xref: /linux/drivers/net/ethernet/marvell/octeontx2/nic/otx2_pf.c (revision 7255fcc80d4b525cc10cfaaf7f485830d4ed2000)
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 
30 #define DRV_NAME	"rvu_nicpf"
31 #define DRV_STRING	"Marvell RVU NIC Physical Function Driver"
32 
33 /* Supported devices */
34 static const struct pci_device_id otx2_pf_id_table[] = {
35 	{ PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, PCI_DEVID_OCTEONTX2_RVU_PF) },
36 	{ 0, }  /* end of table */
37 };
38 
39 MODULE_AUTHOR("Sunil Goutham <sgoutham@marvell.com>");
40 MODULE_DESCRIPTION(DRV_STRING);
41 MODULE_LICENSE("GPL v2");
42 MODULE_DEVICE_TABLE(pci, otx2_pf_id_table);
43 
44 static void otx2_vf_link_event_task(struct work_struct *work);
45 
46 enum {
47 	TYPE_PFAF,
48 	TYPE_PFVF,
49 };
50 
51 static int otx2_config_hw_tx_tstamp(struct otx2_nic *pfvf, bool enable);
52 static int otx2_config_hw_rx_tstamp(struct otx2_nic *pfvf, bool enable);
53 
54 static int otx2_change_mtu(struct net_device *netdev, int new_mtu)
55 {
56 	struct otx2_nic *pf = netdev_priv(netdev);
57 	bool if_up = netif_running(netdev);
58 	int err = 0;
59 
60 	if (pf->xdp_prog && new_mtu > MAX_XDP_MTU) {
61 		netdev_warn(netdev, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
62 			    netdev->mtu);
63 		return -EINVAL;
64 	}
65 	if (if_up)
66 		otx2_stop(netdev);
67 
68 	netdev_info(netdev, "Changing MTU from %d to %d\n",
69 		    netdev->mtu, new_mtu);
70 	netdev->mtu = new_mtu;
71 
72 	if (if_up)
73 		err = otx2_open(netdev);
74 
75 	return err;
76 }
77 
78 static void otx2_disable_flr_me_intr(struct otx2_nic *pf)
79 {
80 	int irq, vfs = pf->total_vfs;
81 
82 	/* Disable VFs ME interrupts */
83 	otx2_write64(pf, RVU_PF_VFME_INT_ENA_W1CX(0), INTR_MASK(vfs));
84 	irq = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFME0);
85 	free_irq(irq, pf);
86 
87 	/* Disable VFs FLR interrupts */
88 	otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1CX(0), INTR_MASK(vfs));
89 	irq = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFFLR0);
90 	free_irq(irq, pf);
91 
92 	if (vfs <= 64)
93 		return;
94 
95 	otx2_write64(pf, RVU_PF_VFME_INT_ENA_W1CX(1), INTR_MASK(vfs - 64));
96 	irq = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFME1);
97 	free_irq(irq, pf);
98 
99 	otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1CX(1), INTR_MASK(vfs - 64));
100 	irq = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFFLR1);
101 	free_irq(irq, pf);
102 }
103 
104 static void otx2_flr_wq_destroy(struct otx2_nic *pf)
105 {
106 	if (!pf->flr_wq)
107 		return;
108 	destroy_workqueue(pf->flr_wq);
109 	pf->flr_wq = NULL;
110 	devm_kfree(pf->dev, pf->flr_wrk);
111 }
112 
113 static void otx2_flr_handler(struct work_struct *work)
114 {
115 	struct flr_work *flrwork = container_of(work, struct flr_work, work);
116 	struct otx2_nic *pf = flrwork->pf;
117 	struct mbox *mbox = &pf->mbox;
118 	struct msg_req *req;
119 	int vf, reg = 0;
120 
121 	vf = flrwork - pf->flr_wrk;
122 
123 	mutex_lock(&mbox->lock);
124 	req = otx2_mbox_alloc_msg_vf_flr(mbox);
125 	if (!req) {
126 		mutex_unlock(&mbox->lock);
127 		return;
128 	}
129 	req->hdr.pcifunc &= RVU_PFVF_FUNC_MASK;
130 	req->hdr.pcifunc |= (vf + 1) & RVU_PFVF_FUNC_MASK;
131 
132 	if (!otx2_sync_mbox_msg(&pf->mbox)) {
133 		if (vf >= 64) {
134 			reg = 1;
135 			vf = vf - 64;
136 		}
137 		/* clear transcation pending bit */
138 		otx2_write64(pf, RVU_PF_VFTRPENDX(reg), BIT_ULL(vf));
139 		otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1SX(reg), BIT_ULL(vf));
140 	}
141 
142 	mutex_unlock(&mbox->lock);
143 }
144 
145 static irqreturn_t otx2_pf_flr_intr_handler(int irq, void *pf_irq)
146 {
147 	struct otx2_nic *pf = (struct otx2_nic *)pf_irq;
148 	int reg, dev, vf, start_vf, num_reg = 1;
149 	u64 intr;
150 
151 	if (pf->total_vfs > 64)
152 		num_reg = 2;
153 
154 	for (reg = 0; reg < num_reg; reg++) {
155 		intr = otx2_read64(pf, RVU_PF_VFFLR_INTX(reg));
156 		if (!intr)
157 			continue;
158 		start_vf = 64 * reg;
159 		for (vf = 0; vf < 64; vf++) {
160 			if (!(intr & BIT_ULL(vf)))
161 				continue;
162 			dev = vf + start_vf;
163 			queue_work(pf->flr_wq, &pf->flr_wrk[dev].work);
164 			/* Clear interrupt */
165 			otx2_write64(pf, RVU_PF_VFFLR_INTX(reg), BIT_ULL(vf));
166 			/* Disable the interrupt */
167 			otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1CX(reg),
168 				     BIT_ULL(vf));
169 		}
170 	}
171 	return IRQ_HANDLED;
172 }
173 
174 static irqreturn_t otx2_pf_me_intr_handler(int irq, void *pf_irq)
175 {
176 	struct otx2_nic *pf = (struct otx2_nic *)pf_irq;
177 	int vf, reg, num_reg = 1;
178 	u64 intr;
179 
180 	if (pf->total_vfs > 64)
181 		num_reg = 2;
182 
183 	for (reg = 0; reg < num_reg; reg++) {
184 		intr = otx2_read64(pf, RVU_PF_VFME_INTX(reg));
185 		if (!intr)
186 			continue;
187 		for (vf = 0; vf < 64; vf++) {
188 			if (!(intr & BIT_ULL(vf)))
189 				continue;
190 			/* clear trpend bit */
191 			otx2_write64(pf, RVU_PF_VFTRPENDX(reg), BIT_ULL(vf));
192 			/* clear interrupt */
193 			otx2_write64(pf, RVU_PF_VFME_INTX(reg), BIT_ULL(vf));
194 		}
195 	}
196 	return IRQ_HANDLED;
197 }
198 
199 static int otx2_register_flr_me_intr(struct otx2_nic *pf, int numvfs)
200 {
201 	struct otx2_hw *hw = &pf->hw;
202 	char *irq_name;
203 	int ret;
204 
205 	/* Register ME interrupt handler*/
206 	irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFME0 * NAME_SIZE];
207 	snprintf(irq_name, NAME_SIZE, "RVUPF%d_ME0", rvu_get_pf(pf->pcifunc));
208 	ret = request_irq(pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFME0),
209 			  otx2_pf_me_intr_handler, 0, irq_name, pf);
210 	if (ret) {
211 		dev_err(pf->dev,
212 			"RVUPF: IRQ registration failed for ME0\n");
213 	}
214 
215 	/* Register FLR interrupt handler */
216 	irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFFLR0 * NAME_SIZE];
217 	snprintf(irq_name, NAME_SIZE, "RVUPF%d_FLR0", rvu_get_pf(pf->pcifunc));
218 	ret = request_irq(pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFFLR0),
219 			  otx2_pf_flr_intr_handler, 0, irq_name, pf);
220 	if (ret) {
221 		dev_err(pf->dev,
222 			"RVUPF: IRQ registration failed for FLR0\n");
223 		return ret;
224 	}
225 
226 	if (numvfs > 64) {
227 		irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFME1 * NAME_SIZE];
228 		snprintf(irq_name, NAME_SIZE, "RVUPF%d_ME1",
229 			 rvu_get_pf(pf->pcifunc));
230 		ret = request_irq(pci_irq_vector
231 				  (pf->pdev, RVU_PF_INT_VEC_VFME1),
232 				  otx2_pf_me_intr_handler, 0, irq_name, pf);
233 		if (ret) {
234 			dev_err(pf->dev,
235 				"RVUPF: IRQ registration failed for ME1\n");
236 		}
237 		irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFFLR1 * NAME_SIZE];
238 		snprintf(irq_name, NAME_SIZE, "RVUPF%d_FLR1",
239 			 rvu_get_pf(pf->pcifunc));
240 		ret = request_irq(pci_irq_vector
241 				  (pf->pdev, RVU_PF_INT_VEC_VFFLR1),
242 				  otx2_pf_flr_intr_handler, 0, irq_name, pf);
243 		if (ret) {
244 			dev_err(pf->dev,
245 				"RVUPF: IRQ registration failed for FLR1\n");
246 			return ret;
247 		}
248 	}
249 
250 	/* Enable ME interrupt for all VFs*/
251 	otx2_write64(pf, RVU_PF_VFME_INTX(0), INTR_MASK(numvfs));
252 	otx2_write64(pf, RVU_PF_VFME_INT_ENA_W1SX(0), INTR_MASK(numvfs));
253 
254 	/* Enable FLR interrupt for all VFs*/
255 	otx2_write64(pf, RVU_PF_VFFLR_INTX(0), INTR_MASK(numvfs));
256 	otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1SX(0), INTR_MASK(numvfs));
257 
258 	if (numvfs > 64) {
259 		numvfs -= 64;
260 
261 		otx2_write64(pf, RVU_PF_VFME_INTX(1), INTR_MASK(numvfs));
262 		otx2_write64(pf, RVU_PF_VFME_INT_ENA_W1SX(1),
263 			     INTR_MASK(numvfs));
264 
265 		otx2_write64(pf, RVU_PF_VFFLR_INTX(1), INTR_MASK(numvfs));
266 		otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1SX(1),
267 			     INTR_MASK(numvfs));
268 	}
269 	return 0;
270 }
271 
272 static int otx2_pf_flr_init(struct otx2_nic *pf, int num_vfs)
273 {
274 	int vf;
275 
276 	pf->flr_wq = alloc_ordered_workqueue("otx2_pf_flr_wq", WQ_HIGHPRI);
277 	if (!pf->flr_wq)
278 		return -ENOMEM;
279 
280 	pf->flr_wrk = devm_kcalloc(pf->dev, num_vfs,
281 				   sizeof(struct flr_work), GFP_KERNEL);
282 	if (!pf->flr_wrk) {
283 		destroy_workqueue(pf->flr_wq);
284 		return -ENOMEM;
285 	}
286 
287 	for (vf = 0; vf < num_vfs; vf++) {
288 		pf->flr_wrk[vf].pf = pf;
289 		INIT_WORK(&pf->flr_wrk[vf].work, otx2_flr_handler);
290 	}
291 
292 	return 0;
293 }
294 
295 static void otx2_queue_vf_work(struct mbox *mw, struct workqueue_struct *mbox_wq,
296 			       int first, int mdevs, u64 intr)
297 {
298 	struct otx2_mbox_dev *mdev;
299 	struct otx2_mbox *mbox;
300 	struct mbox_hdr *hdr;
301 	int i;
302 
303 	for (i = first; i < mdevs; i++) {
304 		/* start from 0 */
305 		if (!(intr & BIT_ULL(i - first)))
306 			continue;
307 
308 		mbox = &mw->mbox;
309 		mdev = &mbox->dev[i];
310 		hdr = mdev->mbase + mbox->rx_start;
311 		/* The hdr->num_msgs is set to zero immediately in the interrupt
312 		 * handler to ensure that it holds a correct value next time
313 		 * when the interrupt handler is called. pf->mw[i].num_msgs
314 		 * holds the data for use in otx2_pfvf_mbox_handler and
315 		 * pf->mw[i].up_num_msgs holds the data for use in
316 		 * otx2_pfvf_mbox_up_handler.
317 		 */
318 		if (hdr->num_msgs) {
319 			mw[i].num_msgs = hdr->num_msgs;
320 			hdr->num_msgs = 0;
321 			queue_work(mbox_wq, &mw[i].mbox_wrk);
322 		}
323 
324 		mbox = &mw->mbox_up;
325 		mdev = &mbox->dev[i];
326 		hdr = mdev->mbase + mbox->rx_start;
327 		if (hdr->num_msgs) {
328 			mw[i].up_num_msgs = hdr->num_msgs;
329 			hdr->num_msgs = 0;
330 			queue_work(mbox_wq, &mw[i].mbox_up_wrk);
331 		}
332 	}
333 }
334 
335 static void otx2_forward_msg_pfvf(struct otx2_mbox_dev *mdev,
336 				  struct otx2_mbox *pfvf_mbox, void *bbuf_base,
337 				  int devid)
338 {
339 	struct otx2_mbox_dev *src_mdev = mdev;
340 	int offset;
341 
342 	/* Msgs are already copied, trigger VF's mbox irq */
343 	smp_wmb();
344 
345 	otx2_mbox_wait_for_zero(pfvf_mbox, devid);
346 
347 	offset = pfvf_mbox->trigger | (devid << pfvf_mbox->tr_shift);
348 	writeq(MBOX_DOWN_MSG, (void __iomem *)pfvf_mbox->reg_base + offset);
349 
350 	/* Restore VF's mbox bounce buffer region address */
351 	src_mdev->mbase = bbuf_base;
352 }
353 
354 static int otx2_forward_vf_mbox_msgs(struct otx2_nic *pf,
355 				     struct otx2_mbox *src_mbox,
356 				     int dir, int vf, int num_msgs)
357 {
358 	struct otx2_mbox_dev *src_mdev, *dst_mdev;
359 	struct mbox_hdr *mbox_hdr;
360 	struct mbox_hdr *req_hdr;
361 	struct mbox *dst_mbox;
362 	int dst_size, err;
363 
364 	if (dir == MBOX_DIR_PFAF) {
365 		/* Set VF's mailbox memory as PF's bounce buffer memory, so
366 		 * that explicit copying of VF's msgs to PF=>AF mbox region
367 		 * and AF=>PF responses to VF's mbox region can be avoided.
368 		 */
369 		src_mdev = &src_mbox->dev[vf];
370 		mbox_hdr = src_mbox->hwbase +
371 				src_mbox->rx_start + (vf * MBOX_SIZE);
372 
373 		dst_mbox = &pf->mbox;
374 		dst_size = dst_mbox->mbox.tx_size -
375 				ALIGN(sizeof(*mbox_hdr), MBOX_MSG_ALIGN);
376 		/* Check if msgs fit into destination area and has valid size */
377 		if (mbox_hdr->msg_size > dst_size || !mbox_hdr->msg_size)
378 			return -EINVAL;
379 
380 		dst_mdev = &dst_mbox->mbox.dev[0];
381 
382 		mutex_lock(&pf->mbox.lock);
383 		dst_mdev->mbase = src_mdev->mbase;
384 		dst_mdev->msg_size = mbox_hdr->msg_size;
385 		dst_mdev->num_msgs = num_msgs;
386 		err = otx2_sync_mbox_msg(dst_mbox);
387 		/* Error code -EIO indicate there is a communication failure
388 		 * to the AF. Rest of the error codes indicate that AF processed
389 		 * VF messages and set the error codes in response messages
390 		 * (if any) so simply forward responses to VF.
391 		 */
392 		if (err == -EIO) {
393 			dev_warn(pf->dev,
394 				 "AF not responding to VF%d messages\n", vf);
395 			/* restore PF mbase and exit */
396 			dst_mdev->mbase = pf->mbox.bbuf_base;
397 			mutex_unlock(&pf->mbox.lock);
398 			return err;
399 		}
400 		/* At this point, all the VF messages sent to AF are acked
401 		 * with proper responses and responses are copied to VF
402 		 * mailbox hence raise interrupt to VF.
