xref: /linux/drivers/net/ethernet/marvell/octeontx2/nic/otx2_pf.c (revision a056db30de92945ff8ee6033096678bfbae878e3)
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 	mbox_data = otx2_read64(pf, RVU_PF_PFAF_MBOX0);
1057 
1058 	if (mbox_data & MBOX_UP_MSG) {
1059 		mbox_data &= ~MBOX_UP_MSG;
1060 		otx2_write64(pf, RVU_PF_PFAF_MBOX0, mbox_data);
1061 
1062 		mbox = &mw->mbox_up;
1063 		mdev = &mbox->dev[0];
1064 		otx2_sync_mbox_bbuf(mbox, 0);
1065 
1066 		hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
1067 		if (hdr->num_msgs)
1068 			queue_work(pf->mbox_wq, &mw->mbox_up_wrk);
1069 
1070 		trace_otx2_msg_interrupt(pf->pdev, "UP message from AF to PF",
1071 					 BIT_ULL(0));
1072 
1073 		trace_otx2_msg_status(pf->pdev, "PF-AF up work queued(interrupt)",
1074 				      hdr->num_msgs);
1075 	}
1076 
1077 	if (mbox_data & MBOX_DOWN_MSG) {
1078 		mbox_data &= ~MBOX_DOWN_MSG;
1079 		otx2_write64(pf, RVU_PF_PFAF_MBOX0, mbox_data);
1080 
1081 		mbox = &mw->mbox;
1082 		mdev = &mbox->dev[0];
1083 		otx2_sync_mbox_bbuf(mbox, 0);
1084 
1085 		hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
1086 		if (hdr->num_msgs)
1087 			queue_work(pf->mbox_wq, &mw->mbox_wrk);
1088 
1089 		trace_otx2_msg_interrupt(pf->pdev, "DOWN reply from AF to PF",
1090 					 BIT_ULL(0));
1091 
1092 		trace_otx2_msg_status(pf->pdev, "PF-AF down work queued(interrupt)",
1093 				      hdr->num_msgs);
1094 	}
1095 
1096 	return IRQ_HANDLED;
1097 }
1098 
1099 void otx2_disable_mbox_intr(struct otx2_nic *pf)
1100 {
1101 	int vector;
1102 
1103 	/* Disable AF => PF mailbox IRQ */
1104 	if (!is_cn20k(pf->pdev)) {
1105 		vector = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_AFPF_MBOX);
1106 		otx2_write64(pf, RVU_PF_INT_ENA_W1C, BIT_ULL(0));
1107 	} else {
1108 		vector = pci_irq_vector(pf->pdev,
1109 					RVU_MBOX_PF_INT_VEC_AFPF_MBOX);
1110 		otx2_write64(pf, RVU_PF_INT_ENA_W1C,
1111 			     BIT_ULL(0) | BIT_ULL(1));
1112 	}
1113 	free_irq(vector, pf);
1114 }
1115 EXPORT_SYMBOL(otx2_disable_mbox_intr);
1116 
1117 int otx2_register_mbox_intr(struct otx2_nic *pf, bool probe_af)
1118 {
1119 	struct otx2_hw *hw = &pf->hw;
1120 	struct msg_req *req;
1121 	u64 mbox_int_mask;
1122 	char *irq_name;
1123 	int err;
1124 
1125 	mbox_int_mask = !is_cn20k(pf->pdev) ? BIT_ULL(0) :
1126 				BIT_ULL(0) | BIT_ULL(1);
1127 
1128 	/* Clear stale mailbox interrupt state before installing the handler. */
1129 	otx2_write64(pf, RVU_PF_INT, mbox_int_mask);
1130 
1131 	/* Register mailbox interrupt handler */
1132 	if (!is_cn20k(pf->pdev)) {
1133 		irq_name = &hw->irq_name[RVU_PF_INT_VEC_AFPF_MBOX * NAME_SIZE];
1134 		snprintf(irq_name, NAME_SIZE, "RVUPF%d AFPF Mbox",
1135 			 rvu_get_pf(pf->pdev, pf->pcifunc));
1136 		err = request_irq(pci_irq_vector
1137 				  (pf->pdev, RVU_PF_INT_VEC_AFPF_MBOX),
1138 				  pf->hw_ops->pfaf_mbox_intr_handler,
1139 				  0, irq_name, pf);
1140 	} else {
1141 		irq_name = &hw->irq_name[RVU_MBOX_PF_INT_VEC_AFPF_MBOX *
1142 						NAME_SIZE];
1143 		snprintf(irq_name, NAME_SIZE, "RVUPF%d AFPF Mbox",
1144 			 rvu_get_pf(pf->pdev, pf->pcifunc));
1145 		err = request_irq(pci_irq_vector
1146 				  (pf->pdev, RVU_MBOX_PF_INT_VEC_AFPF_MBOX),
1147 				  pf->hw_ops->pfaf_mbox_intr_handler,
1148 				  0, irq_name, pf);
1149 	}
1150 	if (err) {
1151 		dev_err(pf->dev,
1152 			"RVUPF: IRQ registration failed for PFAF mbox irq\n");
1153 		return err;
1154 	}
1155 
1156 	/* Enable mailbox interrupt for msgs coming from AF. */
1157 	otx2_write64(pf, RVU_PF_INT_ENA_W1S, mbox_int_mask);
1158 
1159 	if (!probe_af)
1160 		return 0;
1161 
1162 	/* Check mailbox communication with AF */
1163 	req = otx2_mbox_alloc_msg_ready(&pf->mbox);
1164 	if (!req) {
1165 		otx2_disable_mbox_intr(pf);
1166 		return -ENOMEM;
1167 	}
1168 	err = otx2_sync_mbox_msg(&pf->mbox);
1169 	if (err) {
1170 		dev_warn(pf->dev,
1171 			 "AF not responding to mailbox, deferring probe\n");
1172 		otx2_disable_mbox_intr(pf);
1173 		return -EPROBE_DEFER;
1174 	}
1175 
1176 	return 0;
1177 }
1178 
1179 void otx2_pfaf_mbox_destroy(struct otx2_nic *pf)
1180 {
1181 	struct mbox *mbox = &pf->mbox;
1182 
1183 	if (pf->mbox_wq) {
1184 		destroy_workqueue(pf->mbox_wq);
1185 		pf->mbox_wq = NULL;
1186 	}
1187 
1188 	if (mbox->mbox.hwbase && !is_cn20k(pf->pdev))
1189 		iounmap((void __iomem *)mbox->mbox.hwbase);
1190 
1191 	otx2_mbox_destroy(&mbox->mbox);
1192 	otx2_mbox_destroy(&mbox->mbox_up);
1193 }
1194 EXPORT_SYMBOL(otx2_pfaf_mbox_destroy);
1195 
1196 int otx2_pfaf_mbox_init(struct otx2_nic *pf)
1197 {
1198 	struct mbox *mbox = &pf->mbox;
1199 	void __iomem *hwbase;
1200 	int err;
1201 
1202 	mbox->pfvf = pf;
1203 	pf->mbox_wq = alloc_ordered_workqueue("otx2_pfaf_mailbox",
1204 					      WQ_HIGHPRI | WQ_MEM_RECLAIM);
1205 	if (!pf->mbox_wq)
1206 		return -ENOMEM;
1207 
1208 	/* For CN20K, AF allocates mbox memory in DRAM and writes PF
1209 	 * regions/offsets in RVU_MBOX_AF_PFX_ADDR, the RVU_PFX_FUNC_PFAF_MBOX
1210 	 * gives the aliased address to access AF/PF mailbox regions.
1211 	 */
1212 	if (is_cn20k(pf->pdev))
1213 		hwbase = pf->reg_base + RVU_PFX_FUNC_PFAF_MBOX +
1214 			((u64)BLKADDR_MBOX << RVU_FUNC_BLKADDR_SHIFT);
1215 	else
1216 		/* Mailbox is a reserved memory (in RAM) region shared between
1217 		 * admin function (i.e AF) and this PF, shouldn't be mapped as
1218 		 * device memory to allow unaligned accesses.
1219 		 */
1220 		hwbase = ioremap_wc(pci_resource_start
1221 				    (pf->pdev, PCI_MBOX_BAR_NUM), MBOX_SIZE);
1222 	if (!hwbase) {
1223 		dev_err(pf->dev, "Unable to map PFAF mailbox region\n");
1224 		err = -ENOMEM;
1225 		goto exit;
1226 	}
1227 
1228 	err = otx2_mbox_init(&mbox->mbox, hwbase, pf->pdev, pf->reg_base,
1229 			     MBOX_DIR_PFAF, 1);
1230 	if (err)
1231 		goto exit;
1232 
1233 	err = otx2_mbox_init(&mbox->mbox_up, hwbase, pf->pdev, pf->reg_base,
1234 			     MBOX_DIR_PFAF_UP, 1);
1235 	if (err)
1236 		goto exit;
1237 
1238 	err = otx2_mbox_bbuf_init(mbox, pf->pdev);
1239 	if (err)
1240 		goto exit;
1241 
1242 	INIT_WORK(&mbox->mbox_wrk, otx2_pfaf_mbox_handler);
1243 	INIT_WORK(&mbox->mbox_up_wrk, otx2_pfaf_mbox_up_handler);
1244 	mutex_init(&mbox->lock);
1245 
1246 	return 0;
1247 exit:
1248 	otx2_pfaf_mbox_destroy(pf);
1249 	return err;
1250 }
1251 
1252 static int otx2_cgx_config_linkevents(struct otx2_nic *pf, bool enable)
1253 {
1254 	struct msg_req *msg;
1255 	int err;
1256 
1257 	mutex_lock(&pf->mbox.lock);
1258 	if (enable)
1259 		msg = otx2_mbox_alloc_msg_cgx_start_linkevents(&pf->mbox);
1260 	else
1261 		msg = otx2_mbox_alloc_msg_cgx_stop_linkevents(&pf->mbox);
1262 
1263 	if (!msg) {
1264 		mutex_unlock(&pf->mbox.lock);
1265 		return -ENOMEM;
1266 	}
1267 
1268 	err = otx2_sync_mbox_msg(&pf->mbox);
1269 	mutex_unlock(&pf->mbox.lock);
1270 	return err;
1271 }
1272 
1273 int otx2_reset_mac_stats(struct otx2_nic *pfvf)
1274 {
1275 	struct msg_req *req;
1276 	int err;
1277 
1278 	mutex_lock(&pfvf->mbox.lock);
1279 	req = otx2_mbox_alloc_msg_cgx_stats_rst(&pfvf->mbox);
1280 	if (!req) {
1281 		mutex_unlock(&pfvf->mbox.lock);
1282 		return -ENOMEM;
1283 	}
1284 
1285 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1286 	mutex_unlock(&pfvf->mbox.lock);
1287 	return err;
1288 }
1289 
1290 static int otx2_cgx_config_loopback(struct otx2_nic *pf, bool enable)
1291 {
1292 	struct msg_req *msg;
1293 	int err;
1294 
1295 	if (enable && !bitmap_empty(pf->flow_cfg->dmacflt_bmap,
1296 				    pf->flow_cfg->dmacflt_max_flows))
1297 		netdev_warn(pf->netdev,
1298 			    "CGX/RPM internal loopback might not work as DMAC filters are active\n");
1299 
1300 	mutex_lock(&pf->mbox.lock);
1301 	if (enable)
1302 		msg = otx2_mbox_alloc_msg_cgx_intlbk_enable(&pf->mbox);
1303 	else
1304 		msg = otx2_mbox_alloc_msg_cgx_intlbk_disable(&pf->mbox);
1305 
1306 	if (!msg) {
1307 		mutex_unlock(&pf->mbox.lock);
1308 		return -ENOMEM;
1309 	}
1310 
1311 	err = otx2_sync_mbox_msg(&pf->mbox);
1312 	mutex_unlock(&pf->mbox.lock);
1313 	return err;
1314 }
1315 
1316 int otx2_set_real_num_queues(struct net_device *netdev,
1317 			     int tx_queues, int rx_queues)
1318 {
1319 	int err;
1320 
1321 	err = netif_set_real_num_tx_queues(netdev, tx_queues);
1322 	if (err) {
1323 		netdev_err(netdev,
1324 			   "Failed to set no of Tx queues: %d\n", tx_queues);
1325 		return err;
1326 	}
1327 
1328 	err = netif_set_real_num_rx_queues(netdev, rx_queues);
1329 	if (err)
1330 		netdev_err(netdev,
1331 			   "Failed to set no of Rx queues: %d\n", rx_queues);
1332 	return err;
1333 }
1334 EXPORT_SYMBOL(otx2_set_real_num_queues);
1335 
1336 static char *nix_sqoperr_e_str[NIX_SQOPERR_MAX] = {
1337 	"NIX_SQOPERR_OOR",
1338 	"NIX_SQOPERR_CTX_FAULT",
1339 	"NIX_SQOPERR_CTX_POISON",
1340 	"NIX_SQOPERR_DISABLED",
1341 	"NIX_SQOPERR_SIZE_ERR",
1342 	"NIX_SQOPERR_OFLOW",
1343 	"NIX_SQOPERR_SQB_NULL",
1344 	"NIX_SQOPERR_SQB_FAULT",
1345 	"NIX_SQOPERR_SQE_SZ_ZERO",
1346 };
1347 
1348 static char *nix_mnqerr_e_str[NIX_MNQERR_MAX] = {
