xref: /linux/drivers/net/ethernet/marvell/octeontx2/nic/otx2_pf.c (revision 23ca32e4ead48f68e37000f2552b973ef1439acb)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Marvell RVU Physical Function ethernet driver
3  *
4  * Copyright (C) 2020 Marvell.
5  *
6  */
7 
8 #include <linux/module.h>
9 #include <linux/interrupt.h>
10 #include <linux/pci.h>
11 #include <linux/etherdevice.h>
12 #include <linux/of.h>
13 #include <linux/if_vlan.h>
14 #include <linux/iommu.h>
15 #include <net/ip.h>
16 #include <linux/bpf.h>
17 #include <linux/bpf_trace.h>
18 #include <linux/bitfield.h>
19 #include <net/page_pool/types.h>
20 
21 #include "otx2_reg.h"
22 #include "otx2_common.h"
23 #include "otx2_txrx.h"
24 #include "otx2_struct.h"
25 #include "otx2_ptp.h"
26 #include "cn10k.h"
27 #include "qos.h"
28 #include <rvu_trace.h>
29 #include "cn10k_ipsec.h"
30 #include "otx2_xsk.h"
31 
32 #define DRV_NAME	"rvu_nicpf"
33 #define DRV_STRING	"Marvell RVU NIC Physical Function Driver"
34 
35 /* Supported devices */
36 static const struct pci_device_id otx2_pf_id_table[] = {
37 	{ PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, PCI_DEVID_OCTEONTX2_RVU_PF) },
38 	{ 0, }  /* end of table */
39 };
40 
41 MODULE_AUTHOR("Sunil Goutham <sgoutham@marvell.com>");
42 MODULE_DESCRIPTION(DRV_STRING);
43 MODULE_LICENSE("GPL v2");
44 MODULE_DEVICE_TABLE(pci, otx2_pf_id_table);
45 
46 static void otx2_vf_link_event_task(struct work_struct *work);
47 
48 enum {
49 	TYPE_PFAF,
50 	TYPE_PFVF,
51 };
52 
53 static int otx2_config_hw_tx_tstamp(struct otx2_nic *pfvf, bool enable);
54 static int otx2_config_hw_rx_tstamp(struct otx2_nic *pfvf, bool enable);
55 
56 static int otx2_change_mtu(struct net_device *netdev, int new_mtu)
57 {
58 	struct otx2_nic *pf = netdev_priv(netdev);
59 	bool if_up = netif_running(netdev);
60 	int err = 0;
61 
62 	if (pf->xdp_prog && new_mtu > MAX_XDP_MTU) {
63 		netdev_warn(netdev, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
64 			    netdev->mtu);
65 		return -EINVAL;
66 	}
67 	if (if_up)
68 		otx2_stop(netdev);
69 
70 	netdev_info(netdev, "Changing MTU from %d to %d\n",
71 		    netdev->mtu, new_mtu);
72 	WRITE_ONCE(netdev->mtu, new_mtu);
73 
74 	if (if_up)
75 		err = otx2_open(netdev);
76 
77 	return err;
78 }
79 
80 static void otx2_disable_flr_me_intr(struct otx2_nic *pf)
81 {
82 	int irq, vfs = pf->total_vfs;
83 
84 	/* Disable VFs ME interrupts */
85 	otx2_write64(pf, RVU_PF_VFME_INT_ENA_W1CX(0), INTR_MASK(vfs));
86 	irq = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFME0);
87 	free_irq(irq, pf);
88 
89 	/* Disable VFs FLR interrupts */
90 	otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1CX(0), INTR_MASK(vfs));
91 	irq = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFFLR0);
92 	free_irq(irq, pf);
93 
94 	if (vfs <= 64)
95 		return;
96 
97 	otx2_write64(pf, RVU_PF_VFME_INT_ENA_W1CX(1), INTR_MASK(vfs - 64));
98 	irq = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFME1);
99 	free_irq(irq, pf);
100 
101 	otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1CX(1), INTR_MASK(vfs - 64));
102 	irq = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFFLR1);
103 	free_irq(irq, pf);
104 }
105 
106 static void otx2_flr_wq_destroy(struct otx2_nic *pf)
107 {
108 	if (!pf->flr_wq)
109 		return;
110 	destroy_workqueue(pf->flr_wq);
111 	pf->flr_wq = NULL;
112 	devm_kfree(pf->dev, pf->flr_wrk);
113 }
114 
115 static void otx2_flr_handler(struct work_struct *work)
116 {
117 	struct flr_work *flrwork = container_of(work, struct flr_work, work);
118 	struct otx2_nic *pf = flrwork->pf;
119 	struct mbox *mbox = &pf->mbox;
120 	struct msg_req *req;
121 	int vf, reg = 0;
122 
123 	vf = flrwork - pf->flr_wrk;
124 
125 	mutex_lock(&mbox->lock);
126 	req = otx2_mbox_alloc_msg_vf_flr(mbox);
127 	if (!req) {
128 		mutex_unlock(&mbox->lock);
129 		return;
130 	}
131 	req->hdr.pcifunc &= RVU_PFVF_FUNC_MASK;
132 	req->hdr.pcifunc |= (vf + 1) & RVU_PFVF_FUNC_MASK;
133 
134 	if (!otx2_sync_mbox_msg(&pf->mbox)) {
135 		if (vf >= 64) {
136 			reg = 1;
137 			vf = vf - 64;
138 		}
139 		/* clear transcation pending bit */
140 		otx2_write64(pf, RVU_PF_VFTRPENDX(reg), BIT_ULL(vf));
141 		otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1SX(reg), BIT_ULL(vf));
142 	}
143 
144 	mutex_unlock(&mbox->lock);
145 }
146 
147 static irqreturn_t otx2_pf_flr_intr_handler(int irq, void *pf_irq)
148 {
149 	struct otx2_nic *pf = (struct otx2_nic *)pf_irq;
150 	int reg, dev, vf, start_vf, num_reg = 1;
151 	u64 intr;
152 
153 	if (pf->total_vfs > 64)
154 		num_reg = 2;
155 
156 	for (reg = 0; reg < num_reg; reg++) {
157 		intr = otx2_read64(pf, RVU_PF_VFFLR_INTX(reg));
158 		if (!intr)
159 			continue;
160 		start_vf = 64 * reg;
161 		for (vf = 0; vf < 64; vf++) {
162 			if (!(intr & BIT_ULL(vf)))
163 				continue;
164 			dev = vf + start_vf;
165 			queue_work(pf->flr_wq, &pf->flr_wrk[dev].work);
166 			/* Clear interrupt */
167 			otx2_write64(pf, RVU_PF_VFFLR_INTX(reg), BIT_ULL(vf));
168 			/* Disable the interrupt */
169 			otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1CX(reg),
170 				     BIT_ULL(vf));
171 		}
172 	}
173 	return IRQ_HANDLED;
174 }
175 
176 static irqreturn_t otx2_pf_me_intr_handler(int irq, void *pf_irq)
177 {
178 	struct otx2_nic *pf = (struct otx2_nic *)pf_irq;
179 	int vf, reg, num_reg = 1;
180 	u64 intr;
181 
182 	if (pf->total_vfs > 64)
183 		num_reg = 2;
184 
185 	for (reg = 0; reg < num_reg; reg++) {
186 		intr = otx2_read64(pf, RVU_PF_VFME_INTX(reg));
187 		if (!intr)
188 			continue;
189 		for (vf = 0; vf < 64; vf++) {
190 			if (!(intr & BIT_ULL(vf)))
191 				continue;
192 			/* clear trpend bit */
193 			otx2_write64(pf, RVU_PF_VFTRPENDX(reg), BIT_ULL(vf));
194 			/* clear interrupt */
195 			otx2_write64(pf, RVU_PF_VFME_INTX(reg), BIT_ULL(vf));
196 		}
197 	}
198 	return IRQ_HANDLED;
199 }
200 
201 static int otx2_register_flr_me_intr(struct otx2_nic *pf, int numvfs)
202 {
203 	struct otx2_hw *hw = &pf->hw;
204 	char *irq_name;
205 	int ret;
206 
207 	/* Register ME interrupt handler*/
208 	irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFME0 * NAME_SIZE];
209 	snprintf(irq_name, NAME_SIZE, "RVUPF%d_ME0",
210 		 rvu_get_pf(pf->pdev, pf->pcifunc));
211 	ret = request_irq(pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFME0),
212 			  otx2_pf_me_intr_handler, 0, irq_name, pf);
213 	if (ret) {
214 		dev_err(pf->dev,
215 			"RVUPF: IRQ registration failed for ME0\n");
216 	}
217 
218 	/* Register FLR interrupt handler */
219 	irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFFLR0 * NAME_SIZE];
220 	snprintf(irq_name, NAME_SIZE, "RVUPF%d_FLR0",
221 		 rvu_get_pf(pf->pdev, pf->pcifunc));
222 	ret = request_irq(pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFFLR0),
223 			  otx2_pf_flr_intr_handler, 0, irq_name, pf);
224 	if (ret) {
225 		dev_err(pf->dev,
226 			"RVUPF: IRQ registration failed for FLR0\n");
227 		return ret;
228 	}
229 
230 	if (numvfs > 64) {
231 		irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFME1 * NAME_SIZE];
232 		snprintf(irq_name, NAME_SIZE, "RVUPF%d_ME1",
233 			 rvu_get_pf(pf->pdev, pf->pcifunc));
234 		ret = request_irq(pci_irq_vector
235 				  (pf->pdev, RVU_PF_INT_VEC_VFME1),
236 				  otx2_pf_me_intr_handler, 0, irq_name, pf);
237 		if (ret) {
238 			dev_err(pf->dev,
239 				"RVUPF: IRQ registration failed for ME1\n");
240 		}
241 		irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFFLR1 * NAME_SIZE];
242 		snprintf(irq_name, NAME_SIZE, "RVUPF%d_FLR1",
243 			 rvu_get_pf(pf->pdev, pf->pcifunc));
244 		ret = request_irq(pci_irq_vector
245 				  (pf->pdev, RVU_PF_INT_VEC_VFFLR1),
246 				  otx2_pf_flr_intr_handler, 0, irq_name, pf);
247 		if (ret) {
248 			dev_err(pf->dev,
249 				"RVUPF: IRQ registration failed for FLR1\n");
250 			return ret;
251 		}
252 	}
253 
254 	/* Enable ME interrupt for all VFs*/
255 	otx2_write64(pf, RVU_PF_VFME_INTX(0), INTR_MASK(numvfs));
256 	otx2_write64(pf, RVU_PF_VFME_INT_ENA_W1SX(0), INTR_MASK(numvfs));
257 
258 	/* Enable FLR interrupt for all VFs*/
259 	otx2_write64(pf, RVU_PF_VFFLR_INTX(0), INTR_MASK(numvfs));
260 	otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1SX(0), INTR_MASK(numvfs));
261 
262 	if (numvfs > 64) {
263 		numvfs -= 64;
264 
265 		otx2_write64(pf, RVU_PF_VFME_INTX(1), INTR_MASK(numvfs));
266 		otx2_write64(pf, RVU_PF_VFME_INT_ENA_W1SX(1),
267 			     INTR_MASK(numvfs));
268 
269 		otx2_write64(pf, RVU_PF_VFFLR_INTX(1), INTR_MASK(numvfs));
270 		otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1SX(1),
271 			     INTR_MASK(numvfs));
272 	}
273 	return 0;
274 }
275 
276 static int otx2_pf_flr_init(struct otx2_nic *pf, int num_vfs)
277 {
278 	int vf;
279 
280 	pf->flr_wq = alloc_ordered_workqueue("otx2_pf_flr_wq", WQ_HIGHPRI);
281 	if (!pf->flr_wq)
282 		return -ENOMEM;
283 
284 	pf->flr_wrk = devm_kcalloc(pf->dev, num_vfs,
285 				   sizeof(struct flr_work), GFP_KERNEL);
286 	if (!pf->flr_wrk) {
287 		destroy_workqueue(pf->flr_wq);
288 		return -ENOMEM;
289 	}
290 
291 	for (vf = 0; vf < num_vfs; vf++) {
292 		pf->flr_wrk[vf].pf = pf;
293 		INIT_WORK(&pf->flr_wrk[vf].work, otx2_flr_handler);
294 	}
295 
296 	return 0;
297 }
298 
299 void otx2_queue_vf_work(struct mbox *mw, struct workqueue_struct *mbox_wq,
300 			int first, int mdevs, u64 intr)
301 {
302 	struct otx2_mbox_dev *mdev;
303 	struct otx2_mbox *mbox;
304 	struct mbox_hdr *hdr;
305 	int i;
306 
307 	for (i = first; i < mdevs; i++) {
308 		/* start from 0 */
309 		if (!(intr & BIT_ULL(i - first)))
310 			continue;
311 
312 		mbox = &mw->mbox;
313 		mdev = &mbox->dev[i];
314 		hdr = mdev->mbase + mbox->rx_start;
315 		/* The hdr->num_msgs is set to zero immediately in the interrupt
316 		 * handler to ensure that it holds a correct value next time
317 		 * when the interrupt handler is called. pf->mw[i].num_msgs
318 		 * holds the data for use in otx2_pfvf_mbox_handler and
319 		 * pf->mw[i].up_num_msgs holds the data for use in
320 		 * otx2_pfvf_mbox_up_handler.
321 		 */
322 		if (hdr->num_msgs) {
323 			mw[i].num_msgs = hdr->num_msgs;
324 			hdr->num_msgs = 0;
325 			queue_work(mbox_wq, &mw[i].mbox_wrk);
326 		}
327 
328 		mbox = &mw->mbox_up;
329 		mdev = &mbox->dev[i];
330 		hdr = mdev->mbase + mbox->rx_start;
331 		if (hdr->num_msgs) {
332 			mw[i].up_num_msgs = hdr->num_msgs;
333 			hdr->num_msgs = 0;
334 			queue_work(mbox_wq, &mw[i].mbox_up_wrk);
335 		}
336 	}
337 }
338 
339 static void otx2_forward_msg_pfvf(struct otx2_mbox_dev *mdev,
340 				  struct otx2_mbox *pfvf_mbox, void *bbuf_base,
341 				  int devid)
342 {
343 	struct otx2_mbox_dev *src_mdev = mdev;
344 	int offset;
345 
346 	/* Msgs are already copied, trigger VF's mbox irq */
347 	smp_wmb();
348 
349 	otx2_mbox_wait_for_zero(pfvf_mbox, devid);
350 
351 	offset = pfvf_mbox->trigger | (devid << pfvf_mbox->tr_shift);
352 	writeq(MBOX_DOWN_MSG, (void __iomem *)pfvf_mbox->reg_base + offset);
353 
354 	/* Restore VF's mbox bounce buffer region address */
355 	src_mdev->mbase = bbuf_base;
356 }
357 
358 static int otx2_forward_vf_mbox_msgs(struct otx2_nic *pf,
359 				     struct otx2_mbox *src_mbox,
360 				     int dir, int vf, int num_msgs)
361 {
362 	struct otx2_mbox_dev *src_mdev, *dst_mdev;
363 	struct mbox_hdr *mbox_hdr;
364 	struct mbox_hdr *req_hdr;
365 	struct mbox *dst_mbox;
366 	int dst_size, err;
367 
368 	if (dir == MBOX_DIR_PFAF) {
369 		/* Set VF's mailbox memory as PF's bounce buffer memory, so
370 		 * that explicit copying of VF's msgs to PF=>AF mbox region
371 		 * and AF=>PF responses to VF's mbox region can be avoided.
372 		 */
373 		src_mdev = &src_mbox->dev[vf];
374 		mbox_hdr = src_mbox->hwbase +
375 				src_mbox->rx_start + (vf * MBOX_SIZE);
376 
377 		dst_mbox = &pf->mbox;
378 		dst_size = dst_mbox->mbox.tx_size -
379 				ALIGN(sizeof(*mbox_hdr), MBOX_MSG_ALIGN);
380 		/* Check if msgs fit into destination area and has valid size */
381 		if (mbox_hdr->msg_size > dst_size || !mbox_hdr->msg_size)
382 			return -EINVAL;
383 
384 		dst_mdev = &dst_mbox->mbox.dev[0];
385 
386 		mutex_lock(&pf->mbox.lock);
387 		dst_mdev->mbase = src_mdev->mbase;
388 		dst_mdev->msg_size = mbox_hdr->msg_size;
389 		dst_mdev->num_msgs = num_msgs;
390 		err = otx2_sync_mbox_msg(dst_mbox);
391 		/* Error code -EIO indicate there is a communication failure
392 		 * to the AF. Rest of the error codes indicate that AF processed
393 		 * VF messages and set the error codes in response messages
394 		 * (if any) so simply forward responses to VF.
395 		 */
396 		if (err == -EIO) {
397 			dev_warn(pf->dev,
398 				 "AF not responding to VF%d messages\n", vf);
399 			/* restore PF mbase and exit */
400 			dst_mdev->mbase = pf->mbox.bbuf_base;
401 			mutex_unlock(&pf->mbox.lock);
402 			return err;
403 		}
404 		/* At this point, all the VF messages sent to AF are acked
405 		 * with proper responses and responses are copied to VF
406 		 * mailbox hence raise interrupt to VF.
