xref: /linux/drivers/net/ethernet/engleder/tsnep_main.c (revision f5c31bcf604db54470868f3118a60dc4a9ba8813)
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 /* mapping type */
55 #define TSNEP_TX_TYPE_MAP		BIT(0)
56 #define TSNEP_TX_TYPE_MAP_PAGE		BIT(1)
57 #define TSNEP_TX_TYPE_INLINE		BIT(2)
58 /* buffer type */
59 #define TSNEP_TX_TYPE_SKB		BIT(8)
60 #define TSNEP_TX_TYPE_SKB_MAP		(TSNEP_TX_TYPE_SKB | TSNEP_TX_TYPE_MAP)
61 #define TSNEP_TX_TYPE_SKB_INLINE	(TSNEP_TX_TYPE_SKB | TSNEP_TX_TYPE_INLINE)
62 #define TSNEP_TX_TYPE_SKB_FRAG		BIT(9)
63 #define TSNEP_TX_TYPE_SKB_FRAG_MAP_PAGE	(TSNEP_TX_TYPE_SKB_FRAG | TSNEP_TX_TYPE_MAP_PAGE)
64 #define TSNEP_TX_TYPE_SKB_FRAG_INLINE	(TSNEP_TX_TYPE_SKB_FRAG | TSNEP_TX_TYPE_INLINE)
65 #define TSNEP_TX_TYPE_XDP_TX		BIT(10)
66 #define TSNEP_TX_TYPE_XDP_NDO		BIT(11)
67 #define TSNEP_TX_TYPE_XDP_NDO_MAP_PAGE	(TSNEP_TX_TYPE_XDP_NDO | TSNEP_TX_TYPE_MAP_PAGE)
68 #define TSNEP_TX_TYPE_XDP		(TSNEP_TX_TYPE_XDP_TX | TSNEP_TX_TYPE_XDP_NDO)
69 #define TSNEP_TX_TYPE_XSK		BIT(12)
70 
71 #define TSNEP_XDP_TX		BIT(0)
72 #define TSNEP_XDP_REDIRECT	BIT(1)
73 
74 static void tsnep_enable_irq(struct tsnep_adapter *adapter, u32 mask)
75 {
76 	iowrite32(mask, adapter->addr + ECM_INT_ENABLE);
77 }
78 
79 static void tsnep_disable_irq(struct tsnep_adapter *adapter, u32 mask)
80 {
81 	mask |= ECM_INT_DISABLE;
82 	iowrite32(mask, adapter->addr + ECM_INT_ENABLE);
83 }
84 
85 static irqreturn_t tsnep_irq(int irq, void *arg)
86 {
87 	struct tsnep_adapter *adapter = arg;
88 	u32 active = ioread32(adapter->addr + ECM_INT_ACTIVE);
89 
90 	/* acknowledge interrupt */
91 	if (active != 0)
92 		iowrite32(active, adapter->addr + ECM_INT_ACKNOWLEDGE);
93 
94 	/* handle link interrupt */
95 	if ((active & ECM_INT_LINK) != 0)
96 		phy_mac_interrupt(adapter->netdev->phydev);
97 
98 	/* handle TX/RX queue 0 interrupt */
99 	if ((active & adapter->queue[0].irq_mask) != 0) {
100 		if (napi_schedule_prep(&adapter->queue[0].napi)) {
101 			tsnep_disable_irq(adapter, adapter->queue[0].irq_mask);
102 			/* schedule after masking to avoid races */
103 			__napi_schedule(&adapter->queue[0].napi);
104 		}
105 	}
106 
107 	return IRQ_HANDLED;
108 }
109 
110 static irqreturn_t tsnep_irq_txrx(int irq, void *arg)
111 {
112 	struct tsnep_queue *queue = arg;
113 
114 	/* handle TX/RX queue interrupt */
115 	if (napi_schedule_prep(&queue->napi)) {
116 		tsnep_disable_irq(queue->adapter, queue->irq_mask);
117 		/* schedule after masking to avoid races */
118 		__napi_schedule(&queue->napi);
119 	}
120 
121 	return IRQ_HANDLED;
122 }
123 
124 int tsnep_set_irq_coalesce(struct tsnep_queue *queue, u32 usecs)
125 {
126 	if (usecs > TSNEP_COALESCE_USECS_MAX)
127 		return -ERANGE;
128 
129 	usecs /= ECM_INT_DELAY_BASE_US;
130 	usecs <<= ECM_INT_DELAY_SHIFT;
131 	usecs &= ECM_INT_DELAY_MASK;
132 
133 	queue->irq_delay &= ~ECM_INT_DELAY_MASK;
134 	queue->irq_delay |= usecs;
135 	iowrite8(queue->irq_delay, queue->irq_delay_addr);
136 
137 	return 0;
138 }
139 
140 u32 tsnep_get_irq_coalesce(struct tsnep_queue *queue)
141 {
142 	u32 usecs;
143 
144 	usecs = (queue->irq_delay & ECM_INT_DELAY_MASK);
145 	usecs >>= ECM_INT_DELAY_SHIFT;
146 	usecs *= ECM_INT_DELAY_BASE_US;
