1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (C) 2021 Gerhard Engleder <gerhard@engleder-embedded.com> */
3 
4 /* TSN endpoint Ethernet MAC driver
5  *
6  * The TSN endpoint Ethernet MAC is a FPGA based network device for real-time
7  * communication. It is designed for endpoints within TSN (Time Sensitive
8  * Networking) networks; e.g., for PLCs in the industrial automation case.
9  *
10  * It supports multiple TX/RX queue pairs. The first TX/RX queue pair is used
11  * by the driver.
12  *
13  * More information can be found here:
14  * - www.embedded-experts.at/tsn
15  * - www.engleder-embedded.com
16  */
17 
18 #include "tsnep.h"
19 #include "tsnep_hw.h"
20 
21 #include <linux/module.h>
22 #include <linux/of.h>
23 #include <linux/of_net.h>
24 #include <linux/of_mdio.h>
25 #include <linux/interrupt.h>
26 #include <linux/etherdevice.h>
27 #include <linux/phy.h>
28 #include <linux/iopoll.h>
29 #include <linux/bpf.h>
30 #include <linux/bpf_trace.h>
31 #include <net/page_pool/helpers.h>
32 #include <net/xdp_sock_drv.h>
33 
34 #define TSNEP_RX_OFFSET (max(NET_SKB_PAD, XDP_PACKET_HEADROOM) + NET_IP_ALIGN)
35 #define TSNEP_HEADROOM ALIGN(TSNEP_RX_OFFSET, 4)
36 #define TSNEP_MAX_RX_BUF_SIZE (PAGE_SIZE - TSNEP_HEADROOM - \
37 			       SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
38 /* XSK buffer shall store at least Q-in-Q frame */
39 #define TSNEP_XSK_RX_BUF_SIZE (ALIGN(TSNEP_RX_INLINE_METADATA_SIZE + \
40 				     ETH_FRAME_LEN + ETH_FCS_LEN + \
41 				     VLAN_HLEN * 2, 4))
42 
43 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
44 #define DMA_ADDR_HIGH(dma_addr) ((u32)(((dma_addr) >> 32) & 0xFFFFFFFF))
45 #else
46 #define DMA_ADDR_HIGH(dma_addr) ((u32)(0))
47 #endif
48 #define DMA_ADDR_LOW(dma_addr) ((u32)((dma_addr) & 0xFFFFFFFF))
49 
50 #define TSNEP_COALESCE_USECS_DEFAULT 64
51 #define TSNEP_COALESCE_USECS_MAX     ((ECM_INT_DELAY_MASK >> ECM_INT_DELAY_SHIFT) * \
52 				      ECM_INT_DELAY_BASE_US + ECM_INT_DELAY_BASE_US - 1)
53 
54 #define TSNEP_TX_TYPE_SKB	BIT(0)
55 #define TSNEP_TX_TYPE_SKB_FRAG	BIT(1)
56 #define TSNEP_TX_TYPE_XDP_TX	BIT(2)
57 #define TSNEP_TX_TYPE_XDP_NDO	BIT(3)
58 #define TSNEP_TX_TYPE_XDP	(TSNEP_TX_TYPE_XDP_TX | TSNEP_TX_TYPE_XDP_NDO)
59 #define TSNEP_TX_TYPE_XSK	BIT(4)
60 
61 #define TSNEP_XDP_TX		BIT(0)
62 #define TSNEP_XDP_REDIRECT	BIT(1)
63 
64 static void tsnep_enable_irq(struct tsnep_adapter *adapter, u32 mask)
65 {
66 	iowrite32(mask, adapter->addr + ECM_INT_ENABLE);
67 }
68 
69 static void tsnep_disable_irq(struct tsnep_adapter *adapter, u32 mask)
70 {
71 	mask |= ECM_INT_DISABLE;
72 	iowrite32(mask, adapter->addr + ECM_INT_ENABLE);
73 }
74 
75 static irqreturn_t tsnep_irq(int irq, void *arg)
76 {
77 	struct tsnep_adapter *adapter = arg;
78 	u32 active = ioread32(adapter->addr + ECM_INT_ACTIVE);
79 
80 	/* acknowledge interrupt */
81 	if (active != 0)
82 		iowrite32(active, adapter->addr + ECM_INT_ACKNOWLEDGE);
83 
84 	/* handle link interrupt */
85 	if ((active & ECM_INT_LINK) != 0)
86 		phy_mac_interrupt(adapter->netdev->phydev);
87 
88 	/* handle TX/RX queue 0 interrupt */
89 	if ((active & adapter->queue[0].irq_mask) != 0) {
90 		if (napi_schedule_prep(&adapter->queue[0].napi)) {
91 			tsnep_disable_irq(adapter, adapter->queue[0].irq_mask);
92 			/* schedule after masking to avoid races */
93 			__napi_schedule(&adapter->queue[0].napi);
94 		}
95 	}
96 
97 	return IRQ_HANDLED;
98 }
99 
100 static irqreturn_t tsnep_irq_txrx(int irq, void *arg)
101 {
102 	struct tsnep_queue *queue = arg;
103 
104 	/* handle TX/RX queue interrupt */
105 	if (napi_schedule_prep(&queue->napi)) {
106 		tsnep_disable_irq(queue->adapter, queue->irq_mask);
107 		/* schedule after masking to avoid races */
108 		__napi_schedule(&queue->napi);
109 	}
110 
111 	return IRQ_HANDLED;
112 }
113 
114 int tsnep_set_irq_coalesce(struct tsnep_queue *queue, u32 usecs)
115 {
116 	if (usecs > TSNEP_COALESCE_USECS_MAX)
117 		return -ERANGE;
118 
119 	usecs /= ECM_INT_DELAY_BASE_US;
120 	usecs <<= ECM_INT_DELAY_SHIFT;
121 	usecs &= ECM_INT_DELAY_MASK;
122 
123 	queue->irq_delay &= ~ECM_INT_DELAY_MASK;
124 	queue->irq_delay |= usecs;
125 	iowrite8(queue->irq_delay, queue->irq_delay_addr);
126 
127 	return 0;
128 }
129 
130 u32 tsnep_get_irq_coalesce(struct tsnep_queue *queue)
131 {
132 	u32 usecs;
133 
134 	usecs = (queue->irq_delay & ECM_INT_DELAY_MASK);
135 	usecs >>= ECM_INT_DELAY_SHIFT;
136 	usecs *= ECM_INT_DELAY_BASE_US;
137 
138 	return usecs;
139 }
140 
141 static int tsnep_mdiobus_read(struct mii_bus *bus, int addr, int regnum)
142 {
143 	struct tsnep_adapter *adapter = bus->priv;
144 	u32 md;
145 	int retval;
146 
147 	md = ECM_MD_READ;
148 	if (!adapter->suppress_preamble)
149 		md |= ECM_MD_PREAMBLE;
150 	md |= (regnum << ECM_MD_ADDR_SHIFT) & ECM_MD_ADDR_MASK;
151 	md |= (addr << ECM_MD_PHY_ADDR_SHIFT) & ECM_MD_PHY_ADDR_MASK;
152 	iowrite32(md, adapter->addr + ECM_MD_CONTROL);
153 	retval = readl_poll_timeout_atomic(adapter->addr + ECM_MD_STATUS, md,
154 					   !(md & ECM_MD_BUSY), 16, 1000);
155 	if (retval != 0)
156 		return retval;
157 
158 	return (md & ECM_MD_DATA_MASK) >> ECM_MD_DATA_SHIFT;
159 }
160 
161 static int tsnep_mdiobus_write(struct mii_bus *bus, int addr, int regnum,
162 			       u16 val)
163 {
164 	struct tsnep_adapter *adapter = bus->priv;
165 	u32 md;
166 	int retval;
167 
168 	md = ECM_MD_WRITE;
169 	if (!adapter->suppress_preamble)
170 		md |= ECM_MD_PREAMBLE;
171 	md |= (regnum << ECM_MD_ADDR_SHIFT) & ECM_MD_ADDR_MASK;
172 	md |= (addr << ECM_MD_PHY_ADDR_SHIFT) & ECM_MD_PHY_ADDR_MASK;
173 	md |= ((u32)val << ECM_MD_DATA_SHIFT) & ECM_MD_DATA_MASK;
174 	iowrite32(md, adapter->addr + ECM_MD_CONTROL);
175 	retval = readl_poll_timeout_atomic(adapter->addr + ECM_MD_STATUS, md,
176 					   !(md & ECM_MD_BUSY), 16, 1000);
177 	if (retval != 0)
178 		return retval;
179 
180 	return 0;
181 }
182 
183 static void tsnep_set_link_mode(struct tsnep_adapter *adapter)
184 {
185 	u32 mode;
186 
187 	switch (adapter->phydev->speed) {
188 	case SPEED_100:
189 		mode = ECM_LINK_MODE_100;
190 		break;
191 	case SPEED_1000:
192 		mode = ECM_LINK_MODE_1000;
193 		break;
194 	default:
195 		mode = ECM_LINK_MODE_OFF;
196 		break;
197 	}
198 	iowrite32(mode, adapter->addr + ECM_STATUS);
199 }
200 
201 static void tsnep_phy_link_status_change(struct net_device *netdev)
202 {
203 	struct tsnep_adapter *adapter = netdev_priv(netdev);
204 	struct phy_device *phydev = netdev->phydev;
205 
206 	if (phydev->link)
207 		tsnep_set_link_mode(adapter);
208 
209 	phy_print_status(netdev->phydev);
210 }
211 
212 static int tsnep_phy_loopback(struct tsnep_adapter *adapter, bool enable)
213 {
214 	int retval;
215 
216 	retval = phy_loopback(adapter->phydev, enable);
217 
218 	/* PHY link state change is not signaled if loopback is enabled, it
219 	 * would delay a working loopback anyway, let's ensure that loopback
220 	 * is working immediately by setting link mode directly
221 	 */
222 	if (!retval && enable)
223 		tsnep_set_link_mode(adapter);
224 
225 	return retval;
226 }
227 
228 static int tsnep_phy_open(struct tsnep_adapter *adapter)
229 {
230 	struct phy_device *phydev;
231 	struct ethtool_eee ethtool_eee;
232 	int retval;
233 
234 	retval = phy_connect_direct(adapter->netdev, adapter->phydev,
235 				    tsnep_phy_link_status_change,
236 				    adapter->phy_mode);
237 	if (retval)
238 		return retval;
239 	phydev = adapter->netdev->phydev;
240 
241 	/* MAC supports only 100Mbps|1000Mbps full duplex
242 	 * SPE (Single Pair Ethernet) is also an option but not implemented yet
243 	 */
244 	phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_10baseT_Half_BIT);
245 	phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_10baseT_Full_BIT);
246 	phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_100baseT_Half_BIT);
247 	phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_1000baseT_Half_BIT);
248 
249 	/* disable EEE autoneg, EEE not supported by TSNEP */
250 	memset(&ethtool_eee, 0, sizeof(ethtool_eee));
251 	phy_ethtool_set_eee(adapter->phydev, &ethtool_eee);
252 
253 	adapter->phydev->irq = PHY_MAC_INTERRUPT;
254 	phy_start(adapter->phydev);
255 
256 	return 0;
257 }
258 
259 static void tsnep_phy_close(struct tsnep_adapter *adapter)
260 {
261 	phy_stop(adapter->netdev->phydev);
262 	phy_disconnect(adapter->netdev->phydev);
263 }
264 
265 static void tsnep_tx_ring_cleanup(struct tsnep_tx *tx)
266 {
267 	struct device *dmadev = tx->adapter->dmadev;
268 	int i;
269 
270 	memset(tx->entry, 0, sizeof(tx->entry));
271 
272 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
273 		if (tx->page[i]) {
274 			dma_free_coherent(dmadev, PAGE_SIZE, tx->page[i],
275 					  tx->page_dma[i]);
276 			tx->page[i] = NULL;
277 			tx->page_dma[i] = 0;
278 		}
279 	}
280 }
281 
282 static int tsnep_tx_ring_create(struct tsnep_tx *tx)
283 {
284 	struct device *dmadev = tx->adapter->dmadev;
285 	struct tsnep_tx_entry *entry;
286 	struct tsnep_tx_entry *next_entry;
287 	int i, j;
288 	int retval;
289 
290 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
291 		tx->page[i] =
292 			dma_alloc_coherent(dmadev, PAGE_SIZE, &tx->page_dma[i],
293 					   GFP_KERNEL);
294 		if (!tx->page[i]) {
295 			retval = -ENOMEM;
296 			goto alloc_failed;
297 		}
298 		for (j = 0; j < TSNEP_RING_ENTRIES_PER_PAGE; j++) {
299 			entry = &tx->entry[TSNEP_RING_ENTRIES_PER_PAGE * i + j];
300 			entry->desc_wb = (struct tsnep_tx_desc_wb *)
301 				(((u8 *)tx->page[i]) + TSNEP_DESC_SIZE * j);
302 			entry->desc = (struct tsnep_tx_desc *)
303 				(((u8 *)entry->desc_wb) + TSNEP_DESC_OFFSET);
304 			entry->desc_dma = tx->page_dma[i] + TSNEP_DESC_SIZE * j;
305 			entry->owner_user_flag = false;
306 		}
307 	}
308 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
309 		entry = &tx->entry[i];
310 		next_entry = &tx->entry[(i + 1) & TSNEP_RING_MASK];
311 		entry->desc->next = __cpu_to_le64(next_entry->desc_dma);
312 	}
313 
314 	return 0;
315 
316 alloc_failed:
317 	tsnep_tx_ring_cleanup(tx);
318 	return retval;
319 }
320 
321 static void tsnep_tx_init(struct tsnep_tx *tx)
322 {
323 	dma_addr_t dma;
324 
325 	dma = tx->entry[0].desc_dma | TSNEP_RESET_OWNER_COUNTER;
326 	iowrite32(DMA_ADDR_LOW(dma), tx->addr + TSNEP_TX_DESC_ADDR_LOW);
327 	iowrite32(DMA_ADDR_HIGH(dma), tx->addr + TSNEP_TX_DESC_ADDR_HIGH);
328 	tx->write = 0;
329 	tx->read = 0;
330 	tx->owner_counter = 1;
331 	tx->increment_owner_counter = TSNEP_RING_SIZE - 1;
332 }
333 
334 static void tsnep_tx_enable(struct tsnep_tx *tx)
335 {
336 	struct netdev_queue *nq;
337 
338 	nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
339 
340 	__netif_tx_lock_bh(nq);
341 	netif_tx_wake_queue(nq);
342 	__netif_tx_unlock_bh(nq);
343 }
344 
345 static void tsnep_tx_disable(struct tsnep_tx *tx, struct napi_struct *napi)
346 {
347 	struct netdev_queue *nq;
348 	u32 val;
349 
350 	nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
351 
352 	__netif_tx_lock_bh(nq);
353 	netif_tx_stop_queue(nq);
354 	__netif_tx_unlock_bh(nq);
355 
356 	/* wait until TX is done in hardware */
357 	readx_poll_timeout(ioread32, tx->addr + TSNEP_CONTROL, val,
358 			   ((val & TSNEP_CONTROL_TX_ENABLE) == 0), 10000,
359 			   1000000);
360 
361 	/* wait until TX is also done in software */
362 	while (READ_ONCE(tx->read) != tx->write) {
363 		napi_schedule(napi);
364 		napi_synchronize(napi);
365 	}
366 }
367 
368 static void tsnep_tx_activate(struct tsnep_tx *tx, int index, int length,
369 			      bool last)
370 {
371 	struct tsnep_tx_entry *entry = &tx->entry[index];
372 
373 	entry->properties = 0;
374 	/* xdpf and zc are union with skb */
375 	if (entry->skb) {
376 		entry->properties = length & TSNEP_DESC_LENGTH_MASK;
377 		entry->properties |= TSNEP_DESC_INTERRUPT_FLAG;
378 		if ((entry->type & TSNEP_TX_TYPE_SKB) &&
379 		    (skb_shinfo(entry->skb)->tx_flags & SKBTX_IN_PROGRESS))
380 			entry->properties |= TSNEP_DESC_EXTENDED_WRITEBACK_FLAG;
381 
382 		/* toggle user flag to prevent false acknowledge
383 		 *
384 		 * Only the first fragment is acknowledged. For all other
385 		 * fragments no acknowledge is done and the last written owner
386 		 * counter stays in the writeback descriptor. Therefore, it is
387 		 * possible that the last written owner counter is identical to
388 		 * the new incremented owner counter and a false acknowledge is
389 		 * detected before the real acknowledge has been done by
390 		 * hardware.