407 		 */
408 		req_hdr = (struct mbox_hdr *)(dst_mdev->mbase +
409 					      dst_mbox->mbox.rx_start);
410 		req_hdr->num_msgs = num_msgs;
411 
412 		otx2_forward_msg_pfvf(dst_mdev, &pf->mbox_pfvf[0].mbox,
413 				      pf->mbox.bbuf_base, vf);
414 		mutex_unlock(&pf->mbox.lock);
415 	} else if (dir == MBOX_DIR_PFVF_UP) {
416 		src_mdev = &src_mbox->dev[0];
417 		mbox_hdr = src_mbox->hwbase + src_mbox->rx_start;
418 		req_hdr = (struct mbox_hdr *)(src_mdev->mbase +
419 					      src_mbox->rx_start);
420 		req_hdr->num_msgs = num_msgs;
421 
422 		dst_mbox = &pf->mbox_pfvf[0];
423 		dst_size = dst_mbox->mbox_up.tx_size -
424 				ALIGN(sizeof(*mbox_hdr), MBOX_MSG_ALIGN);
425 		/* Check if msgs fit into destination area */
426 		if (mbox_hdr->msg_size > dst_size)
427 			return -EINVAL;
428 
429 		dst_mdev = &dst_mbox->mbox_up.dev[vf];
430 		dst_mdev->mbase = src_mdev->mbase;
431 		dst_mdev->msg_size = mbox_hdr->msg_size;
432 		dst_mdev->num_msgs = mbox_hdr->num_msgs;
433 		err = otx2_sync_mbox_up_msg(dst_mbox, vf);
434 		if (err) {
435 			dev_warn(pf->dev,
436 				 "VF%d is not responding to mailbox\n", vf);
437 			return err;
438 		}
439 	} else if (dir == MBOX_DIR_VFPF_UP) {
440 		req_hdr = (struct mbox_hdr *)(src_mbox->dev[0].mbase +
441 					      src_mbox->rx_start);
442 		req_hdr->num_msgs = num_msgs;
443 		otx2_forward_msg_pfvf(&pf->mbox_pfvf->mbox_up.dev[vf],
444 				      &pf->mbox.mbox_up,
445 				      pf->mbox_pfvf[vf].bbuf_base,
446 				      0);
447 	}
448 
449 	return 0;
450 }
451 
452 static void otx2_pfvf_mbox_handler(struct work_struct *work)
453 {
454 	struct mbox_msghdr *msg = NULL;
455 	int offset, vf_idx, id, err;
456 	struct otx2_mbox_dev *mdev;
457 	struct otx2_mbox *mbox;
458 	struct mbox *vf_mbox;
459 	struct otx2_nic *pf;
460 
461 	vf_mbox = container_of(work, struct mbox, mbox_wrk);
462 	pf = vf_mbox->pfvf;
463 	vf_idx = vf_mbox - pf->mbox_pfvf;
464 
465 	mbox = &pf->mbox_pfvf[0].mbox;
466 	mdev = &mbox->dev[vf_idx];
467 
468 	offset = ALIGN(sizeof(struct mbox_hdr), MBOX_MSG_ALIGN);
469 
470 	trace_otx2_msg_status(pf->pdev, "PF-VF down queue handler(forwarding)",
471 			      vf_mbox->num_msgs);
472 
473 	for (id = 0; id < vf_mbox->num_msgs; id++) {
474 		msg = (struct mbox_msghdr *)(mdev->mbase + mbox->rx_start +
475 					     offset);
476 
477 		if (msg->sig != OTX2_MBOX_REQ_SIG)
478 			goto inval_msg;
479 
480 		/* Set VF's number in each of the msg */
481 		msg->pcifunc &= ~RVU_PFVF_FUNC_MASK;
482 		msg->pcifunc |= (vf_idx + 1) & RVU_PFVF_FUNC_MASK;
483 		offset = msg->next_msgoff;
484 	}
485 	err = otx2_forward_vf_mbox_msgs(pf, mbox, MBOX_DIR_PFAF, vf_idx,
486 					vf_mbox->num_msgs);
487 	if (err)
488 		goto inval_msg;
489 	return;
490 
491 inval_msg:
492 	otx2_reply_invalid_msg(mbox, vf_idx, 0, msg->id);
493 	otx2_mbox_msg_send(mbox, vf_idx);
494 }
495 
496 static void otx2_pfvf_mbox_up_handler(struct work_struct *work)
497 {
498 	struct mbox *vf_mbox = container_of(work, struct mbox, mbox_up_wrk);
499 	struct otx2_nic *pf = vf_mbox->pfvf;
500 	struct otx2_mbox_dev *mdev;
501 	int offset, id, vf_idx = 0;
502 	struct mbox_msghdr *msg;
503 	struct otx2_mbox *mbox;
504 
505 	vf_idx = vf_mbox - pf->mbox_pfvf;
506 	mbox = &pf->mbox_pfvf[0].mbox_up;
507 	mdev = &mbox->dev[vf_idx];
508 
509 	offset = mbox->rx_start + ALIGN(sizeof(struct mbox_hdr), MBOX_MSG_ALIGN);
510 
511 	trace_otx2_msg_status(pf->pdev, "PF-VF up queue handler(response)",
512 			      vf_mbox->up_num_msgs);
513 
514 	for (id = 0; id < vf_mbox->up_num_msgs; id++) {
515 		msg = mdev->mbase + offset;
516 
517 		if (msg->id >= MBOX_MSG_MAX) {
518 			dev_err(pf->dev,
519 				"Mbox msg with unknown ID 0x%x\n", msg->id);
520 			goto end;
521 		}
522 
523 		if (msg->sig != OTX2_MBOX_RSP_SIG) {
524 			dev_err(pf->dev,
525 				"Mbox msg with wrong signature %x, ID 0x%x\n",
526 				msg->sig, msg->id);
527 			goto end;
528 		}
529 
530 		switch (msg->id) {
531 		case MBOX_MSG_CGX_LINK_EVENT:
532 		case MBOX_MSG_REP_EVENT_UP_NOTIFY:
533 			break;
534 		default:
535 			if (msg->rc)
536 				dev_err(pf->dev,
537 					"Mbox msg response has err %d, ID 0x%x\n",
538 					msg->rc, msg->id);
539 			break;
540 		}
541 
542 end:
543 		offset = mbox->rx_start + msg->next_msgoff;
544 		if (mdev->msgs_acked == (vf_mbox->up_num_msgs - 1))
545 			__otx2_mbox_reset(mbox, vf_idx);
546 		mdev->msgs_acked++;
547 	}
548 }
549 
550 irqreturn_t otx2_pfvf_mbox_intr_handler(int irq, void *pf_irq)
551 {
552 	struct otx2_nic *pf = (struct otx2_nic *)(pf_irq);
553 	int vfs = pf->total_vfs;
554 	struct mbox *mbox;
555 	u64 intr;
556 
557 	mbox = pf->mbox_pfvf;
558 	/* Handle VF interrupts */
559 	if (vfs > 64) {
560 		intr = otx2_read64(pf, RVU_PF_VFPF_MBOX_INTX(1));
561 		otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(1), intr);
562 		otx2_queue_vf_work(mbox, pf->mbox_pfvf_wq, 64, vfs, intr);
563 		if (intr)
564 			trace_otx2_msg_interrupt(mbox->mbox.pdev, "VF(s) to PF", intr);
565 		vfs = 64;
566 	}
567 
568 	intr = otx2_read64(pf, RVU_PF_VFPF_MBOX_INTX(0));
569 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(0), intr);
570 
571 	otx2_queue_vf_work(mbox, pf->mbox_pfvf_wq, 0, vfs, intr);
572 
573 	if (intr)
574 		trace_otx2_msg_interrupt(mbox->mbox.pdev, "VF(s) to PF", intr);
575 
576 	return IRQ_HANDLED;
577 }
578 
579 static void *cn20k_pfvf_mbox_alloc(struct otx2_nic *pf, int numvfs)
580 {
581 	struct qmem *mbox_addr;
582 	int err;
583 
584 	err = qmem_alloc(&pf->pdev->dev, &mbox_addr, numvfs, MBOX_SIZE);
585 	if (err) {
586 		dev_err(pf->dev, "qmem alloc fail\n");
587 		return ERR_PTR(-ENOMEM);
588 	}
589 
590 	otx2_write64(pf, RVU_PF_VF_MBOX_ADDR, (u64)mbox_addr->iova);
591 	pf->pfvf_mbox_addr = mbox_addr;
592 
593 	return mbox_addr->base;
594 }
595 
596 static int otx2_pfvf_mbox_init(struct otx2_nic *pf, int numvfs)
597 {
598 	void __iomem *hwbase;
599 	struct mbox *mbox;
600 	int err, vf;
601 	u64 base;
602 
603 	if (!numvfs)
604 		return -EINVAL;
605 
606 	pf->mbox_pfvf = devm_kcalloc(&pf->pdev->dev, numvfs,
607 				     sizeof(struct mbox), GFP_KERNEL);
608 	if (!pf->mbox_pfvf)
609 		return -ENOMEM;
610 
611 	pf->mbox_pfvf_wq = alloc_workqueue("otx2_pfvf_mailbox",
612 					   WQ_UNBOUND | WQ_HIGHPRI |
613 					   WQ_MEM_RECLAIM, 0);
614 	if (!pf->mbox_pfvf_wq)
615 		return -ENOMEM;
616 
617 	/* For CN20K, PF allocates mbox memory in DRAM and writes PF/VF
618 	 * regions/offsets in RVU_PF_VF_MBOX_ADDR, the RVU_PFX_FUNC_PFAF_MBOX
619 	 * gives the aliased address to access PF/VF mailbox regions.
620 	 */
621 	if (is_cn20k(pf->pdev)) {
622 		hwbase = (void __iomem *)cn20k_pfvf_mbox_alloc(pf, numvfs);
623 	} else {
624 		/* On CN10K platform, PF <-> VF mailbox region follows after
625 		 * PF <-> AF mailbox region.
626 		 */
627 		if (test_bit(CN10K_MBOX, &pf->hw.cap_flag))
628 			base = pci_resource_start(pf->pdev, PCI_MBOX_BAR_NUM) +
629 						  MBOX_SIZE;
630 		else
631 			base = readq(pf->reg_base + RVU_PF_VF_BAR4_ADDR);
632 
633 		hwbase = ioremap_wc(base, MBOX_SIZE * pf->total_vfs);
634 		if (!hwbase) {
635 			err = -ENOMEM;
636 			goto free_wq;
637 		}
638 	}
639 
640 	mbox = &pf->mbox_pfvf[0];
641 	err = otx2_mbox_init(&mbox->mbox, hwbase, pf->pdev, pf->reg_base,
642 			     MBOX_DIR_PFVF, numvfs);
643 	if (err)
644 		goto free_iomem;
645 
646 	err = otx2_mbox_init(&mbox->mbox_up, hwbase, pf->pdev, pf->reg_base,
647 			     MBOX_DIR_PFVF_UP, numvfs);
648 	if (err)
649 		goto free_iomem;
650 
651 	for (vf = 0; vf < numvfs; vf++) {
652 		mbox->pfvf = pf;
653 		INIT_WORK(&mbox->mbox_wrk, otx2_pfvf_mbox_handler);
654 		INIT_WORK(&mbox->mbox_up_wrk, otx2_pfvf_mbox_up_handler);
655 		mbox++;
656 	}
657 
658 	return 0;
659 
660 free_iomem:
661 	if (hwbase && !(is_cn20k(pf->pdev)))
662 		iounmap(hwbase);
663 free_wq:
664 	destroy_workqueue(pf->mbox_pfvf_wq);
665 	return err;
666 }
667 
668 static void otx2_pfvf_mbox_destroy(struct otx2_nic *pf)
669 {
670 	struct mbox *mbox = &pf->mbox_pfvf[0];
671 
672 	if (!mbox)
673 		return;
674 
675 	if (pf->mbox_pfvf_wq) {
676 		destroy_workqueue(pf->mbox_pfvf_wq);
677 		pf->mbox_pfvf_wq = NULL;
678 	}
679 
680 	if (mbox->mbox.hwbase && !is_cn20k(pf->pdev))
681 		iounmap(mbox->mbox.hwbase);
682 	else
683 		qmem_free(&pf->pdev->dev, pf->pfvf_mbox_addr);
684 
685 	otx2_mbox_destroy(&mbox->mbox);
686 }
687 
688 static void otx2_enable_pfvf_mbox_intr(struct otx2_nic *pf, int numvfs)
689 {
690 	/* Clear PF <=> VF mailbox IRQ */
691 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(0), ~0ull);
692 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(1), ~0ull);
693 
694 	/* Enable PF <=> VF mailbox IRQ */
695 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INT_ENA_W1SX(0), INTR_MASK(numvfs));
696 	if (numvfs > 64) {
697 		numvfs -= 64;
698 		otx2_write64(pf, RVU_PF_VFPF_MBOX_INT_ENA_W1SX(1),
699 			     INTR_MASK(numvfs));
700 	}
701 }
702 
703 static void otx2_disable_pfvf_mbox_intr(struct otx2_nic *pf, int numvfs)
704 {
705 	int vector;
706 
707 	if (is_cn20k(pf->pdev))
708 		return cn20k_disable_pfvf_mbox_intr(pf, numvfs);
709 
710 	/* Disable PF <=> VF mailbox IRQ */
711 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INT_ENA_W1CX(0), ~0ull);
712 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INT_ENA_W1CX(1), ~0ull);
713 
714 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(0), ~0ull);
715 	vector = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFPF_MBOX0);
716 	free_irq(vector, pf);
717 
718 	if (numvfs > 64) {
719 		otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(1), ~0ull);
720 		vector = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFPF_MBOX1);
721 		free_irq(vector, pf);
722 	}
723 }
724 
725 static int otx2_register_pfvf_mbox_intr(struct otx2_nic *pf, int numvfs)
726 {
727 	struct otx2_hw *hw = &pf->hw;
728 	char *irq_name;
729 	int err;
730 
731 	if (is_cn20k(pf->pdev))
732 		return cn20k_register_pfvf_mbox_intr(pf, numvfs);
733 
734 	/* Register MBOX0 interrupt handler */
735 	irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFPF_MBOX0 * NAME_SIZE];
736 	if (pf->pcifunc)
737 		snprintf(irq_name, NAME_SIZE,
738 			 "RVUPF%d_VF Mbox0", rvu_get_pf(pf->pdev, pf->pcifunc));
739 	else
740 		snprintf(irq_name, NAME_SIZE, "RVUPF_VF Mbox0");
741 	err = request_irq(pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFPF_MBOX0),
742 			  otx2_pfvf_mbox_intr_handler, 0, irq_name, pf);
743 	if (err) {
744 		dev_err(pf->dev,
745 			"RVUPF: IRQ registration failed for PFVF mbox0 irq\n");
746 		return err;
747 	}
748 
749 	if (numvfs > 64) {
750 		/* Register MBOX1 interrupt handler */
751 		irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFPF_MBOX1 * NAME_SIZE];
752 		if (pf->pcifunc)
753 			snprintf(irq_name, NAME_SIZE,
754 				 "RVUPF%d_VF Mbox1",
755 				 rvu_get_pf(pf->pdev, pf->pcifunc));
756 		else
757 			snprintf(irq_name, NAME_SIZE, "RVUPF_VF Mbox1");
758 		err = request_irq(pci_irq_vector(pf->pdev,
759 						 RVU_PF_INT_VEC_VFPF_MBOX1),
760 						 otx2_pfvf_mbox_intr_handler,
761 						 0, irq_name, pf);
762 		if (err) {
763 			dev_err(pf->dev,
764 				"RVUPF: IRQ registration failed for PFVF mbox1 irq\n");
765 			return err;
766 		}
767 	}
768 
769 	otx2_enable_pfvf_mbox_intr(pf, numvfs);
770 
771 	return 0;
772 }
773 
774 static void otx2_process_pfaf_mbox_msg(struct otx2_nic *pf,
775 				       struct mbox_msghdr *msg)
776 {
777 	int devid;
778 
779 	if (msg->id >= MBOX_MSG_MAX) {
780 		dev_err(pf->dev,
781 			"Mbox msg with unknown ID 0x%x\n", msg->id);
782 		return;
783 	}
784 
785 	if (msg->sig != OTX2_MBOX_RSP_SIG) {
786 		dev_err(pf->dev,
787 			"Mbox msg with wrong signature %x, ID 0x%x\n",
788 			 msg->sig, msg->id);
789 		return;
790 	}
791 
792 	/* message response heading VF */
793 	devid = msg->pcifunc & RVU_PFVF_FUNC_MASK;
794 	if (devid) {
795 		struct otx2_vf_config *config = &pf->vf_configs[devid - 1];
796 		struct delayed_work *dwork;
797 
798 		switch (msg->id) {
799 		case MBOX_MSG_NIX_LF_START_RX:
800 			config->intf_down = false;
801 			dwork = &config->link_event_work;
802 			schedule_delayed_work(dwork, msecs_to_jiffies(100));
803 			break;
804 		case MBOX_MSG_NIX_LF_STOP_RX:
805 			config->intf_down = true;
806 			break;
807 		}
808 
809 		return;
810 	}
811 
812 	switch (msg->id) {
813 	case MBOX_MSG_READY:
814 		pf->pcifunc = msg->pcifunc;
815 		break;
816 	case MBOX_MSG_MSIX_OFFSET:
817 		mbox_handler_msix_offset(pf, (struct msix_offset_rsp *)msg);
818 		break;
819 	case MBOX_MSG_NPA_LF_ALLOC:
820 		mbox_handler_npa_lf_alloc(pf, (struct npa_lf_alloc_rsp *)msg);
821 		break;
822 	case MBOX_MSG_NIX_LF_ALLOC:
823 		mbox_handler_nix_lf_alloc(pf, (struct nix_lf_alloc_rsp *)msg);
824 		break;
825 	case MBOX_MSG_NIX_BP_ENABLE:
826 		mbox_handler_nix_bp_enable(pf, (struct nix_bp_cfg_rsp *)msg);
827 		break;
828 	case MBOX_MSG_CGX_STATS:
829 		mbox_handler_cgx_stats(pf, (struct cgx_stats_rsp *)msg);
830 		break;
831 	case MBOX_MSG_CGX_FEC_STATS:
832 		mbox_handler_cgx_fec_stats(pf, (struct cgx_fec_stats_rsp *)msg);
833 		break;
834 	default:
835 		if (msg->rc)
836 			dev_err(pf->dev,
837 				"Mbox msg response has err %d, ID 0x%x\n",
838 				msg->rc, msg->id);
839 		break;
840 	}
841 }
842 
843 static void otx2_pfaf_mbox_handler(struct work_struct *work)
844 {
845 	struct otx2_mbox_dev *mdev;
846 	struct mbox_hdr *rsp_hdr;
847 	struct mbox_msghdr *msg;
848 	struct otx2_mbox *mbox;
849 	struct mbox *af_mbox;
850 	struct otx2_nic *pf;
851 	int offset, id;
852 	u16 num_msgs;
853 
854 	af_mbox = container_of(work, struct mbox, mbox_wrk);
855 	mbox = &af_mbox->mbox;
856 	mdev = &mbox->dev[0];
857 	rsp_hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
858 	num_msgs = rsp_hdr->num_msgs;
859 
860 	offset = mbox->rx_start + ALIGN(sizeof(*rsp_hdr), MBOX_MSG_ALIGN);
861 	pf = af_mbox->pfvf;
862 
863 	trace_otx2_msg_status(pf->pdev, "PF-AF down queue handler(response)",
864 			      num_msgs);
865 
866 	for (id = 0; id < num_msgs; id++) {
867 		msg = (struct mbox_msghdr *)(mdev->mbase + offset);
868 		otx2_process_pfaf_mbox_msg(pf, msg);
869 		offset = mbox->rx_start + msg->next_msgoff;
870 		if (mdev->msgs_acked == (num_msgs - 1))
871 			__otx2_mbox_reset(mbox, 0);
872 		mdev->msgs_acked++;
873 	}
874 
875 }
876 
877 static void otx2_handle_link_event(struct otx2_nic *pf)
878 {
879 	struct cgx_link_user_info *linfo = &pf->linfo;
880 	struct net_device *netdev = pf->netdev;
881 
882 	if (pf->flags & OTX2_FLAG_PORT_UP)
883 		return;
884 
885 	pr_info("%s NIC Link is %s %d Mbps %s duplex\n", netdev->name,
886 		linfo->link_up ? "UP" : "DOWN", linfo->speed,
887 		linfo->full_duplex ? "Full" : "Half");
888 	if (linfo->link_up) {
889 		netif_carrier_on(netdev);
890 		netif_tx_start_all_queues(netdev);
891 	} else {
892 		netif_tx_stop_all_queues(netdev);
893 		netif_carrier_off(netdev);
894 	}
895 }
896 
897 static int otx2_mbox_up_handler_rep_event_up_notify(struct otx2_nic *pf,
898 						    struct rep_event *info,
899 						    struct msg_rsp *rsp)
900 {
901 	struct net_device *netdev = pf->netdev;
902 
903 	if (info->event == RVU_EVENT_MTU_CHANGE) {
904 		netdev->mtu = info->evt_data.mtu;
905 		return 0;
906 	}
907 
908 	if (info->event == RVU_EVENT_PORT_STATE) {
909 		if (info->evt_data.port_state) {
910 			pf->flags |= OTX2_FLAG_PORT_UP;
911 			netif_carrier_on(netdev);
912 			netif_tx_start_all_queues(netdev);
913 		} else {
914 			pf->flags &= ~OTX2_FLAG_PORT_UP;
915 			netif_tx_stop_all_queues(netdev);
916 			netif_carrier_off(netdev);
917 		}
918 		return 0;
919 	}
920 #ifdef CONFIG_RVU_ESWITCH
921 	rvu_event_up_notify(pf, info);
922 #endif
923 	return 0;
924 }
925 
926 int otx2_mbox_up_handler_mcs_intr_notify(struct otx2_nic *pf,
927 					 struct mcs_intr_info *event,
928 					 struct msg_rsp *rsp)
929 {
930 	cn10k_handle_mcs_event(pf, event);
931 
932 	return 0;
933 }
934 
935 int otx2_mbox_up_handler_cgx_link_event(struct otx2_nic *pf,
936 					struct cgx_link_info_msg *msg,
937 					struct msg_rsp *rsp)
938 {
939 	int i;
940 
941 	/* Copy the link info sent by AF */
942 	pf->linfo = msg->link_info;
943 
944 	/* notify VFs about link event */
945 	for (i = 0; i < pci_num_vf(pf->pdev); i++) {
946 		struct otx2_vf_config *config = &pf->vf_configs[i];
947 		struct delayed_work *dwork = &config->link_event_work;
948 
949 		if (config->intf_down)
950 			continue;
951 
952 		schedule_delayed_work(dwork, msecs_to_jiffies(100));
953 	}
954 
955 	/* interface has not been fully configured yet */
956 	if (pf->flags & OTX2_FLAG_INTF_DOWN)
957 		return 0;
958 
959 	otx2_handle_link_event(pf);
960 	return 0;
961 }
962 
963 static int otx2_process_mbox_msg_up(struct otx2_nic *pf,
964 				    struct mbox_msghdr *req)
965 {
966 	/* Check if valid, if not reply with a invalid msg */
967 	if (req->sig != OTX2_MBOX_REQ_SIG) {
968 		otx2_reply_invalid_msg(&pf->mbox.mbox_up, 0, 0, req->id);
969 		return -ENODEV;
970 	}
971 
972 	switch (req->id) {
973 #define M(_name, _id, _fn_name, _req_type, _rsp_type)			\
974 	case _id: {							\
975 		struct _rsp_type *rsp;					\
976 		int err;						\
977 									\
978 		rsp = (struct _rsp_type *)otx2_mbox_alloc_msg(		\
979 			&pf->mbox.mbox_up, 0,				\
980 			sizeof(struct _rsp_type));			\
981 		if (!rsp)						\
982 			return -ENOMEM;					\
983 									\
984 		rsp->hdr.id = _id;					\
985 		rsp->hdr.sig = OTX2_MBOX_RSP_SIG;			\
986 		rsp->hdr.pcifunc = 0;					\
987 		rsp->hdr.rc = 0;					\
988 									\
989 		err = otx2_mbox_up_handler_ ## _fn_name(		\
990 			pf, (struct _req_type *)req, rsp);		\
991 		return err;						\
992 	}
993 MBOX_UP_CGX_MESSAGES
994 MBOX_UP_MCS_MESSAGES
995 MBOX_UP_REP_MESSAGES
996 #undef M
997 		break;
998 	default:
999 		otx2_reply_invalid_msg(&pf->mbox.mbox_up, 0, 0, req->id);
1000 		return -ENODEV;
1001 	}
1002 	return 0;
1003 }
1004 
1005 static void otx2_pfaf_mbox_up_handler(struct work_struct *work)
1006 {
1007 	struct mbox *af_mbox = container_of(work, struct mbox, mbox_up_wrk);
1008 	struct otx2_mbox *mbox = &af_mbox->mbox_up;
1009 	struct otx2_mbox_dev *mdev = &mbox->dev[0];
1010 	struct otx2_nic *pf = af_mbox->pfvf;
1011 	int offset, id, devid = 0;
1012 	struct mbox_hdr *rsp_hdr;
1013 	struct mbox_msghdr *msg;
1014 	u16 num_msgs;
1015 
1016 	rsp_hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
1017 	num_msgs = rsp_hdr->num_msgs;
1018 
1019 	offset = mbox->rx_start + ALIGN(sizeof(*rsp_hdr), MBOX_MSG_ALIGN);
1020 
1021 	trace_otx2_msg_status(pf->pdev, "PF-AF up queue handler(notification)",
1022 			      num_msgs);
1023 
1024 	for (id = 0; id < num_msgs; id++) {
1025 		msg = (struct mbox_msghdr *)(mdev->mbase + offset);
1026 
1027 		devid = msg->pcifunc & RVU_PFVF_FUNC_MASK;
1028 		/* Skip processing VF's messages */
1029 		if (!devid)
1030 			otx2_process_mbox_msg_up(pf, msg);
1031 		offset = mbox->rx_start + msg->next_msgoff;
1032 	}
1033 	/* Forward to VF iff VFs are really present */
1034 	if (devid && pci_num_vf(pf->pdev)) {
1035 		otx2_forward_vf_mbox_msgs(pf, &pf->mbox.mbox_up,
1036 					  MBOX_DIR_PFVF_UP, devid - 1,
1037 					  num_msgs);
1038 		return;
1039 	}
1040 
1041 	otx2_mbox_msg_send(mbox, 0);
1042 }
1043 
1044 irqreturn_t otx2_pfaf_mbox_intr_handler(int irq, void *pf_irq)
1045 {
1046 	struct otx2_nic *pf = (struct otx2_nic *)pf_irq;
1047 	struct mbox *mw = &pf->mbox;
1048 	struct otx2_mbox_dev *mdev;
1049 	struct otx2_mbox *mbox;
1050 	struct mbox_hdr *hdr;
1051 	u64 mbox_data;
1052 
1053 	/* Clear the IRQ */
1054 	otx2_write64(pf, RVU_PF_INT, BIT_ULL(0));
1055 
1056 
1057 	mbox_data = otx2_read64(pf, RVU_PF_PFAF_MBOX0);
1058 
1059 	if (mbox_data & MBOX_UP_MSG) {
1060 		mbox_data &= ~MBOX_UP_MSG;
1061 		otx2_write64(pf, RVU_PF_PFAF_MBOX0, mbox_data);
1062 
1063 		mbox = &mw->mbox_up;
1064 		mdev = &mbox->dev[0];
1065 		otx2_sync_mbox_bbuf(mbox, 0);
1066 
1067 		hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
1068 		if (hdr->num_msgs)
1069 			queue_work(pf->mbox_wq, &mw->mbox_up_wrk);
1070 
1071 		trace_otx2_msg_interrupt(pf->pdev, "UP message from AF to PF",
1072 					 BIT_ULL(0));
1073 
1074 		trace_otx2_msg_status(pf->pdev, "PF-AF up work queued(interrupt)",
1075 				      hdr->num_msgs);
1076 	}
1077 
1078 	if (mbox_data & MBOX_DOWN_MSG) {
1079 		mbox_data &= ~MBOX_DOWN_MSG;
1080 		otx2_write64(pf, RVU_PF_PFAF_MBOX0, mbox_data);
1081 
1082 		mbox = &mw->mbox;
1083 		mdev = &mbox->dev[0];
1084 		otx2_sync_mbox_bbuf(mbox, 0);
1085 
1086 		hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
1087 		if (hdr->num_msgs)
1088 			queue_work(pf->mbox_wq, &mw->mbox_wrk);
1089 
1090 		trace_otx2_msg_interrupt(pf->pdev, "DOWN reply from AF to PF",
1091 					 BIT_ULL(0));
1092 
1093 		trace_otx2_msg_status(pf->pdev, "PF-AF down work queued(interrupt)",
1094 				      hdr->num_msgs);
1095 	}
1096 
1097 	return IRQ_HANDLED;
1098 }
1099 
1100 void otx2_disable_mbox_intr(struct otx2_nic *pf)
1101 {
1102 	int vector;
1103 
1104 	/* Disable AF => PF mailbox IRQ */
1105 	if (!is_cn20k(pf->pdev)) {
1106 		vector = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_AFPF_MBOX);
1107 		otx2_write64(pf, RVU_PF_INT_ENA_W1C, BIT_ULL(0));
1108 	} else {
1109 		vector = pci_irq_vector(pf->pdev,
1110 					RVU_MBOX_PF_INT_VEC_AFPF_MBOX);
1111 		otx2_write64(pf, RVU_PF_INT_ENA_W1C,
1112 			     BIT_ULL(0) | BIT_ULL(1));
1113 	}
1114 	free_irq(vector, pf);
1115 }
1116 EXPORT_SYMBOL(otx2_disable_mbox_intr);
1117 
1118 int otx2_register_mbox_intr(struct otx2_nic *pf, bool probe_af)
1119 {
1120 	struct otx2_hw *hw = &pf->hw;
1121 	struct msg_req *req;
1122 	char *irq_name;
1123 	int err;
1124 
1125 	/* Register mailbox interrupt handler */
1126 	if (!is_cn20k(pf->pdev)) {
1127 		irq_name = &hw->irq_name[RVU_PF_INT_VEC_AFPF_MBOX * NAME_SIZE];
1128 		snprintf(irq_name, NAME_SIZE, "RVUPF%d AFPF Mbox",
1129 			 rvu_get_pf(pf->pdev, pf->pcifunc));
1130 		err = request_irq(pci_irq_vector
1131 				  (pf->pdev, RVU_PF_INT_VEC_AFPF_MBOX),
1132 				  pf->hw_ops->pfaf_mbox_intr_handler,
1133 				  0, irq_name, pf);
1134 	} else {
1135 		irq_name = &hw->irq_name[RVU_MBOX_PF_INT_VEC_AFPF_MBOX *
1136 						NAME_SIZE];
1137 		snprintf(irq_name, NAME_SIZE, "RVUPF%d AFPF Mbox",
1138 			 rvu_get_pf(pf->pdev, pf->pcifunc));
1139 		err = request_irq(pci_irq_vector
1140 				  (pf->pdev, RVU_MBOX_PF_INT_VEC_AFPF_MBOX),
1141 				  pf->hw_ops->pfaf_mbox_intr_handler,
1142 				  0, irq_name, pf);
1143 	}
1144 	if (err) {
1145 		dev_err(pf->dev,
1146 			"RVUPF: IRQ registration failed for PFAF mbox irq\n");
1147 		return err;
1148 	}
1149 
1150 	/* Enable mailbox interrupt for msgs coming from AF.
1151 	 * First clear to avoid spurious interrupts, if any.
1152 	 */
1153 	if (!is_cn20k(pf->pdev)) {
1154 		otx2_write64(pf, RVU_PF_INT, BIT_ULL(0));
1155 		otx2_write64(pf, RVU_PF_INT_ENA_W1S, BIT_ULL(0));
1156 	} else {
1157 		otx2_write64(pf, RVU_PF_INT, BIT_ULL(0) | BIT_ULL(1));
1158 		otx2_write64(pf, RVU_PF_INT_ENA_W1S, BIT_ULL(0) |
1159 			     BIT_ULL(1));
1160 	}
1161 
1162 	if (!probe_af)
1163 		return 0;
1164 
1165 	/* Check mailbox communication with AF */
1166 	req = otx2_mbox_alloc_msg_ready(&pf->mbox);
1167 	if (!req) {
1168 		otx2_disable_mbox_intr(pf);
1169 		return -ENOMEM;
1170 	}
1171 	err = otx2_sync_mbox_msg(&pf->mbox);
1172 	if (err) {
1173 		dev_warn(pf->dev,
1174 			 "AF not responding to mailbox, deferring probe\n");
1175 		otx2_disable_mbox_intr(pf);
1176 		return -EPROBE_DEFER;
1177 	}
1178 
1179 	return 0;
1180 }
1181 
1182 void otx2_pfaf_mbox_destroy(struct otx2_nic *pf)
1183 {
1184 	struct mbox *mbox = &pf->mbox;
1185 
1186 	if (pf->mbox_wq) {
1187 		destroy_workqueue(pf->mbox_wq);
1188 		pf->mbox_wq = NULL;
1189 	}
1190 
1191 	if (mbox->mbox.hwbase && !is_cn20k(pf->pdev))
1192 		iounmap((void __iomem *)mbox->mbox.hwbase);
1193 
1194 	otx2_mbox_destroy(&mbox->mbox);
1195 	otx2_mbox_destroy(&mbox->mbox_up);
1196 }
1197 EXPORT_SYMBOL(otx2_pfaf_mbox_destroy);
1198 
1199 int otx2_pfaf_mbox_init(struct otx2_nic *pf)
1200 {
1201 	struct mbox *mbox = &pf->mbox;
1202 	void __iomem *hwbase;
1203 	int err;
1204 
1205 	mbox->pfvf = pf;
1206 	pf->mbox_wq = alloc_ordered_workqueue("otx2_pfaf_mailbox",
1207 					      WQ_HIGHPRI | WQ_MEM_RECLAIM);
1208 	if (!pf->mbox_wq)
1209 		return -ENOMEM;
1210 
1211 	/* For CN20K, AF allocates mbox memory in DRAM and writes PF
1212 	 * regions/offsets in RVU_MBOX_AF_PFX_ADDR, the RVU_PFX_FUNC_PFAF_MBOX
1213 	 * gives the aliased address to access AF/PF mailbox regions.
1214 	 */
1215 	if (is_cn20k(pf->pdev))
1216 		hwbase = pf->reg_base + RVU_PFX_FUNC_PFAF_MBOX +
1217 			((u64)BLKADDR_MBOX << RVU_FUNC_BLKADDR_SHIFT);
1218 	else
1219 		/* Mailbox is a reserved memory (in RAM) region shared between
1220 		 * admin function (i.e AF) and this PF, shouldn't be mapped as
1221 		 * device memory to allow unaligned accesses.
