAN reg should be configured with channel
667 	 * within LMAC i.e 0-7 and not the actual channel number like on 88xx
668 	 */
669 	chan = (lmac * hw->chans_per_lmac) + (bgx * hw->chans_per_bgx);
670 	if (hw->tl1_per_bgx)
671 		nic_reg_write(nic, NIC_PF_TL3_0_255_CHAN | (tl3 << 3), chan);
672 	else
673 		nic_reg_write(nic, NIC_PF_TL3_0_255_CHAN | (tl3 << 3), 0);
674 
675 	/* Enable backpressure on the channel */
676 	nic_reg_write(nic, NIC_PF_CHAN_0_255_TX_CFG | (chan << 3), 1);
677 
678 	tl2 = tl3 >> 2;
679 	nic_reg_write(nic, NIC_PF_TL3A_0_63_CFG | (tl2 << 3), tl2);
680 	nic_reg_write(nic, NIC_PF_TL2_0_63_CFG | (tl2 << 3), rr_quantum);
681 	/* No priorities as of now */
682 	nic_reg_write(nic, NIC_PF_TL2_0_63_PRI | (tl2 << 3), 0x00);
683 
684 	/* Unlike 88xx where TL2s 0-31 transmits to TL1 '0' and rest to TL1 '1'
685 	 * on 81xx/83xx TL2 needs to be configured to transmit to one of the
686 	 * possible LMACs.
687 	 *
688 	 * This register doesn't exist on 88xx.
689 	 */
690 	if (!hw->tl1_per_bgx)
691 		nic_reg_write(nic, NIC_PF_TL2_LMAC | (tl2 << 3),
692 			      lmac + (bgx * MAX_LMAC_PER_BGX));
693 }
694 
695 /* Send primary nicvf pointer to secondary QS's VF */
696 static void nic_send_pnicvf(struct nicpf *nic, int sqs)
697 {
698 	union nic_mbx mbx = {};
699 
700 	mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
701 	mbx.nicvf.nicvf = nic->nicvf[nic->pqs_vf[sqs]];
702 	nic_send_msg_to_vf(nic, sqs, &mbx);
703 }
704 
705 /* Send SQS's nicvf pointer to primary QS's VF */
706 static void nic_send_snicvf(struct nicpf *nic, struct nicvf_ptr *nicvf)
707 {
708 	union nic_mbx mbx = {};
709 	int sqs_id = nic->vf_sqs[nicvf->vf_id][nicvf->sqs_id];
710 
711 	mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
712 	mbx.nicvf.sqs_id = nicvf->sqs_id;
713 	mbx.nicvf.nicvf = nic->nicvf[sqs_id];
714 	nic_send_msg_to_vf(nic, nicvf->vf_id, &mbx);
715 }
716 
717 /* Find next available Qset that can be assigned as a
718  * secondary Qset to a VF.
719  */
720 static int nic_nxt_avail_sqs(struct nicpf *nic)
721 {
722 	int sqs;
723 
724 	for (sqs = 0; sqs < nic->num_sqs_en; sqs++) {
725 		if (!nic->sqs_used[sqs])
726 			nic->sqs_used[sqs] = true;
727 		else
728 			continue;
729 		return sqs + nic->num_vf_en;
730 	}
731 	return -1;
732 }
733 
734 /* Allocate additional Qsets for requested VF */
735 static void nic_alloc_sqs(struct nicpf *nic, struct sqs_alloc *sqs)
736 {
737 	union nic_mbx mbx = {};
738 	int idx, alloc_qs = 0;
739 	int sqs_id;
740 
741 	if (!nic->num_sqs_en)
742 		goto send_mbox;
743 
744 	for (idx = 0; idx < sqs->qs_count; idx++) {
745 		sqs_id = nic_nxt_avail_sqs(nic);
746 		if (sqs_id < 0)
747 			break;
748 		nic->vf_sqs[sqs->vf_id][idx] = sqs_id;
749 		nic->pqs_vf[sqs_id] = sqs->vf_id;
750 		alloc_qs++;
751 	}
752 
753 send_mbox:
754 	mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
755 	mbx.sqs_alloc.vf_id = sqs->vf_id;
756 	mbx.sqs_alloc.qs_count = alloc_qs;
757 	nic_send_msg_to_vf(nic, sqs->vf_id, &mbx);
758 }
759 
760 static int nic_config_loopback(struct nicpf *nic, struct set_loopback *lbk)
761 {
762 	int bgx_idx, lmac_idx;
763 
764 	if (lbk->vf_id >= nic->num_vf_en)
765 		return -1;
766 
767 	bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
768 	lmac_idx = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lbk->vf_id]);
769 
770 	bgx_lmac_internal_loopback(nic->node, bgx_idx, lmac_idx, lbk->enable);
771 
772 	/* Enable moving average calculation.
