xref: /openbmc/linux/drivers/net/ethernet/marvell/octeontx2/nic/otx2_pf.c (revision 630dce2810b9f09d312aed4189300e785254c24b)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Marvell OcteonTx2 RVU Physcial Function ethernet driver
3  *
4  * Copyright (C) 2020 Marvell International Ltd.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  */
10 
11 #include <linux/module.h>
12 #include <linux/interrupt.h>
13 #include <linux/pci.h>
14 #include <linux/etherdevice.h>
15 #include <linux/of.h>
16 #include <linux/if_vlan.h>
17 #include <linux/iommu.h>
18 #include <net/ip.h>
19 
20 #include "otx2_reg.h"
21 #include "otx2_common.h"
22 #include "otx2_txrx.h"
23 #include "otx2_struct.h"
24 #include "otx2_ptp.h"
25 #include <rvu_trace.h>
26 
27 #define DRV_NAME	"octeontx2-nicpf"
28 #define DRV_STRING	"Marvell OcteonTX2 NIC Physical Function Driver"
29 
30 /* Supported devices */
31 static const struct pci_device_id otx2_pf_id_table[] = {
32 	{ PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, PCI_DEVID_OCTEONTX2_RVU_PF) },
33 	{ 0, }  /* end of table */
34 };
35 
36 MODULE_AUTHOR("Sunil Goutham <sgoutham@marvell.com>");
37 MODULE_DESCRIPTION(DRV_STRING);
38 MODULE_LICENSE("GPL v2");
39 MODULE_DEVICE_TABLE(pci, otx2_pf_id_table);
40 
41 enum {
42 	TYPE_PFAF,
43 	TYPE_PFVF,
44 };
45 
46 static int otx2_config_hw_tx_tstamp(struct otx2_nic *pfvf, bool enable);
47 static int otx2_config_hw_rx_tstamp(struct otx2_nic *pfvf, bool enable);
48 
49 static int otx2_change_mtu(struct net_device *netdev, int new_mtu)
50 {
51 	bool if_up = netif_running(netdev);
52 	int err = 0;
53 
54 	if (if_up)
55 		otx2_stop(netdev);
56 
57 	netdev_info(netdev, "Changing MTU from %d to %d\n",
58 		    netdev->mtu, new_mtu);
59 	netdev->mtu = new_mtu;
60 
61 	if (if_up)
62 		err = otx2_open(netdev);
63 
64 	return err;
65 }
66 
67 static void otx2_disable_flr_me_intr(struct otx2_nic *pf)
68 {
69 	int irq, vfs = pf->total_vfs;
70 
71 	/* Disable VFs ME interrupts */
72 	otx2_write64(pf, RVU_PF_VFME_INT_ENA_W1CX(0), INTR_MASK(vfs));
73 	irq = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFME0);
74 	free_irq(irq, pf);
75 
76 	/* Disable VFs FLR interrupts */
77 	otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1CX(0), INTR_MASK(vfs));
78 	irq = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFFLR0);
79 	free_irq(irq, pf);
80 
81 	if (vfs <= 64)
82 		return;
83 
84 	otx2_write64(pf, RVU_PF_VFME_INT_ENA_W1CX(1), INTR_MASK(vfs - 64));
85 	irq = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFME1);
86 	free_irq(irq, pf);
87 
88 	otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1CX(1), INTR_MASK(vfs - 64));
89 	irq = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFFLR1);
90 	free_irq(irq, pf);
91 }
92 
93 static void otx2_flr_wq_destroy(struct otx2_nic *pf)
94 {
95 	if (!pf->flr_wq)
96 		return;
97 	destroy_workqueue(pf->flr_wq);
98 	pf->flr_wq = NULL;
99 	devm_kfree(pf->dev, pf->flr_wrk);
100 }
101 
102 static void otx2_flr_handler(struct work_struct *work)
103 {
104 	struct flr_work *flrwork = container_of(work, struct flr_work, work);
105 	struct otx2_nic *pf = flrwork->pf;
106 	struct mbox *mbox = &pf->mbox;
107 	struct msg_req *req;
108 	int vf, reg = 0;
109 
110 	vf = flrwork - pf->flr_wrk;
111 
112 	mutex_lock(&mbox->lock);
113 	req = otx2_mbox_alloc_msg_vf_flr(mbox);
114 	if (!req) {
115 		mutex_unlock(&mbox->lock);
116 		return;
117 	}
118 	req->hdr.pcifunc &= RVU_PFVF_FUNC_MASK;
119 	req->hdr.pcifunc |= (vf + 1) & RVU_PFVF_FUNC_MASK;
120 
121 	if (!otx2_sync_mbox_msg(&pf->mbox)) {
122 		if (vf >= 64) {
123 			reg = 1;
124 			vf = vf - 64;
125 		}
126 		/* clear transcation pending bit */
127 		otx2_write64(pf, RVU_PF_VFTRPENDX(reg), BIT_ULL(vf));
128 		otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1SX(reg), BIT_ULL(vf));
129 	}
130 
131 	mutex_unlock(&mbox->lock);
132 }
133 
134 static irqreturn_t otx2_pf_flr_intr_handler(int irq, void *pf_irq)
135 {
136 	struct otx2_nic *pf = (struct otx2_nic *)pf_irq;
137 	int reg, dev, vf, start_vf, num_reg = 1;
138 	u64 intr;
139 
140 	if (pf->total_vfs > 64)
141 		num_reg = 2;
142 
143 	for (reg = 0; reg < num_reg; reg++) {
144 		intr = otx2_read64(pf, RVU_PF_VFFLR_INTX(reg));
145 		if (!intr)
146 			continue;
147 		start_vf = 64 * reg;
148 		for (vf = 0; vf < 64; vf++) {
149 			if (!(intr & BIT_ULL(vf)))
150 				continue;
151 			dev = vf + start_vf;
152 			queue_work(pf->flr_wq, &pf->flr_wrk[dev].work);
153 			/* Clear interrupt */
154 			otx2_write64(pf, RVU_PF_VFFLR_INTX(reg), BIT_ULL(vf));
155 			/* Disable the interrupt */
156 			otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1CX(reg),
157 				     BIT_ULL(vf));
158 		}
159 	}
160 	return IRQ_HANDLED;
161 }
162 
163 static irqreturn_t otx2_pf_me_intr_handler(int irq, void *pf_irq)
164 {
165 	struct otx2_nic *pf = (struct otx2_nic *)pf_irq;
166 	int vf, reg, num_reg = 1;
167 	u64 intr;
168 
169 	if (pf->total_vfs > 64)
170 		num_reg = 2;
171 
172 	for (reg = 0; reg < num_reg; reg++) {
173 		intr = otx2_read64(pf, RVU_PF_VFME_INTX(reg));
174 		if (!intr)
175 			continue;
176 		for (vf = 0; vf < 64; vf++) {
177 			if (!(intr & BIT_ULL(vf)))
178 				continue;
179 			/* clear trpend bit */
180 			otx2_write64(pf, RVU_PF_VFTRPENDX(reg), BIT_ULL(vf));
181 			/* clear interrupt */
182 			otx2_write64(pf, RVU_PF_VFME_INTX(reg), BIT_ULL(vf));
183 		}
184 	}
185 	return IRQ_HANDLED;
186 }
187 
188 static int otx2_register_flr_me_intr(struct otx2_nic *pf, int numvfs)
189 {
190 	struct otx2_hw *hw = &pf->hw;
191 	char *irq_name;
192 	int ret;
193 
194 	/* Register ME interrupt handler*/
195 	irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFME0 * NAME_SIZE];
196 	snprintf(irq_name, NAME_SIZE, "RVUPF%d_ME0", rvu_get_pf(pf->pcifunc));
197 	ret = request_irq(pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFME0),
198 			  otx2_pf_me_intr_handler, 0, irq_name, pf);
199 	if (ret) {
200 		dev_err(pf->dev,
201 			"RVUPF: IRQ registration failed for ME0\n");
202 	}
203 
204 	/* Register FLR interrupt handler */
205 	irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFFLR0 * NAME_SIZE];
206 	snprintf(irq_name, NAME_SIZE, "RVUPF%d_FLR0", rvu_get_pf(pf->pcifunc));
207 	ret = request_irq(pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFFLR0),
208 			  otx2_pf_flr_intr_handler, 0, irq_name, pf);
209 	if (ret) {
210 		dev_err(pf->dev,
211 			"RVUPF: IRQ registration failed for FLR0\n");
212 		return ret;
213 	}
214 
215 	if (numvfs > 64) {
216 		irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFME1 * NAME_SIZE];
217 		snprintf(irq_name, NAME_SIZE, "RVUPF%d_ME1",
218 			 rvu_get_pf(pf->pcifunc));
219 		ret = request_irq(pci_irq_vector
220 				  (pf->pdev, RVU_PF_INT_VEC_VFME1),
221 				  otx2_pf_me_intr_handler, 0, irq_name, pf);
222 		if (ret) {
223 			dev_err(pf->dev,
224 				"RVUPF: IRQ registration failed for ME1\n");
225 		}
226 		irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFFLR1 * NAME_SIZE];
227 		snprintf(irq_name, NAME_SIZE, "RVUPF%d_FLR1",
228 			 rvu_get_pf(pf->pcifunc));
229 		ret = request_irq(pci_irq_vector
230 				  (pf->pdev, RVU_PF_INT_VEC_VFFLR1),
231 				  otx2_pf_flr_intr_handler, 0, irq_name, pf);
232 		if (ret) {
233 			dev_err(pf->dev,
234 				"RVUPF: IRQ registration failed for FLR1\n");
235 			return ret;
236 		}
237 	}
238 
239 	/* Enable ME interrupt for all VFs*/
240 	otx2_write64(pf, RVU_PF_VFME_INTX(0), INTR_MASK(numvfs));
241 	otx2_write64(pf, RVU_PF_VFME_INT_ENA_W1SX(0), INTR_MASK(numvfs));
242 
243 	/* Enable FLR interrupt for all VFs*/
244 	otx2_write64(pf, RVU_PF_VFFLR_INTX(0), INTR_MASK(numvfs));
245 	otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1SX(0), INTR_MASK(numvfs));
246 
247 	if (numvfs > 64) {
248 		numvfs -= 64;
249 
250 		otx2_write64(pf, RVU_PF_VFME_INTX(1), INTR_MASK(numvfs));
251 		otx2_write64(pf, RVU_PF_VFME_INT_ENA_W1SX(1),
252 			     INTR_MASK(numvfs));
253 
254 		otx2_write64(pf, RVU_PF_VFFLR_INTX(1), INTR_MASK(numvfs));
255 		otx2_write64(pf, RVU_PF_VFFLR_INT_ENA_W1SX(1),
256 			     INTR_MASK(numvfs));
257 	}
258 	return 0;
259 }
260 
261 static int otx2_pf_flr_init(struct otx2_nic *pf, int num_vfs)
262 {
263 	int vf;
264 
265 	pf->flr_wq = alloc_workqueue("otx2_pf_flr_wq",
266 				     WQ_UNBOUND | WQ_HIGHPRI, 1);
267 	if (!pf->flr_wq)
268 		return -ENOMEM;
269 
270 	pf->flr_wrk = devm_kcalloc(pf->dev, num_vfs,
271 				   sizeof(struct flr_work), GFP_KERNEL);
272 	if (!pf->flr_wrk) {
273 		destroy_workqueue(pf->flr_wq);
274 		return -ENOMEM;
275 	}
276 
277 	for (vf = 0; vf < num_vfs; vf++) {
278 		pf->flr_wrk[vf].pf = pf;
279 		INIT_WORK(&pf->flr_wrk[vf].work, otx2_flr_handler);
280 	}
281 
282 	return 0;
283 }
284 
285 static void otx2_queue_work(struct mbox *mw, struct workqueue_struct *mbox_wq,
286 			    int first, int mdevs, u64 intr, int type)
287 {
288 	struct otx2_mbox_dev *mdev;
289 	struct otx2_mbox *mbox;
290 	struct mbox_hdr *hdr;
291 	int i;
292 
293 	for (i = first; i < mdevs; i++) {
294 		/* start from 0 */
295 		if (!(intr & BIT_ULL(i - first)))
296 			continue;
297 
298 		mbox = &mw->mbox;
299 		mdev = &mbox->dev[i];
300 		if (type == TYPE_PFAF)
301 			otx2_sync_mbox_bbuf(mbox, i);
302 		hdr = mdev->mbase + mbox->rx_start;
303 		/* The hdr->num_msgs is set to zero immediately in the interrupt
304 		 * handler to  ensure that it holds a correct value next time
305 		 * when the interrupt handler is called.
306 		 * pf->mbox.num_msgs holds the data for use in pfaf_mbox_handler
307 		 * pf>mbox.up_num_msgs holds the data for use in
308 		 * pfaf_mbox_up_handler.