1222 		 */
1223 		hwbase = ioremap_wc(pci_resource_start
1224 				    (pf->pdev, PCI_MBOX_BAR_NUM), MBOX_SIZE);
1225 	if (!hwbase) {
1226 		dev_err(pf->dev, "Unable to map PFAF mailbox region\n");
1227 		err = -ENOMEM;
1228 		goto exit;
1229 	}
1230 
1231 	err = otx2_mbox_init(&mbox->mbox, hwbase, pf->pdev, pf->reg_base,
1232 			     MBOX_DIR_PFAF, 1);
1233 	if (err)
1234 		goto exit;
1235 
1236 	err = otx2_mbox_init(&mbox->mbox_up, hwbase, pf->pdev, pf->reg_base,
1237 			     MBOX_DIR_PFAF_UP, 1);
1238 	if (err)
1239 		goto exit;
1240 
1241 	err = otx2_mbox_bbuf_init(mbox, pf->pdev);
1242 	if (err)
1243 		goto exit;
1244 
1245 	INIT_WORK(&mbox->mbox_wrk, otx2_pfaf_mbox_handler);
1246 	INIT_WORK(&mbox->mbox_up_wrk, otx2_pfaf_mbox_up_handler);
1247 	mutex_init(&mbox->lock);
1248 
1249 	return 0;
1250 exit:
1251 	otx2_pfaf_mbox_destroy(pf);
1252 	return err;
1253 }
1254 
1255 static int otx2_cgx_config_linkevents(struct otx2_nic *pf, bool enable)
1256 {
1257 	struct msg_req *msg;
1258 	int err;
1259 
1260 	mutex_lock(&pf->mbox.lock);
1261 	if (enable)
1262 		msg = otx2_mbox_alloc_msg_cgx_start_linkevents(&pf->mbox);
1263 	else
1264 		msg = otx2_mbox_alloc_msg_cgx_stop_linkevents(&pf->mbox);
1265 
1266 	if (!msg) {
1267 		mutex_unlock(&pf->mbox.lock);
1268 		return -ENOMEM;
1269 	}
1270 
1271 	err = otx2_sync_mbox_msg(&pf->mbox);
1272 	mutex_unlock(&pf->mbox.lock);
1273 	return err;
1274 }
1275 
1276 int otx2_reset_mac_stats(struct otx2_nic *pfvf)
1277 {
1278 	struct msg_req *req;
1279 	int err;
1280 
1281 	mutex_lock(&pfvf->mbox.lock);
1282 	req = otx2_mbox_alloc_msg_cgx_stats_rst(&pfvf->mbox);
1283 	if (!req) {
1284 		mutex_unlock(&pfvf->mbox.lock);
1285 		return -ENOMEM;
1286 	}
1287 
1288 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1289 	mutex_unlock(&pfvf->mbox.lock);
1290 	return err;
1291 }
1292 
1293 static int otx2_cgx_config_loopback(struct otx2_nic *pf, bool enable)
1294 {
1295 	struct msg_req *msg;
1296 	int err;
1297 
1298 	if (enable && !bitmap_empty(pf->flow_cfg->dmacflt_bmap,
1299 				    pf->flow_cfg->dmacflt_max_flows))
1300 		netdev_warn(pf->netdev,
1301 			    "CGX/RPM internal loopback might not work as DMAC filters are active\n");
1302 
1303 	mutex_lock(&pf->mbox.lock);
1304 	if (enable)
1305 		msg = otx2_mbox_alloc_msg_cgx_intlbk_enable(&pf->mbox);
1306 	else
1307 		msg = otx2_mbox_alloc_msg_cgx_intlbk_disable(&pf->mbox);
1308 
1309 	if (!msg) {
1310 		mutex_unlock(&pf->mbox.lock);
1311 		return -ENOMEM;
1312 	}
1313 
1314 	err = otx2_sync_mbox_msg(&pf->mbox);
1315 	mutex_unlock(&pf->mbox.lock);
1316 	return err;
1317 }
1318 
1319 int otx2_set_real_num_queues(struct net_device *netdev,
1320 			     int tx_queues, int rx_queues)
1321 {
1322 	int err;
1323 
1324 	err = netif_set_real_num_tx_queues(netdev, tx_queues);
1325 	if (err) {
1326 		netdev_err(netdev,
1327 			   "Failed to set no of Tx queues: %d\n", tx_queues);
1328 		return err;
1329 	}
1330 
1331 	err = netif_set_real_num_rx_queues(netdev, rx_queues);
1332 	if (err)
1333 		netdev_err(netdev,
1334 			   "Failed to set no of Rx queues: %d\n", rx_queues);
1335 	return err;
1336 }
1337 EXPORT_SYMBOL(otx2_set_real_num_queues);
1338 
1339 static char *nix_sqoperr_e_str[NIX_SQOPERR_MAX] = {
1340 	"NIX_SQOPERR_OOR",
1341 	"NIX_SQOPERR_CTX_FAULT",
1342 	"NIX_SQOPERR_CTX_POISON",
1343 	"NIX_SQOPERR_DISABLED",
1344 	"NIX_SQOPERR_SIZE_ERR",
1345 	"NIX_SQOPERR_OFLOW",
1346 	"NIX_SQOPERR_SQB_NULL",
1347 	"NIX_SQOPERR_SQB_FAULT",
1348 	"NIX_SQOPERR_SQE_SZ_ZERO",
1349 };
1350 
1351 static char *nix_mnqerr_e_str[NIX_MNQERR_MAX] = {
1352 	"NIX_MNQERR_SQ_CTX_FAULT",
1353 	"NIX_MNQERR_SQ_CTX_POISON",
1354 	"NIX_MNQERR_SQB_FAULT",
1355 	"NIX_MNQERR_SQB_POISON",
1356 	"NIX_MNQERR_TOTAL_ERR",
1357 	"NIX_MNQERR_LSO_ERR",
1358 	"NIX_MNQERR_CQ_QUERY_ERR",
1359 	"NIX_MNQERR_MAX_SQE_SIZE_ERR",
1360 	"NIX_MNQERR_MAXLEN_ERR",
1361 	"NIX_MNQERR_SQE_SIZEM1_ZERO",
1362 };
1363 
1364 static char *nix_snd_status_e_str[NIX_SND_STATUS_MAX] =  {
1365 	[NIX_SND_STATUS_GOOD] = "NIX_SND_STATUS_GOOD",
1366 	[NIX_SND_STATUS_SQ_CTX_FAULT] = "NIX_SND_STATUS_SQ_CTX_FAULT",
1367 	[NIX_SND_STATUS_SQ_CTX_POISON] = "NIX_SND_STATUS_SQ_CTX_POISON",
1368 	[NIX_SND_STATUS_SQB_FAULT] = "NIX_SND_STATUS_SQB_FAULT",
1369 	[NIX_SND_STATUS_SQB_POISON] = "NIX_SND_STATUS_SQB_POISON",
1370 	[NIX_SND_STATUS_HDR_ERR] = "NIX_SND_STATUS_HDR_ERR",
1371 	[NIX_SND_STATUS_EXT_ERR] = "NIX_SND_STATUS_EXT_ERR",
1372 	[NIX_SND_STATUS_JUMP_FAULT] = "NIX_SND_STATUS_JUMP_FAULT",
1373 	[NIX_SND_STATUS_JUMP_POISON] = "NIX_SND_STATUS_JUMP_POISON",
1374 	[NIX_SND_STATUS_CRC_ERR] = "NIX_SND_STATUS_CRC_ERR",
1375 	[NIX_SND_STATUS_IMM_ERR] = "NIX_SND_STATUS_IMM_ERR",
1376 	[NIX_SND_STATUS_SG_ERR] = "NIX_SND_STATUS_SG_ERR",
1377 	[NIX_SND_STATUS_MEM_ERR] = "NIX_SND_STATUS_MEM_ERR",
1378 	[NIX_SND_STATUS_INVALID_SUBDC] = "NIX_SND_STATUS_INVALID_SUBDC",
1379 	[NIX_SND_STATUS_SUBDC_ORDER_ERR] = "NIX_SND_STATUS_SUBDC_ORDER_ERR",
1380 	[NIX_SND_STATUS_DATA_FAULT] = "NIX_SND_STATUS_DATA_FAULT",
1381 	[NIX_SND_STATUS_DATA_POISON] = "NIX_SND_STATUS_DATA_POISON",
1382 	[NIX_SND_STATUS_NPC_DROP_ACTION] = "NIX_SND_STATUS_NPC_DROP_ACTION",
1383 	[NIX_SND_STATUS_LOCK_VIOL] = "NIX_SND_STATUS_LOCK_VIOL",
1384 	[NIX_SND_STATUS_NPC_UCAST_CHAN_ERR] = "NIX_SND_STAT_NPC_UCAST_CHAN_ERR",
1385 	[NIX_SND_STATUS_NPC_MCAST_CHAN_ERR] = "NIX_SND_STAT_NPC_MCAST_CHAN_ERR",
1386 	[NIX_SND_STATUS_NPC_MCAST_ABORT] = "NIX_SND_STATUS_NPC_MCAST_ABORT",
1387 	[NIX_SND_STATUS_NPC_VTAG_PTR_ERR] = "NIX_SND_STATUS_NPC_VTAG_PTR_ERR",
1388 	[NIX_SND_STATUS_NPC_VTAG_SIZE_ERR] = "NIX_SND_STATUS_NPC_VTAG_SIZE_ERR",
1389 	[NIX_SND_STATUS_SEND_MEM_FAULT] = "NIX_SND_STATUS_SEND_MEM_FAULT",
1390 	[NIX_SND_STATUS_SEND_STATS_ERR] = "NIX_SND_STATUS_SEND_STATS_ERR",
1391 };
1392 
1393 static irqreturn_t otx2_q_intr_handler(int irq, void *data)
1394 {
1395 	struct otx2_nic *pf = data;
1396 	struct otx2_snd_queue *sq;
1397 	void __iomem *ptr;
1398 	u64 val, qidx = 0;
1399 
1400 	/* CQ */
1401 	for (qidx = 0; qidx < pf->qset.cq_cnt; qidx++) {
1402 		ptr = otx2_get_regaddr(pf, NIX_LF_CQ_OP_INT);
1403 		val = otx2_atomic64_add((qidx << 44), ptr);
1404 
1405 		otx2_write64(pf, NIX_LF_CQ_OP_INT, (qidx << 44) |
1406 			     (val & NIX_CQERRINT_BITS));
1407 		if (!(val & (NIX_CQERRINT_BITS | BIT_ULL(42))))
1408 			continue;
1409 
1410 		if (val & BIT_ULL(42)) {
1411 			netdev_err(pf->netdev,
1412 				   "CQ%lld: error reading NIX_LF_CQ_OP_INT, NIX_LF_ERR_INT 0x%llx\n",
1413 				   qidx, otx2_read64(pf, NIX_LF_ERR_INT));
1414 		} else {
1415 			if (val & BIT_ULL(NIX_CQERRINT_DOOR_ERR))
1416 				netdev_err(pf->netdev, "CQ%lld: Doorbell error",
1417 					   qidx);
1418 			if (val & BIT_ULL(NIX_CQERRINT_CQE_FAULT))
1419 				netdev_err(pf->netdev,
1420 					   "CQ%lld: Memory fault on CQE write to LLC/DRAM",
1421 					   qidx);
1422 		}
1423 
1424 		schedule_work(&pf->reset_task);
1425 	}
1426 
1427 	/* SQ */
1428 	for (qidx = 0; qidx < otx2_get_total_tx_queues(pf); qidx++) {
1429 		u64 sq_op_err_dbg, mnq_err_dbg, snd_err_dbg;
1430 		u8 sq_op_err_code, mnq_err_code, snd_err_code;
1431 
1432 		sq = &pf->qset.sq[qidx];
1433 		if (!sq->sqb_ptrs)
1434 			continue;
1435 
1436 		/* Below debug registers captures first errors corresponding to
1437 		 * those registers. We don't have to check against SQ qid as
1438 		 * these are fatal errors.
1439 		 */
1440 
1441 		ptr = otx2_get_regaddr(pf, NIX_LF_SQ_OP_INT);
1442 		val = otx2_atomic64_add((qidx << 44), ptr);
1443 		otx2_write64(pf, NIX_LF_SQ_OP_INT, (qidx << 44) |
1444 			     (val & NIX_SQINT_BITS));
1445 
1446 		if (val & BIT_ULL(42)) {
1447 			netdev_err(pf->netdev,
1448 				   "SQ%lld: error reading NIX_LF_SQ_OP_INT, NIX_LF_ERR_INT 0x%llx\n",
1449 				   qidx, otx2_read64(pf, NIX_LF_ERR_INT));
1450 			goto done;
1451 		}
1452 
1453 		sq_op_err_dbg = otx2_read64(pf, NIX_LF_SQ_OP_ERR_DBG);
1454 		if (!(sq_op_err_dbg & BIT(44)))
1455 			goto chk_mnq_err_dbg;
1456 
1457 		sq_op_err_code = FIELD_GET(GENMASK(7, 0), sq_op_err_dbg);
1458 		netdev_err(pf->netdev,
1459 			   "SQ%lld: NIX_LF_SQ_OP_ERR_DBG(0x%llx)  err=%s(%#x)\n",
1460 			   qidx, sq_op_err_dbg,
1461 			   nix_sqoperr_e_str[sq_op_err_code],
1462 			   sq_op_err_code);
1463 
1464 		otx2_write64(pf, NIX_LF_SQ_OP_ERR_DBG, BIT_ULL(44));
1465 
1466 		if (sq_op_err_code == NIX_SQOPERR_SQB_NULL)
1467 			goto chk_mnq_err_dbg;
1468 
1469 		/* Err is not NIX_SQOPERR_SQB_NULL, call aq function to read SQ structure.
1470 		 * TODO: But we are in irq context. How to call mbox functions which does sleep
1471 		 */
1472 
1473 chk_mnq_err_dbg:
1474 		mnq_err_dbg = otx2_read64(pf, NIX_LF_MNQ_ERR_DBG);
1475 		if (!(mnq_err_dbg & BIT(44)))
1476 			goto chk_snd_err_dbg;
1477 
1478 		mnq_err_code = FIELD_GET(GENMASK(7, 0), mnq_err_dbg);
1479 		netdev_err(pf->netdev,
1480 			   "SQ%lld: NIX_LF_MNQ_ERR_DBG(0x%llx)  err=%s(%#x)\n",
1481 			   qidx, mnq_err_dbg,  nix_mnqerr_e_str[mnq_err_code],
1482 			   mnq_err_code);
1483 		otx2_write64(pf, NIX_LF_MNQ_ERR_DBG, BIT_ULL(44));
1484 
1485 chk_snd_err_dbg:
1486 		snd_err_dbg = otx2_read64(pf, NIX_LF_SEND_ERR_DBG);
1487 		if (snd_err_dbg & BIT(44)) {
1488 			snd_err_code = FIELD_GET(GENMASK(7, 0), snd_err_dbg);
1489 			netdev_err(pf->netdev,
1490 				   "SQ%lld: NIX_LF_SND_ERR_DBG:0x%llx err=%s(%#x)\n",
1491 				   qidx, snd_err_dbg,
1492 				   nix_snd_status_e_str[snd_err_code],
1493 				   snd_err_code);
1494 			otx2_write64(pf, NIX_LF_SEND_ERR_DBG, BIT_ULL(44));
1495 		}
1496 
1497 done:
1498 		/* Print values and reset */
1499 		if (val & BIT_ULL(NIX_SQINT_SQB_ALLOC_FAIL))
1500 			netdev_err(pf->netdev, "SQ%lld: SQB allocation failed",
1501 				   qidx);
1502 
1503 		schedule_work(&pf->reset_task);
1504 	}
1505 
1506 	return IRQ_HANDLED;
1507 }
1508 
1509 irqreturn_t otx2_cq_intr_handler(int irq, void *cq_irq)
1510 {
1511 	struct otx2_cq_poll *cq_poll = (struct otx2_cq_poll *)cq_irq;
1512 	struct otx2_nic *pf = (struct otx2_nic *)cq_poll->dev;
1513 	int qidx = cq_poll->cint_idx;
1514 
1515 	/* Disable interrupts.
1516 	 *
1517 	 * Completion interrupts behave in a level-triggered interrupt
1518 	 * fashion, and hence have to be cleared only after it is serviced.
1519 	 */
1520 	otx2_write64(pf, NIX_LF_CINTX_ENA_W1C(qidx), BIT_ULL(0));
1521 
1522 	/* Schedule NAPI */
1523 	pf->napi_events++;
1524 	napi_schedule_irqoff(&cq_poll->napi);
1525 
1526 	return IRQ_HANDLED;
1527 }
1528 EXPORT_SYMBOL(otx2_cq_intr_handler);
1529 
1530 void otx2_disable_napi(struct otx2_nic *pf)
1531 {
1532 	struct otx2_qset *qset = &pf->qset;
1533 	struct otx2_cq_poll *cq_poll;
1534 	struct work_struct *work;
1535 	int qidx;
1536 
1537 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
1538 		cq_poll = &qset->napi[qidx];
1539 		work = &cq_poll->dim.work;
1540 		if (work->func)
1541 			cancel_work_sync(work);
1542 		napi_disable(&cq_poll->napi);
1543 		netif_napi_del(&cq_poll->napi);
1544 	}
1545 }
1546 EXPORT_SYMBOL(otx2_disable_napi);
1547 
1548 static void otx2_free_cq_res(struct otx2_nic *pf)
1549 {
1550 	struct otx2_qset *qset = &pf->qset;
1551 	struct otx2_cq_queue *cq;
1552 	int qidx;
1553 
1554 	/* Disable CQs */
1555 	otx2_ctx_disable(&pf->mbox, NIX_AQ_CTYPE_CQ, false);
1556 	for (qidx = 0; qidx < qset->cq_cnt; qidx++) {
1557 		cq = &qset->cq[qidx];
1558 		qmem_free(pf->dev, cq->cqe);
1559 	}
1560 }
1561 
1562 static void otx2_free_sq_res(struct otx2_nic *pf)
1563 {
1564 	struct otx2_qset *qset = &pf->qset;
1565 	struct otx2_snd_queue *sq;
1566 	int qidx;
1567 
1568 	/* Disable SQs */
1569 	otx2_ctx_disable(&pf->mbox, NIX_AQ_CTYPE_SQ, false);
1570 	/* Free SQB pointers */
1571 	otx2_sq_free_sqbs(pf);
1572 	for (qidx = 0; qidx < otx2_get_total_tx_queues(pf); qidx++) {
1573 		sq = &qset->sq[qidx];
1574 		/* Skip freeing Qos queues if they are not initialized */
1575 		if (!sq->sqe)
1576 			continue;
1577 		qmem_free(pf->dev, sq->sqe);
1578 		qmem_free(pf->dev, sq->sqe_ring);
1579 		qmem_free(pf->dev, sq->cpt_resp);
1580 		qmem_free(pf->dev, sq->tso_hdrs);
1581 		kfree(sq->sg);
1582 		kfree(sq->sqb_ptrs);
1583 	}
1584 }
1585 
1586 static int otx2_get_rbuf_size(struct otx2_nic *pf, int mtu)
1587 {
1588 	int frame_size;
1589 	int total_size;
1590 	int rbuf_size;
1591 
1592 	if (pf->hw.rbuf_len)
1593 		return ALIGN(pf->hw.rbuf_len, OTX2_ALIGN) + OTX2_HEAD_ROOM;
1594 
1595 	/* The data transferred by NIX to memory consists of actual packet
1596 	 * plus additional data which has timestamp and/or EDSA/HIGIG2
1597 	 * headers if interface is configured in corresponding modes.
1598 	 * NIX transfers entire data using 6 segments/buffers and writes
1599 	 * a CQE_RX descriptor with those segment addresses. First segment
1600 	 * has additional data prepended to packet. Also software omits a
1601 	 * headroom of 128 bytes in each segment. Hence the total size of
1602 	 * memory needed to receive a packet with 'mtu' is:
1603 	 * frame size =  mtu + additional data;
1604 	 * memory = frame_size + headroom * 6;
1605 	 * each receive buffer size = memory / 6;
1606 	 */
1607 	frame_size = mtu + OTX2_ETH_HLEN + OTX2_HW_TIMESTAMP_LEN;
1608 	total_size = frame_size + OTX2_HEAD_ROOM * 6;
1609 	rbuf_size = total_size / 6;
1610 
1611 	return ALIGN(rbuf_size, 2048);
1612 }
1613 
1614 int otx2_init_hw_resources(struct otx2_nic *pf)
1615 {
1616 	struct nix_lf_free_req *free_req;
1617 	struct mbox *mbox = &pf->mbox;
1618 	struct otx2_hw *hw = &pf->hw;
1619 	struct msg_req *req;
1620 	int err = 0, lvl;
1621 
1622 	/* Set required NPA LF's pool counts
1623 	 * Auras and Pools are used in a 1:1 mapping,
1624 	 * so, aura count = pool count.