1349 	"NIX_MNQERR_SQ_CTX_FAULT",
1350 	"NIX_MNQERR_SQ_CTX_POISON",
1351 	"NIX_MNQERR_SQB_FAULT",
1352 	"NIX_MNQERR_SQB_POISON",
1353 	"NIX_MNQERR_TOTAL_ERR",
1354 	"NIX_MNQERR_LSO_ERR",
1355 	"NIX_MNQERR_CQ_QUERY_ERR",
1356 	"NIX_MNQERR_MAX_SQE_SIZE_ERR",
1357 	"NIX_MNQERR_MAXLEN_ERR",
1358 	"NIX_MNQERR_SQE_SIZEM1_ZERO",
1359 };
1360 
1361 static char *nix_snd_status_e_str[NIX_SND_STATUS_MAX] =  {
1362 	[NIX_SND_STATUS_GOOD] = "NIX_SND_STATUS_GOOD",
1363 	[NIX_SND_STATUS_SQ_CTX_FAULT] = "NIX_SND_STATUS_SQ_CTX_FAULT",
1364 	[NIX_SND_STATUS_SQ_CTX_POISON] = "NIX_SND_STATUS_SQ_CTX_POISON",
1365 	[NIX_SND_STATUS_SQB_FAULT] = "NIX_SND_STATUS_SQB_FAULT",
1366 	[NIX_SND_STATUS_SQB_POISON] = "NIX_SND_STATUS_SQB_POISON",
1367 	[NIX_SND_STATUS_HDR_ERR] = "NIX_SND_STATUS_HDR_ERR",
1368 	[NIX_SND_STATUS_EXT_ERR] = "NIX_SND_STATUS_EXT_ERR",
1369 	[NIX_SND_STATUS_JUMP_FAULT] = "NIX_SND_STATUS_JUMP_FAULT",
1370 	[NIX_SND_STATUS_JUMP_POISON] = "NIX_SND_STATUS_JUMP_POISON",
1371 	[NIX_SND_STATUS_CRC_ERR] = "NIX_SND_STATUS_CRC_ERR",
1372 	[NIX_SND_STATUS_IMM_ERR] = "NIX_SND_STATUS_IMM_ERR",
1373 	[NIX_SND_STATUS_SG_ERR] = "NIX_SND_STATUS_SG_ERR",
1374 	[NIX_SND_STATUS_MEM_ERR] = "NIX_SND_STATUS_MEM_ERR",
1375 	[NIX_SND_STATUS_INVALID_SUBDC] = "NIX_SND_STATUS_INVALID_SUBDC",
1376 	[NIX_SND_STATUS_SUBDC_ORDER_ERR] = "NIX_SND_STATUS_SUBDC_ORDER_ERR",
1377 	[NIX_SND_STATUS_DATA_FAULT] = "NIX_SND_STATUS_DATA_FAULT",
1378 	[NIX_SND_STATUS_DATA_POISON] = "NIX_SND_STATUS_DATA_POISON",
1379 	[NIX_SND_STATUS_NPC_DROP_ACTION] = "NIX_SND_STATUS_NPC_DROP_ACTION",
1380 	[NIX_SND_STATUS_LOCK_VIOL] = "NIX_SND_STATUS_LOCK_VIOL",
1381 	[NIX_SND_STATUS_NPC_UCAST_CHAN_ERR] = "NIX_SND_STAT_NPC_UCAST_CHAN_ERR",
1382 	[NIX_SND_STATUS_NPC_MCAST_CHAN_ERR] = "NIX_SND_STAT_NPC_MCAST_CHAN_ERR",
1383 	[NIX_SND_STATUS_NPC_MCAST_ABORT] = "NIX_SND_STATUS_NPC_MCAST_ABORT",
1384 	[NIX_SND_STATUS_NPC_VTAG_PTR_ERR] = "NIX_SND_STATUS_NPC_VTAG_PTR_ERR",
1385 	[NIX_SND_STATUS_NPC_VTAG_SIZE_ERR] = "NIX_SND_STATUS_NPC_VTAG_SIZE_ERR",
1386 	[NIX_SND_STATUS_SEND_MEM_FAULT] = "NIX_SND_STATUS_SEND_MEM_FAULT",
1387 	[NIX_SND_STATUS_SEND_STATS_ERR] = "NIX_SND_STATUS_SEND_STATS_ERR",
1388 };
1389 
1390 static irqreturn_t otx2_q_intr_handler(int irq, void *data)
1391 {
1392 	struct otx2_nic *pf = data;
1393 	struct otx2_snd_queue *sq;
1394 	void __iomem *ptr;
1395 	u64 val, qidx = 0;
1396 
1397 	/* CQ */
1398 	for (qidx = 0; qidx < pf->qset.cq_cnt; qidx++) {
1399 		ptr = otx2_get_regaddr(pf, NIX_LF_CQ_OP_INT);
1400 		val = otx2_atomic64_add((qidx << 44), ptr);
1401 
1402 		otx2_write64(pf, NIX_LF_CQ_OP_INT, (qidx << 44) |
1403 			     (val & NIX_CQERRINT_BITS));
1404 		if (!(val & (NIX_CQERRINT_BITS | BIT_ULL(42))))
1405 			continue;
1406 
1407 		if (val & BIT_ULL(42)) {
1408 			netdev_err(pf->netdev,
1409 				   "CQ%lld: error reading NIX_LF_CQ_OP_INT, NIX_LF_ERR_INT 0x%llx\n",
1410 				   qidx, otx2_read64(pf, NIX_LF_ERR_INT));
1411 		} else {
1412 			if (val & BIT_ULL(NIX_CQERRINT_DOOR_ERR))
1413 				netdev_err(pf->netdev, "CQ%lld: Doorbell error",
1414 					   qidx);
1415 			if (val & BIT_ULL(NIX_CQERRINT_CQE_FAULT))
1416 				netdev_err(pf->netdev,
1417 					   "CQ%lld: Memory fault on CQE write to LLC/DRAM",
1418 					   qidx);
1419 		}
1420 
1421 		schedule_work(&pf->reset_task);
1422 	}
1423 
1424 	/* SQ */
1425 	for (qidx = 0; qidx < otx2_get_total_tx_queues(pf); qidx++) {
1426 		u64 sq_op_err_dbg, mnq_err_dbg, snd_err_dbg;
1427 		u8 sq_op_err_code, mnq_err_code, snd_err_code;
1428 
1429 		sq = &pf->qset.sq[qidx];
1430 		if (!sq->sqb_ptrs)
1431 			continue;
1432 
1433 		/* Below debug registers captures first errors corresponding to
1434 		 * those registers. We don't have to check against SQ qid as
1435 		 * these are fatal errors.
1436 		 */
1437 
1438 		ptr = otx2_get_regaddr(pf, NIX_LF_SQ_OP_INT);
1439 		val = otx2_atomic64_add((qidx << 44), ptr);
1440 		otx2_write64(pf, NIX_LF_SQ_OP_INT, (qidx << 44) |
1441 			     (val & NIX_SQINT_BITS));
1442 
1443 		if (val & BIT_ULL(42)) {
1444 			netdev_err(pf->netdev,
1445 				   "SQ%lld: error reading NIX_LF_SQ_OP_INT, NIX_LF_ERR_INT 0x%llx\n",
1446 				   qidx, otx2_read64(pf, NIX_LF_ERR_INT));
1447 			goto done;
1448 		}
1449 
1450 		sq_op_err_dbg = otx2_read64(pf, NIX_LF_SQ_OP_ERR_DBG);
1451 		if (!(sq_op_err_dbg & BIT(44)))
1452 			goto chk_mnq_err_dbg;
1453 
1454 		sq_op_err_code = FIELD_GET(GENMASK(7, 0), sq_op_err_dbg);
1455 		netdev_err(pf->netdev,
1456 			   "SQ%lld: NIX_LF_SQ_OP_ERR_DBG(0x%llx)  err=%s(%#x)\n",
1457 			   qidx, sq_op_err_dbg,
1458 			   nix_sqoperr_e_str[sq_op_err_code],
1459 			   sq_op_err_code);
1460 
1461 		otx2_write64(pf, NIX_LF_SQ_OP_ERR_DBG, BIT_ULL(44));
1462 
1463 		if (sq_op_err_code == NIX_SQOPERR_SQB_NULL)
1464 			goto chk_mnq_err_dbg;
1465 
1466 		/* Err is not NIX_SQOPERR_SQB_NULL, call aq function to read SQ structure.
1467 		 * TODO: But we are in irq context. How to call mbox functions which does sleep
1468 		 */
1469 
1470 chk_mnq_err_dbg:
1471 		mnq_err_dbg = otx2_read64(pf, NIX_LF_MNQ_ERR_DBG);
1472 		if (!(mnq_err_dbg & BIT(44)))
1473 			goto chk_snd_err_dbg;
1474 
1475 		mnq_err_code = FIELD_GET(GENMASK(7, 0), mnq_err_dbg);
1476 		netdev_err(pf->netdev,
1477 			   "SQ%lld: NIX_LF_MNQ_ERR_DBG(0x%llx)  err=%s(%#x)\n",
1478 			   qidx, mnq_err_dbg,  nix_mnqerr_e_str[mnq_err_code],
1479 			   mnq_err_code);
1480 		otx2_write64(pf, NIX_LF_MNQ_ERR_DBG, BIT_ULL(44));
1481 
1482 chk_snd_err_dbg:
1483 		snd_err_dbg = otx2_read64(pf, NIX_LF_SEND_ERR_DBG);
1484 		if (snd_err_dbg & BIT(44)) {
1485 			snd_err_code = FIELD_GET(GENMASK(7, 0), snd_err_dbg);
1486 			netdev_err(pf->netdev,
1487 				   "SQ%lld: NIX_LF_SND_ERR_DBG:0x%llx err=%s(%#x)\n",
1488 				   qidx, snd_err_dbg,
1489 				   nix_snd_status_e_str[snd_err_code],
1490 				   snd_err_code);
1491 			otx2_write64(pf, NIX_LF_SEND_ERR_DBG, BIT_ULL(44));
1492 		}
1493 
1494 done:
1495 		/* Print values and reset */
1496 		if (val & BIT_ULL(NIX_SQINT_SQB_ALLOC_FAIL))
1497 			netdev_err(pf->netdev, "SQ%lld: SQB allocation failed",
1498 				   qidx);
1499 
1500 		schedule_work(&pf->reset_task);
1501 	}
1502 
1503 	return IRQ_HANDLED;
1504 }
1505 
1506 irqreturn_t otx2_cq_intr_handler(int irq, void *cq_irq)
1507 {
1508 	struct otx2_cq_poll *cq_poll = (struct otx2_cq_poll *)cq_irq;
1509 	struct otx2_nic *pf = (struct otx2_nic *)cq_poll->dev;
1510 	int qidx = cq_poll->cint_idx;
1511 
1512 	/* Disable interrupts.
1513 	 *
1514 	 * Completion interrupts behave in a level-triggered interrupt
1515 	 * fashion, and hence have to be cleared only after it is serviced.
1516 	 */
1517 	otx2_write64(pf, NIX_LF_CINTX_ENA_W1C(qidx), BIT_ULL(0));
1518 
1519 	/* Schedule NAPI */
1520 	pf->napi_events++;
1521 	napi_schedule_irqoff(&cq_poll->napi);
1522 
1523 	return IRQ_HANDLED;
1524 }
1525 EXPORT_SYMBOL(otx2_cq_intr_handler);
1526 
1527 void otx2_disable_napi(struct otx2_nic *pf)
1528 {
1529 	struct otx2_qset *qset = &pf->qset;
1530 	struct otx2_cq_poll *cq_poll;
1531 	struct work_struct *work;
1532 	int qidx;
1533 
1534 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
1535 		cq_poll = &qset->napi[qidx];
1536 		work = &cq_poll->dim.work;
1537 		if (work->func)
1538 			cancel_work_sync(work);
1539 		napi_disable(&cq_poll->napi);
1540 		netif_napi_del(&cq_poll->napi);
1541 	}
1542 }
1543 EXPORT_SYMBOL(otx2_disable_napi);
1544 
1545 static void otx2_free_cq_res(struct otx2_nic *pf)
1546 {
1547 	struct otx2_qset *qset = &pf->qset;
1548 	struct otx2_cq_queue *cq;
1549 	int qidx;
1550 
1551 	/* Disable CQs */
1552 	otx2_ctx_disable(&pf->mbox, NIX_AQ_CTYPE_CQ, false);
1553 	for (qidx = 0; qidx < qset->cq_cnt; qidx++) {
1554 		cq = &qset->cq[qidx];
1555 		qmem_free(pf->dev, cq->cqe);
1556 	}
1557 }
1558 
1559 static void otx2_free_sq_res(struct otx2_nic *pf)
1560 {
1561 	struct otx2_qset *qset = &pf->qset;
1562 	struct otx2_snd_queue *sq;
1563 	int qidx;
1564 
1565 	/* Disable SQs */
1566 	otx2_ctx_disable(&pf->mbox, NIX_AQ_CTYPE_SQ, false);
1567 	/* Free SQB pointers */
1568 	otx2_sq_free_sqbs(pf);
1569 	for (qidx = 0; qidx < otx2_get_total_tx_queues(pf); qidx++) {
1570 		sq = &qset->sq[qidx];
1571 		/* Skip freeing Qos queues if they are not initialized */
1572 		if (!sq->sqe)
1573 			continue;
1574 		qmem_free(pf->dev, sq->sqe);
1575 		qmem_free(pf->dev, sq->sqe_ring);
1576 		qmem_free(pf->dev, sq->cpt_resp);
1577 		qmem_free(pf->dev, sq->tso_hdrs);
1578 		qmem_free(pf->dev, sq->timestamps);
1579 		kfree(sq->sg);
1580 		kfree(sq->sqb_ptrs);
1581 	}
1582 }
1583 
1584 static int otx2_get_rbuf_size(struct otx2_nic *pf, int mtu)
1585 {
1586 	int frame_size;
1587 	int total_size;
1588 	int rbuf_size;
1589 
1590 	if (pf->hw.rbuf_len)
1591 		return ALIGN(pf->hw.rbuf_len, OTX2_ALIGN) + OTX2_HEAD_ROOM;
1592 
1593 	/* The data transferred by NIX to memory consists of actual packet
1594 	 * plus additional data which has timestamp and/or EDSA/HIGIG2
1595 	 * headers if interface is configured in corresponding modes.