147 
148 	return usecs;
149 }
150 
151 static int tsnep_mdiobus_read(struct mii_bus *bus, int addr, int regnum)
152 {
153 	struct tsnep_adapter *adapter = bus->priv;
154 	u32 md;
155 	int retval;
156 
157 	md = ECM_MD_READ;
158 	if (!adapter->suppress_preamble)
159 		md |= ECM_MD_PREAMBLE;
160 	md |= (regnum << ECM_MD_ADDR_SHIFT) & ECM_MD_ADDR_MASK;
161 	md |= (addr << ECM_MD_PHY_ADDR_SHIFT) & ECM_MD_PHY_ADDR_MASK;
162 	iowrite32(md, adapter->addr + ECM_MD_CONTROL);
163 	retval = readl_poll_timeout_atomic(adapter->addr + ECM_MD_STATUS, md,
164 					   !(md & ECM_MD_BUSY), 16, 1000);
165 	if (retval != 0)
166 		return retval;
167 
168 	return (md & ECM_MD_DATA_MASK) >> ECM_MD_DATA_SHIFT;
169 }
170 
171 static int tsnep_mdiobus_write(struct mii_bus *bus, int addr, int regnum,
172 			       u16 val)
173 {
174 	struct tsnep_adapter *adapter = bus->priv;
175 	u32 md;
176 	int retval;
177 
178 	md = ECM_MD_WRITE;
179 	if (!adapter->suppress_preamble)
180 		md |= ECM_MD_PREAMBLE;
181 	md |= (regnum << ECM_MD_ADDR_SHIFT) & ECM_MD_ADDR_MASK;
182 	md |= (addr << ECM_MD_PHY_ADDR_SHIFT) & ECM_MD_PHY_ADDR_MASK;
183 	md |= ((u32)val << ECM_MD_DATA_SHIFT) & ECM_MD_DATA_MASK;
184 	iowrite32(md, adapter->addr + ECM_MD_CONTROL);
185 	retval = readl_poll_timeout_atomic(adapter->addr + ECM_MD_STATUS, md,
186 					   !(md & ECM_MD_BUSY), 16, 1000);
187 	if (retval != 0)
188 		return retval;
189 
190 	return 0;
191 }
192 
193 static void tsnep_set_link_mode(struct tsnep_adapter *adapter)
194 {
195 	u32 mode;
196 
197 	switch (adapter->phydev->speed) {
198 	case SPEED_100:
199 		mode = ECM_LINK_MODE_100;
200 		break;
201 	case SPEED_1000:
202 		mode = ECM_LINK_MODE_1000;
203 		break;
204 	default:
205 		mode = ECM_LINK_MODE_OFF;
206 		break;
207 	}
208 	iowrite32(mode, adapter->addr + ECM_STATUS);
209 }
210 
211 static void tsnep_phy_link_status_change(struct net_device *netdev)
212 {
213 	struct tsnep_adapter *adapter = netdev_priv(netdev);
214 	struct phy_device *phydev = netdev->phydev;
215 
216 	if (phydev->link)
217 		tsnep_set_link_mode(adapter);
218 
219 	phy_print_status(netdev->phydev);
220 }
221 
222 static int tsnep_phy_loopback(struct tsnep_adapter *adapter, bool enable)
223 {
224 	int retval;
225 
226 	retval = phy_loopback(adapter->phydev, enable);
227 
228 	/* PHY link state change is not signaled if loopback is enabled, it
229 	 * would delay a working loopback anyway, let's ensure that loopback
230 	 * is working immediately by setting link mode directly
231 	 */
232 	if (!retval && enable) {
233 		netif_carrier_on(adapter->netdev);
234 		tsnep_set_link_mode(adapter);
235 	}
236 
237 	return retval;
238 }
239 
240 static int tsnep_phy_open(struct tsnep_adapter *adapter)
241 {
242 	struct phy_device *phydev;
243 	struct ethtool_keee ethtool_keee;
244 	int retval;
245 
246 	retval = phy_connect_direct(adapter->netdev, adapter->phydev,
247 				    tsnep_phy_link_status_change,
248 				    adapter->phy_mode);
249 	if (retval)
250 		return retval;
251 	phydev = adapter->netdev->phydev;
252 
253 	/* MAC supports only 100Mbps|1000Mbps full duplex
254 	 * SPE (Single Pair Ethernet) is also an option but not implemented yet
255 	 */