309 		 */
310 		if (hdr->num_msgs) {
311 			mw[i].num_msgs = hdr->num_msgs;
312 			hdr->num_msgs = 0;
313 			if (type == TYPE_PFAF)
314 				memset(mbox->hwbase + mbox->rx_start, 0,
315 				       ALIGN(sizeof(struct mbox_hdr),
316 					     sizeof(u64)));
317 
318 			queue_work(mbox_wq, &mw[i].mbox_wrk);
319 		}
320 
321 		mbox = &mw->mbox_up;
322 		mdev = &mbox->dev[i];
323 		if (type == TYPE_PFAF)
324 			otx2_sync_mbox_bbuf(mbox, i);
325 		hdr = mdev->mbase + mbox->rx_start;
326 		if (hdr->num_msgs) {
327 			mw[i].up_num_msgs = hdr->num_msgs;
328 			hdr->num_msgs = 0;
329 			if (type == TYPE_PFAF)
330 				memset(mbox->hwbase + mbox->rx_start, 0,
331 				       ALIGN(sizeof(struct mbox_hdr),
332 					     sizeof(u64)));
333 
334 			queue_work(mbox_wq, &mw[i].mbox_up_wrk);
335 		}
336 	}
337 }
338 
339 static void otx2_forward_msg_pfvf(struct otx2_mbox_dev *mdev,
340 				  struct otx2_mbox *pfvf_mbox, void *bbuf_base,
341 				  int devid)
342 {
343 	struct otx2_mbox_dev *src_mdev = mdev;
344 	int offset;
345 
346 	/* Msgs are already copied, trigger VF's mbox irq */
347 	smp_wmb();
348 
349 	offset = pfvf_mbox->trigger | (devid << pfvf_mbox->tr_shift);
350 	writeq(1, (void __iomem *)pfvf_mbox->reg_base + offset);
351 
352 	/* Restore VF's mbox bounce buffer region address */
353 	src_mdev->mbase = bbuf_base;
354 }
355 
356 static int otx2_forward_vf_mbox_msgs(struct otx2_nic *pf,
357 				     struct otx2_mbox *src_mbox,
358 				     int dir, int vf, int num_msgs)
359 {
360 	struct otx2_mbox_dev *src_mdev, *dst_mdev;
361 	struct mbox_hdr *mbox_hdr;
362 	struct mbox_hdr *req_hdr;
363 	struct mbox *dst_mbox;
364 	int dst_size, err;
365 
366 	if (dir == MBOX_DIR_PFAF) {
367 		/* Set VF's mailbox memory as PF's bounce buffer memory, so
368 		 * that explicit copying of VF's msgs to PF=>AF mbox region
369 		 * and AF=>PF responses to VF's mbox region can be avoided.
370 		 */
371 		src_mdev = &src_mbox->dev[vf];
372 		mbox_hdr = src_mbox->hwbase +
373 				src_mbox->rx_start + (vf * MBOX_SIZE);
374 
375 		dst_mbox = &pf->mbox;
376 		dst_size = dst_mbox->mbox.tx_size -
377 				ALIGN(sizeof(*mbox_hdr), MBOX_MSG_ALIGN);
378 		/* Check if msgs fit into destination area and has valid size */
379 		if (mbox_hdr->msg_size > dst_size || !mbox_hdr->msg_size)
380 			return -EINVAL;
381 
382 		dst_mdev = &dst_mbox->mbox.dev[0];
383 
384 		mutex_lock(&pf->mbox.lock);
385 		dst_mdev->mbase = src_mdev->mbase;
386 		dst_mdev->msg_size = mbox_hdr->msg_size;
387 		dst_mdev->num_msgs = num_msgs;
388 		err = otx2_sync_mbox_msg(dst_mbox);
389 		if (err) {
390 			dev_warn(pf->dev,
391 				 "AF not responding to VF%d messages\n", vf);
392 			/* restore PF mbase and exit */
393 			dst_mdev->mbase = pf->mbox.bbuf_base;
394 			mutex_unlock(&pf->mbox.lock);
395 			return err;
396 		}
397 		/* At this point, all the VF messages sent to AF are acked
398 		 * with proper responses and responses are copied to VF
399 		 * mailbox hence raise interrupt to VF.
400 		 */
401 		req_hdr = (struct mbox_hdr *)(dst_mdev->mbase +
402 					      dst_mbox->mbox.rx_start);
403 		req_hdr->num_msgs = num_msgs;
404 
405 		otx2_forward_msg_pfvf(dst_mdev, &pf->mbox_pfvf[0].mbox,
406 				      pf->mbox.bbuf_base, vf);
407 		mutex_unlock(&pf->mbox.lock);
408 	} else if (dir == MBOX_DIR_PFVF_UP) {
409 		src_mdev = &src_mbox->dev[0];
410 		mbox_hdr = src_mbox->hwbase + src_mbox->rx_start;
411 		req_hdr = (struct mbox_hdr *)(src_mdev->mbase +
412 					      src_mbox->rx_start);
413 		req_hdr->num_msgs = num_msgs;
414 
415 		dst_mbox = &pf->mbox_pfvf[0];
416 		dst_size = dst_mbox->mbox_up.tx_size -
417 				ALIGN(sizeof(*mbox_hdr), MBOX_MSG_ALIGN);
418 		/* Check if msgs fit into destination area */
419 		if (mbox_hdr->msg_size > dst_size)
420 			return -EINVAL;
421 
422 		dst_mdev = &dst_mbox->mbox_up.dev[vf];
423 		dst_mdev->mbase = src_mdev->mbase;
424 		dst_mdev->msg_size = mbox_hdr->msg_size;
425 		dst_mdev->num_msgs = mbox_hdr->num_msgs;
426 		err = otx2_sync_mbox_up_msg(dst_mbox, vf);
427 		if (err) {
428 			dev_warn(pf->dev,
429 				 "VF%d is not responding to mailbox\n", vf);
430 			return err;
431 		}
432 	} else if (dir == MBOX_DIR_VFPF_UP) {
433 		req_hdr = (struct mbox_hdr *)(src_mbox->dev[0].mbase +
434 					      src_mbox->rx_start);
435 		req_hdr->num_msgs = num_msgs;
436 		otx2_forward_msg_pfvf(&pf->mbox_pfvf->mbox_up.dev[vf],
437 				      &pf->mbox.mbox_up,
438 				      pf->mbox_pfvf[vf].bbuf_base,
439 				      0);
440 	}
441 
442 	return 0;
443 }
444 
445 static void otx2_pfvf_mbox_handler(struct work_struct *work)
446 {
447 	struct mbox_msghdr *msg = NULL;
448 	int offset, vf_idx, id, err;
449 	struct otx2_mbox_dev *mdev;
450 	struct mbox_hdr *req_hdr;
451 	struct otx2_mbox *mbox;
452 	struct mbox *vf_mbox;
453 	struct otx2_nic *pf;
454 
455 	vf_mbox = container_of(work, struct mbox, mbox_wrk);
456 	pf = vf_mbox->pfvf;
457 	vf_idx = vf_mbox - pf->mbox_pfvf;
458 
459 	mbox = &pf->mbox_pfvf[0].mbox;
460 	mdev = &mbox->dev[vf_idx];
461 	req_hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
462 
463 	offset = ALIGN(sizeof(*req_hdr), MBOX_MSG_ALIGN);
464 
465 	for (id = 0; id < vf_mbox->num_msgs; id++) {
466 		msg = (struct mbox_msghdr *)(mdev->mbase + mbox->rx_start +
467 					     offset);
468 
469 		if (msg->sig != OTX2_MBOX_REQ_SIG)
470 			goto inval_msg;
471 
472 		/* Set VF's number in each of the msg */
473 		msg->pcifunc &= RVU_PFVF_FUNC_MASK;
474 		msg->pcifunc |= (vf_idx + 1) & RVU_PFVF_FUNC_MASK;
475 		offset = msg->next_msgoff;
476 	}
477 	err = otx2_forward_vf_mbox_msgs(pf, mbox, MBOX_DIR_PFAF, vf_idx,
478 					vf_mbox->num_msgs);
479 	if (err)
480 		goto inval_msg;
481 	return;
482 
483 inval_msg:
484 	otx2_reply_invalid_msg(mbox, vf_idx, 0, msg->id);
485 	otx2_mbox_msg_send(mbox, vf_idx);
486 }
487 
488 static void otx2_pfvf_mbox_up_handler(struct work_struct *work)
489 {
490 	struct mbox *vf_mbox = container_of(work, struct mbox, mbox_up_wrk);
491 	struct otx2_nic *pf = vf_mbox->pfvf;
492 	struct otx2_mbox_dev *mdev;
493 	int offset, id, vf_idx = 0;
494 	struct mbox_hdr *rsp_hdr;
495 	struct mbox_msghdr *msg;
496 	struct otx2_mbox *mbox;
497 
498 	vf_idx = vf_mbox - pf->mbox_pfvf;
499 	mbox = &pf->mbox_pfvf[0].mbox_up;
500 	mdev = &mbox->dev[vf_idx];
501 
502 	rsp_hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
503 	offset = mbox->rx_start + ALIGN(sizeof(*rsp_hdr), MBOX_MSG_ALIGN);
504 
505 	for (id = 0; id < vf_mbox->up_num_msgs; id++) {
506 		msg = mdev->mbase + offset;
507 
508 		if (msg->id >= MBOX_MSG_MAX) {
509 			dev_err(pf->dev,
510 				"Mbox msg with unknown ID 0x%x\n", msg->id);
511 			goto end;
512 		}
513 
514 		if (msg->sig != OTX2_MBOX_RSP_SIG) {
515 			dev_err(pf->dev,
516 				"Mbox msg with wrong signature %x, ID 0x%x\n",
517 				msg->sig, msg->id);
518 			goto end;
519 		}
520 
521 		switch (msg->id) {
522 		case MBOX_MSG_CGX_LINK_EVENT:
523 			break;
524 		default:
525 			if (msg->rc)
526 				dev_err(pf->dev,
527 					"Mbox msg response has err %d, ID 0x%x\n",
528 					msg->rc, msg->id);
529 			break;
530 		}
531 
532 end:
533 		offset = mbox->rx_start + msg->next_msgoff;
534 		if (mdev->msgs_acked == (vf_mbox->up_num_msgs - 1))
535 			__otx2_mbox_reset(mbox, 0);
536 		mdev->msgs_acked++;
537 	}
538 }
539 
540 static irqreturn_t otx2_pfvf_mbox_intr_handler(int irq, void *pf_irq)
541 {
542 	struct otx2_nic *pf = (struct otx2_nic *)(pf_irq);
543 	int vfs = pf->total_vfs;
544 	struct mbox *mbox;
545 	u64 intr;
546 
547 	mbox = pf->mbox_pfvf;
548 	/* Handle VF interrupts */
549 	if (vfs > 64) {
550 		intr = otx2_read64(pf, RVU_PF_VFPF_MBOX_INTX(1));
551 		otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(1), intr);
552 		otx2_queue_work(mbox, pf->mbox_pfvf_wq, 64, vfs, intr,
553 				TYPE_PFVF);
554 		vfs -= 64;
555 	}
556 
557 	intr = otx2_read64(pf, RVU_PF_VFPF_MBOX_INTX(0));
558 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(0), intr);
559 
560 	otx2_queue_work(mbox, pf->mbox_pfvf_wq, 0, vfs, intr, TYPE_PFVF);
561 
562 	trace_otx2_msg_interrupt(mbox->mbox.pdev, "VF(s) to PF", intr);
563 
564 	return IRQ_HANDLED;
565 }
566 
567 static int otx2_pfvf_mbox_init(struct otx2_nic *pf, int numvfs)
568 {
569 	void __iomem *hwbase;
570 	struct mbox *mbox;
571 	int err, vf;
572 	u64 base;
573 
574 	if (!numvfs)
575 		return -EINVAL;
576 
577 	pf->mbox_pfvf = devm_kcalloc(&pf->pdev->dev, numvfs,
578 				     sizeof(struct mbox), GFP_KERNEL);
579 	if (!pf->mbox_pfvf)
580 		return -ENOMEM;
581 
582 	pf->mbox_pfvf_wq = alloc_workqueue("otx2_pfvf_mailbox",
583 					   WQ_UNBOUND | WQ_HIGHPRI |
584 					   WQ_MEM_RECLAIM, 1);
585 	if (!pf->mbox_pfvf_wq)
586 		return -ENOMEM;
587 
588 	base = readq((void __iomem *)((u64)pf->reg_base + RVU_PF_VF_BAR4_ADDR));
589 	hwbase = ioremap_wc(base, MBOX_SIZE * pf->total_vfs);
590 
591 	if (!hwbase) {
592 		err = -ENOMEM;
593 		goto free_wq;
594 	}
595 
596 	mbox = &pf->mbox_pfvf[0];
597 	err = otx2_mbox_init(&mbox->mbox, hwbase, pf->pdev, pf->reg_base,
598 			     MBOX_DIR_PFVF, numvfs);
599 	if (err)
600 		goto free_iomem;
601 
602 	err = otx2_mbox_init(&mbox->mbox_up, hwbase, pf->pdev, pf->reg_base,
603 			     MBOX_DIR_PFVF_UP, numvfs);
604 	if (err)
605 		goto free_iomem;
606 
607 	for (vf = 0; vf < numvfs; vf++) {
608 		mbox->pfvf = pf;
609 		INIT_WORK(&mbox->mbox_wrk, otx2_pfvf_mbox_handler);
610 		INIT_WORK(&mbox->mbox_up_wrk, otx2_pfvf_mbox_up_handler);
611 		mbox++;
612 	}
613 
614 	return 0;
615 
616 free_iomem:
617 	if (hwbase)
618 		iounmap(hwbase);
619 free_wq:
620 	destroy_workqueue(pf->mbox_pfvf_wq);
621 	return err;
622 }
623 
624 static void otx2_pfvf_mbox_destroy(struct otx2_nic *pf)
625 {
626 	struct mbox *mbox = &pf->mbox_pfvf[0];
627 
628 	if (!mbox)
629 		return;
630 
631 	if (pf->mbox_pfvf_wq) {
632 		destroy_workqueue(pf->mbox_pfvf_wq);
633 		pf->mbox_pfvf_wq = NULL;
634 	}
635 
636 	if (mbox->mbox.hwbase)
637 		iounmap(mbox->mbox.hwbase);
638 
639 	otx2_mbox_destroy(&mbox->mbox);
640 }
641 
642 static void otx2_enable_pfvf_mbox_intr(struct otx2_nic *pf, int numvfs)
643 {
644 	/* Clear PF <=> VF mailbox IRQ */
645 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(0), ~0ull);
646 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(1), ~0ull);
647 