391 		 *
392 		 * The user flag is used to prevent this situation. The user
393 		 * flag is copied to the writeback descriptor by the hardware
394 		 * and is used as additional acknowledge data. By toggeling the
395 		 * user flag only for the first fragment (which is
396 		 * acknowledged), it is guaranteed that the last acknowledge
397 		 * done for this descriptor has used a different user flag and
398 		 * cannot be detected as false acknowledge.
399 		 */
400 		entry->owner_user_flag = !entry->owner_user_flag;
401 	}
402 	if (last)
403 		entry->properties |= TSNEP_TX_DESC_LAST_FRAGMENT_FLAG;
404 	if (index == tx->increment_owner_counter) {
405 		tx->owner_counter++;
406 		if (tx->owner_counter == 4)
407 			tx->owner_counter = 1;
408 		tx->increment_owner_counter--;
409 		if (tx->increment_owner_counter < 0)
410 			tx->increment_owner_counter = TSNEP_RING_SIZE - 1;
411 	}
412 	entry->properties |=
413 		(tx->owner_counter << TSNEP_DESC_OWNER_COUNTER_SHIFT) &
414 		TSNEP_DESC_OWNER_COUNTER_MASK;
415 	if (entry->owner_user_flag)
416 		entry->properties |= TSNEP_TX_DESC_OWNER_USER_FLAG;
417 	entry->desc->more_properties =
418 		__cpu_to_le32(entry->len & TSNEP_DESC_LENGTH_MASK);
419 
420 	/* descriptor properties shall be written last, because valid data is
421 	 * signaled there
422 	 */
423 	dma_wmb();
424 
425 	entry->desc->properties = __cpu_to_le32(entry->properties);
426 }
427 
428 static int tsnep_tx_desc_available(struct tsnep_tx *tx)
429 {
430 	if (tx->read <= tx->write)
431 		return TSNEP_RING_SIZE - tx->write + tx->read - 1;
432 	else
433 		return tx->read - tx->write - 1;
434 }
435 
436 static int tsnep_tx_map(struct sk_buff *skb, struct tsnep_tx *tx, int count)
437 {
438 	struct device *dmadev = tx->adapter->dmadev;
439 	struct tsnep_tx_entry *entry;
440 	unsigned int len;
441 	dma_addr_t dma;
442 	int map_len = 0;
443 	int i;
444 
445 	for (i = 0; i < count; i++) {
446 		entry = &tx->entry[(tx->write + i) & TSNEP_RING_MASK];
447 
448 		if (!i) {
449 			len = skb_headlen(skb);
450 			dma = dma_map_single(dmadev, skb->data, len,
451 					     DMA_TO_DEVICE);
452 
453 			entry->type = TSNEP_TX_TYPE_SKB;
454 		} else {
455 			len = skb_frag_size(&skb_shinfo(skb)->frags[i - 1]);
456 			dma = skb_frag_dma_map(dmadev,
457 					       &skb_shinfo(skb)->frags[i - 1],
458 					       0, len, DMA_TO_DEVICE);
459 
460 			entry->type = TSNEP_TX_TYPE_SKB_FRAG;
461 		}
462 		if (dma_mapping_error(dmadev, dma))
463 			return -ENOMEM;
464 
465 		entry->len = len;
466 		dma_unmap_addr_set(entry, dma, dma);
467 
468 		entry->desc->tx = __cpu_to_le64(dma);
469 
470 		map_len += len;
471 	}
472 
473 	return map_len;
474 }
475 
476 static int tsnep_tx_unmap(struct tsnep_tx *tx, int index, int count)
477 {
478 	struct device *dmadev = tx->adapter->dmadev;
479 	struct tsnep_tx_entry *entry;
480 	int map_len = 0;
481 	int i;
482 
483 	for (i = 0; i < count; i++) {
484 		entry = &tx->entry[(index + i) & TSNEP_RING_MASK];
485 
486 		if (entry->len) {
487 			if (entry->type & TSNEP_TX_TYPE_SKB)
488 				dma_unmap_single(dmadev,
489 						 dma_unmap_addr(entry, dma),
490 						 dma_unmap_len(entry, len),
491 						 DMA_TO_DEVICE);
492 			else if (entry->type &
493 				 (TSNEP_TX_TYPE_SKB_FRAG | TSNEP_TX_TYPE_XDP_NDO))
494 				dma_unmap_page(dmadev,
495 					       dma_unmap_addr(entry, dma),
496 					       dma_unmap_len(entry, len),
497 					       DMA_TO_DEVICE);
498 			map_len += entry->len;
499 			entry->len = 0;
500 		}
501 	}
502 
503 	return map_len;
504 }
505 
506 static netdev_tx_t tsnep_xmit_frame_ring(struct sk_buff *skb,
507 					 struct tsnep_tx *tx)
508 {
509 	int count = 1;
510 	struct tsnep_tx_entry *entry;
511 	int length;
512 	int i;
513 	int retval;
514 
515 	if (skb_shinfo(skb)->nr_frags > 0)
516 		count += skb_shinfo(skb)->nr_frags;
517 
518 	if (tsnep_tx_desc_available(tx) < count) {
519 		/* ring full, shall not happen because queue is stopped if full
520 		 * below
521 		 */
522 		netif_stop_subqueue(tx->adapter->netdev, tx->queue_index);
523 
524 		return NETDEV_TX_BUSY;
525 	}
526 
527 	entry = &tx->entry[tx->write];
528 	entry->skb = skb;
529 
530 	retval = tsnep_tx_map(skb, tx, count);
531 	if (retval < 0) {
532 		tsnep_tx_unmap(tx, tx->write, count);
533 		dev_kfree_skb_any(entry->skb);
534 		entry->skb = NULL;
535 
536 		tx->dropped++;
537 
538 		return NETDEV_TX_OK;
539 	}
540 	length = retval;
541 
542 	if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)
543 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
544 
545 	for (i = 0; i < count; i++)
546 		tsnep_tx_activate(tx, (tx->write + i) & TSNEP_RING_MASK, length,
547 				  i == count - 1);
548 	tx->write = (tx->write + count) & TSNEP_RING_MASK;
549 
550 	skb_tx_timestamp(skb);
551 
552 	/* descriptor properties shall be valid before hardware is notified */
553 	dma_wmb();
554 
555 	iowrite32(TSNEP_CONTROL_TX_ENABLE, tx->addr + TSNEP_CONTROL);
556 
557 	if (tsnep_tx_desc_available(tx) < (MAX_SKB_FRAGS + 1)) {
558 		/* ring can get full with next frame */
559 		netif_stop_subqueue(tx->adapter->netdev, tx->queue_index);
560 	}
561 
562 	return NETDEV_TX_OK;
563 }
564 
565 static int tsnep_xdp_tx_map(struct xdp_frame *xdpf, struct tsnep_tx *tx,
566 			    struct skb_shared_info *shinfo, int count, u32 type)
567 {
568 	struct device *dmadev = tx->adapter->dmadev;
569 	struct tsnep_tx_entry *entry;
570 	struct page *page;
571 	skb_frag_t *frag;
572 	unsigned int len;
573 	int map_len = 0;
574 	dma_addr_t dma;
575 	void *data;
576 	int i;
577 
578 	frag = NULL;
579 	len = xdpf->len;
580 	for (i = 0; i < count; i++) {
581 		entry = &tx->entry[(tx->write + i) & TSNEP_RING_MASK];
582 		if (type & TSNEP_TX_TYPE_XDP_NDO) {
583 			data = unlikely(frag) ? skb_frag_address(frag) :
584 						xdpf->data;
585 			dma = dma_map_single(dmadev, data, len, DMA_TO_DEVICE);
586 			if (dma_mapping_error(dmadev, dma))
587 				return -ENOMEM;
588 
589 			entry->type = TSNEP_TX_TYPE_XDP_NDO;
590 		} else {
591 			page = unlikely(frag) ? skb_frag_page(frag) :
592 						virt_to_page(xdpf->data);
593 			dma = page_pool_get_dma_addr(page);
594 			if (unlikely(frag))
595 				dma += skb_frag_off(frag);
596 			else
597 				dma += sizeof(*xdpf) + xdpf->headroom;
598 			dma_sync_single_for_device(dmadev, dma, len,
599 						   DMA_BIDIRECTIONAL);
600 
601 			entry->type = TSNEP_TX_TYPE_XDP_TX;
602 		}
603 
604 		entry->len = len;
605 		dma_unmap_addr_set(entry, dma, dma);
606 
607 		entry->desc->tx = __cpu_to_le64(dma);
608 
609 		map_len += len;
610 
611 		if (i + 1 < count) {
612 			frag = &shinfo->frags[i];
613 			len = skb_frag_size(frag);
614 		}
615 	}
616 
617 	return map_len;
618 }
619 
620 /* This function requires __netif_tx_lock is held by the caller. */
621 static bool tsnep_xdp_xmit_frame_ring(struct xdp_frame *xdpf,
622 				      struct tsnep_tx *tx, u32 type)
623 {
624 	struct skb_shared_info *shinfo = xdp_get_shared_info_from_frame(xdpf);
625 	struct tsnep_tx_entry *entry;
626 	int count, length, retval, i;
627 
628 	count = 1;
629 	if (unlikely(xdp_frame_has_frags(xdpf)))
630 		count += shinfo->nr_frags;
631 
632 	/* ensure that TX ring is not filled up by XDP, always MAX_SKB_FRAGS
633 	 * will be available for normal TX path and queue is stopped there if
634 	 * necessary
635 	 */
636 	if (tsnep_tx_desc_available(tx) < (MAX_SKB_FRAGS + 1 + count))
637 		return false;
638 
639 	entry = &tx->entry[tx->write];
640 	entry->xdpf = xdpf;
641 
642 	retval = tsnep_xdp_tx_map(xdpf, tx, shinfo, count, type);
643 	if (retval < 0) {
644 		tsnep_tx_unmap(tx, tx->write, count);
645 		entry->xdpf = NULL;
646 
647 		tx->dropped++;
648 
649 		return false;
650 	}
651 	length = retval;
652 
653 	for (i = 0; i < count; i++)
654 		tsnep_tx_activate(tx, (tx->write + i) & TSNEP_RING_MASK, length,
655 				  i == count - 1);
656 	tx->write = (tx->write + count) & TSNEP_RING_MASK;
657 
658 	/* descriptor properties shall be valid before hardware is notified */
659 	dma_wmb();
660 
661 	return true;
662 }
663 
664 static void tsnep_xdp_xmit_flush(struct tsnep_tx *tx)
665 {
666 	iowrite32(TSNEP_CONTROL_TX_ENABLE, tx->addr + TSNEP_CONTROL);
667 }
668 
669 static bool tsnep_xdp_xmit_back(struct tsnep_adapter *adapter,
670 				struct xdp_buff *xdp,
671 				struct netdev_queue *tx_nq, struct tsnep_tx *tx,
672 				bool zc)
673 {
674 	struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
675 	bool xmit;
676 	u32 type;
677 
678 	if (unlikely(!xdpf))
679 		return false;
680 
681 	/* no page pool for zero copy */
682 	if (zc)
683 		type = TSNEP_TX_TYPE_XDP_NDO;
684 	else
685 		type = TSNEP_TX_TYPE_XDP_TX;
686 
687 	__netif_tx_lock(tx_nq, smp_processor_id());
688 
689 	xmit = tsnep_xdp_xmit_frame_ring(xdpf, tx, type);
690 
691 	/* Avoid transmit queue timeout since we share it with the slow path */
692 	if (xmit)
693 		txq_trans_cond_update(tx_nq);
694 
695 	__netif_tx_unlock(tx_nq);
696 
697 	return xmit;
698 }
699 
700 static int tsnep_xdp_tx_map_zc(struct xdp_desc *xdpd, struct tsnep_tx *tx)
701 {
702 	struct tsnep_tx_entry *entry;
703 	dma_addr_t dma;
704 
705 	entry = &tx->entry[tx->write];
706 	entry->zc = true;
707 
708 	dma = xsk_buff_raw_get_dma(tx->xsk_pool, xdpd->addr);
709 	xsk_buff_raw_dma_sync_for_device(tx->xsk_pool, dma, xdpd->len);
710 
711 	entry->type = TSNEP_TX_TYPE_XSK;
712 	entry->len = xdpd->len;
713 
714 	entry->desc->tx = __cpu_to_le64(dma);
715 
716 	return xdpd->len;
717 }
718 
719 static void tsnep_xdp_xmit_frame_ring_zc(struct xdp_desc *xdpd,
720 					 struct tsnep_tx *tx)
721 {
722 	int length;
723 
724 	length = tsnep_xdp_tx_map_zc(xdpd, tx);
725 
726 	tsnep_tx_activate(tx, tx->write, length, true);
727 	tx->write = (tx->write + 1) & TSNEP_RING_MASK;
728 }
729 
730 static void tsnep_xdp_xmit_zc(struct tsnep_tx *tx)
731 {
732 	int desc_available = tsnep_tx_desc_available(tx);
733 	struct xdp_desc *descs = tx->xsk_pool->tx_descs;
734 	int batch, i;
735 