1596 	 * NIX transfers entire data using 6 segments/buffers and writes
1597 	 * a CQE_RX descriptor with those segment addresses. First segment
1598 	 * has additional data prepended to packet. Also software omits a
1599 	 * headroom of 128 bytes in each segment. Hence the total size of
1600 	 * memory needed to receive a packet with 'mtu' is:
1601 	 * frame size =  mtu + additional data;
1602 	 * memory = frame_size + headroom * 6;
1603 	 * each receive buffer size = memory / 6;
1604 	 */
1605 	frame_size = mtu + OTX2_ETH_HLEN + OTX2_HW_TIMESTAMP_LEN;
1606 	total_size = frame_size + OTX2_HEAD_ROOM * 6;
1607 	rbuf_size = total_size / 6;
1608 
1609 	return ALIGN(rbuf_size, 2048);
1610 }
1611 
1612 int otx2_init_hw_resources(struct otx2_nic *pf)
1613 {
1614 	struct nix_lf_free_req *free_req;
1615 	struct mbox *mbox = &pf->mbox;
1616 	struct otx2_hw *hw = &pf->hw;
1617 	struct msg_req *req;
1618 	int err = 0, lvl;
1619 
1620 	/* Set required NPA LF's pool counts
1621 	 * Auras and Pools are used in a 1:1 mapping,
1622 	 * so, aura count = pool count.
1623 	 */
1624 	hw->rqpool_cnt = hw->rx_queues;
1625 	hw->sqpool_cnt = otx2_get_total_tx_queues(pf);
1626 	hw->pool_cnt = hw->rqpool_cnt + hw->sqpool_cnt;
1627 
1628 	if (!otx2_rep_dev(pf->pdev)) {
1629 		/* Maximum hardware supported transmit length */
1630 		pf->tx_max_pktlen = pf->netdev->max_mtu + OTX2_ETH_HLEN;
1631 		pf->rbsize = otx2_get_rbuf_size(pf, pf->netdev->mtu);
1632 	}
1633 
1634 	mutex_lock(&mbox->lock);
1635 	/* NPA init */
1636 	err = otx2_config_npa(pf);
1637 	if (err)
1638 		goto exit;
1639 
1640 	/* NIX init */
1641 	err = otx2_config_nix(pf);
1642 	if (err)
1643 		goto err_free_npa_lf;
1644 
1645 	/* Default disable backpressure on NIX-CPT */
1646 	otx2_nix_cpt_config_bp(pf, false);
1647 
1648 	/* Enable backpressure for CGX mapped PF/VFs */
1649 	if (!is_otx2_lbkvf(pf->pdev))
1650 		otx2_nix_config_bp(pf, true);
1651 
1652 	/* Init Auras and pools used by NIX RQ, for free buffer ptrs */
1653 	err = otx2_rq_aura_pool_init(pf);
1654 	if (err) {
1655 		mutex_unlock(&mbox->lock);
1656 		goto err_free_nix_lf;
1657 	}
1658 	/* Init Auras and pools used by NIX SQ, for queueing SQEs */
1659 	err = otx2_sq_aura_pool_init(pf);
1660 	if (err) {
1661 		mutex_unlock(&mbox->lock);
1662 		goto err_free_rq_ptrs;
1663 	}
1664 
1665 	err = otx2_txsch_alloc(pf);
1666 	if (err) {
1667 		mutex_unlock(&mbox->lock);
1668 		goto err_free_sq_ptrs;
1669 	}
1670 
1671 #ifdef CONFIG_DCB
1672 	if (pf->pfc_en) {
1673 		err = otx2_pfc_txschq_alloc(pf);
1674 		if (err) {
1675 			mutex_unlock(&mbox->lock);
1676 			goto err_free_sq_ptrs;
1677 		}
1678 	}
1679 #endif
1680 
1681 	err = otx2_config_nix_queues(pf);
1682 	if (err) {
1683 		mutex_unlock(&mbox->lock);
1684 		goto err_free_txsch;
1685 	}
1686 
1687 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
1688 		int idx;
1689 
1690 		for (idx = 0; idx < pf->hw.txschq_cnt[lvl]; idx++) {
1691 			err = otx2_txschq_config(pf, lvl, idx, false);
1692 			if (err) {
1693 				dev_err(pf->dev, "Failed to config TXSCH\n");
1694 				mutex_unlock(&mbox->lock);
1695 				goto err_free_nix_queues;
1696 			}
1697 		}
1698 	}
1699 
1700 #ifdef CONFIG_DCB
1701 	if (pf->pfc_en) {
1702 		err = otx2_pfc_txschq_config(pf);
1703 		if (err) {
1704 			mutex_unlock(&mbox->lock);
1705 			goto err_free_nix_queues;
1706 		}
1707 	}
1708 #endif
1709 
1710 	mutex_unlock(&mbox->lock);
1711 	return err;
1712 
1713 err_free_nix_queues:
1714 	otx2_free_sq_res(pf);
1715 	otx2_free_cq_res(pf);
1716 	otx2_ctx_disable(mbox, NIX_AQ_CTYPE_RQ, false);
1717 err_free_txsch:
1718 	otx2_txschq_stop(pf);
1719 err_free_sq_ptrs:
1720 	otx2_sq_free_sqbs(pf);
1721 err_free_rq_ptrs:
1722 	otx2_free_aura_ptr(pf, AURA_NIX_RQ);
1723 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_POOL, true);
1724 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_AURA, true);
1725 	otx2_aura_pool_free(pf);
1726 err_free_nix_lf:
1727 	mutex_lock(&mbox->lock);
1728 	free_req = otx2_mbox_alloc_msg_nix_lf_free(mbox);
1729 	if (free_req) {
1730 		free_req->flags = NIX_LF_DISABLE_FLOWS | NIX_LF_DONT_FREE_DFT_IDXS;
1731 		if (otx2_sync_mbox_msg(mbox))
1732 			dev_err(pf->dev, "%s failed to free nixlf\n", __func__);
1733 	}
1734 err_free_npa_lf:
1735 	/* Reset NPA LF */
1736 	req = otx2_mbox_alloc_msg_npa_lf_free(mbox);
1737 	if (req) {
1738 		if (otx2_sync_mbox_msg(mbox))
1739 			dev_err(pf->dev, "%s failed to free npalf\n", __func__);
1740 	}
1741 exit:
1742 	mutex_unlock(&mbox->lock);
1743 	return err;
1744 }
1745 EXPORT_SYMBOL(otx2_init_hw_resources);
1746 
1747 void otx2_free_hw_resources(struct otx2_nic *pf)
1748 {
1749 	struct otx2_qset *qset = &pf->qset;
1750 	struct nix_lf_free_req *free_req;
1751 	struct mbox *mbox = &pf->mbox;
1752 	struct otx2_cq_queue *cq;
1753 	struct msg_req *req;
1754 	int qidx;
1755 
1756 	/* Ensure all SQE are processed */
1757 	otx2_sqb_flush(pf);
1758 
1759 	/* Stop transmission */
1760 	otx2_txschq_stop(pf);
1761 
1762 #ifdef CONFIG_DCB
1763 	if (pf->pfc_en)
1764 		otx2_pfc_txschq_stop(pf);
1765 #endif
1766 
1767 	if (!otx2_rep_dev(pf->pdev))
1768 		otx2_clean_qos_queues(pf);
1769 
1770 	mutex_lock(&mbox->lock);
1771 	/* Disable backpressure */
1772 	if (!is_otx2_lbkvf(pf->pdev))
1773 		otx2_nix_config_bp(pf, false);
1774 	mutex_unlock(&mbox->lock);
1775 
1776 	/* Disable RQs */
1777 	otx2_ctx_disable(mbox, NIX_AQ_CTYPE_RQ, false);
1778 
1779 	/*Dequeue all CQEs */
1780 	for (qidx = 0; qidx < qset->cq_cnt; qidx++) {
1781 		cq = &qset->cq[qidx];
1782 		if (cq->cq_type == CQ_RX)
1783 			otx2_cleanup_rx_cqes(pf, cq, qidx);
1784 		else
1785 			otx2_cleanup_tx_cqes(pf, cq);
1786 	}
1787 	otx2_free_pending_sqe(pf);
1788 
1789 	otx2_free_sq_res(pf);
1790 
1791 	/* Free RQ buffer pointers*/
1792 	otx2_free_aura_ptr(pf, AURA_NIX_RQ);
1793 
1794 	otx2_free_cq_res(pf);
1795 
1796 	/* Free all ingress bandwidth profiles allocated */
1797 	if (!otx2_rep_dev(pf->pdev))
1798 		cn10k_free_all_ipolicers(pf);
1799 
1800 	mutex_lock(&mbox->lock);
1801 	/* Reset NIX LF */
1802 	free_req = otx2_mbox_alloc_msg_nix_lf_free(mbox);
1803 	if (free_req) {
1804 		free_req->flags = NIX_LF_DISABLE_FLOWS | NIX_LF_DONT_FREE_DFT_IDXS;
1805 		if (!(pf->flags & OTX2_FLAG_PF_SHUTDOWN))
1806 			free_req->flags |= NIX_LF_DONT_FREE_TX_VTAG;
1807 		if (otx2_sync_mbox_msg(mbox))
1808 			dev_err(pf->dev, "%s failed to free nixlf\n", __func__);
1809 	}
1810 	mutex_unlock(&mbox->lock);
1811 
1812 	/* Disable NPA Pool and Aura hw context */
1813 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_POOL, true);
1814 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_AURA, true);
1815 	otx2_aura_pool_free(pf);
1816 
1817 	mutex_lock(&mbox->lock);
1818 	/* Reset NPA LF */
1819 	req = otx2_mbox_alloc_msg_npa_lf_free(mbox);
1820 	if (req) {
1821 		if (otx2_sync_mbox_msg(mbox))
1822 			dev_err(pf->dev, "%s failed to free npalf\n", __func__);
1823 	}
1824 	mutex_unlock(&mbox->lock);
1825 }
1826 EXPORT_SYMBOL(otx2_free_hw_resources);
1827 
1828 static bool otx2_promisc_use_mce_list(struct otx2_nic *pfvf)
1829 {
1830 	int vf;
1831 
1832 	/* The AF driver will determine whether to allow the VF netdev or not */
1833 	if (is_otx2_vf(pfvf->pcifunc))
1834 		return true;
1835 
1836 	/* check if there are any trusted VFs associated with the PF netdev */
1837 	for (vf = 0; vf < pci_num_vf(pfvf->pdev); vf++)
1838 		if (pfvf->vf_configs[vf].trusted)
1839 			return true;
1840 	return false;
1841 }
1842 
1843 static void otx2_do_set_rx_mode(struct otx2_nic *pf)
1844 {
1845 	struct net_device *netdev = pf->netdev;
1846 	struct nix_rx_mode *req;
1847 	bool promisc = false;
1848 
1849 	if (!(netdev->flags & IFF_UP))
1850 		return;
1851 
1852 	if ((netdev->flags & IFF_PROMISC) ||
1853 	    (netdev_uc_count(netdev) > pf->flow_cfg->ucast_flt_cnt)) {
1854 		promisc = true;
1855 	}
1856 
1857 	/* Write unicast address to mcam entries or del from mcam */
1858 	if (!promisc && netdev->priv_flags & IFF_UNICAST_FLT)
1859 		__dev_uc_sync(netdev, otx2_add_macfilter, otx2_del_macfilter);
1860 
1861 	mutex_lock(&pf->mbox.lock);
1862 	req = otx2_mbox_alloc_msg_nix_set_rx_mode(&pf->mbox);
1863 	if (!req) {
1864 		mutex_unlock(&pf->mbox.lock);
1865 		return;
1866 	}
1867 
1868 	req->mode = NIX_RX_MODE_UCAST;
1869 
1870 	if (promisc)
1871 		req->mode |= NIX_RX_MODE_PROMISC;
1872 	if (netdev->flags & (IFF_ALLMULTI | IFF_MULTICAST))
1873 		req->mode |= NIX_RX_MODE_ALLMULTI;
1874 
1875 	if (otx2_promisc_use_mce_list(pf))
1876 		req->mode |= NIX_RX_MODE_USE_MCE;
1877 
1878 	otx2_sync_mbox_msg(&pf->mbox);
1879 	mutex_unlock(&pf->mbox.lock);
1880 }
1881 
1882 static void otx2_set_irq_coalesce(struct otx2_nic *pfvf)
1883 {
1884 	int cint;
1885 
1886 	for (cint = 0; cint < pfvf->hw.cint_cnt; cint++)
1887 		otx2_config_irq_coalescing(pfvf, cint);
1888 }
1889 
1890 static void otx2_dim_work(struct work_struct *w)
1891 {
1892 	struct dim_cq_moder cur_moder;
1893 	struct otx2_cq_poll *cq_poll;
1894 	struct otx2_nic *pfvf;
1895 	struct dim *dim;
1896 
1897 	dim = container_of(w, struct dim, work);
1898 	cur_moder = net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1899 	cq_poll = container_of(dim, struct otx2_cq_poll, dim);
1900 	pfvf = (struct otx2_nic *)cq_poll->dev;
1901 	pfvf->hw.cq_time_wait = (cur_moder.usec > CQ_TIMER_THRESH_MAX) ?
1902 		CQ_TIMER_THRESH_MAX : cur_moder.usec;
1903 	pfvf->hw.cq_ecount_wait = (cur_moder.pkts > NAPI_POLL_WEIGHT) ?