648 	/* Enable PF <=> VF mailbox IRQ */
649 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INT_ENA_W1SX(0), INTR_MASK(numvfs));
650 	if (numvfs > 64) {
651 		numvfs -= 64;
652 		otx2_write64(pf, RVU_PF_VFPF_MBOX_INT_ENA_W1SX(1),
653 			     INTR_MASK(numvfs));
654 	}
655 }
656 
657 static void otx2_disable_pfvf_mbox_intr(struct otx2_nic *pf, int numvfs)
658 {
659 	int vector;
660 
661 	/* Disable PF <=> VF mailbox IRQ */
662 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INT_ENA_W1CX(0), ~0ull);
663 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INT_ENA_W1CX(1), ~0ull);
664 
665 	otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(0), ~0ull);
666 	vector = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFPF_MBOX0);
667 	free_irq(vector, pf);
668 
669 	if (numvfs > 64) {
670 		otx2_write64(pf, RVU_PF_VFPF_MBOX_INTX(1), ~0ull);
671 		vector = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFPF_MBOX1);
672 		free_irq(vector, pf);
673 	}
674 }
675 
676 static int otx2_register_pfvf_mbox_intr(struct otx2_nic *pf, int numvfs)
677 {
678 	struct otx2_hw *hw = &pf->hw;
679 	char *irq_name;
680 	int err;
681 
682 	/* Register MBOX0 interrupt handler */
683 	irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFPF_MBOX0 * NAME_SIZE];
684 	if (pf->pcifunc)
685 		snprintf(irq_name, NAME_SIZE,
686 			 "RVUPF%d_VF Mbox0", rvu_get_pf(pf->pcifunc));
687 	else
688 		snprintf(irq_name, NAME_SIZE, "RVUPF_VF Mbox0");
689 	err = request_irq(pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_VFPF_MBOX0),
690 			  otx2_pfvf_mbox_intr_handler, 0, irq_name, pf);
691 	if (err) {
692 		dev_err(pf->dev,
693 			"RVUPF: IRQ registration failed for PFVF mbox0 irq\n");
694 		return err;
695 	}
696 
697 	if (numvfs > 64) {
698 		/* Register MBOX1 interrupt handler */
699 		irq_name = &hw->irq_name[RVU_PF_INT_VEC_VFPF_MBOX1 * NAME_SIZE];
700 		if (pf->pcifunc)
701 			snprintf(irq_name, NAME_SIZE,
702 				 "RVUPF%d_VF Mbox1", rvu_get_pf(pf->pcifunc));
703 		else
704 			snprintf(irq_name, NAME_SIZE, "RVUPF_VF Mbox1");
705 		err = request_irq(pci_irq_vector(pf->pdev,
706 						 RVU_PF_INT_VEC_VFPF_MBOX1),
707 						 otx2_pfvf_mbox_intr_handler,
708 						 0, irq_name, pf);
709 		if (err) {
710 			dev_err(pf->dev,
711 				"RVUPF: IRQ registration failed for PFVF mbox1 irq\n");
712 			return err;
713 		}
714 	}
715 
716 	otx2_enable_pfvf_mbox_intr(pf, numvfs);
717 
718 	return 0;
719 }
720 
721 static void otx2_process_pfaf_mbox_msg(struct otx2_nic *pf,
722 				       struct mbox_msghdr *msg)
723 {
724 	int devid;
725 
726 	if (msg->id >= MBOX_MSG_MAX) {
727 		dev_err(pf->dev,
728 			"Mbox msg with unknown ID 0x%x\n", msg->id);
729 		return;
730 	}
731 
732 	if (msg->sig != OTX2_MBOX_RSP_SIG) {
733 		dev_err(pf->dev,
734 			"Mbox msg with wrong signature %x, ID 0x%x\n",
735 			 msg->sig, msg->id);
736 		return;
737 	}
738 
739 	/* message response heading VF */
740 	devid = msg->pcifunc & RVU_PFVF_FUNC_MASK;
741 	if (devid) {
742 		struct otx2_vf_config *config = &pf->vf_configs[devid - 1];
743 		struct delayed_work *dwork;
744 
745 		switch (msg->id) {
746 		case MBOX_MSG_NIX_LF_START_RX:
747 			config->intf_down = false;
748 			dwork = &config->link_event_work;
749 			schedule_delayed_work(dwork, msecs_to_jiffies(100));
750 			break;
751 		case MBOX_MSG_NIX_LF_STOP_RX:
752 			config->intf_down = true;
753 			break;
754 		}
755 
756 		return;
757 	}
758 
759 	switch (msg->id) {
760 	case MBOX_MSG_READY:
761 		pf->pcifunc = msg->pcifunc;
762 		break;
763 	case MBOX_MSG_MSIX_OFFSET:
764 		mbox_handler_msix_offset(pf, (struct msix_offset_rsp *)msg);
765 		break;
766 	case MBOX_MSG_NPA_LF_ALLOC:
767 		mbox_handler_npa_lf_alloc(pf, (struct npa_lf_alloc_rsp *)msg);
768 		break;
769 	case MBOX_MSG_NIX_LF_ALLOC:
770 		mbox_handler_nix_lf_alloc(pf, (struct nix_lf_alloc_rsp *)msg);
771 		break;
772 	case MBOX_MSG_NIX_TXSCH_ALLOC:
773 		mbox_handler_nix_txsch_alloc(pf,
774 					     (struct nix_txsch_alloc_rsp *)msg);
775 		break;
776 	case MBOX_MSG_NIX_BP_ENABLE:
777 		mbox_handler_nix_bp_enable(pf, (struct nix_bp_cfg_rsp *)msg);
778 		break;
779 	case MBOX_MSG_CGX_STATS:
780 		mbox_handler_cgx_stats(pf, (struct cgx_stats_rsp *)msg);
781 		break;
782 	default:
783 		if (msg->rc)
784 			dev_err(pf->dev,
785 				"Mbox msg response has err %d, ID 0x%x\n",
786 				msg->rc, msg->id);
787 		break;
788 	}
789 }
790 
791 static void otx2_pfaf_mbox_handler(struct work_struct *work)
792 {
793 	struct otx2_mbox_dev *mdev;
794 	struct mbox_hdr *rsp_hdr;
795 	struct mbox_msghdr *msg;
796 	struct otx2_mbox *mbox;
797 	struct mbox *af_mbox;
798 	struct otx2_nic *pf;
799 	int offset, id;
800 
801 	af_mbox = container_of(work, struct mbox, mbox_wrk);
802 	mbox = &af_mbox->mbox;
803 	mdev = &mbox->dev[0];
804 	rsp_hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
805 
806 	offset = mbox->rx_start + ALIGN(sizeof(*rsp_hdr), MBOX_MSG_ALIGN);
807 	pf = af_mbox->pfvf;
808 
809 	for (id = 0; id < af_mbox->num_msgs; id++) {
810 		msg = (struct mbox_msghdr *)(mdev->mbase + offset);
811 		otx2_process_pfaf_mbox_msg(pf, msg);
812 		offset = mbox->rx_start + msg->next_msgoff;
813 		if (mdev->msgs_acked == (af_mbox->num_msgs - 1))
814 			__otx2_mbox_reset(mbox, 0);
815 		mdev->msgs_acked++;
816 	}
817 
818 }
819 
820 static void otx2_handle_link_event(struct otx2_nic *pf)
821 {
822 	struct cgx_link_user_info *linfo = &pf->linfo;
823 	struct net_device *netdev = pf->netdev;
824 
825 	pr_info("%s NIC Link is %s %d Mbps %s duplex\n", netdev->name,
826 		linfo->link_up ? "UP" : "DOWN", linfo->speed,
827 		linfo->full_duplex ? "Full" : "Half");
828 	if (linfo->link_up) {
829 		netif_carrier_on(netdev);
830 		netif_tx_start_all_queues(netdev);
831 	} else {
832 		netif_tx_stop_all_queues(netdev);
833 		netif_carrier_off(netdev);
834 	}
835 }
836 
837 int otx2_mbox_up_handler_cgx_link_event(struct otx2_nic *pf,
838 					struct cgx_link_info_msg *msg,
839 					struct msg_rsp *rsp)
840 {
841 	int i;
842 
843 	/* Copy the link info sent by AF */
844 	pf->linfo = msg->link_info;
845 
846 	/* notify VFs about link event */
847 	for (i = 0; i < pci_num_vf(pf->pdev); i++) {
848 		struct otx2_vf_config *config = &pf->vf_configs[i];
849 		struct delayed_work *dwork = &config->link_event_work;
850 
851 		if (config->intf_down)
852 			continue;
853 
854 		schedule_delayed_work(dwork, msecs_to_jiffies(100));
855 	}
856 
857 	/* interface has not been fully configured yet */
858 	if (pf->flags & OTX2_FLAG_INTF_DOWN)
859 		return 0;
860 
861 	otx2_handle_link_event(pf);
862 	return 0;
863 }
864 
865 static int otx2_process_mbox_msg_up(struct otx2_nic *pf,
866 				    struct mbox_msghdr *req)
867 {
868 	/* Check if valid, if not reply with a invalid msg */
869 	if (req->sig != OTX2_MBOX_REQ_SIG) {
870 		otx2_reply_invalid_msg(&pf->mbox.mbox_up, 0, 0, req->id);
871 		return -ENODEV;
872 	}
873 
874 	switch (req->id) {
875 #define M(_name, _id, _fn_name, _req_type, _rsp_type)			\
876 	case _id: {							\
877 		struct _rsp_type *rsp;					\
878 		int err;						\
879 									\
880 		rsp = (struct _rsp_type *)otx2_mbox_alloc_msg(		\
881 			&pf->mbox.mbox_up, 0,				\
882 			sizeof(struct _rsp_type));			\
883 		if (!rsp)						\
884 			return -ENOMEM;					\
885 									\
886 		rsp->hdr.id = _id;					\
887 		rsp->hdr.sig = OTX2_MBOX_RSP_SIG;			\
888 		rsp->hdr.pcifunc = 0;					\
889 		rsp->hdr.rc = 0;					\
890 									\
891 		err = otx2_mbox_up_handler_ ## _fn_name(		\
892 			pf, (struct _req_type *)req, rsp);		\
893 		return err;						\
894 	}
895 MBOX_UP_CGX_MESSAGES
896 #undef M
897 		break;
898 	default:
899 		otx2_reply_invalid_msg(&pf->mbox.mbox_up, 0, 0, req->id);
900 		return -ENODEV;
901 	}
902 	return 0;
903 }
904 
905 static void otx2_pfaf_mbox_up_handler(struct work_struct *work)
906 {
907 	struct mbox *af_mbox = container_of(work, struct mbox, mbox_up_wrk);
908 	struct otx2_mbox *mbox = &af_mbox->mbox_up;
909 	struct otx2_mbox_dev *mdev = &mbox->dev[0];
910 	struct otx2_nic *pf = af_mbox->pfvf;
911 	int offset, id, devid = 0;
912 	struct mbox_hdr *rsp_hdr;
913 	struct mbox_msghdr *msg;
914 
915 	rsp_hdr = (struct mbox_hdr *)(mdev->mbase + mbox->rx_start);
916 
917 	offset = mbox->rx_start + ALIGN(sizeof(*rsp_hdr), MBOX_MSG_ALIGN);
918 
919 	for (id = 0; id < af_mbox->up_num_msgs; id++) {
920 		msg = (struct mbox_msghdr *)(mdev->mbase + offset);
921 
922 		devid = msg->pcifunc & RVU_PFVF_FUNC_MASK;
923 		/* Skip processing VF's messages */
924 		if (!devid)
925 			otx2_process_mbox_msg_up(pf, msg);
926 		offset = mbox->rx_start + msg->next_msgoff;
927 	}
928 	if (devid) {
929 		otx2_forward_vf_mbox_msgs(pf, &pf->mbox.mbox_up,
930 					  MBOX_DIR_PFVF_UP, devid - 1,
931 					  af_mbox->up_num_msgs);
932 		return;
933 	}
934 
935 	otx2_mbox_msg_send(mbox, 0);
936 }
937 
938 static irqreturn_t otx2_pfaf_mbox_intr_handler(int irq, void *pf_irq)
939 {
940 	struct otx2_nic *pf = (struct otx2_nic *)pf_irq;
941 	struct mbox *mbox;
942 
943 	/* Clear the IRQ */
944 	otx2_write64(pf, RVU_PF_INT, BIT_ULL(0));
945 
946 	mbox = &pf->mbox;
947 
948 	trace_otx2_msg_interrupt(mbox->mbox.pdev, "AF to PF", BIT_ULL(0));
949 
950 	otx2_queue_work(mbox, pf->mbox_wq, 0, 1, 1, TYPE_PFAF);
951 
952 	return IRQ_HANDLED;
953 }
954 
955 static void otx2_disable_mbox_intr(struct otx2_nic *pf)
956 {
957 	int vector = pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_AFPF_MBOX);
958 
959 	/* Disable AF => PF mailbox IRQ */
960 	otx2_write64(pf, RVU_PF_INT_ENA_W1C, BIT_ULL(0));
961 	free_irq(vector, pf);
962 }
963 
964 static int otx2_register_mbox_intr(struct otx2_nic *pf, bool probe_af)
965 {
966 	struct otx2_hw *hw = &pf->hw;
967 	struct msg_req *req;
968 	char *irq_name;
969 	int err;
970 
971 	/* Register mailbox interrupt handler */
972 	irq_name = &hw->irq_name[RVU_PF_INT_VEC_AFPF_MBOX * NAME_SIZE];
973 	snprintf(irq_name, NAME_SIZE, "RVUPFAF Mbox");
974 	err = request_irq(pci_irq_vector(pf->pdev, RVU_PF_INT_VEC_AFPF_MBOX),
975 			  otx2_pfaf_mbox_intr_handler, 0, irq_name, pf);
976 	if (err) {
977 		dev_err(pf->dev,
978 			"RVUPF: IRQ registration failed for PFAF mbox irq\n");
979 		return err;
980 	}
981 
982 	/* Enable mailbox interrupt for msgs coming from AF.
983 	 * First clear to avoid spurious interrupts, if any.