736 	/* ensure that TX ring is not filled up by XDP, always MAX_SKB_FRAGS
737 	 * will be available for normal TX path and queue is stopped there if
738 	 * necessary
739 	 */
740 	if (desc_available <= (MAX_SKB_FRAGS + 1))
741 		return;
742 	desc_available -= MAX_SKB_FRAGS + 1;
743 
744 	batch = xsk_tx_peek_release_desc_batch(tx->xsk_pool, desc_available);
745 	for (i = 0; i < batch; i++)
746 		tsnep_xdp_xmit_frame_ring_zc(&descs[i], tx);
747 
748 	if (batch) {
749 		/* descriptor properties shall be valid before hardware is
750 		 * notified
751 		 */
752 		dma_wmb();
753 
754 		tsnep_xdp_xmit_flush(tx);
755 	}
756 }
757 
758 static bool tsnep_tx_poll(struct tsnep_tx *tx, int napi_budget)
759 {
760 	struct tsnep_tx_entry *entry;
761 	struct netdev_queue *nq;
762 	int xsk_frames = 0;
763 	int budget = 128;
764 	int length;
765 	int count;
766 
767 	nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
768 	__netif_tx_lock(nq, smp_processor_id());
769 
770 	do {
771 		if (tx->read == tx->write)
772 			break;
773 
774 		entry = &tx->entry[tx->read];
775 		if ((__le32_to_cpu(entry->desc_wb->properties) &
776 		     TSNEP_TX_DESC_OWNER_MASK) !=
777 		    (entry->properties & TSNEP_TX_DESC_OWNER_MASK))
778 			break;
779 
780 		/* descriptor properties shall be read first, because valid data
781 		 * is signaled there
782 		 */
783 		dma_rmb();
784 
785 		count = 1;
786 		if ((entry->type & TSNEP_TX_TYPE_SKB) &&
787 		    skb_shinfo(entry->skb)->nr_frags > 0)
788 			count += skb_shinfo(entry->skb)->nr_frags;
789 		else if ((entry->type & TSNEP_TX_TYPE_XDP) &&
790 			 xdp_frame_has_frags(entry->xdpf))
791 			count += xdp_get_shared_info_from_frame(entry->xdpf)->nr_frags;
792 
793 		length = tsnep_tx_unmap(tx, tx->read, count);
794 
795 		if ((entry->type & TSNEP_TX_TYPE_SKB) &&
796 		    (skb_shinfo(entry->skb)->tx_flags & SKBTX_IN_PROGRESS) &&
797 		    (__le32_to_cpu(entry->desc_wb->properties) &
798 		     TSNEP_DESC_EXTENDED_WRITEBACK_FLAG)) {
799 			struct skb_shared_hwtstamps hwtstamps;
800 			u64 timestamp;
801 
802 			if (skb_shinfo(entry->skb)->tx_flags &
803 			    SKBTX_HW_TSTAMP_USE_CYCLES)
804 				timestamp =
805 					__le64_to_cpu(entry->desc_wb->counter);
806 			else
807 				timestamp =
808 					__le64_to_cpu(entry->desc_wb->timestamp);
809 
810 			memset(&hwtstamps, 0, sizeof(hwtstamps));
811 			hwtstamps.hwtstamp = ns_to_ktime(timestamp);
812 
813 			skb_tstamp_tx(entry->skb, &hwtstamps);
814 		}
815 
816 		if (entry->type & TSNEP_TX_TYPE_SKB)
817 			napi_consume_skb(entry->skb, napi_budget);
818 		else if (entry->type & TSNEP_TX_TYPE_XDP)
819 			xdp_return_frame_rx_napi(entry->xdpf);
820 		else
821 			xsk_frames++;
822 		/* xdpf and zc are union with skb */
823 		entry->skb = NULL;
824 
825 		tx->read = (tx->read + count) & TSNEP_RING_MASK;
826 
827 		tx->packets++;
828 		tx->bytes += length + ETH_FCS_LEN;
829 
830 		budget--;
831 	} while (likely(budget));
832 
833 	if (tx->xsk_pool) {
834 		if (xsk_frames)
835 			xsk_tx_completed(tx->xsk_pool, xsk_frames);
836 		if (xsk_uses_need_wakeup(tx->xsk_pool))
837 			xsk_set_tx_need_wakeup(tx->xsk_pool);
838 		tsnep_xdp_xmit_zc(tx);
839 	}
840 
841 	if ((tsnep_tx_desc_available(tx) >= ((MAX_SKB_FRAGS + 1) * 2)) &&
842 	    netif_tx_queue_stopped(nq)) {
843 		netif_tx_wake_queue(nq);
844 	}
845 
846 	__netif_tx_unlock(nq);
847 
848 	return budget != 0;
849 }
850 
851 static bool tsnep_tx_pending(struct tsnep_tx *tx)
852 {
853 	struct tsnep_tx_entry *entry;
854 	struct netdev_queue *nq;
855 	bool pending = false;
856 
857 	nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
858 	__netif_tx_lock(nq, smp_processor_id());
859 
860 	if (tx->read != tx->write) {
861 		entry = &tx->entry[tx->read];
862 		if ((__le32_to_cpu(entry->desc_wb->properties) &
863 		     TSNEP_TX_DESC_OWNER_MASK) ==
864 		    (entry->properties & TSNEP_TX_DESC_OWNER_MASK))
865 			pending = true;
866 	}
867 
868 	__netif_tx_unlock(nq);
869 
870 	return pending;
871 }
872 
873 static int tsnep_tx_open(struct tsnep_tx *tx)
874 {
875 	int retval;
876 
877 	retval = tsnep_tx_ring_create(tx);
878 	if (retval)
879 		return retval;
880 
881 	tsnep_tx_init(tx);
882 
883 	return 0;
884 }
885 
886 static void tsnep_tx_close(struct tsnep_tx *tx)
887 {
888 	tsnep_tx_ring_cleanup(tx);
889 }
890 
891 static void tsnep_rx_ring_cleanup(struct tsnep_rx *rx)
892 {
893 	struct device *dmadev = rx->adapter->dmadev;
894 	struct tsnep_rx_entry *entry;
895 	int i;
896 
897 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
898 		entry = &rx->entry[i];
899 		if (!rx->xsk_pool && entry->page)
900 			page_pool_put_full_page(rx->page_pool, entry->page,
901 						false);
902 		if (rx->xsk_pool && entry->xdp)
903 			xsk_buff_free(entry->xdp);
904 		/* xdp is union with page */
905 		entry->page = NULL;
906 	}
907 
908 	if (rx->page_pool)
909 		page_pool_destroy(rx->page_pool);
910 
911 	memset(rx->entry, 0, sizeof(rx->entry));
912 
913 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
914 		if (rx->page[i]) {
915 			dma_free_coherent(dmadev, PAGE_SIZE, rx->page[i],
916 					  rx->page_dma[i]);
917 			rx->page[i] = NULL;
918 			rx->page_dma[i] = 0;
919 		}
920 	}
921 }
922 
923 static int tsnep_rx_ring_create(struct tsnep_rx *rx)
924 {
925 	struct device *dmadev = rx->adapter->dmadev;
926 	struct tsnep_rx_entry *entry;
927 	struct page_pool_params pp_params = { 0 };
928 	struct tsnep_rx_entry *next_entry;
929 	int i, j;
930 	int retval;
931 
932 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
933 		rx->page[i] =
934 			dma_alloc_coherent(dmadev, PAGE_SIZE, &rx->page_dma[i],
935 					   GFP_KERNEL);
936 		if (!rx->page[i]) {
937 			retval = -ENOMEM;
938 			goto failed;
939 		}
940 		for (j = 0; j < TSNEP_RING_ENTRIES_PER_PAGE; j++) {
941 			entry = &rx->entry[TSNEP_RING_ENTRIES_PER_PAGE * i + j];
942 			entry->desc_wb = (struct tsnep_rx_desc_wb *)
943 				(((u8 *)rx->page[i]) + TSNEP_DESC_SIZE * j);
944 			entry->desc = (struct tsnep_rx_desc *)
945 				(((u8 *)entry->desc_wb) + TSNEP_DESC_OFFSET);
946 			entry->desc_dma = rx->page_dma[i] + TSNEP_DESC_SIZE * j;
947 		}
948 	}
949 
950 	pp_params.flags = PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV;
951 	pp_params.order = 0;
952 	pp_params.pool_size = TSNEP_RING_SIZE;
953 	pp_params.nid = dev_to_node(dmadev);
954 	pp_params.dev = dmadev;
955 	pp_params.dma_dir = DMA_BIDIRECTIONAL;
956 	pp_params.max_len = TSNEP_MAX_RX_BUF_SIZE;
957 	pp_params.offset = TSNEP_RX_OFFSET;
958 	rx->page_pool = page_pool_create(&pp_params);
959 	if (IS_ERR(rx->page_pool)) {
960 		retval = PTR_ERR(rx->page_pool);
961 		rx->page_pool = NULL;
962 		goto failed;
963 	}
964 
965 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
966 		entry = &rx->entry[i];
967 		next_entry = &rx->entry[(i + 1) & TSNEP_RING_MASK];
968 		entry->desc->next = __cpu_to_le64(next_entry->desc_dma);
969 	}
970 
971 	return 0;
972 
973 failed:
974 	tsnep_rx_ring_cleanup(rx);
975 	return retval;
976 }
977 
978 static void tsnep_rx_init(struct tsnep_rx *rx)
979 {
980 	dma_addr_t dma;
981 
982 	dma = rx->entry[0].desc_dma | TSNEP_RESET_OWNER_COUNTER;
983 	iowrite32(DMA_ADDR_LOW(dma), rx->addr + TSNEP_RX_DESC_ADDR_LOW);
984 	iowrite32(DMA_ADDR_HIGH(dma), rx->addr + TSNEP_RX_DESC_ADDR_HIGH);
985 	rx->write = 0;
986 	rx->read = 0;
987 	rx->owner_counter = 1;
988 	rx->increment_owner_counter = TSNEP_RING_SIZE - 1;
989 }
990 
991 static void tsnep_rx_enable(struct tsnep_rx *rx)
992 {
993 	/* descriptor properties shall be valid before hardware is notified */
994 	dma_wmb();
995 
996 	iowrite32(TSNEP_CONTROL_RX_ENABLE, rx->addr + TSNEP_CONTROL);
997 }
998 
999 static void tsnep_rx_disable(struct tsnep_rx *rx)
1000 {
1001 	u32 val;
1002 
1003 	iowrite32(TSNEP_CONTROL_RX_DISABLE, rx->addr + TSNEP_CONTROL);
1004 	readx_poll_timeout(ioread32, rx->addr + TSNEP_CONTROL, val,
1005 			   ((val & TSNEP_CONTROL_RX_ENABLE) == 0), 10000,
1006 			   1000000);
1007 }
1008 
1009 static int tsnep_rx_desc_available(struct tsnep_rx *rx)
1010 {
1011 	if (rx->read <= rx->write)
1012 		return TSNEP_RING_SIZE - rx->write + rx->read - 1;
1013 	else
1014 		return rx->read - rx->write - 1;
1015 }
1016 
1017 static void tsnep_rx_free_page_buffer(struct tsnep_rx *rx)
1018 {
1019 	struct page **page;
1020 
1021 	/* last entry of page_buffer is always zero, because ring cannot be
1022 	 * filled completely
1023 	 */
1024 	page = rx->page_buffer;
1025 	while (*page) {
1026 		page_pool_put_full_page(rx->page_pool, *page, false);
1027 		*page = NULL;
1028 		page++;
1029 	}
1030 }
1031 
1032 static int tsnep_rx_alloc_page_buffer(struct tsnep_rx *rx)
1033 {
1034 	int i;
1035 
1036 	/* alloc for all ring entries except the last one, because ring cannot
1037 	 * be filled completely
1038 	 */
1039 	for (i = 0; i < TSNEP_RING_SIZE - 1; i++) {
1040 		rx->page_buffer[i] = page_pool_dev_alloc_pages(rx->page_pool);
1041 		if (!rx->page_buffer[i]) {
1042 			tsnep_rx_free_page_buffer(rx);
1043 
1044 			return -ENOMEM;
1045 		}
1046 	}
1047 
1048 	return 0;
1049 }
1050 
1051 static void tsnep_rx_set_page(struct tsnep_rx *rx, struct tsnep_rx_entry *entry,
1052 			      struct page *page)
1053 {
1054 	entry->page = page;
1055 	entry->len = TSNEP_MAX_RX_BUF_SIZE;
1056 	entry->dma = page_pool_get_dma_addr(entry->page);
1057 	entry->desc->rx = __cpu_to_le64(entry->dma + TSNEP_RX_OFFSET);
1058 }
1059 
1060 static int tsnep_rx_alloc_buffer(struct tsnep_rx *rx, int index)
1061 {
1062 	struct tsnep_rx_entry *entry = &rx->entry[index];
1063 	struct page *page;
1064 
1065 	page = page_pool_dev_alloc_pages(rx->page_pool);
1066 	if (unlikely(!page))
1067 		return -ENOMEM;
1068 	tsnep_rx_set_page(rx, entry, page);
1069 
1070 	return 0;
1071 }
1072 
1073 static void tsnep_rx_reuse_buffer(struct tsnep_rx *rx, int index)
1074 {
1075 	struct tsnep_rx_entry *entry = &rx->entry[index];
1076 	struct tsnep_rx_entry *read = &rx->entry[rx->read];
1077 
1078 	tsnep_rx_set_page(rx, entry, read->page);