1904 		NAPI_POLL_WEIGHT : cur_moder.pkts;
1905 	otx2_set_irq_coalesce(pfvf);
1906 	dim->state = DIM_START_MEASURE;
1907 }
1908 
1909 void otx2_free_queue_mem(struct otx2_qset *qset)
1910 {
1911 	kfree(qset->sq);
1912 	qset->sq = NULL;
1913 	kfree(qset->cq);
1914 	qset->cq = NULL;
1915 	kfree(qset->rq);
1916 	qset->rq = NULL;
1917 	kfree(qset->napi);
1918 	qset->napi = NULL;
1919 }
1920 EXPORT_SYMBOL(otx2_free_queue_mem);
1921 
1922 int otx2_alloc_queue_mem(struct otx2_nic *pf)
1923 {
1924 	struct otx2_qset *qset = &pf->qset;
1925 	struct otx2_cq_poll *cq_poll;
1926 
1927 	/* RQ and SQs are mapped to different CQs,
1928 	 * so find out max CQ IRQs (i.e CINTs) needed.
1929 	 */
1930 	pf->hw.non_qos_queues =  pf->hw.tx_queues + pf->hw.xdp_queues;
1931 	pf->hw.cint_cnt = max3(pf->hw.rx_queues, pf->hw.tx_queues,
1932 			       pf->hw.tc_tx_queues);
1933 
1934 	pf->qset.cq_cnt = pf->hw.rx_queues + otx2_get_total_tx_queues(pf);
1935 
1936 	qset->napi = kzalloc_objs(*cq_poll, pf->hw.cint_cnt);
1937 	if (!qset->napi)
1938 		return -ENOMEM;
1939 
1940 	/* CQ size of RQ */
1941 	qset->rqe_cnt = qset->rqe_cnt ? qset->rqe_cnt : Q_COUNT(Q_SIZE_256);
1942 	/* CQ size of SQ */
1943 	qset->sqe_cnt = qset->sqe_cnt ? qset->sqe_cnt : Q_COUNT(Q_SIZE_4K);
1944 
1945 	qset->cq = kzalloc_objs(struct otx2_cq_queue, pf->qset.cq_cnt);
1946 	if (!qset->cq)
1947 		goto err_free_mem;
1948 
1949 	qset->sq = kzalloc_objs(struct otx2_snd_queue,
1950 				otx2_get_total_tx_queues(pf));
1951 	if (!qset->sq)
1952 		goto err_free_mem;
1953 
1954 	qset->rq = kzalloc_objs(struct otx2_rcv_queue, pf->hw.rx_queues);
1955 	if (!qset->rq)
1956 		goto err_free_mem;
1957 
1958 	return 0;
1959 
1960 err_free_mem:
1961 	otx2_free_queue_mem(qset);
1962 	return -ENOMEM;
1963 }
1964 EXPORT_SYMBOL(otx2_alloc_queue_mem);
1965 
1966 int otx2_open(struct net_device *netdev)
1967 {
1968 	struct otx2_nic *pf = netdev_priv(netdev);
1969 	struct otx2_cq_poll *cq_poll = NULL;
1970 	struct otx2_qset *qset = &pf->qset;
1971 	int err = 0, qidx, vec;
1972 	char *irq_name;
1973 
1974 	netif_carrier_off(netdev);
1975 
1976 	err = otx2_alloc_queue_mem(pf);
1977 	if (err)
1978 		return err;
1979 
1980 	err = otx2_init_hw_resources(pf);
1981 	if (err)
1982 		goto err_free_mem;
1983 
1984 	/* Register NAPI handler */
1985 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
1986 		cq_poll = &qset->napi[qidx];
1987 		cq_poll->cint_idx = qidx;
1988 		/* RQ0 & SQ0 are mapped to CINT0 and so on..
1989 		 * 'cq_ids[0]' points to RQ's CQ and
1990 		 * 'cq_ids[1]' points to SQ's CQ and
1991 		 * 'cq_ids[2]' points to XDP's CQ and
1992 		 */
1993 		cq_poll->cq_ids[CQ_RX] =
1994 			(qidx <  pf->hw.rx_queues) ? qidx : CINT_INVALID_CQ;
1995 		cq_poll->cq_ids[CQ_TX] = (qidx < pf->hw.tx_queues) ?
1996 				      qidx + pf->hw.rx_queues : CINT_INVALID_CQ;
1997 		if (pf->xdp_prog)
1998 			cq_poll->cq_ids[CQ_XDP] = (qidx < pf->hw.xdp_queues) ?
1999 						  (qidx + pf->hw.rx_queues +
2000 						  pf->hw.tx_queues) :
2001 						  CINT_INVALID_CQ;
2002 		else
2003 			cq_poll->cq_ids[CQ_XDP] = CINT_INVALID_CQ;
2004 
2005 		cq_poll->cq_ids[CQ_QOS] = (qidx < pf->hw.tc_tx_queues) ?
2006 					  (qidx + pf->hw.rx_queues +
2007 					   pf->hw.non_qos_queues) :
2008 					  CINT_INVALID_CQ;
2009 
2010 		cq_poll->dev = (void *)pf;
2011 		cq_poll->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_CQE;
2012 		INIT_WORK(&cq_poll->dim.work, otx2_dim_work);
2013 		netif_napi_add(netdev, &cq_poll->napi, otx2_napi_handler);
2014 		napi_enable(&cq_poll->napi);
2015 	}
2016 
2017 	/* Set maximum frame size allowed in HW */
2018 	err = otx2_hw_set_mtu(pf, netdev->mtu);
2019 	if (err)
2020 		goto err_disable_napi;
2021 
2022 	/* Setup segmentation algorithms, if failed, clear offload capability */
2023 	otx2_setup_segmentation(pf);
2024 
2025 	/* Initialize RSS */
2026 	err = otx2_rss_init(pf);
2027 	if (err)
2028 		goto err_disable_napi;
2029 
2030 	/* Register Queue IRQ handlers */
2031 	vec = pf->hw.nix_msixoff + NIX_LF_QINT_VEC_START;
2032 	irq_name = &pf->hw.irq_name[vec * NAME_SIZE];
2033 
2034 	snprintf(irq_name, NAME_SIZE, "%s-qerr", pf->netdev->name);
2035 
2036 	err = request_irq(pci_irq_vector(pf->pdev, vec),
2037 			  otx2_q_intr_handler, 0, irq_name, pf);
2038 	if (err) {
2039 		dev_err(pf->dev,
2040 			"RVUPF%d: IRQ registration failed for QERR\n",
2041 			rvu_get_pf(pf->pdev, pf->pcifunc));
2042 		goto err_disable_napi;
2043 	}
2044 
2045 	/* Enable QINT IRQ */
2046 	otx2_write64(pf, NIX_LF_QINTX_ENA_W1S(0), BIT_ULL(0));
2047 
2048 	/* Register CQ IRQ handlers */
2049 	vec = pf->hw.nix_msixoff + NIX_LF_CINT_VEC_START;
2050 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
2051 		irq_name = &pf->hw.irq_name[vec * NAME_SIZE];
2052 		int name_len;
2053 
2054 		name_len = snprintf(irq_name, NAME_SIZE, "%s-rxtx-%d",
2055 				    pf->netdev->name, qidx);
2056 		if (name_len >= NAME_SIZE) {
2057 			dev_err(pf->dev,
2058 				"RVUPF%d: IRQ registration failed for CQ%d, irq name is too long\n",
2059 				rvu_get_pf(pf->pdev, pf->pcifunc), qidx);
2060 			err = -EINVAL;
2061 			goto err_free_cints;
2062 		}
2063 
2064 		err = request_irq(pci_irq_vector(pf->pdev, vec),
2065 				  otx2_cq_intr_handler, 0, irq_name,
2066 				  &qset->napi[qidx]);
2067 		if (err) {
2068 			dev_err(pf->dev,
2069 				"RVUPF%d: IRQ registration failed for CQ%d\n",
2070 				rvu_get_pf(pf->pdev, pf->pcifunc), qidx);
2071 			goto err_free_cints;
2072 		}
2073 		vec++;
2074 
2075 		otx2_config_irq_coalescing(pf, qidx);
2076 
2077 		/* Enable CQ IRQ */
2078 		otx2_write64(pf, NIX_LF_CINTX_INT(qidx), BIT_ULL(0));
2079 		otx2_write64(pf, NIX_LF_CINTX_ENA_W1S(qidx), BIT_ULL(0));
2080 	}
2081 
2082 	otx2_set_cints_affinity(pf);
2083 
2084 	if (pf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
2085 		otx2_enable_rxvlan(pf, true);
2086 
2087 	/* When reinitializing enable time stamping if it is enabled before */
2088 	if (pf->flags & OTX2_FLAG_TX_TSTAMP_ENABLED) {
2089 		pf->flags &= ~OTX2_FLAG_TX_TSTAMP_ENABLED;
2090 		otx2_config_hw_tx_tstamp(pf, true);
2091 	}
2092 	if (pf->flags & OTX2_FLAG_RX_TSTAMP_ENABLED) {
2093 		pf->flags &= ~OTX2_FLAG_RX_TSTAMP_ENABLED;
2094 		otx2_config_hw_rx_tstamp(pf, true);
2095 	}
2096 
2097 	pf->flags &= ~OTX2_FLAG_INTF_DOWN;
2098 	pf->flags &= ~OTX2_FLAG_PORT_UP;
2099 	/* 'intf_down' may be checked on any cpu */
2100 	smp_wmb();
2101 
2102 	/* Enable QoS configuration before starting tx queues */
2103 	otx2_qos_config_txschq(pf);
2104 
2105 	/* we have already received link status notification */
2106 	if (pf->linfo.link_up && !(pf->pcifunc & RVU_PFVF_FUNC_MASK))
2107 		otx2_handle_link_event(pf);
2108 
2109 	/* Install DMAC Filters */
2110 	if (pf->flags & OTX2_FLAG_DMACFLTR_SUPPORT)
2111 		otx2_dmacflt_reinstall_flows(pf);
2112 
2113 	otx2_tc_apply_ingress_police_rules(pf);
2114 
2115 	err = otx2_rxtx_enable(pf, true);
2116 	/* If a mbox communication error happens at this point then interface
2117 	 * will end up in a state such that it is in down state but hardware
2118 	 * mcam entries are enabled to receive the packets. Hence disable the
2119 	 * packet I/O.