1625 	 */
1626 	hw->rqpool_cnt = hw->rx_queues;
1627 	hw->sqpool_cnt = otx2_get_total_tx_queues(pf);
1628 	hw->pool_cnt = hw->rqpool_cnt + hw->sqpool_cnt;
1629 
1630 	if (!otx2_rep_dev(pf->pdev)) {
1631 		/* Maximum hardware supported transmit length */
1632 		pf->tx_max_pktlen = pf->netdev->max_mtu + OTX2_ETH_HLEN;
1633 		pf->rbsize = otx2_get_rbuf_size(pf, pf->netdev->mtu);
1634 	}
1635 
1636 	mutex_lock(&mbox->lock);
1637 	/* NPA init */
1638 	err = otx2_config_npa(pf);
1639 	if (err)
1640 		goto exit;
1641 
1642 	/* NIX init */
1643 	err = otx2_config_nix(pf);
1644 	if (err)
1645 		goto err_free_npa_lf;
1646 
1647 	/* Default disable backpressure on NIX-CPT */
1648 	otx2_nix_cpt_config_bp(pf, false);
1649 
1650 	/* Enable backpressure for CGX mapped PF/VFs */
1651 	if (!is_otx2_lbkvf(pf->pdev))
1652 		otx2_nix_config_bp(pf, true);
1653 
1654 	/* Init Auras and pools used by NIX RQ, for free buffer ptrs */
1655 	err = otx2_rq_aura_pool_init(pf);
1656 	if (err) {
1657 		mutex_unlock(&mbox->lock);
1658 		goto err_free_nix_lf;
1659 	}
1660 	/* Init Auras and pools used by NIX SQ, for queueing SQEs */
1661 	err = otx2_sq_aura_pool_init(pf);
1662 	if (err) {
1663 		mutex_unlock(&mbox->lock);
1664 		goto err_free_rq_ptrs;
1665 	}
1666 
1667 	err = otx2_txsch_alloc(pf);
1668 	if (err) {
1669 		mutex_unlock(&mbox->lock);
1670 		goto err_free_sq_ptrs;
1671 	}
1672 
1673 #ifdef CONFIG_DCB
1674 	if (pf->pfc_en) {
1675 		err = otx2_pfc_txschq_alloc(pf);
1676 		if (err) {
1677 			mutex_unlock(&mbox->lock);
1678 			goto err_free_sq_ptrs;
1679 		}
1680 	}
1681 #endif
1682 
1683 	err = otx2_config_nix_queues(pf);
1684 	if (err) {
1685 		mutex_unlock(&mbox->lock);
1686 		goto err_free_txsch;
1687 	}
1688 
1689 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
1690 		int idx;
1691 
1692 		for (idx = 0; idx < pf->hw.txschq_cnt[lvl]; idx++) {
1693 			err = otx2_txschq_config(pf, lvl, idx, false);
1694 			if (err) {
1695 				dev_err(pf->dev, "Failed to config TXSCH\n");
1696 				mutex_unlock(&mbox->lock);
1697 				goto err_free_nix_queues;
1698 			}
1699 		}
1700 	}
1701 
1702 #ifdef CONFIG_DCB
1703 	if (pf->pfc_en) {
1704 		err = otx2_pfc_txschq_config(pf);
1705 		if (err) {
1706 			mutex_unlock(&mbox->lock);
1707 			goto err_free_nix_queues;
1708 		}
1709 	}
1710 #endif
1711 
1712 	mutex_unlock(&mbox->lock);
1713 	return err;
1714 
1715 err_free_nix_queues:
1716 	otx2_free_sq_res(pf);
1717 	otx2_free_cq_res(pf);
1718 	otx2_ctx_disable(mbox, NIX_AQ_CTYPE_RQ, false);
1719 err_free_txsch:
1720 	otx2_txschq_stop(pf);
1721 err_free_sq_ptrs:
1722 	otx2_sq_free_sqbs(pf);
1723 err_free_rq_ptrs:
1724 	otx2_free_aura_ptr(pf, AURA_NIX_RQ);
1725 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_POOL, true);
1726 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_AURA, true);
1727 	otx2_aura_pool_free(pf);
1728 err_free_nix_lf:
1729 	mutex_lock(&mbox->lock);
1730 	free_req = otx2_mbox_alloc_msg_nix_lf_free(mbox);
1731 	if (free_req) {
1732 		free_req->flags = NIX_LF_DISABLE_FLOWS;
1733 		if (otx2_sync_mbox_msg(mbox))
1734 			dev_err(pf->dev, "%s failed to free nixlf\n", __func__);
1735 	}
1736 err_free_npa_lf:
1737 	/* Reset NPA LF */
1738 	req = otx2_mbox_alloc_msg_npa_lf_free(mbox);
1739 	if (req) {
1740 		if (otx2_sync_mbox_msg(mbox))
1741 			dev_err(pf->dev, "%s failed to free npalf\n", __func__);
1742 	}
1743 exit:
1744 	mutex_unlock(&mbox->lock);
1745 	return err;
1746 }
1747 EXPORT_SYMBOL(otx2_init_hw_resources);
1748 
1749 void otx2_free_hw_resources(struct otx2_nic *pf)
1750 {
1751 	struct otx2_qset *qset = &pf->qset;
1752 	struct nix_lf_free_req *free_req;
1753 	struct mbox *mbox = &pf->mbox;
1754 	struct otx2_cq_queue *cq;
1755 	struct msg_req *req;
1756 	int qidx;
1757 
1758 	/* Ensure all SQE are processed */
1759 	otx2_sqb_flush(pf);
1760 
1761 	/* Stop transmission */
1762 	otx2_txschq_stop(pf);
1763 
1764 #ifdef CONFIG_DCB
1765 	if (pf->pfc_en)
1766 		otx2_pfc_txschq_stop(pf);
1767 #endif
1768 
1769 	if (!otx2_rep_dev(pf->pdev))
1770 		otx2_clean_qos_queues(pf);
1771 
1772 	mutex_lock(&mbox->lock);
1773 	/* Disable backpressure */
1774 	if (!is_otx2_lbkvf(pf->pdev))
1775 		otx2_nix_config_bp(pf, false);
1776 	mutex_unlock(&mbox->lock);
1777 
1778 	/* Disable RQs */
1779 	otx2_ctx_disable(mbox, NIX_AQ_CTYPE_RQ, false);
1780 
1781 	/*Dequeue all CQEs */
1782 	for (qidx = 0; qidx < qset->cq_cnt; qidx++) {
1783 		cq = &qset->cq[qidx];
1784 		if (cq->cq_type == CQ_RX)
1785 			otx2_cleanup_rx_cqes(pf, cq, qidx);
1786 		else
1787 			otx2_cleanup_tx_cqes(pf, cq);
1788 	}
1789 	otx2_free_pending_sqe(pf);
1790 
1791 	otx2_free_sq_res(pf);
1792 
1793 	/* Free RQ buffer pointers*/
1794 	otx2_free_aura_ptr(pf, AURA_NIX_RQ);
1795 
1796 	otx2_free_cq_res(pf);
1797 
1798 	/* Free all ingress bandwidth profiles allocated */
1799 	if (!otx2_rep_dev(pf->pdev))
1800 		cn10k_free_all_ipolicers(pf);
1801 
1802 	mutex_lock(&mbox->lock);
1803 	/* Reset NIX LF */
1804 	free_req = otx2_mbox_alloc_msg_nix_lf_free(mbox);
1805 	if (free_req) {
1806 		free_req->flags = NIX_LF_DISABLE_FLOWS;
1807 		if (!(pf->flags & OTX2_FLAG_PF_SHUTDOWN))
1808 			free_req->flags |= NIX_LF_DONT_FREE_TX_VTAG;
1809 		if (otx2_sync_mbox_msg(mbox))
1810 			dev_err(pf->dev, "%s failed to free nixlf\n", __func__);
1811 	}
1812 	mutex_unlock(&mbox->lock);
1813 
1814 	/* Disable NPA Pool and Aura hw context */
1815 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_POOL, true);
1816 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_AURA, true);
1817 	otx2_aura_pool_free(pf);
1818 
1819 	mutex_lock(&mbox->lock);
1820 	/* Reset NPA LF */
1821 	req = otx2_mbox_alloc_msg_npa_lf_free(mbox);
1822 	if (req) {
1823 		if (otx2_sync_mbox_msg(mbox))
1824 			dev_err(pf->dev, "%s failed to free npalf\n", __func__);
1825 	}
1826 	mutex_unlock(&mbox->lock);
1827 }
1828 EXPORT_SYMBOL(otx2_free_hw_resources);
1829 
1830 static bool otx2_promisc_use_mce_list(struct otx2_nic *pfvf)
1831 {
1832 	int vf;
1833 
1834 	/* The AF driver will determine whether to allow the VF netdev or not */
1835 	if (is_otx2_vf(pfvf->pcifunc))
1836 		return true;
1837 
1838 	/* check if there are any trusted VFs associated with the PF netdev */
1839 	for (vf = 0; vf < pci_num_vf(pfvf->pdev); vf++)
1840 		if (pfvf->vf_configs[vf].trusted)
1841 			return true;
1842 	return false;
1843 }
1844 
1845 static void otx2_do_set_rx_mode(struct otx2_nic *pf)
1846 {
1847 	struct net_device *netdev = pf->netdev;
1848 	struct nix_rx_mode *req;
1849 	bool promisc = false;
1850 
1851 	if (!(netdev->flags & IFF_UP))
1852 		return;
1853 
1854 	if ((netdev->flags & IFF_PROMISC) ||
1855 	    (netdev_uc_count(netdev) > pf->flow_cfg->ucast_flt_cnt)) {
1856 		promisc = true;
1857 	}
1858 
1859 	/* Write unicast address to mcam entries or del from mcam */
1860 	if (!promisc && netdev->priv_flags & IFF_UNICAST_FLT)
1861 		__dev_uc_sync(netdev, otx2_add_macfilter, otx2_del_macfilter);
1862 
1863 	mutex_lock(&pf->mbox.lock);
1864 	req = otx2_mbox_alloc_msg_nix_set_rx_mode(&pf->mbox);
1865 	if (!req) {
1866 		mutex_unlock(&pf->mbox.lock);
1867 		return;
1868 	}
1869 
1870 	req->mode = NIX_RX_MODE_UCAST;
1871 
1872 	if (promisc)
1873 		req->mode |= NIX_RX_MODE_PROMISC;
1874 	if (netdev->flags & (IFF_ALLMULTI | IFF_MULTICAST))
1875 		req->mode |= NIX_RX_MODE_ALLMULTI;
1876 
1877 	if (otx2_promisc_use_mce_list(pf))
1878 		req->mode |= NIX_RX_MODE_USE_MCE;
1879 
1880 	otx2_sync_mbox_msg(&pf->mbox);
1881 	mutex_unlock(&pf->mbox.lock);
1882 }
1883 
1884 static void otx2_set_irq_coalesce(struct otx2_nic *pfvf)
1885 {
1886 	int cint;
1887 
1888 	for (cint = 0; cint < pfvf->hw.cint_cnt; cint++)
1889 		otx2_config_irq_coalescing(pfvf, cint);
1890 }
1891 
1892 static void otx2_dim_work(struct work_struct *w)
1893 {
1894 	struct dim_cq_moder cur_moder;
1895 	struct otx2_cq_poll *cq_poll;
1896 	struct otx2_nic *pfvf;
1897 	struct dim *dim;
1898 
1899 	dim = container_of(w, struct dim, work);
1900 	cur_moder = net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1901 	cq_poll = container_of(dim, struct otx2_cq_poll, dim);
1902 	pfvf = (struct otx2_nic *)cq_poll->dev;
1903 	pfvf->hw.cq_time_wait = (cur_moder.usec > CQ_TIMER_THRESH_MAX) ?
1904 		CQ_TIMER_THRESH_MAX : cur_moder.usec;
1905 	pfvf->hw.cq_ecount_wait = (cur_moder.pkts > NAPI_POLL_WEIGHT) ?
1906 		NAPI_POLL_WEIGHT : cur_moder.pkts;
1907 	otx2_set_irq_coalesce(pfvf);
1908 	dim->state = DIM_START_MEASURE;
1909 }
1910 
1911 void otx2_free_queue_mem(struct otx2_qset *qset)
1912 {
1913 	kfree(qset->sq);
1914 	qset->sq = NULL;
1915 	kfree(qset->cq);
1916 	qset->cq = NULL;
1917 	kfree(qset->rq);
1918 	qset->rq = NULL;
1919 	kfree(qset->napi);
1920 	qset->napi = NULL;
1921 }
1922 EXPORT_SYMBOL(otx2_free_queue_mem);
1923 
1924 int otx2_alloc_queue_mem(struct otx2_nic *pf)
1925 {
1926 	struct otx2_qset *qset = &pf->qset;
1927 	struct otx2_cq_poll *cq_poll;
1928 
1929 
1930 	/* RQ and SQs are mapped to different CQs,
1931 	 * so find out max CQ IRQs (i.e CINTs) needed.
1932 	 */
1933 	pf->hw.non_qos_queues =  pf->hw.tx_queues + pf->hw.xdp_queues;
1934 	pf->hw.cint_cnt = max3(pf->hw.rx_queues, pf->hw.tx_queues,
1935 			       pf->hw.tc_tx_queues);
1936 
1937 	pf->qset.cq_cnt = pf->hw.rx_queues + otx2_get_total_tx_queues(pf);
1938 
1939 	qset->napi = kcalloc(pf->hw.cint_cnt, sizeof(*cq_poll), GFP_KERNEL);
1940 	if (!qset->napi)
1941 		return -ENOMEM;
1942 
1943 	/* CQ size of RQ */
1944 	qset->rqe_cnt = qset->rqe_cnt ? qset->rqe_cnt : Q_COUNT(Q_SIZE_256);
1945 	/* CQ size of SQ */
1946 	qset->sqe_cnt = qset->sqe_cnt ? qset->sqe_cnt : Q_COUNT(Q_SIZE_4K);
1947 
1948 	qset->cq = kcalloc(pf->qset.cq_cnt,
1949 			   sizeof(struct otx2_cq_queue), GFP_KERNEL);
1950 	if (!qset->cq)
1951 		goto err_free_mem;
1952 
1953 	qset->sq = kcalloc(otx2_get_total_tx_queues(pf),
1954 			   sizeof(struct otx2_snd_queue), GFP_KERNEL);
1955 	if (!qset->sq)
1956 		goto err_free_mem;
1957 
1958 	qset->rq = kcalloc(pf->hw.rx_queues,
1959 			   sizeof(struct otx2_rcv_queue), GFP_KERNEL);
1960 	if (!qset->rq)
1961 		goto err_free_mem;
1962 
1963 	return 0;
1964 
1965 err_free_mem:
1966 	otx2_free_queue_mem(qset);
1967 	return -ENOMEM;
1968 }
1969 EXPORT_SYMBOL(otx2_alloc_queue_mem);
1970 
1971 int otx2_open(struct net_device *netdev)
1972 {
1973 	struct otx2_nic *pf = netdev_priv(netdev);
1974 	struct otx2_cq_poll *cq_poll = NULL;
1975 	struct otx2_qset *qset = &pf->qset;
1976 	int err = 0, qidx, vec;
1977 	char *irq_name;
1978 
1979 	netif_carrier_off(netdev);
1980 
1981 	err = otx2_alloc_queue_mem(pf);
1982 	if (err)
1983 		return err;
1984 
1985 	err = otx2_init_hw_resources(pf);
1986 	if (err)
1987 		goto err_free_mem;
1988 
1989 	/* Register NAPI handler */
1990 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
1991 		cq_poll = &qset->napi[qidx];
1992 		cq_poll->cint_idx = qidx;
1993 		/* RQ0 & SQ0 are mapped to CINT0 and so on..
1994 		 * 'cq_ids[0]' points to RQ's CQ and
1995 		 * 'cq_ids[1]' points to SQ's CQ and
1996 		 * 'cq_ids[2]' points to XDP's CQ and
1997 		 */
1998 		cq_poll->cq_ids[CQ_RX] =
1999 			(qidx <  pf->hw.rx_queues) ? qidx : CINT_INVALID_CQ;
2000 		cq_poll->cq_ids[CQ_TX] = (qidx < pf->hw.tx_queues) ?
2001 				      qidx + pf->hw.rx_queues : CINT_INVALID_CQ;
2002 		if (pf->xdp_prog)
2003 			cq_poll->cq_ids[CQ_XDP] = (qidx < pf->hw.xdp_queues) ?
2004 						  (qidx + pf->hw.rx_queues +
2005 						  pf->hw.tx_queues) :
2006 						  CINT_INVALID_CQ;
2007 		else
2008 			cq_poll->cq_ids[CQ_XDP] = CINT_INVALID_CQ;
2009 
2010 		cq_poll->cq_ids[CQ_QOS] = (qidx < pf->hw.tc_tx_queues) ?
2011 					  (qidx + pf->hw.rx_queues +
2012 					   pf->hw.non_qos_queues) :
2013 					  CINT_INVALID_CQ;
2014 
2015 		cq_poll->dev = (void *)pf;
2016 		cq_poll->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_CQE;
2017 		INIT_WORK(&cq_poll->dim.work, otx2_dim_work);
2018 		netif_napi_add(netdev, &cq_poll->napi, otx2_napi_handler);
2019 		napi_enable(&cq_poll->napi);
2020 	}
2021 
2022 	/* Set maximum frame size allowed in HW */
2023 	err = otx2_hw_set_mtu(pf, netdev->mtu);
2024 	if (err)
2025 		goto err_disable_napi;
2026 
2027 	/* Setup segmentation algorithms, if failed, clear offload capability */
2028 	otx2_setup_segmentation(pf);
2029 
2030 	/* Initialize RSS */
2031 	err = otx2_rss_init(pf);
2032 	if (err)
2033 		goto err_disable_napi;
2034 
2035 	/* Register Queue IRQ handlers */
2036 	vec = pf->hw.nix_msixoff + NIX_LF_QINT_VEC_START;
2037 	irq_name = &pf->hw.irq_name[vec * NAME_SIZE];
2038 
2039 	snprintf(irq_name, NAME_SIZE, "%s-qerr", pf->netdev->name);
2040 
2041 	err = request_irq(pci_irq_vector(pf->pdev, vec),
2042 			  otx2_q_intr_handler, 0, irq_name, pf);
2043 	if (err) {
2044 		dev_err(pf->dev,
2045 			"RVUPF%d: IRQ registration failed for QERR\n",
2046 			rvu_get_pf(pf->pdev, pf->pcifunc));
2047 		goto err_disable_napi;
2048 	}
2049 
2050 	/* Enable QINT IRQ */
2051 	otx2_write64(pf, NIX_LF_QINTX_ENA_W1S(0), BIT_ULL(0));
2052 
2053 	/* Register CQ IRQ handlers */
2054 	vec = pf->hw.nix_msixoff + NIX_LF_CINT_VEC_START;
2055 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
2056 		irq_name = &pf->hw.irq_name[vec * NAME_SIZE];
2057 		int name_len;
2058 
2059 		name_len = snprintf(irq_name, NAME_SIZE, "%s-rxtx-%d",
2060 				    pf->netdev->name, qidx);
2061 		if (name_len >= NAME_SIZE) {
2062 			dev_err(pf->dev,
2063 				"RVUPF%d: IRQ registration failed for CQ%d, irq name is too long\n",
2064 				rvu_get_pf(pf->pdev, pf->pcifunc), qidx);
2065 			err = -EINVAL;
2066 			goto err_free_cints;
2067 		}
2068 
2069 		err = request_irq(pci_irq_vector(pf->pdev, vec),
2070 				  otx2_cq_intr_handler, 0, irq_name,
2071 				  &qset->napi[qidx]);
2072 		if (err) {
2073 			dev_err(pf->dev,
2074 				"RVUPF%d: IRQ registration failed for CQ%d\n",
2075 				rvu_get_pf(pf->pdev, pf->pcifunc), qidx);
2076 			goto err_free_cints;
2077 		}
2078 		vec++;
2079 
2080 		otx2_config_irq_coalescing(pf, qidx);
2081 
2082 		/* Enable CQ IRQ */
2083 		otx2_write64(pf, NIX_LF_CINTX_INT(qidx), BIT_ULL(0));
2084 		otx2_write64(pf, NIX_LF_CINTX_ENA_W1S(qidx), BIT_ULL(0));
2085 	}
2086 
2087 	otx2_set_cints_affinity(pf);
2088 
2089 	if (pf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
2090 		otx2_enable_rxvlan(pf, true);
2091 
2092 	/* When reinitializing enable time stamping if it is enabled before */
2093 	if (pf->flags & OTX2_FLAG_TX_TSTAMP_ENABLED) {
2094 		pf->flags &= ~OTX2_FLAG_TX_TSTAMP_ENABLED;
2095 		otx2_config_hw_tx_tstamp(pf, true);
2096 	}
2097 	if (pf->flags & OTX2_FLAG_RX_TSTAMP_ENABLED) {
2098 		pf->flags &= ~OTX2_FLAG_RX_TSTAMP_ENABLED;
2099 		otx2_config_hw_rx_tstamp(pf, true);
2100 	}
2101 
2102 	pf->flags &= ~OTX2_FLAG_INTF_DOWN;
2103 	pf->flags &= ~OTX2_FLAG_PORT_UP;
2104 	/* 'intf_down' may be checked on any cpu */
2105 	smp_wmb();
2106 
2107 	/* Enable QoS configuration before starting tx queues */
2108 	otx2_qos_config_txschq(pf);
2109 
2110 	/* we have already received link status notification */
2111 	if (pf->linfo.link_up && !(pf->pcifunc & RVU_PFVF_FUNC_MASK))
2112 		otx2_handle_link_event(pf);
2113 
2114 	/* Install DMAC Filters */
2115 	if (pf->flags & OTX2_FLAG_DMACFLTR_SUPPORT)
2116 		otx2_dmacflt_reinstall_flows(pf);
2117 
2118 	otx2_tc_apply_ingress_police_rules(pf);
2119 
2120 	err = otx2_rxtx_enable(pf, true);
2121 	/* If a mbox communication error happens at this point then interface
2122 	 * will end up in a state such that it is in down state but hardware
2123 	 * mcam entries are enabled to receive the packets. Hence disable the
2124 	 * packet I/O.