773 	 * Keep the LVL/AVG delay to HW enforced minimum so that, not too many
774 	 * packets sneek in between average calculations.
775 	 */
776 	nic_reg_write(nic, NIC_PF_CQ_AVG_CFG,
777 		      (BIT_ULL(20) | 0x2ull << 14 | 0x1));
778 	nic_reg_write(nic, NIC_PF_RRM_AVG_CFG,
779 		      (BIT_ULL(20) | 0x3ull << 14 | 0x1));
780 
781 	return 0;
782 }
783 
784 /* Reset statistics counters */
785 static int nic_reset_stat_counters(struct nicpf *nic,
786 				   int vf, struct reset_stat_cfg *cfg)
787 {
788 	int i, stat, qnum;
789 	u64 reg_addr;
790 
791 	for (i = 0; i < RX_STATS_ENUM_LAST; i++) {
792 		if (cfg->rx_stat_mask & BIT(i)) {
793 			reg_addr = NIC_PF_VNIC_0_127_RX_STAT_0_13 |
794 				   (vf << NIC_QS_ID_SHIFT) |
795 				   (i << 3);
796 			nic_reg_write(nic, reg_addr, 0);
797 		}
798 	}
799 
800 	for (i = 0; i < TX_STATS_ENUM_LAST; i++) {
801 		if (cfg->tx_stat_mask & BIT(i)) {
802 			reg_addr = NIC_PF_VNIC_0_127_TX_STAT_0_4 |
803 				   (vf << NIC_QS_ID_SHIFT) |
804 				   (i << 3);
805 			nic_reg_write(nic, reg_addr, 0);
806 		}
807 	}
808 
809 	for (i = 0; i <= 15; i++) {
810 		qnum = i >> 1;
811 		stat = i & 1 ? 1 : 0;
812 		reg_addr = (vf << NIC_QS_ID_SHIFT) |
813 			   (qnum << NIC_Q_NUM_SHIFT) | (stat << 3);
814 		if (cfg->rq_stat_mask & BIT(i)) {
815 			reg_addr |= NIC_PF_QSET_0_127_RQ_0_7_STAT_0_1;
816 			nic_reg_write(nic, reg_addr, 0);
817 		}
818 		if (cfg->sq_stat_mask & BIT(i)) {
819 			reg_addr |= NIC_PF_QSET_0_127_SQ_0_7_STAT_0_1;
820 			nic_reg_write(nic, reg_addr, 0);
821 		}
822 	}
823 
824 	return 0;
825 }
826 
827 static void nic_enable_tunnel_parsing(struct nicpf *nic, int vf)
828 {
829 	u64 prot_def = (IPV6_PROT << 32) | (IPV4_PROT << 16) | ET_PROT;
830 	u64 vxlan_prot_def = (IPV6_PROT_DEF << 32) |
831 			      (IPV4_PROT_DEF) << 16 | ET_PROT_DEF;
832 
833 	/* Configure tunnel parsing parameters */
834 	nic_reg_write(nic, NIC_PF_RX_GENEVE_DEF,
835 		      (1ULL << 63 | UDP_GENEVE_PORT_NUM));
836 	nic_reg_write(nic, NIC_PF_RX_GENEVE_PROT_DEF,
837 		      ((7ULL << 61) | prot_def));
838 	nic_reg_write(nic, NIC_PF_RX_NVGRE_PROT_DEF,
839 		      ((7ULL << 61) | prot_def));
840 	nic_reg_write(nic, NIC_PF_RX_VXLAN_DEF_0_1,
841 		      ((1ULL << 63) | UDP_VXLAN_PORT_NUM));
842 	nic_reg_write(nic, NIC_PF_RX_VXLAN_PROT_DEF,
843 		      ((0xfULL << 60) | vxlan_prot_def));
844 }
845 
846 static void nic_enable_vf(struct nicpf *nic, int vf, bool enable)
847 {
848 	int bgx, lmac;
849 
850 	nic->vf_enabled[vf] = enable;