2120 	 */
2121 	if (err == -EIO)
2122 		goto err_disable_rxtx;
2123 	else if (err)
2124 		goto err_tx_stop_queues;
2125 
2126 	otx2_do_set_rx_mode(pf);
2127 
2128 	return 0;
2129 
2130 err_disable_rxtx:
2131 	otx2_rxtx_enable(pf, false);
2132 err_tx_stop_queues:
2133 	netif_tx_stop_all_queues(netdev);
2134 	netif_carrier_off(netdev);
2135 	pf->flags |= OTX2_FLAG_INTF_DOWN;
2136 err_free_cints:
2137 	otx2_free_cints(pf, qidx);
2138 	vec = pci_irq_vector(pf->pdev,
2139 			     pf->hw.nix_msixoff + NIX_LF_QINT_VEC_START);
2140 	otx2_write64(pf, NIX_LF_QINTX_ENA_W1C(0), BIT_ULL(0));
2141 	free_irq(vec, pf);
2142 err_disable_napi:
2143 	otx2_disable_napi(pf);
2144 	otx2_free_hw_resources(pf);
2145 err_free_mem:
2146 	otx2_free_queue_mem(qset);
2147 	return err;
2148 }
2149 EXPORT_SYMBOL(otx2_open);
2150 
2151 int otx2_stop(struct net_device *netdev)
2152 {
2153 	struct otx2_nic *pf = netdev_priv(netdev);
2154 	struct otx2_cq_poll *cq_poll = NULL;
2155 	struct otx2_qset *qset = &pf->qset;
2156 	int qidx, vec, wrk;
2157 
2158 	/* If the DOWN flag is set resources are already freed */
2159 	if (pf->flags & OTX2_FLAG_INTF_DOWN)
2160 		return 0;
2161 
2162 	netif_carrier_off(netdev);
2163 	netif_tx_stop_all_queues(netdev);
2164 
2165 	pf->flags |= OTX2_FLAG_INTF_DOWN;
2166 	/* 'intf_down' may be checked on any cpu */
2167 	smp_wmb();
2168 
2169 	/* First stop packet Rx/Tx */
2170 	otx2_rxtx_enable(pf, false);
2171 
2172 	/* Clear RSS enable flag */
2173 	pf->hw.rss_info.enable = false;
2174 
2175 	/* Cleanup Queue IRQ */
2176 	vec = pci_irq_vector(pf->pdev,
2177 			     pf->hw.nix_msixoff + NIX_LF_QINT_VEC_START);
2178 	otx2_write64(pf, NIX_LF_QINTX_ENA_W1C(0), BIT_ULL(0));
2179 	free_irq(vec, pf);
2180 
2181 	/* Cleanup CQ NAPI and IRQ */
2182 	vec = pf->hw.nix_msixoff + NIX_LF_CINT_VEC_START;
2183 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
2184 		/* Disable interrupt */
2185 		otx2_write64(pf, NIX_LF_CINTX_ENA_W1C(qidx), BIT_ULL(0));
2186 
2187 		synchronize_irq(pci_irq_vector(pf->pdev, vec));
2188 
2189 		cq_poll = &qset->napi[qidx];
2190 		napi_synchronize(&cq_poll->napi);
2191 		vec++;
2192 	}
2193 
2194 	netif_tx_disable(netdev);
2195 
2196 	for (wrk = 0; wrk < pf->qset.cq_cnt; wrk++)
2197 		cancel_delayed_work_sync(&pf->refill_wrk[wrk].pool_refill_work);
2198 	devm_kfree(pf->dev, pf->refill_wrk);
2199 
2200 	otx2_free_hw_resources(pf);
2201 	otx2_free_cints(pf, pf->hw.cint_cnt);
2202 	otx2_disable_napi(pf);
2203 
2204 	for (qidx = 0; qidx < netdev->num_tx_queues; qidx++)
2205 		netdev_tx_reset_queue(netdev_get_tx_queue(netdev, qidx));
2206 
2207 	otx2_free_queue_mem(qset);
2208 	/* Do not clear RQ/SQ ringsize settings */
2209 	memset_startat(qset, 0, sqe_cnt);
2210 	return 0;
2211 }
2212 EXPORT_SYMBOL(otx2_stop);
2213 
2214 static netdev_tx_t otx2_xmit(struct sk_buff *skb, struct net_device *netdev)
2215 {
2216 	struct otx2_nic *pf = netdev_priv(netdev);
2217 	int qidx = skb_get_queue_mapping(skb);
2218 	struct otx2_dev_stats *dev_stats;
2219 	struct otx2_snd_queue *sq;
2220 	struct netdev_queue *txq;
2221 	int sq_idx;
2222 
2223 	/* XDP SQs are not mapped with TXQs
2224 	 * advance qid to derive correct sq mapped with QOS
2225 	 */
2226 	sq_idx = (qidx >= pf->hw.tx_queues) ? (qidx + pf->hw.xdp_queues) : qidx;
2227 
2228 	/* Check for minimum and maximum packet length */
2229 	if (skb->len <= ETH_HLEN ||
2230 	    (!skb_shinfo(skb)->gso_size && skb->len > pf->tx_max_pktlen)) {
2231 		dev_stats = &pf->hw.dev_stats;
2232 		atomic_long_inc(&dev_stats->tx_discards);
2233 		dev_kfree_skb(skb);
2234 		return NETDEV_TX_OK;
2235 	}
2236 
2237 	sq = &pf->qset.sq[sq_idx];
2238 	txq = netdev_get_tx_queue(netdev, qidx);
2239 
2240 	if (!otx2_sq_append_skb(pf, txq, sq, skb, qidx)) {
2241 		netif_tx_stop_queue(txq);
2242 
2243 		/* Check again, incase SQBs got freed up */
2244 		smp_mb();
2245 		if (((sq->num_sqbs - *sq->aura_fc_addr) * sq->sqe_per_sqb)
2246 							> sq->sqe_thresh)
2247 			netif_tx_wake_queue(txq);
2248 
2249 		return NETDEV_TX_BUSY;
2250 	}
2251 
2252 	return NETDEV_TX_OK;
2253 }
2254 
2255 static int otx2_qos_select_htb_queue(struct otx2_nic *pf, struct sk_buff *skb,
2256 				     u16 htb_maj_id)
2257 {
2258 	u16 classid;
2259 
2260 	if ((TC_H_MAJ(skb->priority) >> 16) == htb_maj_id)
2261 		classid = TC_H_MIN(skb->priority);
2262 	else
2263 		classid = READ_ONCE(pf->qos.defcls);
2264 
2265 	if (!classid)
2266 		return 0;
2267 
2268 	return otx2_get_txq_by_classid(pf, classid);
2269 }
2270 
2271 u16 otx2_select_queue(struct net_device *netdev, struct sk_buff *skb,
2272 		      struct net_device *sb_dev)
2273 {
2274 	struct otx2_nic *pf = netdev_priv(netdev);
2275 	bool qos_enabled;
2276 #ifdef CONFIG_DCB
2277 	u8 vlan_prio;
2278 #endif
2279 	int txq;
2280 
2281 	qos_enabled = netdev->real_num_tx_queues > pf->hw.tx_queues;
2282 	if (unlikely(qos_enabled)) {
2283 		/* This smp_load_acquire() pairs with smp_store_release() in
2284 		 * otx2_qos_root_add() called from htb offload root creation
2285 		 */
2286 		u16 htb_maj_id = smp_load_acquire(&pf->qos.maj_id);
2287 
2288 		if (unlikely(htb_maj_id)) {
2289 			txq = otx2_qos_select_htb_queue(pf, skb, htb_maj_id);
2290 			if (txq > 0)
2291 				return txq;
2292 			goto process_pfc;
2293 		}
2294 	}
2295 
2296 process_pfc:
2297 #ifdef CONFIG_DCB
2298 	if (!skb_vlan_tag_present(skb))
2299 		goto pick_tx;
2300 
2301 	vlan_prio = skb->vlan_tci >> 13;
2302 	if ((vlan_prio > pf->hw.tx_queues - 1) ||
2303 	    !pf->pfc_alloc_status[vlan_prio])
2304 		goto pick_tx;
2305 
2306 	return vlan_prio;
2307 
2308 pick_tx:
2309 #endif
2310 	txq = netdev_pick_tx(netdev, skb, NULL);
2311 	if (unlikely(qos_enabled))
2312 		return txq % pf->hw.tx_queues;
2313 
2314 	return txq;
2315 }
2316 EXPORT_SYMBOL(otx2_select_queue);
2317 
2318 static netdev_features_t otx2_fix_features(struct net_device *dev,
2319 					   netdev_features_t features)
2320 {
2321 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
2322 		features |= NETIF_F_HW_VLAN_STAG_RX;
2323 	else
2324 		features &= ~NETIF_F_HW_VLAN_STAG_RX;
2325 
2326 	return features;
2327 }
2328 
2329 static void otx2_set_rx_mode(struct net_device *netdev)
2330 {
2331 	struct otx2_nic *pf = netdev_priv(netdev);
2332 
2333 	queue_work(pf->otx2_wq, &pf->rx_mode_work);
2334 }
2335 
2336 static void otx2_rx_mode_wrk_handler(struct work_struct *work)
2337 {
2338 	struct otx2_nic *pf = container_of(work, struct otx2_nic, rx_mode_work);
2339 
2340 	otx2_do_set_rx_mode(pf);
2341 }
2342 
2343 static int otx2_set_features(struct net_device *netdev,
2344 			     netdev_features_t features)
2345 {
2346 	netdev_features_t changed = features ^ netdev->features;
2347 	struct otx2_nic *pf = netdev_priv(netdev);
2348 
2349 	if ((changed & NETIF_F_LOOPBACK) && netif_running(netdev))
2350 		return otx2_cgx_config_loopback(pf,
2351 						features & NETIF_F_LOOPBACK);
2352 
2353 	if ((changed & NETIF_F_HW_VLAN_CTAG_RX) && netif_running(netdev))
2354 		return otx2_enable_rxvlan(pf,
2355 					  features & NETIF_F_HW_VLAN_CTAG_RX);
2356 
2357 	if (changed & NETIF_F_HW_ESP)
2358 		return cn10k_ipsec_ethtool_init(netdev,
2359 						features & NETIF_F_HW_ESP);
2360 
2361 	return otx2_handle_ntuple_tc_features(netdev, features);
2362 }
2363 
2364 static void otx2_reset_task(struct work_struct *work)
2365 {
2366 	struct otx2_nic *pf = container_of(work, struct otx2_nic, reset_task);
2367 
2368 	if (!netif_running(pf->netdev))
2369 		return;
2370 
2371 	rtnl_lock();
2372 	otx2_stop(pf->netdev);
2373 	pf->reset_count++;
2374 	otx2_open(pf->netdev);
2375 	netif_trans_update(pf->netdev);
2376 	rtnl_unlock();
2377 }
2378 
2379 static int otx2_config_hw_rx_tstamp(struct otx2_nic *pfvf, bool enable)
2380 {
2381 	struct msg_req *req;
2382 	int err;
2383 
2384 	if (pfvf->flags & OTX2_FLAG_RX_TSTAMP_ENABLED && enable)
2385 		return 0;
2386 
2387 	mutex_lock(&pfvf->mbox.lock);
2388 	if (enable)
2389 		req = otx2_mbox_alloc_msg_cgx_ptp_rx_enable(&pfvf->mbox);
2390 	else
2391 		req = otx2_mbox_alloc_msg_cgx_ptp_rx_disable(&pfvf->mbox);
2392 	if (!req) {
2393 		mutex_unlock(&pfvf->mbox.lock);
2394 		return -ENOMEM;
2395 	}
2396 
2397 	err = otx2_sync_mbox_msg(&pfvf->mbox);
2398 	if (err) {
2399 		mutex_unlock(&pfvf->mbox.lock);
2400 		return err;
2401 	}
2402 
2403 	mutex_unlock(&pfvf->mbox.lock);
2404 	if (enable)
2405 		pfvf->flags |= OTX2_FLAG_RX_TSTAMP_ENABLED;
2406 	else
2407 		pfvf->flags &= ~OTX2_FLAG_RX_TSTAMP_ENABLED;
2408 	return 0;
2409 }
2410 
2411 static int otx2_config_hw_tx_tstamp(struct otx2_nic *pfvf, bool enable)
2412 {
2413 	struct msg_req *req;
2414 	int err;
2415 
2416 	if (pfvf->flags & OTX2_FLAG_TX_TSTAMP_ENABLED && enable)
2417 		return 0;
2418 
2419 	mutex_lock(&pfvf->mbox.lock);
2420 	if (enable)
2421 		req = otx2_mbox_alloc_msg_nix_lf_ptp_tx_enable(&pfvf->mbox);
2422 	else
2423 		req = otx2_mbox_alloc_msg_nix_lf_ptp_tx_disable(&pfvf->mbox);
2424 	if (!req) {
2425 		mutex_unlock(&pfvf->mbox.lock);
2426 		return -ENOMEM;
2427 	}
2428 
2429 	err = otx2_sync_mbox_msg(&pfvf->mbox);
2430 	if (err) {
2431 		mutex_unlock(&pfvf->mbox.lock);
2432 		return err;
2433 	}
2434 
2435 	mutex_unlock(&pfvf->mbox.lock);
2436 	if (enable)
2437 		pfvf->flags |= OTX2_FLAG_TX_TSTAMP_ENABLED;
2438 	else
2439 		pfvf->flags &= ~OTX2_FLAG_TX_TSTAMP_ENABLED;
2440 	return 0;
2441 }
2442 
2443 int otx2_config_hwtstamp_get(struct net_device *netdev,
2444 			     struct kernel_hwtstamp_config *config)
2445 {
2446 	struct otx2_nic *pfvf = netdev_priv(netdev);
2447 
2448 	*config = pfvf->tstamp;
2449 	return 0;
2450 }
2451 EXPORT_SYMBOL(otx2_config_hwtstamp_get);
2452 
2453 int otx2_config_hwtstamp_set(struct net_device *netdev,
2454 			     struct kernel_hwtstamp_config *config,
2455 			     struct netlink_ext_ack *extack)
2456 {
2457 	struct otx2_nic *pfvf = netdev_priv(netdev);
2458 
2459 	if (!pfvf->ptp)
2460 		return -ENODEV;
2461 
2462 	switch (config->tx_type) {
2463 	case HWTSTAMP_TX_OFF:
2464 		if (pfvf->flags & OTX2_FLAG_PTP_ONESTEP_SYNC)
2465 			pfvf->flags &= ~OTX2_FLAG_PTP_ONESTEP_SYNC;
2466 
2467 		cancel_delayed_work(&pfvf->ptp->synctstamp_work);
2468 		otx2_config_hw_tx_tstamp(pfvf, false);
2469 		break;
2470 	case HWTSTAMP_TX_ONESTEP_SYNC:
2471 		if (!test_bit(CN10K_PTP_ONESTEP, &pfvf->hw.cap_flag)) {
2472 			NL_SET_ERR_MSG_MOD(extack,
2473 					   "One-step time stamping is not supported");
2474 			return -ERANGE;
2475 		}
2476 		pfvf->flags |= OTX2_FLAG_PTP_ONESTEP_SYNC;
2477 		schedule_delayed_work(&pfvf->ptp->synctstamp_work,
2478 				      msecs_to_jiffies(500));
2479 		fallthrough;
2480 	case HWTSTAMP_TX_ON:
2481 		otx2_config_hw_tx_tstamp(pfvf, true);
2482 		break;
2483 	default:
2484 		return -ERANGE;
2485 	}
2486 
2487 	switch (config->rx_filter) {
2488 	case HWTSTAMP_FILTER_NONE:
2489 		otx2_config_hw_rx_tstamp(pfvf, false);
2490 		break;
2491 	case HWTSTAMP_FILTER_ALL:
2492 	case HWTSTAMP_FILTER_SOME:
2493 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2494 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2495 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2496 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2497 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2498 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2499 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2500 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2501 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2502 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2503 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2504 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2505 		otx2_config_hw_rx_tstamp(pfvf, true);
2506 		config->rx_filter = HWTSTAMP_FILTER_ALL;
2507 		break;
2508 	default:
2509 		return -ERANGE;
2510 	}
2511 
2512 	pfvf->tstamp = *config;
2513 
2514 	return 0;
2515 }
2516 EXPORT_SYMBOL(otx2_config_hwtstamp_set);
2517 
2518 static int otx2_do_set_vf_mac(struct otx2_nic *pf, int vf, const u8 *mac)
2519 {
2520 	struct npc_install_flow_req *req;
2521 	int err;
2522 
2523 	mutex_lock(&pf->mbox.lock);
2524 	req = otx2_mbox_alloc_msg_npc_install_flow(&pf->mbox);
2525 	if (!req) {
2526 		err = -ENOMEM;
2527 		goto out;
2528 	}
2529 
2530 	ether_addr_copy(req->packet.dmac, mac);