403 		 */
404 		req_hdr = (struct mbox_hdr *)(dst_mdev->mbase +
405 					      dst_mbox->mbox.rx_start);
406 		req_hdr->num_msgs = num_msgs;
407 
408 		otx2_forward_msg_pfvf(dst_mdev, &pf->mbox_pfvf[0].mbox,
409 				      pf->mbox.bbuf_base, vf);
410 		mutex_unlock(&pf->mbox.lock);
411 	} else if (dir == MBOX_DIR_PFVF_UP) {
412 		src_mdev = &src_mbox->dev[0];
413 		mbox_hdr = src_mbox->hwbase + src_mbox->rx_start;
414 		req_hdr = (struct mbox_hdr *)(src_mdev->mbase +
415 					      src_mbox->rx_start);
416 		req_hdr->num_msgs = num_msgs;
417 
418 		dst_mbox = &pf->mbox_pfvf[0];
419 		dst_size = dst_mbox->mbox_up.tx_size -
420 				ALIGN(sizeof(*mbox_hdr), MBOX_MSG_ALIGN);
421 		/* Check if msgs fit into destination area */
422 		if (mbox_hdr->msg_size > dst_size)
423 			return -EINVAL;
424 
425 		dst_mdev = &dst_mbox->mbox_up.dev[vf];
426 		dst_mdev->mbase = src_mdev->mbase;
427 		dst_mdev->msg_size = mbox_hdr->msg_size;
428 		dst_mdev->num_msgs = mbox_hdr->num_msgs;
429 		err = otx2_sync_mbox_up_msg(dst_mbox, vf);
430 		if (err) {
431 			dev_warn(pf->dev,
432 				 "VF%d is not responding to mailbox\n", vf);
433 			return err;
434 		}
435 	} else if (dir == MBOX_DIR_VFPF_UP) {
436 		req_hdr = (struct mbox_hdr *)(src_mbox->dev[0].mbase +
437 					      src_mbox->rx_start);
438 		req_hdr->num_msgs = num_msgs;
439 		otx2_forward_msg_pfvf(&pf->mbox_pfvf->mbox_up.dev[vf],
440 				      &pf->mbox.mbox_up,
441 				      pf->mbox_pfvf[vf].bbuf_base,
442 				      0);
443 	}
444 
445 	return 0;
446 }
447 
448 static void otx2_pfvf_mbox_handler(struct work_struct *work)
449 {
450 	struct mbox_msghdr *msg = NULL;
451 	int offset, vf_idx, id, err;
452 	struct otx2_mbox_dev *mdev;
453 	struct mbox_hdr *req_hdr;
454 	struct otx2_mbox *mbox;
455 	struct mbox *vf_mbox;
456 	struct otx2_nic *pf;
457 
458 	vf_mbox = container_of(work, struct mbox, mbox_wrk);
459 	pf = vf_mbox->pfvf;
460 	vf_idx = vf_mbox - pf->mbox_pfvf;
461 
462 	mbox = &pf->mbox_pfvf[0].mbox;
463 	mdev = &mbox->dev[vf_idx];
464 	req_hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
465 
466 	offset = ALIGN(sizeof(*req_hdr), MBOX_MSG_ALIGN);
467 
468 	for (id = 0; id < vf_mbox->num_msgs; id++) {
469 		msg = (struct mbox_msghdr *)(mdev->mbase + mbox->rx_start +
470 					     offset);
471 
472 		if (msg->sig != OTX2_MBOX_REQ_SIG)
473 			goto inval_msg;
474 
475 		/* Set VF's number in each of the msg */
476 		msg->pcifunc &= RVU_PFVF_FUNC_MASK;
477 		msg->pcifunc |= (vf_idx + 1) & RVU_PFVF_FUNC_MASK;
478 		offset = msg->next_msgoff;
479 	}
480 	err = otx2_forward_vf_mbox_msgs(pf, mbox, MBOX_DIR_PFAF, vf_idx,
481 					vf_mbox->num_msgs);
482 	if (err)
483 		goto inval_msg;
484 	return;
485 
486 inval_msg:
487 	otx2_reply_invalid_msg(mbox, vf_idx, 0, msg->id);
488 	otx2_mbox_msg_send(mbox, vf_idx);
489 }
490 
491 static void otx2_pfvf_mbox_up_handler(struct work_struct *work)
492 {
493 	struct mbox *vf_mbox = container_of(work, struct mbox, mbox_up_wrk);
494 	struct otx2_nic *pf = vf_mbox->pfvf;
495 	struct otx2_mbox_dev *mdev;
496 	int offset, id, vf_idx = 0;
497 	struct mbox_hdr *rsp_hdr;
498 	struct mbox_msghdr *msg;
499 	struct otx2_mbox *mbox;
500 
501 	vf_idx = vf_mbox - pf->mbox_pfvf;
502 	mbox = &pf->mbox_pfvf[0].mbox_up;
503 	mdev = &mbox->dev[vf_idx];
504 
505 	rsp_hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
506 	offset = mbox->rx_start + ALIGN(sizeof(*rsp_hdr), MBOX_MSG_ALIGN);
507 
508 	for (id = 0; id < vf_mbox->up_num_msgs; id++) {
509 		msg = mdev->mbase + offset;
510 
511 		if (msg->id >= MBOX_MSG_MAX) {
512 			dev_err(pf->dev,
513 				"Mbox msg with unknown ID 0x%x\n", msg->id);
514 			goto end;
515 		}
516 
517 		if (msg->sig != OTX2_MBOX_RSP_SIG) {
518 			dev_err(pf->dev,
519 				"Mbox msg with wrong signature %x, ID 0x%x\n",
520 				msg->sig, msg->id);
521 			goto end;
522 		}
523 
524 		switch (msg->id) {
525 		case MBOX_MSG_CGX_LINK_EVENT:
526 			break;
527 		default:
528 			if (msg->rc)
529 				dev_err(pf->dev,
530 					"Mbox msg response has err %d, ID 0x%x\n",
531 					msg->rc, msg->id);
532 			break;
533 		}
534 
535 end:
536 		offset = mbox->rx_start + msg->next_msgoff;
537 		if (mdev->msgs_acked == (vf_mbox->up_num_msgs - 1))
538 			__otx2_mbox_reset(mbox, vf_idx);
539 		mdev->msgs_acked++;
540 	}
541 }
542 
543 static irqreturn_t otx2_pfvf_mbox_intr_handler(int irq, void *pf_irq)
544 {
545 	struct otx2_nic *pf = (struct otx2_nic *)(pf_irq);
546 	int vfs = pf->total_vfs;
547 	struct mbox *mbox;
548 	u64 intr;
549 
550 	mbox = pf->mbox_pfvf;
551 	/* Handle VF interrupts */
552 	if (vfs > 64) {
553 		intr = otx2_read64(pf, RVU_PF_VFPF_MBOX_INTX(1));
554 		otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(1), intr);
555 		otx2_queue_vf_work(mbox, pf->mbox_pfvf_wq, 64, vfs, intr);
556 		if (intr)
557 			trace_otx2_msg_interrupt(mbox->mbox.pdev, "VF(s) to PF", intr);
558 		vfs = 64;
559 	}
560 
561 	intr = otx2_read64(pf, RVU_PF_VFPF_MBOX_INTX(0));
562 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(0), intr);
563 
564 	otx2_queue_vf_work(mbox, pf->mbox_pfvf_wq, 0, vfs, intr);
565 
566 	if (intr)
567 		trace_otx2_msg_interrupt(mbox->mbox.pdev, "VF(s) to PF", intr);
568 
569 	return IRQ_HANDLED;
570 }
571 
572 static int otx2_pfvf_mbox_init(struct otx2_nic *pf, int numvfs)
573 {
574 	void __iomem *hwbase;
575 	struct mbox *mbox;
576 	int err, vf;
577 	u64 base;
578 
579 	if (!numvfs)
580 		return -EINVAL;
581 
582 	pf->mbox_pfvf = devm_kcalloc(&pf->pdev->dev, numvfs,
583 				     sizeof(struct mbox), GFP_KERNEL);
584 	if (!pf->mbox_pfvf)
585 		return -ENOMEM;
586 
587 	pf->mbox_pfvf_wq = alloc_workqueue("otx2_pfvf_mailbox",
588 					   WQ_UNBOUND | WQ_HIGHPRI |
589 					   WQ_MEM_RECLAIM, 0);
590 	if (!pf->mbox_pfvf_wq)
591 		return -ENOMEM;
592 
593 	/* On CN10K platform, PF <-> VF mailbox region follows after
594 	 * PF <-> AF mailbox region.
595 	 */
596 	if (test_bit(CN10K_MBOX, &pf->hw.cap_flag))
597 		base = pci_resource_start(pf->pdev, PCI_MBOX_BAR_NUM) +
598 		       MBOX_SIZE;
599 	else
600 		base = readq((void __iomem *)((u64)pf->reg_base +
601 					      RVU_PF_VF_BAR4_ADDR));
602 
603 	hwbase = ioremap_wc(base, MBOX_SIZE * pf->total_vfs);
604 	if (!hwbase) {
605 		err = -ENOMEM;
606 		goto free_wq;
607 	}
608 
609 	mbox = &pf->mbox_pfvf[0];
610 	err = otx2_mbox_init(&mbox->mbox, hwbase, pf->pdev, pf->reg_base,
611 			     MBOX_DIR_PFVF, numvfs);
612 	if (err)
613 		goto free_iomem;
614 
615 	err = otx2_mbox_init(&mbox->mbox_up, hwbase, pf->pdev, pf->reg_base,
616 			     MBOX_DIR_PFVF_UP, numvfs);
617 	if (err)
618 		goto free_iomem;
619 
620 	for (vf = 0; vf < numvfs; vf++) {
621 		mbox->pfvf = pf;
622 		INIT_WORK(&mbox->mbox_wrk, otx2_pfvf_mbox_handler);
623 		INIT_WORK(&mbox->mbox_up_wrk, otx2_pfvf_mbox_up_handler);
624 		mbox++;
625 	}
626 
627 	return 0;
628 
629 free_iomem:
630 	if (hwbase)
631 		iounmap(hwbase);
632 free_wq:
633 	destroy_workqueue(pf->mbox_pfvf_wq);
634 	return err;
635 }
636 
637 static void otx2_pfvf_mbox_destroy(struct otx2_nic *pf)
638 {
639 	struct mbox *mbox = &pf->mbox_pfvf[0];
640 
641 	if (!mbox)
642 		return;
643 
644 	if (pf->mbox_pfvf_wq) {
645 		destroy_workqueue(pf->mbox_pfvf_wq);
646 		pf->mbox_pfvf_wq = NULL;
647 	}
648 
649 	if (mbox->mbox.hwbase)
650 		iounmap(mbox->mbox.hwbase);
651 
652 	otx2_mbox_destroy(&mbox->mbox);
653 }
654 
655 static void otx2_enable_pfvf_mbox_intr(struct otx2_nic *pf, int numvfs)
656 {
657 	/* Clear PF <=> VF mailbox IRQ */
658 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(0), ~0ull);
659 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(1), ~0ull);
660 
661 	/* Enable PF <=> VF mailbox IRQ */
662 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INT_ENA_W1SX(0), INTR_MASK(numvfs));
663 	if (numvfs > 64) {
664 		numvfs -= 64;
665 		otx2_write64(pf, RVU_PF_VFPF_MBOX_INT_ENA_W1SX(1),
666 			     INTR_MASK(numvfs));
667 	}
668 }
669 
670 static void otx2_disable_pfvf_mbox_intr(struct otx2_nic *pf, int numvfs)
671 {
672 	int vector;
673 
674 	/* Disable PF <=> VF mailbox IRQ */
675 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INT_ENA_W1CX(0), ~0ull);
676 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INT_ENA_W1CX(1), ~0ull);
677 
678 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(0), ~0ull);
679 	vector = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFPF_MBOX0);
680 	free_irq(vector, pf);
681 
682 	if (numvfs > 64) {
683 		otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(1), ~0ull);
684 		vector = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFPF_MBOX1);
685 		free_irq(vector, pf);
686 	}
687 }
688 
689 static int otx2_register_pfvf_mbox_intr(struct otx2_nic *pf, int numvfs)
690 {
691 	struct otx2_hw *hw = &pf->hw;
692 	char *irq_name;
693 	int err;
694 
695 	/* Register MBOX0 interrupt handler */
696 	irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFPF_MBOX0 * NAME_SIZE];
697 	if (pf->pcifunc)
698 		snprintf(irq_name, NAME_SIZE,
699 			 "RVUPF%d_VF Mbox0", rvu_get_pf(pf->pcifunc));
700 	else
701 		snprintf(irq_name, NAME_SIZE, "RVUPF_VF Mbox0");
702 	err = request_irq(pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFPF_MBOX0),
703 			  otx2_pfvf_mbox_intr_handler, 0, irq_name, pf);
704 	if (err) {
705 		dev_err(pf->dev,
706 			"RVUPF: IRQ registration failed for PFVF mbox0 irq\n");
707 		return err;
708 	}
709 
710 	if (numvfs > 64) {
711 		/* Register MBOX1 interrupt handler */
712 		irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFPF_MBOX1 * NAME_SIZE];
713 		if (pf->pcifunc)
714 			snprintf(irq_name, NAME_SIZE,
715 				 "RVUPF%d_VF Mbox1", rvu_get_pf(pf->pcifunc));
716 		else
717 			snprintf(irq_name, NAME_SIZE, "RVUPF_VF Mbox1");
718 		err = request_irq(pci_irq_vector(pf->pdev,
719 						 RVU_PF_INT_VEC_VFPF_MBOX1),
720 						 otx2_pfvf_mbox_intr_handler,
721 						 0, irq_name, pf);
722 		if (err) {
723 			dev_err(pf->dev,
724 				"RVUPF: IRQ registration failed for PFVF mbox1 irq\n");
725 			return err;
726 		}
727 	}
728 
729 	otx2_enable_pfvf_mbox_intr(pf, numvfs);
730 
731 	return 0;
732 }
733 
734 static void otx2_process_pfaf_mbox_msg(struct otx2_nic *pf,
735 				       struct mbox_msghdr *msg)
736 {
737 	int devid;
738 
739 	if (msg->id >= MBOX_MSG_MAX) {
740 		dev_err(pf->dev,
741 			"Mbox msg with unknown ID 0x%x\n", msg->id);
742 		return;
743 	}
744 
745 	if (msg->sig != OTX2_MBOX_RSP_SIG) {
746 		dev_err(pf->dev,
747 			"Mbox msg with wrong signature %x, ID 0x%x\n",
748 			 msg->sig, msg->id);
749 		return;
750 	}
751 
752 	/* message response heading VF */
753 	devid = msg->pcifunc & RVU_PFVF_FUNC_MASK;
754 	if (devid) {
755 		struct otx2_vf_config *config = &pf->vf_configs[devid - 1];
756 		struct delayed_work *dwork;
757 
758 		switch (msg->id) {
759 		case MBOX_MSG_NIX_LF_START_RX:
760 			config->intf_down = false;
761 			dwork = &config->link_event_work;
762 			schedule_delayed_work(dwork, msecs_to_jiffies(100));
763 			break;
764 		case MBOX_MSG_NIX_LF_STOP_RX:
765 			config->intf_down = true;
766 			break;
767 		}
768 
769 		return;
770 	}
771 
772 	switch (msg->id) {
773 	case MBOX_MSG_READY:
774 		pf->pcifunc = msg->pcifunc;
775 		break;
776 	case MBOX_MSG_MSIX_OFFSET:
777 		mbox_handler_msix_offset(pf, (struct msix_offset_rsp *)msg);
778 		break;
779 	case MBOX_MSG_NPA_LF_ALLOC:
780 		mbox_handler_npa_lf_alloc(pf, (struct npa_lf_alloc_rsp *)msg);
781 		break;
782 	case MBOX_MSG_NIX_LF_ALLOC:
783 		mbox_handler_nix_lf_alloc(pf, (struct nix_lf_alloc_rsp *)msg);
784 		break;
785 	case MBOX_MSG_NIX_BP_ENABLE:
786 		mbox_handler_nix_bp_enable(pf, (struct nix_bp_cfg_rsp *)msg);
787 		break;
788 	case MBOX_MSG_CGX_STATS:
789 		mbox_handler_cgx_stats(pf, (struct cgx_stats_rsp *)msg);
790 		break;
791 	case MBOX_MSG_CGX_FEC_STATS:
792 		mbox_handler_cgx_fec_stats(pf, (struct cgx_fec_stats_rsp *)msg);
793 		break;
794 	default:
795 		if (msg->rc)
796 			dev_err(pf->dev,
797 				"Mbox msg response has err %d, ID 0x%x\n",
798 				msg->rc, msg->id);
799 		break;
800 	}
801 }
802 
803 static void otx2_pfaf_mbox_handler(struct work_struct *work)
804 {
805 	struct otx2_mbox_dev *mdev;
806 	struct mbox_hdr *rsp_hdr;
807 	struct mbox_msghdr *msg;
808 	struct otx2_mbox *mbox;
809 	struct mbox *af_mbox;
810 	struct otx2_nic *pf;
811 	int offset, id;
812 	u16 num_msgs;
813 
814 	af_mbox = container_of(work, struct mbox, mbox_wrk);
815 	mbox = &af_mbox->mbox;
816 	mdev = &mbox->dev[0];
817 	rsp_hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
818 	num_msgs = rsp_hdr->num_msgs;
819 
820 	offset = mbox->rx_start + ALIGN(sizeof(*rsp_hdr), MBOX_MSG_ALIGN);
821 	pf = af_mbox->pfvf;
822 
823 	for (id = 0; id < num_msgs; id++) {
824 		msg = (struct mbox_msghdr *)(mdev->mbase + offset);
825 		otx2_process_pfaf_mbox_msg(pf, msg);
826 		offset = mbox->rx_start + msg->next_msgoff;
827 		if (mdev->msgs_acked == (num_msgs - 1))
828 			__otx2_mbox_reset(mbox, 0);
829 		mdev->msgs_acked++;
830 	}
831 
832 }
833 
834 static void otx2_handle_link_event(struct otx2_nic *pf)
835 {
836 	struct cgx_link_user_info *linfo = &pf->linfo;
837 	struct net_device *netdev = pf->netdev;
838 
839 	pr_info("%s NIC Link is %s %d Mbps %s duplex\n", netdev->name,
840 		linfo->link_up ? "UP" : "DOWN", linfo->speed,
841 		linfo->full_duplex ? "Full" : "Half");
842 	if (linfo->link_up) {
843 		netif_carrier_on(netdev);
844 		netif_tx_start_all_queues(netdev);
845 	} else {
846 		netif_tx_stop_all_queues(netdev);
847 		netif_carrier_off(netdev);
848 	}
849 }
850 
851 int otx2_mbox_up_handler_mcs_intr_notify(struct otx2_nic *pf,
852 					 struct mcs_intr_info *event,
853 					 struct msg_rsp *rsp)
854 {
855 	cn10k_handle_mcs_event(pf, event);
856 
857 	return 0;
858 }
859 
860 int otx2_mbox_up_handler_cgx_link_event(struct otx2_nic *pf,
861 					struct cgx_link_info_msg *msg,
862 					struct msg_rsp *rsp)
863 {
864 	int i;