851 
852 	if (vf >= nic->num_vf_en)
853 		return;
854 
855 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
856 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
857 
858 	bgx_lmac_rx_tx_enable(nic->node, bgx, lmac, enable);
859 }
860 
861 static void nic_pause_frame(struct nicpf *nic, int vf, struct pfc *cfg)
862 {
863 	int bgx, lmac;
864 	struct pfc pfc;
865 	union nic_mbx mbx = {};
866 
867 	if (vf >= nic->num_vf_en)
868 		return;
869 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
870 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
871 
872 	if (cfg->get) {
873 		bgx_lmac_get_pfc(nic->node, bgx, lmac, &pfc);
874 		mbx.pfc.msg = NIC_MBOX_MSG_PFC;
875 		mbx.pfc.autoneg = pfc.autoneg;
876 		mbx.pfc.fc_rx = pfc.fc_rx;
877 		mbx.pfc.fc_tx = pfc.fc_tx;
878 		nic_send_msg_to_vf(nic, vf, &mbx);
879 	} else {
880 		bgx_lmac_set_pfc(nic->node, bgx, lmac, cfg);
881 		nic_mbx_send_ack(nic, vf);
882 	}
883 }
884 
885 /* Enable or disable HW timestamping by BGX for pkts received on a LMAC */
886 static void nic_config_timestamp(struct nicpf *nic, int vf, struct set_ptp *ptp)
887 {
888 	struct pkind_cfg *pkind;
889 	u8 lmac, bgx_idx;
890 	u64 pkind_val, pkind_idx;
891 
892 	if (vf >= nic->num_vf_en)
893 		return;
894 
895 	bgx_idx = 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 	pkind_idx = lmac + bgx_idx * MAX_LMAC_PER_BGX;
899 	pkind_val = nic_reg_read(nic, NIC_PF_PKIND_0_15_CFG | (pkind_idx << 3));
900 	pkind = (struct pkind_cfg *)&pkind_val;
901 
902 	if (ptp->enable && !pkind->hdr_sl) {
903 		/* Skiplen to exclude 8byte timestamp while parsing pkt
904 		 * If not configured, will result in L2 errors.
905 		 */
906 		pkind->hdr_sl = 4;
907 		/* Adjust max packet length allowed */
908 		pkind->maxlen += (pkind->hdr_sl * 2);
909 		bgx_config_timestamping(nic->node, bgx_idx, lmac, true);
910 		nic_reg_write(nic, NIC_PF_RX_ETYPE_0_7 | (1 << 3),
911 			      (ETYPE_ALG_ENDPARSE << 16) | ETH_P_1588);
912 	} else if (!ptp->enable && pkind->hdr_sl) {
913 		pkind->maxlen -= (pkind->hdr_sl * 2);
914 		pkind->hdr_sl = 0;
915 		bgx_config_timestamping(nic->node, bgx_idx, lmac, false);
916 		nic_reg_write(nic, NIC_PF_RX_ETYPE_0_7 | (1 << 3),
917 			      (ETYPE_ALG_SKIP << 16) | ETH_P_8021Q);
918 	}
919 
920 	nic_reg_write(nic, NIC_PF_PKIND_0_15_CFG | (pkind_idx << 3), pkind_val);
921 }
922 
923 /* Get BGX LMAC link status and update corresponding VF
924  * if there is a change, valid only if internal L2 switch