984 	 */
985 	otx2_write64(pf, RVU_PF_INT, BIT_ULL(0));
986 	otx2_write64(pf, RVU_PF_INT_ENA_W1S, BIT_ULL(0));
987 
988 	if (!probe_af)
989 		return 0;
990 
991 	/* Check mailbox communication with AF */
992 	req = otx2_mbox_alloc_msg_ready(&pf->mbox);
993 	if (!req) {
994 		otx2_disable_mbox_intr(pf);
995 		return -ENOMEM;
996 	}
997 	err = otx2_sync_mbox_msg(&pf->mbox);
998 	if (err) {
999 		dev_warn(pf->dev,
1000 			 "AF not responding to mailbox, deferring probe\n");
1001 		otx2_disable_mbox_intr(pf);
1002 		return -EPROBE_DEFER;
1003 	}
1004 
1005 	return 0;
1006 }
1007 
1008 static void otx2_pfaf_mbox_destroy(struct otx2_nic *pf)
1009 {
1010 	struct mbox *mbox = &pf->mbox;
1011 
1012 	if (pf->mbox_wq) {
1013 		destroy_workqueue(pf->mbox_wq);
1014 		pf->mbox_wq = NULL;
1015 	}
1016 
1017 	if (mbox->mbox.hwbase)
1018 		iounmap((void __iomem *)mbox->mbox.hwbase);
1019 
1020 	otx2_mbox_destroy(&mbox->mbox);
1021 	otx2_mbox_destroy(&mbox->mbox_up);
1022 }
1023 
1024 static int otx2_pfaf_mbox_init(struct otx2_nic *pf)
1025 {
1026 	struct mbox *mbox = &pf->mbox;
1027 	void __iomem *hwbase;
1028 	int err;
1029 
1030 	mbox->pfvf = pf;
1031 	pf->mbox_wq = alloc_workqueue("otx2_pfaf_mailbox",
1032 				      WQ_UNBOUND | WQ_HIGHPRI |
1033 				      WQ_MEM_RECLAIM, 1);
1034 	if (!pf->mbox_wq)
1035 		return -ENOMEM;
1036 
1037 	/* Mailbox is a reserved memory (in RAM) region shared between
1038 	 * admin function (i.e AF) and this PF, shouldn't be mapped as
1039 	 * device memory to allow unaligned accesses.
1040 	 */
1041 	hwbase = ioremap_wc(pci_resource_start(pf->pdev, PCI_MBOX_BAR_NUM),
1042 			    pci_resource_len(pf->pdev, PCI_MBOX_BAR_NUM));
1043 	if (!hwbase) {
1044 		dev_err(pf->dev, "Unable to map PFAF mailbox region\n");
1045 		err = -ENOMEM;
1046 		goto exit;
1047 	}
1048 
1049 	err = otx2_mbox_init(&mbox->mbox, hwbase, pf->pdev, pf->reg_base,
1050 			     MBOX_DIR_PFAF, 1);
1051 	if (err)
1052 		goto exit;
1053 
1054 	err = otx2_mbox_init(&mbox->mbox_up, hwbase, pf->pdev, pf->reg_base,
1055 			     MBOX_DIR_PFAF_UP, 1);
1056 	if (err)
1057 		goto exit;
1058 
1059 	err = otx2_mbox_bbuf_init(mbox, pf->pdev);
1060 	if (err)
1061 		goto exit;
1062 
1063 	INIT_WORK(&mbox->mbox_wrk, otx2_pfaf_mbox_handler);
1064 	INIT_WORK(&mbox->mbox_up_wrk, otx2_pfaf_mbox_up_handler);
1065 	mutex_init(&mbox->lock);
1066 
1067 	return 0;
1068 exit:
1069 	otx2_pfaf_mbox_destroy(pf);
1070 	return err;
1071 }
1072 
1073 static int otx2_cgx_config_linkevents(struct otx2_nic *pf, bool enable)
1074 {
1075 	struct msg_req *msg;
1076 	int err;
1077 
1078 	mutex_lock(&pf->mbox.lock);
1079 	if (enable)
1080 		msg = otx2_mbox_alloc_msg_cgx_start_linkevents(&pf->mbox);
1081 	else
1082 		msg = otx2_mbox_alloc_msg_cgx_stop_linkevents(&pf->mbox);
1083 
1084 	if (!msg) {
1085 		mutex_unlock(&pf->mbox.lock);
1086 		return -ENOMEM;
1087 	}
1088 
1089 	err = otx2_sync_mbox_msg(&pf->mbox);
1090 	mutex_unlock(&pf->mbox.lock);
1091 	return err;
1092 }
1093 
1094 static int otx2_cgx_config_loopback(struct otx2_nic *pf, bool enable)
1095 {
1096 	struct msg_req *msg;
1097 	int err;
1098 
1099 	mutex_lock(&pf->mbox.lock);
1100 	if (enable)
1101 		msg = otx2_mbox_alloc_msg_cgx_intlbk_enable(&pf->mbox);
1102 	else
1103 		msg = otx2_mbox_alloc_msg_cgx_intlbk_disable(&pf->mbox);
1104 
1105 	if (!msg) {
1106 		mutex_unlock(&pf->mbox.lock);
1107 		return -ENOMEM;
1108 	}
1109 
1110 	err = otx2_sync_mbox_msg(&pf->mbox);
1111 	mutex_unlock(&pf->mbox.lock);
1112 	return err;
1113 }
1114 
1115 int otx2_set_real_num_queues(struct net_device *netdev,
1116 			     int tx_queues, int rx_queues)
1117 {
1118 	int err;
1119 
1120 	err = netif_set_real_num_tx_queues(netdev, tx_queues);
1121 	if (err) {
1122 		netdev_err(netdev,
1123 			   "Failed to set no of Tx queues: %d\n", tx_queues);
1124 		return err;
1125 	}
1126 
1127 	err = netif_set_real_num_rx_queues(netdev, rx_queues);
1128 	if (err)
1129 		netdev_err(netdev,
1130 			   "Failed to set no of Rx queues: %d\n", rx_queues);
1131 	return err;
1132 }
1133 EXPORT_SYMBOL(otx2_set_real_num_queues);
1134 
1135 static irqreturn_t otx2_q_intr_handler(int irq, void *data)
1136 {
1137 	struct otx2_nic *pf = data;
1138 	u64 val, *ptr;
1139 	u64 qidx = 0;
1140 
1141 	/* CQ */
1142 	for (qidx = 0; qidx < pf->qset.cq_cnt; qidx++) {
1143 		ptr = otx2_get_regaddr(pf, NIX_LF_CQ_OP_INT);
1144 		val = otx2_atomic64_add((qidx << 44), ptr);
1145 
1146 		otx2_write64(pf, NIX_LF_CQ_OP_INT, (qidx << 44) |
1147 			     (val & NIX_CQERRINT_BITS));
1148 		if (!(val & (NIX_CQERRINT_BITS | BIT_ULL(42))))
1149 			continue;
1150 
1151 		if (val & BIT_ULL(42)) {
1152 			netdev_err(pf->netdev, "CQ%lld: error reading NIX_LF_CQ_OP_INT, NIX_LF_ERR_INT 0x%llx\n",
1153 				   qidx, otx2_read64(pf, NIX_LF_ERR_INT));
1154 		} else {
1155 			if (val & BIT_ULL(NIX_CQERRINT_DOOR_ERR))
1156 				netdev_err(pf->netdev, "CQ%lld: Doorbell error",
1157 					   qidx);
1158 			if (val & BIT_ULL(NIX_CQERRINT_CQE_FAULT))
1159 				netdev_err(pf->netdev, "CQ%lld: Memory fault on CQE write to LLC/DRAM",
1160 					   qidx);
1161 		}
1162 
1163 		schedule_work(&pf->reset_task);
1164 	}
1165 
1166 	/* SQ */
1167 	for (qidx = 0; qidx < pf->hw.tx_queues; qidx++) {
1168 		ptr = otx2_get_regaddr(pf, NIX_LF_SQ_OP_INT);
1169 		val = otx2_atomic64_add((qidx << 44), ptr);
1170 		otx2_write64(pf, NIX_LF_SQ_OP_INT, (qidx << 44) |
1171 			     (val & NIX_SQINT_BITS));
1172 
1173 		if (!(val & (NIX_SQINT_BITS | BIT_ULL(42))))
1174 			continue;
1175 
1176 		if (val & BIT_ULL(42)) {
1177 			netdev_err(pf->netdev, "SQ%lld: error reading NIX_LF_SQ_OP_INT, NIX_LF_ERR_INT 0x%llx\n",
1178 				   qidx, otx2_read64(pf, NIX_LF_ERR_INT));
1179 		} else {
1180 			if (val & BIT_ULL(NIX_SQINT_LMT_ERR)) {
1181 				netdev_err(pf->netdev, "SQ%lld: LMT store error NIX_LF_SQ_OP_ERR_DBG:0x%llx",
1182 					   qidx,
1183 					   otx2_read64(pf,
1184 						       NIX_LF_SQ_OP_ERR_DBG));
1185 				otx2_write64(pf, NIX_LF_SQ_OP_ERR_DBG,
1186 					     BIT_ULL(44));
1187 			}
1188 			if (val & BIT_ULL(NIX_SQINT_MNQ_ERR)) {
1189 				netdev_err(pf->netdev, "SQ%lld: Meta-descriptor enqueue error NIX_LF_MNQ_ERR_DGB:0x%llx\n",
1190 					   qidx,
1191 					   otx2_read64(pf, NIX_LF_MNQ_ERR_DBG));
1192 				otx2_write64(pf, NIX_LF_MNQ_ERR_DBG,
1193 					     BIT_ULL(44));
1194 			}
1195 			if (val & BIT_ULL(NIX_SQINT_SEND_ERR)) {
1196 				netdev_err(pf->netdev, "SQ%lld: Send error, NIX_LF_SEND_ERR_DBG 0x%llx",
1197 					   qidx,
1198 					   otx2_read64(pf,
1199 						       NIX_LF_SEND_ERR_DBG));
1200 				otx2_write64(pf, NIX_LF_SEND_ERR_DBG,
1201 					     BIT_ULL(44));
1202 			}
1203 			if (val & BIT_ULL(NIX_SQINT_SQB_ALLOC_FAIL))
1204 				netdev_err(pf->netdev, "SQ%lld: SQB allocation failed",
1205 					   qidx);
1206 		}
1207 
1208 		schedule_work(&pf->reset_task);
1209 	}
1210 
1211 	return IRQ_HANDLED;
1212 }
1213 
1214 static irqreturn_t otx2_cq_intr_handler(int irq, void *cq_irq)
1215 {
1216 	struct otx2_cq_poll *cq_poll = (struct otx2_cq_poll *)cq_irq;
1217 	struct otx2_nic *pf = (struct otx2_nic *)cq_poll->dev;
1218 	int qidx = cq_poll->cint_idx;
1219 
1220 	/* Disable interrupts.
1221 	 *
1222 	 * Completion interrupts behave in a level-triggered interrupt
1223 	 * fashion, and hence have to be cleared only after it is serviced.
1224 	 */
1225 	otx2_write64(pf, NIX_LF_CINTX_ENA_W1C(qidx), BIT_ULL(0));
1226 
1227 	/* Schedule NAPI */
1228 	napi_schedule_irqoff(&cq_poll->napi);
1229 
1230 	return IRQ_HANDLED;
1231 }
1232 
1233 static void otx2_disable_napi(struct otx2_nic *pf)
1234 {
1235 	struct otx2_qset *qset = &pf->qset;
1236 	struct otx2_cq_poll *cq_poll;
1237 	int qidx;
1238 
1239 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
1240 		cq_poll = &qset->napi[qidx];
1241 		napi_disable(&cq_poll->napi);
1242 		netif_napi_del(&cq_poll->napi);
1243 	}
1244 }
1245 
1246 static void otx2_free_cq_res(struct otx2_nic *pf)
1247 {
1248 	struct otx2_qset *qset = &pf->qset;
1249 	struct otx2_cq_queue *cq;
1250 	int qidx;
1251 
1252 	/* Disable CQs */
1253 	otx2_ctx_disable(&pf->mbox, NIX_AQ_CTYPE_CQ, false);
1254 	for (qidx = 0; qidx < qset->cq_cnt; qidx++) {
1255 		cq = &qset->cq[qidx];
1256 		qmem_free(pf->dev, cq->cqe);
1257 	}
1258 }
1259 
1260 static void otx2_free_sq_res(struct otx2_nic *pf)
1261 {
1262 	struct otx2_qset *qset = &pf->qset;
1263 	struct otx2_snd_queue *sq;
1264 	int qidx;
1265 
1266 	/* Disable SQs */
1267 	otx2_ctx_disable(&pf->mbox, NIX_AQ_CTYPE_SQ, false);
1268 	/* Free SQB pointers */
1269 	otx2_sq_free_sqbs(pf);
1270 	for (qidx = 0; qidx < pf->hw.tx_queues; qidx++) {
1271 		sq = &qset->sq[qidx];
1272 		qmem_free(pf->dev, sq->sqe);
1273 		qmem_free(pf->dev, sq->tso_hdrs);
1274 		kfree(sq->sg);
1275 		kfree(sq->sqb_ptrs);
1276 	}
1277 }
1278 
1279 static int otx2_init_hw_resources(struct otx2_nic *pf)
1280 {
1281 	struct nix_lf_free_req *free_req;
1282 	struct mbox *mbox = &pf->mbox;
1283 	struct otx2_hw *hw = &pf->hw;
1284 	struct msg_req *req;
1285 	int err = 0, lvl;
1286 
1287 	/* Set required NPA LF's pool counts
1288 	 * Auras and Pools are used in a 1:1 mapping,
1289 	 * so, aura count = pool count.