865 
866 	/* Copy the link info sent by AF */
867 	pf->linfo = msg->link_info;
868 
869 	/* notify VFs about link event */
870 	for (i = 0; i < pci_num_vf(pf->pdev); i++) {
871 		struct otx2_vf_config *config = &pf->vf_configs[i];
872 		struct delayed_work *dwork = &config->link_event_work;
873 
874 		if (config->intf_down)
875 			continue;
876 
877 		schedule_delayed_work(dwork, msecs_to_jiffies(100));
878 	}
879 
880 	/* interface has not been fully configured yet */
881 	if (pf->flags & OTX2_FLAG_INTF_DOWN)
882 		return 0;
883 
884 	otx2_handle_link_event(pf);
885 	return 0;
886 }
887 
888 static int otx2_process_mbox_msg_up(struct otx2_nic *pf,
889 				    struct mbox_msghdr *req)
890 {
891 	/* Check if valid, if not reply with a invalid msg */
892 	if (req->sig != OTX2_MBOX_REQ_SIG) {
893 		otx2_reply_invalid_msg(&pf->mbox.mbox_up, 0, 0, req->id);
894 		return -ENODEV;
895 	}
896 
897 	switch (req->id) {
898 #define M(_name, _id, _fn_name, _req_type, _rsp_type)			\
899 	case _id: {							\
900 		struct _rsp_type *rsp;					\
901 		int err;						\
902 									\
903 		rsp = (struct _rsp_type *)otx2_mbox_alloc_msg(		\
904 			&pf->mbox.mbox_up, 0,				\
905 			sizeof(struct _rsp_type));			\
906 		if (!rsp)						\
907 			return -ENOMEM;					\
908 									\
909 		rsp->hdr.id = _id;					\
910 		rsp->hdr.sig = OTX2_MBOX_RSP_SIG;			\
911 		rsp->hdr.pcifunc = 0;					\
912 		rsp->hdr.rc = 0;					\
913 									\
914 		err = otx2_mbox_up_handler_ ## _fn_name(		\
915 			pf, (struct _req_type *)req, rsp);		\
916 		return err;						\
917 	}
918 MBOX_UP_CGX_MESSAGES
919 MBOX_UP_MCS_MESSAGES
920 #undef M
921 		break;
922 	default:
923 		otx2_reply_invalid_msg(&pf->mbox.mbox_up, 0, 0, req->id);
924 		return -ENODEV;
925 	}
926 	return 0;
927 }
928 
929 static void otx2_pfaf_mbox_up_handler(struct work_struct *work)
930 {
931 	struct mbox *af_mbox = container_of(work, struct mbox, mbox_up_wrk);
932 	struct otx2_mbox *mbox = &af_mbox->mbox_up;
933 	struct otx2_mbox_dev *mdev = &mbox->dev[0];
934 	struct otx2_nic *pf = af_mbox->pfvf;
935 	int offset, id, devid = 0;
936 	struct mbox_hdr *rsp_hdr;
937 	struct mbox_msghdr *msg;
938 	u16 num_msgs;
939 
940 	rsp_hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
941 	num_msgs = rsp_hdr->num_msgs;
942 
943 	offset = mbox->rx_start + ALIGN(sizeof(*rsp_hdr), MBOX_MSG_ALIGN);
944 
945 	for (id = 0; id < num_msgs; id++) {
946 		msg = (struct mbox_msghdr *)(mdev->mbase + offset);
947 
948 		devid = msg->pcifunc & RVU_PFVF_FUNC_MASK;
949 		/* Skip processing VF's messages */
950 		if (!devid)
951 			otx2_process_mbox_msg_up(pf, msg);
952 		offset = mbox->rx_start + msg->next_msgoff;
953 	}
954 	/* Forward to VF iff VFs are really present */
955 	if (devid && pci_num_vf(pf->pdev)) {
956 		otx2_forward_vf_mbox_msgs(pf, &pf->mbox.mbox_up,
957 					  MBOX_DIR_PFVF_UP, devid - 1,
958 					  num_msgs);
959 		return;
960 	}
961 
962 	otx2_mbox_msg_send(mbox, 0);
963 }
964 
965 static irqreturn_t otx2_pfaf_mbox_intr_handler(int irq, void *pf_irq)
966 {
967 	struct otx2_nic *pf = (struct otx2_nic *)pf_irq;
968 	struct mbox *mw = &pf->mbox;
969 	struct otx2_mbox_dev *mdev;
970 	struct otx2_mbox *mbox;
971 	struct mbox_hdr *hdr;
972 	u64 mbox_data;
973 
974 	/* Clear the IRQ */
975 	otx2_write64(pf, RVU_PF_INT, BIT_ULL(0));
976 
977 
978 	mbox_data = otx2_read64(pf, RVU_PF_PFAF_MBOX0);
979 
980 	if (mbox_data & MBOX_UP_MSG) {
981 		mbox_data &= ~MBOX_UP_MSG;
982 		otx2_write64(pf, RVU_PF_PFAF_MBOX0, mbox_data);
983 
984 		mbox = &mw->mbox_up;
985 		mdev = &mbox->dev[0];
986 		otx2_sync_mbox_bbuf(mbox, 0);
987 
988 		hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
989 		if (hdr->num_msgs)
990 			queue_work(pf->mbox_wq, &mw->mbox_up_wrk);
991 
992 		trace_otx2_msg_interrupt(pf->pdev, "UP message from AF to PF",
993 					 BIT_ULL(0));
994 	}
995 
996 	if (mbox_data & MBOX_DOWN_MSG) {
997 		mbox_data &= ~MBOX_DOWN_MSG;
998 		otx2_write64(pf, RVU_PF_PFAF_MBOX0, mbox_data);
999 
1000 		mbox = &mw->mbox;
1001 		mdev = &mbox->dev[0];
1002 		otx2_sync_mbox_bbuf(mbox, 0);
1003 
1004 		hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
1005 		if (hdr->num_msgs)
1006 			queue_work(pf->mbox_wq, &mw->mbox_wrk);
1007 
1008 		trace_otx2_msg_interrupt(pf->pdev, "DOWN reply from AF to PF",
1009 					 BIT_ULL(0));
1010 	}
1011 
1012 	return IRQ_HANDLED;
1013 }
1014 
1015 static void otx2_disable_mbox_intr(struct otx2_nic *pf)
1016 {
1017 	int vector = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_AFPF_MBOX);
1018 
1019 	/* Disable AF => PF mailbox IRQ */
1020 	otx2_write64(pf, RVU_PF_INT_ENA_W1C, BIT_ULL(0));
1021 	free_irq(vector, pf);
1022 }
1023 
1024 static int otx2_register_mbox_intr(struct otx2_nic *pf, bool probe_af)
1025 {
1026 	struct otx2_hw *hw = &pf->hw;
1027 	struct msg_req *req;
1028 	char *irq_name;
1029 	int err;
1030 
1031 	/* Register mailbox interrupt handler */
1032 	irq_name = &hw->irq_name[RVU_PF_INT_VEC_AFPF_MBOX * NAME_SIZE];
1033 	snprintf(irq_name, NAME_SIZE, "RVUPFAF Mbox");
1034 	err = request_irq(pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_AFPF_MBOX),
1035 			  otx2_pfaf_mbox_intr_handler, 0, irq_name, pf);
1036 	if (err) {
1037 		dev_err(pf->dev,
1038 			"RVUPF: IRQ registration failed for PFAF mbox irq\n");
1039 		return err;
1040 	}
1041 
1042 	/* Enable mailbox interrupt for msgs coming from AF.
1043 	 * First clear to avoid spurious interrupts, if any.
1044 	 */
1045 	otx2_write64(pf, RVU_PF_INT, BIT_ULL(0));
1046 	otx2_write64(pf, RVU_PF_INT_ENA_W1S, BIT_ULL(0));
1047 
1048 	if (!probe_af)
1049 		return 0;
1050 
1051 	/* Check mailbox communication with AF */
1052 	req = otx2_mbox_alloc_msg_ready(&pf->mbox);
1053 	if (!req) {
1054 		otx2_disable_mbox_intr(pf);
1055 		return -ENOMEM;
1056 	}
1057 	err = otx2_sync_mbox_msg(&pf->mbox);
1058 	if (err) {
1059 		dev_warn(pf->dev,
1060 			 "AF not responding to mailbox, deferring probe\n");
1061 		otx2_disable_mbox_intr(pf);
1062 		return -EPROBE_DEFER;
1063 	}
1064 
1065 	return 0;
1066 }
1067 
1068 static void otx2_pfaf_mbox_destroy(struct otx2_nic *pf)
1069 {
1070 	struct mbox *mbox = &pf->mbox;
1071 
1072 	if (pf->mbox_wq) {
1073 		destroy_workqueue(pf->mbox_wq);
1074 		pf->mbox_wq = NULL;
1075 	}
1076 
1077 	if (mbox->mbox.hwbase)
1078 		iounmap((void __iomem *)mbox->mbox.hwbase);
1079 
1080 	otx2_mbox_destroy(&mbox->mbox);
1081 	otx2_mbox_destroy(&mbox->mbox_up);
1082 }
1083 
1084 static int otx2_pfaf_mbox_init(struct otx2_nic *pf)
1085 {
1086 	struct mbox *mbox = &pf->mbox;
1087 	void __iomem *hwbase;
1088 	int err;
1089 
1090 	mbox->pfvf = pf;
1091 	pf->mbox_wq = alloc_ordered_workqueue("otx2_pfaf_mailbox",
1092 					      WQ_HIGHPRI | WQ_MEM_RECLAIM);
1093 	if (!pf->mbox_wq)
1094 		return -ENOMEM;
1095 
1096 	/* Mailbox is a reserved memory (in RAM) region shared between
1097 	 * admin function (i.e AF) and this PF, shouldn't be mapped as
1098 	 * device memory to allow unaligned accesses.
1099 	 */
1100 	hwbase = ioremap_wc(pci_resource_start(pf->pdev, PCI_MBOX_BAR_NUM),
1101 			    MBOX_SIZE);
1102 	if (!hwbase) {
1103 		dev_err(pf->dev, "Unable to map PFAF mailbox region\n");
1104 		err = -ENOMEM;
1105 		goto exit;
1106 	}
1107 
1108 	err = otx2_mbox_init(&mbox->mbox, hwbase, pf->pdev, pf->reg_base,
1109 			     MBOX_DIR_PFAF, 1);
1110 	if (err)
1111 		goto exit;
1112 
1113 	err = otx2_mbox_init(&mbox->mbox_up, hwbase, pf->pdev, pf->reg_base,
1114 			     MBOX_DIR_PFAF_UP, 1);
1115 	if (err)
1116 		goto exit;
1117 
1118 	err = otx2_mbox_bbuf_init(mbox, pf->pdev);
1119 	if (err)
1120 		goto exit;
1121 
1122 	INIT_WORK(&mbox->mbox_wrk, otx2_pfaf_mbox_handler);
1123 	INIT_WORK(&mbox->mbox_up_wrk, otx2_pfaf_mbox_up_handler);
1124 	mutex_init(&mbox->lock);
1125 
1126 	return 0;
1127 exit:
1128 	otx2_pfaf_mbox_destroy(pf);
1129 	return err;
1130 }
1131 
1132 static int otx2_cgx_config_linkevents(struct otx2_nic *pf, bool enable)
1133 {
1134 	struct msg_req *msg;
1135 	int err;
1136 
1137 	mutex_lock(&pf->mbox.lock);
1138 	if (enable)
1139 		msg = otx2_mbox_alloc_msg_cgx_start_linkevents(&pf->mbox);
1140 	else
1141 		msg = otx2_mbox_alloc_msg_cgx_stop_linkevents(&pf->mbox);
1142 
1143 	if (!msg) {
1144 		mutex_unlock(&pf->mbox.lock);
1145 		return -ENOMEM;
1146 	}
1147 
1148 	err = otx2_sync_mbox_msg(&pf->mbox);
1149 	mutex_unlock(&pf->mbox.lock);
1150 	return err;
1151 }
1152 
1153 static int otx2_cgx_config_loopback(struct otx2_nic *pf, bool enable)
1154 {
1155 	struct msg_req *msg;
1156 	int err;
1157 
1158 	if (enable && !bitmap_empty(pf->flow_cfg->dmacflt_bmap,
1159 				    pf->flow_cfg->dmacflt_max_flows))
1160 		netdev_warn(pf->netdev,
1161 			    "CGX/RPM internal loopback might not work as DMAC filters are active\n");
1162 
1163 	mutex_lock(&pf->mbox.lock);
1164 	if (enable)
1165 		msg = otx2_mbox_alloc_msg_cgx_intlbk_enable(&pf->mbox);
1166 	else
1167 		msg = otx2_mbox_alloc_msg_cgx_intlbk_disable(&pf->mbox);
1168 
1169 	if (!msg) {
1170 		mutex_unlock(&pf->mbox.lock);
1171 		return -ENOMEM;
1172 	}
1173 
1174 	err = otx2_sync_mbox_msg(&pf->mbox);
1175 	mutex_unlock(&pf->mbox.lock);
1176 	return err;
1177 }
1178 
1179 int otx2_set_real_num_queues(struct net_device *netdev,
1180 			     int tx_queues, int rx_queues)
1181 {
1182 	int err;
1183 
1184 	err = netif_set_real_num_tx_queues(netdev, tx_queues);
1185 	if (err) {
1186 		netdev_err(netdev,
1187 			   "Failed to set no of Tx queues: %d\n", tx_queues);
1188 		return err;
1189 	}
1190 
1191 	err = netif_set_real_num_rx_queues(netdev, rx_queues);
1192 	if (err)
1193 		netdev_err(netdev,
1194 			   "Failed to set no of Rx queues: %d\n", rx_queues);
1195 	return err;
1196 }
1197 EXPORT_SYMBOL(otx2_set_real_num_queues);
1198 
1199 static char *nix_sqoperr_e_str[NIX_SQOPERR_MAX] = {
1200 	"NIX_SQOPERR_OOR",
1201 	"NIX_SQOPERR_CTX_FAULT",
1202 	"NIX_SQOPERR_CTX_POISON",
1203 	"NIX_SQOPERR_DISABLED",
1204 	"NIX_SQOPERR_SIZE_ERR",
1205 	"NIX_SQOPERR_OFLOW",
1206 	"NIX_SQOPERR_SQB_NULL",
1207 	"NIX_SQOPERR_SQB_FAULT",
1208 	"NIX_SQOPERR_SQE_SZ_ZERO",
1209 };
1210 
1211 static char *nix_mnqerr_e_str[NIX_MNQERR_MAX] = {
1212 	"NIX_MNQERR_SQ_CTX_FAULT",
1213 	"NIX_MNQERR_SQ_CTX_POISON",
1214 	"NIX_MNQERR_SQB_FAULT",
1215 	"NIX_MNQERR_SQB_POISON",
1216 	"NIX_MNQERR_TOTAL_ERR",
1217 	"NIX_MNQERR_LSO_ERR",
1218 	"NIX_MNQERR_CQ_QUERY_ERR",
1219 	"NIX_MNQERR_MAX_SQE_SIZE_ERR",
1220 	"NIX_MNQERR_MAXLEN_ERR",
1221 	"NIX_MNQERR_SQE_SIZEM1_ZERO",
1222 };
1223 
1224 static char *nix_snd_status_e_str[NIX_SND_STATUS_MAX] =  {
1225 	[NIX_SND_STATUS_GOOD] = "NIX_SND_STATUS_GOOD",
1226 	[NIX_SND_STATUS_SQ_CTX_FAULT] = "NIX_SND_STATUS_SQ_CTX_FAULT",
1227 	[NIX_SND_STATUS_SQ_CTX_POISON] = "NIX_SND_STATUS_SQ_CTX_POISON",
1228 	[NIX_SND_STATUS_SQB_FAULT] = "NIX_SND_STATUS_SQB_FAULT",
1229 	[NIX_SND_STATUS_SQB_POISON] = "NIX_SND_STATUS_SQB_POISON",
1230 	[NIX_SND_STATUS_HDR_ERR] = "NIX_SND_STATUS_HDR_ERR",
1231 	[NIX_SND_STATUS_EXT_ERR] = "NIX_SND_STATUS_EXT_ERR",
1232 	[NIX_SND_STATUS_JUMP_FAULT] = "NIX_SND_STATUS_JUMP_FAULT",
1233 	[NIX_SND_STATUS_JUMP_POISON] = "NIX_SND_STATUS_JUMP_POISON",
1234 	[NIX_SND_STATUS_CRC_ERR] = "NIX_SND_STATUS_CRC_ERR",
1235 	[NIX_SND_STATUS_IMM_ERR] = "NIX_SND_STATUS_IMM_ERR",
1236 	[NIX_SND_STATUS_SG_ERR] = "NIX_SND_STATUS_SG_ERR",
1237 	[NIX_SND_STATUS_MEM_ERR] = "NIX_SND_STATUS_MEM_ERR",
1238 	[NIX_SND_STATUS_INVALID_SUBDC] = "NIX_SND_STATUS_INVALID_SUBDC",
1239 	[NIX_SND_STATUS_SUBDC_ORDER_ERR] = "NIX_SND_STATUS_SUBDC_ORDER_ERR",
1240 	[NIX_SND_STATUS_DATA_FAULT] = "NIX_SND_STATUS_DATA_FAULT",
1241 	[NIX_SND_STATUS_DATA_POISON] = "NIX_SND_STATUS_DATA_POISON",
1242 	[NIX_SND_STATUS_NPC_DROP_ACTION] = "NIX_SND_STATUS_NPC_DROP_ACTION",
1243 	[NIX_SND_STATUS_LOCK_VIOL] = "NIX_SND_STATUS_LOCK_VIOL",
1244 	[NIX_SND_STATUS_NPC_UCAST_CHAN_ERR] = "NIX_SND_STAT_NPC_UCAST_CHAN_ERR",
1245 	[NIX_SND_STATUS_NPC_MCAST_CHAN_ERR] = "NIX_SND_STAT_NPC_MCAST_CHAN_ERR",
1246 	[NIX_SND_STATUS_NPC_MCAST_ABORT] = "NIX_SND_STATUS_NPC_MCAST_ABORT",
1247 	[NIX_SND_STATUS_NPC_VTAG_PTR_ERR] = "NIX_SND_STATUS_NPC_VTAG_PTR_ERR",
1248 	[NIX_SND_STATUS_NPC_VTAG_SIZE_ERR] = "NIX_SND_STATUS_NPC_VTAG_SIZE_ERR",
1249 	[NIX_SND_STATUS_SEND_MEM_FAULT] = "NIX_SND_STATUS_SEND_MEM_FAULT",
1250 	[NIX_SND_STATUS_SEND_STATS_ERR] = "NIX_SND_STATUS_SEND_STATS_ERR",
1251 };
1252 
1253 static irqreturn_t otx2_q_intr_handler(int irq, void *data)
1254 {
1255 	struct otx2_nic *pf = data;
1256 	struct otx2_snd_queue *sq;
1257 	u64 val, *ptr;
1258 	u64 qidx = 0;
1259 
1260 	/* CQ */
1261 	for (qidx = 0; qidx < pf->qset.cq_cnt; qidx++) {
1262 		ptr = otx2_get_regaddr(pf, NIX_LF_CQ_OP_INT);
1263 		val = otx2_atomic64_add((qidx << 44), ptr);
1264 
1265 		otx2_write64(pf, NIX_LF_CQ_OP_INT, (qidx << 44) |
1266 			     (val & NIX_CQERRINT_BITS));
1267 		if (!(val & (NIX_CQERRINT_BITS | BIT_ULL(42))))
1268 			continue;
1269 
1270 		if (val & BIT_ULL(42)) {
1271 			netdev_err(pf->netdev,
1272 				   "CQ%lld: error reading NIX_LF_CQ_OP_INT, NIX_LF_ERR_INT 0x%llx\n",
1273 				   qidx, otx2_read64(pf, NIX_LF_ERR_INT));
1274 		} else {
1275 			if (val & BIT_ULL(NIX_CQERRINT_DOOR_ERR))
1276 				netdev_err(pf->netdev, "CQ%lld: Doorbell error",
1277 					   qidx);
1278 			if (val & BIT_ULL(NIX_CQERRINT_CQE_FAULT))
1279 				netdev_err(pf->netdev,
1280 					   "CQ%lld: Memory fault on CQE write to LLC/DRAM",
1281 					   qidx);
1282 		}
1283 
1284 		schedule_work(&pf->reset_task);
1285 	}
1286 
1287 	/* SQ */
1288 	for (qidx = 0; qidx < otx2_get_total_tx_queues(pf); qidx++) {
1289 		u64 sq_op_err_dbg, mnq_err_dbg, snd_err_dbg;
1290 		u8 sq_op_err_code, mnq_err_code, snd_err_code;
1291 
1292 		sq = &pf->qset.sq[qidx];
1293 		if (!sq->sqb_ptrs)
1294 			continue;
1295 
1296 		/* Below debug registers captures first errors corresponding to
1297 		 * those registers. We don't have to check against SQ qid as
1298 		 * these are fatal errors.