256 	phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_10baseT_Half_BIT);
257 	phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_10baseT_Full_BIT);
258 	phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_100baseT_Half_BIT);
259 	phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_1000baseT_Half_BIT);
260 
261 	/* disable EEE autoneg, EEE not supported by TSNEP */
262 	memset(&ethtool_keee, 0, sizeof(ethtool_keee));
263 	phy_ethtool_set_eee(adapter->phydev, &ethtool_keee);
264 
265 	adapter->phydev->irq = PHY_MAC_INTERRUPT;
266 	phy_start(adapter->phydev);
267 
268 	return 0;
269 }
270 
271 static void tsnep_phy_close(struct tsnep_adapter *adapter)
272 {
273 	phy_stop(adapter->netdev->phydev);
274 	phy_disconnect(adapter->netdev->phydev);
275 }
276 
277 static void tsnep_tx_ring_cleanup(struct tsnep_tx *tx)
278 {
279 	struct device *dmadev = tx->adapter->dmadev;
280 	int i;
281 
282 	memset(tx->entry, 0, sizeof(tx->entry));
283 
284 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
285 		if (tx->page[i]) {
286 			dma_free_coherent(dmadev, PAGE_SIZE, tx->page[i],
287 					  tx->page_dma[i]);
288 			tx->page[i] = NULL;
289 			tx->page_dma[i] = 0;
290 		}
291 	}
292 }
293 
294 static int tsnep_tx_ring_create(struct tsnep_tx *tx)
295 {
296 	struct device *dmadev = tx->adapter->dmadev;
297 	struct tsnep_tx_entry *entry;
298 	struct tsnep_tx_entry *next_entry;
299 	int i, j;
300 	int retval;
301 
302 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
303 		tx->page[i] =
304 			dma_alloc_coherent(dmadev, PAGE_SIZE, &tx->page_dma[i],
305 					   GFP_KERNEL);
306 		if (!tx->page[i]) {
307 			retval = -ENOMEM;
308 			goto alloc_failed;
309 		}
310 		for (j = 0; j < TSNEP_RING_ENTRIES_PER_PAGE; j++) {
311 			entry = &tx->entry[TSNEP_RING_ENTRIES_PER_PAGE * i + j];
312 			entry->desc_wb = (struct tsnep_tx_desc_wb *)
313 				(((u8 *)tx->page[i]) + TSNEP_DESC_SIZE * j);
314 			entry->desc = (struct tsnep_tx_desc *)
315 				(((u8 *)entry->desc_wb) + TSNEP_DESC_OFFSET);
316 			entry->desc_dma = tx->page_dma[i] + TSNEP_DESC_SIZE * j;
317 			entry->owner_user_flag = false;
318 		}
319 	}
320 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
321 		entry = &tx->entry[i];
322 		next_entry = &tx->entry[(i + 1) & TSNEP_RING_MASK];
323 		entry->desc->next = __cpu_to_le64(next_entry->desc_dma);
324 	}
325 
326 	return 0;
327 
328 alloc_failed:
329 	tsnep_tx_ring_cleanup(tx);
330 	return retval;
331 }
332 
333 static void tsnep_tx_init(struct tsnep_tx *tx)
334 {
335 	dma_addr_t dma;
336 
337 	dma = tx->entry[0].desc_dma | TSNEP_RESET_OWNER_COUNTER;
338 	iowrite32(DMA_ADDR_LOW(dma), tx->addr + TSNEP_TX_DESC_ADDR_LOW);
339 	iowrite32(DMA_ADDR_HIGH(dma), tx->addr + TSNEP_TX_DESC_ADDR_HIGH);
340 	tx->write = 0;
341 	tx->read = 0;
342 	tx->owner_counter = 1;
343 	tx->increment_owner_counter = TSNEP_RING_SIZE - 1;
344 }
345 
346 static void tsnep_tx_enable(struct tsnep_tx *tx)
347 {
348 	struct netdev_queue *nq;
349 
350 	nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
351 
352 	__netif_tx_lock_bh(nq);
353 	netif_tx_wake_queue(nq);
354 	__netif_tx_unlock_bh(nq);
355 }
356 
357 static void tsnep_tx_disable(struct tsnep_tx *tx, struct napi_struct *napi)
358 {