1079 	read->page = NULL;
1080 }
1081 
1082 static void tsnep_rx_activate(struct tsnep_rx *rx, int index)
1083 {
1084 	struct tsnep_rx_entry *entry = &rx->entry[index];
1085 
1086 	/* TSNEP_MAX_RX_BUF_SIZE and TSNEP_XSK_RX_BUF_SIZE are multiple of 4 */
1087 	entry->properties = entry->len & TSNEP_DESC_LENGTH_MASK;
1088 	entry->properties |= TSNEP_DESC_INTERRUPT_FLAG;
1089 	if (index == rx->increment_owner_counter) {
1090 		rx->owner_counter++;
1091 		if (rx->owner_counter == 4)
1092 			rx->owner_counter = 1;
1093 		rx->increment_owner_counter--;
1094 		if (rx->increment_owner_counter < 0)
1095 			rx->increment_owner_counter = TSNEP_RING_SIZE - 1;
1096 	}
1097 	entry->properties |=
1098 		(rx->owner_counter << TSNEP_DESC_OWNER_COUNTER_SHIFT) &
1099 		TSNEP_DESC_OWNER_COUNTER_MASK;
1100 
1101 	/* descriptor properties shall be written last, because valid data is
1102 	 * signaled there
1103 	 */
1104 	dma_wmb();
1105 
1106 	entry->desc->properties = __cpu_to_le32(entry->properties);
1107 }
1108 
1109 static int tsnep_rx_alloc(struct tsnep_rx *rx, int count, bool reuse)
1110 {
1111 	bool alloc_failed = false;
1112 	int i, index;
1113 
1114 	for (i = 0; i < count && !alloc_failed; i++) {
1115 		index = (rx->write + i) & TSNEP_RING_MASK;
1116 
1117 		if (unlikely(tsnep_rx_alloc_buffer(rx, index))) {
1118 			rx->alloc_failed++;
1119 			alloc_failed = true;
1120 
1121 			/* reuse only if no other allocation was successful */
1122 			if (i == 0 && reuse)
1123 				tsnep_rx_reuse_buffer(rx, index);
1124 			else
1125 				break;
1126 		}
1127 
1128 		tsnep_rx_activate(rx, index);
1129 	}
1130 
1131 	if (i)
1132 		rx->write = (rx->write + i) & TSNEP_RING_MASK;
1133 
1134 	return i;
1135 }
1136 
1137 static int tsnep_rx_refill(struct tsnep_rx *rx, int count, bool reuse)
1138 {
1139 	int desc_refilled;
1140 
1141 	desc_refilled = tsnep_rx_alloc(rx, count, reuse);
1142 	if (desc_refilled)
1143 		tsnep_rx_enable(rx);
1144 
1145 	return desc_refilled;
1146 }
1147 
1148 static void tsnep_rx_set_xdp(struct tsnep_rx *rx, struct tsnep_rx_entry *entry,
1149 			     struct xdp_buff *xdp)
1150 {
1151 	entry->xdp = xdp;
1152 	entry->len = TSNEP_XSK_RX_BUF_SIZE;
1153 	entry->dma = xsk_buff_xdp_get_dma(entry->xdp);
1154 	entry->desc->rx = __cpu_to_le64(entry->dma);
1155 }
1156 
1157 static void tsnep_rx_reuse_buffer_zc(struct tsnep_rx *rx, int index)
1158 {
1159 	struct tsnep_rx_entry *entry = &rx->entry[index];
1160 	struct tsnep_rx_entry *read = &rx->entry[rx->read];
1161 
1162 	tsnep_rx_set_xdp(rx, entry, read->xdp);
1163 	read->xdp = NULL;
1164 }
1165 
1166 static int tsnep_rx_alloc_zc(struct tsnep_rx *rx, int count, bool reuse)
1167 {
1168 	u32 allocated;
1169 	int i;
1170 
1171 	allocated = xsk_buff_alloc_batch(rx->xsk_pool, rx->xdp_batch, count);
1172 	for (i = 0; i < allocated; i++) {
1173 		int index = (rx->write + i) & TSNEP_RING_MASK;
1174 		struct tsnep_rx_entry *entry = &rx->entry[index];
1175 
1176 		tsnep_rx_set_xdp(rx, entry, rx->xdp_batch[i]);
1177 		tsnep_rx_activate(rx, index);
1178 	}
1179 	if (i == 0) {
1180 		rx->alloc_failed++;
1181 
1182 		if (reuse) {
1183 			tsnep_rx_reuse_buffer_zc(rx, rx->write);
1184 			tsnep_rx_activate(rx, rx->write);
1185 		}
1186 	}
1187 
1188 	if (i)
1189 		rx->write = (rx->write + i) & TSNEP_RING_MASK;
1190 
1191 	return i;
1192 }
1193 
1194 static void tsnep_rx_free_zc(struct tsnep_rx *rx)
1195 {
1196 	int i;
1197 
1198 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
1199 		struct tsnep_rx_entry *entry = &rx->entry[i];
1200 
1201 		if (entry->xdp)
1202 			xsk_buff_free(entry->xdp);
1203 		entry->xdp = NULL;
1204 	}
1205 }
1206 
1207 static int tsnep_rx_refill_zc(struct tsnep_rx *rx, int count, bool reuse)
1208 {
1209 	int desc_refilled;
1210 
1211 	desc_refilled = tsnep_rx_alloc_zc(rx, count, reuse);
1212 	if (desc_refilled)
1213 		tsnep_rx_enable(rx);
1214 
1215 	return desc_refilled;
1216 }
1217 
1218 static bool tsnep_xdp_run_prog(struct tsnep_rx *rx, struct bpf_prog *prog,
1219 			       struct xdp_buff *xdp, int *status,
1220 			       struct netdev_queue *tx_nq, struct tsnep_tx *tx)
1221 {
1222 	unsigned int length;
1223 	unsigned int sync;
1224 	u32 act;
1225 
1226 	length = xdp->data_end - xdp->data_hard_start - XDP_PACKET_HEADROOM;
1227 
1228 	act = bpf_prog_run_xdp(prog, xdp);
1229 	switch (act) {
1230 	case XDP_PASS:
1231 		return false;
1232 	case XDP_TX:
1233 		if (!tsnep_xdp_xmit_back(rx->adapter, xdp, tx_nq, tx, false))
1234 			goto out_failure;
1235 		*status |= TSNEP_XDP_TX;
1236 		return true;
1237 	case XDP_REDIRECT:
1238 		if (xdp_do_redirect(rx->adapter->netdev, xdp, prog) < 0)
1239 			goto out_failure;
1240 		*status |= TSNEP_XDP_REDIRECT;
1241 		return true;
1242 	default:
1243 		bpf_warn_invalid_xdp_action(rx->adapter->netdev, prog, act);
1244 		fallthrough;
1245 	case XDP_ABORTED:
1246 out_failure:
1247 		trace_xdp_exception(rx->adapter->netdev, prog, act);
1248 		fallthrough;
1249 	case XDP_DROP:
1250 		/* Due xdp_adjust_tail: DMA sync for_device cover max len CPU
1251 		 * touch
1252 		 */
1253 		sync = xdp->data_end - xdp->data_hard_start -
1254 		       XDP_PACKET_HEADROOM;
1255 		sync = max(sync, length);
1256 		page_pool_put_page(rx->page_pool, virt_to_head_page(xdp->data),
1257 				   sync, true);
1258 		return true;
1259 	}
1260 }
1261 
1262 static bool tsnep_xdp_run_prog_zc(struct tsnep_rx *rx, struct bpf_prog *prog,
1263 				  struct xdp_buff *xdp, int *status,
1264 				  struct netdev_queue *tx_nq,
1265 				  struct tsnep_tx *tx)
1266 {
1267 	u32 act;
1268 
1269 	act = bpf_prog_run_xdp(prog, xdp);
1270 
1271 	/* XDP_REDIRECT is the main action for zero-copy */
1272 	if (likely(act == XDP_REDIRECT)) {
1273 		if (xdp_do_redirect(rx->adapter->netdev, xdp, prog) < 0)
1274 			goto out_failure;
1275 		*status |= TSNEP_XDP_REDIRECT;
1276 		return true;
1277 	}
1278 
1279 	switch (act) {
1280 	case XDP_PASS:
1281 		return false;
1282 	case XDP_TX:
1283 		if (!tsnep_xdp_xmit_back(rx->adapter, xdp, tx_nq, tx, true))
1284 			goto out_failure;
1285 		*status |= TSNEP_XDP_TX;
1286 		return true;
1287 	default:
1288 		bpf_warn_invalid_xdp_action(rx->adapter->netdev, prog, act);
1289 		fallthrough;
1290 	case XDP_ABORTED:
1291 out_failure:
1292 		trace_xdp_exception(rx->adapter->netdev, prog, act);
1293 		fallthrough;
1294 	case XDP_DROP:
1295 		xsk_buff_free(xdp);
1296 		return true;
1297 	}
1298 }
1299 
1300 static void tsnep_finalize_xdp(struct tsnep_adapter *adapter, int status,
1301 			       struct netdev_queue *tx_nq, struct tsnep_tx *tx)
1302 {
1303 	if (status & TSNEP_XDP_TX) {
1304 		__netif_tx_lock(tx_nq, smp_processor_id());
1305 		tsnep_xdp_xmit_flush(tx);
1306 		__netif_tx_unlock(tx_nq);
1307 	}
1308 
1309 	if (status & TSNEP_XDP_REDIRECT)
1310 		xdp_do_flush();
1311 }
1312 
1313 static struct sk_buff *tsnep_build_skb(struct tsnep_rx *rx, struct page *page,
1314 				       int length)
1315 {
1316 	struct sk_buff *skb;
1317 
1318 	skb = napi_build_skb(page_address(page), PAGE_SIZE);
1319 	if (unlikely(!skb))
1320 		return NULL;
1321 
1322 	/* update pointers within the skb to store the data */
1323 	skb_reserve(skb, TSNEP_RX_OFFSET + TSNEP_RX_INLINE_METADATA_SIZE);
1324 	__skb_put(skb, length - ETH_FCS_LEN);
1325 
1326 	if (rx->adapter->hwtstamp_config.rx_filter == HWTSTAMP_FILTER_ALL) {
1327 		struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
1328 		struct tsnep_rx_inline *rx_inline =
1329 			(struct tsnep_rx_inline *)(page_address(page) +
1330 						   TSNEP_RX_OFFSET);
1331 
1332 		skb_shinfo(skb)->tx_flags |=
1333 			SKBTX_HW_TSTAMP_NETDEV;
1334 		memset(hwtstamps, 0, sizeof(*hwtstamps));
1335 		hwtstamps->netdev_data = rx_inline;
1336 	}
1337 
1338 	skb_record_rx_queue(skb, rx->queue_index);
1339 	skb->protocol = eth_type_trans(skb, rx->adapter->netdev);
1340 
1341 	return skb;
1342 }
1343 
1344 static void tsnep_rx_page(struct tsnep_rx *rx, struct napi_struct *napi,
1345 			  struct page *page, int length)
1346 {
1347 	struct sk_buff *skb;
1348 
1349 	skb = tsnep_build_skb(rx, page, length);
1350 	if (skb) {
1351 		skb_mark_for_recycle(skb);
1352 
1353 		rx->packets++;
1354 		rx->bytes += length;
1355 		if (skb->pkt_type == PACKET_MULTICAST)
1356 			rx->multicast++;
1357 
1358 		napi_gro_receive(napi, skb);
1359 	} else {
1360 		page_pool_recycle_direct(rx->page_pool, page);
1361 
1362 		rx->dropped++;
1363 	}
1364 }
1365 
1366 static int tsnep_rx_poll(struct tsnep_rx *rx, struct napi_struct *napi,
1367 			 int budget)
1368 {
1369 	struct device *dmadev = rx->adapter->dmadev;
1370 	enum dma_data_direction dma_dir;
1371 	struct tsnep_rx_entry *entry;
1372 	struct netdev_queue *tx_nq;
1373 	struct bpf_prog *prog;
1374 	struct xdp_buff xdp;
1375 	struct tsnep_tx *tx;
1376 	int desc_available;
1377 	int xdp_status = 0;
1378 	int done = 0;
1379 	int length;
1380 
1381 	desc_available = tsnep_rx_desc_available(rx);
1382 	dma_dir = page_pool_get_dma_dir(rx->page_pool);
1383 	prog = READ_ONCE(rx->adapter->xdp_prog);
1384 	if (prog) {
1385 		tx_nq = netdev_get_tx_queue(rx->adapter->netdev,
1386 					    rx->tx_queue_index);
1387 		tx = &rx->adapter->tx[rx->tx_queue_index];
1388 
1389 		xdp_init_buff(&xdp, PAGE_SIZE, &rx->xdp_rxq);
1390 	}
1391 
1392 	while (likely(done < budget) && (rx->read != rx->write)) {
1393 		entry = &rx->entry[rx->read];
1394 		if ((__le32_to_cpu(entry->desc_wb->properties) &
1395 		     TSNEP_DESC_OWNER_COUNTER_MASK) !=
1396 		    (entry->properties & TSNEP_DESC_OWNER_COUNTER_MASK))
1397 			break;
1398 		done++;
1399 
1400 		if (desc_available >= TSNEP_RING_RX_REFILL) {
1401 			bool reuse = desc_available >= TSNEP_RING_RX_REUSE;
1402 
1403 			desc_available -= tsnep_rx_refill(rx, desc_available,
1404 							  reuse);
1405 			if (!entry->page) {
1406 				/* buffer has been reused for refill to prevent
1407 				 * empty RX ring, thus buffer cannot be used for
1408 				 * RX processing
1409 				 */
1410 				rx->read = (rx->read + 1) & TSNEP_RING_MASK;