925  * is not present otherwise VF link is always treated as up
926  */
927 static void nic_link_status_get(struct nicpf *nic, u8 vf)
928 {
929 	union nic_mbx mbx = {};
930 	struct bgx_link_status link;
931 	u8 bgx, lmac;
932 
933 	mbx.link_status.msg = NIC_MBOX_MSG_BGX_LINK_CHANGE;
934 
935 	/* Get BGX, LMAC indices for the VF */
936 	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
937 	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
938 
939 	/* Get interface link status */
940 	bgx_get_lmac_link_state(nic->node, bgx, lmac, &link);
941 
942 	/* Send a mbox message to VF with current link status */
943 	mbx.link_status.link_up = link.link_up;
944 	mbx.link_status.duplex = link.duplex;
945 	mbx.link_status.speed = link.speed;
946 	mbx.link_status.mac_type = link.mac_type;
947 
948 	/* reply with link status */
949 	nic_send_msg_to_vf(nic, vf, &mbx);
950 }
951 
952 /* Interrupt handler to handle mailbox messages from VFs */
953 static void nic_handle_mbx_intr(struct nicpf *nic, int vf)
954 {
955 	union nic_mbx mbx = {};
956 	u64 *mbx_data;
957 	u64 mbx_addr;
958 	u64 reg_addr;
959 	u64 cfg;
960 	int bgx, lmac;
961 	int i;
962 	int ret = 0;
963 
964 	mbx_addr = nic_get_mbx_addr(vf);
965 	mbx_data = (u64 *)&mbx;
966 
967 	for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
968 		*mbx_data = nic_reg_read(nic, mbx_addr);
969 		mbx_data++;
970 		mbx_addr += sizeof(u64);
971 	}
972 
973 	dev_dbg(&nic->pdev->dev, "%s: Mailbox msg 0x%02x from VF%d\n",
974 		__func__, mbx.msg.msg, vf);
975 	switch (mbx.msg.msg) {
976 	case NIC_MBOX_MSG_READY:
977 		nic_mbx_send_ready(nic, vf);
978 		return;
979 	case NIC_MBOX_MSG_QS_CFG:
980 		reg_addr = NIC_PF_QSET_0_127_CFG |
981 			   (mbx.qs.num << NIC_QS_ID_SHIFT);
982 		cfg = mbx.qs.cfg;
983 		/* Check if its a secondary Qset */
984 		if (vf >= nic->num_vf_en) {
985 			cfg = cfg & (~0x7FULL);
986 			/* Assign this Qset to primary Qset's VF */
987 			cfg |= nic->pqs_vf[vf];
988 		}
989 		nic_reg_write(nic, reg_addr, cfg);
990 		break;
991 	case NIC_MBOX_MSG_RQ_CFG:
992 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_CFG |
993 			   (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
994 			   (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
995 		nic_reg_write(nic, reg_addr, mbx.rq.cfg);
996 		/* Enable CQE_RX2_S extension in CQE_RX descriptor.
997 		 * This gets appended by default on 81xx/83xx chips,
998 		 * for consistency enabling the same on 88xx pass2
999 		 * where this is introduced.