1299 		 */
1300 
1301 		ptr = otx2_get_regaddr(pf, NIX_LF_SQ_OP_INT);
1302 		val = otx2_atomic64_add((qidx << 44), ptr);
1303 		otx2_write64(pf, NIX_LF_SQ_OP_INT, (qidx << 44) |
1304 			     (val & NIX_SQINT_BITS));
1305 
1306 		if (val & BIT_ULL(42)) {
1307 			netdev_err(pf->netdev,
1308 				   "SQ%lld: error reading NIX_LF_SQ_OP_INT, NIX_LF_ERR_INT 0x%llx\n",
1309 				   qidx, otx2_read64(pf, NIX_LF_ERR_INT));
1310 			goto done;
1311 		}
1312 
1313 		sq_op_err_dbg = otx2_read64(pf, NIX_LF_SQ_OP_ERR_DBG);
1314 		if (!(sq_op_err_dbg & BIT(44)))
1315 			goto chk_mnq_err_dbg;
1316 
1317 		sq_op_err_code = FIELD_GET(GENMASK(7, 0), sq_op_err_dbg);
1318 		netdev_err(pf->netdev,
1319 			   "SQ%lld: NIX_LF_SQ_OP_ERR_DBG(0x%llx)  err=%s(%#x)\n",
1320 			   qidx, sq_op_err_dbg,
1321 			   nix_sqoperr_e_str[sq_op_err_code],
1322 			   sq_op_err_code);
1323 
1324 		otx2_write64(pf, NIX_LF_SQ_OP_ERR_DBG, BIT_ULL(44));
1325 
1326 		if (sq_op_err_code == NIX_SQOPERR_SQB_NULL)
1327 			goto chk_mnq_err_dbg;
1328 
1329 		/* Err is not NIX_SQOPERR_SQB_NULL, call aq function to read SQ structure.
1330 		 * TODO: But we are in irq context. How to call mbox functions which does sleep
1331 		 */
1332 
1333 chk_mnq_err_dbg:
1334 		mnq_err_dbg = otx2_read64(pf, NIX_LF_MNQ_ERR_DBG);
1335 		if (!(mnq_err_dbg & BIT(44)))
1336 			goto chk_snd_err_dbg;
1337 
1338 		mnq_err_code = FIELD_GET(GENMASK(7, 0), mnq_err_dbg);
1339 		netdev_err(pf->netdev,
1340 			   "SQ%lld: NIX_LF_MNQ_ERR_DBG(0x%llx)  err=%s(%#x)\n",
1341 			   qidx, mnq_err_dbg,  nix_mnqerr_e_str[mnq_err_code],
1342 			   mnq_err_code);
1343 		otx2_write64(pf, NIX_LF_MNQ_ERR_DBG, BIT_ULL(44));
1344 
1345 chk_snd_err_dbg:
1346 		snd_err_dbg = otx2_read64(pf, NIX_LF_SEND_ERR_DBG);
1347 		if (snd_err_dbg & BIT(44)) {
1348 			snd_err_code = FIELD_GET(GENMASK(7, 0), snd_err_dbg);
1349 			netdev_err(pf->netdev,
1350 				   "SQ%lld: NIX_LF_SND_ERR_DBG:0x%llx err=%s(%#x)\n",
1351 				   qidx, snd_err_dbg,
1352 				   nix_snd_status_e_str[snd_err_code],
1353 				   snd_err_code);
1354 			otx2_write64(pf, NIX_LF_SEND_ERR_DBG, BIT_ULL(44));
1355 		}
1356 
1357 done:
1358 		/* Print values and reset */
1359 		if (val & BIT_ULL(NIX_SQINT_SQB_ALLOC_FAIL))
1360 			netdev_err(pf->netdev, "SQ%lld: SQB allocation failed",
1361 				   qidx);
1362 
1363 		schedule_work(&pf->reset_task);
1364 	}
1365 
1366 	return IRQ_HANDLED;
1367 }
1368 
1369 static irqreturn_t otx2_cq_intr_handler(int irq, void *cq_irq)
1370 {
1371 	struct otx2_cq_poll *cq_poll = (struct otx2_cq_poll *)cq_irq;
1372 	struct otx2_nic *pf = (struct otx2_nic *)cq_poll->dev;
1373 	int qidx = cq_poll->cint_idx;
1374 
1375 	/* Disable interrupts.
1376 	 *
1377 	 * Completion interrupts behave in a level-triggered interrupt
1378 	 * fashion, and hence have to be cleared only after it is serviced.
1379 	 */
1380 	otx2_write64(pf, NIX_LF_CINTX_ENA_W1C(qidx), BIT_ULL(0));
1381 
1382 	/* Schedule NAPI */
1383 	pf->napi_events++;
1384 	napi_schedule_irqoff(&cq_poll->napi);
1385 
1386 	return IRQ_HANDLED;
1387 }
1388 
1389 static void otx2_disable_napi(struct otx2_nic *pf)
1390 {
1391 	struct otx2_qset *qset = &pf->qset;
1392 	struct otx2_cq_poll *cq_poll;
1393 	int qidx;
1394 
1395 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
1396 		cq_poll = &qset->napi[qidx];
1397 		cancel_work_sync(&cq_poll->dim.work);
1398 		napi_disable(&cq_poll->napi);
1399 		netif_napi_del(&cq_poll->napi);
1400 	}
1401 }
1402 
1403 static void otx2_free_cq_res(struct otx2_nic *pf)
1404 {
1405 	struct otx2_qset *qset = &pf->qset;
1406 	struct otx2_cq_queue *cq;
1407 	int qidx;
1408 
1409 	/* Disable CQs */
1410 	otx2_ctx_disable(&pf->mbox, NIX_AQ_CTYPE_CQ, false);
1411 	for (qidx = 0; qidx < qset->cq_cnt; qidx++) {
1412 		cq = &qset->cq[qidx];
1413 		qmem_free(pf->dev, cq->cqe);
1414 	}
1415 }
1416 
1417 static void otx2_free_sq_res(struct otx2_nic *pf)
1418 {
1419 	struct otx2_qset *qset = &pf->qset;
1420 	struct otx2_snd_queue *sq;
1421 	int qidx;
1422 
1423 	/* Disable SQs */
1424 	otx2_ctx_disable(&pf->mbox, NIX_AQ_CTYPE_SQ, false);
1425 	/* Free SQB pointers */
1426 	otx2_sq_free_sqbs(pf);
1427 	for (qidx = 0; qidx < otx2_get_total_tx_queues(pf); qidx++) {
1428 		sq = &qset->sq[qidx];
1429 		/* Skip freeing Qos queues if they are not initialized */
1430 		if (!sq->sqe)
1431 			continue;
1432 		qmem_free(pf->dev, sq->sqe);
1433 		qmem_free(pf->dev, sq->tso_hdrs);
1434 		kfree(sq->sg);
1435 		kfree(sq->sqb_ptrs);
1436 	}
1437 }
1438 
1439 static int otx2_get_rbuf_size(struct otx2_nic *pf, int mtu)
1440 {
1441 	int frame_size;
1442 	int total_size;
1443 	int rbuf_size;
1444 
1445 	if (pf->hw.rbuf_len)
1446 		return ALIGN(pf->hw.rbuf_len, OTX2_ALIGN) + OTX2_HEAD_ROOM;
1447 
1448 	/* The data transferred by NIX to memory consists of actual packet
1449 	 * plus additional data which has timestamp and/or EDSA/HIGIG2
1450 	 * headers if interface is configured in corresponding modes.
1451 	 * NIX transfers entire data using 6 segments/buffers and writes
1452 	 * a CQE_RX descriptor with those segment addresses. First segment
1453 	 * has additional data prepended to packet. Also software omits a
1454 	 * headroom of 128 bytes in each segment. Hence the total size of
1455 	 * memory needed to receive a packet with 'mtu' is:
1456 	 * frame size =  mtu + additional data;
1457 	 * memory = frame_size + headroom * 6;
1458 	 * each receive buffer size = memory / 6;
1459 	 */
1460 	frame_size = mtu + OTX2_ETH_HLEN + OTX2_HW_TIMESTAMP_LEN;
1461 	total_size = frame_size + OTX2_HEAD_ROOM * 6;
1462 	rbuf_size = total_size / 6;
1463 
1464 	return ALIGN(rbuf_size, 2048);
1465 }
1466 
1467 static int otx2_init_hw_resources(struct otx2_nic *pf)
1468 {
1469 	struct nix_lf_free_req *free_req;
1470 	struct mbox *mbox = &pf->mbox;
1471 	struct otx2_hw *hw = &pf->hw;
1472 	struct msg_req *req;
1473 	int err = 0, lvl;
1474 
1475 	/* Set required NPA LF's pool counts
1476 	 * Auras and Pools are used in a 1:1 mapping,
1477 	 * so, aura count = pool count.
1478 	 */
1479 	hw->rqpool_cnt = hw->rx_queues;
1480 	hw->sqpool_cnt = otx2_get_total_tx_queues(pf);
1481 	hw->pool_cnt = hw->rqpool_cnt + hw->sqpool_cnt;
1482 
1483 	/* Maximum hardware supported transmit length */
1484 	pf->tx_max_pktlen = pf->netdev->max_mtu + OTX2_ETH_HLEN;
1485 
1486 	pf->rbsize = otx2_get_rbuf_size(pf, pf->netdev->mtu);
1487 
1488 	mutex_lock(&mbox->lock);
1489 	/* NPA init */
1490 	err = otx2_config_npa(pf);
1491 	if (err)
1492 		goto exit;
1493 
1494 	/* NIX init */
1495 	err = otx2_config_nix(pf);
1496 	if (err)
1497 		goto err_free_npa_lf;
1498 
1499 	/* Enable backpressure for CGX mapped PF/VFs */
1500 	if (!is_otx2_lbkvf(pf->pdev))
1501 		otx2_nix_config_bp(pf, true);
1502 
1503 	/* Init Auras and pools used by NIX RQ, for free buffer ptrs */
1504 	err = otx2_rq_aura_pool_init(pf);
1505 	if (err) {
1506 		mutex_unlock(&mbox->lock);
1507 		goto err_free_nix_lf;
1508 	}
1509 	/* Init Auras and pools used by NIX SQ, for queueing SQEs */
1510 	err = otx2_sq_aura_pool_init(pf);
1511 	if (err) {
1512 		mutex_unlock(&mbox->lock);
1513 		goto err_free_rq_ptrs;
1514 	}
1515 
1516 	err = otx2_txsch_alloc(pf);
1517 	if (err) {
1518 		mutex_unlock(&mbox->lock);
1519 		goto err_free_sq_ptrs;
1520 	}
1521 
1522 #ifdef CONFIG_DCB
1523 	if (pf->pfc_en) {
1524 		err = otx2_pfc_txschq_alloc(pf);
1525 		if (err) {
1526 			mutex_unlock(&mbox->lock);
1527 			goto err_free_sq_ptrs;
1528 		}
1529 	}
1530 #endif
1531 
1532 	err = otx2_config_nix_queues(pf);
1533 	if (err) {
1534 		mutex_unlock(&mbox->lock);
1535 		goto err_free_txsch;
1536 	}
1537 
1538 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
1539 		err = otx2_txschq_config(pf, lvl, 0, false);
1540 		if (err) {
1541 			mutex_unlock(&mbox->lock);
1542 			goto err_free_nix_queues;
1543 		}
1544 	}
1545 
1546 #ifdef CONFIG_DCB
1547 	if (pf->pfc_en) {
1548 		err = otx2_pfc_txschq_config(pf);
1549 		if (err) {
1550 			mutex_unlock(&mbox->lock);
1551 			goto err_free_nix_queues;
1552 		}
1553 	}
1554 #endif
1555 
1556 	mutex_unlock(&mbox->lock);
1557 	return err;
1558 
1559 err_free_nix_queues:
1560 	otx2_free_sq_res(pf);
1561 	otx2_free_cq_res(pf);
1562 	otx2_ctx_disable(mbox, NIX_AQ_CTYPE_RQ, false);
1563 err_free_txsch:
1564 	otx2_txschq_stop(pf);
1565 err_free_sq_ptrs:
1566 	otx2_sq_free_sqbs(pf);
1567 err_free_rq_ptrs:
1568 	otx2_free_aura_ptr(pf, AURA_NIX_RQ);
1569 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_POOL, true);
1570 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_AURA, true);
1571 	otx2_aura_pool_free(pf);
1572 err_free_nix_lf:
1573 	mutex_lock(&mbox->lock);
1574 	free_req = otx2_mbox_alloc_msg_nix_lf_free(mbox);
1575 	if (free_req) {
1576 		free_req->flags = NIX_LF_DISABLE_FLOWS;
1577 		if (otx2_sync_mbox_msg(mbox))
1578 			dev_err(pf->dev, "%s failed to free nixlf\n", __func__);
1579 	}
1580 err_free_npa_lf:
1581 	/* Reset NPA LF */
1582 	req = otx2_mbox_alloc_msg_npa_lf_free(mbox);
1583 	if (req) {
1584 		if (otx2_sync_mbox_msg(mbox))
1585 			dev_err(pf->dev, "%s failed to free npalf\n", __func__);
1586 	}
1587 exit:
1588 	mutex_unlock(&mbox->lock);
1589 	return err;
1590 }
1591 
1592 static void otx2_free_hw_resources(struct otx2_nic *pf)
1593 {
1594 	struct otx2_qset *qset = &pf->qset;
1595 	struct nix_lf_free_req *free_req;
1596 	struct mbox *mbox = &pf->mbox;
1597 	struct otx2_cq_queue *cq;
1598 	struct otx2_pool *pool;
1599 	struct msg_req *req;
1600 	int pool_id;
1601 	int qidx;
1602 
1603 	/* Ensure all SQE are processed */
1604 	otx2_sqb_flush(pf);
1605 
1606 	/* Stop transmission */
1607 	otx2_txschq_stop(pf);
1608 
1609 #ifdef CONFIG_DCB
1610 	if (pf->pfc_en)
1611 		otx2_pfc_txschq_stop(pf);
1612 #endif
1613 
1614 	otx2_clean_qos_queues(pf);
1615 
1616 	mutex_lock(&mbox->lock);
1617 	/* Disable backpressure */
1618 	if (!(pf->pcifunc & RVU_PFVF_FUNC_MASK))
1619 		otx2_nix_config_bp(pf, false);
1620 	mutex_unlock(&mbox->lock);
1621 
1622 	/* Disable RQs */
1623 	otx2_ctx_disable(mbox, NIX_AQ_CTYPE_RQ, false);
1624 
1625 	/*Dequeue all CQEs */
1626 	for (qidx = 0; qidx < qset->cq_cnt; qidx++) {
1627 		cq = &qset->cq[qidx];
1628 		if (cq->cq_type == CQ_RX)
1629 			otx2_cleanup_rx_cqes(pf, cq, qidx);