1290 	 */
1291 	hw->rqpool_cnt = hw->rx_queues;
1292 	hw->sqpool_cnt = hw->tx_queues;
1293 	hw->pool_cnt = hw->rqpool_cnt + hw->sqpool_cnt;
1294 
1295 	/* Get the size of receive buffers to allocate */
1296 	pf->rbsize = RCV_FRAG_LEN(OTX2_HW_TIMESTAMP_LEN + pf->netdev->mtu +
1297 				  OTX2_ETH_HLEN);
1298 
1299 	mutex_lock(&mbox->lock);
1300 	/* NPA init */
1301 	err = otx2_config_npa(pf);
1302 	if (err)
1303 		goto exit;
1304 
1305 	/* NIX init */
1306 	err = otx2_config_nix(pf);
1307 	if (err)
1308 		goto err_free_npa_lf;
1309 
1310 	/* Enable backpressure */
1311 	otx2_nix_config_bp(pf, true);
1312 
1313 	/* Init Auras and pools used by NIX RQ, for free buffer ptrs */
1314 	err = otx2_rq_aura_pool_init(pf);
1315 	if (err) {
1316 		mutex_unlock(&mbox->lock);
1317 		goto err_free_nix_lf;
1318 	}
1319 	/* Init Auras and pools used by NIX SQ, for queueing SQEs */
1320 	err = otx2_sq_aura_pool_init(pf);
1321 	if (err) {
1322 		mutex_unlock(&mbox->lock);
1323 		goto err_free_rq_ptrs;
1324 	}
1325 
1326 	err = otx2_txsch_alloc(pf);
1327 	if (err) {
1328 		mutex_unlock(&mbox->lock);
1329 		goto err_free_sq_ptrs;
1330 	}
1331 
1332 	err = otx2_config_nix_queues(pf);
1333 	if (err) {
1334 		mutex_unlock(&mbox->lock);
1335 		goto err_free_txsch;
1336 	}
1337 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
1338 		err = otx2_txschq_config(pf, lvl);
1339 		if (err) {
1340 			mutex_unlock(&mbox->lock);
1341 			goto err_free_nix_queues;
1342 		}
1343 	}
1344 	mutex_unlock(&mbox->lock);
1345 	return err;
1346 
1347 err_free_nix_queues:
1348 	otx2_free_sq_res(pf);
1349 	otx2_free_cq_res(pf);
1350 	otx2_ctx_disable(mbox, NIX_AQ_CTYPE_RQ, false);
1351 err_free_txsch:
1352 	if (otx2_txschq_stop(pf))
1353 		dev_err(pf->dev, "%s failed to stop TX schedulers\n", __func__);
1354 err_free_sq_ptrs:
1355 	otx2_sq_free_sqbs(pf);
1356 err_free_rq_ptrs:
1357 	otx2_free_aura_ptr(pf, AURA_NIX_RQ);
1358 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_POOL, true);
1359 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_AURA, true);
1360 	otx2_aura_pool_free(pf);
1361 err_free_nix_lf:
1362 	mutex_lock(&mbox->lock);
1363 	free_req = otx2_mbox_alloc_msg_nix_lf_free(mbox);
1364 	if (free_req) {
1365 		free_req->flags = NIX_LF_DISABLE_FLOWS;
1366 		if (otx2_sync_mbox_msg(mbox))
1367 			dev_err(pf->dev, "%s failed to free nixlf\n", __func__);
1368 	}
1369 err_free_npa_lf:
1370 	/* Reset NPA LF */
1371 	req = otx2_mbox_alloc_msg_npa_lf_free(mbox);
1372 	if (req) {
1373 		if (otx2_sync_mbox_msg(mbox))
1374 			dev_err(pf->dev, "%s failed to free npalf\n", __func__);
1375 	}
1376 exit:
1377 	mutex_unlock(&mbox->lock);
1378 	return err;
1379 }
1380 
1381 static void otx2_free_hw_resources(struct otx2_nic *pf)
1382 {
1383 	struct otx2_qset *qset = &pf->qset;
1384 	struct nix_lf_free_req *free_req;
1385 	struct mbox *mbox = &pf->mbox;
1386 	struct otx2_cq_queue *cq;
1387 	struct msg_req *req;
1388 	int qidx, err;
1389 
1390 	/* Ensure all SQE are processed */
1391 	otx2_sqb_flush(pf);
1392 
1393 	/* Stop transmission */
1394 	err = otx2_txschq_stop(pf);
1395 	if (err)
1396 		dev_err(pf->dev, "RVUPF: Failed to stop/free TX schedulers\n");
1397 
1398 	mutex_lock(&mbox->lock);
1399 	/* Disable backpressure */
1400 	if (!(pf->pcifunc & RVU_PFVF_FUNC_MASK))
1401 		otx2_nix_config_bp(pf, false);
1402 	mutex_unlock(&mbox->lock);
1403 
1404 	/* Disable RQs */
1405 	otx2_ctx_disable(mbox, NIX_AQ_CTYPE_RQ, false);
1406 
1407 	/*Dequeue all CQEs */
1408 	for (qidx = 0; qidx < qset->cq_cnt; qidx++) {
1409 		cq = &qset->cq[qidx];
1410 		if (cq->cq_type == CQ_RX)
1411 			otx2_cleanup_rx_cqes(pf, cq);
1412 		else
1413 			otx2_cleanup_tx_cqes(pf, cq);
1414 	}
1415 
1416 	otx2_free_sq_res(pf);
1417 
1418 	/* Free RQ buffer pointers*/
1419 	otx2_free_aura_ptr(pf, AURA_NIX_RQ);
1420 
1421 	otx2_free_cq_res(pf);
1422 
1423 	mutex_lock(&mbox->lock);
1424 	/* Reset NIX LF */
1425 	free_req = otx2_mbox_alloc_msg_nix_lf_free(mbox);
1426 	if (free_req) {
1427 		free_req->flags = NIX_LF_DISABLE_FLOWS;
1428 		if (!(pf->flags & OTX2_FLAG_PF_SHUTDOWN))
1429 			free_req->flags |= NIX_LF_DONT_FREE_TX_VTAG;
1430 		if (otx2_sync_mbox_msg(mbox))
1431 			dev_err(pf->dev, "%s failed to free nixlf\n", __func__);
1432 	}
1433 	mutex_unlock(&mbox->lock);
1434 
1435 	/* Disable NPA Pool and Aura hw context */
1436 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_POOL, true);
1437 	otx2_ctx_disable(mbox, NPA_AQ_CTYPE_AURA, true);
1438 	otx2_aura_pool_free(pf);
1439 
1440 	mutex_lock(&mbox->lock);
1441 	/* Reset NPA LF */
1442 	req = otx2_mbox_alloc_msg_npa_lf_free(mbox);
1443 	if (req) {
1444 		if (otx2_sync_mbox_msg(mbox))
1445 			dev_err(pf->dev, "%s failed to free npalf\n", __func__);
1446 	}
1447 	mutex_unlock(&mbox->lock);
1448 }
1449 
1450 int otx2_open(struct net_device *netdev)
1451 {
1452 	struct otx2_nic *pf = netdev_priv(netdev);
1453 	struct otx2_cq_poll *cq_poll = NULL;
1454 	struct otx2_qset *qset = &pf->qset;
1455 	int err = 0, qidx, vec;
1456 	char *irq_name;
1457 
1458 	netif_carrier_off(netdev);
1459 
1460 	pf->qset.cq_cnt = pf->hw.rx_queues + pf->hw.tx_queues;
1461 	/* RQ and SQs are mapped to different CQs,
1462 	 * so find out max CQ IRQs (i.e CINTs) needed.
1463 	 */
1464 	pf->hw.cint_cnt = max(pf->hw.rx_queues, pf->hw.tx_queues);
1465 	qset->napi = kcalloc(pf->hw.cint_cnt, sizeof(*cq_poll), GFP_KERNEL);
1466 	if (!qset->napi)
1467 		return -ENOMEM;
1468 
1469 	/* CQ size of RQ */
1470 	qset->rqe_cnt = qset->rqe_cnt ? qset->rqe_cnt : Q_COUNT(Q_SIZE_256);
1471 	/* CQ size of SQ */
1472 	qset->sqe_cnt = qset->sqe_cnt ? qset->sqe_cnt : Q_COUNT(Q_SIZE_4K);
1473 
1474 	err = -ENOMEM;
1475 	qset->cq = kcalloc(pf->qset.cq_cnt,
1476 			   sizeof(struct otx2_cq_queue), GFP_KERNEL);
1477 	if (!qset->cq)
1478 		goto err_free_mem;
1479 
1480 	qset->sq = kcalloc(pf->hw.tx_queues,
1481 			   sizeof(struct otx2_snd_queue), GFP_KERNEL);
1482 	if (!qset->sq)
1483 		goto err_free_mem;
1484 
1485 	qset->rq = kcalloc(pf->hw.rx_queues,
1486 			   sizeof(struct otx2_rcv_queue), GFP_KERNEL);
1487 	if (!qset->rq)
1488 		goto err_free_mem;
1489 
1490 	err = otx2_init_hw_resources(pf);
1491 	if (err)
1492 		goto err_free_mem;
1493 
1494 	/* Register NAPI handler */
1495 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
1496 		cq_poll = &qset->napi[qidx];
1497 		cq_poll->cint_idx = qidx;
1498 		/* RQ0 & SQ0 are mapped to CINT0 and so on..
1499 		 * 'cq_ids[0]' points to RQ's CQ and
1500 		 * 'cq_ids[1]' points to SQ's CQ and
1501 		 */
1502 		cq_poll->cq_ids[CQ_RX] =
1503 			(qidx <  pf->hw.rx_queues) ? qidx : CINT_INVALID_CQ;
1504 		cq_poll->cq_ids[CQ_TX] = (qidx < pf->hw.tx_queues) ?
1505 				      qidx + pf->hw.rx_queues : CINT_INVALID_CQ;
1506 		cq_poll->dev = (void *)pf;
1507 		netif_napi_add(netdev, &cq_poll->napi,
1508 			       otx2_napi_handler, NAPI_POLL_WEIGHT);
1509 		napi_enable(&cq_poll->napi);
1510 	}
1511 
1512 	/* Set maximum frame size allowed in HW */
1513 	err = otx2_hw_set_mtu(pf, netdev->mtu);
1514 	if (err)
1515 		goto err_disable_napi;
1516 
1517 	/* Setup segmentation algorithms, if failed, clear offload capability */
1518 	otx2_setup_segmentation(pf);
1519 
1520 	/* Initialize RSS */
1521 	err = otx2_rss_init(pf);
1522 	if (err)
1523 		goto err_disable_napi;
1524 
1525 	/* Register Queue IRQ handlers */
1526 	vec = pf->hw.nix_msixoff + NIX_LF_QINT_VEC_START;
1527 	irq_name = &pf->hw.irq_name[vec * NAME_SIZE];
1528 
1529 	snprintf(irq_name, NAME_SIZE, "%s-qerr", pf->netdev->name);
1530 
1531 	err = request_irq(pci_irq_vector(pf->pdev, vec),
1532 			  otx2_q_intr_handler, 0, irq_name, pf);
1533 	if (err) {
1534 		dev_err(pf->dev,
1535 			"RVUPF%d: IRQ registration failed for QERR\n",
1536 			rvu_get_pf(pf->pcifunc));
1537 		goto err_disable_napi;
1538 	}
1539 
1540 	/* Enable QINT IRQ */
1541 	otx2_write64(pf, NIX_LF_QINTX_ENA_W1S(0), BIT_ULL(0));
1542 
1543 	/* Register CQ IRQ handlers */
1544 	vec = pf->hw.nix_msixoff + NIX_LF_CINT_VEC_START;
1545 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
1546 		irq_name = &pf->hw.irq_name[vec * NAME_SIZE];
1547 
1548 		snprintf(irq_name, NAME_SIZE, "%s-rxtx-%d", pf->netdev->name,
1549 			 qidx);
1550 
1551 		err = request_irq(pci_irq_vector(pf->pdev, vec),
1552 				  otx2_cq_intr_handler, 0, irq_name,
1553 				  &qset->napi[qidx]);
1554 		if (err) {
1555 			dev_err(pf->dev,
1556 				"RVUPF%d: IRQ registration failed for CQ%d\n",
1557 				rvu_get_pf(pf->pcifunc), qidx);
1558 			goto err_free_cints;
1559 		}
1560 		vec++;
1561 
1562 		otx2_config_irq_coalescing(pf, qidx);
1563 
1564 		/* Enable CQ IRQ */
1565 		otx2_write64(pf, NIX_LF_CINTX_INT(qidx), BIT_ULL(0));
1566 		otx2_write64(pf, NIX_LF_CINTX_ENA_W1S(qidx), BIT_ULL(0));
1567 	}
1568 
1569 	otx2_set_cints_affinity(pf);
1570 
1571 	if (pf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
1572 		otx2_enable_rxvlan(pf, true);
1573 
1574 	/* When reinitializing enable time stamping if it is enabled before */
1575 	if (pf->flags & OTX2_FLAG_TX_TSTAMP_ENABLED) {
1576 		pf->flags &= ~OTX2_FLAG_TX_TSTAMP_ENABLED;
1577 		otx2_config_hw_tx_tstamp(pf, true);
1578 	}
1579 	if (pf->flags & OTX2_FLAG_RX_TSTAMP_ENABLED) {
1580 		pf->flags &= ~OTX2_FLAG_RX_TSTAMP_ENABLED;
1581 		otx2_config_hw_rx_tstamp(pf, true);
1582 	}
1583 
1584 	pf->flags &= ~OTX2_FLAG_INTF_DOWN;
1585 	/* 'intf_down' may be checked on any cpu */
1586 	smp_wmb();
1587 
1588 	/* we have already received link status notification */
1589 	if (pf->linfo.link_up && !(pf->pcifunc & RVU_PFVF_FUNC_MASK))
1590 		otx2_handle_link_event(pf);
1591 
1592 	/* Restore pause frame settings */
1593 	otx2_config_pause_frm(pf);
1594 
1595 	err = otx2_rxtx_enable(pf, true);
1596 	if (err)
1597 		goto err_tx_stop_queues;
1598 
1599 	return 0;
1600 
1601 err_tx_stop_queues:
1602 	netif_tx_stop_all_queues(netdev);
1603 	netif_carrier_off(netdev);
1604 err_free_cints:
1605 	otx2_free_cints(pf, qidx);
1606 	vec = pci_irq_vector(pf->pdev,
1607 			     pf->hw.nix_msixoff + NIX_LF_QINT_VEC_START);
1608 	otx2_write64(pf, NIX_LF_QINTX_ENA_W1C(0), BIT_ULL(0));
1609 	synchronize_irq(vec);
1610 	free_irq(vec, pf);
1611 err_disable_napi:
1612 	otx2_disable_napi(pf);
1613 	otx2_free_hw_resources(pf);
1614 err_free_mem:
1615 	kfree(qset->sq);
1616 	kfree(qset->cq);
1617 	kfree(qset->rq);
1618 	kfree(qset->napi);
1619 	return err;
1620 }
1621 EXPORT_SYMBOL(otx2_open);
1622 
1623 int otx2_stop(struct net_device *netdev)
1624 {
1625 	struct otx2_nic *pf = netdev_priv(netdev);
1626 	struct otx2_cq_poll *cq_poll = NULL;