1000 		 */
1001 		if (pass2_silicon(nic->pdev))
1002 			nic_reg_write(nic, NIC_PF_RX_CFG, 0x01);
1003 		if (!pass1_silicon(nic->pdev))
1004 			nic_enable_tunnel_parsing(nic, vf);
1005 		break;
1006 	case NIC_MBOX_MSG_RQ_BP_CFG:
1007 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_BP_CFG |
1008 			   (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
1009 			   (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
1010 		nic_reg_write(nic, reg_addr, mbx.rq.cfg);
1011 		break;
1012 	case NIC_MBOX_MSG_RQ_SW_SYNC:
1013 		ret = nic_rcv_queue_sw_sync(nic);
1014 		break;
1015 	case NIC_MBOX_MSG_RQ_DROP_CFG:
1016 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_DROP_CFG |
1017 			   (mbx.rq.qs_num << NIC_QS_ID_SHIFT) |
1018 			   (mbx.rq.rq_num << NIC_Q_NUM_SHIFT);
1019 		nic_reg_write(nic, reg_addr, mbx.rq.cfg);
1020 		break;
1021 	case NIC_MBOX_MSG_SQ_CFG:
1022 		reg_addr = NIC_PF_QSET_0_127_SQ_0_7_CFG |
1023 			   (mbx.sq.qs_num << NIC_QS_ID_SHIFT) |
1024 			   (mbx.sq.sq_num << NIC_Q_NUM_SHIFT);
1025 		nic_reg_write(nic, reg_addr, mbx.sq.cfg);
1026 		nic_tx_channel_cfg(nic, mbx.qs.num, &mbx.sq);
1027 		break;
1028 	case NIC_MBOX_MSG_SET_MAC:
1029 		if (vf >= nic->num_vf_en) {
1030 			ret = -1; /* NACK */
1031 			break;
1032 		}
1033 		lmac = mbx.mac.vf_id;
1034 		bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
1035 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
1036 		bgx_set_lmac_mac(nic->node, bgx, lmac, mbx.mac.mac_addr);
1037 		break;
1038 	case NIC_MBOX_MSG_SET_MAX_FRS:
1039 		ret = nic_update_hw_frs(nic, mbx.frs.max_frs,
1040 					mbx.frs.vf_id);
1041 		break;
1042 	case NIC_MBOX_MSG_CPI_CFG:
1043 		nic_config_cpi(nic, &mbx.cpi_cfg);
1044 		break;
1045 	case NIC_MBOX_MSG_RSS_SIZE:
1046 		nic_send_rss_size(nic, vf);
1047 		return;
1048 	case NIC_MBOX_MSG_RSS_CFG:
1049 	case NIC_MBOX_MSG_RSS_CFG_CONT:
1050 		nic_config_rss(nic, &mbx.rss_cfg);
1051 		break;
1052 	case NIC_MBOX_MSG_CFG_DONE:
1053 		/* Last message of VF config msg sequence */
1054 		nic_enable_vf(nic, vf, true);
1055 		break;
1056 	case NIC_MBOX_MSG_SHUTDOWN:
1057 		/* First msg in VF teardown sequence */
1058 		if (vf >= nic->num_vf_en)
1059 			nic->sqs_used[vf - nic->num_vf_en] = false;
1060 		nic->pqs_vf[vf] = 0;
1061 		nic_enable_vf(nic, vf, false);
1062 		break;
1063 	case NIC_MBOX_MSG_ALLOC_SQS:
1064 		nic_alloc_sqs(nic, &mbx.sqs_alloc);
1065 		return;
1066 	case NIC_MBOX_MSG_NICVF_PTR:
1067 		nic->nicvf[vf] = mbx.nicvf.nicvf;
1068 		break;
1069 	case NIC_MBOX_MSG_PNICVF_PTR:
1070 		nic_send_pnicvf(nic, vf);
1071 		return;
1072 	case NIC_MBOX_MSG_SNICVF_PTR:
1073 		nic_send_snicvf(nic, &mbx.nicvf);
1074 		return;
1075 	case NIC_MBOX_MSG_BGX_STATS:
1076 		nic_get_bgx_stats(nic, &mbx.bgx_stats);
1077 		return;
1078 	case NIC_MBOX_MSG_LOOPBACK:
1079 		ret = nic_config_loopback(nic, &mbx.lbk);
1080 		break;
1081 	case NIC_MBOX_MSG_RESET_STAT_COUNTER:
1082 		ret = nic_reset_stat_counters(nic, vf, &mbx.reset_stat);
1083 		break;
1084 	case NIC_MBOX_MSG_PFC:
1085 		nic_pause_frame(nic, vf, &mbx.pfc);
1086 		return;
1087 	case NIC_MBOX_MSG_PTP_CFG:
1088 		nic_config_timestamp(nic, vf, &mbx.ptp);
1089 		break;
1090 	case NIC_MBOX_MSG_RESET_XCAST:
1091 		if (vf >= nic->num_vf_en) {
1092 			ret = -1; /* NACK */
1093 			break;
1094 		}
1095 		bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1096 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1097 		bgx_reset_xcast_mode(nic->node, bgx, lmac,
1098 				     vf < NIC_VF_PER_MBX_REG ? vf :
1099 				     vf - NIC_VF_PER_MBX_REG);
1100 		break;
1101 
1102 	case NIC_MBOX_MSG_ADD_MCAST:
1103 		if (vf >= nic->num_vf_en) {
1104 			ret = -1; /* NACK */
1105 			break;
1106 		}
1107 		bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1108 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1109 		bgx_set_dmac_cam_filter(nic->node, bgx, lmac,
1110 					mbx.xcast.mac,
1111 					vf < NIC_VF_PER_MBX_REG ? vf :
1112 					vf - NIC_VF_PER_MBX_REG);
1113 		break;
1114 
1115 	case NIC_MBOX_MSG_SET_XCAST:
1116 		if (vf >= nic->num_vf_en) {
1117 			ret = -1; /* NACK */
1118 			break;
1119 		}
1120 		bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1121 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
1122 		bgx_set_xcast_mode(nic->node, bgx, lmac, mbx.xcast.mode);
1123 		break;
1124 	case NIC_MBOX_MSG_BGX_LINK_CHANGE:
1125 		if (vf >= nic->num_vf_en) {
1126 			ret = -1; /* NACK */
1127 			break;
1128 		}
1129 		nic_link_status_get(nic, vf);
1130 		return;
1131 	default:
1132 		dev_err(&nic->pdev->dev,
1133 			"Invalid msg from VF%d, msg 0x%x\n", vf, mbx.msg.msg);
1134 		break;
1135 	}
1136 
1137 	if (!ret) {
1138 		nic_mbx_send_ack(nic, vf);
1139 	} else if (mbx.msg.msg != NIC_MBOX_MSG_READY) {
1140 		dev_err(&nic->pdev->dev, "NACK for MBOX 0x%02x from VF %d\n",
1141 			mbx.msg.msg, vf);
1142 		nic_mbx_send_nack(nic, vf);
1143 	}
1144 }
1145 
1146 static irqreturn_t nic_mbx_intr_handler(int irq, void *nic_irq)
1147 {
1148 	struct nicpf *nic = (struct nicpf *)nic_irq;
1149 	int mbx;
1150 	u64 intr;
1151 	u8  vf;
1152 
1153 	if (irq == pci_irq_vector(nic->pdev, NIC_PF_INTR_ID_MBOX0))
1154 		mbx = 0;
1155 	else
1156 		mbx = 1;
1157 
1158 	intr = nic_reg_read(nic, NIC_PF_MAILBOX_INT + (mbx << 3));
1159 	dev_dbg(&nic->pdev->dev, "PF interrupt Mbox%d 0x%llx\n", mbx, intr);
1160 	for (vf = 0; vf < NIC_VF_PER_MBX_REG; vf++) {
1161 		if (intr & (1ULL << vf)) {
1162 			dev_dbg(&nic->pdev->dev, "Intr from VF %d\n",
1163 				vf + (mbx * NIC_VF_PER_MBX_REG));
1164 
1165 			nic_handle_mbx_intr(nic, vf +
1166 					    (mbx * NIC_VF_PER_MBX_REG));
1167 			nic_clear_mbx_intr(nic, vf, mbx);
1168 		}
1169 	}
1170 	return IRQ_HANDLED;
1171 }
1172 
1173 static void nic_free_all_interrupts(struct nicpf *nic)
1174 {
1175 	int irq;
1176 
1177 	for (irq = 0; irq < nic->num_vec; irq++) {
1178 		if (nic->irq_allocated[irq])
1179 			free_irq(pci_irq_vector(nic->pdev, irq), nic);
1180 		nic->irq_allocated[irq] = false;
1181 	}
1182 }
1183 
1184 static int nic_register_interrupts(struct nicpf *nic)
1185 {
1186 	int i, ret;
1187 	nic->num_vec = pci_msix_vec_count(nic->pdev);
1188 
1189 	/* Enable MSI-X */
1190 	ret = pci_alloc_irq_vectors(nic->pdev, nic->num_vec, nic->num_vec,