359 	struct netdev_queue *nq;
360 	u32 val;
361 
362 	nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
363 
364 	__netif_tx_lock_bh(nq);
365 	netif_tx_stop_queue(nq);
366 	__netif_tx_unlock_bh(nq);
367 
368 	/* wait until TX is done in hardware */
369 	readx_poll_timeout(ioread32, tx->addr + TSNEP_CONTROL, val,
370 			   ((val & TSNEP_CONTROL_TX_ENABLE) == 0), 10000,
371 			   1000000);
372 
373 	/* wait until TX is also done in software */
374 	while (READ_ONCE(tx->read) != tx->write) {
375 		napi_schedule(napi);
376 		napi_synchronize(napi);
377 	}
378 }
379 
380 static void tsnep_tx_activate(struct tsnep_tx *tx, int index, int length,
381 			      bool last)
382 {
383 	struct tsnep_tx_entry *entry = &tx->entry[index];
384 
385 	entry->properties = 0;
386 	/* xdpf and zc are union with skb */
387 	if (entry->skb) {
388 		entry->properties = length & TSNEP_DESC_LENGTH_MASK;
389 		entry->properties |= TSNEP_DESC_INTERRUPT_FLAG;
390 		if ((entry->type & TSNEP_TX_TYPE_SKB) &&
391 		    (skb_shinfo(entry->skb)->tx_flags & SKBTX_IN_PROGRESS))
392 			entry->properties |= TSNEP_DESC_EXTENDED_WRITEBACK_FLAG;
393 
394 		/* toggle user flag to prevent false acknowledge
395 		 *
396 		 * Only the first fragment is acknowledged. For all other
397 		 * fragments no acknowledge is done and the last written owner
398 		 * counter stays in the writeback descriptor. Therefore, it is
399 		 * possible that the last written owner counter is identical to
400 		 * the new incremented owner counter and a false acknowledge is
401 		 * detected before the real acknowledge has been done by
402 		 * hardware.
403 		 *
404 		 * The user flag is used to prevent this situation. The user
405 		 * flag is copied to the writeback descriptor by the hardware
406 		 * and is used as additional acknowledge data. By toggeling the
407 		 * user flag only for the first fragment (which is
408 		 * acknowledged), it is guaranteed that the last acknowledge
409 		 * done for this descriptor has used a different user flag and
410 		 * cannot be detected as false acknowledge.
411 		 */
412 		entry->owner_user_flag = !entry->owner_user_flag;
413 	}
414 	if (last)
415 		entry->properties |= TSNEP_TX_DESC_LAST_FRAGMENT_FLAG;
416 	if (index == tx->increment_owner_counter) {
417 		tx->owner_counter++;
418 		if (tx->owner_counter == 4)
419 			tx->owner_counter = 1;
420 		tx->increment_owner_counter--;
421 		if (tx->increment_owner_counter < 0)
422 			tx->increment_owner_counter = TSNEP_RING_SIZE - 1;
423 	}
424 	entry->properties |=
425 		(tx->owner_counter << TSNEP_DESC_OWNER_COUNTER_SHIFT) &
426 		TSNEP_DESC_OWNER_COUNTER_MASK;
427 	if (entry->owner_user_flag)
428 		entry->properties |= TSNEP_TX_DESC_OWNER_USER_FLAG;
429 	entry->desc->more_properties =
430 		__cpu_to_le32(entry->len & TSNEP_DESC_LENGTH_MASK);
431 	if (entry->type & TSNEP_TX_TYPE_INLINE)
432 		entry->properties |= TSNEP_TX_DESC_DATA_AFTER_DESC_FLAG;
433 
434 	/* descriptor properties shall be written last, because valid data is
435 	 * signaled there
436 	 */
437 	dma_wmb();
438 
439 	entry->desc->properties = __cpu_to_le32(entry->properties);
440 }
441 
442 static int tsnep_tx_desc_available(struct tsnep_tx *tx)
443 {
444 	if (tx->read <= tx->write)