2125 	 */
2126 	if (err == -EIO)
2127 		goto err_disable_rxtx;
2128 	else if (err)
2129 		goto err_tx_stop_queues;
2130 
2131 	otx2_do_set_rx_mode(pf);
2132 
2133 	return 0;
2134 
2135 err_disable_rxtx:
2136 	otx2_rxtx_enable(pf, false);
2137 err_tx_stop_queues:
2138 	netif_tx_stop_all_queues(netdev);
2139 	netif_carrier_off(netdev);
2140 	pf->flags |= OTX2_FLAG_INTF_DOWN;
2141 err_free_cints:
2142 	otx2_free_cints(pf, qidx);
2143 	vec = pci_irq_vector(pf->pdev,
2144 			     pf->hw.nix_msixoff + NIX_LF_QINT_VEC_START);
2145 	otx2_write64(pf, NIX_LF_QINTX_ENA_W1C(0), BIT_ULL(0));
2146 	free_irq(vec, pf);
2147 err_disable_napi:
2148 	otx2_disable_napi(pf);
2149 	otx2_free_hw_resources(pf);
2150 err_free_mem:
2151 	otx2_free_queue_mem(qset);
2152 	return err;
2153 }
2154 EXPORT_SYMBOL(otx2_open);
2155 
2156 int otx2_stop(struct net_device *netdev)
2157 {
2158 	struct otx2_nic *pf = netdev_priv(netdev);
2159 	struct otx2_cq_poll *cq_poll = NULL;
2160 	struct otx2_qset *qset = &pf->qset;
2161 	struct otx2_rss_info *rss;
2162 	int qidx, vec, wrk;
2163 
2164 	/* If the DOWN flag is set resources are already freed */
2165 	if (pf->flags & OTX2_FLAG_INTF_DOWN)
2166 		return 0;
2167 
2168 	netif_carrier_off(netdev);
2169 	netif_tx_stop_all_queues(netdev);
2170 
2171 	pf->flags |= OTX2_FLAG_INTF_DOWN;
2172 	/* 'intf_down' may be checked on any cpu */
2173 	smp_wmb();
2174 
2175 	/* First stop packet Rx/Tx */
2176 	otx2_rxtx_enable(pf, false);
2177 
2178 	/* Clear RSS enable flag */
2179 	rss = &pf->hw.rss_info;
2180 	rss->enable = false;
2181 	if (!netif_is_rxfh_configured(netdev))
2182 		kfree(rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP]);
2183 
2184 	/* Cleanup Queue IRQ */
2185 	vec = pci_irq_vector(pf->pdev,
2186 			     pf->hw.nix_msixoff + NIX_LF_QINT_VEC_START);
2187 	otx2_write64(pf, NIX_LF_QINTX_ENA_W1C(0), BIT_ULL(0));
2188 	free_irq(vec, pf);
2189 
2190 	/* Cleanup CQ NAPI and IRQ */
2191 	vec = pf->hw.nix_msixoff + NIX_LF_CINT_VEC_START;
2192 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
2193 		/* Disable interrupt */
2194 		otx2_write64(pf, NIX_LF_CINTX_ENA_W1C(qidx), BIT_ULL(0));
2195 
2196 		synchronize_irq(pci_irq_vector(pf->pdev, vec));
2197 
2198 		cq_poll = &qset->napi[qidx];
2199 		napi_synchronize(&cq_poll->napi);
2200 		vec++;
2201 	}
2202 
2203 	netif_tx_disable(netdev);
2204 
2205 	for (wrk = 0; wrk < pf->qset.cq_cnt; wrk++)
2206 		cancel_delayed_work_sync(&pf->refill_wrk[wrk].pool_refill_work);
2207 	devm_kfree(pf->dev, pf->refill_wrk);
2208 
2209 	otx2_free_hw_resources(pf);
2210 	otx2_free_cints(pf, pf->hw.cint_cnt);
2211 	otx2_disable_napi(pf);
2212 
2213 	for (qidx = 0; qidx < netdev->num_tx_queues; qidx++)
2214 		netdev_tx_reset_queue(netdev_get_tx_queue(netdev, qidx));
2215 
2216 	otx2_free_queue_mem(qset);
2217 	/* Do not clear RQ/SQ ringsize settings */
2218 	memset_startat(qset, 0, sqe_cnt);
2219 	return 0;
2220 }
2221 EXPORT_SYMBOL(otx2_stop);
2222 
2223 static netdev_tx_t otx2_xmit(struct sk_buff *skb, struct net_device *netdev)
2224 {
2225 	struct otx2_nic *pf = netdev_priv(netdev);
2226 	int qidx = skb_get_queue_mapping(skb);
2227 	struct otx2_snd_queue *sq;
2228 	struct netdev_queue *txq;
2229 	int sq_idx;
2230 
2231 	/* XDP SQs are not mapped with TXQs
2232 	 * advance qid to derive correct sq mapped with QOS
2233 	 */
2234 	sq_idx = (qidx >= pf->hw.tx_queues) ? (qidx + pf->hw.xdp_queues) : qidx;
2235 
2236 	/* Check for minimum and maximum packet length */
2237 	if (skb->len <= ETH_HLEN ||
2238 	    (!skb_shinfo(skb)->gso_size && skb->len > pf->tx_max_pktlen)) {
2239 		dev_kfree_skb(skb);
2240 		return NETDEV_TX_OK;
2241 	}
2242 
2243 	sq = &pf->qset.sq[sq_idx];
2244 	txq = netdev_get_tx_queue(netdev, qidx);
2245 
2246 	if (!otx2_sq_append_skb(pf, txq, sq, skb, qidx)) {
2247 		netif_tx_stop_queue(txq);
2248 
2249 		/* Check again, incase SQBs got freed up */
2250 		smp_mb();
2251 		if (((sq->num_sqbs - *sq->aura_fc_addr) * sq->sqe_per_sqb)
2252 							> sq->sqe_thresh)
2253 			netif_tx_wake_queue(txq);
2254 
2255 		return NETDEV_TX_BUSY;
2256 	}
2257 
2258 	return NETDEV_TX_OK;
2259 }
2260 
2261 static int otx2_qos_select_htb_queue(struct otx2_nic *pf, struct sk_buff *skb,
2262 				     u16 htb_maj_id)
2263 {
2264 	u16 classid;
2265 
2266 	if ((TC_H_MAJ(skb->priority) >> 16) == htb_maj_id)
2267 		classid = TC_H_MIN(skb->priority);
2268 	else
2269 		classid = READ_ONCE(pf->qos.defcls);
2270 
2271 	if (!classid)
2272 		return 0;
2273 
2274 	return otx2_get_txq_by_classid(pf, classid);
2275 }
2276 
2277 u16 otx2_select_queue(struct net_device *netdev, struct sk_buff *skb,
2278 		      struct net_device *sb_dev)
2279 {
2280 	struct otx2_nic *pf = netdev_priv(netdev);
2281 	bool qos_enabled;
2282 #ifdef CONFIG_DCB
2283 	u8 vlan_prio;
2284 #endif
2285 	int txq;
2286 
2287 	qos_enabled = netdev->real_num_tx_queues > pf->hw.tx_queues;
2288 	if (unlikely(qos_enabled)) {
2289 		/* This smp_load_acquire() pairs with smp_store_release() in
2290 		 * otx2_qos_root_add() called from htb offload root creation
2291 		 */
2292 		u16 htb_maj_id = smp_load_acquire(&pf->qos.maj_id);
2293 
2294 		if (unlikely(htb_maj_id)) {
2295 			txq = otx2_qos_select_htb_queue(pf, skb, htb_maj_id);
2296 			if (txq > 0)
2297 				return txq;
2298 			goto process_pfc;
2299 		}
2300 	}
2301 
2302 process_pfc:
2303 #ifdef CONFIG_DCB
2304 	if (!skb_vlan_tag_present(skb))
2305 		goto pick_tx;
2306 
2307 	vlan_prio = skb->vlan_tci >> 13;
2308 	if ((vlan_prio > pf->hw.tx_queues - 1) ||
2309 	    !pf->pfc_alloc_status[vlan_prio])
2310 		goto pick_tx;
2311 
2312 	return vlan_prio;
2313 
2314 pick_tx:
2315 #endif
2316 	txq = netdev_pick_tx(netdev, skb, NULL);
2317 	if (unlikely(qos_enabled))
2318 		return txq % pf->hw.tx_queues;
2319 
2320 	return txq;
2321 }
2322 EXPORT_SYMBOL(otx2_select_queue);
2323 
2324 static netdev_features_t otx2_fix_features(struct net_device *dev,
2325 					   netdev_features_t features)
2326 {
2327 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
2328 		features |= NETIF_F_HW_VLAN_STAG_RX;
2329 	else
2330 		features &= ~NETIF_F_HW_VLAN_STAG_RX;
2331 
2332 	return features;
2333 }
2334 
2335 static void otx2_set_rx_mode(struct net_device *netdev)
2336 {
2337 	struct otx2_nic *pf = netdev_priv(netdev);
2338 
2339 	queue_work(pf->otx2_wq, &pf->rx_mode_work);
2340 }
2341 
2342 static void otx2_rx_mode_wrk_handler(struct work_struct *work)
2343 {
2344 	struct otx2_nic *pf = container_of(work, struct otx2_nic, rx_mode_work);
2345 
2346 	otx2_do_set_rx_mode(pf);
2347 }
2348 
2349 static int otx2_set_features(struct net_device *netdev,
2350 			     netdev_features_t features)
2351 {
2352 	netdev_features_t changed = features ^ netdev->features;
2353 	struct otx2_nic *pf = netdev_priv(netdev);
2354 
2355 	if ((changed & NETIF_F_LOOPBACK) && netif_running(netdev))
2356 		return otx2_cgx_config_loopback(pf,
2357 						features & NETIF_F_LOOPBACK);
2358 
2359 	if ((changed & NETIF_F_HW_VLAN_CTAG_RX) && netif_running(netdev))
2360 		return otx2_enable_rxvlan(pf,
2361 					  features & NETIF_F_HW_VLAN_CTAG_RX);
2362 
2363 	if (changed & NETIF_F_HW_ESP)
2364 		return cn10k_ipsec_ethtool_init(netdev,
2365 						features & NETIF_F_HW_ESP);
2366 
2367 	return otx2_handle_ntuple_tc_features(netdev, features);
2368 }
2369 
2370 static void otx2_reset_task(struct work_struct *work)
2371 {
2372 	struct otx2_nic *pf = container_of(work, struct otx2_nic, reset_task);
2373 
2374 	if (!netif_running(pf->netdev))
2375 		return;
2376 
2377 	rtnl_lock();
2378 	otx2_stop(pf->netdev);
2379 	pf->reset_count++;
2380 	otx2_open(pf->netdev);
2381 	netif_trans_update(pf->netdev);
2382 	rtnl_unlock();
2383 }
2384 
2385 static int otx2_config_hw_rx_tstamp(struct otx2_nic *pfvf, bool enable)
2386 {
2387 	struct msg_req *req;
2388 	int err;
2389 
2390 	if (pfvf->flags & OTX2_FLAG_RX_TSTAMP_ENABLED && enable)
2391 		return 0;
2392 
2393 	mutex_lock(&pfvf->mbox.lock);
2394 	if (enable)
2395 		req = otx2_mbox_alloc_msg_cgx_ptp_rx_enable(&pfvf->mbox);
2396 	else
2397 		req = otx2_mbox_alloc_msg_cgx_ptp_rx_disable(&pfvf->mbox);
2398 	if (!req) {
2399 		mutex_unlock(&pfvf->mbox.lock);
2400 		return -ENOMEM;
2401 	}
2402 
2403 	err = otx2_sync_mbox_msg(&pfvf->mbox);
2404 	if (err) {
2405 		mutex_unlock(&pfvf->mbox.lock);
2406 		return err;
2407 	}
2408 
2409 	mutex_unlock(&pfvf->mbox.lock);
2410 	if (enable)
2411 		pfvf->flags |= OTX2_FLAG_RX_TSTAMP_ENABLED;
2412 	else
2413 		pfvf->flags &= ~OTX2_FLAG_RX_TSTAMP_ENABLED;
2414 	return 0;
2415 }
2416 
2417 static int otx2_config_hw_tx_tstamp(struct otx2_nic *pfvf, bool enable)
2418 {
2419 	struct msg_req *req;
2420 	int err;
2421 
2422 	if (pfvf->flags & OTX2_FLAG_TX_TSTAMP_ENABLED && enable)
2423 		return 0;
2424 
2425 	mutex_lock(&pfvf->mbox.lock);
2426 	if (enable)
2427 		req = otx2_mbox_alloc_msg_nix_lf_ptp_tx_enable(&pfvf->mbox);
2428 	else
2429 		req = otx2_mbox_alloc_msg_nix_lf_ptp_tx_disable(&pfvf->mbox);
2430 	if (!req) {
2431 		mutex_unlock(&pfvf->mbox.lock);
2432 		return -ENOMEM;
2433 	}
2434 
2435 	err = otx2_sync_mbox_msg(&pfvf->mbox);
2436 	if (err) {
2437 		mutex_unlock(&pfvf->mbox.lock);
2438 		return err;
2439 	}
2440 
2441 	mutex_unlock(&pfvf->mbox.lock);
2442 	if (enable)
2443 		pfvf->flags |= OTX2_FLAG_TX_TSTAMP_ENABLED;
2444 	else
2445 		pfvf->flags &= ~OTX2_FLAG_TX_TSTAMP_ENABLED;
2446 	return 0;
2447 }
2448 
2449 int otx2_config_hwtstamp(struct net_device *netdev, struct ifreq *ifr)
2450 {
2451 	struct otx2_nic *pfvf = netdev_priv(netdev);
2452 	struct hwtstamp_config config;
2453 
2454 	if (!pfvf->ptp)
2455 		return -ENODEV;
2456 
2457 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2458 		return -EFAULT;
2459 
2460 	switch (config.tx_type) {
2461 	case HWTSTAMP_TX_OFF:
2462 		if (pfvf->flags & OTX2_FLAG_PTP_ONESTEP_SYNC)
2463 			pfvf->flags &= ~OTX2_FLAG_PTP_ONESTEP_SYNC;
2464 
2465 		cancel_delayed_work(&pfvf->ptp->synctstamp_work);
2466 		otx2_config_hw_tx_tstamp(pfvf, false);
2467 		break;
2468 	case HWTSTAMP_TX_ONESTEP_SYNC:
2469 		if (!test_bit(CN10K_PTP_ONESTEP, &pfvf->hw.cap_flag))
2470 			return -ERANGE;
2471 		pfvf->flags |= OTX2_FLAG_PTP_ONESTEP_SYNC;
2472 		schedule_delayed_work(&pfvf->ptp->synctstamp_work,
2473 				      msecs_to_jiffies(500));
2474 		fallthrough;
2475 	case HWTSTAMP_TX_ON:
2476 		otx2_config_hw_tx_tstamp(pfvf, true);
2477 		break;
2478 	default:
2479 		return -ERANGE;
2480 	}
2481 
2482 	switch (config.rx_filter) {
2483 	case HWTSTAMP_FILTER_NONE:
2484 		otx2_config_hw_rx_tstamp(pfvf, false);
2485 		break;
2486 	case HWTSTAMP_FILTER_ALL:
2487 	case HWTSTAMP_FILTER_SOME:
2488 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2489 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2490 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2491 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2492 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2493 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2494 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2495 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2496 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2497 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2498 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2499 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2500 		otx2_config_hw_rx_tstamp(pfvf, true);
2501 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2502 		break;
2503 	default:
2504 		return -ERANGE;
2505 	}
2506 
2507 	memcpy(&pfvf->tstamp, &config, sizeof(config));
2508 
2509 	return copy_to_user(ifr->ifr_data, &config,
2510 			    sizeof(config)) ? -EFAULT : 0;
2511 }
2512 EXPORT_SYMBOL(otx2_config_hwtstamp);
2513 
2514 int otx2_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
2515 {
2516 	struct otx2_nic *pfvf = netdev_priv(netdev);
2517 	struct hwtstamp_config *cfg = &pfvf->tstamp;
2518 
2519 	switch (cmd) {
2520 	case SIOCSHWTSTAMP:
2521 		return otx2_config_hwtstamp(netdev, req);
2522 	case SIOCGHWTSTAMP:
2523 		return copy_to_user(req->ifr_data, cfg,
2524 				    sizeof(*cfg)) ? -EFAULT : 0;
2525 	default:
2526 		return -EOPNOTSUPP;
2527 	}
2528 }
2529 EXPORT_SYMBOL(otx2_ioctl);
2530 
2531 static int otx2_do_set_vf_mac(struct otx2_nic *pf, int vf, const u8 *mac)
2532 {
2533 	struct npc_install_flow_req *req;
2534 	int err;
2535 
2536 	mutex_lock(&pf->mbox.lock);
2537 	req = otx2_mbox_alloc_msg_npc_install_flow(&pf->mbox);
2538 	if (!req) {
2539 		err = -ENOMEM;
2540 		goto out;
2541 	}
2542 
2543 	ether_addr_copy(req->packet.dmac, mac);
2544 	eth_broadcast_addr((u8 *)&req->mask.dmac);
2545 	req->features = BIT_ULL(NPC_DMAC);
2546 	req->channel = pf->hw.rx_chan_base;
2547 	req->intf = NIX_INTF_RX;
2548 	req->default_rule = 1;
2549 	req->append = 1;
2550 	req->vf = vf + 1;
2551 	req->op = NIX_RX_ACTION_DEFAULT;
2552 
2553 	err = otx2_sync_mbox_msg(&pf->mbox);
2554 out:
2555 	mutex_unlock(&pf->mbox.lock);
2556 	return err;
2557 }
2558 
2559 static int otx2_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
2560 {
2561 	struct otx2_nic *pf = netdev_priv(netdev);
2562 	struct pci_dev *pdev = pf->pdev;
2563 	struct otx2_vf_config *config;
2564 	int ret;
2565 
2566 	if (!netif_running(netdev))
2567 		return -EAGAIN;
2568 
2569 	if (vf >= pf->total_vfs)
2570 		return -EINVAL;
2571 
2572 	if (!is_valid_ether_addr(mac))
2573 		return -EINVAL;
2574 
2575 	config = &pf->vf_configs[vf];
2576 	ether_addr_copy(config->mac, mac);
2577 
2578 	ret = otx2_do_set_vf_mac(pf, vf, mac);
2579 	if (ret == 0)
2580 		dev_info(&pdev->dev,
2581 			 "Load/Reload VF driver\n");
2582 
2583 	return ret;
2584 }
2585 
2586 static int otx2_do_set_vf_vlan(struct otx2_nic *pf, int vf, u16 vlan, u8 qos,
2587 			       __be16 proto)
2588 {
2589 	struct otx2_flow_config *flow_cfg = pf->flow_cfg;
2590 	struct nix_vtag_config_rsp *vtag_rsp;
2591 	struct npc_delete_flow_req *del_req;
2592 	struct nix_vtag_config *vtag_req;
2593 	struct npc_install_flow_req *req;
2594 	struct otx2_vf_config *config;
2595 	int err = 0;
2596 	u32 idx;
2597 
2598 	config = &pf->vf_configs[vf];
2599 
2600 	if (!vlan && !config->vlan)
2601 		goto out;
2602 
2603 	mutex_lock(&pf->mbox.lock);
2604 
2605 	/* free old tx vtag entry */
2606 	if (config->vlan) {
2607 		vtag_req = otx2_mbox_alloc_msg_nix_vtag_cfg(&pf->mbox);