1191 				    PCI_IRQ_MSIX);
1192 	if (ret < 0) {
1193 		dev_err(&nic->pdev->dev,
1194 			"Request for #%d msix vectors failed, returned %d\n",
1195 			   nic->num_vec, ret);
1196 		return 1;
1197 	}
1198 
1199 	/* Register mailbox interrupt handler */
1200 	for (i = NIC_PF_INTR_ID_MBOX0; i < nic->num_vec; i++) {
1201 		sprintf(nic->irq_name[i],
1202 			"NICPF Mbox%d", (i - NIC_PF_INTR_ID_MBOX0));
1203 
1204 		ret = request_irq(pci_irq_vector(nic->pdev, i),
1205 				  nic_mbx_intr_handler, 0,
1206 				  nic->irq_name[i], nic);
1207 		if (ret)
1208 			goto fail;
1209 
1210 		nic->irq_allocated[i] = true;
1211 	}
1212 
1213 	/* Enable mailbox interrupt */
1214 	nic_enable_mbx_intr(nic);
1215 	return 0;
1216 
1217 fail:
1218 	dev_err(&nic->pdev->dev, "Request irq failed\n");
1219 	nic_free_all_interrupts(nic);
1220 	pci_free_irq_vectors(nic->pdev);
1221 	nic->num_vec = 0;
1222 	return ret;
1223 }
1224 
1225 static void nic_unregister_interrupts(struct nicpf *nic)
1226 {
1227 	nic_free_all_interrupts(nic);
1228 	pci_free_irq_vectors(nic->pdev);
1229 	nic->num_vec = 0;
1230 }
1231 
1232 static int nic_num_sqs_en(struct nicpf *nic, int vf_en)
1233 {
1234 	int pos, sqs_per_vf = MAX_SQS_PER_VF_SINGLE_NODE;
1235 	u16 total_vf;
1236 
1237 	/* Secondary Qsets are needed only if CPU count is
1238 	 * morethan MAX_QUEUES_PER_QSET.
1239 	 */
1240 	if (num_online_cpus() <= MAX_QUEUES_PER_QSET)
1241 		return 0;
1242 
1243 	/* Check if its a multi-node environment */
1244 	if (nr_node_ids > 1)
1245 		sqs_per_vf = MAX_SQS_PER_VF;
1246 
1247 	pos = pci_find_ext_capability(nic->pdev, PCI_EXT_CAP_ID_SRIOV);
1248 	pci_read_config_word(nic->pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf);
1249 	return min(total_vf - vf_en, vf_en * sqs_per_vf);
1250 }
1251 
1252 static int nic_sriov_init(struct pci_dev *pdev, struct nicpf *nic)
1253 {
1254 	int pos = 0;
1255 	int vf_en;
1256 	int err;
1257 	u16 total_vf_cnt;
1258 
1259 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
1260 	if (!pos) {
1261 		dev_err(&pdev->dev, "SRIOV capability is not found in PCIe config space\n");
1262 		return -ENODEV;
1263 	}
1264 
1265 	pci_read_config_word(pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf_cnt);
1266 	if (total_vf_cnt < nic->num_vf_en)
1267 		nic->num_vf_en = total_vf_cnt;
1268 
1269 	if (!total_vf_cnt)
1270 		return 0;
1271 
1272 	vf_en = nic->num_vf_en;
1273 	nic->num_sqs_en = nic_num_sqs_en(nic, nic->num_vf_en);
1274 	vf_en += nic->num_sqs_en;
1275 
1276 	err = pci_enable_sriov(pdev, vf_en);
1277 	if (err) {
1278 		dev_err(&pdev->dev, "SRIOV enable failed, num VF is %d\n",
1279 			vf_en);
1280 		nic->num_vf_en = 0;
1281 		return err;
1282 	}
1283 
1284 	dev_info(&pdev->dev, "SRIOV enabled, number of VF available %d\n",
1285 		 vf_en);
1286 
1287 	nic->flags |= NIC_SRIOV_ENABLED;
1288 	return 0;
1289 }
1290 
1291 static int nic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1292 {
1293 	struct device *dev = &pdev->dev;
1294 	struct nicpf *nic;
1295 	u8     max_lmac;
1296 	int    err;
1297 
1298 	BUILD_BUG_ON(sizeof(union nic_mbx) > 16);