1630 		else
1631 			otx2_cleanup_tx_cqes(pf, cq);
1632 	}
1633 	otx2_free_pending_sqe(pf);
1634 
1635 	otx2_free_sq_res(pf);
1636 
1637 	/* Free RQ buffer pointers*/
1638 	otx2_free_aura_ptr(pf, AURA_NIX_RQ);
1639 
1640 	for (qidx = 0; qidx < pf->hw.rx_queues; qidx++) {
1641 		pool_id = otx2_get_pool_idx(pf, AURA_NIX_RQ, qidx);
1642 		pool = &pf->qset.pool[pool_id];
1643 		page_pool_destroy(pool->page_pool);
1644 		pool->page_pool = NULL;
1645 	}
1646 
1647 	otx2_free_cq_res(pf);
1648 
1649 	/* Free all ingress bandwidth profiles allocated */
1650 	cn10k_free_all_ipolicers(pf);
1651 
1652 	mutex_lock(&mbox->lock);
1653 	/* Reset NIX LF */
1654 	free_req = otx2_mbox_alloc_msg_nix_lf_free(mbox);
1655 	if (free_req) {
1656 		free_req->flags = NIX_LF_DISABLE_FLOWS;
1657 		if (!(pf->flags & OTX2_FLAG_PF_SHUTDOWN))
1658 			free_req->flags |= NIX_LF_DONT_FREE_TX_VTAG;
1659 		if (otx2_sync_mbox_msg(mbox))
1660 			dev_err(pf->dev, "%s failed to free nixlf\n", __func__);
1661 	}
1662 	mutex_unlock(&mbox->lock);
1663 
1664 	/* Disable NPA Pool and Aura hw context */
1665 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_POOL, true);
1666 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_AURA, true);
1667 	otx2_aura_pool_free(pf);
1668 
1669 	mutex_lock(&mbox->lock);
1670 	/* Reset NPA LF */
1671 	req = otx2_mbox_alloc_msg_npa_lf_free(mbox);
1672 	if (req) {
1673 		if (otx2_sync_mbox_msg(mbox))
1674 			dev_err(pf->dev, "%s failed to free npalf\n", __func__);
1675 	}
1676 	mutex_unlock(&mbox->lock);
1677 }
1678 
1679 static bool otx2_promisc_use_mce_list(struct otx2_nic *pfvf)
1680 {
1681 	int vf;
1682 
1683 	/* The AF driver will determine whether to allow the VF netdev or not */
1684 	if (is_otx2_vf(pfvf->pcifunc))
1685 		return true;
1686 
1687 	/* check if there are any trusted VFs associated with the PF netdev */
1688 	for (vf = 0; vf < pci_num_vf(pfvf->pdev); vf++)
1689 		if (pfvf->vf_configs[vf].trusted)
1690 			return true;
1691 	return false;
1692 }
1693 
1694 static void otx2_do_set_rx_mode(struct otx2_nic *pf)
1695 {
1696 	struct net_device *netdev = pf->netdev;
1697 	struct nix_rx_mode *req;
1698 	bool promisc = false;
1699 
1700 	if (!(netdev->flags & IFF_UP))
1701 		return;
1702 
1703 	if ((netdev->flags & IFF_PROMISC) ||
1704 	    (netdev_uc_count(netdev) > OTX2_MAX_UNICAST_FLOWS)) {
1705 		promisc = true;
1706 	}
1707 
1708 	/* Write unicast address to mcam entries or del from mcam */
1709 	if (!promisc && netdev->priv_flags & IFF_UNICAST_FLT)
1710 		__dev_uc_sync(netdev, otx2_add_macfilter, otx2_del_macfilter);
1711 
1712 	mutex_lock(&pf->mbox.lock);
1713 	req = otx2_mbox_alloc_msg_nix_set_rx_mode(&pf->mbox);
1714 	if (!req) {
1715 		mutex_unlock(&pf->mbox.lock);
1716 		return;
1717 	}
1718 
1719 	req->mode = NIX_RX_MODE_UCAST;
1720 
1721 	if (promisc)
1722 		req->mode |= NIX_RX_MODE_PROMISC;
1723 	if (netdev->flags & (IFF_ALLMULTI | IFF_MULTICAST))
1724 		req->mode |= NIX_RX_MODE_ALLMULTI;
1725 
1726 	if (otx2_promisc_use_mce_list(pf))
1727 		req->mode |= NIX_RX_MODE_USE_MCE;
1728 
1729 	otx2_sync_mbox_msg(&pf->mbox);
1730 	mutex_unlock(&pf->mbox.lock);
1731 }
1732 
1733 static void otx2_set_irq_coalesce(struct otx2_nic *pfvf)
1734 {
1735 	int cint;
1736 
1737 	for (cint = 0; cint < pfvf->hw.cint_cnt; cint++)
1738 		otx2_config_irq_coalescing(pfvf, cint);
1739 }
1740 
1741 static void otx2_dim_work(struct work_struct *w)
1742 {
1743 	struct dim_cq_moder cur_moder;
1744 	struct otx2_cq_poll *cq_poll;
1745 	struct otx2_nic *pfvf;
1746 	struct dim *dim;
1747 
1748 	dim = container_of(w, struct dim, work);
1749 	cur_moder = net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1750 	cq_poll = container_of(dim, struct otx2_cq_poll, dim);
1751 	pfvf = (struct otx2_nic *)cq_poll->dev;
1752 	pfvf->hw.cq_time_wait = (cur_moder.usec > CQ_TIMER_THRESH_MAX) ?
1753 		CQ_TIMER_THRESH_MAX : cur_moder.usec;
1754 	pfvf->hw.cq_ecount_wait = (cur_moder.pkts > NAPI_POLL_WEIGHT) ?
1755 		NAPI_POLL_WEIGHT : cur_moder.pkts;
1756 	otx2_set_irq_coalesce(pfvf);
1757 	dim->state = DIM_START_MEASURE;
1758 }
1759 
1760 int otx2_open(struct net_device *netdev)
1761 {
1762 	struct otx2_nic *pf = netdev_priv(netdev);
1763 	struct otx2_cq_poll *cq_poll = NULL;
1764 	struct otx2_qset *qset = &pf->qset;
1765 	int err = 0, qidx, vec;
1766 	char *irq_name;
1767 
1768 	netif_carrier_off(netdev);
1769 
1770 	/* RQ and SQs are mapped to different CQs,
1771 	 * so find out max CQ IRQs (i.e CINTs) needed.
1772 	 */
1773 	pf->hw.non_qos_queues =  pf->hw.tx_queues + pf->hw.xdp_queues;
1774 	pf->hw.cint_cnt = max3(pf->hw.rx_queues, pf->hw.tx_queues,
1775 			       pf->hw.tc_tx_queues);
1776 
1777 	pf->qset.cq_cnt = pf->hw.rx_queues + otx2_get_total_tx_queues(pf);
1778 
1779 	qset->napi = kcalloc(pf->hw.cint_cnt, sizeof(*cq_poll), GFP_KERNEL);
1780 	if (!qset->napi)
1781 		return -ENOMEM;
1782 
1783 	/* CQ size of RQ */
1784 	qset->rqe_cnt = qset->rqe_cnt ? qset->rqe_cnt : Q_COUNT(Q_SIZE_256);
1785 	/* CQ size of SQ */
1786 	qset->sqe_cnt = qset->sqe_cnt ? qset->sqe_cnt : Q_COUNT(Q_SIZE_4K);
1787 
1788 	err = -ENOMEM;
1789 	qset->cq = kcalloc(pf->qset.cq_cnt,
1790 			   sizeof(struct otx2_cq_queue), GFP_KERNEL);
1791 	if (!qset->cq)
1792 		goto err_free_mem;
1793 
1794 	qset->sq = kcalloc(otx2_get_total_tx_queues(pf),
1795 			   sizeof(struct otx2_snd_queue), GFP_KERNEL);
1796 	if (!qset->sq)
1797 		goto err_free_mem;
1798 
1799 	qset->rq = kcalloc(pf->hw.rx_queues,
1800 			   sizeof(struct otx2_rcv_queue), GFP_KERNEL);
1801 	if (!qset->rq)
1802 		goto err_free_mem;
1803 
1804 	err = otx2_init_hw_resources(pf);
1805 	if (err)
1806 		goto err_free_mem;
1807 
1808 	/* Register NAPI handler */
1809 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
1810 		cq_poll = &qset->napi[qidx];
1811 		cq_poll->cint_idx = qidx;
1812 		/* RQ0 & SQ0 are mapped to CINT0 and so on..
1813 		 * 'cq_ids[0]' points to RQ's CQ and
1814 		 * 'cq_ids[1]' points to SQ's CQ and
1815 		 * 'cq_ids[2]' points to XDP's CQ and
1816 		 */
1817 		cq_poll->cq_ids[CQ_RX] =
1818 			(qidx <  pf->hw.rx_queues) ? qidx : CINT_INVALID_CQ;
1819 		cq_poll->cq_ids[CQ_TX] = (qidx < pf->hw.tx_queues) ?
1820 				      qidx + pf->hw.rx_queues : CINT_INVALID_CQ;
1821 		if (pf->xdp_prog)
1822 			cq_poll->cq_ids[CQ_XDP] = (qidx < pf->hw.xdp_queues) ?
1823 						  (qidx + pf->hw.rx_queues +
1824 						  pf->hw.tx_queues) :
1825 						  CINT_INVALID_CQ;
1826 		else
1827 			cq_poll->cq_ids[CQ_XDP] = CINT_INVALID_CQ;
1828 
1829 		cq_poll->cq_ids[CQ_QOS] = (qidx < pf->hw.tc_tx_queues) ?
1830 					  (qidx + pf->hw.rx_queues +
1831 					   pf->hw.non_qos_queues) :
1832 					  CINT_INVALID_CQ;
1833 
1834 		cq_poll->dev = (void *)pf;
1835 		cq_poll->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_CQE;
1836 		INIT_WORK(&cq_poll->dim.work, otx2_dim_work);
1837 		netif_napi_add(netdev, &cq_poll->napi, otx2_napi_handler);
1838 		napi_enable(&cq_poll->napi);
1839 	}
1840 
1841 	/* Set maximum frame size allowed in HW */
1842 	err = otx2_hw_set_mtu(pf, netdev->mtu);
1843 	if (err)
1844 		goto err_disable_napi;
1845 
1846 	/* Setup segmentation algorithms, if failed, clear offload capability */
1847 	otx2_setup_segmentation(pf);
1848 
1849 	/* Initialize RSS */
1850 	err = otx2_rss_init(pf);
1851 	if (err)
1852 		goto err_disable_napi;
1853 
1854 	/* Register Queue IRQ handlers */
1855 	vec = pf->hw.nix_msixoff + NIX_LF_QINT_VEC_START;
1856 	irq_name = &pf->hw.irq_name[vec * NAME_SIZE];
1857 
1858 	snprintf(irq_name, NAME_SIZE, "%s-qerr", pf->netdev->name);
1859 
1860 	err = request_irq(pci_irq_vector(pf->pdev, vec),
1861 			  otx2_q_intr_handler, 0, irq_name, pf);
1862 	if (err) {
1863 		dev_err(pf->dev,
1864 			"RVUPF%d: IRQ registration failed for QERR\n",
1865 			rvu_get_pf(pf->pcifunc));
1866 		goto err_disable_napi;
1867 	}
1868 
1869 	/* Enable QINT IRQ */
1870 	otx2_write64(pf, NIX_LF_QINTX_ENA_W1S(0), BIT_ULL(0));
1871 
1872 	/* Register CQ IRQ handlers */
1873 	vec = pf->hw.nix_msixoff + NIX_LF_CINT_VEC_START;
1874 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
1875 		irq_name = &pf->hw.irq_name[vec * NAME_SIZE];
1876 
1877 		snprintf(irq_name, NAME_SIZE, "%s-rxtx-%d", pf->netdev->name,
1878 			 qidx);
1879 
1880 		err = request_irq(pci_irq_vector(pf->pdev, vec),
1881 				  otx2_cq_intr_handler, 0, irq_name,
1882 				  &qset->napi[qidx]);
1883 		if (err) {
1884 			dev_err(pf->dev,
1885 				"RVUPF%d: IRQ registration failed for CQ%d\n",
1886 				rvu_get_pf(pf->pcifunc), qidx);
1887 			goto err_free_cints;
1888 		}
1889 		vec++;
1890 
1891 		otx2_config_irq_coalescing(pf, qidx);
1892 
1893 		/* Enable CQ IRQ */
1894 		otx2_write64(pf, NIX_LF_CINTX_INT(qidx), BIT_ULL(0));
1895 		otx2_write64(pf, NIX_LF_CINTX_ENA_W1S(qidx), BIT_ULL(0));
1896 	}
1897 
1898 	otx2_set_cints_affinity(pf);
1899 
1900 	if (pf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
1901 		otx2_enable_rxvlan(pf, true);
1902 
1903 	/* When reinitializing enable time stamping if it is enabled before */
1904 	if (pf->flags & OTX2_FLAG_TX_TSTAMP_ENABLED) {
1905 		pf->flags &= ~OTX2_FLAG_TX_TSTAMP_ENABLED;
1906 		otx2_config_hw_tx_tstamp(pf, true);
1907 	}
1908 	if (pf->flags & OTX2_FLAG_RX_TSTAMP_ENABLED) {
1909 		pf->flags &= ~OTX2_FLAG_RX_TSTAMP_ENABLED;
1910 		otx2_config_hw_rx_tstamp(pf, true);
1911 	}
1912 
1913 	pf->flags &= ~OTX2_FLAG_INTF_DOWN;
1914 	/* 'intf_down' may be checked on any cpu */
1915 	smp_wmb();
1916 
1917 	/* Enable QoS configuration before starting tx queues */
1918 	otx2_qos_config_txschq(pf);
1919 
1920 	/* we have already received link status notification */
1921 	if (pf->linfo.link_up && !(pf->pcifunc & RVU_PFVF_FUNC_MASK))
1922 		otx2_handle_link_event(pf);
1923 
1924 	/* Install DMAC Filters */
1925 	if (pf->flags & OTX2_FLAG_DMACFLTR_SUPPORT)
1926 		otx2_dmacflt_reinstall_flows(pf);
1927 
1928 	otx2_tc_apply_ingress_police_rules(pf);
1929 
1930 	err = otx2_rxtx_enable(pf, true);
1931 	/* If a mbox communication error happens at this point then interface
1932 	 * will end up in a state such that it is in down state but hardware
1933 	 * mcam entries are enabled to receive the packets. Hence disable the
1934 	 * packet I/O.