1411 				desc_available++;
1412 
1413 				rx->dropped++;
1414 
1415 				continue;
1416 			}
1417 		}
1418 
1419 		/* descriptor properties shall be read first, because valid data
1420 		 * is signaled there
1421 		 */
1422 		dma_rmb();
1423 
1424 		prefetch(page_address(entry->page) + TSNEP_RX_OFFSET);
1425 		length = __le32_to_cpu(entry->desc_wb->properties) &
1426 			 TSNEP_DESC_LENGTH_MASK;
1427 		dma_sync_single_range_for_cpu(dmadev, entry->dma,
1428 					      TSNEP_RX_OFFSET, length, dma_dir);
1429 
1430 		/* RX metadata with timestamps is in front of actual data,
1431 		 * subtract metadata size to get length of actual data and
1432 		 * consider metadata size as offset of actual data during RX
1433 		 * processing
1434 		 */
1435 		length -= TSNEP_RX_INLINE_METADATA_SIZE;
1436 
1437 		rx->read = (rx->read + 1) & TSNEP_RING_MASK;
1438 		desc_available++;
1439 
1440 		if (prog) {
1441 			bool consume;
1442 
1443 			xdp_prepare_buff(&xdp, page_address(entry->page),
1444 					 XDP_PACKET_HEADROOM + TSNEP_RX_INLINE_METADATA_SIZE,
1445 					 length - ETH_FCS_LEN, false);
1446 
1447 			consume = tsnep_xdp_run_prog(rx, prog, &xdp,
1448 						     &xdp_status, tx_nq, tx);
1449 			if (consume) {
1450 				rx->packets++;
1451 				rx->bytes += length;
1452 
1453 				entry->page = NULL;
1454 
1455 				continue;
1456 			}
1457 		}
1458 
1459 		tsnep_rx_page(rx, napi, entry->page, length);
1460 		entry->page = NULL;
1461 	}
1462 
1463 	if (xdp_status)
1464 		tsnep_finalize_xdp(rx->adapter, xdp_status, tx_nq, tx);
1465 
1466 	if (desc_available)
1467 		tsnep_rx_refill(rx, desc_available, false);
1468 
1469 	return done;
1470 }
1471 
1472 static int tsnep_rx_poll_zc(struct tsnep_rx *rx, struct napi_struct *napi,
1473 			    int budget)
1474 {
1475 	struct tsnep_rx_entry *entry;
1476 	struct netdev_queue *tx_nq;
1477 	struct bpf_prog *prog;
1478 	struct tsnep_tx *tx;
1479 	int desc_available;
1480 	int xdp_status = 0;
1481 	struct page *page;
1482 	int done = 0;
1483 	int length;
1484 
1485 	desc_available = tsnep_rx_desc_available(rx);
1486 	prog = READ_ONCE(rx->adapter->xdp_prog);
1487 	if (prog) {
1488 		tx_nq = netdev_get_tx_queue(rx->adapter->netdev,
1489 					    rx->tx_queue_index);
1490 		tx = &rx->adapter->tx[rx->tx_queue_index];
1491 	}
1492 
1493 	while (likely(done < budget) && (rx->read != rx->write)) {
1494 		entry = &rx->entry[rx->read];
1495 		if ((__le32_to_cpu(entry->desc_wb->properties) &
1496 		     TSNEP_DESC_OWNER_COUNTER_MASK) !=
1497 		    (entry->properties & TSNEP_DESC_OWNER_COUNTER_MASK))
1498 			break;
1499 		done++;
1500 
1501 		if (desc_available >= TSNEP_RING_RX_REFILL) {
1502 			bool reuse = desc_available >= TSNEP_RING_RX_REUSE;
1503 
1504 			desc_available -= tsnep_rx_refill_zc(rx, desc_available,
1505 							     reuse);
1506 			if (!entry->xdp) {
1507 				/* buffer has been reused for refill to prevent
1508 				 * empty RX ring, thus buffer cannot be used for
1509 				 * RX processing
1510 				 */
1511 				rx->read = (rx->read + 1) & TSNEP_RING_MASK;
1512 				desc_available++;
1513 
1514 				rx->dropped++;
1515 
1516 				continue;
1517 			}
1518 		}
1519 
1520 		/* descriptor properties shall be read first, because valid data
1521 		 * is signaled there
1522 		 */
1523 		dma_rmb();
1524 
1525 		prefetch(entry->xdp->data);
1526 		length = __le32_to_cpu(entry->desc_wb->properties) &
1527 			 TSNEP_DESC_LENGTH_MASK;
1528 		xsk_buff_set_size(entry->xdp, length - ETH_FCS_LEN);
1529 		xsk_buff_dma_sync_for_cpu(entry->xdp, rx->xsk_pool);
1530 
1531 		/* RX metadata with timestamps is in front of actual data,
1532 		 * subtract metadata size to get length of actual data and
1533 		 * consider metadata size as offset of actual data during RX
1534 		 * processing
1535 		 */
1536 		length -= TSNEP_RX_INLINE_METADATA_SIZE;
1537 
1538 		rx->read = (rx->read + 1) & TSNEP_RING_MASK;
1539 		desc_available++;
1540 
1541 		if (prog) {
1542 			bool consume;
1543 
1544 			entry->xdp->data += TSNEP_RX_INLINE_METADATA_SIZE;
1545 			entry->xdp->data_meta += TSNEP_RX_INLINE_METADATA_SIZE;
1546 
1547 			consume = tsnep_xdp_run_prog_zc(rx, prog, entry->xdp,
1548 							&xdp_status, tx_nq, tx);
1549 			if (consume) {
1550 				rx->packets++;
1551 				rx->bytes += length;
1552 
1553 				entry->xdp = NULL;
1554 
1555 				continue;
1556 			}
1557 		}
1558 
1559 		page = page_pool_dev_alloc_pages(rx->page_pool);
1560 		if (page) {
1561 			memcpy(page_address(page) + TSNEP_RX_OFFSET,
1562 			       entry->xdp->data - TSNEP_RX_INLINE_METADATA_SIZE,
1563 			       length + TSNEP_RX_INLINE_METADATA_SIZE);
1564 			tsnep_rx_page(rx, napi, page, length);
1565 		} else {
1566 			rx->dropped++;
1567 		}
1568 		xsk_buff_free(entry->xdp);
1569 		entry->xdp = NULL;
1570 	}
1571 
1572 	if (xdp_status)
1573 		tsnep_finalize_xdp(rx->adapter, xdp_status, tx_nq, tx);
1574 
1575 	if (desc_available)
1576 		desc_available -= tsnep_rx_refill_zc(rx, desc_available, false);
1577 
1578 	if (xsk_uses_need_wakeup(rx->xsk_pool)) {
1579 		if (desc_available)
1580 			xsk_set_rx_need_wakeup(rx->xsk_pool);
1581 		else
1582 			xsk_clear_rx_need_wakeup(rx->xsk_pool);
1583 
1584 		return done;
1585 	}
1586 
1587 	return desc_available ? budget : done;
1588 }
1589 
1590 static bool tsnep_rx_pending(struct tsnep_rx *rx)
1591 {
1592 	struct tsnep_rx_entry *entry;
1593 
1594 	if (rx->read != rx->write) {
1595 		entry = &rx->entry[rx->read];
1596 		if ((__le32_to_cpu(entry->desc_wb->properties) &
1597 		     TSNEP_DESC_OWNER_COUNTER_MASK) ==
1598 		    (entry->properties & TSNEP_DESC_OWNER_COUNTER_MASK))
1599 			return true;
1600 	}
1601 
1602 	return false;
1603 }
1604 
1605 static int tsnep_rx_open(struct tsnep_rx *rx)
1606 {
1607 	int desc_available;
1608 	int retval;
1609 
1610 	retval = tsnep_rx_ring_create(rx);
1611 	if (retval)
1612 		return retval;
1613 
1614 	tsnep_rx_init(rx);
1615 
1616 	desc_available = tsnep_rx_desc_available(rx);
1617 	if (rx->xsk_pool)
1618 		retval = tsnep_rx_alloc_zc(rx, desc_available, false);
1619 	else
1620 		retval = tsnep_rx_alloc(rx, desc_available, false);
1621 	if (retval != desc_available) {
1622 		retval = -ENOMEM;
1623 
1624 		goto alloc_failed;
1625 	}
1626 
1627 	/* prealloc pages to prevent allocation failures when XSK pool is
1628 	 * disabled at runtime
1629 	 */
1630 	if (rx->xsk_pool) {
1631 		retval = tsnep_rx_alloc_page_buffer(rx);
1632 		if (retval)
1633 			goto alloc_failed;
1634 	}
1635 
1636 	return 0;
1637 
1638 alloc_failed:
1639 	tsnep_rx_ring_cleanup(rx);
1640 	return retval;
1641 }
1642 
1643 static void tsnep_rx_close(struct tsnep_rx *rx)
1644 {
1645 	if (rx->xsk_pool)
1646 		tsnep_rx_free_page_buffer(rx);
1647 
1648 	tsnep_rx_ring_cleanup(rx);
1649 }
1650 
1651 static void tsnep_rx_reopen(struct tsnep_rx *rx)
1652 {
1653 	struct page **page = rx->page_buffer;
1654 	int i;
1655 
1656 	tsnep_rx_init(rx);
1657 
1658 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
1659 		struct tsnep_rx_entry *entry = &rx->entry[i];
1660 
1661 		/* defined initial values for properties are required for
1662 		 * correct owner counter checking
1663 		 */
1664 		entry->desc->properties = 0;
1665 		entry->desc_wb->properties = 0;
1666 
1667 		/* prevent allocation failures by reusing kept pages */
1668 		if (*page) {
1669 			tsnep_rx_set_page(rx, entry, *page);
1670 			tsnep_rx_activate(rx, rx->write);
1671 			rx->write++;
1672 
1673 			*page = NULL;
1674 			page++;
1675 		}
1676 	}
1677 }
1678 
1679 static void tsnep_rx_reopen_xsk(struct tsnep_rx *rx)
1680 {
1681 	struct page **page = rx->page_buffer;
1682 	u32 allocated;
1683 	int i;
1684 
1685 	tsnep_rx_init(rx);
1686 
1687 	/* alloc all ring entries except the last one, because ring cannot be
1688 	 * filled completely, as many buffers as possible is enough as wakeup is
1689 	 * done if new buffers are available
1690 	 */
1691 	allocated = xsk_buff_alloc_batch(rx->xsk_pool, rx->xdp_batch,
1692 					 TSNEP_RING_SIZE - 1);
1693 
1694 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
1695 		struct tsnep_rx_entry *entry = &rx->entry[i];
1696 
1697 		/* keep pages to prevent allocation failures when xsk is
1698 		 * disabled
1699 		 */
1700 		if (entry->page) {
1701 			*page = entry->page;
1702 			entry->page = NULL;
1703 
1704 			page++;
1705 		}
1706 
1707 		/* defined initial values for properties are required for
1708 		 * correct owner counter checking
1709 		 */
1710 		entry->desc->properties = 0;
1711 		entry->desc_wb->properties = 0;
1712 
1713 		if (allocated) {
1714 			tsnep_rx_set_xdp(rx, entry,
1715 					 rx->xdp_batch[allocated - 1]);
1716 			tsnep_rx_activate(rx, rx->write);
1717 			rx->write++;
1718 
1719 			allocated--;
1720 		}
1721 	}
1722 
1723 	/* set need wakeup flag immediately if ring is not filled completely,
1724 	 * first polling would be too late as need wakeup signalisation would
1725 	 * be delayed for an indefinite time
1726 	 */
1727 	if (xsk_uses_need_wakeup(rx->xsk_pool)) {
1728 		int desc_available = tsnep_rx_desc_available(rx);
1729 
1730 		if (desc_available)
1731 			xsk_set_rx_need_wakeup(rx->xsk_pool);
1732 		else
1733 			xsk_clear_rx_need_wakeup(rx->xsk_pool);
1734 	}
1735 }
1736 
1737 static bool tsnep_pending(struct tsnep_queue *queue)
1738 {
1739 	if (queue->tx && tsnep_tx_pending(queue->tx))
1740 		return true;
1741 
1742 	if (queue->rx && tsnep_rx_pending(queue->rx))
1743 		return true;
1744 
1745 	return false;
1746 }
1747 
1748 static int tsnep_poll(struct napi_struct *napi, int budget)
1749 {
1750 	struct tsnep_queue *queue = container_of(napi, struct tsnep_queue,
1751 						 napi);
1752 	bool complete = true;
1753 	int done = 0;
1754 
1755 	if (queue->tx)
1756 		complete = tsnep_tx_poll(queue->tx, budget);
1757 
1758 	/* handle case where we are called by netpoll with a budget of 0 */
1759 	if (unlikely(budget <= 0))
1760 		return budget;
1761 
1762 	if (queue->rx) {
1763 		done = queue->rx->xsk_pool ?