1299 
1300 	nic = devm_kzalloc(dev, sizeof(*nic), GFP_KERNEL);
1301 	if (!nic)
1302 		return -ENOMEM;
1303 
1304 	nic->hw = devm_kzalloc(dev, sizeof(struct hw_info), GFP_KERNEL);
1305 	if (!nic->hw)
1306 		return -ENOMEM;
1307 
1308 	pci_set_drvdata(pdev, nic);
1309 
1310 	nic->pdev = pdev;
1311 
1312 	err = pci_enable_device(pdev);
1313 	if (err) {
1314 		dev_err(dev, "Failed to enable PCI device\n");
1315 		pci_set_drvdata(pdev, NULL);
1316 		return err;
1317 	}
1318 
1319 	err = pci_request_regions(pdev, DRV_NAME);
1320 	if (err) {
1321 		dev_err(dev, "PCI request regions failed 0x%x\n", err);
1322 		goto err_disable_device;
1323 	}
1324 
1325 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(48));
1326 	if (err) {
1327 		dev_err(dev, "Unable to get usable DMA configuration\n");
1328 		goto err_release_regions;
1329 	}
1330 
1331 	/* MAP PF's configuration registers */
1332 	nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
1333 	if (!nic->reg_base) {
1334 		dev_err(dev, "Cannot map config register space, aborting\n");
1335 		err = -ENOMEM;
1336 		goto err_release_regions;
1337 	}
1338 
1339 	nic->node = nic_get_node_id(pdev);
1340 
1341 	/* Get HW capability info */
1342 	nic_get_hw_info(nic);
1343 
1344 	/* Allocate memory for LMAC tracking elements */
1345 	err = -ENOMEM;
1346 	max_lmac = nic->hw->bgx_cnt * MAX_LMAC_PER_BGX;
1347 
1348 	nic->vf_lmac_map = devm_kmalloc_array(dev, max_lmac, sizeof(u8),
1349 					      GFP_KERNEL);
1350 	if (!nic->vf_lmac_map)
1351 		goto err_release_regions;
1352 
1353 	/* Initialize hardware */
1354 	nic_init_hw(nic);
1355 
1356 	nic_set_lmac_vf_mapping(nic);
1357 
1358 	/* Register interrupts */
1359 	err = nic_register_interrupts(nic);
1360 	if (err)
1361 		goto err_release_regions;
1362 
1363 	/* Configure SRIOV */
1364 	err = nic_sriov_init(pdev, nic);
1365 	if (err)
1366 		goto err_unregister_interrupts;
1367 
1368 	return 0;
1369 
1370 err_unregister_interrupts:
1371 	nic_unregister_interrupts(nic);
1372 err_release_regions:
1373 	pci_release_regions(pdev);
1374 err_disable_device:
1375 	pci_disable_device(pdev);
1376 	pci_set_drvdata(pdev, NULL);
1377 	return err;
1378 }
1379 
1380 static void nic_remove(struct pci_dev *pdev)
1381 {
1382 	struct nicpf *nic = pci_get_drvdata(pdev);
1383 
1384 	if (!nic)
1385 		return;
1386 
1387 	if (nic->flags & NIC_SRIOV_ENABLED)
1388 		pci_disable_sriov(pdev);
1389 
1390 	nic_unregister_interrupts(nic);
1391 	pci_release_regions(pdev);
1392 
1393 	pci_disable_device(pdev);
1394 	pci_set_drvdata(pdev, NULL);
1395 }
1396 
1397 static struct pci_driver nic_driver = {
1398 	.name = DRV_NAME,
1399 	.id_table = nic_id_table,
1400 	.probe = nic_probe,
1401 	.remove = nic_remove,
1402 };
1403 
1404 static int __init nic_init_module(void)
1405 {
1406 	pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
1407 
1408 	return pci_register_driver(&nic_driver);
1409 }
1410 
1411 static void __exit nic_cleanup_module(void)
1412 {
1413 	pci_unregister_driver(&nic_driver);
1414 }
1415 
1416 module_init(nic_init_module);
1417 module_exit(nic_cleanup_module);
1418