1935 	 */
1936 	if (err == -EIO)
1937 		goto err_disable_rxtx;
1938 	else if (err)
1939 		goto err_tx_stop_queues;
1940 
1941 	otx2_do_set_rx_mode(pf);
1942 
1943 	return 0;
1944 
1945 err_disable_rxtx:
1946 	otx2_rxtx_enable(pf, false);
1947 err_tx_stop_queues:
1948 	netif_tx_stop_all_queues(netdev);
1949 	netif_carrier_off(netdev);
1950 	pf->flags |= OTX2_FLAG_INTF_DOWN;
1951 err_free_cints:
1952 	otx2_free_cints(pf, qidx);
1953 	vec = pci_irq_vector(pf->pdev,
1954 			     pf->hw.nix_msixoff + NIX_LF_QINT_VEC_START);
1955 	otx2_write64(pf, NIX_LF_QINTX_ENA_W1C(0), BIT_ULL(0));
1956 	free_irq(vec, pf);
1957 err_disable_napi:
1958 	otx2_disable_napi(pf);
1959 	otx2_free_hw_resources(pf);
1960 err_free_mem:
1961 	kfree(qset->sq);
1962 	kfree(qset->cq);
1963 	kfree(qset->rq);
1964 	kfree(qset->napi);
1965 	return err;
1966 }
1967 EXPORT_SYMBOL(otx2_open);
1968 
1969 int otx2_stop(struct net_device *netdev)
1970 {
1971 	struct otx2_nic *pf = netdev_priv(netdev);
1972 	struct otx2_cq_poll *cq_poll = NULL;
1973 	struct otx2_qset *qset = &pf->qset;
1974 	struct otx2_rss_info *rss;
1975 	int qidx, vec, wrk;
1976 
1977 	/* If the DOWN flag is set resources are already freed */
1978 	if (pf->flags & OTX2_FLAG_INTF_DOWN)
1979 		return 0;
1980 
1981 	netif_carrier_off(netdev);
1982 	netif_tx_stop_all_queues(netdev);
1983 
1984 	pf->flags |= OTX2_FLAG_INTF_DOWN;
1985 	/* 'intf_down' may be checked on any cpu */
1986 	smp_wmb();
1987 
1988 	/* First stop packet Rx/Tx */
1989 	otx2_rxtx_enable(pf, false);
1990 
1991 	/* Clear RSS enable flag */
1992 	rss = &pf->hw.rss_info;
1993 	rss->enable = false;
1994 	if (!netif_is_rxfh_configured(netdev))
1995 		kfree(rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP]);
1996 
1997 	/* Cleanup Queue IRQ */
1998 	vec = pci_irq_vector(pf->pdev,
1999 			     pf->hw.nix_msixoff + NIX_LF_QINT_VEC_START);
2000 	otx2_write64(pf, NIX_LF_QINTX_ENA_W1C(0), BIT_ULL(0));
2001 	free_irq(vec, pf);
2002 
2003 	/* Cleanup CQ NAPI and IRQ */
2004 	vec = pf->hw.nix_msixoff + NIX_LF_CINT_VEC_START;
2005 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
2006 		/* Disable interrupt */
2007 		otx2_write64(pf, NIX_LF_CINTX_ENA_W1C(qidx), BIT_ULL(0));
2008 
2009 		synchronize_irq(pci_irq_vector(pf->pdev, vec));
2010 
2011 		cq_poll = &qset->napi[qidx];
2012 		napi_synchronize(&cq_poll->napi);
2013 		vec++;
2014 	}
2015 
2016 	netif_tx_disable(netdev);
2017 
2018 	for (wrk = 0; wrk < pf->qset.cq_cnt; wrk++)
2019 		cancel_delayed_work_sync(&pf->refill_wrk[wrk].pool_refill_work);
2020 	devm_kfree(pf->dev, pf->refill_wrk);
2021 
2022 	otx2_free_hw_resources(pf);
2023 	otx2_free_cints(pf, pf->hw.cint_cnt);
2024 	otx2_disable_napi(pf);
2025 
2026 	for (qidx = 0; qidx < netdev->num_tx_queues; qidx++)
2027 		netdev_tx_reset_queue(netdev_get_tx_queue(netdev, qidx));
2028 
2029 
2030 	kfree(qset->sq);
2031 	kfree(qset->cq);
2032 	kfree(qset->rq);
2033 	kfree(qset->napi);
2034 	/* Do not clear RQ/SQ ringsize settings */
2035 	memset_startat(qset, 0, sqe_cnt);
2036 	return 0;
2037 }
2038 EXPORT_SYMBOL(otx2_stop);
2039 
2040 static netdev_tx_t otx2_xmit(struct sk_buff *skb, struct net_device *netdev)
2041 {
2042 	struct otx2_nic *pf = netdev_priv(netdev);
2043 	int qidx = skb_get_queue_mapping(skb);
2044 	struct otx2_snd_queue *sq;
2045 	struct netdev_queue *txq;
2046 	int sq_idx;
2047 
2048 	/* XDP SQs are not mapped with TXQs
2049 	 * advance qid to derive correct sq mapped with QOS
2050 	 */
2051 	sq_idx = (qidx >= pf->hw.tx_queues) ? (qidx + pf->hw.xdp_queues) : qidx;
2052 
2053 	/* Check for minimum and maximum packet length */
2054 	if (skb->len <= ETH_HLEN ||
2055 	    (!skb_shinfo(skb)->gso_size && skb->len > pf->tx_max_pktlen)) {
2056 		dev_kfree_skb(skb);
2057 		return NETDEV_TX_OK;
2058 	}
2059 
2060 	sq = &pf->qset.sq[sq_idx];
2061 	txq = netdev_get_tx_queue(netdev, qidx);
2062 
2063 	if (!otx2_sq_append_skb(netdev, sq, skb, qidx)) {
2064 		netif_tx_stop_queue(txq);
2065 
2066 		/* Check again, incase SQBs got freed up */
2067 		smp_mb();
2068 		if (((sq->num_sqbs - *sq->aura_fc_addr) * sq->sqe_per_sqb)
2069 							> sq->sqe_thresh)
2070 			netif_tx_wake_queue(txq);
2071 
2072 		return NETDEV_TX_BUSY;
2073 	}
2074 
2075 	return NETDEV_TX_OK;
2076 }
2077 
2078 static int otx2_qos_select_htb_queue(struct otx2_nic *pf, struct sk_buff *skb,
2079 				     u16 htb_maj_id)
2080 {
2081 	u16 classid;
2082 
2083 	if ((TC_H_MAJ(skb->priority) >> 16) == htb_maj_id)
2084 		classid = TC_H_MIN(skb->priority);
2085 	else
2086 		classid = READ_ONCE(pf->qos.defcls);
2087 
2088 	if (!classid)
2089 		return 0;
2090 
2091 	return otx2_get_txq_by_classid(pf, classid);
2092 }
2093 
2094 u16 otx2_select_queue(struct net_device *netdev, struct sk_buff *skb,
2095 		      struct net_device *sb_dev)
2096 {
2097 	struct otx2_nic *pf = netdev_priv(netdev);
2098 	bool qos_enabled;
2099 #ifdef CONFIG_DCB
2100 	u8 vlan_prio;
2101 #endif
2102 	int txq;
2103 
2104 	qos_enabled = netdev->real_num_tx_queues > pf->hw.tx_queues;
2105 	if (unlikely(qos_enabled)) {
2106 		/* This smp_load_acquire() pairs with smp_store_release() in
2107 		 * otx2_qos_root_add() called from htb offload root creation
2108 		 */
2109 		u16 htb_maj_id = smp_load_acquire(&pf->qos.maj_id);
2110 
2111 		if (unlikely(htb_maj_id)) {
2112 			txq = otx2_qos_select_htb_queue(pf, skb, htb_maj_id);
2113 			if (txq > 0)
2114 				return txq;
2115 			goto process_pfc;
2116 		}
2117 	}
2118 
2119 process_pfc:
2120 #ifdef CONFIG_DCB
2121 	if (!skb_vlan_tag_present(skb))
2122 		goto pick_tx;
2123 
2124 	vlan_prio = skb->vlan_tci >> 13;
2125 	if ((vlan_prio > pf->hw.tx_queues - 1) ||
2126 	    !pf->pfc_alloc_status[vlan_prio])
2127 		goto pick_tx;
2128 
2129 	return vlan_prio;
2130 
2131 pick_tx:
2132 #endif
2133 	txq = netdev_pick_tx(netdev, skb, NULL);
2134 	if (unlikely(qos_enabled))
2135 		return txq % pf->hw.tx_queues;
2136 
2137 	return txq;
2138 }
2139 EXPORT_SYMBOL(otx2_select_queue);
2140 
2141 static netdev_features_t otx2_fix_features(struct net_device *dev,
2142 					   netdev_features_t features)
2143 {
2144 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
2145 		features |= NETIF_F_HW_VLAN_STAG_RX;
2146 	else
2147 		features &= ~NETIF_F_HW_VLAN_STAG_RX;
2148 
2149 	return features;
2150 }
2151 
2152 static void otx2_set_rx_mode(struct net_device *netdev)
2153 {
2154 	struct otx2_nic *pf = netdev_priv(netdev);
2155 
2156 	queue_work(pf->otx2_wq, &pf->rx_mode_work);
2157 }
2158 
2159 static void otx2_rx_mode_wrk_handler(struct work_struct *work)
2160 {
2161 	struct otx2_nic *pf = container_of(work, struct otx2_nic, rx_mode_work);
2162 
2163 	otx2_do_set_rx_mode(pf);
2164 }
2165 
2166 static int otx2_set_features(struct net_device *netdev,
2167 			     netdev_features_t features)
2168 {
2169 	netdev_features_t changed = features ^ netdev->features;
2170 	struct otx2_nic *pf = netdev_priv(netdev);
2171 
2172 	if ((changed & NETIF_F_LOOPBACK) && netif_running(netdev))
2173 		return otx2_cgx_config_loopback(pf,
2174 						features & NETIF_F_LOOPBACK);
2175 
2176 	if ((changed & NETIF_F_HW_VLAN_CTAG_RX) && netif_running(netdev))
2177 		return otx2_enable_rxvlan(pf,
2178 					  features & NETIF_F_HW_VLAN_CTAG_RX);
2179 
2180 	return otx2_handle_ntuple_tc_features(netdev, features);
2181 }
2182 
2183 static void otx2_reset_task(struct work_struct *work)
2184 {
2185 	struct otx2_nic *pf = container_of(work, struct otx2_nic, reset_task);
2186 
2187 	if (!netif_running(pf->netdev))
2188 		return;
2189 
2190 	rtnl_lock();
2191 	otx2_stop(pf->netdev);
2192 	pf->reset_count++;
2193 	otx2_open(pf->netdev);
2194 	netif_trans_update(pf->netdev);
2195 	rtnl_unlock();
2196 }
2197 
2198 static int otx2_config_hw_rx_tstamp(struct otx2_nic *pfvf, bool enable)
2199 {
2200 	struct msg_req *req;
2201 	int err;
2202 
2203 	if (pfvf->flags & OTX2_FLAG_RX_TSTAMP_ENABLED && enable)
2204 		return 0;
2205 
2206 	mutex_lock(&pfvf->mbox.lock);
2207 	if (enable)
2208 		req = otx2_mbox_alloc_msg_cgx_ptp_rx_enable(&pfvf->mbox);
2209 	else
2210 		req = otx2_mbox_alloc_msg_cgx_ptp_rx_disable(&pfvf->mbox);
2211 	if (!req) {
2212 		mutex_unlock(&pfvf->mbox.lock);
2213 		return -ENOMEM;
2214 	}
2215 
2216 	err = otx2_sync_mbox_msg(&pfvf->mbox);
2217 	if (err) {
2218 		mutex_unlock(&pfvf->mbox.lock);
2219 		return err;
2220 	}
2221 
2222 	mutex_unlock(&pfvf->mbox.lock);
2223 	if (enable)
2224 		pfvf->flags |= OTX2_FLAG_RX_TSTAMP_ENABLED;
2225 	else
2226 		pfvf->flags &= ~OTX2_FLAG_RX_TSTAMP_ENABLED;
2227 	return 0;
2228 }
2229 
2230 static int otx2_config_hw_tx_tstamp(struct otx2_nic *pfvf, bool enable)
2231 {
2232 	struct msg_req *req;
2233 	int err;
2234 
2235 	if (pfvf->flags & OTX2_FLAG_TX_TSTAMP_ENABLED && enable)
2236 		return 0;
2237 
2238 	mutex_lock(&pfvf->mbox.lock);
2239 	if (enable)
2240 		req = otx2_mbox_alloc_msg_nix_lf_ptp_tx_enable(&pfvf->mbox);
2241 	else
2242 		req = otx2_mbox_alloc_msg_nix_lf_ptp_tx_disable(&pfvf->mbox);
2243 	if (!req) {
2244 		mutex_unlock(&pfvf->mbox.lock);
2245 		return -ENOMEM;
2246 	}
2247 
2248 	err = otx2_sync_mbox_msg(&pfvf->mbox);
2249 	if (err) {
2250 		mutex_unlock(&pfvf->mbox.lock);
2251 		return err;
2252 	}
2253 
2254 	mutex_unlock(&pfvf->mbox.lock);
2255 	if (enable)
2256 		pfvf->flags |= OTX2_FLAG_TX_TSTAMP_ENABLED;
2257 	else
2258 		pfvf->flags &= ~OTX2_FLAG_TX_TSTAMP_ENABLED;
2259 	return 0;
2260 }
2261 
2262 int otx2_config_hwtstamp(struct net_device *netdev, struct ifreq *ifr)
2263 {
2264 	struct otx2_nic *pfvf = netdev_priv(netdev);
2265 	struct hwtstamp_config config;
2266 
2267 	if (!pfvf->ptp)
2268 		return -ENODEV;
2269 
2270 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2271 		return -EFAULT;
2272 
2273 	switch (config.tx_type) {
2274 	case HWTSTAMP_TX_OFF:
2275 		if (pfvf->flags & OTX2_FLAG_PTP_ONESTEP_SYNC)
2276 			pfvf->flags &= ~OTX2_FLAG_PTP_ONESTEP_SYNC;
2277 
2278 		cancel_delayed_work(&pfvf->ptp->synctstamp_work);
2279 		otx2_config_hw_tx_tstamp(pfvf, false);
2280 		break;
2281 	case HWTSTAMP_TX_ONESTEP_SYNC:
2282 		if (!test_bit(CN10K_PTP_ONESTEP, &pfvf->hw.cap_flag))
2283 			return -ERANGE;
2284 		pfvf->flags |= OTX2_FLAG_PTP_ONESTEP_SYNC;
2285 		schedule_delayed_work(&pfvf->ptp->synctstamp_work,
2286 				      msecs_to_jiffies(500));
2287 		fallthrough;
2288 	case HWTSTAMP_TX_ON:
2289 		otx2_config_hw_tx_tstamp(pfvf, true);
2290 		break;
2291 	default:
2292 		return -ERANGE;
2293 	}
2294 
2295 	switch (config.rx_filter) {
2296 	case HWTSTAMP_FILTER_NONE:
2297 		otx2_config_hw_rx_tstamp(pfvf, false);
2298 		break;
2299 	case HWTSTAMP_FILTER_ALL:
2300 	case HWTSTAMP_FILTER_SOME:
2301 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2302 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2303 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2304 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2305 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2306 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2307 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2308 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2309 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2310 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2311 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2312 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2313 		otx2_config_hw_rx_tstamp(pfvf, true);
2314 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2315 		break;
2316 	default:
2317 		return -ERANGE;
2318 	}
2319 
2320 	memcpy(&pfvf->tstamp, &config, sizeof(config));
2321 
2322 	return copy_to_user(ifr->ifr_data, &config,
2323 			    sizeof(config)) ? -EFAULT : 0;
2324 }
2325 EXPORT_SYMBOL(otx2_config_hwtstamp);
2326 
2327 int otx2_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
2328 {
2329 	struct otx2_nic *pfvf = netdev_priv(netdev);
2330 	struct hwtstamp_config *cfg = &pfvf->tstamp;
2331 
2332 	switch (cmd) {
2333 	case SIOCSHWTSTAMP:
2334 		return otx2_config_hwtstamp(netdev, req);
2335 	case SIOCGHWTSTAMP:
2336 		return copy_to_user(req->ifr_data, cfg,
2337 				    sizeof(*cfg)) ? -EFAULT : 0;
2338 	default:
2339 		return -EOPNOTSUPP;
2340 	}
2341 }
2342 EXPORT_SYMBOL(otx2_ioctl);
2343 
2344 static int otx2_do_set_vf_mac(struct otx2_nic *pf, int vf, const u8 *mac)
2345 {
2346 	struct npc_install_flow_req *req;
2347 	int err;
2348 
2349 	mutex_lock(&pf->mbox.lock);
2350 	req = otx2_mbox_alloc_msg_npc_install_flow(&pf->mbox);
2351 	if (!req) {
2352 		err = -ENOMEM;
2353 		goto out;
2354 	}
2355 
2356 	ether_addr_copy(req->packet.dmac, mac);
2357 	eth_broadcast_addr((u8 *)&req->mask.dmac);
2358 	req->features = BIT_ULL(NPC_DMAC);