1627 	struct otx2_qset *qset = &pf->qset;
1628 	int qidx, vec, wrk;
1629 
1630 	netif_carrier_off(netdev);
1631 	netif_tx_stop_all_queues(netdev);
1632 
1633 	pf->flags |= OTX2_FLAG_INTF_DOWN;
1634 	/* 'intf_down' may be checked on any cpu */
1635 	smp_wmb();
1636 
1637 	/* First stop packet Rx/Tx */
1638 	otx2_rxtx_enable(pf, false);
1639 
1640 	/* Cleanup Queue IRQ */
1641 	vec = pci_irq_vector(pf->pdev,
1642 			     pf->hw.nix_msixoff + NIX_LF_QINT_VEC_START);
1643 	otx2_write64(pf, NIX_LF_QINTX_ENA_W1C(0), BIT_ULL(0));
1644 	synchronize_irq(vec);
1645 	free_irq(vec, pf);
1646 
1647 	/* Cleanup CQ NAPI and IRQ */
1648 	vec = pf->hw.nix_msixoff + NIX_LF_CINT_VEC_START;
1649 	for (qidx = 0; qidx < pf->hw.cint_cnt; qidx++) {
1650 		/* Disable interrupt */
1651 		otx2_write64(pf, NIX_LF_CINTX_ENA_W1C(qidx), BIT_ULL(0));
1652 
1653 		synchronize_irq(pci_irq_vector(pf->pdev, vec));
1654 
1655 		cq_poll = &qset->napi[qidx];
1656 		napi_synchronize(&cq_poll->napi);
1657 		vec++;
1658 	}
1659 
1660 	netif_tx_disable(netdev);
1661 
1662 	otx2_free_hw_resources(pf);
1663 	otx2_free_cints(pf, pf->hw.cint_cnt);
1664 	otx2_disable_napi(pf);
1665 
1666 	for (qidx = 0; qidx < netdev->num_tx_queues; qidx++)
1667 		netdev_tx_reset_queue(netdev_get_tx_queue(netdev, qidx));
1668 
1669 	for (wrk = 0; wrk < pf->qset.cq_cnt; wrk++)
1670 		cancel_delayed_work_sync(&pf->refill_wrk[wrk].pool_refill_work);
1671 	devm_kfree(pf->dev, pf->refill_wrk);
1672 
1673 	kfree(qset->sq);
1674 	kfree(qset->cq);
1675 	kfree(qset->rq);
1676 	kfree(qset->napi);
1677 	/* Do not clear RQ/SQ ringsize settings */
1678 	memset((void *)qset + offsetof(struct otx2_qset, sqe_cnt), 0,
1679 	       sizeof(*qset) - offsetof(struct otx2_qset, sqe_cnt));
1680 	return 0;
1681 }
1682 EXPORT_SYMBOL(otx2_stop);
1683 
1684 static netdev_tx_t otx2_xmit(struct sk_buff *skb, struct net_device *netdev)
1685 {
1686 	struct otx2_nic *pf = netdev_priv(netdev);
1687 	int qidx = skb_get_queue_mapping(skb);
1688 	struct otx2_snd_queue *sq;
1689 	struct netdev_queue *txq;
1690 
1691 	/* Check for minimum and maximum packet length */
1692 	if (skb->len <= ETH_HLEN ||
1693 	    (!skb_shinfo(skb)->gso_size && skb->len > pf->max_frs)) {
1694 		dev_kfree_skb(skb);
1695 		return NETDEV_TX_OK;
1696 	}
1697 
1698 	sq = &pf->qset.sq[qidx];
1699 	txq = netdev_get_tx_queue(netdev, qidx);
1700 
1701 	if (!otx2_sq_append_skb(netdev, sq, skb, qidx)) {
1702 		netif_tx_stop_queue(txq);
1703 
1704 		/* Check again, incase SQBs got freed up */
1705 		smp_mb();
1706 		if (((sq->num_sqbs - *sq->aura_fc_addr) * sq->sqe_per_sqb)
1707 							> sq->sqe_thresh)
1708 			netif_tx_wake_queue(txq);
1709 
1710 		return NETDEV_TX_BUSY;
1711 	}
1712 
1713 	return NETDEV_TX_OK;
1714 }
1715 
1716 static void otx2_set_rx_mode(struct net_device *netdev)
1717 {
1718 	struct otx2_nic *pf = netdev_priv(netdev);
1719 
1720 	queue_work(pf->otx2_wq, &pf->rx_mode_work);
1721 }
1722 
1723 static void otx2_do_set_rx_mode(struct work_struct *work)
1724 {
1725 	struct otx2_nic *pf = container_of(work, struct otx2_nic, rx_mode_work);
1726 	struct net_device *netdev = pf->netdev;
1727 	struct nix_rx_mode *req;
1728 	bool promisc = false;
1729 
1730 	if (!(netdev->flags & IFF_UP))
1731 		return;
1732 
1733 	if ((netdev->flags & IFF_PROMISC) ||
1734 	    (netdev_uc_count(netdev) > OTX2_MAX_UNICAST_FLOWS)) {
1735 		promisc = true;
1736 	}
1737 
1738 	/* Write unicast address to mcam entries or del from mcam */
1739 	if (!promisc && netdev->priv_flags & IFF_UNICAST_FLT)
1740 		__dev_uc_sync(netdev, otx2_add_macfilter, otx2_del_macfilter);
1741 
1742 	mutex_lock(&pf->mbox.lock);
1743 	req = otx2_mbox_alloc_msg_nix_set_rx_mode(&pf->mbox);
1744 	if (!req) {
1745 		mutex_unlock(&pf->mbox.lock);
1746 		return;
1747 	}
1748 
1749 	req->mode = NIX_RX_MODE_UCAST;
1750 
1751 	if (promisc)
1752 		req->mode |= NIX_RX_MODE_PROMISC;
1753 	else if (netdev->flags & (IFF_ALLMULTI | IFF_MULTICAST))
1754 		req->mode |= NIX_RX_MODE_ALLMULTI;
1755 
1756 	otx2_sync_mbox_msg(&pf->mbox);
1757 	mutex_unlock(&pf->mbox.lock);
1758 }
1759 
1760 static int otx2_set_features(struct net_device *netdev,
1761 			     netdev_features_t features)
1762 {
1763 	netdev_features_t changed = features ^ netdev->features;
1764 	bool ntuple = !!(features & NETIF_F_NTUPLE);
1765 	struct otx2_nic *pf = netdev_priv(netdev);
1766 
1767 	if ((changed & NETIF_F_LOOPBACK) && netif_running(netdev))
1768 		return otx2_cgx_config_loopback(pf,
1769 						features & NETIF_F_LOOPBACK);
1770 
1771 	if ((changed & NETIF_F_HW_VLAN_CTAG_RX) && netif_running(netdev))
1772 		return otx2_enable_rxvlan(pf,
1773 					  features & NETIF_F_HW_VLAN_CTAG_RX);
1774 
1775 	if ((changed & NETIF_F_NTUPLE) && !ntuple)
1776 		otx2_destroy_ntuple_flows(pf);
1777 
1778 	return 0;
1779 }
1780 
1781 static void otx2_reset_task(struct work_struct *work)
1782 {
1783 	struct otx2_nic *pf = container_of(work, struct otx2_nic, reset_task);
1784 
1785 	if (!netif_running(pf->netdev))
1786 		return;
1787 
1788 	rtnl_lock();
1789 	otx2_stop(pf->netdev);
1790 	pf->reset_count++;
1791 	otx2_open(pf->netdev);
1792 	netif_trans_update(pf->netdev);
1793 	rtnl_unlock();
1794 }
1795 
1796 static int otx2_config_hw_rx_tstamp(struct otx2_nic *pfvf, bool enable)
1797 {
1798 	struct msg_req *req;
1799 	int err;
1800 
1801 	if (pfvf->flags & OTX2_FLAG_RX_TSTAMP_ENABLED && enable)
1802 		return 0;
1803 
1804 	mutex_lock(&pfvf->mbox.lock);
1805 	if (enable)
1806 		req = otx2_mbox_alloc_msg_cgx_ptp_rx_enable(&pfvf->mbox);
1807 	else
1808 		req = otx2_mbox_alloc_msg_cgx_ptp_rx_disable(&pfvf->mbox);
1809 	if (!req) {
1810 		mutex_unlock(&pfvf->mbox.lock);
1811 		return -ENOMEM;
1812 	}
1813 
1814 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1815 	if (err) {
1816 		mutex_unlock(&pfvf->mbox.lock);
1817 		return err;
1818 	}
1819 
1820 	mutex_unlock(&pfvf->mbox.lock);
1821 	if (enable)
1822 		pfvf->flags |= OTX2_FLAG_RX_TSTAMP_ENABLED;
1823 	else
1824 		pfvf->flags &= ~OTX2_FLAG_RX_TSTAMP_ENABLED;
1825 	return 0;
1826 }
1827 
1828 static int otx2_config_hw_tx_tstamp(struct otx2_nic *pfvf, bool enable)
1829 {
1830 	struct msg_req *req;
1831 	int err;
1832 
1833 	if (pfvf->flags & OTX2_FLAG_TX_TSTAMP_ENABLED && enable)
1834 		return 0;
1835 
1836 	mutex_lock(&pfvf->mbox.lock);
1837 	if (enable)
1838 		req = otx2_mbox_alloc_msg_nix_lf_ptp_tx_enable(&pfvf->mbox);
1839 	else
1840 		req = otx2_mbox_alloc_msg_nix_lf_ptp_tx_disable(&pfvf->mbox);
1841 	if (!req) {
1842 		mutex_unlock(&pfvf->mbox.lock);
1843 		return -ENOMEM;
1844 	}
1845 
1846 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1847 	if (err) {
1848 		mutex_unlock(&pfvf->mbox.lock);
1849 		return err;
1850 	}
1851 
1852 	mutex_unlock(&pfvf->mbox.lock);
1853 	if (enable)
1854 		pfvf->flags |= OTX2_FLAG_TX_TSTAMP_ENABLED;
1855 	else
1856 		pfvf->flags &= ~OTX2_FLAG_TX_TSTAMP_ENABLED;
1857 	return 0;
1858 }
1859 
1860 static int otx2_config_hwtstamp(struct net_device *netdev, struct ifreq *ifr)
1861 {
1862 	struct otx2_nic *pfvf = netdev_priv(netdev);
1863 	struct hwtstamp_config config;
1864 
1865 	if (!pfvf->ptp)
1866 		return -ENODEV;
1867 
1868 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
1869 		return -EFAULT;
1870 
1871 	/* reserved for future extensions */
1872 	if (config.flags)
1873 		return -EINVAL;
1874 
1875 	switch (config.tx_type) {
1876 	case HWTSTAMP_TX_OFF:
1877 		otx2_config_hw_tx_tstamp(pfvf, false);
1878 		break;
1879 	case HWTSTAMP_TX_ON:
1880 		otx2_config_hw_tx_tstamp(pfvf, true);
1881 		break;
1882 	default:
1883 		return -ERANGE;
1884 	}
1885 
1886 	switch (config.rx_filter) {
1887 	case HWTSTAMP_FILTER_NONE:
1888 		otx2_config_hw_rx_tstamp(pfvf, false);
1889 		break;
1890 	case HWTSTAMP_FILTER_ALL:
1891 	case HWTSTAMP_FILTER_SOME:
1892 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1893 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1894 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1895 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1896 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1897 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1898 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1899 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1900 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1901 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
1902 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
1903 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1904 		otx2_config_hw_rx_tstamp(pfvf, true);
1905 		config.rx_filter = HWTSTAMP_FILTER_ALL;
1906 		break;
1907 	default:
1908 		return -ERANGE;
1909 	}
1910 
1911 	memcpy(&pfvf->tstamp, &config, sizeof(config));
1912 
1913 	return copy_to_user(ifr->ifr_data, &config,
1914 			    sizeof(config)) ? -EFAULT : 0;
1915 }
1916 
1917 static int otx2_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
1918 {
1919 	struct otx2_nic *pfvf = netdev_priv(netdev);
1920 	struct hwtstamp_config *cfg = &pfvf->tstamp;
1921 
1922 	switch (cmd) {
1923 	case SIOCSHWTSTAMP:
1924 		return otx2_config_hwtstamp(netdev, req);
1925 	case SIOCGHWTSTAMP:
1926 		return copy_to_user(req->ifr_data, cfg,
1927 				    sizeof(*cfg)) ? -EFAULT : 0;
1928 	default:
1929 		return -EOPNOTSUPP;
1930 	}
1931 }
1932 
1933 static int otx2_do_set_vf_mac(struct otx2_nic *pf, int vf, const u8 *mac)
1934 {
1935 	struct npc_install_flow_req *req;
1936 	int err;
1937 
1938 	mutex_lock(&pf->mbox.lock);
1939 	req = otx2_mbox_alloc_msg_npc_install_flow(&pf->mbox);
1940 	if (!req) {
1941 		err = -ENOMEM;
1942 		goto out;
1943 	}
1944 
1945 	ether_addr_copy(req->packet.dmac, mac);
1946 	eth_broadcast_addr((u8 *)&req->mask.dmac);
1947 	req->features = BIT_ULL(NPC_DMAC);
1948 	req->channel = pf->hw.rx_chan_base;
1949 	req->intf = NIX_INTF_RX;
1950 	req->default_rule = 1;
1951 	req->append = 1;
1952 	req->vf = vf + 1;
1953 	req->op = NIX_RX_ACTION_DEFAULT;
1954 
1955 	err = otx2_sync_mbox_msg(&pf->mbox);
1956 out:
1957 	mutex_unlock(&pf->mbox.lock);
1958 	return err;
1959 }
1960 
1961 static int otx2_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1962 {
1963 	struct otx2_nic *pf = netdev_priv(netdev);
1964 	struct pci_dev *pdev = pf->pdev;
1965 	struct otx2_vf_config *config;
1966 	int ret;
1967 
1968 	if (!netif_running(netdev))
1969 		return -EAGAIN;
1970 
1971 	if (vf >= pci_num_vf(pdev))
1972 		return -EINVAL;
1973 
1974 	if (!is_valid_ether_addr(mac))
1975 		return -EINVAL;
1976 
1977 	config = &pf->vf_configs[vf];