2608 		if (!vtag_req) {
2609 			err = -ENOMEM;
2610 			goto out;
2611 		}
2612 		vtag_req->cfg_type = 0;
2613 		vtag_req->tx.free_vtag0 = 1;
2614 		vtag_req->tx.vtag0_idx = config->tx_vtag_idx;
2615 
2616 		err = otx2_sync_mbox_msg(&pf->mbox);
2617 		if (err)
2618 			goto out;
2619 	}
2620 
2621 	if (!vlan && config->vlan) {
2622 		/* rx */
2623 		del_req = otx2_mbox_alloc_msg_npc_delete_flow(&pf->mbox);
2624 		if (!del_req) {
2625 			err = -ENOMEM;
2626 			goto out;
2627 		}
2628 		idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_RX_INDEX);
2629 		del_req->entry =
2630 			flow_cfg->def_ent[flow_cfg->vf_vlan_offset + idx];
2631 		err = otx2_sync_mbox_msg(&pf->mbox);
2632 		if (err)
2633 			goto out;
2634 
2635 		/* tx */
2636 		del_req = otx2_mbox_alloc_msg_npc_delete_flow(&pf->mbox);
2637 		if (!del_req) {
2638 			err = -ENOMEM;
2639 			goto out;
2640 		}
2641 		idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_TX_INDEX);
2642 		del_req->entry =
2643 			flow_cfg->def_ent[flow_cfg->vf_vlan_offset + idx];
2644 		err = otx2_sync_mbox_msg(&pf->mbox);
2645 
2646 		goto out;
2647 	}
2648 
2649 	/* rx */
2650 	req = otx2_mbox_alloc_msg_npc_install_flow(&pf->mbox);
2651 	if (!req) {
2652 		err = -ENOMEM;
2653 		goto out;
2654 	}
2655 
2656 	idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_RX_INDEX);
2657 	req->entry = flow_cfg->def_ent[flow_cfg->vf_vlan_offset + idx];
2658 	req->packet.vlan_tci = htons(vlan);
2659 	req->mask.vlan_tci = htons(VLAN_VID_MASK);
2660 	/* af fills the destination mac addr */
2661 	eth_broadcast_addr((u8 *)&req->mask.dmac);
2662 	req->features = BIT_ULL(NPC_OUTER_VID) | BIT_ULL(NPC_DMAC);
2663 	req->channel = pf->hw.rx_chan_base;
2664 	req->intf = NIX_INTF_RX;
2665 	req->vf = vf + 1;
2666 	req->op = NIX_RX_ACTION_DEFAULT;
2667 	req->vtag0_valid = true;
2668 	req->vtag0_type = NIX_AF_LFX_RX_VTAG_TYPE7;
2669 	req->set_cntr = 1;
2670 
2671 	err = otx2_sync_mbox_msg(&pf->mbox);
2672 	if (err)
2673 		goto out;
2674 
2675 	/* tx */
2676 	vtag_req = otx2_mbox_alloc_msg_nix_vtag_cfg(&pf->mbox);
2677 	if (!vtag_req) {
2678 		err = -ENOMEM;
2679 		goto out;
2680 	}
2681 
2682 	/* configure tx vtag params */
2683 	vtag_req->vtag_size = VTAGSIZE_T4;
2684 	vtag_req->cfg_type = 0; /* tx vlan cfg */
2685 	vtag_req->tx.cfg_vtag0 = 1;
2686 	vtag_req->tx.vtag0 = ((u64)ntohs(proto) << 16) | vlan;
2687 
2688 	err = otx2_sync_mbox_msg(&pf->mbox);
2689 	if (err)
2690 		goto out;
2691 
2692 	vtag_rsp = (struct nix_vtag_config_rsp *)otx2_mbox_get_rsp
2693 			(&pf->mbox.mbox, 0, &vtag_req->hdr);
2694 	if (IS_ERR(vtag_rsp)) {
2695 		err = PTR_ERR(vtag_rsp);
2696 		goto out;
2697 	}
2698 	config->tx_vtag_idx = vtag_rsp->vtag0_idx;
2699 
2700 	req = otx2_mbox_alloc_msg_npc_install_flow(&pf->mbox);
2701 	if (!req) {
2702 		err = -ENOMEM;
2703 		goto out;
2704 	}
2705 
2706 	eth_zero_addr((u8 *)&req->mask.dmac);
2707 	idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_TX_INDEX);
2708 	req->entry = flow_cfg->def_ent[flow_cfg->vf_vlan_offset + idx];
2709 	req->features = BIT_ULL(NPC_DMAC);
2710 	req->channel = pf->hw.tx_chan_base;
2711 	req->intf = NIX_INTF_TX;
2712 	req->vf = vf + 1;
2713 	req->op = NIX_TX_ACTIONOP_UCAST_DEFAULT;
2714 	req->vtag0_def = vtag_rsp->vtag0_idx;
2715 	req->vtag0_op = VTAG_INSERT;
2716 	req->set_cntr = 1;
2717 
2718 	err = otx2_sync_mbox_msg(&pf->mbox);
2719 out:
2720 	config->vlan = vlan;
2721 	mutex_unlock(&pf->mbox.lock);
2722 	return err;
2723 }
2724 
2725 static int otx2_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
2726 			    __be16 proto)
2727 {
2728 	struct otx2_nic *pf = netdev_priv(netdev);
2729 	struct pci_dev *pdev = pf->pdev;
2730 
2731 	if (!netif_running(netdev))
2732 		return -EAGAIN;
2733 
2734 	if (vf >= pci_num_vf(pdev))
2735 		return -EINVAL;
2736 
2737 	/* qos is currently unsupported */
2738 	if (vlan >= VLAN_N_VID || qos)
2739 		return -EINVAL;
2740 
2741 	if (proto != htons(ETH_P_8021Q))
2742 		return -EPROTONOSUPPORT;
2743 
2744 	if (!(pf->flags & OTX2_FLAG_VF_VLAN_SUPPORT))
2745 		return -EOPNOTSUPP;
2746 
2747 	return otx2_do_set_vf_vlan(pf, vf, vlan, qos, proto);
2748 }
2749 
2750 static int otx2_get_vf_config(struct net_device *netdev, int vf,
2751 			      struct ifla_vf_info *ivi)
2752 {
2753 	struct otx2_nic *pf = netdev_priv(netdev);
2754 	struct pci_dev *pdev = pf->pdev;
2755 	struct otx2_vf_config *config;
2756 
2757 	if (!netif_running(netdev))
2758 		return -EAGAIN;
2759 
2760 	if (vf >= pci_num_vf(pdev))
2761 		return -EINVAL;
2762 
2763 	config = &pf->vf_configs[vf];
2764 	ivi->vf = vf;
2765 	ether_addr_copy(ivi->mac, config->mac);
2766 	ivi->vlan = config->vlan;
2767 	ivi->trusted = config->trusted;
2768 
2769 	return 0;
2770 }
2771 
2772 static int otx2_xdp_xmit_tx(struct otx2_nic *pf, struct xdp_frame *xdpf,
2773 			    int qidx)
2774 {
2775 	u64 dma_addr;
2776 	int err = 0;
2777 
2778 	dma_addr = otx2_dma_map_page(pf, virt_to_page(xdpf->data),
2779 				     offset_in_page(xdpf->data), xdpf->len,
2780 				     DMA_TO_DEVICE);
2781 	if (dma_mapping_error(pf->dev, dma_addr))
2782 		return -ENOMEM;
2783 
2784 	err = otx2_xdp_sq_append_pkt(pf, xdpf, dma_addr, xdpf->len,
2785 				     qidx, OTX2_XDP_REDIRECT);
2786 	if (!err) {
2787 		otx2_dma_unmap_page(pf, dma_addr, xdpf->len, DMA_TO_DEVICE);
2788 		xdp_return_frame(xdpf);
2789 		return -ENOMEM;
2790 	}
2791 	return 0;
2792 }
2793 
2794 static int otx2_xdp_xmit(struct net_device *netdev, int n,
2795 			 struct xdp_frame **frames, u32 flags)
2796 {
2797 	struct otx2_nic *pf = netdev_priv(netdev);
2798 	int qidx = smp_processor_id();
2799 	struct otx2_snd_queue *sq;
2800 	int drops = 0, i;
2801 
2802 	if (!netif_running(netdev))
2803 		return -ENETDOWN;
2804 
2805 	qidx += pf->hw.tx_queues;
2806 	sq = pf->xdp_prog ? &pf->qset.sq[qidx] : NULL;
2807 
2808 	/* Abort xmit if xdp queue is not */
2809 	if (unlikely(!sq))
2810 		return -ENXIO;
2811 
2812 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2813 		return -EINVAL;
2814 
2815 	for (i = 0; i < n; i++) {
2816 		struct xdp_frame *xdpf = frames[i];
2817 		int err;
2818 
2819 		err = otx2_xdp_xmit_tx(pf, xdpf, qidx);
2820 		if (err)
2821 			drops++;
2822 	}
2823 	return n - drops;
2824 }
2825 
2826 static int otx2_xdp_setup(struct otx2_nic *pf, struct bpf_prog *prog)
2827 {
2828 	struct net_device *dev = pf->netdev;
2829 	bool if_up = netif_running(pf->netdev);
2830 	struct bpf_prog *old_prog;
2831 
2832 	if (prog && dev->mtu > MAX_XDP_MTU) {
2833 		netdev_warn(dev, "Jumbo frames not yet supported with XDP\n");
2834 		return -EOPNOTSUPP;
2835 	}
2836 
2837 	if (if_up)
2838 		otx2_stop(pf->netdev);
2839 
2840 	old_prog = xchg(&pf->xdp_prog, prog);
2841 
2842 	if (old_prog)
2843 		bpf_prog_put(old_prog);
2844 
2845 	if (pf->xdp_prog)
2846 		bpf_prog_add(pf->xdp_prog, pf->hw.rx_queues - 1);
2847 
2848 	/* Network stack and XDP shared same rx queues.
2849 	 * Use separate tx queues for XDP and network stack.
2850 	 */
2851 	if (pf->xdp_prog) {
2852 		pf->hw.xdp_queues = pf->hw.rx_queues;
2853 		xdp_features_set_redirect_target(dev, false);
2854 	} else {
2855 		pf->hw.xdp_queues = 0;
2856 		xdp_features_clear_redirect_target(dev);
2857 	}
2858 
2859 	if (if_up)
2860 		otx2_open(pf->netdev);
2861 
2862 	return 0;
2863 }
2864 
2865 static int otx2_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
2866 {
2867 	struct otx2_nic *pf = netdev_priv(netdev);
2868 
2869 	switch (xdp->command) {
2870 	case XDP_SETUP_PROG:
2871 		return otx2_xdp_setup(pf, xdp->prog);
2872 	case XDP_SETUP_XSK_POOL:
2873 		return otx2_xsk_pool_setup(pf, xdp->xsk.pool, xdp->xsk.queue_id);
2874 	default:
2875 		return -EINVAL;
2876 	}
2877 }
2878 
2879 static int otx2_set_vf_permissions(struct otx2_nic *pf, int vf,
2880 				   int req_perm)
2881 {
2882 	struct set_vf_perm *req;
2883 	int rc;
2884 
2885 	mutex_lock(&pf->mbox.lock);
2886 	req = otx2_mbox_alloc_msg_set_vf_perm(&pf->mbox);
2887 	if (!req) {
2888 		rc = -ENOMEM;
2889 		goto out;
2890 	}
2891 
2892 	/* Let AF reset VF permissions as sriov is disabled */
2893 	if (req_perm == OTX2_RESET_VF_PERM) {
2894 		req->flags |= RESET_VF_PERM;
2895 	} else if (req_perm == OTX2_TRUSTED_VF) {
2896 		if (pf->vf_configs[vf].trusted)
2897 			req->flags |= VF_TRUSTED;
2898 	}
2899 
2900 	req->vf = vf;
2901 	rc = otx2_sync_mbox_msg(&pf->mbox);
2902 out:
2903 	mutex_unlock(&pf->mbox.lock);
2904 	return rc;
2905 }
2906 
2907 static int otx2_ndo_set_vf_trust(struct net_device *netdev, int vf,
2908 				 bool enable)
2909 {
2910 	struct otx2_nic *pf = netdev_priv(netdev);
2911 	struct pci_dev *pdev = pf->pdev;
2912 	int rc;
2913 
2914 	if (vf >= pci_num_vf(pdev))
2915 		return -EINVAL;
2916 
2917 	if (pf->vf_configs[vf].trusted == enable)
2918 		return 0;
2919 
2920 	pf->vf_configs[vf].trusted = enable;
2921 	rc = otx2_set_vf_permissions(pf, vf, OTX2_TRUSTED_VF);
2922 
2923 	if (rc) {
2924 		pf->vf_configs[vf].trusted = !enable;
2925 	} else {
2926 		netdev_info(pf->netdev, "VF %d is %strusted\n",
2927 			    vf, enable ? "" : "not ");
2928 		otx2_set_rx_mode(netdev);
2929 	}
2930 
2931 	return rc;
2932 }
2933 
2934 static const struct net_device_ops otx2_netdev_ops = {
2935 	.ndo_open		= otx2_open,
2936 	.ndo_stop		= otx2_stop,
2937 	.ndo_start_xmit		= otx2_xmit,
2938 	.ndo_select_queue	= otx2_select_queue,
2939 	.ndo_fix_features	= otx2_fix_features,
2940 	.ndo_set_mac_address    = otx2_set_mac_address,
2941 	.ndo_change_mtu		= otx2_change_mtu,
2942 	.ndo_set_rx_mode	= otx2_set_rx_mode,
2943 	.ndo_set_features	= otx2_set_features,
2944 	.ndo_tx_timeout		= otx2_tx_timeout,
2945 	.ndo_get_stats64	= otx2_get_stats64,
2946 	.ndo_eth_ioctl		= otx2_ioctl,
2947 	.ndo_set_vf_mac		= otx2_set_vf_mac,
2948 	.ndo_set_vf_vlan	= otx2_set_vf_vlan,
2949 	.ndo_get_vf_config	= otx2_get_vf_config,
2950 	.ndo_bpf		= otx2_xdp,
2951 	.ndo_xsk_wakeup		= otx2_xsk_wakeup,
2952 	.ndo_xdp_xmit           = otx2_xdp_xmit,
2953 	.ndo_setup_tc		= otx2_setup_tc,
2954 	.ndo_set_vf_trust	= otx2_ndo_set_vf_trust,
2955 };
2956 
2957 int otx2_wq_init(struct otx2_nic *pf)
2958 {
2959 	pf->otx2_wq = create_singlethread_workqueue("otx2_wq");
2960 	if (!pf->otx2_wq)
2961 		return -ENOMEM;
2962 
2963 	INIT_WORK(&pf->rx_mode_work, otx2_rx_mode_wrk_handler);
2964 	INIT_WORK(&pf->reset_task, otx2_reset_task);
2965 	return 0;
2966 }
2967 
2968 int otx2_check_pf_usable(struct otx2_nic *nic)
2969 {
2970 	u64 rev;
2971 
2972 	rev = otx2_read64(nic, RVU_PF_BLOCK_ADDRX_DISC(BLKADDR_RVUM));
2973 	rev = (rev >> 12) & 0xFF;
2974 	/* Check if AF has setup revision for RVUM block,
2975 	 * otherwise this driver probe should be deferred
2976 	 * until AF driver comes up.
2977 	 */
2978 	if (!rev) {
2979 		dev_warn(nic->dev,
2980 			 "AF is not initialized, deferring probe\n");
2981 		return -EPROBE_DEFER;
2982 	}
2983 	return 0;
2984 }
2985 
2986 int otx2_realloc_msix_vectors(struct otx2_nic *pf)
2987 {
2988 	struct otx2_hw *hw = &pf->hw;
2989 	int num_vec, err;
2990 
2991 	/* NPA interrupts are inot registered, so alloc only
2992 	 * upto NIX vector offset.
2993 	 */
2994 	num_vec = hw->nix_msixoff;
2995 	num_vec += NIX_LF_CINT_VEC_START + hw->max_queues;
2996 
2997 	otx2_disable_mbox_intr(pf);
2998 	pci_free_irq_vectors(hw->pdev);
2999 	err = pci_alloc_irq_vectors(hw->pdev, num_vec, num_vec, PCI_IRQ_MSIX);
3000 	if (err < 0) {
3001 		dev_err(pf->dev, "%s: Failed to realloc %d IRQ vectors\n",
3002 			__func__, num_vec);
3003 		return err;
3004 	}
3005 
3006 	return otx2_register_mbox_intr(pf, false);
3007 }
3008 EXPORT_SYMBOL(otx2_realloc_msix_vectors);
3009 
3010 static int otx2_sriov_vfcfg_init(struct otx2_nic *pf)
3011 {
3012 	int i;
3013 
3014 	pf->vf_configs = devm_kcalloc(pf->dev, pf->total_vfs,
3015 				      sizeof(struct otx2_vf_config),
3016 				      GFP_KERNEL);
3017 	if (!pf->vf_configs)
3018 		return -ENOMEM;
3019 
3020 	for (i = 0; i < pf->total_vfs; i++) {
3021 		pf->vf_configs[i].pf = pf;
3022 		pf->vf_configs[i].intf_down = true;
3023 		pf->vf_configs[i].trusted = false;
3024 		INIT_DELAYED_WORK(&pf->vf_configs[i].link_event_work,
3025 				  otx2_vf_link_event_task);
3026 	}
3027 
3028 	return 0;
3029 }
3030 
3031 static void otx2_sriov_vfcfg_cleanup(struct otx2_nic *pf)
3032 {
3033 	int i;
3034 
3035 	if (!pf->vf_configs)
3036 		return;
3037 
3038 	for (i = 0; i < pf->total_vfs; i++) {
3039 		cancel_delayed_work_sync(&pf->vf_configs[i].link_event_work);
3040 		otx2_set_vf_permissions(pf, i, OTX2_RESET_VF_PERM);
3041 	}
3042 }
3043 
3044 int otx2_init_rsrc(struct pci_dev *pdev, struct otx2_nic *pf)
3045 {
3046 	struct device *dev = &pdev->dev;
3047 	struct otx2_hw *hw = &pf->hw;
3048 	int num_vec, err;
3049 
3050 	num_vec = pci_msix_vec_count(pdev);
3051 	hw->irq_name = devm_kmalloc_array(&hw->pdev->dev, num_vec, NAME_SIZE,
3052 					  GFP_KERNEL);
3053 	if (!hw->irq_name)
3054 		return -ENOMEM;
3055 
3056 	hw->affinity_mask = devm_kcalloc(&hw->pdev->dev, num_vec,
3057 					 sizeof(cpumask_var_t), GFP_KERNEL);
3058 	if (!hw->affinity_mask)
3059 		return -ENOMEM;
3060 
3061 	/* Map CSRs */
3062 	pf->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
3063 	if (!pf->reg_base) {
3064 		dev_err(dev, "Unable to map physical function CSRs, aborting\n");
3065 		return -ENOMEM;
3066 	}
3067 
3068 	err = otx2_check_pf_usable(pf);
3069 	if (err)
3070 		return err;
3071 
3072 	if (!is_cn20k(pf->pdev))
3073 		err = pci_alloc_irq_vectors(hw->pdev, RVU_PF_INT_VEC_CNT,
3074 					    RVU_PF_INT_VEC_CNT, PCI_IRQ_MSIX);
3075 	else
3076 		err = pci_alloc_irq_vectors(hw->pdev, RVU_MBOX_PF_INT_VEC_CNT,
3077 					    RVU_MBOX_PF_INT_VEC_CNT,
3078 					    PCI_IRQ_MSIX);
3079 	if (err < 0) {
3080 		dev_err(dev, "%s: Failed to alloc %d IRQ vectors\n",
3081 			__func__, num_vec);
3082 		return err;
3083 	}
3084 
3085 	otx2_setup_dev_hw_settings(pf);
3086 
3087 	if (is_cn20k(pf->pdev))
3088 		cn20k_init(pf);
3089 	else
3090 		otx2_init_hw_ops(pf);
3091 
3092 	/* Init PF <=> AF mailbox stuff */
3093 	err = otx2_pfaf_mbox_init(pf);
3094 	if (err)
3095 		goto err_free_irq_vectors;
3096 
3097 	/* Register mailbox interrupt */
3098 	err = otx2_register_mbox_intr(pf, true);
3099 	if (err)
3100 		goto err_mbox_destroy;
3101 
3102 	/* Request AF to attach NPA and NIX LFs to this PF.
3103 	 * NIX and NPA LFs are needed for this PF to function as a NIC.