1764 		       tsnep_rx_poll_zc(queue->rx, napi, budget) :
1765 		       tsnep_rx_poll(queue->rx, napi, budget);
1766 		if (done >= budget)
1767 			complete = false;
1768 	}
1769 
1770 	/* if all work not completed, return budget and keep polling */
1771 	if (!complete)
1772 		return budget;
1773 
1774 	if (likely(napi_complete_done(napi, done))) {
1775 		tsnep_enable_irq(queue->adapter, queue->irq_mask);
1776 
1777 		/* reschedule if work is already pending, prevent rotten packets
1778 		 * which are transmitted or received after polling but before
1779 		 * interrupt enable
1780 		 */
1781 		if (tsnep_pending(queue)) {
1782 			tsnep_disable_irq(queue->adapter, queue->irq_mask);
1783 			napi_schedule(napi);
1784 		}
1785 	}
1786 
1787 	return min(done, budget - 1);
1788 }
1789 
1790 static int tsnep_request_irq(struct tsnep_queue *queue, bool first)
1791 {
1792 	const char *name = netdev_name(queue->adapter->netdev);
1793 	irq_handler_t handler;
1794 	void *dev;
1795 	int retval;
1796 
1797 	if (first) {
1798 		sprintf(queue->name, "%s-mac", name);
1799 		handler = tsnep_irq;
1800 		dev = queue->adapter;
1801 	} else {
1802 		if (queue->tx && queue->rx)
1803 			snprintf(queue->name, sizeof(queue->name), "%s-txrx-%d",
1804 				 name, queue->rx->queue_index);
1805 		else if (queue->tx)
1806 			snprintf(queue->name, sizeof(queue->name), "%s-tx-%d",
1807 				 name, queue->tx->queue_index);
1808 		else
1809 			snprintf(queue->name, sizeof(queue->name), "%s-rx-%d",
1810 				 name, queue->rx->queue_index);
1811 		handler = tsnep_irq_txrx;
1812 		dev = queue;
1813 	}
1814 
1815 	retval = request_irq(queue->irq, handler, 0, queue->name, dev);
1816 	if (retval) {
1817 		/* if name is empty, then interrupt won't be freed */
1818 		memset(queue->name, 0, sizeof(queue->name));
1819 	}
1820 
1821 	return retval;
1822 }
1823 
1824 static void tsnep_free_irq(struct tsnep_queue *queue, bool first)
1825 {
1826 	void *dev;
1827 
1828 	if (!strlen(queue->name))
1829 		return;
1830 
1831 	if (first)
1832 		dev = queue->adapter;
1833 	else
1834 		dev = queue;
1835 
1836 	free_irq(queue->irq, dev);
1837 	memset(queue->name, 0, sizeof(queue->name));
1838 }
1839 
1840 static void tsnep_queue_close(struct tsnep_queue *queue, bool first)
1841 {
1842 	struct tsnep_rx *rx = queue->rx;
1843 
1844 	tsnep_free_irq(queue, first);
1845 
1846 	if (rx) {
1847 		if (xdp_rxq_info_is_reg(&rx->xdp_rxq))
1848 			xdp_rxq_info_unreg(&rx->xdp_rxq);
1849 		if (xdp_rxq_info_is_reg(&rx->xdp_rxq_zc))
1850 			xdp_rxq_info_unreg(&rx->xdp_rxq_zc);
1851 	}
1852 
1853 	netif_napi_del(&queue->napi);
1854 }
1855 
1856 static int tsnep_queue_open(struct tsnep_adapter *adapter,
1857 			    struct tsnep_queue *queue, bool first)
1858 {
1859 	struct tsnep_rx *rx = queue->rx;
1860 	struct tsnep_tx *tx = queue->tx;
1861 	int retval;
1862 
1863 	netif_napi_add(adapter->netdev, &queue->napi, tsnep_poll);
1864 
1865 	if (rx) {
1866 		/* choose TX queue for XDP_TX */
1867 		if (tx)
1868 			rx->tx_queue_index = tx->queue_index;
1869 		else if (rx->queue_index < adapter->num_tx_queues)
1870 			rx->tx_queue_index = rx->queue_index;
1871 		else
1872 			rx->tx_queue_index = 0;
1873 
1874 		/* prepare both memory models to eliminate possible registration
1875 		 * errors when memory model is switched between page pool and
1876 		 * XSK pool during runtime
1877 		 */
1878 		retval = xdp_rxq_info_reg(&rx->xdp_rxq, adapter->netdev,
1879 					  rx->queue_index, queue->napi.napi_id);
1880 		if (retval)
1881 			goto failed;
1882 		retval = xdp_rxq_info_reg_mem_model(&rx->xdp_rxq,
1883 						    MEM_TYPE_PAGE_POOL,
1884 						    rx->page_pool);
1885 		if (retval)
1886 			goto failed;
1887 		retval = xdp_rxq_info_reg(&rx->xdp_rxq_zc, adapter->netdev,
1888 					  rx->queue_index, queue->napi.napi_id);
1889 		if (retval)
1890 			goto failed;
1891 		retval = xdp_rxq_info_reg_mem_model(&rx->xdp_rxq_zc,
1892 						    MEM_TYPE_XSK_BUFF_POOL,
1893 						    NULL);
1894 		if (retval)
1895 			goto failed;
1896 		if (rx->xsk_pool)
1897 			xsk_pool_set_rxq_info(rx->xsk_pool, &rx->xdp_rxq_zc);
1898 	}
1899 
1900 	retval = tsnep_request_irq(queue, first);
1901 	if (retval) {
1902 		netif_err(adapter, drv, adapter->netdev,
1903 			  "can't get assigned irq %d.\n", queue->irq);
1904 		goto failed;
1905 	}
1906 
1907 	return 0;
1908 
1909 failed:
1910 	tsnep_queue_close(queue, first);
1911 
1912 	return retval;
1913 }
1914 
1915 static void tsnep_queue_enable(struct tsnep_queue *queue)
1916 {
1917 	napi_enable(&queue->napi);
1918 	tsnep_enable_irq(queue->adapter, queue->irq_mask);
1919 
1920 	if (queue->tx)
1921 		tsnep_tx_enable(queue->tx);
1922 
1923 	if (queue->rx)
1924 		tsnep_rx_enable(queue->rx);
1925 }
1926 
1927 static void tsnep_queue_disable(struct tsnep_queue *queue)
1928 {
1929 	if (queue->tx)
1930 		tsnep_tx_disable(queue->tx, &queue->napi);
1931 
1932 	napi_disable(&queue->napi);
1933 	tsnep_disable_irq(queue->adapter, queue->irq_mask);
1934 
1935 	/* disable RX after NAPI polling has been disabled, because RX can be
1936 	 * enabled during NAPI polling
1937 	 */
1938 	if (queue->rx)
1939 		tsnep_rx_disable(queue->rx);
1940 }
1941 
1942 static int tsnep_netdev_open(struct net_device *netdev)
1943 {
1944 	struct tsnep_adapter *adapter = netdev_priv(netdev);
1945 	int i, retval;
1946 
1947 	for (i = 0; i < adapter->num_queues; i++) {
1948 		if (adapter->queue[i].tx) {
1949 			retval = tsnep_tx_open(adapter->queue[i].tx);
1950 			if (retval)
1951 				goto failed;
1952 		}
1953 		if (adapter->queue[i].rx) {
1954 			retval = tsnep_rx_open(adapter->queue[i].rx);
1955 			if (retval)
1956 				goto failed;
1957 		}
1958 
1959 		retval = tsnep_queue_open(adapter, &adapter->queue[i], i == 0);
1960 		if (retval)
1961 			goto failed;
1962 	}
1963 
1964 	retval = netif_set_real_num_tx_queues(adapter->netdev,
1965 					      adapter->num_tx_queues);
1966 	if (retval)
1967 		goto failed;
1968 	retval = netif_set_real_num_rx_queues(adapter->netdev,
1969 					      adapter->num_rx_queues);
1970 	if (retval)
1971 		goto failed;
1972 
1973 	tsnep_enable_irq(adapter, ECM_INT_LINK);
1974 	retval = tsnep_phy_open(adapter);
1975 	if (retval)
1976 		goto phy_failed;
1977 
1978 	for (i = 0; i < adapter->num_queues; i++)
1979 		tsnep_queue_enable(&adapter->queue[i]);
1980 
1981 	return 0;
1982 
1983 phy_failed:
1984 	tsnep_disable_irq(adapter, ECM_INT_LINK);
1985 failed:
1986 	for (i = 0; i < adapter->num_queues; i++) {
1987 		tsnep_queue_close(&adapter->queue[i], i == 0);
1988 
1989 		if (adapter->queue[i].rx)
1990 			tsnep_rx_close(adapter->queue[i].rx);
1991 		if (adapter->queue[i].tx)
1992 			tsnep_tx_close(adapter->queue[i].tx);
1993 	}
1994 	return retval;
1995 }
1996 
1997 static int tsnep_netdev_close(struct net_device *netdev)
1998 {
1999 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2000 	int i;
2001 
2002 	tsnep_disable_irq(adapter, ECM_INT_LINK);
2003 	tsnep_phy_close(adapter);
2004 
2005 	for (i = 0; i < adapter->num_queues; i++) {
2006 		tsnep_queue_disable(&adapter->queue[i]);
2007 
2008 		tsnep_queue_close(&adapter->queue[i], i == 0);
2009 
2010 		if (adapter->queue[i].rx)
2011 			tsnep_rx_close(adapter->queue[i].rx);
2012 		if (adapter->queue[i].tx)
2013 			tsnep_tx_close(adapter->queue[i].tx);
2014 	}
2015 
2016 	return 0;
2017 }
2018 
2019 int tsnep_enable_xsk(struct tsnep_queue *queue, struct xsk_buff_pool *pool)
2020 {
2021 	bool running = netif_running(queue->adapter->netdev);
2022 	u32 frame_size;
2023 
2024 	frame_size = xsk_pool_get_rx_frame_size(pool);
2025 	if (frame_size < TSNEP_XSK_RX_BUF_SIZE)
2026 		return -EOPNOTSUPP;
2027 
2028 	queue->rx->page_buffer = kcalloc(TSNEP_RING_SIZE,
2029 					 sizeof(*queue->rx->page_buffer),
2030 					 GFP_KERNEL);
2031 	if (!queue->rx->page_buffer)
2032 		return -ENOMEM;
2033 	queue->rx->xdp_batch = kcalloc(TSNEP_RING_SIZE,
2034 				       sizeof(*queue->rx->xdp_batch),
2035 				       GFP_KERNEL);
2036 	if (!queue->rx->xdp_batch) {
2037 		kfree(queue->rx->page_buffer);
2038 		queue->rx->page_buffer = NULL;
2039 
2040 		return -ENOMEM;
2041 	}
2042 
2043 	xsk_pool_set_rxq_info(pool, &queue->rx->xdp_rxq_zc);
2044 
2045 	if (running)
2046 		tsnep_queue_disable(queue);
2047 
2048 	queue->tx->xsk_pool = pool;
2049 	queue->rx->xsk_pool = pool;
2050 
2051 	if (running) {
2052 		tsnep_rx_reopen_xsk(queue->rx);
2053 		tsnep_queue_enable(queue);
2054 	}
2055 
2056 	return 0;
2057 }
2058 
2059 void tsnep_disable_xsk(struct tsnep_queue *queue)
2060 {
2061 	bool running = netif_running(queue->adapter->netdev);
2062 
2063 	if (running)
2064 		tsnep_queue_disable(queue);
2065 
2066 	tsnep_rx_free_zc(queue->rx);
2067 
2068 	queue->rx->xsk_pool = NULL;
2069 	queue->tx->xsk_pool = NULL;
2070 
2071 	if (running) {
2072 		tsnep_rx_reopen(queue->rx);
2073 		tsnep_queue_enable(queue);
2074 	}
2075 
2076 	kfree(queue->rx->xdp_batch);
2077 	queue->rx->xdp_batch = NULL;
2078 	kfree(queue->rx->page_buffer);
2079 	queue->rx->page_buffer = NULL;
2080 }
2081 
2082 static netdev_tx_t tsnep_netdev_xmit_frame(struct sk_buff *skb,
2083 					   struct net_device *netdev)
2084 {