445 		return TSNEP_RING_SIZE - tx->write + tx->read - 1;
446 	else
447 		return tx->read - tx->write - 1;
448 }
449 
450 static int tsnep_tx_map_frag(skb_frag_t *frag, struct tsnep_tx_entry *entry,
451 			     struct device *dmadev, dma_addr_t *dma)
452 {
453 	unsigned int len;
454 	int mapped;
455 
456 	len = skb_frag_size(frag);
457 	if (likely(len > TSNEP_DESC_SIZE_DATA_AFTER_INLINE)) {
458 		*dma = skb_frag_dma_map(dmadev, frag, 0, len, DMA_TO_DEVICE);
459 		if (dma_mapping_error(dmadev, *dma))
460 			return -ENOMEM;
461 		entry->type = TSNEP_TX_TYPE_SKB_FRAG_MAP_PAGE;
462 		mapped = 1;
463 	} else {
464 		void *fragdata = skb_frag_address_safe(frag);
465 
466 		if (likely(fragdata)) {
467 			memcpy(&entry->desc->tx, fragdata, len);
468 		} else {
469 			struct page *page = skb_frag_page(frag);
470 
471 			fragdata = kmap_local_page(page);
472 			memcpy(&entry->desc->tx, fragdata + skb_frag_off(frag),
473 			       len);
474 			kunmap_local(fragdata);
475 		}
476 		entry->type = TSNEP_TX_TYPE_SKB_FRAG_INLINE;
477 		mapped = 0;
478 	}
479 
480 	return mapped;
481 }
482 
483 static int tsnep_tx_map(struct sk_buff *skb, struct tsnep_tx *tx, int count)
484 {
485 	struct device *dmadev = tx->adapter->dmadev;
486 	struct tsnep_tx_entry *entry;
487 	unsigned int len;
488 	int map_len = 0;
489 	dma_addr_t dma;
490 	int i, mapped;
491 
492 	for (i = 0; i < count; i++) {
493 		entry = &tx->entry[(tx->write + i) & TSNEP_RING_MASK];
494 
495 		if (!i) {
496 			len = skb_headlen(skb);
497 			if (likely(len > TSNEP_DESC_SIZE_DATA_AFTER_INLINE)) {
498 				dma = dma_map_single(dmadev, skb->data, len,
499 						     DMA_TO_DEVICE);
500 				if (dma_mapping_error(dmadev, dma))
501 					return -ENOMEM;
502 				entry->type = TSNEP_TX_TYPE_SKB_MAP;
503 				mapped = 1;
504 			} else {
505 				memcpy(&entry->desc->tx, skb->data, len);
506 				entry->type = TSNEP_TX_TYPE_SKB_INLINE;
507 				mapped = 0;
508 			}
509 		} else {
510 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
511 
512 			len = skb_frag_size(frag);
513 			mapped = tsnep_tx_map_frag(frag, entry, dmadev, &dma);
514 			if (mapped < 0)
515 				return mapped;
516 		}
517 
518 		entry->len = len;
519 		if (likely(mapped)) {
520 			dma_unmap_addr_set(entry, dma, dma);
521 			entry->desc->tx = __cpu_to_le64(dma);
522 		}
523 
524 		map_len += len;
525 	}
526 
527 	return map_len;
528 }
529 
530 static int tsnep_tx_unmap(struct tsnep_tx *tx, int index, int count)
531 {
532 	struct device *dmadev = tx->adapter->dmadev;
533 	struct tsnep_tx_entry *entry;
534 	int map_len = 0;
535 	int i;
536 
537 	for (i = 0; i < count; i++) {
538 		entry = &tx->entry[(index + i) & TSNEP_RING_MASK];
539 
540 		if (entry->len) {
541 			if (entry->type & TSNEP_TX_TYPE_MAP)
542 				dma_unmap_single(dmadev,
543 						 dma_unmap_addr(entry, dma),
544 						 dma_unmap_len(entry, len),
545 						 DMA_TO_DEVICE);
546 			else if (entry->type & TSNEP_TX_TYPE_MAP_PAGE)
547 				dma_unmap_page(dmadev,
548 					       dma_unmap_addr(entry, dma),
549 					       dma_unmap_len(entry, len),
550 					       DMA_TO_DEVICE);
551 			map_len += entry->len;
552 			entry->len = 0;
553 		}
554 	}
555 
556 	return map_len;
557 }
558 