1978 	ether_addr_copy(config->mac, mac);
1979 
1980 	ret = otx2_do_set_vf_mac(pf, vf, mac);
1981 	if (ret == 0)
1982 		dev_info(&pdev->dev, "Reload VF driver to apply the changes\n");
1983 
1984 	return ret;
1985 }
1986 
1987 static int otx2_do_set_vf_vlan(struct otx2_nic *pf, int vf, u16 vlan, u8 qos,
1988 			       __be16 proto)
1989 {
1990 	struct otx2_flow_config *flow_cfg = pf->flow_cfg;
1991 	struct nix_vtag_config_rsp *vtag_rsp;
1992 	struct npc_delete_flow_req *del_req;
1993 	struct nix_vtag_config *vtag_req;
1994 	struct npc_install_flow_req *req;
1995 	struct otx2_vf_config *config;
1996 	int err = 0;
1997 	u32 idx;
1998 
1999 	config = &pf->vf_configs[vf];
2000 
2001 	if (!vlan && !config->vlan)
2002 		goto out;
2003 
2004 	mutex_lock(&pf->mbox.lock);
2005 
2006 	/* free old tx vtag entry */
2007 	if (config->vlan) {
2008 		vtag_req = otx2_mbox_alloc_msg_nix_vtag_cfg(&pf->mbox);
2009 		if (!vtag_req) {
2010 			err = -ENOMEM;
2011 			goto out;
2012 		}
2013 		vtag_req->cfg_type = 0;
2014 		vtag_req->tx.free_vtag0 = 1;
2015 		vtag_req->tx.vtag0_idx = config->tx_vtag_idx;
2016 
2017 		err = otx2_sync_mbox_msg(&pf->mbox);
2018 		if (err)
2019 			goto out;
2020 	}
2021 
2022 	if (!vlan && config->vlan) {
2023 		/* rx */
2024 		del_req = otx2_mbox_alloc_msg_npc_delete_flow(&pf->mbox);
2025 		if (!del_req) {
2026 			err = -ENOMEM;
2027 			goto out;
2028 		}
2029 		idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_RX_INDEX);
2030 		del_req->entry =
2031 			flow_cfg->entry[flow_cfg->vf_vlan_offset + idx];
2032 		err = otx2_sync_mbox_msg(&pf->mbox);
2033 		if (err)
2034 			goto out;
2035 
2036 		/* tx */
2037 		del_req = otx2_mbox_alloc_msg_npc_delete_flow(&pf->mbox);
2038 		if (!del_req) {
2039 			err = -ENOMEM;
2040 			goto out;
2041 		}
2042 		idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_TX_INDEX);
2043 		del_req->entry =
2044 			flow_cfg->entry[flow_cfg->vf_vlan_offset + idx];
2045 		err = otx2_sync_mbox_msg(&pf->mbox);
2046 
2047 		goto out;
2048 	}
2049 
2050 	/* rx */
2051 	req = otx2_mbox_alloc_msg_npc_install_flow(&pf->mbox);
2052 	if (!req) {
2053 		err = -ENOMEM;
2054 		goto out;
2055 	}
2056 
2057 	idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_RX_INDEX);
2058 	req->entry = flow_cfg->entry[flow_cfg->vf_vlan_offset + idx];
2059 	req->packet.vlan_tci = htons(vlan);
2060 	req->mask.vlan_tci = htons(VLAN_VID_MASK);
2061 	/* af fills the destination mac addr */
2062 	eth_broadcast_addr((u8 *)&req->mask.dmac);
2063 	req->features = BIT_ULL(NPC_OUTER_VID) | BIT_ULL(NPC_DMAC);
2064 	req->channel = pf->hw.rx_chan_base;
2065 	req->intf = NIX_INTF_RX;
2066 	req->vf = vf + 1;
2067 	req->op = NIX_RX_ACTION_DEFAULT;
2068 	req->vtag0_valid = true;
2069 	req->vtag0_type = NIX_AF_LFX_RX_VTAG_TYPE7;
2070 	req->set_cntr = 1;
2071 
2072 	err = otx2_sync_mbox_msg(&pf->mbox);
2073 	if (err)
2074 		goto out;
2075 
2076 	/* tx */
2077 	vtag_req = otx2_mbox_alloc_msg_nix_vtag_cfg(&pf->mbox);
2078 	if (!vtag_req) {
2079 		err = -ENOMEM;
2080 		goto out;
2081 	}
2082 
2083 	/* configure tx vtag params */
2084 	vtag_req->vtag_size = VTAGSIZE_T4;
2085 	vtag_req->cfg_type = 0; /* tx vlan cfg */
2086 	vtag_req->tx.cfg_vtag0 = 1;
2087 	vtag_req->tx.vtag0 = ((u64)ntohs(proto) << 16) | vlan;
2088 
2089 	err = otx2_sync_mbox_msg(&pf->mbox);
2090 	if (err)
2091 		goto out;
2092 
2093 	vtag_rsp = (struct nix_vtag_config_rsp *)otx2_mbox_get_rsp
2094 			(&pf->mbox.mbox, 0, &vtag_req->hdr);
2095 	if (IS_ERR(vtag_rsp)) {
2096 		err = PTR_ERR(vtag_rsp);
2097 		goto out;
2098 	}
2099 	config->tx_vtag_idx = vtag_rsp->vtag0_idx;
2100 
2101 	req = otx2_mbox_alloc_msg_npc_install_flow(&pf->mbox);
2102 	if (!req) {
2103 		err = -ENOMEM;
2104 		goto out;
2105 	}
2106 
2107 	eth_zero_addr((u8 *)&req->mask.dmac);
2108 	idx = ((vf * OTX2_PER_VF_VLAN_FLOWS) + OTX2_VF_VLAN_TX_INDEX);
2109 	req->entry = flow_cfg->entry[flow_cfg->vf_vlan_offset + idx];
2110 	req->features = BIT_ULL(NPC_DMAC);
2111 	req->channel = pf->hw.tx_chan_base;
2112 	req->intf = NIX_INTF_TX;
2113 	req->vf = vf + 1;
2114 	req->op = NIX_TX_ACTIONOP_UCAST_DEFAULT;
2115 	req->vtag0_def = vtag_rsp->vtag0_idx;
2116 	req->vtag0_op = VTAG_INSERT;
2117 	req->set_cntr = 1;
2118 
2119 	err = otx2_sync_mbox_msg(&pf->mbox);
2120 out:
2121 	config->vlan = vlan;
2122 	mutex_unlock(&pf->mbox.lock);
2123 	return err;
2124 }
2125 
2126 static int otx2_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
2127 			    __be16 proto)
2128 {
2129 	struct otx2_nic *pf = netdev_priv(netdev);
2130 	struct pci_dev *pdev = pf->pdev;
2131 
2132 	if (!netif_running(netdev))
2133 		return -EAGAIN;
2134 
2135 	if (vf >= pci_num_vf(pdev))
2136 		return -EINVAL;
2137 
2138 	/* qos is currently unsupported */
2139 	if (vlan >= VLAN_N_VID || qos)
2140 		return -EINVAL;
2141 
2142 	if (proto != htons(ETH_P_8021Q))
2143 		return -EPROTONOSUPPORT;
2144 
2145 	if (!(pf->flags & OTX2_FLAG_VF_VLAN_SUPPORT))
2146 		return -EOPNOTSUPP;
2147 
2148 	return otx2_do_set_vf_vlan(pf, vf, vlan, qos, proto);
2149 }
2150 
2151 static int otx2_get_vf_config(struct net_device *netdev, int vf,
2152 			      struct ifla_vf_info *ivi)
2153 {
2154 	struct otx2_nic *pf = netdev_priv(netdev);
2155 	struct pci_dev *pdev = pf->pdev;
2156 	struct otx2_vf_config *config;
2157 
2158 	if (!netif_running(netdev))
2159 		return -EAGAIN;
2160 
2161 	if (vf >= pci_num_vf(pdev))
2162 		return -EINVAL;
2163 
2164 	config = &pf->vf_configs[vf];
2165 	ivi->vf = vf;
2166 	ether_addr_copy(ivi->mac, config->mac);
2167 	ivi->vlan = config->vlan;
2168 
2169 	return 0;
2170 }
2171 
2172 static const struct net_device_ops otx2_netdev_ops = {
2173 	.ndo_open		= otx2_open,
2174 	.ndo_stop		= otx2_stop,
2175 	.ndo_start_xmit		= otx2_xmit,
2176 	.ndo_set_mac_address    = otx2_set_mac_address,
2177 	.ndo_change_mtu		= otx2_change_mtu,
2178 	.ndo_set_rx_mode	= otx2_set_rx_mode,
2179 	.ndo_set_features	= otx2_set_features,
2180 	.ndo_tx_timeout		= otx2_tx_timeout,
2181 	.ndo_get_stats64	= otx2_get_stats64,
2182 	.ndo_do_ioctl		= otx2_ioctl,
2183 	.ndo_set_vf_mac		= otx2_set_vf_mac,
2184 	.ndo_set_vf_vlan	= otx2_set_vf_vlan,
2185 	.ndo_get_vf_config	= otx2_get_vf_config,
2186 };
2187 
2188 static int otx2_wq_init(struct otx2_nic *pf)
2189 {
2190 	pf->otx2_wq = create_singlethread_workqueue("otx2_wq");
2191 	if (!pf->otx2_wq)
2192 		return -ENOMEM;
2193 
2194 	INIT_WORK(&pf->rx_mode_work, otx2_do_set_rx_mode);
2195 	INIT_WORK(&pf->reset_task, otx2_reset_task);
2196 	return 0;
2197 }
2198 
2199 static int otx2_check_pf_usable(struct otx2_nic *nic)
2200 {
2201 	u64 rev;
2202 
2203 	rev = otx2_read64(nic, RVU_PF_BLOCK_ADDRX_DISC(BLKADDR_RVUM));
2204 	rev = (rev >> 12) & 0xFF;
2205 	/* Check if AF has setup revision for RVUM block,
2206 	 * otherwise this driver probe should be deferred
2207 	 * until AF driver comes up.
2208 	 */
2209 	if (!rev) {
2210 		dev_warn(nic->dev,
2211 			 "AF is not initialized, deferring probe\n");
2212 		return -EPROBE_DEFER;
2213 	}
2214 	return 0;
2215 }
2216 
2217 static int otx2_realloc_msix_vectors(struct otx2_nic *pf)
2218 {
2219 	struct otx2_hw *hw = &pf->hw;
2220 	int num_vec, err;
2221 
2222 	/* NPA interrupts are inot registered, so alloc only
2223 	 * upto NIX vector offset.
2224 	 */
2225 	num_vec = hw->nix_msixoff;
2226 	num_vec += NIX_LF_CINT_VEC_START + hw->max_queues;
2227 
2228 	otx2_disable_mbox_intr(pf);
2229 	pci_free_irq_vectors(hw->pdev);
2230 	err = pci_alloc_irq_vectors(hw->pdev, num_vec, num_vec, PCI_IRQ_MSIX);
2231 	if (err < 0) {
2232 		dev_err(pf->dev, "%s: Failed to realloc %d IRQ vectors\n",
2233 			__func__, num_vec);
2234 		return err;
2235 	}
2236 
2237 	return otx2_register_mbox_intr(pf, false);
2238 }
2239 
2240 static int otx2_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2241 {
2242 	struct device *dev = &pdev->dev;
2243 	struct net_device *netdev;
2244 	struct otx2_nic *pf;
2245 	struct otx2_hw *hw;
2246 	int err, qcount;
2247 	int num_vec;
2248 
2249 	err = pcim_enable_device(pdev);
2250 	if (err) {
2251 		dev_err(dev, "Failed to enable PCI device\n");
2252 		return err;
2253 	}
2254 
2255 	err = pci_request_regions(pdev, DRV_NAME);
2256 	if (err) {
2257 		dev_err(dev, "PCI request regions failed 0x%x\n", err);
2258 		return err;
2259 	}
2260 
2261 	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(48));
2262 	if (err) {
2263 		dev_err(dev, "DMA mask config failed, abort\n");
2264 		goto err_release_regions;
2265 	}
2266 
2267 	pci_set_master(pdev);
2268 
2269 	/* Set number of queues */
2270 	qcount = min_t(int, num_online_cpus(), OTX2_MAX_CQ_CNT);
2271 
2272 	netdev = alloc_etherdev_mqs(sizeof(*pf), qcount, qcount);
2273 	if (!netdev) {
2274 		err = -ENOMEM;
2275 		goto err_release_regions;
2276 	}
2277 
2278 	pci_set_drvdata(pdev, netdev);
2279 	SET_NETDEV_DEV(netdev, &pdev->dev);
2280 	pf = netdev_priv(netdev);
2281 	pf->netdev = netdev;
2282 	pf->pdev = pdev;
2283 	pf->dev = dev;
2284 	pf->total_vfs = pci_sriov_get_totalvfs(pdev);
2285 	pf->flags |= OTX2_FLAG_INTF_DOWN;
2286 
2287 	hw = &pf->hw;
2288 	hw->pdev = pdev;
2289 	hw->rx_queues = qcount;
2290 	hw->tx_queues = qcount;
2291 	hw->max_queues = qcount;
2292 
2293 	num_vec = pci_msix_vec_count(pdev);
2294 	hw->irq_name = devm_kmalloc_array(&hw->pdev->dev, num_vec, NAME_SIZE,
2295 					  GFP_KERNEL);
2296 	if (!hw->irq_name) {
2297 		err = -ENOMEM;
2298 		goto err_free_netdev;
2299 	}
2300 
2301 	hw->affinity_mask = devm_kcalloc(&hw->pdev->dev, num_vec,
2302 					 sizeof(cpumask_var_t), GFP_KERNEL);
2303 	if (!hw->affinity_mask) {
2304 		err = -ENOMEM;
2305 		goto err_free_netdev;
2306 	}
2307 
2308 	/* Map CSRs */
2309 	pf->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
2310 	if (!pf->reg_base) {
2311 		dev_err(dev, "Unable to map physical function CSRs, aborting\n");
2312 		err = -ENOMEM;
2313 		goto err_free_netdev;
2314 	}
2315 
2316 	err = otx2_check_pf_usable(pf);
2317 	if (err)
2318 		goto err_free_netdev;
2319 
2320 	err = pci_alloc_irq_vectors(hw->pdev, RVU_PF_INT_VEC_CNT,
2321 				    RVU_PF_INT_VEC_CNT, PCI_IRQ_MSIX);
2322 	if (err < 0) {
2323 		dev_err(dev, "%s: Failed to alloc %d IRQ vectors\n",
2324 			__func__, num_vec);
2325 		goto err_free_netdev;
2326 	}
2327 
2328 	/* Init PF <=> AF mailbox stuff */
2329 	err = otx2_pfaf_mbox_init(pf);
2330 	if (err)
2331 		goto err_free_irq_vectors;
2332 
2333 	/* Register mailbox interrupt */
2334 	err = otx2_register_mbox_intr(pf, true);
2335 	if (err)
2336 		goto err_mbox_destroy;
2337 
2338 	/* Request AF to attach NPA and NIX LFs to this PF.