2359 	req->channel = pf->hw.rx_chan_base;
2360 	req->intf = NIX_INTF_RX;
2361 	req->default_rule = 1;
2362 	req->append = 1;
2363 	req->vf = vf + 1;
2364 	req->op = NIX_RX_ACTION_DEFAULT;
2365 
2366 	err = otx2_sync_mbox_msg(&pf->mbox);
2367 out:
2368 	mutex_unlock(&pf->mbox.lock);
2369 	return err;
2370 }
2371 
2372 static int otx2_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
2373 {
2374 	struct otx2_nic *pf = netdev_priv(netdev);
2375 	struct pci_dev *pdev = pf->pdev;
2376 	struct otx2_vf_config *config;
2377 	int ret;
2378 
2379 	if (!netif_running(netdev))
2380 		return -EAGAIN;
2381 
2382 	if (vf >= pf->total_vfs)
2383 		return -EINVAL;
2384 
2385 	if (!is_valid_ether_addr(mac))
2386 		return -EINVAL;
2387 
2388 	config = &pf->vf_configs[vf];
2389 	ether_addr_copy(config->mac, mac);
2390 
2391 	ret = otx2_do_set_vf_mac(pf, vf, mac);
2392 	if (ret == 0)
2393 		dev_info(&pdev->dev,
2394 			 "Load/Reload VF driver\n");
2395 
2396 	return ret;
2397 }
2398 
2399 static int otx2_do_set_vf_vlan(struct otx2_nic *pf, int vf, u16 vlan, u8 qos,
2400 			       __be16 proto)
2401 {
2402 	struct otx2_flow_config *flow_cfg = pf->flow_cfg;
2403 	struct nix_vtag_config_rsp *vtag_rsp;
2404 	struct npc_delete_flow_req *del_req;
2405 	struct nix_vtag_config *vtag_req;
2406 	struct npc_install_flow_req *req;
2407 	struct otx2_vf_config *config;
2408 	int err = 0;
2409 	u32 idx;
2410 
2411 	config = &pf->vf_configs[vf];
2412 
2413 	if (!vlan && !config->vlan)
2414 		goto out;
2415 
2416 	mutex_lock(&pf->mbox.lock);
2417 
2418 	/* free old tx vtag entry */
2419 	if (config->vlan) {
2420 		vtag_req = otx2_mbox_alloc_msg_nix_vtag_cfg(&pf->mbox);
2421 		if (!vtag_req) {
2422 			err = -ENOMEM;
2423 			goto out;
2424 		}
2425 		vtag_req->cfg_type = 0;
2426 		vtag_req->tx.free_vtag0 = 1;
2427 		vtag_req->tx.vtag0_idx = config->tx_vtag_idx;
2428 
2429 		err = otx2_sync_mbox_msg(&pf->mbox);
2430 		if (err)
2431 			goto out;
2432 	}
2433 
2434 	if (!vlan && config->vlan) {
2435 		/* rx */
2436 		del_req = otx2_mbox_alloc_msg_npc_delete_flow(&pf->mbox);
2437 		if (!del_req) {
2438 			err = -ENOMEM;
2439 			goto out;
2440 		}
2441 		idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_RX_INDEX);
2442 		del_req->entry =
2443 			flow_cfg->def_ent[flow_cfg->vf_vlan_offset + idx];
2444 		err = otx2_sync_mbox_msg(&pf->mbox);
2445 		if (err)
2446 			goto out;
2447 
2448 		/* tx */
2449 		del_req = otx2_mbox_alloc_msg_npc_delete_flow(&pf->mbox);
2450 		if (!del_req) {
2451 			err = -ENOMEM;
2452 			goto out;
2453 		}
2454 		idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_TX_INDEX);
2455 		del_req->entry =
2456 			flow_cfg->def_ent[flow_cfg->vf_vlan_offset + idx];
2457 		err = otx2_sync_mbox_msg(&pf->mbox);
2458 
2459 		goto out;
2460 	}
2461 
2462 	/* rx */
2463 	req = otx2_mbox_alloc_msg_npc_install_flow(&pf->mbox);
2464 	if (!req) {
2465 		err = -ENOMEM;
2466 		goto out;
2467 	}
2468 
2469 	idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_RX_INDEX);
2470 	req->entry = flow_cfg->def_ent[flow_cfg->vf_vlan_offset + idx];
2471 	req->packet.vlan_tci = htons(vlan);
2472 	req->mask.vlan_tci = htons(VLAN_VID_MASK);
2473 	/* af fills the destination mac addr */
2474 	eth_broadcast_addr((u8 *)&req->mask.dmac);
2475 	req->features = BIT_ULL(NPC_OUTER_VID) | BIT_ULL(NPC_DMAC);
2476 	req->channel = pf->hw.rx_chan_base;
2477 	req->intf = NIX_INTF_RX;
2478 	req->vf = vf + 1;
2479 	req->op = NIX_RX_ACTION_DEFAULT;
2480 	req->vtag0_valid = true;
2481 	req->vtag0_type = NIX_AF_LFX_RX_VTAG_TYPE7;
2482 	req->set_cntr = 1;
2483 
2484 	err = otx2_sync_mbox_msg(&pf->mbox);
2485 	if (err)
2486 		goto out;
2487 
2488 	/* tx */
2489 	vtag_req = otx2_mbox_alloc_msg_nix_vtag_cfg(&pf->mbox);
2490 	if (!vtag_req) {
2491 		err = -ENOMEM;
2492 		goto out;
2493 	}
2494 
2495 	/* configure tx vtag params */
2496 	vtag_req->vtag_size = VTAGSIZE_T4;
2497 	vtag_req->cfg_type = 0; /* tx vlan cfg */
2498 	vtag_req->tx.cfg_vtag0 = 1;
2499 	vtag_req->tx.vtag0 = ((u64)ntohs(proto) << 16) | vlan;
2500 
2501 	err = otx2_sync_mbox_msg(&pf->mbox);
2502 	if (err)
2503 		goto out;
2504 
2505 	vtag_rsp = (struct nix_vtag_config_rsp *)otx2_mbox_get_rsp
2506 			(&pf->mbox.mbox, 0, &vtag_req->hdr);
2507 	if (IS_ERR(vtag_rsp)) {
2508 		err = PTR_ERR(vtag_rsp);
2509 		goto out;
2510 	}
2511 	config->tx_vtag_idx = vtag_rsp->vtag0_idx;
2512 
2513 	req = otx2_mbox_alloc_msg_npc_install_flow(&pf->mbox);
2514 	if (!req) {
2515 		err = -ENOMEM;
2516 		goto out;
2517 	}
2518 
2519 	eth_zero_addr((u8 *)&req->mask.dmac);
2520 	idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_TX_INDEX);
2521 	req->entry = flow_cfg->def_ent[flow_cfg->vf_vlan_offset + idx];
2522 	req->features = BIT_ULL(NPC_DMAC);
2523 	req->channel = pf->hw.tx_chan_base;
2524 	req->intf = NIX_INTF_TX;
2525 	req->vf = vf + 1;
2526 	req->op = NIX_TX_ACTIONOP_UCAST_DEFAULT;
2527 	req->vtag0_def = vtag_rsp->vtag0_idx;
2528 	req->vtag0_op = VTAG_INSERT;
2529 	req->set_cntr = 1;
2530 
2531 	err = otx2_sync_mbox_msg(&pf->mbox);
2532 out:
2533 	config->vlan = vlan;
2534 	mutex_unlock(&pf->mbox.lock);
2535 	return err;
2536 }
2537 
2538 static int otx2_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
2539 			    __be16 proto)
2540 {
2541 	struct otx2_nic *pf = netdev_priv(netdev);
2542 	struct pci_dev *pdev = pf->pdev;
2543 
2544 	if (!netif_running(netdev))
2545 		return -EAGAIN;
2546 
2547 	if (vf >= pci_num_vf(pdev))
2548 		return -EINVAL;
2549 
2550 	/* qos is currently unsupported */
2551 	if (vlan >= VLAN_N_VID || qos)
2552 		return -EINVAL;
2553 
2554 	if (proto != htons(ETH_P_8021Q))
2555 		return -EPROTONOSUPPORT;
2556 
2557 	if (!(pf->flags & OTX2_FLAG_VF_VLAN_SUPPORT))
2558 		return -EOPNOTSUPP;
2559 
2560 	return otx2_do_set_vf_vlan(pf, vf, vlan, qos, proto);
2561 }
2562 
2563 static int otx2_get_vf_config(struct net_device *netdev, int vf,
2564 			      struct ifla_vf_info *ivi)
2565 {
2566 	struct otx2_nic *pf = netdev_priv(netdev);
2567 	struct pci_dev *pdev = pf->pdev;
2568 	struct otx2_vf_config *config;
2569 
2570 	if (!netif_running(netdev))
2571 		return -EAGAIN;
2572 
2573 	if (vf >= pci_num_vf(pdev))
2574 		return -EINVAL;
2575 
2576 	config = &pf->vf_configs[vf];
2577 	ivi->vf = vf;
2578 	ether_addr_copy(ivi->mac, config->mac);
2579 	ivi->vlan = config->vlan;
2580 	ivi->trusted = config->trusted;
2581 
2582 	return 0;
2583 }
2584 
2585 static int otx2_xdp_xmit_tx(struct otx2_nic *pf, struct xdp_frame *xdpf,
2586 			    int qidx)
2587 {
2588 	struct page *page;
2589 	u64 dma_addr;
2590 	int err = 0;
2591 
2592 	dma_addr = otx2_dma_map_page(pf, virt_to_page(xdpf->data),
2593 				     offset_in_page(xdpf->data), xdpf->len,
2594 				     DMA_TO_DEVICE);
2595 	if (dma_mapping_error(pf->dev, dma_addr))
2596 		return -ENOMEM;
2597 
2598 	err = otx2_xdp_sq_append_pkt(pf, dma_addr, xdpf->len, qidx);
2599 	if (!err) {
2600 		otx2_dma_unmap_page(pf, dma_addr, xdpf->len, DMA_TO_DEVICE);
2601 		page = virt_to_page(xdpf->data);
2602 		put_page(page);
2603 		return -ENOMEM;
2604 	}
2605 	return 0;
2606 }
2607 
2608 static int otx2_xdp_xmit(struct net_device *netdev, int n,
2609 			 struct xdp_frame **frames, u32 flags)
2610 {
2611 	struct otx2_nic *pf = netdev_priv(netdev);
2612 	int qidx = smp_processor_id();
2613 	struct otx2_snd_queue *sq;
2614 	int drops = 0, i;
2615 
2616 	if (!netif_running(netdev))
2617 		return -ENETDOWN;
2618 
2619 	qidx += pf->hw.tx_queues;
2620 	sq = pf->xdp_prog ? &pf->qset.sq[qidx] : NULL;
2621 
2622 	/* Abort xmit if xdp queue is not */
2623 	if (unlikely(!sq))
2624 		return -ENXIO;
2625 
2626 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2627 		return -EINVAL;
2628 
2629 	for (i = 0; i < n; i++) {
2630 		struct xdp_frame *xdpf = frames[i];
2631 		int err;
2632 
2633 		err = otx2_xdp_xmit_tx(pf, xdpf, qidx);
2634 		if (err)
2635 			drops++;
2636 	}
2637 	return n - drops;
2638 }
2639 
2640 static int otx2_xdp_setup(struct otx2_nic *pf, struct bpf_prog *prog)
2641 {
2642 	struct net_device *dev = pf->netdev;
2643 	bool if_up = netif_running(pf->netdev);
2644 	struct bpf_prog *old_prog;
2645 
2646 	if (prog && dev->mtu > MAX_XDP_MTU) {
2647 		netdev_warn(dev, "Jumbo frames not yet supported with XDP\n");
2648 		return -EOPNOTSUPP;
2649 	}
2650 
2651 	if (if_up)
2652 		otx2_stop(pf->netdev);
2653 
2654 	old_prog = xchg(&pf->xdp_prog, prog);
2655 
2656 	if (old_prog)
2657 		bpf_prog_put(old_prog);
2658 
2659 	if (pf->xdp_prog)
2660 		bpf_prog_add(pf->xdp_prog, pf->hw.rx_queues - 1);
2661 
2662 	/* Network stack and XDP shared same rx queues.
2663 	 * Use separate tx queues for XDP and network stack.
2664 	 */
2665 	if (pf->xdp_prog) {
2666 		pf->hw.xdp_queues = pf->hw.rx_queues;
2667 		xdp_features_set_redirect_target(dev, false);
2668 	} else {
2669 		pf->hw.xdp_queues = 0;
2670 		xdp_features_clear_redirect_target(dev);
2671 	}
2672 
2673 	if (if_up)
2674 		otx2_open(pf->netdev);
2675 
2676 	return 0;
2677 }
2678 
2679 static int otx2_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
2680 {
2681 	struct otx2_nic *pf = netdev_priv(netdev);
2682 
2683 	switch (xdp->command) {
2684 	case XDP_SETUP_PROG:
2685 		return otx2_xdp_setup(pf, xdp->prog);
2686 	default:
2687 		return -EINVAL;
2688 	}
2689 }
2690 
2691 static int otx2_set_vf_permissions(struct otx2_nic *pf, int vf,
2692 				   int req_perm)
2693 {
2694 	struct set_vf_perm *req;
2695 	int rc;
2696 
2697 	mutex_lock(&pf->mbox.lock);
2698 	req = otx2_mbox_alloc_msg_set_vf_perm(&pf->mbox);
2699 	if (!req) {
2700 		rc = -ENOMEM;
2701 		goto out;
2702 	}
2703 
2704 	/* Let AF reset VF permissions as sriov is disabled */
2705 	if (req_perm == OTX2_RESET_VF_PERM) {
2706 		req->flags |= RESET_VF_PERM;
2707 	} else if (req_perm == OTX2_TRUSTED_VF) {
2708 		if (pf->vf_configs[vf].trusted)
2709 			req->flags |= VF_TRUSTED;
2710 	}
2711 
2712 	req->vf = vf;
2713 	rc = otx2_sync_mbox_msg(&pf->mbox);
2714 out:
2715 	mutex_unlock(&pf->mbox.lock);
2716 	return rc;
2717 }
2718 
2719 static int otx2_ndo_set_vf_trust(struct net_device *netdev, int vf,
2720 				 bool enable)
2721 {
2722 	struct otx2_nic *pf = netdev_priv(netdev);
2723 	struct pci_dev *pdev = pf->pdev;
2724 	int rc;
2725 
2726 	if (vf >= pci_num_vf(pdev))
2727 		return -EINVAL;
2728 
2729 	if (pf->vf_configs[vf].trusted == enable)
2730 		return 0;
2731 
2732 	pf->vf_configs[vf].trusted = enable;
2733 	rc = otx2_set_vf_permissions(pf, vf, OTX2_TRUSTED_VF);
2734 
2735 	if (rc) {
2736 		pf->vf_configs[vf].trusted = !enable;
2737 	} else {
2738 		netdev_info(pf->netdev, "VF %d is %strusted\n",
2739 			    vf, enable ? "" : "not ");
2740 		otx2_set_rx_mode(netdev);
2741 	}
2742 
2743 	return rc;
2744 }
2745 
2746 static const struct net_device_ops otx2_netdev_ops = {
2747 	.ndo_open		= otx2_open,
2748 	.ndo_stop		= otx2_stop,
2749 	.ndo_start_xmit		= otx2_xmit,
2750 	.ndo_select_queue	= otx2_select_queue,
2751 	.ndo_fix_features	= otx2_fix_features,
2752 	.ndo_set_mac_address    = otx2_set_mac_address,
2753 	.ndo_change_mtu		= otx2_change_mtu,
2754 	.ndo_set_rx_mode	= otx2_set_rx_mode,
2755 	.ndo_set_features	= otx2_set_features,
2756 	.ndo_tx_timeout		= otx2_tx_timeout,
2757 	.ndo_get_stats64	= otx2_get_stats64,
2758 	.ndo_eth_ioctl		= otx2_ioctl,
2759 	.ndo_set_vf_mac		= otx2_set_vf_mac,
2760 	.ndo_set_vf_vlan	= otx2_set_vf_vlan,
2761 	.ndo_get_vf_config	= otx2_get_vf_config,
2762 	.ndo_bpf		= otx2_xdp,
2763 	.ndo_xdp_xmit           = otx2_xdp_xmit,
2764 	.ndo_setup_tc		= otx2_setup_tc,
2765 	.ndo_set_vf_trust	= otx2_ndo_set_vf_trust,
2766 };
2767 
2768 static int otx2_wq_init(struct otx2_nic *pf)
2769 {
2770 	pf->otx2_wq = create_singlethread_workqueue("otx2_wq");
2771 	if (!pf->otx2_wq)
2772 		return -ENOMEM;
2773 
2774 	INIT_WORK(&pf->rx_mode_work, otx2_rx_mode_wrk_handler);
2775 	INIT_WORK(&pf->reset_task, otx2_reset_task);
2776 	return 0;
2777 }
2778 
2779 static int otx2_check_pf_usable(struct otx2_nic *nic)
2780 {
2781 	u64 rev;
2782 
2783 	rev = otx2_read64(nic, RVU_PF_BLOCK_ADDRX_DISC(BLKADDR_RVUM));
2784 	rev = (rev >> 12) & 0xFF;
2785 	/* Check if AF has setup revision for RVUM block,
2786 	 * otherwise this driver probe should be deferred
2787 	 * until AF driver comes up.
2788 	 */
2789 	if (!rev) {
2790 		dev_warn(nic->dev,
2791 			 "AF is not initialized, deferring probe\n");
2792 		return -EPROBE_DEFER;
2793 	}
2794 	return 0;
2795 }
2796 
2797 static int otx2_realloc_msix_vectors(struct otx2_nic *pf)
2798 {
2799 	struct otx2_hw *hw = &pf->hw;
2800 	int num_vec, err;
2801 
2802 	/* NPA interrupts are inot registered, so alloc only
2803 	 * upto NIX vector offset.