559 static netdev_tx_t tsnep_xmit_frame_ring(struct sk_buff *skb,
560 					 struct tsnep_tx *tx)
561 {
562 	int count = 1;
563 	struct tsnep_tx_entry *entry;
564 	int length;
565 	int i;
566 	int retval;
567 
568 	if (skb_shinfo(skb)->nr_frags > 0)
569 		count += skb_shinfo(skb)->nr_frags;
570 
571 	if (tsnep_tx_desc_available(tx) < count) {
572 		/* ring full, shall not happen because queue is stopped if full
573 		 * below
574 		 */
575 		netif_stop_subqueue(tx->adapter->netdev, tx->queue_index);
576 
577 		return NETDEV_TX_BUSY;
578 	}
579 
580 	entry = &tx->entry[tx->write];
581 	entry->skb = skb;
582 
583 	retval = tsnep_tx_map(skb, tx, count);
584 	if (retval < 0) {
585 		tsnep_tx_unmap(tx, tx->write, count);
586 		dev_kfree_skb_any(entry->skb);
587 		entry->skb = NULL;
588 
589 		tx->dropped++;
590 
591 		return NETDEV_TX_OK;
592 	}
593 	length = retval;
594 
595 	if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)
596 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
597 
598 	for (i = 0; i < count; i++)
599 		tsnep_tx_activate(tx, (tx->write + i) & TSNEP_RING_MASK, length,
600 				  i == count - 1);
601 	tx->write = (tx->write + count) & TSNEP_RING_MASK;
602 
603 	skb_tx_timestamp(skb);
604 
605 	/* descriptor properties shall be valid before hardware is notified */
606 	dma_wmb();
607 
608 	iowrite32(TSNEP_CONTROL_TX_ENABLE, tx->addr + TSNEP_CONTROL);
609 
610 	if (tsnep_tx_desc_available(tx) < (MAX_SKB_FRAGS + 1)) {
611 		/* ring can get full with next frame */
612 		netif_stop_subqueue(tx->adapter->netdev, tx->queue_index);
613 	}
614 
615 	return NETDEV_TX_OK;
616 }
617 
618 static int tsnep_xdp_tx_map(struct xdp_frame *xdpf, struct tsnep_tx *tx,
619 			    struct skb_shared_info *shinfo, int count, u32 type)
620 {
621 	struct device *dmadev = tx->adapter->dmadev;
622 	struct tsnep_tx_entry *entry;
623 	struct page *page;
624 	skb_frag_t *frag;
625 	unsigned int len;
626 	int map_len = 0;
627 	dma_addr_t dma;
628 	void *data;
629 	int i;
630 
631 	frag = NULL;
632 	len = xdpf->len;
633 	for (i = 0; i < count; i++) {
634 		entry = &tx->entry[(tx->write + i) & TSNEP_RING_MASK];
635 		if (type & TSNEP_TX_TYPE_XDP_NDO) {
636 			data = unlikely(frag) ? skb_frag_address(frag) :
637 						xdpf->data;
638 			dma = dma_map_single(dmadev, data, len, DMA_TO_DEVICE);
639 			if (dma_mapping_error(dmadev, dma))
640 				return -ENOMEM;
641 
642 			entry->type = TSNEP_TX_TYPE_XDP_NDO_MAP_PAGE;
643 		} else {
644 			page = unlikely(frag) ? skb_frag_page(frag) :
645 						virt_to_page(xdpf->data);
646 			dma = page_pool_get_dma_addr(page);
647 			if (unlikely(frag))
648 				dma += skb_frag_off(frag);
649 			else
650 				dma += sizeof(*xdpf) + xdpf->headroom;
651 			dma_sync_single_for_device(dmadev, dma, len,
652 						   DMA_BIDIRECTIONAL);
653 
654 			entry->type = TSNEP_TX_TYPE_XDP_TX;
655 		}
656 
657 		entry->len = len;
658 		dma_unmap_addr_set(entry, dma, dma);
659 
660 		entry->desc->tx = __cpu_to_le64(dma);
661 
662 		map_len += len;
663 
664 		if (i + 1 < count) {
665 			frag = &shinfo->frags[i];
666 			len = skb_frag_size(frag);
667 		}
668 	}
669 
670 	return map_len;
671 }
672 
673 /* This function requires __netif_tx_lock is held by the caller. */