2339 	 * NIX and NPA LFs are needed for this PF to function as a NIC.
2340 	 */
2341 	err = otx2_attach_npa_nix(pf);
2342 	if (err)
2343 		goto err_disable_mbox_intr;
2344 
2345 	err = otx2_realloc_msix_vectors(pf);
2346 	if (err)
2347 		goto err_detach_rsrc;
2348 
2349 	err = otx2_set_real_num_queues(netdev, hw->tx_queues, hw->rx_queues);
2350 	if (err)
2351 		goto err_detach_rsrc;
2352 
2353 	otx2_setup_dev_hw_settings(pf);
2354 
2355 	/* Assign default mac address */
2356 	otx2_get_mac_from_af(netdev);
2357 
2358 	/* Don't check for error.  Proceed without ptp */
2359 	otx2_ptp_init(pf);
2360 
2361 	/* NPA's pool is a stack to which SW frees buffer pointers via Aura.
2362 	 * HW allocates buffer pointer from stack and uses it for DMA'ing
2363 	 * ingress packet. In some scenarios HW can free back allocated buffer
2364 	 * pointers to pool. This makes it impossible for SW to maintain a
2365 	 * parallel list where physical addresses of buffer pointers (IOVAs)
2366 	 * given to HW can be saved for later reference.
2367 	 *
2368 	 * So the only way to convert Rx packet's buffer address is to use
2369 	 * IOMMU's iova_to_phys() handler which translates the address by
2370 	 * walking through the translation tables.
2371 	 */
2372 	pf->iommu_domain = iommu_get_domain_for_dev(dev);
2373 
2374 	netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_IP_CSUM |
2375 			       NETIF_F_IPV6_CSUM | NETIF_F_RXHASH |
2376 			       NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
2377 			       NETIF_F_GSO_UDP_L4);
2378 	netdev->features |= netdev->hw_features;
2379 
2380 	netdev->hw_features |= NETIF_F_LOOPBACK | NETIF_F_RXALL;
2381 
2382 	err = otx2_mcam_flow_init(pf);
2383 	if (err)
2384 		goto err_ptp_destroy;
2385 
2386 	if (pf->flags & OTX2_FLAG_NTUPLE_SUPPORT)
2387 		netdev->hw_features |= NETIF_F_NTUPLE;
2388 
2389 	if (pf->flags & OTX2_FLAG_UCAST_FLTR_SUPPORT)
2390 		netdev->priv_flags |= IFF_UNICAST_FLT;
2391 
2392 	/* Support TSO on tag interface */
2393 	netdev->vlan_features |= netdev->features;
2394 	netdev->hw_features  |= NETIF_F_HW_VLAN_CTAG_TX |
2395 				NETIF_F_HW_VLAN_STAG_TX;
2396 	if (pf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
2397 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX |
2398 				       NETIF_F_HW_VLAN_STAG_RX;
2399 	netdev->features |= netdev->hw_features;
2400 
2401 	netdev->gso_max_segs = OTX2_MAX_GSO_SEGS;
2402 	netdev->watchdog_timeo = OTX2_TX_TIMEOUT;
2403 
2404 	netdev->netdev_ops = &otx2_netdev_ops;
2405 
2406 	/* MTU range: 64 - 9190 */
2407 	netdev->min_mtu = OTX2_MIN_MTU;
2408 	netdev->max_mtu = OTX2_MAX_MTU;
2409 
2410 	err = register_netdev(netdev);
2411 	if (err) {
2412 		dev_err(dev, "Failed to register netdevice\n");
2413 		goto err_del_mcam_entries;
2414 	}
2415 
2416 	err = otx2_wq_init(pf);
2417 	if (err)
2418 		goto err_unreg_netdev;
2419 
2420 	otx2_set_ethtool_ops(netdev);
2421 
2422 	/* Enable link notifications */
2423 	otx2_cgx_config_linkevents(pf, true);
2424 
2425 	/* Enable pause frames by default */
2426 	pf->flags |= OTX2_FLAG_RX_PAUSE_ENABLED;
2427 	pf->flags |= OTX2_FLAG_TX_PAUSE_ENABLED;
2428 
2429 	return 0;
2430 
2431 err_unreg_netdev:
2432 	unregister_netdev(netdev);
2433 err_del_mcam_entries:
2434 	otx2_mcam_flow_del(pf);
2435 err_ptp_destroy:
2436 	otx2_ptp_destroy(pf);
2437 err_detach_rsrc:
2438 	otx2_detach_resources(&pf->mbox);
2439 err_disable_mbox_intr:
2440 	otx2_disable_mbox_intr(pf);
2441 err_mbox_destroy:
2442 	otx2_pfaf_mbox_destroy(pf);
2443 err_free_irq_vectors:
2444 	pci_free_irq_vectors(hw->pdev);
2445 err_free_netdev:
2446 	pci_set_drvdata(pdev, NULL);
2447 	free_netdev(netdev);
2448 err_release_regions:
2449 	pci_release_regions(pdev);
2450 	return err;
2451 }
2452 
2453 static void otx2_vf_link_event_task(struct work_struct *work)
2454 {
2455 	struct otx2_vf_config *config;
2456 	struct cgx_link_info_msg *req;
2457 	struct mbox_msghdr *msghdr;
2458 	struct otx2_nic *pf;
2459 	int vf_idx;
2460 
2461 	config = container_of(work, struct otx2_vf_config,
2462 			      link_event_work.work);
2463 	vf_idx = config - config->pf->vf_configs;
2464 	pf = config->pf;
2465 
2466 	msghdr = otx2_mbox_alloc_msg_rsp(&pf->mbox_pfvf[0].mbox_up, vf_idx,
2467 					 sizeof(*req), sizeof(struct msg_rsp));
2468 	if (!msghdr) {
2469 		dev_err(pf->dev, "Failed to create VF%d link event\n", vf_idx);
2470 		return;
2471 	}
2472 
2473 	req = (struct cgx_link_info_msg *)msghdr;
2474 	req->hdr.id = MBOX_MSG_CGX_LINK_EVENT;
2475 	req->hdr.sig = OTX2_MBOX_REQ_SIG;
2476 	memcpy(&req->link_info, &pf->linfo, sizeof(req->link_info));
2477 
2478 	otx2_sync_mbox_up_msg(&pf->mbox_pfvf[0], vf_idx);
2479 }
2480 
2481 static int otx2_sriov_enable(struct pci_dev *pdev, int numvfs)
2482 {
2483 	struct net_device *netdev = pci_get_drvdata(pdev);
2484 	struct otx2_nic *pf = netdev_priv(netdev);
2485 	int ret, i;
2486 
2487 	/* Init PF <=> VF mailbox stuff */
2488 	ret = otx2_pfvf_mbox_init(pf, numvfs);
2489 	if (ret)
2490 		return ret;
2491 
2492 	ret = otx2_register_pfvf_mbox_intr(pf, numvfs);
2493 	if (ret)
2494 		goto free_mbox;
2495 
2496 	pf->vf_configs = kcalloc(numvfs, sizeof(struct otx2_vf_config),
2497 				 GFP_KERNEL);
2498 	if (!pf->vf_configs) {
2499 		ret = -ENOMEM;
2500 		goto free_intr;
2501 	}
2502 
2503 	for (i = 0; i < numvfs; i++) {
2504 		pf->vf_configs[i].pf = pf;
2505 		pf->vf_configs[i].intf_down = true;
2506 		INIT_DELAYED_WORK(&pf->vf_configs[i].link_event_work,
2507 				  otx2_vf_link_event_task);
2508 	}
2509 
2510 	ret = otx2_pf_flr_init(pf, numvfs);
2511 	if (ret)
2512 		goto free_configs;
2513 
2514 	ret = otx2_register_flr_me_intr(pf, numvfs);
2515 	if (ret)
2516 		goto free_flr;
2517 
2518 	ret = pci_enable_sriov(pdev, numvfs);
2519 	if (ret)
2520 		goto free_flr_intr;
2521 
2522 	return numvfs;
2523 free_flr_intr:
2524 	otx2_disable_flr_me_intr(pf);
2525 free_flr:
2526 	otx2_flr_wq_destroy(pf);
2527 free_configs:
2528 	kfree(pf->vf_configs);
2529 free_intr:
2530 	otx2_disable_pfvf_mbox_intr(pf, numvfs);
2531 free_mbox:
2532 	otx2_pfvf_mbox_destroy(pf);
2533 	return ret;
2534 }
2535 
2536 static int otx2_sriov_disable(struct pci_dev *pdev)
2537 {
2538 	struct net_device *netdev = pci_get_drvdata(pdev);
2539 	struct otx2_nic *pf = netdev_priv(netdev);
2540 	int numvfs = pci_num_vf(pdev);
2541 	int i;
2542 
2543 	if (!numvfs)
2544 		return 0;
2545 
2546 	pci_disable_sriov(pdev);
2547 
2548 	for (i = 0; i < pci_num_vf(pdev); i++)
2549 		cancel_delayed_work_sync(&pf->vf_configs[i].link_event_work);
2550 	kfree(pf->vf_configs);
2551 
2552 	otx2_disable_flr_me_intr(pf);
2553 	otx2_flr_wq_destroy(pf);
2554 	otx2_disable_pfvf_mbox_intr(pf, numvfs);
2555 	otx2_pfvf_mbox_destroy(pf);
2556 
2557 	return 0;
2558 }
2559 
2560 static int otx2_sriov_configure(struct pci_dev *pdev, int numvfs)
2561 {
2562 	if (numvfs == 0)
2563 		return otx2_sriov_disable(pdev);
2564 	else
2565 		return otx2_sriov_enable(pdev, numvfs);
2566 }
2567 
2568 static void otx2_remove(struct pci_dev *pdev)
2569 {
2570 	struct net_device *netdev = pci_get_drvdata(pdev);
2571 	struct otx2_nic *pf;
2572 
2573 	if (!netdev)
2574 		return;
2575 
2576 	pf = netdev_priv(netdev);
2577 
2578 	pf->flags |= OTX2_FLAG_PF_SHUTDOWN;
2579 
2580 	if (pf->flags & OTX2_FLAG_TX_TSTAMP_ENABLED)
2581 		otx2_config_hw_tx_tstamp(pf, false);
2582 	if (pf->flags & OTX2_FLAG_RX_TSTAMP_ENABLED)
2583 		otx2_config_hw_rx_tstamp(pf, false);
2584 
2585 	cancel_work_sync(&pf->reset_task);
2586 	/* Disable link notifications */
2587 	otx2_cgx_config_linkevents(pf, false);
2588 
2589 	unregister_netdev(netdev);
2590 	otx2_sriov_disable(pf->pdev);
2591 	if (pf->otx2_wq)
2592 		destroy_workqueue(pf->otx2_wq);
2593 
2594 	otx2_ptp_destroy(pf);
2595 	otx2_mcam_flow_del(pf);
2596 	otx2_detach_resources(&pf->mbox);
2597 	otx2_disable_mbox_intr(pf);
2598 	otx2_pfaf_mbox_destroy(pf);
2599 	pci_free_irq_vectors(pf->pdev);
2600 	pci_set_drvdata(pdev, NULL);
2601 	free_netdev(netdev);
2602 
2603 	pci_release_regions(pdev);
2604 }
2605 
2606 static struct pci_driver otx2_pf_driver = {
2607 	.name = DRV_NAME,
2608 	.id_table = otx2_pf_id_table,
2609 	.probe = otx2_probe,
2610 	.shutdown = otx2_remove,
2611 	.remove = otx2_remove,
2612 	.sriov_configure = otx2_sriov_configure
2613 };
2614 
2615 static int __init otx2_rvupf_init_module(void)
2616 {
2617 	pr_info("%s: %s\n", DRV_NAME, DRV_STRING);
2618 
2619 	return pci_register_driver(&otx2_pf_driver);
2620 }
2621 
2622 static void __exit otx2_rvupf_cleanup_module(void)
2623 {
2624 	pci_unregister_driver(&otx2_pf_driver);
2625 }
2626 
2627 module_init(otx2_rvupf_init_module);
2628 module_exit(otx2_rvupf_cleanup_module);
2629