2531 	eth_broadcast_addr((u8 *)&req->mask.dmac);
2532 	req->features = BIT_ULL(NPC_DMAC);
2533 	req->channel = pf->hw.rx_chan_base;
2534 	req->intf = NIX_INTF_RX;
2535 	req->default_rule = 1;
2536 	req->append = 1;
2537 	req->vf = vf + 1;
2538 	req->op = NIX_RX_ACTION_DEFAULT;
2539 
2540 	err = otx2_sync_mbox_msg(&pf->mbox);
2541 out:
2542 	mutex_unlock(&pf->mbox.lock);
2543 	return err;
2544 }
2545 
2546 static int otx2_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
2547 {
2548 	struct otx2_nic *pf = netdev_priv(netdev);
2549 	struct pci_dev *pdev = pf->pdev;
2550 	struct otx2_vf_config *config;
2551 	int ret;
2552 
2553 	if (!netif_running(netdev))
2554 		return -EAGAIN;
2555 
2556 	if (vf >= pf->total_vfs)
2557 		return -EINVAL;
2558 
2559 	if (!is_valid_ether_addr(mac))
2560 		return -EINVAL;
2561 
2562 	config = &pf->vf_configs[vf];
2563 	ether_addr_copy(config->mac, mac);
2564 
2565 	ret = otx2_do_set_vf_mac(pf, vf, mac);
2566 	if (ret == 0)
2567 		dev_info(&pdev->dev,
2568 			 "Load/Reload VF driver\n");
2569 
2570 	return ret;
2571 }
2572 
2573 static int otx2_do_set_vf_vlan(struct otx2_nic *pf, int vf, u16 vlan, u8 qos,
2574 			       __be16 proto)
2575 {
2576 	struct otx2_flow_config *flow_cfg = pf->flow_cfg;
2577 	struct nix_vtag_config_rsp *vtag_rsp;
2578 	struct npc_delete_flow_req *del_req;
2579 	struct nix_vtag_config *vtag_req;
2580 	struct npc_install_flow_req *req;
2581 	struct otx2_vf_config *config;
2582 	int err = 0;
2583 	u32 idx;
2584 
2585 	config = &pf->vf_configs[vf];
2586 
2587 	if (!vlan && !config->vlan)
2588 		goto out;
2589 
2590 	mutex_lock(&pf->mbox.lock);
2591 
2592 	/* free old tx vtag entry */
2593 	if (config->vlan) {
2594 		vtag_req = otx2_mbox_alloc_msg_nix_vtag_cfg(&pf->mbox);
2595 		if (!vtag_req) {
2596 			err = -ENOMEM;
2597 			goto out;
2598 		}
2599 		vtag_req->cfg_type = 0;
2600 		vtag_req->tx.free_vtag0 = 1;
2601 		vtag_req->tx.vtag0_idx = config->tx_vtag_idx;
2602 
2603 		err = otx2_sync_mbox_msg(&pf->mbox);
2604 		if (err)
2605 			goto out;
2606 	}
2607 
2608 	if (!vlan && config->vlan) {
2609 		/* rx */
2610 		del_req = otx2_mbox_alloc_msg_npc_delete_flow(&pf->mbox);
2611 		if (!del_req) {
2612 			err = -ENOMEM;
2613 			goto out;
2614 		}
2615 		idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_RX_INDEX);
2616 		del_req->entry =
2617 			flow_cfg->def_ent[flow_cfg->vf_vlan_offset + idx];
2618 		err = otx2_sync_mbox_msg(&pf->mbox);
2619 		if (err)
2620 			goto out;
2621 
2622 		/* tx */
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_TX_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 
2633 		goto out;
2634 	}
2635 
2636 	/* rx */
2637 	req = otx2_mbox_alloc_msg_npc_install_flow(&pf->mbox);
2638 	if (!req) {
2639 		err = -ENOMEM;
2640 		goto out;
2641 	}
2642 
2643 	idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_RX_INDEX);
2644 	req->entry = flow_cfg->def_ent[flow_cfg->vf_vlan_offset + idx];
2645 	req->packet.vlan_tci = htons(vlan);
2646 	req->mask.vlan_tci = htons(VLAN_VID_MASK);
2647 	/* af fills the destination mac addr */
2648 	eth_broadcast_addr((u8 *)&req->mask.dmac);
2649 	req->features = BIT_ULL(NPC_OUTER_VID) | BIT_ULL(NPC_DMAC);
2650 	req->channel = pf->hw.rx_chan_base;
2651 	req->intf = NIX_INTF_RX;
2652 	req->vf = vf + 1;
2653 	req->op = NIX_RX_ACTION_DEFAULT;
2654 	req->vtag0_valid = true;
2655 	req->vtag0_type = NIX_AF_LFX_RX_VTAG_TYPE7;
2656 	req->set_cntr = 1;
2657 
2658 	err = otx2_sync_mbox_msg(&pf->mbox);
2659 	if (err)
2660 		goto out;
2661 
2662 	/* tx */
2663 	vtag_req = otx2_mbox_alloc_msg_nix_vtag_cfg(&pf->mbox);
2664 	if (!vtag_req) {
2665 		err = -ENOMEM;
2666 		goto out;
2667 	}
2668 
2669 	/* configure tx vtag params */
2670 	vtag_req->vtag_size = VTAGSIZE_T4;
2671 	vtag_req->cfg_type = 0; /* tx vlan cfg */
2672 	vtag_req->tx.cfg_vtag0 = 1;
2673 	vtag_req->tx.vtag0 = ((u64)ntohs(proto) << 16) | vlan;
2674 
2675 	err = otx2_sync_mbox_msg(&pf->mbox);
2676 	if (err)
2677 		goto out;
2678 
2679 	vtag_rsp = (struct nix_vtag_config_rsp *)otx2_mbox_get_rsp
2680 			(&pf->mbox.mbox, 0, &vtag_req->hdr);
2681 	if (IS_ERR(vtag_rsp)) {
2682 		err = PTR_ERR(vtag_rsp);
2683 		goto out;
2684 	}
2685 	config->tx_vtag_idx = vtag_rsp->vtag0_idx;
2686 
2687 	req = otx2_mbox_alloc_msg_npc_install_flow(&pf->mbox);
2688 	if (!req) {
2689 		err = -ENOMEM;
2690 		goto out;
2691 	}
2692 
2693 	eth_zero_addr((u8 *)&req->mask.dmac);
2694 	idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_TX_INDEX);
2695 	req->entry = flow_cfg->def_ent[flow_cfg->vf_vlan_offset + idx];
2696 	req->features = BIT_ULL(NPC_DMAC);
2697 	req->channel = pf->hw.tx_chan_base;
2698 	req->intf = NIX_INTF_TX;
2699 	req->vf = vf + 1;
2700 	req->op = NIX_TX_ACTIONOP_UCAST_DEFAULT;
2701 	req->vtag0_def = vtag_rsp->vtag0_idx;
2702 	req->vtag0_op = VTAG_INSERT;
2703 	req->set_cntr = 1;
2704 
2705 	err = otx2_sync_mbox_msg(&pf->mbox);
2706 out:
2707 	config->vlan = vlan;
2708 	mutex_unlock(&pf->mbox.lock);
2709 	return err;
2710 }
2711 
2712 static int otx2_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
2713 			    __be16 proto)
2714 {
2715 	struct otx2_nic *pf = netdev_priv(netdev);
2716 	struct pci_dev *pdev = pf->pdev;
2717 
2718 	if (!netif_running(netdev))
2719 		return -EAGAIN;
2720 
2721 	if (vf >= pci_num_vf(pdev))
2722 		return -EINVAL;
2723 
2724 	/* qos is currently unsupported */
2725 	if (vlan >= VLAN_N_VID || qos)
2726 		return -EINVAL;
2727 
2728 	if (proto != htons(ETH_P_8021Q))
2729 		return -EPROTONOSUPPORT;
2730 
2731 	if (!(pf->flags & OTX2_FLAG_VF_VLAN_SUPPORT))
2732 		return -EOPNOTSUPP;
2733 
2734 	return otx2_do_set_vf_vlan(pf, vf, vlan, qos, proto);
2735 }
2736 
2737 static int otx2_get_vf_config(struct net_device *netdev, int vf,
2738 			      struct ifla_vf_info *ivi)
2739 {
2740 	struct otx2_nic *pf = netdev_priv(netdev);
2741 	struct pci_dev *pdev = pf->pdev;
2742 	struct otx2_vf_config *config;
2743 
2744 	if (!netif_running(netdev))
2745 		return -EAGAIN;
2746 
2747 	if (vf >= pci_num_vf(pdev))
2748 		return -EINVAL;
2749 
2750 	config = &pf->vf_configs[vf];
2751 	ivi->vf = vf;
2752 	ether_addr_copy(ivi->mac, config->mac);
2753 	ivi->vlan = config->vlan;
2754 	ivi->trusted = config->trusted;
2755 
2756 	return 0;
2757 }
2758 
2759 static int otx2_xdp_xmit_tx(struct otx2_nic *pf, struct xdp_frame *xdpf,
2760 			    int qidx)
2761 {
2762 	u64 dma_addr;
2763 	int err = 0;
2764 
2765 	dma_addr = otx2_dma_map_page(pf, virt_to_page(xdpf->data),
2766 				     offset_in_page(xdpf->data), xdpf->len,
2767 				     DMA_TO_DEVICE);
2768 	if (dma_mapping_error(pf->dev, dma_addr))
2769 		return -ENOMEM;
2770 
2771 	err = otx2_xdp_sq_append_pkt(pf, xdpf, dma_addr, xdpf->len,
2772 				     qidx, OTX2_XDP_REDIRECT);
2773 	if (!err) {
2774 		otx2_dma_unmap_page(pf, dma_addr, xdpf->len, DMA_TO_DEVICE);
2775 		xdp_return_frame(xdpf);
2776 		return -ENOMEM;
2777 	}
2778 	return 0;
2779 }
2780 
2781 static int otx2_xdp_xmit(struct net_device *netdev, int n,
2782 			 struct xdp_frame **frames, u32 flags)
2783 {
2784 	struct otx2_nic *pf = netdev_priv(netdev);
2785 	int qidx = smp_processor_id();
2786 	struct otx2_snd_queue *sq;
2787 	int drops = 0, i;
2788 
2789 	if (!netif_running(netdev))
2790 		return -ENETDOWN;
2791 
2792 	qidx += pf->hw.tx_queues;
2793 	sq = pf->xdp_prog ? &pf->qset.sq[qidx] : NULL;
2794 
2795 	/* Abort xmit if xdp queue is not */
2796 	if (unlikely(!sq))
2797 		return -ENXIO;
2798 
2799 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2800 		return -EINVAL;
2801 
2802 	for (i = 0; i < n; i++) {
2803 		struct xdp_frame *xdpf = frames[i];
2804 		int err;
2805 
2806 		err = otx2_xdp_xmit_tx(pf, xdpf, qidx);
2807 		if (err)
2808 			drops++;
2809 	}
2810 	return n - drops;
2811 }
2812 
2813 static int otx2_xdp_setup(struct otx2_nic *pf, struct bpf_prog *prog)
2814 {
2815 	struct net_device *dev = pf->netdev;
2816 	bool if_up = netif_running(pf->netdev);
2817 	struct bpf_prog *old_prog;
2818 
2819 	if (prog && dev->mtu > MAX_XDP_MTU) {
2820 		netdev_warn(dev, "Jumbo frames not yet supported with XDP\n");
2821 		return -EOPNOTSUPP;
2822 	}
2823 
2824 	if (if_up)
2825 		otx2_stop(pf->netdev);
2826 
2827 	old_prog = xchg(&pf->xdp_prog, prog);
2828 
2829 	if (old_prog)
2830 		bpf_prog_put(old_prog);
2831 
2832 	if (pf->xdp_prog)
2833 		bpf_prog_add(pf->xdp_prog, pf->hw.rx_queues - 1);
2834 
2835 	/* Network stack and XDP shared same rx queues.
2836 	 * Use separate tx queues for XDP and network stack.
2837 	 */
2838 	if (pf->xdp_prog) {
2839 		pf->hw.xdp_queues = pf->hw.rx_queues;
2840 		xdp_features_set_redirect_target(dev, false);
2841 	} else {
2842 		pf->hw.xdp_queues = 0;
2843 		xdp_features_clear_redirect_target(dev);
2844 	}
2845 
2846 	if (if_up)
2847 		otx2_open(pf->netdev);
2848 
2849 	return 0;
2850 }
2851 
2852 static int otx2_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
2853 {
2854 	struct otx2_nic *pf = netdev_priv(netdev);
2855 
2856 	switch (xdp->command) {
2857 	case XDP_SETUP_PROG:
2858 		return otx2_xdp_setup(pf, xdp->prog);
2859 	case XDP_SETUP_XSK_POOL:
2860 		return otx2_xsk_pool_setup(pf, xdp->xsk.pool, xdp->xsk.queue_id);
2861 	default:
2862 		return -EINVAL;
2863 	}
2864 }
2865 
2866 static int otx2_set_vf_permissions(struct otx2_nic *pf, int vf,
2867 				   int req_perm)
2868 {
2869 	struct set_vf_perm *req;
2870 	int rc;
2871 
2872 	mutex_lock(&pf->mbox.lock);
2873 	req = otx2_mbox_alloc_msg_set_vf_perm(&pf->mbox);
2874 	if (!req) {
2875 		rc = -ENOMEM;
2876 		goto out;
2877 	}
2878 
2879 	/* Let AF reset VF permissions as sriov is disabled */
2880 	if (req_perm == OTX2_RESET_VF_PERM) {
2881 		req->flags |= RESET_VF_PERM;
2882 	} else if (req_perm == OTX2_TRUSTED_VF) {
2883 		if (pf->vf_configs[vf].trusted)
2884 			req->flags |= VF_TRUSTED;
2885 	}
2886 
2887 	req->vf = vf;
2888 	rc = otx2_sync_mbox_msg(&pf->mbox);
2889 out:
2890 	mutex_unlock(&pf->mbox.lock);
2891 	return rc;
2892 }
2893 
2894 static int otx2_ndo_set_vf_trust(struct net_device *netdev, int vf,
2895 				 bool enable)
2896 {
2897 	struct otx2_nic *pf = netdev_priv(netdev);
2898 	struct pci_dev *pdev = pf->pdev;
2899 	int rc;
2900 
2901 	if (vf >= pci_num_vf(pdev))
2902 		return -EINVAL;
2903 
2904 	if (pf->vf_configs[vf].trusted == enable)
2905 		return 0;
2906 
2907 	pf->vf_configs[vf].trusted = enable;
2908 	rc = otx2_set_vf_permissions(pf, vf, OTX2_TRUSTED_VF);
2909 
2910 	if (rc) {
2911 		pf->vf_configs[vf].trusted = !enable;
2912 	} else {
2913 		netdev_info(pf->netdev, "VF %d is %strusted\n",
2914 			    vf, enable ? "" : "not ");
2915 		otx2_set_rx_mode(netdev);
2916 	}
2917 
2918 	return rc;
2919 }
2920 
2921 static const struct net_device_ops otx2_netdev_ops = {
2922 	.ndo_open		= otx2_open,
2923 	.ndo_stop		= otx2_stop,
2924 	.ndo_start_xmit		= otx2_xmit,
2925 	.ndo_select_queue	= otx2_select_queue,
2926 	.ndo_fix_features	= otx2_fix_features,
2927 	.ndo_set_mac_address    = otx2_set_mac_address,
2928 	.ndo_change_mtu		= otx2_change_mtu,
2929 	.ndo_set_rx_mode	= otx2_set_rx_mode,
2930 	.ndo_set_features	= otx2_set_features,
2931 	.ndo_tx_timeout		= otx2_tx_timeout,
2932 	.ndo_get_stats64	= otx2_get_stats64,
2933 	.ndo_set_vf_mac		= otx2_set_vf_mac,
2934 	.ndo_set_vf_vlan	= otx2_set_vf_vlan,
2935 	.ndo_get_vf_config	= otx2_get_vf_config,
2936 	.ndo_bpf		= otx2_xdp,
2937 	.ndo_xsk_wakeup		= otx2_xsk_wakeup,
2938 	.ndo_xdp_xmit           = otx2_xdp_xmit,
2939 	.ndo_setup_tc		= otx2_setup_tc,
2940 	.ndo_set_vf_trust	= otx2_ndo_set_vf_trust,
2941 	.ndo_hwtstamp_get	= otx2_config_hwtstamp_get,
2942 	.ndo_hwtstamp_set	= otx2_config_hwtstamp_set,
2943 };
2944 
2945 int otx2_wq_init(struct otx2_nic *pf)
2946 {
2947 	pf->otx2_wq = create_singlethread_workqueue("otx2_wq");
2948 	if (!pf->otx2_wq)
2949 		return -ENOMEM;
2950 
2951 	INIT_WORK(&pf->rx_mode_work, otx2_rx_mode_wrk_handler);
2952 	INIT_WORK(&pf->reset_task, otx2_reset_task);
2953 	return 0;
2954 }
2955 
2956 int otx2_check_pf_usable(struct otx2_nic *nic)
2957 {
2958 	u64 rev;
2959 
2960 	rev = otx2_read64(nic, RVU_PF_BLOCK_ADDRX_DISC(BLKADDR_RVUM));
2961 	rev = (rev >> 12) & 0xFF;
2962 	/* Check if AF has setup revision for RVUM block,
2963 	 * otherwise this driver probe should be deferred
2964 	 * until AF driver comes up.