3104 	 */
3105 	err = otx2_attach_npa_nix(pf);
3106 	if (err)
3107 		goto err_disable_mbox_intr;
3108 
3109 	err = otx2_realloc_msix_vectors(pf);
3110 	if (err)
3111 		goto err_detach_rsrc;
3112 
3113 	err = cn10k_lmtst_init(pf);
3114 	if (err)
3115 		goto err_detach_rsrc;
3116 
3117 	return 0;
3118 
3119 err_detach_rsrc:
3120 	if (pf->hw.lmt_info)
3121 		free_percpu(pf->hw.lmt_info);
3122 	if (test_bit(CN10K_LMTST, &pf->hw.cap_flag))
3123 		qmem_free(pf->dev, pf->dync_lmt);
3124 	otx2_detach_resources(&pf->mbox);
3125 err_disable_mbox_intr:
3126 	otx2_disable_mbox_intr(pf);
3127 err_mbox_destroy:
3128 	otx2_pfaf_mbox_destroy(pf);
3129 err_free_irq_vectors:
3130 	pci_free_irq_vectors(hw->pdev);
3131 
3132 	return err;
3133 }
3134 EXPORT_SYMBOL(otx2_init_rsrc);
3135 
3136 static int otx2_probe(struct pci_dev *pdev, const struct pci_device_id *id)
3137 {
3138 	struct device *dev = &pdev->dev;
3139 	int err, qcount, qos_txqs;
3140 	struct net_device *netdev;
3141 	struct otx2_nic *pf;
3142 	struct otx2_hw *hw;
3143 
3144 	err = pcim_enable_device(pdev);
3145 	if (err) {
3146 		dev_err(dev, "Failed to enable PCI device\n");
3147 		return err;
3148 	}
3149 
3150 	err = pcim_request_all_regions(pdev, DRV_NAME);
3151 	if (err) {
3152 		dev_err(dev, "PCI request regions failed 0x%x\n", err);
3153 		return err;
3154 	}
3155 
3156 	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(48));
3157 	if (err) {
3158 		dev_err(dev, "DMA mask config failed, abort\n");
3159 		return err;
3160 	}
3161 
3162 	pci_set_master(pdev);
3163 
3164 	/* Set number of queues */
3165 	qcount = min_t(int, num_online_cpus(), OTX2_MAX_CQ_CNT);
3166 	qos_txqs = min_t(int, qcount, OTX2_QOS_MAX_LEAF_NODES);
3167 
3168 	netdev = alloc_etherdev_mqs(sizeof(*pf), qcount + qos_txqs, qcount);
3169 	if (!netdev)
3170 		return -ENOMEM;
3171 
3172 	pci_set_drvdata(pdev, netdev);
3173 	SET_NETDEV_DEV(netdev, &pdev->dev);
3174 	pf = netdev_priv(netdev);
3175 	pf->netdev = netdev;
3176 	pf->pdev = pdev;
3177 	pf->dev = dev;
3178 	pf->total_vfs = pci_sriov_get_totalvfs(pdev);
3179 	pf->flags |= OTX2_FLAG_INTF_DOWN;
3180 
3181 	hw = &pf->hw;
3182 	hw->pdev = pdev;
3183 	hw->rx_queues = qcount;
3184 	hw->tx_queues = qcount;
3185 	hw->non_qos_queues = qcount;
3186 	hw->max_queues = qcount;
3187 	hw->rbuf_len = OTX2_DEFAULT_RBUF_LEN;
3188 	/* Use CQE of 128 byte descriptor size by default */
3189 	hw->xqe_size = 128;
3190 
3191 	err = otx2_init_rsrc(pdev, pf);
3192 	if (err)
3193 		goto err_free_netdev;
3194 
3195 	err = otx2_set_real_num_queues(netdev, hw->tx_queues, hw->rx_queues);
3196 	if (err)
3197 		goto err_detach_rsrc;
3198 
3199 	/* Assign default mac address */
3200 	otx2_get_mac_from_af(netdev);
3201 
3202 	/* Don't check for error.  Proceed without ptp */
3203 	otx2_ptp_init(pf);
3204 
3205 	/* NPA's pool is a stack to which SW frees buffer pointers via Aura.
3206 	 * HW allocates buffer pointer from stack and uses it for DMA'ing
3207 	 * ingress packet. In some scenarios HW can free back allocated buffer
3208 	 * pointers to pool. This makes it impossible for SW to maintain a
3209 	 * parallel list where physical addresses of buffer pointers (IOVAs)
3210 	 * given to HW can be saved for later reference.
3211 	 *
3212 	 * So the only way to convert Rx packet's buffer address is to use
3213 	 * IOMMU's iova_to_phys() handler which translates the address by
3214 	 * walking through the translation tables.
3215 	 */
3216 	pf->iommu_domain = iommu_get_domain_for_dev(dev);
3217 
3218 	netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_IP_CSUM |
3219 			       NETIF_F_IPV6_CSUM | NETIF_F_RXHASH |
3220 			       NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
3221 			       NETIF_F_GSO_UDP_L4);
3222 	netdev->features |= netdev->hw_features;
3223 
3224 	err = otx2_mcam_flow_init(pf);
3225 	if (err)
3226 		goto err_ptp_destroy;
3227 
3228 	otx2_set_hw_capabilities(pf);
3229 
3230 	err = cn10k_mcs_init(pf);
3231 	if (err)
3232 		goto err_del_mcam_entries;
3233 
3234 	if (pf->flags & OTX2_FLAG_NTUPLE_SUPPORT)
3235 		netdev->hw_features |= NETIF_F_NTUPLE;
3236 
3237 	if (pf->flags & OTX2_FLAG_UCAST_FLTR_SUPPORT)
3238 		netdev->priv_flags |= IFF_UNICAST_FLT;
3239 
3240 	/* Support TSO on tag interface */
3241 	netdev->vlan_features |= netdev->features;
3242 	netdev->hw_features  |= NETIF_F_HW_VLAN_CTAG_TX |
3243 				NETIF_F_HW_VLAN_STAG_TX;
3244 	if (pf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
3245 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX |
3246 				       NETIF_F_HW_VLAN_STAG_RX;
3247 	netdev->features |= netdev->hw_features;
3248 
3249 	/* HW supports tc offload but mutually exclusive with n-tuple filters */
3250 	if (pf->flags & OTX2_FLAG_TC_FLOWER_SUPPORT)
3251 		netdev->hw_features |= NETIF_F_HW_TC;
3252 
3253 	netdev->hw_features |= NETIF_F_LOOPBACK | NETIF_F_RXALL;
3254 
3255 	netif_set_tso_max_segs(netdev, OTX2_MAX_GSO_SEGS);
3256 	netdev->watchdog_timeo = OTX2_TX_TIMEOUT;
3257 
3258 	netdev->netdev_ops = &otx2_netdev_ops;
3259 	netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT;
3260 
3261 	netdev->min_mtu = OTX2_MIN_MTU;
3262 	netdev->max_mtu = otx2_get_max_mtu(pf);
3263 	hw->max_mtu = netdev->max_mtu;
3264 
3265 	/* reset CGX/RPM MAC stats */
3266 	otx2_reset_mac_stats(pf);
3267 
3268 	err = cn10k_ipsec_init(netdev);
3269 	if (err)
3270 		goto err_mcs_free;
3271 
3272 	err = register_netdev(netdev);
3273 	if (err) {
3274 		dev_err(dev, "Failed to register netdevice\n");
3275 		goto err_ipsec_clean;
3276 	}
3277 
3278 	err = otx2_wq_init(pf);
3279 	if (err)
3280 		goto err_unreg_netdev;
3281 
3282 	otx2_set_ethtool_ops(netdev);
3283 
3284 	err = otx2_init_tc(pf);
3285 	if (err)
3286 		goto err_mcam_flow_del;
3287 
3288 	err = otx2_register_dl(pf);
3289 	if (err)
3290 		goto err_mcam_flow_del;
3291 
3292 	/* Initialize SR-IOV resources */
3293 	err = otx2_sriov_vfcfg_init(pf);
3294 	if (err)
3295 		goto err_pf_sriov_init;
3296 
3297 	/* Enable link notifications */
3298 	otx2_cgx_config_linkevents(pf, true);
3299 
3300 	pf->af_xdp_zc_qidx = bitmap_zalloc(qcount, GFP_KERNEL);
3301 	if (!pf->af_xdp_zc_qidx) {
3302 		err = -ENOMEM;
3303 		goto err_sriov_cleannup;
3304 	}
3305 
3306 #ifdef CONFIG_DCB
3307 	err = otx2_dcbnl_set_ops(netdev);
3308 	if (err)
3309 		goto err_free_zc_bmap;
3310 #endif
3311 
3312 	otx2_qos_init(pf, qos_txqs);
3313 
3314 	return 0;
3315 
3316 #ifdef CONFIG_DCB
3317 err_free_zc_bmap:
3318 	bitmap_free(pf->af_xdp_zc_qidx);
3319 #endif
3320 err_sriov_cleannup:
3321 	otx2_sriov_vfcfg_cleanup(pf);
3322 err_pf_sriov_init:
3323 	otx2_shutdown_tc(pf);
3324 err_mcam_flow_del:
3325 	otx2_mcam_flow_del(pf);
3326 err_unreg_netdev:
3327 	unregister_netdev(netdev);
3328 err_ipsec_clean:
3329 	cn10k_ipsec_clean(pf);
3330 err_mcs_free:
3331 	cn10k_mcs_free(pf);
3332 err_del_mcam_entries:
3333 	otx2_mcam_flow_del(pf);
3334 err_ptp_destroy:
3335 	otx2_ptp_destroy(pf);
3336 err_detach_rsrc:
3337 	if (pf->hw.lmt_info)
3338 		free_percpu(pf->hw.lmt_info);
3339 	if (test_bit(CN10K_LMTST, &pf->hw.cap_flag))
3340 		qmem_free(pf->dev, pf->dync_lmt);
3341 	otx2_detach_resources(&pf->mbox);
3342 	otx2_disable_mbox_intr(pf);
3343 	otx2_pfaf_mbox_destroy(pf);
3344 	pci_free_irq_vectors(hw->pdev);
3345 err_free_netdev:
3346 	pci_set_drvdata(pdev, NULL);
3347 	free_netdev(netdev);
3348 	return err;
3349 }
3350 
3351 static void otx2_vf_link_event_task(struct work_struct *work)
3352 {
3353 	struct otx2_vf_config *config;
3354 	struct cgx_link_info_msg *req;
3355 	struct mbox_msghdr *msghdr;
3356 	struct delayed_work *dwork;
3357 	struct otx2_nic *pf;
3358 	int vf_idx;
3359 
3360 	config = container_of(work, struct otx2_vf_config,
3361 			      link_event_work.work);
3362 	vf_idx = config - config->pf->vf_configs;
3363 	pf = config->pf;
3364 
3365 	if (config->intf_down)
3366 		return;
3367 
3368 	mutex_lock(&pf->mbox.lock);
3369 
3370 	dwork = &config->link_event_work;
3371 
3372 	if (!otx2_mbox_wait_for_zero(&pf->mbox_pfvf[0].mbox_up, vf_idx)) {
3373 		schedule_delayed_work(dwork, msecs_to_jiffies(100));
3374 		mutex_unlock(&pf->mbox.lock);
3375 		return;
3376 	}
3377 
3378 	msghdr = otx2_mbox_alloc_msg_rsp(&pf->mbox_pfvf[0].mbox_up, vf_idx,
3379 					 sizeof(*req), sizeof(struct msg_rsp));
3380 	if (!msghdr) {
3381 		dev_err(pf->dev, "Failed to create VF%d link event\n", vf_idx);
3382 		mutex_unlock(&pf->mbox.lock);
3383 		return;
3384 	}
3385 
3386 	req = (struct cgx_link_info_msg *)msghdr;
3387 	req->hdr.id = MBOX_MSG_CGX_LINK_EVENT;
3388 	req->hdr.sig = OTX2_MBOX_REQ_SIG;
3389 	req->hdr.pcifunc = pf->pcifunc;
3390 	memcpy(&req->link_info, &pf->linfo, sizeof(req->link_info));
3391 
3392 	otx2_mbox_wait_for_zero(&pf->mbox_pfvf[0].mbox_up, vf_idx);
3393 
3394 	otx2_sync_mbox_up_msg(&pf->mbox_pfvf[0], vf_idx);
3395 
3396 	mutex_unlock(&pf->mbox.lock);
3397 }
3398 
3399 static int otx2_sriov_enable(struct pci_dev *pdev, int numvfs)
3400 {
3401 	struct net_device *netdev = pci_get_drvdata(pdev);
3402 	struct otx2_nic *pf = netdev_priv(netdev);
3403 	int ret;
3404 
3405 	/* Init PF <=> VF mailbox stuff */
3406 	ret = otx2_pfvf_mbox_init(pf, numvfs);
3407 	if (ret)
3408 		return ret;
3409 
3410 	ret = otx2_register_pfvf_mbox_intr(pf, numvfs);
3411 	if (ret)
3412 		goto free_mbox;
3413 
3414 	ret = otx2_pf_flr_init(pf, numvfs);
3415 	if (ret)
3416 		goto free_intr;
3417 
3418 	ret = otx2_register_flr_me_intr(pf, numvfs);
3419 	if (ret)
3420 		goto free_flr;
3421 
3422 	ret = pci_enable_sriov(pdev, numvfs);
3423 	if (ret)
3424 		goto free_flr_intr;
3425 
3426 	return numvfs;
3427 free_flr_intr:
3428 	otx2_disable_flr_me_intr(pf);
3429 free_flr:
3430 	otx2_flr_wq_destroy(pf);
3431 free_intr:
3432 	otx2_disable_pfvf_mbox_intr(pf, numvfs);
3433 free_mbox:
3434 	otx2_pfvf_mbox_destroy(pf);
3435 	return ret;
3436 }
3437 
3438 static int otx2_sriov_disable(struct pci_dev *pdev)
3439 {
3440 	struct net_device *netdev = pci_get_drvdata(pdev);
3441 	struct otx2_nic *pf = netdev_priv(netdev);
3442 	int numvfs = pci_num_vf(pdev);
3443 
3444 	if (!numvfs)
3445 		return 0;
3446 
3447 	pci_disable_sriov(pdev);
3448 
3449 	otx2_disable_flr_me_intr(pf);
3450 	otx2_flr_wq_destroy(pf);
3451 	otx2_disable_pfvf_mbox_intr(pf, numvfs);
3452 	otx2_pfvf_mbox_destroy(pf);
3453 
3454 	return 0;
3455 }
3456 
3457 static int otx2_sriov_configure(struct pci_dev *pdev, int numvfs)
3458 {
3459 	if (numvfs == 0)
3460 		return otx2_sriov_disable(pdev);
3461 	else
3462 		return otx2_sriov_enable(pdev, numvfs);
3463 }
3464 
3465 static void otx2_ndc_sync(struct otx2_nic *pf)
3466 {
3467 	struct mbox *mbox = &pf->mbox;
3468 	struct ndc_sync_op *req;
3469 
3470 	mutex_lock(&mbox->lock);
3471 
3472 	req = otx2_mbox_alloc_msg_ndc_sync_op(mbox);
3473 	if (!req) {
3474 		mutex_unlock(&mbox->lock);
3475 		return;
3476 	}
3477 
3478 	req->nix_lf_tx_sync = 1;
3479 	req->nix_lf_rx_sync = 1;
3480 	req->npa_lf_sync = 1;
3481 
3482 	if (!otx2_sync_mbox_msg(mbox))
3483 		dev_err(pf->dev, "NDC sync operation failed\n");
3484 
3485 	mutex_unlock(&mbox->lock);
3486 }
3487 
3488 static void otx2_remove(struct pci_dev *pdev)
3489 {
3490 	struct net_device *netdev = pci_get_drvdata(pdev);
3491 	struct otx2_nic *pf;
3492 
3493 	if (!netdev)
3494 		return;
3495 
3496 	pf = netdev_priv(netdev);
3497 
3498 	pf->flags |= OTX2_FLAG_PF_SHUTDOWN;
3499 
3500 	if (pf->flags & OTX2_FLAG_TX_TSTAMP_ENABLED)
3501 		otx2_config_hw_tx_tstamp(pf, false);
3502 	if (pf->flags & OTX2_FLAG_RX_TSTAMP_ENABLED)
3503 		otx2_config_hw_rx_tstamp(pf, false);
3504 
3505 	/* Disable 802.3x pause frames */
3506 	if (pf->flags & OTX2_FLAG_RX_PAUSE_ENABLED ||
3507 	    (pf->flags & OTX2_FLAG_TX_PAUSE_ENABLED)) {
3508 		pf->flags &= ~OTX2_FLAG_RX_PAUSE_ENABLED;
3509 		pf->flags &= ~OTX2_FLAG_TX_PAUSE_ENABLED;
3510 		otx2_config_pause_frm(pf);
3511 	}
3512 
3513 #ifdef CONFIG_DCB
3514 	/* Disable PFC config */
3515 	if (pf->pfc_en) {
3516 		pf->pfc_en = 0;
3517 		otx2_config_priority_flow_ctrl(pf);
3518 	}
3519 #endif
3520 	cancel_work_sync(&pf->reset_task);
3521 	/* Disable link notifications */
3522 	otx2_cgx_config_linkevents(pf, false);
3523 
3524 	otx2_unregister_dl(pf);
3525 	unregister_netdev(netdev);
3526 	cn10k_ipsec_clean(pf);
3527 	cn10k_mcs_free(pf);
3528 	otx2_sriov_disable(pf->pdev);
3529 	otx2_sriov_vfcfg_cleanup(pf);
3530 	if (pf->otx2_wq)
3531 		destroy_workqueue(pf->otx2_wq);
3532 
3533 	otx2_ptp_destroy(pf);
3534 	otx2_mcam_flow_del(pf);
3535 	otx2_shutdown_tc(pf);
3536 	otx2_shutdown_qos(pf);
3537 	otx2_ndc_sync(pf);
3538 	otx2_detach_resources(&pf->mbox);
3539 	if (pf->hw.lmt_info)
3540 		free_percpu(pf->hw.lmt_info);
3541 	if (test_bit(CN10K_LMTST, &pf->hw.cap_flag))
3542 		qmem_free(pf->dev, pf->dync_lmt);
3543 	otx2_disable_mbox_intr(pf);
3544 	otx2_pfaf_mbox_destroy(pf);
3545 	pci_free_irq_vectors(pf->pdev);
3546 	pci_set_drvdata(pdev, NULL);
3547 	free_netdev(netdev);
3548 }
3549 
3550 static struct pci_driver otx2_pf_driver = {
3551 	.name = DRV_NAME,
3552 	.id_table = otx2_pf_id_table,
3553 	.probe = otx2_probe,
3554 	.shutdown = otx2_remove,
3555 	.remove = otx2_remove,
3556 	.sriov_configure = otx2_sriov_configure
3557 };
3558 
3559 static int __init otx2_rvupf_init_module(void)
3560 {
3561 	pr_info("%s: %s\n", DRV_NAME, DRV_STRING);
3562 
3563 	return pci_register_driver(&otx2_pf_driver);
3564 }
3565 
3566 static void __exit otx2_rvupf_cleanup_module(void)
3567 {
3568 	pci_unregister_driver(&otx2_pf_driver);
3569 }
3570 
3571 module_init(otx2_rvupf_init_module);
3572 module_exit(otx2_rvupf_cleanup_module);
3573