2085 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2086 	u16 queue_mapping = skb_get_queue_mapping(skb);
2087 
2088 	if (queue_mapping >= adapter->num_tx_queues)
2089 		queue_mapping = 0;
2090 
2091 	return tsnep_xmit_frame_ring(skb, &adapter->tx[queue_mapping]);
2092 }
2093 
2094 static int tsnep_netdev_ioctl(struct net_device *netdev, struct ifreq *ifr,
2095 			      int cmd)
2096 {
2097 	if (!netif_running(netdev))
2098 		return -EINVAL;
2099 	if (cmd == SIOCSHWTSTAMP || cmd == SIOCGHWTSTAMP)
2100 		return tsnep_ptp_ioctl(netdev, ifr, cmd);
2101 	return phy_mii_ioctl(netdev->phydev, ifr, cmd);
2102 }
2103 
2104 static void tsnep_netdev_set_multicast(struct net_device *netdev)
2105 {
2106 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2107 
2108 	u16 rx_filter = 0;
2109 
2110 	/* configured MAC address and broadcasts are never filtered */
2111 	if (netdev->flags & IFF_PROMISC) {
2112 		rx_filter |= TSNEP_RX_FILTER_ACCEPT_ALL_MULTICASTS;
2113 		rx_filter |= TSNEP_RX_FILTER_ACCEPT_ALL_UNICASTS;
2114 	} else if (!netdev_mc_empty(netdev) || (netdev->flags & IFF_ALLMULTI)) {
2115 		rx_filter |= TSNEP_RX_FILTER_ACCEPT_ALL_MULTICASTS;
2116 	}
2117 	iowrite16(rx_filter, adapter->addr + TSNEP_RX_FILTER);
2118 }
2119 
2120 static void tsnep_netdev_get_stats64(struct net_device *netdev,
2121 				     struct rtnl_link_stats64 *stats)
2122 {
2123 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2124 	u32 reg;
2125 	u32 val;
2126 	int i;
2127 
2128 	for (i = 0; i < adapter->num_tx_queues; i++) {
2129 		stats->tx_packets += adapter->tx[i].packets;
2130 		stats->tx_bytes += adapter->tx[i].bytes;
2131 		stats->tx_dropped += adapter->tx[i].dropped;
2132 	}
2133 	for (i = 0; i < adapter->num_rx_queues; i++) {
2134 		stats->rx_packets += adapter->rx[i].packets;
2135 		stats->rx_bytes += adapter->rx[i].bytes;
2136 		stats->rx_dropped += adapter->rx[i].dropped;
2137 		stats->multicast += adapter->rx[i].multicast;
2138 
2139 		reg = ioread32(adapter->addr + TSNEP_QUEUE(i) +
2140 			       TSNEP_RX_STATISTIC);
2141 		val = (reg & TSNEP_RX_STATISTIC_NO_DESC_MASK) >>
2142 		      TSNEP_RX_STATISTIC_NO_DESC_SHIFT;
2143 		stats->rx_dropped += val;
2144 		val = (reg & TSNEP_RX_STATISTIC_BUFFER_TOO_SMALL_MASK) >>
2145 		      TSNEP_RX_STATISTIC_BUFFER_TOO_SMALL_SHIFT;
2146 		stats->rx_dropped += val;
2147 		val = (reg & TSNEP_RX_STATISTIC_FIFO_OVERFLOW_MASK) >>
2148 		      TSNEP_RX_STATISTIC_FIFO_OVERFLOW_SHIFT;
2149 		stats->rx_errors += val;
2150 		stats->rx_fifo_errors += val;
2151 		val = (reg & TSNEP_RX_STATISTIC_INVALID_FRAME_MASK) >>
2152 		      TSNEP_RX_STATISTIC_INVALID_FRAME_SHIFT;
2153 		stats->rx_errors += val;
2154 		stats->rx_frame_errors += val;
2155 	}
2156 
2157 	reg = ioread32(adapter->addr + ECM_STAT);
2158 	val = (reg & ECM_STAT_RX_ERR_MASK) >> ECM_STAT_RX_ERR_SHIFT;
2159 	stats->rx_errors += val;
2160 	val = (reg & ECM_STAT_INV_FRM_MASK) >> ECM_STAT_INV_FRM_SHIFT;
2161 	stats->rx_errors += val;
2162 	stats->rx_crc_errors += val;
2163 	val = (reg & ECM_STAT_FWD_RX_ERR_MASK) >> ECM_STAT_FWD_RX_ERR_SHIFT;
2164 	stats->rx_errors += val;
2165 }
2166 
2167 static void tsnep_mac_set_address(struct tsnep_adapter *adapter, u8 *addr)
2168 {
2169 	iowrite32(*(u32 *)addr, adapter->addr + TSNEP_MAC_ADDRESS_LOW);
2170 	iowrite16(*(u16 *)(addr + sizeof(u32)),
2171 		  adapter->addr + TSNEP_MAC_ADDRESS_HIGH);
2172 
2173 	ether_addr_copy(adapter->mac_address, addr);
2174 	netif_info(adapter, drv, adapter->netdev, "MAC address set to %pM\n",
2175 		   addr);
2176 }
2177 
2178 static int tsnep_netdev_set_mac_address(struct net_device *netdev, void *addr)
2179 {
2180 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2181 	struct sockaddr *sock_addr = addr;
2182 	int retval;
2183 
2184 	retval = eth_prepare_mac_addr_change(netdev, sock_addr);
2185 	if (retval)
2186 		return retval;
2187 	eth_hw_addr_set(netdev, sock_addr->sa_data);
2188 	tsnep_mac_set_address(adapter, sock_addr->sa_data);
2189 
2190 	return 0;
2191 }
2192 
2193 static int tsnep_netdev_set_features(struct net_device *netdev,
2194 				     netdev_features_t features)
2195 {
2196 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2197 	netdev_features_t changed = netdev->features ^ features;
2198 	bool enable;
2199 	int retval = 0;
2200 
2201 	if (changed & NETIF_F_LOOPBACK) {
2202 		enable = !!(features & NETIF_F_LOOPBACK);
2203 		retval = tsnep_phy_loopback(adapter, enable);
2204 	}
2205 
2206 	return retval;
2207 }
2208 
2209 static ktime_t tsnep_netdev_get_tstamp(struct net_device *netdev,
2210 				       const struct skb_shared_hwtstamps *hwtstamps,
2211 				       bool cycles)
2212 {
2213 	struct tsnep_rx_inline *rx_inline = hwtstamps->netdev_data;
2214 	u64 timestamp;
2215 
2216 	if (cycles)
2217 		timestamp = __le64_to_cpu(rx_inline->counter);
2218 	else
2219 		timestamp = __le64_to_cpu(rx_inline->timestamp);
2220 
2221 	return ns_to_ktime(timestamp);
2222 }
2223 
2224 static int tsnep_netdev_bpf(struct net_device *dev, struct netdev_bpf *bpf)
2225 {
2226 	struct tsnep_adapter *adapter = netdev_priv(dev);
2227 
2228 	switch (bpf->command) {
2229 	case XDP_SETUP_PROG:
2230 		return tsnep_xdp_setup_prog(adapter, bpf->prog, bpf->extack);
2231 	case XDP_SETUP_XSK_POOL:
2232 		return tsnep_xdp_setup_pool(adapter, bpf->xsk.pool,
2233 					    bpf->xsk.queue_id);
2234 	default:
2235 		return -EOPNOTSUPP;
2236 	}
2237 }
2238 
2239 static struct tsnep_tx *tsnep_xdp_get_tx(struct tsnep_adapter *adapter, u32 cpu)
2240 {
2241 	if (cpu >= TSNEP_MAX_QUEUES)
2242 		cpu &= TSNEP_MAX_QUEUES - 1;
2243 
2244 	while (cpu >= adapter->num_tx_queues)
2245 		cpu -= adapter->num_tx_queues;
2246 
2247 	return &adapter->tx[cpu];
2248 }
2249 
2250 static int tsnep_netdev_xdp_xmit(struct net_device *dev, int n,
2251 				 struct xdp_frame **xdp, u32 flags)
2252 {
2253 	struct tsnep_adapter *adapter = netdev_priv(dev);
2254 	u32 cpu = smp_processor_id();
2255 	struct netdev_queue *nq;
2256 	struct tsnep_tx *tx;
2257 	int nxmit;
2258 	bool xmit;
2259 
2260 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2261 		return -EINVAL;
2262 
2263 	tx = tsnep_xdp_get_tx(adapter, cpu);
2264 	nq = netdev_get_tx_queue(adapter->netdev, tx->queue_index);
2265 
2266 	__netif_tx_lock(nq, cpu);
2267 
2268 	for (nxmit = 0; nxmit < n; nxmit++) {
2269 		xmit = tsnep_xdp_xmit_frame_ring(xdp[nxmit], tx,
2270 						 TSNEP_TX_TYPE_XDP_NDO);
2271 		if (!xmit)
2272 			break;
2273 
2274 		/* avoid transmit queue timeout since we share it with the slow
2275 		 * path
2276 		 */
2277 		txq_trans_cond_update(nq);
2278 	}
2279 
2280 	if (flags & XDP_XMIT_FLUSH)
2281 		tsnep_xdp_xmit_flush(tx);
2282 
2283 	__netif_tx_unlock(nq);
2284 
2285 	return nxmit;
2286 }
2287 
2288 static int tsnep_netdev_xsk_wakeup(struct net_device *dev, u32 queue_id,
2289 				   u32 flags)
2290 {
2291 	struct tsnep_adapter *adapter = netdev_priv(dev);
2292 	struct tsnep_queue *queue;
2293 
2294 	if (queue_id >= adapter->num_rx_queues ||
2295 	    queue_id >= adapter->num_tx_queues)
2296 		return -EINVAL;
2297 
2298 	queue = &adapter->queue[queue_id];
2299 
2300 	if (!napi_if_scheduled_mark_missed(&queue->napi))
2301 		napi_schedule(&queue->napi);
2302 
2303 	return 0;
2304 }
2305 
2306 static const struct net_device_ops tsnep_netdev_ops = {
2307 	.ndo_open = tsnep_netdev_open,
2308 	.ndo_stop = tsnep_netdev_close,
2309 	.ndo_start_xmit = tsnep_netdev_xmit_frame,
2310 	.ndo_eth_ioctl = tsnep_netdev_ioctl,
2311 	.ndo_set_rx_mode = tsnep_netdev_set_multicast,
2312 	.ndo_get_stats64 = tsnep_netdev_get_stats64,
2313 	.ndo_set_mac_address = tsnep_netdev_set_mac_address,
2314 	.ndo_set_features = tsnep_netdev_set_features,
2315 	.ndo_get_tstamp = tsnep_netdev_get_tstamp,
2316 	.ndo_setup_tc = tsnep_tc_setup,
2317 	.ndo_bpf = tsnep_netdev_bpf,
2318 	.ndo_xdp_xmit = tsnep_netdev_xdp_xmit,
2319 	.ndo_xsk_wakeup = tsnep_netdev_xsk_wakeup,
2320 };
2321 
2322 static int tsnep_mac_init(struct tsnep_adapter *adapter)
2323 {
2324 	int retval;
2325 
2326 	/* initialize RX filtering, at least configured MAC address and
2327 	 * broadcast are not filtered
2328 	 */
2329 	iowrite16(0, adapter->addr + TSNEP_RX_FILTER);
2330 
2331 	/* try to get MAC address in the following order:
2332 	 * - device tree
2333 	 * - valid MAC address already set
2334 	 * - MAC address register if valid
2335 	 * - random MAC address
2336 	 */
2337 	retval = of_get_mac_address(adapter->pdev->dev.of_node,
2338 				    adapter->mac_address);
2339 	if (retval == -EPROBE_DEFER)
2340 		return retval;
2341 	if (retval && !is_valid_ether_addr(adapter->mac_address)) {
2342 		*(u32 *)adapter->mac_address =
2343 			ioread32(adapter->addr + TSNEP_MAC_ADDRESS_LOW);
2344 		*(u16 *)(adapter->mac_address + sizeof(u32)) =
2345 			ioread16(adapter->addr + TSNEP_MAC_ADDRESS_HIGH);
2346 		if (!is_valid_ether_addr(adapter->mac_address))
2347 			eth_random_addr(adapter->mac_address);
2348 	}
2349 
2350 	tsnep_mac_set_address(adapter, adapter->mac_address);
2351 	eth_hw_addr_set(adapter->netdev, adapter->mac_address);
2352 
2353 	return 0;
2354 }
2355 
2356 static int tsnep_mdio_init(struct tsnep_adapter *adapter)
2357 {
2358 	struct device_node *np = adapter->pdev->dev.of_node;
2359 	int retval;
2360 