2965 	 */
2966 	if (!rev) {
2967 		dev_warn(nic->dev,
2968 			 "AF is not initialized, deferring probe\n");
2969 		return -EPROBE_DEFER;
2970 	}
2971 	return 0;
2972 }
2973 
2974 int otx2_realloc_msix_vectors(struct otx2_nic *pf)
2975 {
2976 	struct otx2_hw *hw = &pf->hw;
2977 	int num_vec, err;
2978 
2979 	/* NPA interrupts are inot registered, so alloc only
2980 	 * upto NIX vector offset.
2981 	 */
2982 	num_vec = hw->nix_msixoff;
2983 	num_vec += NIX_LF_CINT_VEC_START + hw->max_queues;
2984 
2985 	otx2_disable_mbox_intr(pf);
2986 	pci_free_irq_vectors(hw->pdev);
2987 	err = pci_alloc_irq_vectors(hw->pdev, num_vec, num_vec, PCI_IRQ_MSIX);
2988 	if (err < 0) {
2989 		dev_err(pf->dev, "%s: Failed to realloc %d IRQ vectors\n",
2990 			__func__, num_vec);
2991 		return err;
2992 	}
2993 
2994 	return otx2_register_mbox_intr(pf, false);
2995 }
2996 EXPORT_SYMBOL(otx2_realloc_msix_vectors);
2997 
2998 static int otx2_sriov_vfcfg_init(struct otx2_nic *pf)
2999 {
3000 	int i;
3001 
3002 	pf->vf_configs = devm_kcalloc(pf->dev, pf->total_vfs,
3003 				      sizeof(struct otx2_vf_config),
3004 				      GFP_KERNEL);
3005 	if (!pf->vf_configs)
3006 		return -ENOMEM;
3007 
3008 	for (i = 0; i < pf->total_vfs; i++) {
3009 		pf->vf_configs[i].pf = pf;
3010 		pf->vf_configs[i].intf_down = true;
3011 		pf->vf_configs[i].trusted = false;
3012 		INIT_DELAYED_WORK(&pf->vf_configs[i].link_event_work,
3013 				  otx2_vf_link_event_task);
3014 	}
3015 
3016 	return 0;
3017 }
3018 
3019 static void otx2_sriov_vfcfg_cleanup(struct otx2_nic *pf)
3020 {
3021 	int i;
3022 
3023 	if (!pf->vf_configs)
3024 		return;
3025 
3026 	for (i = 0; i < pf->total_vfs; i++) {
3027 		cancel_delayed_work_sync(&pf->vf_configs[i].link_event_work);
3028 		otx2_set_vf_permissions(pf, i, OTX2_RESET_VF_PERM);
3029 	}
3030 }
3031 
3032 int otx2_init_rsrc(struct pci_dev *pdev, struct otx2_nic *pf)
3033 {
3034 	struct device *dev = &pdev->dev;
3035 	struct otx2_hw *hw = &pf->hw;
3036 	int num_vec, err;
3037 
3038 	num_vec = pci_msix_vec_count(pdev);
3039 	hw->irq_name = devm_kmalloc_array(&hw->pdev->dev, num_vec, NAME_SIZE,
3040 					  GFP_KERNEL);
3041 	if (!hw->irq_name)
3042 		return -ENOMEM;
3043 
3044 	hw->affinity_mask = devm_kcalloc(&hw->pdev->dev, num_vec,
3045 					 sizeof(cpumask_var_t), GFP_KERNEL);
3046 	if (!hw->affinity_mask)
3047 		return -ENOMEM;
3048 
3049 	/* Map CSRs */
3050 	pf->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
3051 	if (!pf->reg_base) {
3052 		dev_err(dev, "Unable to map physical function CSRs, aborting\n");
3053 		return -ENOMEM;
3054 	}
3055 
3056 	err = otx2_check_pf_usable(pf);
3057 	if (err)
3058 		return err;
3059 
3060 	if (!is_cn20k(pf->pdev))
3061 		err = pci_alloc_irq_vectors(hw->pdev, RVU_PF_INT_VEC_CNT,
3062 					    RVU_PF_INT_VEC_CNT, PCI_IRQ_MSIX);
3063 	else
3064 		err = pci_alloc_irq_vectors(hw->pdev, RVU_MBOX_PF_INT_VEC_CNT,
3065 					    RVU_MBOX_PF_INT_VEC_CNT,
3066 					    PCI_IRQ_MSIX);
3067 	if (err < 0) {
3068 		dev_err(dev, "%s: Failed to alloc %d IRQ vectors\n",
3069 			__func__, num_vec);
3070 		return err;
3071 	}
3072 
3073 	otx2_setup_dev_hw_settings(pf);
3074 
3075 	if (is_cn20k(pf->pdev))
3076 		cn20k_init(pf);
3077 	else
3078 		otx2_init_hw_ops(pf);
3079 
3080 	/* Init PF <=> AF mailbox stuff */
3081 	err = otx2_pfaf_mbox_init(pf);
3082 	if (err)
3083 		goto err_free_irq_vectors;
3084 
3085 	/* Register mailbox interrupt */
3086 	err = otx2_register_mbox_intr(pf, true);
3087 	if (err)
3088 		goto err_mbox_destroy;
3089 
3090 	/* Request AF to attach NPA and NIX LFs to this PF.
3091 	 * NIX and NPA LFs are needed for this PF to function as a NIC.
3092 	 */
3093 	err = otx2_attach_npa_nix(pf);
3094 	if (err)
3095 		goto err_disable_mbox_intr;
3096 
3097 	err = otx2_realloc_msix_vectors(pf);
3098 	if (err)
3099 		goto err_detach_rsrc;
3100 
3101 	err = cn10k_lmtst_init(pf);
3102 	if (err)
3103 		goto err_detach_rsrc;
3104 
3105 	return 0;
3106 
3107 err_detach_rsrc:
3108 	if (pf->hw.lmt_info)
3109 		free_percpu(pf->hw.lmt_info);
3110 	if (test_bit(CN10K_LMTST, &pf->hw.cap_flag))
3111 		qmem_free(pf->dev, pf->dync_lmt);
3112 	otx2_detach_resources(&pf->mbox);
3113 err_disable_mbox_intr:
3114 	otx2_disable_mbox_intr(pf);
3115 err_mbox_destroy:
3116 	otx2_pfaf_mbox_destroy(pf);
3117 err_free_irq_vectors:
3118 	pci_free_irq_vectors(hw->pdev);
3119 
3120 	return err;
3121 }
3122 EXPORT_SYMBOL(otx2_init_rsrc);
3123 
3124 static int otx2_probe(struct pci_dev *pdev, const struct pci_device_id *id)
3125 {
3126 	struct device *dev = &pdev->dev;
3127 	int err, qcount, qos_txqs;
3128 	struct net_device *netdev;
3129 	struct otx2_nic *pf;
3130 	struct otx2_hw *hw;
3131 
3132 	err = pcim_enable_device(pdev);
3133 	if (err) {
3134 		dev_err(dev, "Failed to enable PCI device\n");
3135 		return err;
3136 	}
3137 
3138 	err = pcim_request_all_regions(pdev, DRV_NAME);
3139 	if (err) {
3140 		dev_err(dev, "PCI request regions failed 0x%x\n", err);
3141 		return err;
3142 	}
3143 
3144 	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(48));
3145 	if (err) {
3146 		dev_err(dev, "DMA mask config failed, abort\n");
3147 		return err;
3148 	}
3149 
3150 	pci_set_master(pdev);
3151 
3152 	/* Set number of queues */
3153 	qcount = min_t(int, num_online_cpus(), OTX2_MAX_CQ_CNT);
3154 	qos_txqs = min_t(int, qcount, OTX2_QOS_MAX_LEAF_NODES);
3155 
3156 	netdev = alloc_etherdev_mqs(sizeof(*pf), qcount + qos_txqs, qcount);
3157 	if (!netdev)
3158 		return -ENOMEM;
3159 
3160 	pci_set_drvdata(pdev, netdev);
3161 	SET_NETDEV_DEV(netdev, &pdev->dev);
3162 	pf = netdev_priv(netdev);
3163 	pf->netdev = netdev;
3164 	pf->pdev = pdev;
3165 	pf->dev = dev;
3166 	pf->total_vfs = pci_sriov_get_totalvfs(pdev);
3167 	pf->flags |= OTX2_FLAG_INTF_DOWN;
3168 
3169 	hw = &pf->hw;
3170 	hw->pdev = pdev;
3171 	hw->rx_queues = qcount;
3172 	hw->tx_queues = qcount;
3173 	hw->non_qos_queues = qcount;
3174 	hw->max_queues = qcount;
3175 	hw->rbuf_len = OTX2_DEFAULT_RBUF_LEN;
3176 	/* Use CQE of 128 byte descriptor size by default */
3177 	hw->xqe_size = 128;
3178 
3179 	err = otx2_init_rsrc(pdev, pf);
3180 	if (err)
3181 		goto err_free_netdev;
3182 
3183 	err = otx2_set_real_num_queues(netdev, hw->tx_queues, hw->rx_queues);
3184 	if (err)
3185 		goto err_detach_rsrc;
3186 
3187 	/* Assign default mac address */
3188 	otx2_get_mac_from_af(netdev);
3189 
3190 	/* Don't check for error.  Proceed without ptp */
3191 	otx2_ptp_init(pf);
3192 
3193 	/* NPA's pool is a stack to which SW frees buffer pointers via Aura.
3194 	 * HW allocates buffer pointer from stack and uses it for DMA'ing
3195 	 * ingress packet. In some scenarios HW can free back allocated buffer
3196 	 * pointers to pool. This makes it impossible for SW to maintain a
3197 	 * parallel list where physical addresses of buffer pointers (IOVAs)
3198 	 * given to HW can be saved for later reference.
3199 	 *
3200 	 * So the only way to convert Rx packet's buffer address is to use
3201 	 * IOMMU's iova_to_phys() handler which translates the address by
3202 	 * walking through the translation tables.