674 static bool tsnep_xdp_xmit_frame_ring(struct xdp_frame *xdpf,
675 				      struct tsnep_tx *tx, u32 type)
676 {
677 	struct skb_shared_info *shinfo = xdp_get_shared_info_from_frame(xdpf);
678 	struct tsnep_tx_entry *entry;
679 	int count, length, retval, i;
680 
681 	count = 1;
682 	if (unlikely(xdp_frame_has_frags(xdpf)))
683 		count += shinfo->nr_frags;
684 
685 	/* ensure that TX ring is not filled up by XDP, always MAX_SKB_FRAGS
686 	 * will be available for normal TX path and queue is stopped there if
687 	 * necessary
688 	 */
689 	if (tsnep_tx_desc_available(tx) < (MAX_SKB_FRAGS + 1 + count))
690 		return false;
691 
692 	entry = &tx->entry[tx->write];
693 	entry->xdpf = xdpf;
694 
695 	retval = tsnep_xdp_tx_map(xdpf, tx, shinfo, count, type);
696 	if (retval < 0) {
697 		tsnep_tx_unmap(tx, tx->write, count);
698 		entry->xdpf = NULL;
699 
700 		tx->dropped++;
701 
702 		return false;
703 	}
704 	length = retval;
705 
706 	for (i = 0; i < count; i++)
707 		tsnep_tx_activate(tx, (tx->write + i) & TSNEP_RING_MASK, length,
708 				  i == count - 1);
709 	tx->write = (tx->write + count) & TSNEP_RING_MASK;
710 
711 	/* descriptor properties shall be valid before hardware is notified */
712 	dma_wmb();
713 
714 	return true;
715 }
716 
717 static void tsnep_xdp_xmit_flush(struct tsnep_tx *tx)
718 {
719 	iowrite32(TSNEP_CONTROL_TX_ENABLE, tx->addr + TSNEP_CONTROL);
720 }
721 
722 static bool tsnep_xdp_xmit_back(struct tsnep_adapter *adapter,
723 				struct xdp_buff *xdp,
724 				struct netdev_queue *tx_nq, struct tsnep_tx *tx,
725 				bool zc)
726 {
727 	struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
728 	bool xmit;
729 	u32 type;
730 
731 	if (unlikely(!xdpf))
732 		return false;
733 
734 	/* no page pool for zero copy */
735 	if (zc)
736 		type = TSNEP_TX_TYPE_XDP_NDO;
737 	else
738 		type = TSNEP_TX_TYPE_XDP_TX;
739 
740 	__netif_tx_lock(tx_nq, smp_processor_id());
741 
742 	xmit = tsnep_xdp_xmit_frame_ring(xdpf, tx, type);
743 
744 	/* Avoid transmit queue timeout since we share it with the slow path */
745 	if (xmit)
746 		txq_trans_cond_update(tx_nq);
747 
748 	__netif_tx_unlock(tx_nq);
749 
750 	return xmit;
751 }
752 
753 static int tsnep_xdp_tx_map_zc(struct xdp_desc *xdpd, struct tsnep_tx *tx)
754 {
755 	struct tsnep_tx_entry *entry;
756 	dma_addr_t dma;
757 
758 	entry = &tx->entry[tx->write];
759 	entry->zc = true;
760 
761 	dma = xsk_buff_raw_get_dma(tx->xsk_pool, xdpd->addr);
762 	xsk_buff_raw_dma_sync_for_device(tx->xsk_pool, dma, xdpd->len);
763 
764 	entry->type = TSNEP_TX_TYPE_XSK;
765 	entry->len = xdpd->len;
766 
767 	entry->desc->tx = __cpu_to_le64(dma);
768 
769 	return xdpd->len;
770 }
771 
772 static void tsnep_xdp_xmit_frame_ring_zc(struct xdp_desc *xdpd,
773 					 struct tsnep_tx *tx)
774 {
775 	int length;
776 
777 	length = tsnep_xdp_tx_map_zc(xdpd, tx);
778 
779 	tsnep_tx_activate(tx, tx->write, length, true);
780 	tx->write = (tx->write + 1) & TSNEP_RING_MASK;
781 }
782 
783 static void tsnep_xdp_xmit_zc(struct tsnep_tx *tx)
784 {
785 	int desc_available = tsnep_tx_desc_available(tx);
786 	struct xdp_desc *descs = tx->xsk_pool->tx_descs;
787 	int batch, i;
788 