2361 	if (np) {
2362 		np = of_get_child_by_name(np, "mdio");
2363 		if (!np)
2364 			return 0;
2365 
2366 		adapter->suppress_preamble =
2367 			of_property_read_bool(np, "suppress-preamble");
2368 	}
2369 
2370 	adapter->mdiobus = devm_mdiobus_alloc(&adapter->pdev->dev);
2371 	if (!adapter->mdiobus) {
2372 		retval = -ENOMEM;
2373 
2374 		goto out;
2375 	}
2376 
2377 	adapter->mdiobus->priv = (void *)adapter;
2378 	adapter->mdiobus->parent = &adapter->pdev->dev;
2379 	adapter->mdiobus->read = tsnep_mdiobus_read;
2380 	adapter->mdiobus->write = tsnep_mdiobus_write;
2381 	adapter->mdiobus->name = TSNEP "-mdiobus";
2382 	snprintf(adapter->mdiobus->id, MII_BUS_ID_SIZE, "%s",
2383 		 adapter->pdev->name);
2384 
2385 	/* do not scan broadcast address */
2386 	adapter->mdiobus->phy_mask = 0x0000001;
2387 
2388 	retval = of_mdiobus_register(adapter->mdiobus, np);
2389 
2390 out:
2391 	of_node_put(np);
2392 
2393 	return retval;
2394 }
2395 
2396 static int tsnep_phy_init(struct tsnep_adapter *adapter)
2397 {
2398 	struct device_node *phy_node;
2399 	int retval;
2400 
2401 	retval = of_get_phy_mode(adapter->pdev->dev.of_node,
2402 				 &adapter->phy_mode);
2403 	if (retval)
2404 		adapter->phy_mode = PHY_INTERFACE_MODE_GMII;
2405 
2406 	phy_node = of_parse_phandle(adapter->pdev->dev.of_node, "phy-handle",
2407 				    0);
2408 	adapter->phydev = of_phy_find_device(phy_node);
2409 	of_node_put(phy_node);
2410 	if (!adapter->phydev && adapter->mdiobus)
2411 		adapter->phydev = phy_find_first(adapter->mdiobus);
2412 	if (!adapter->phydev)
2413 		return -EIO;
2414 
2415 	return 0;
2416 }
2417 
2418 static int tsnep_queue_init(struct tsnep_adapter *adapter, int queue_count)
2419 {
2420 	u32 irq_mask = ECM_INT_TX_0 | ECM_INT_RX_0;
2421 	char name[8];
2422 	int i;
2423 	int retval;
2424 
2425 	/* one TX/RX queue pair for netdev is mandatory */
2426 	if (platform_irq_count(adapter->pdev) == 1)
2427 		retval = platform_get_irq(adapter->pdev, 0);
2428 	else
2429 		retval = platform_get_irq_byname(adapter->pdev, "mac");
2430 	if (retval < 0)
2431 		return retval;
2432 	adapter->num_tx_queues = 1;
2433 	adapter->num_rx_queues = 1;
2434 	adapter->num_queues = 1;
2435 	adapter->queue[0].adapter = adapter;
2436 	adapter->queue[0].irq = retval;
2437 	adapter->queue[0].tx = &adapter->tx[0];
2438 	adapter->queue[0].tx->adapter = adapter;
2439 	adapter->queue[0].tx->addr = adapter->addr + TSNEP_QUEUE(0);
2440 	adapter->queue[0].tx->queue_index = 0;
2441 	adapter->queue[0].rx = &adapter->rx[0];
2442 	adapter->queue[0].rx->adapter = adapter;
2443 	adapter->queue[0].rx->addr = adapter->addr + TSNEP_QUEUE(0);
2444 	adapter->queue[0].rx->queue_index = 0;
2445 	adapter->queue[0].irq_mask = irq_mask;
2446 	adapter->queue[0].irq_delay_addr = adapter->addr + ECM_INT_DELAY;
2447 	retval = tsnep_set_irq_coalesce(&adapter->queue[0],
2448 					TSNEP_COALESCE_USECS_DEFAULT);
2449 	if (retval < 0)
2450 		return retval;
2451 
2452 	adapter->netdev->irq = adapter->queue[0].irq;
2453 
2454 	/* add additional TX/RX queue pairs only if dedicated interrupt is
2455 	 * available
2456 	 */
2457 	for (i = 1; i < queue_count; i++) {
2458 		sprintf(name, "txrx-%d", i);
2459 		retval = platform_get_irq_byname_optional(adapter->pdev, name);
2460 		if (retval < 0)
2461 			break;
2462 
2463 		adapter->num_tx_queues++;
2464 		adapter->num_rx_queues++;
2465 		adapter->num_queues++;
2466 		adapter->queue[i].adapter = adapter;
2467 		adapter->queue[i].irq = retval;
2468 		adapter->queue[i].tx = &adapter->tx[i];
2469 		adapter->queue[i].tx->adapter = adapter;
2470 		adapter->queue[i].tx->addr = adapter->addr + TSNEP_QUEUE(i);
2471 		adapter->queue[i].tx->queue_index = i;
2472 		adapter->queue[i].rx = &adapter->rx[i];
2473 		adapter->queue[i].rx->adapter = adapter;
2474 		adapter->queue[i].rx->addr = adapter->addr + TSNEP_QUEUE(i);
2475 		adapter->queue[i].rx->queue_index = i;
2476 		adapter->queue[i].irq_mask =
2477 			irq_mask << (ECM_INT_TXRX_SHIFT * i);
2478 		adapter->queue[i].irq_delay_addr =
2479 			adapter->addr + ECM_INT_DELAY + ECM_INT_DELAY_OFFSET * i;
2480 		retval = tsnep_set_irq_coalesce(&adapter->queue[i],
2481 						TSNEP_COALESCE_USECS_DEFAULT);
2482 		if (retval < 0)
2483 			return retval;
2484 	}
2485 
2486 	return 0;
2487 }
2488 
2489 static int tsnep_probe(struct platform_device *pdev)
2490 {
2491 	struct tsnep_adapter *adapter;
2492 	struct net_device *netdev;
2493 	struct resource *io;
2494 	u32 type;
2495 	int revision;
2496 	int version;
2497 	int queue_count;
2498 	int retval;
2499 
2500 	netdev = devm_alloc_etherdev_mqs(&pdev->dev,
2501 					 sizeof(struct tsnep_adapter),
2502 					 TSNEP_MAX_QUEUES, TSNEP_MAX_QUEUES);
2503 	if (!netdev)
2504 		return -ENODEV;
2505 	SET_NETDEV_DEV(netdev, &pdev->dev);
2506 	adapter = netdev_priv(netdev);
2507 	platform_set_drvdata(pdev, adapter);
2508 	adapter->pdev = pdev;
2509 	adapter->dmadev = &pdev->dev;
2510 	adapter->netdev = netdev;
2511 	adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE |
2512 			      NETIF_MSG_LINK | NETIF_MSG_IFUP |
2513 			      NETIF_MSG_IFDOWN | NETIF_MSG_TX_QUEUED;
2514 
2515 	netdev->min_mtu = ETH_MIN_MTU;
2516 	netdev->max_mtu = TSNEP_MAX_FRAME_SIZE;
2517 
2518 	mutex_init(&adapter->gate_control_lock);
2519 	mutex_init(&adapter->rxnfc_lock);
2520 	INIT_LIST_HEAD(&adapter->rxnfc_rules);
2521 
2522 	io = platform_get_resource(pdev, IORESOURCE_MEM, 0);
2523 	adapter->addr = devm_ioremap_resource(&pdev->dev, io);
2524 	if (IS_ERR(adapter->addr))
2525 		return PTR_ERR(adapter->addr);
2526 	netdev->mem_start = io->start;
2527 	netdev->mem_end = io->end;
2528 
2529 	type = ioread32(adapter->addr + ECM_TYPE);
2530 	revision = (type & ECM_REVISION_MASK) >> ECM_REVISION_SHIFT;
2531 	version = (type & ECM_VERSION_MASK) >> ECM_VERSION_SHIFT;
2532 	queue_count = (type & ECM_QUEUE_COUNT_MASK) >> ECM_QUEUE_COUNT_SHIFT;
2533 	adapter->gate_control = type & ECM_GATE_CONTROL;
2534 	adapter->rxnfc_max = TSNEP_RX_ASSIGN_ETHER_TYPE_COUNT;
2535 
2536 	tsnep_disable_irq(adapter, ECM_INT_ALL);
2537 
2538 	retval = tsnep_queue_init(adapter, queue_count);
2539 	if (retval)
2540 		return retval;
2541 
2542 	retval = dma_set_mask_and_coherent(&adapter->pdev->dev,
2543 					   DMA_BIT_MASK(64));
2544 	if (retval) {
2545 		dev_err(&adapter->pdev->dev, "no usable DMA configuration.\n");
2546 		return retval;
2547 	}
2548 
2549 	retval = tsnep_mac_init(adapter);
2550 	if (retval)
2551 		return retval;
2552 
2553 	retval = tsnep_mdio_init(adapter);
2554 	if (retval)
2555 		goto mdio_init_failed;
2556 
2557 	retval = tsnep_phy_init(adapter);
2558 	if (retval)
2559 		goto phy_init_failed;
2560 
2561 	retval = tsnep_ptp_init(adapter);
2562 	if (retval)
2563 		goto ptp_init_failed;
2564 
2565 	retval = tsnep_tc_init(adapter);
2566 	if (retval)
2567 		goto tc_init_failed;
2568 
2569 	retval = tsnep_rxnfc_init(adapter);
2570 	if (retval)
2571 		goto rxnfc_init_failed;
2572 
2573 	netdev->netdev_ops = &tsnep_netdev_ops;
2574 	netdev->ethtool_ops = &tsnep_ethtool_ops;
2575 	netdev->features = NETIF_F_SG;
2576 	netdev->hw_features = netdev->features | NETIF_F_LOOPBACK;
2577 
2578 	netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
2579 			       NETDEV_XDP_ACT_NDO_XMIT |
2580 			       NETDEV_XDP_ACT_NDO_XMIT_SG |
2581 			       NETDEV_XDP_ACT_XSK_ZEROCOPY;
2582 
2583 	/* carrier off reporting is important to ethtool even BEFORE open */
2584 	netif_carrier_off(netdev);
2585 
2586 	retval = register_netdev(netdev);
2587 	if (retval)
2588 		goto register_failed;
2589 
2590 	dev_info(&adapter->pdev->dev, "device version %d.%02d\n", version,
2591 		 revision);
2592 	if (adapter->gate_control)
2593 		dev_info(&adapter->pdev->dev, "gate control detected\n");
2594 
2595 	return 0;
2596 
2597 register_failed:
2598 	tsnep_rxnfc_cleanup(adapter);
2599 rxnfc_init_failed:
2600 	tsnep_tc_cleanup(adapter);
2601 tc_init_failed:
2602 	tsnep_ptp_cleanup(adapter);
2603 ptp_init_failed:
2604 phy_init_failed:
2605 	if (adapter->mdiobus)
2606 		mdiobus_unregister(adapter->mdiobus);
2607 mdio_init_failed:
2608 	return retval;
2609 }
2610 
2611 static int tsnep_remove(struct platform_device *pdev)
2612 {
2613 	struct tsnep_adapter *adapter = platform_get_drvdata(pdev);
2614 
2615 	unregister_netdev(adapter->netdev);
2616 
2617 	tsnep_rxnfc_cleanup(adapter);
2618 
2619 	tsnep_tc_cleanup(adapter);
2620 
2621 	tsnep_ptp_cleanup(adapter);
2622 
2623 	if (adapter->mdiobus)
2624 		mdiobus_unregister(adapter->mdiobus);
2625 
2626 	tsnep_disable_irq(adapter, ECM_INT_ALL);
2627 
2628 	return 0;
2629 }
2630 
2631 static const struct of_device_id tsnep_of_match[] = {
2632 	{ .compatible = "engleder,tsnep", },
2633 { },
2634 };
2635 MODULE_DEVICE_TABLE(of, tsnep_of_match);
2636 
2637 static struct platform_driver tsnep_driver = {
2638 	.driver = {
2639 		.name = TSNEP,
2640 		.of_match_table = tsnep_of_match,
2641 	},
2642 	.probe = tsnep_probe,
2643 	.remove = tsnep_remove,
2644 };
2645 module_platform_driver(tsnep_driver);
2646 
2647 MODULE_AUTHOR("Gerhard Engleder <gerhard@engleder-embedded.com>");
2648 MODULE_DESCRIPTION("TSN endpoint Ethernet MAC driver");
2649 MODULE_LICENSE("GPL");
2650