3203 	 */
3204 	pf->iommu_domain = iommu_get_domain_for_dev(dev);
3205 
3206 	netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_IP_CSUM |
3207 			       NETIF_F_IPV6_CSUM | NETIF_F_RXHASH |
3208 			       NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
3209 			       NETIF_F_GSO_UDP_L4);
3210 	netdev->features |= netdev->hw_features;
3211 
3212 	err = otx2_mcam_flow_init(pf);
3213 	if (err)
3214 		goto err_ptp_destroy;
3215 
3216 	otx2_set_hw_capabilities(pf);
3217 
3218 	err = cn10k_mcs_init(pf);
3219 	if (err)
3220 		goto err_del_mcam_entries;
3221 
3222 	if (pf->flags & OTX2_FLAG_NTUPLE_SUPPORT)
3223 		netdev->hw_features |= NETIF_F_NTUPLE;
3224 
3225 	if (pf->flags & OTX2_FLAG_UCAST_FLTR_SUPPORT)
3226 		netdev->priv_flags |= IFF_UNICAST_FLT;
3227 
3228 	/* Support TSO on tag interface */
3229 	netdev->vlan_features |= netdev->features;
3230 	netdev->hw_features  |= NETIF_F_HW_VLAN_CTAG_TX |
3231 				NETIF_F_HW_VLAN_STAG_TX;
3232 	if (pf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
3233 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX |
3234 				       NETIF_F_HW_VLAN_STAG_RX;
3235 	netdev->features |= netdev->hw_features;
3236 
3237 	/* HW supports tc offload but mutually exclusive with n-tuple filters */
3238 	if (pf->flags & OTX2_FLAG_TC_FLOWER_SUPPORT)
3239 		netdev->hw_features |= NETIF_F_HW_TC;
3240 
3241 	netdev->hw_features |= NETIF_F_LOOPBACK | NETIF_F_RXALL;
3242 
3243 	netif_set_tso_max_segs(netdev, OTX2_MAX_GSO_SEGS);
3244 	netdev->watchdog_timeo = OTX2_TX_TIMEOUT;
3245 
3246 	netdev->netdev_ops = &otx2_netdev_ops;
3247 	netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
3248 			       NETDEV_XDP_ACT_NDO_XMIT |
3249 			       NETDEV_XDP_ACT_XSK_ZEROCOPY;
3250 
3251 	netdev->min_mtu = OTX2_MIN_MTU;
3252 	netdev->max_mtu = otx2_get_max_mtu(pf);
3253 	hw->max_mtu = netdev->max_mtu;
3254 
3255 	/* reset CGX/RPM MAC stats */
3256 	otx2_reset_mac_stats(pf);
3257 
3258 	err = cn10k_ipsec_init(netdev);
3259 	if (err)
3260 		goto err_mcs_free;
3261 
3262 	err = register_netdev(netdev);
3263 	if (err) {
3264 		dev_err(dev, "Failed to register netdevice\n");
3265 		goto err_ipsec_clean;
3266 	}
3267 
3268 	err = otx2_wq_init(pf);
3269 	if (err)
3270 		goto err_unreg_netdev;
3271 
3272 	otx2_set_ethtool_ops(netdev);
3273 
3274 	err = otx2_init_tc(pf);
3275 	if (err)
3276 		goto err_mcam_flow_del;
3277 
3278 	err = otx2_register_dl(pf);
3279 	if (err)
3280 		goto err_mcam_flow_del;
3281 
3282 	/* Initialize SR-IOV resources */
3283 	err = otx2_sriov_vfcfg_init(pf);
3284 	if (err)
3285 		goto err_pf_sriov_init;
3286 
3287 	/* Enable link notifications */
3288 	otx2_cgx_config_linkevents(pf, true);
3289 
3290 	pf->af_xdp_zc_qidx = bitmap_zalloc(qcount, GFP_KERNEL);
3291 	if (!pf->af_xdp_zc_qidx) {
3292 		err = -ENOMEM;
3293 		goto err_sriov_cleannup;
3294 	}
3295 
3296 #ifdef CONFIG_DCB
3297 	err = otx2_dcbnl_set_ops(netdev);
3298 	if (err)
3299 		goto err_free_zc_bmap;
3300 #endif
3301 
3302 	otx2_qos_init(pf, qos_txqs);
3303 
3304 	return 0;
3305 
3306 #ifdef CONFIG_DCB
3307 err_free_zc_bmap:
3308 	bitmap_free(pf->af_xdp_zc_qidx);
3309 #endif
3310 err_sriov_cleannup:
3311 	otx2_sriov_vfcfg_cleanup(pf);
3312 err_pf_sriov_init:
3313 	otx2_unregister_dl(pf);
3314 	otx2_shutdown_tc(pf);
3315 err_mcam_flow_del:
3316 	otx2_mcam_flow_del(pf);
3317 err_unreg_netdev:
3318 	unregister_netdev(netdev);
3319 err_ipsec_clean:
3320 	cn10k_ipsec_clean(pf);
3321 err_mcs_free:
3322 	cn10k_mcs_free(pf);
3323 err_del_mcam_entries:
3324 	otx2_mcam_flow_del(pf);
3325 err_ptp_destroy:
3326 	otx2_ptp_destroy(pf);
3327 err_detach_rsrc:
3328 	if (pf->hw.lmt_info)
3329 		free_percpu(pf->hw.lmt_info);
3330 	if (test_bit(CN10K_LMTST, &pf->hw.cap_flag))
3331 		qmem_free(pf->dev, pf->dync_lmt);
3332 	otx2_detach_resources(&pf->mbox);
3333 	otx2_disable_mbox_intr(pf);
3334 	otx2_pfaf_mbox_destroy(pf);
3335 	pci_free_irq_vectors(hw->pdev);
3336 err_free_netdev:
3337 	pci_set_drvdata(pdev, NULL);
3338 	free_netdev(netdev);
3339 	return err;
3340 }
3341 
3342 static void otx2_vf_link_event_task(struct work_struct *work)
3343 {
3344 	struct otx2_vf_config *config;
3345 	struct cgx_link_info_msg *req;
3346 	struct mbox_msghdr *msghdr;
3347 	struct delayed_work *dwork;
3348 	struct otx2_nic *pf;
3349 	int vf_idx;
3350 
3351 	config = container_of(work, struct otx2_vf_config,
3352 			      link_event_work.work);
3353 	vf_idx = config - config->pf->vf_configs;
3354 	pf = config->pf;
3355 
3356 	if (config->intf_down)
3357 		return;
3358 
3359 	mutex_lock(&pf->mbox.lock);
3360 
3361 	dwork = &config->link_event_work;
3362 
3363 	if (!otx2_mbox_wait_for_zero(&pf->mbox_pfvf[0].mbox_up, vf_idx)) {
3364 		schedule_delayed_work(dwork, msecs_to_jiffies(100));
3365 		mutex_unlock(&pf->mbox.lock);
3366 		return;
3367 	}
3368 
3369 	msghdr = otx2_mbox_alloc_msg_rsp(&pf->mbox_pfvf[0].mbox_up, vf_idx,
3370 					 sizeof(*req), sizeof(struct msg_rsp));
3371 	if (!msghdr) {
3372 		dev_err(pf->dev, "Failed to create VF%d link event\n", vf_idx);
3373 		mutex_unlock(&pf->mbox.lock);
3374 		return;
3375 	}
3376 
3377 	req = (struct cgx_link_info_msg *)msghdr;
3378 	req->hdr.id = MBOX_MSG_CGX_LINK_EVENT;
3379 	req->hdr.sig = OTX2_MBOX_REQ_SIG;
3380 	req->hdr.pcifunc = pf->pcifunc;
3381 	memcpy(&req->link_info, &pf->linfo, sizeof(req->link_info));
3382 
3383 	otx2_mbox_wait_for_zero(&pf->mbox_pfvf[0].mbox_up, vf_idx);
3384 
3385 	otx2_sync_mbox_up_msg(&pf->mbox_pfvf[0], vf_idx);
3386 
3387 	mutex_unlock(&pf->mbox.lock);
3388 }
3389 
3390 static int otx2_sriov_enable(struct pci_dev *pdev, int numvfs)
3391 {
3392 	struct net_device *netdev = pci_get_drvdata(pdev);
3393 	struct otx2_nic *pf = netdev_priv(netdev);
3394 	int ret;
3395 
3396 	/* Init PF <=> VF mailbox stuff */
3397 	ret = otx2_pfvf_mbox_init(pf, numvfs);
3398 	if (ret)
3399 		return ret;
3400 
3401 	ret = otx2_register_pfvf_mbox_intr(pf, numvfs);
3402 	if (ret)
3403 		goto free_mbox;
3404 
3405 	ret = otx2_pf_flr_init(pf, numvfs);
3406 	if (ret)
3407 		goto free_intr;
3408 
3409 	ret = otx2_register_flr_me_intr(pf, numvfs);
3410 	if (ret)
3411 		goto free_flr;
3412 
3413 	ret = pci_enable_sriov(pdev, numvfs);
3414 	if (ret)
3415 		goto free_flr_intr;
3416 
3417 	return numvfs;
3418 free_flr_intr:
3419 	otx2_disable_flr_me_intr(pf);
3420 free_flr:
3421 	otx2_flr_wq_destroy(pf);
3422 free_intr:
3423 	otx2_disable_pfvf_mbox_intr(pf, numvfs);
3424 free_mbox:
3425 	otx2_pfvf_mbox_destroy(pf);
3426 	return ret;
3427 }
3428 
3429 static int otx2_sriov_disable(struct pci_dev *pdev)
3430 {
3431 	struct net_device *netdev = pci_get_drvdata(pdev);
3432 	struct otx2_nic *pf = netdev_priv(netdev);
3433 	int numvfs = pci_num_vf(pdev);
3434 
3435 	if (!numvfs)
3436 		return 0;
3437 
3438 	pci_disable_sriov(pdev);
3439 
3440 	otx2_disable_flr_me_intr(pf);
3441 	otx2_flr_wq_destroy(pf);
3442 	otx2_disable_pfvf_mbox_intr(pf, numvfs);
3443 	otx2_pfvf_mbox_destroy(pf);
3444 
3445 	return 0;
3446 }
3447 
3448 static int otx2_sriov_configure(struct pci_dev *pdev, int numvfs)
3449 {
3450 	if (numvfs == 0)
3451 		return otx2_sriov_disable(pdev);
3452 	else
3453 		return otx2_sriov_enable(pdev, numvfs);
3454 }
3455 
3456 static void otx2_ndc_sync(struct otx2_nic *pf)
3457 {
3458 	struct mbox *mbox = &pf->mbox;
3459 	struct ndc_sync_op *req;
3460 
3461 	mutex_lock(&mbox->lock);
3462 
3463 	req = otx2_mbox_alloc_msg_ndc_sync_op(mbox);
3464 	if (!req) {
3465 		mutex_unlock(&mbox->lock);
3466 		return;
3467 	}
3468 
3469 	req->nix_lf_tx_sync = 1;
3470 	req->nix_lf_rx_sync = 1;
3471 	req->npa_lf_sync = 1;
3472 
3473 	if (otx2_sync_mbox_msg(mbox))
3474 		dev_err(pf->dev, "NDC sync operation failed\n");
3475 
3476 	mutex_unlock(&mbox->lock);
3477 }
3478 
3479 static void otx2_remove(struct pci_dev *pdev)
3480 {
3481 	struct net_device *netdev = pci_get_drvdata(pdev);
3482 	struct otx2_nic *pf;
3483 
3484 	if (!netdev)
3485 		return;
3486 
3487 	pf = netdev_priv(netdev);
3488 
3489 	pf->flags |= OTX2_FLAG_PF_SHUTDOWN;
3490 
3491 	if (pf->flags & OTX2_FLAG_TX_TSTAMP_ENABLED)
3492 		otx2_config_hw_tx_tstamp(pf, false);
3493 	if (pf->flags & OTX2_FLAG_RX_TSTAMP_ENABLED)
3494 		otx2_config_hw_rx_tstamp(pf, false);
3495 
3496 	/* Disable 802.3x pause frames */
3497 	if (pf->flags & OTX2_FLAG_RX_PAUSE_ENABLED ||
3498 	    (pf->flags & OTX2_FLAG_TX_PAUSE_ENABLED)) {
3499 		pf->flags &= ~OTX2_FLAG_RX_PAUSE_ENABLED;
3500 		pf->flags &= ~OTX2_FLAG_TX_PAUSE_ENABLED;
3501 		otx2_config_pause_frm(pf);
3502 	}
3503 
3504 #ifdef CONFIG_DCB
3505 	/* Disable PFC config */
3506 	if (pf->pfc_en) {
3507 		pf->pfc_en = 0;
3508 		otx2_config_priority_flow_ctrl(pf);
3509 	}
3510 #endif
3511 	cancel_work_sync(&pf->reset_task);
3512 	/* Disable link notifications */
3513 	otx2_cgx_config_linkevents(pf, false);
3514 
3515 	otx2_unregister_dl(pf);
3516 	unregister_netdev(netdev);
3517 	cn10k_ipsec_clean(pf);
3518 	cn10k_mcs_free(pf);
3519 	otx2_sriov_disable(pf->pdev);
3520 	otx2_sriov_vfcfg_cleanup(pf);
3521 	if (pf->otx2_wq)
3522 		destroy_workqueue(pf->otx2_wq);
3523 
3524 	otx2_ptp_destroy(pf);
3525 	otx2_mcam_flow_del(pf);
3526 	otx2_shutdown_tc(pf);
3527 	otx2_shutdown_qos(pf);
3528 	otx2_ndc_sync(pf);
3529 	otx2_detach_resources(&pf->mbox);
3530 	if (pf->hw.lmt_info)
3531 		free_percpu(pf->hw.lmt_info);
3532 	if (test_bit(CN10K_LMTST, &pf->hw.cap_flag))
3533 		qmem_free(pf->dev, pf->dync_lmt);
3534 	otx2_disable_mbox_intr(pf);
3535 	otx2_pfaf_mbox_destroy(pf);
3536 	pci_free_irq_vectors(pf->pdev);
3537 	bitmap_free(pf->af_xdp_zc_qidx);
3538 	pci_set_drvdata(pdev, NULL);
3539 	free_netdev(netdev);
3540 }
3541 
3542 static struct pci_driver otx2_pf_driver = {
3543 	.name = DRV_NAME,
3544 	.id_table = otx2_pf_id_table,
3545 	.probe = otx2_probe,
3546 	.shutdown = otx2_remove,
3547 	.remove = otx2_remove,
3548 	.sriov_configure = otx2_sriov_configure
3549 };
3550 
3551 static int __init otx2_rvupf_init_module(void)
3552 {
3553 	pr_info("%s: %s\n", DRV_NAME, DRV_STRING);
3554 
3555 	return pci_register_driver(&otx2_pf_driver);
3556 }
3557 
3558 static void __exit otx2_rvupf_cleanup_module(void)
3559 {
3560 	pci_unregister_driver(&otx2_pf_driver);
3561 }
3562 
3563 module_init(otx2_rvupf_init_module);
3564 module_exit(otx2_rvupf_cleanup_module);
3565