2804 	 */
2805 	num_vec = hw->nix_msixoff;
2806 	num_vec += NIX_LF_CINT_VEC_START + hw->max_queues;
2807 
2808 	otx2_disable_mbox_intr(pf);
2809 	pci_free_irq_vectors(hw->pdev);
2810 	err = pci_alloc_irq_vectors(hw->pdev, num_vec, num_vec, PCI_IRQ_MSIX);
2811 	if (err < 0) {
2812 		dev_err(pf->dev, "%s: Failed to realloc %d IRQ vectors\n",
2813 			__func__, num_vec);
2814 		return err;
2815 	}
2816 
2817 	return otx2_register_mbox_intr(pf, false);
2818 }
2819 
2820 static int otx2_sriov_vfcfg_init(struct otx2_nic *pf)
2821 {
2822 	int i;
2823 
2824 	pf->vf_configs = devm_kcalloc(pf->dev, pf->total_vfs,
2825 				      sizeof(struct otx2_vf_config),
2826 				      GFP_KERNEL);
2827 	if (!pf->vf_configs)
2828 		return -ENOMEM;
2829 
2830 	for (i = 0; i < pf->total_vfs; i++) {
2831 		pf->vf_configs[i].pf = pf;
2832 		pf->vf_configs[i].intf_down = true;
2833 		pf->vf_configs[i].trusted = false;
2834 		INIT_DELAYED_WORK(&pf->vf_configs[i].link_event_work,
2835 				  otx2_vf_link_event_task);
2836 	}
2837 
2838 	return 0;
2839 }
2840 
2841 static void otx2_sriov_vfcfg_cleanup(struct otx2_nic *pf)
2842 {
2843 	int i;
2844 
2845 	if (!pf->vf_configs)
2846 		return;
2847 
2848 	for (i = 0; i < pf->total_vfs; i++) {
2849 		cancel_delayed_work_sync(&pf->vf_configs[i].link_event_work);
2850 		otx2_set_vf_permissions(pf, i, OTX2_RESET_VF_PERM);
2851 	}
2852 }
2853 
2854 static int otx2_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2855 {
2856 	struct device *dev = &pdev->dev;
2857 	int err, qcount, qos_txqs;
2858 	struct net_device *netdev;
2859 	struct otx2_nic *pf;
2860 	struct otx2_hw *hw;
2861 	int num_vec;
2862 
2863 	err = pcim_enable_device(pdev);
2864 	if (err) {
2865 		dev_err(dev, "Failed to enable PCI device\n");
2866 		return err;
2867 	}
2868 
2869 	err = pci_request_regions(pdev, DRV_NAME);
2870 	if (err) {
2871 		dev_err(dev, "PCI request regions failed 0x%x\n", err);
2872 		return err;
2873 	}
2874 
2875 	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(48));
2876 	if (err) {
2877 		dev_err(dev, "DMA mask config failed, abort\n");
2878 		goto err_release_regions;
2879 	}
2880 
2881 	pci_set_master(pdev);
2882 
2883 	/* Set number of queues */
2884 	qcount = min_t(int, num_online_cpus(), OTX2_MAX_CQ_CNT);
2885 	qos_txqs = min_t(int, qcount, OTX2_QOS_MAX_LEAF_NODES);
2886 
2887 	netdev = alloc_etherdev_mqs(sizeof(*pf), qcount + qos_txqs, qcount);
2888 	if (!netdev) {
2889 		err = -ENOMEM;
2890 		goto err_release_regions;
2891 	}
2892 
2893 	pci_set_drvdata(pdev, netdev);
2894 	SET_NETDEV_DEV(netdev, &pdev->dev);
2895 	pf = netdev_priv(netdev);
2896 	pf->netdev = netdev;
2897 	pf->pdev = pdev;
2898 	pf->dev = dev;
2899 	pf->total_vfs = pci_sriov_get_totalvfs(pdev);
2900 	pf->flags |= OTX2_FLAG_INTF_DOWN;
2901 
2902 	hw = &pf->hw;
2903 	hw->pdev = pdev;
2904 	hw->rx_queues = qcount;
2905 	hw->tx_queues = qcount;
2906 	hw->non_qos_queues = qcount;
2907 	hw->max_queues = qcount;
2908 	hw->rbuf_len = OTX2_DEFAULT_RBUF_LEN;
2909 	/* Use CQE of 128 byte descriptor size by default */
2910 	hw->xqe_size = 128;
2911 
2912 	num_vec = pci_msix_vec_count(pdev);
2913 	hw->irq_name = devm_kmalloc_array(&hw->pdev->dev, num_vec, NAME_SIZE,
2914 					  GFP_KERNEL);
2915 	if (!hw->irq_name) {
2916 		err = -ENOMEM;
2917 		goto err_free_netdev;
2918 	}
2919 
2920 	hw->affinity_mask = devm_kcalloc(&hw->pdev->dev, num_vec,
2921 					 sizeof(cpumask_var_t), GFP_KERNEL);
2922 	if (!hw->affinity_mask) {
2923 		err = -ENOMEM;
2924 		goto err_free_netdev;
2925 	}
2926 
2927 	/* Map CSRs */
2928 	pf->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
2929 	if (!pf->reg_base) {
2930 		dev_err(dev, "Unable to map physical function CSRs, aborting\n");
2931 		err = -ENOMEM;
2932 		goto err_free_netdev;
2933 	}
2934 
2935 	err = otx2_check_pf_usable(pf);
2936 	if (err)
2937 		goto err_free_netdev;
2938 
2939 	err = pci_alloc_irq_vectors(hw->pdev, RVU_PF_INT_VEC_CNT,
2940 				    RVU_PF_INT_VEC_CNT, PCI_IRQ_MSIX);
2941 	if (err < 0) {
2942 		dev_err(dev, "%s: Failed to alloc %d IRQ vectors\n",
2943 			__func__, num_vec);
2944 		goto err_free_netdev;
2945 	}
2946 
2947 	otx2_setup_dev_hw_settings(pf);
2948 
2949 	/* Init PF <=> AF mailbox stuff */
2950 	err = otx2_pfaf_mbox_init(pf);
2951 	if (err)
2952 		goto err_free_irq_vectors;
2953 
2954 	/* Register mailbox interrupt */
2955 	err = otx2_register_mbox_intr(pf, true);
2956 	if (err)
2957 		goto err_mbox_destroy;
2958 
2959 	/* Request AF to attach NPA and NIX LFs to this PF.
2960 	 * NIX and NPA LFs are needed for this PF to function as a NIC.
2961 	 */
2962 	err = otx2_attach_npa_nix(pf);
2963 	if (err)
2964 		goto err_disable_mbox_intr;
2965 
2966 	err = otx2_realloc_msix_vectors(pf);
2967 	if (err)
2968 		goto err_detach_rsrc;
2969 
2970 	err = otx2_set_real_num_queues(netdev, hw->tx_queues, hw->rx_queues);
2971 	if (err)
2972 		goto err_detach_rsrc;
2973 
2974 	err = cn10k_lmtst_init(pf);
2975 	if (err)
2976 		goto err_detach_rsrc;
2977 
2978 	/* Assign default mac address */
2979 	otx2_get_mac_from_af(netdev);
2980 
2981 	/* Don't check for error.  Proceed without ptp */
2982 	otx2_ptp_init(pf);
2983 
2984 	/* NPA's pool is a stack to which SW frees buffer pointers via Aura.
2985 	 * HW allocates buffer pointer from stack and uses it for DMA'ing
2986 	 * ingress packet. In some scenarios HW can free back allocated buffer
2987 	 * pointers to pool. This makes it impossible for SW to maintain a
2988 	 * parallel list where physical addresses of buffer pointers (IOVAs)
2989 	 * given to HW can be saved for later reference.
2990 	 *
2991 	 * So the only way to convert Rx packet's buffer address is to use
2992 	 * IOMMU's iova_to_phys() handler which translates the address by
2993 	 * walking through the translation tables.
2994 	 */
2995 	pf->iommu_domain = iommu_get_domain_for_dev(dev);
2996 
2997 	netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_IP_CSUM |
2998 			       NETIF_F_IPV6_CSUM | NETIF_F_RXHASH |
2999 			       NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
3000 			       NETIF_F_GSO_UDP_L4);
3001 	netdev->features |= netdev->hw_features;
3002 
3003 	err = otx2_mcam_flow_init(pf);
3004 	if (err)
3005 		goto err_ptp_destroy;
3006 
3007 	err = cn10k_mcs_init(pf);
3008 	if (err)
3009 		goto err_del_mcam_entries;
3010 
3011 	if (pf->flags & OTX2_FLAG_NTUPLE_SUPPORT)
3012 		netdev->hw_features |= NETIF_F_NTUPLE;
3013 
3014 	if (pf->flags & OTX2_FLAG_UCAST_FLTR_SUPPORT)
3015 		netdev->priv_flags |= IFF_UNICAST_FLT;
3016 
3017 	/* Support TSO on tag interface */
3018 	netdev->vlan_features |= netdev->features;
3019 	netdev->hw_features  |= NETIF_F_HW_VLAN_CTAG_TX |
3020 				NETIF_F_HW_VLAN_STAG_TX;
3021 	if (pf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
3022 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX |
3023 				       NETIF_F_HW_VLAN_STAG_RX;
3024 	netdev->features |= netdev->hw_features;
3025 
3026 	/* HW supports tc offload but mutually exclusive with n-tuple filters */
3027 	if (pf->flags & OTX2_FLAG_TC_FLOWER_SUPPORT)
3028 		netdev->hw_features |= NETIF_F_HW_TC;
3029 
3030 	netdev->hw_features |= NETIF_F_LOOPBACK | NETIF_F_RXALL;
3031 
3032 	netif_set_tso_max_segs(netdev, OTX2_MAX_GSO_SEGS);
3033 	netdev->watchdog_timeo = OTX2_TX_TIMEOUT;
3034 
3035 	netdev->netdev_ops = &otx2_netdev_ops;
3036 	netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT;
3037 
3038 	netdev->min_mtu = OTX2_MIN_MTU;
3039 	netdev->max_mtu = otx2_get_max_mtu(pf);
3040 
3041 	err = register_netdev(netdev);
3042 	if (err) {
3043 		dev_err(dev, "Failed to register netdevice\n");
3044 		goto err_mcs_free;
3045 	}
3046 
3047 	err = otx2_wq_init(pf);
3048 	if (err)
3049 		goto err_unreg_netdev;
3050 
3051 	otx2_set_ethtool_ops(netdev);
3052 
3053 	err = otx2_init_tc(pf);
3054 	if (err)
3055 		goto err_mcam_flow_del;
3056 
3057 	err = otx2_register_dl(pf);
3058 	if (err)
3059 		goto err_mcam_flow_del;
3060 
3061 	/* Initialize SR-IOV resources */
3062 	err = otx2_sriov_vfcfg_init(pf);
3063 	if (err)
3064 		goto err_pf_sriov_init;
3065 
3066 	/* Enable link notifications */
3067 	otx2_cgx_config_linkevents(pf, true);
3068 
3069 #ifdef CONFIG_DCB
3070 	err = otx2_dcbnl_set_ops(netdev);
3071 	if (err)
3072 		goto err_pf_sriov_init;
3073 #endif
3074 
3075 	otx2_qos_init(pf, qos_txqs);
3076 
3077 	return 0;
3078 
3079 err_pf_sriov_init:
3080 	otx2_shutdown_tc(pf);
3081 err_mcam_flow_del:
3082 	otx2_mcam_flow_del(pf);
3083 err_unreg_netdev:
3084 	unregister_netdev(netdev);
3085 err_mcs_free:
3086 	cn10k_mcs_free(pf);
3087 err_del_mcam_entries:
3088 	otx2_mcam_flow_del(pf);
3089 err_ptp_destroy:
3090 	otx2_ptp_destroy(pf);
3091 err_detach_rsrc:
3092 	if (pf->hw.lmt_info)
3093 		free_percpu(pf->hw.lmt_info);
3094 	if (test_bit(CN10K_LMTST, &pf->hw.cap_flag))
3095 		qmem_free(pf->dev, pf->dync_lmt);
3096 	otx2_detach_resources(&pf->mbox);
3097 err_disable_mbox_intr:
3098 	otx2_disable_mbox_intr(pf);
3099 err_mbox_destroy:
3100 	otx2_pfaf_mbox_destroy(pf);
3101 err_free_irq_vectors:
3102 	pci_free_irq_vectors(hw->pdev);
3103 err_free_netdev:
3104 	pci_set_drvdata(pdev, NULL);
3105 	free_netdev(netdev);
3106 err_release_regions:
3107 	pci_release_regions(pdev);
3108 	return err;
3109 }
3110 
3111 static void otx2_vf_link_event_task(struct work_struct *work)
3112 {
3113 	struct otx2_vf_config *config;
3114 	struct cgx_link_info_msg *req;
3115 	struct mbox_msghdr *msghdr;
3116 	struct delayed_work *dwork;
3117 	struct otx2_nic *pf;
3118 	int vf_idx;
3119 
3120 	config = container_of(work, struct otx2_vf_config,
3121 			      link_event_work.work);
3122 	vf_idx = config - config->pf->vf_configs;
3123 	pf = config->pf;
3124 
3125 	if (config->intf_down)
3126 		return;
3127 
3128 	mutex_lock(&pf->mbox.lock);
3129 
3130 	dwork = &config->link_event_work;
3131 
3132 	if (!otx2_mbox_wait_for_zero(&pf->mbox_pfvf[0].mbox_up, vf_idx)) {
3133 		schedule_delayed_work(dwork, msecs_to_jiffies(100));
3134 		mutex_unlock(&pf->mbox.lock);
3135 		return;
3136 	}
3137 
3138 	msghdr = otx2_mbox_alloc_msg_rsp(&pf->mbox_pfvf[0].mbox_up, vf_idx,
3139 					 sizeof(*req), sizeof(struct msg_rsp));
3140 	if (!msghdr) {
3141 		dev_err(pf->dev, "Failed to create VF%d link event\n", vf_idx);
3142 		mutex_unlock(&pf->mbox.lock);
3143 		return;
3144 	}
3145 
3146 	req = (struct cgx_link_info_msg *)msghdr;
3147 	req->hdr.id = MBOX_MSG_CGX_LINK_EVENT;
3148 	req->hdr.sig = OTX2_MBOX_REQ_SIG;
3149 	memcpy(&req->link_info, &pf->linfo, sizeof(req->link_info));
3150 
3151 	otx2_mbox_wait_for_zero(&pf->mbox_pfvf[0].mbox_up, vf_idx);
3152 
3153 	otx2_sync_mbox_up_msg(&pf->mbox_pfvf[0], vf_idx);
3154 
3155 	mutex_unlock(&pf->mbox.lock);
3156 }
3157 
3158 static int otx2_sriov_enable(struct pci_dev *pdev, int numvfs)
3159 {
3160 	struct net_device *netdev = pci_get_drvdata(pdev);
3161 	struct otx2_nic *pf = netdev_priv(netdev);
3162 	int ret;
3163 
3164 	/* Init PF <=> VF mailbox stuff */
3165 	ret = otx2_pfvf_mbox_init(pf, numvfs);
3166 	if (ret)
3167 		return ret;
3168 
3169 	ret = otx2_register_pfvf_mbox_intr(pf, numvfs);
3170 	if (ret)
3171 		goto free_mbox;
3172 
3173 	ret = otx2_pf_flr_init(pf, numvfs);
3174 	if (ret)
3175 		goto free_intr;
3176 
3177 	ret = otx2_register_flr_me_intr(pf, numvfs);
3178 	if (ret)
3179 		goto free_flr;
3180 
3181 	ret = pci_enable_sriov(pdev, numvfs);
3182 	if (ret)
3183 		goto free_flr_intr;
3184 
3185 	return numvfs;
3186 free_flr_intr:
3187 	otx2_disable_flr_me_intr(pf);
3188 free_flr:
3189 	otx2_flr_wq_destroy(pf);
3190 free_intr:
3191 	otx2_disable_pfvf_mbox_intr(pf, numvfs);
3192 free_mbox:
3193 	otx2_pfvf_mbox_destroy(pf);
3194 	return ret;
3195 }
3196 
3197 static int otx2_sriov_disable(struct pci_dev *pdev)
3198 {
3199 	struct net_device *netdev = pci_get_drvdata(pdev);
3200 	struct otx2_nic *pf = netdev_priv(netdev);
3201 	int numvfs = pci_num_vf(pdev);
3202 
3203 	if (!numvfs)
3204 		return 0;
3205 
3206 	pci_disable_sriov(pdev);
3207 
3208 	otx2_disable_flr_me_intr(pf);
3209 	otx2_flr_wq_destroy(pf);
3210 	otx2_disable_pfvf_mbox_intr(pf, numvfs);
3211 	otx2_pfvf_mbox_destroy(pf);
3212 
3213 	return 0;
3214 }
3215 
3216 static int otx2_sriov_configure(struct pci_dev *pdev, int numvfs)
3217 {
3218 	if (numvfs == 0)
3219 		return otx2_sriov_disable(pdev);
3220 	else
3221 		return otx2_sriov_enable(pdev, numvfs);
3222 }
3223 
3224 static void otx2_remove(struct pci_dev *pdev)
3225 {
3226 	struct net_device *netdev = pci_get_drvdata(pdev);
3227 	struct otx2_nic *pf;
3228 
3229 	if (!netdev)
3230 		return;
3231 
3232 	pf = netdev_priv(netdev);
3233 
3234 	pf->flags |= OTX2_FLAG_PF_SHUTDOWN;
3235 
3236 	if (pf->flags & OTX2_FLAG_TX_TSTAMP_ENABLED)
3237 		otx2_config_hw_tx_tstamp(pf, false);
3238 	if (pf->flags & OTX2_FLAG_RX_TSTAMP_ENABLED)
3239 		otx2_config_hw_rx_tstamp(pf, false);
3240 
3241 	/* Disable 802.3x pause frames */
3242 	if (pf->flags & OTX2_FLAG_RX_PAUSE_ENABLED ||
3243 	    (pf->flags & OTX2_FLAG_TX_PAUSE_ENABLED)) {
3244 		pf->flags &= ~OTX2_FLAG_RX_PAUSE_ENABLED;
3245 		pf->flags &= ~OTX2_FLAG_TX_PAUSE_ENABLED;
3246 		otx2_config_pause_frm(pf);
3247 	}
3248 
3249 #ifdef CONFIG_DCB
3250 	/* Disable PFC config */
3251 	if (pf->pfc_en) {
3252 		pf->pfc_en = 0;
3253 		otx2_config_priority_flow_ctrl(pf);
3254 	}
3255 #endif
3256 	cancel_work_sync(&pf->reset_task);
3257 	/* Disable link notifications */
3258 	otx2_cgx_config_linkevents(pf, false);
3259 
3260 	otx2_unregister_dl(pf);
3261 	unregister_netdev(netdev);
3262 	cn10k_mcs_free(pf);
3263 	otx2_sriov_disable(pf->pdev);
3264 	otx2_sriov_vfcfg_cleanup(pf);
3265 	if (pf->otx2_wq)
3266 		destroy_workqueue(pf->otx2_wq);
3267 
3268 	otx2_ptp_destroy(pf);
3269 	otx2_mcam_flow_del(pf);
3270 	otx2_shutdown_tc(pf);
3271 	otx2_shutdown_qos(pf);
3272 	otx2_detach_resources(&pf->mbox);
3273 	if (pf->hw.lmt_info)
3274 		free_percpu(pf->hw.lmt_info);
3275 	if (test_bit(CN10K_LMTST, &pf->hw.cap_flag))
3276 		qmem_free(pf->dev, pf->dync_lmt);
3277 	otx2_disable_mbox_intr(pf);
3278 	otx2_pfaf_mbox_destroy(pf);
3279 	pci_free_irq_vectors(pf->pdev);
3280 	pci_set_drvdata(pdev, NULL);
3281 	free_netdev(netdev);
3282 
3283 	pci_release_regions(pdev);
3284 }
3285 
3286 static struct pci_driver otx2_pf_driver = {
3287 	.name = DRV_NAME,
3288 	.id_table = otx2_pf_id_table,
3289 	.probe = otx2_probe,
3290 	.shutdown = otx2_remove,
3291 	.remove = otx2_remove,
3292 	.sriov_configure = otx2_sriov_configure
3293 };
3294 
3295 static int __init otx2_rvupf_init_module(void)
3296 {
3297 	pr_info("%s: %s\n", DRV_NAME, DRV_STRING);
3298 
3299 	return pci_register_driver(&otx2_pf_driver);
3300 }
3301 
3302 static void __exit otx2_rvupf_cleanup_module(void)
3303 {
3304 	pci_unregister_driver(&otx2_pf_driver);
3305 }
3306 
3307 module_init(otx2_rvupf_init_module);
3308 module_exit(otx2_rvupf_cleanup_module);
3309