789 	/* ensure that TX ring is not filled up by XDP, always MAX_SKB_FRAGS
790 	 * will be available for normal TX path and queue is stopped there if
791 	 * necessary
792 	 */
793 	if (desc_available <= (MAX_SKB_FRAGS + 1))
794 		return;
795 	desc_available -= MAX_SKB_FRAGS + 1;
796 
797 	batch = xsk_tx_peek_release_desc_batch(tx->xsk_pool, desc_available);
798 	for (i = 0; i < batch; i++)
799 		tsnep_xdp_xmit_frame_ring_zc(&descs[i], tx);
800 
801 	if (batch) {
802 		/* descriptor properties shall be valid before hardware is
803 		 * notified
804 		 */
805 		dma_wmb();
806 
807 		tsnep_xdp_xmit_flush(tx);
808 	}
809 }
810 
811 static bool tsnep_tx_poll(struct tsnep_tx *tx, int napi_budget)
812 {
813 	struct tsnep_tx_entry *entry;
814 	struct netdev_queue *nq;
815 	int xsk_frames = 0;
816 	int budget = 128;
817 	int length;
818 	int count;
819 
820 	nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
821 	__netif_tx_lock(nq, smp_processor_id());
822 
823 	do {
824 		if (tx->read == tx->write)
825 			break;
826 
827 		entry = &tx->entry[tx->read];
828 		if ((__le32_to_cpu(entry->desc_wb->properties) &
829 		     TSNEP_TX_DESC_OWNER_MASK) !=
830 		    (entry->properties & TSNEP_TX_DESC_OWNER_MASK))
831 			break;
832 
833 		/* descriptor properties shall be read first, because valid data
834 		 * is signaled there
835 		 */
836 		dma_rmb();
837 
838 		count = 1;
839 		if ((entry->type & TSNEP_TX_TYPE_SKB) &&
840 		    skb_shinfo(entry->skb)->nr_frags > 0)
841 			count += skb_shinfo(entry->skb)->nr_frags;
842 		else if ((entry->type & TSNEP_TX_TYPE_XDP) &&
843 			 xdp_frame_has_frags(entry->xdpf))
844 			count += xdp_get_shared_info_from_frame(entry->xdpf)->nr_frags;
845 
846 		length = tsnep_tx_unmap(tx, tx->read, count);
847 
848 		if ((entry->type & TSNEP_TX_TYPE_SKB) &&
849 		    (skb_shinfo(entry->skb)->tx_flags & SKBTX_IN_PROGRESS) &&
850 		    (__le32_to_cpu(entry->desc_wb->properties) &
851 		     TSNEP_DESC_EXTENDED_WRITEBACK_FLAG)) {
852 			struct skb_shared_hwtstamps hwtstamps;
853 			u64 timestamp;
854 
855 			if (skb_shinfo(entry->skb)->tx_flags &
856 			    SKBTX_HW_TSTAMP_USE_CYCLES)
857 				timestamp =
858 					__le64_to_cpu(entry->desc_wb->counter);
859 			else
860 				timestamp =
861 					__le64_to_cpu(entry->desc_wb->timestamp);
862 
863 			memset(&hwtstamps, 0, sizeof(hwtstamps));
864 			hwtstamps.hwtstamp = ns_to_ktime(timestamp);
865 
866 			skb_tstamp_tx(entry->skb, &hwtstamps);
867 		}
868 
869 		if (entry->type & TSNEP_TX_TYPE_SKB)
870 			napi_consume_skb(entry->skb, napi_budget);
871 		else if (entry->type & TSNEP_TX_TYPE_XDP)
872 			xdp_return_frame_rx_napi(entry->xdpf);
873 		else
874 			xsk_frames++;
875 		/* xdpf and zc are union with skb */
876 		entry->skb = NULL;
877 
878 		tx->read = (tx->read + count) & TSNEP_RING_MASK;
879 
880 		tx->packets++;
881 		tx->bytes += length + ETH_FCS_LEN;
882 
883 		budget--;
884 	} while (likely(budget));
885 
886 	if (tx->xsk_pool) {
887 		if (xsk_frames)
888 			xsk_tx_completed(tx->xsk_pool, xsk_frames);
889 		if (xsk_uses_need_wakeup(tx->xsk_pool))
890 			xsk_set_tx_need_wakeup(tx->xsk_pool);
891 		tsnep_xdp_xmit_zc(tx);
892 	}
893 
894 	if ((tsnep_tx_desc_available(tx) >= ((MAX_SKB_FRAGS + 1) * 2)) &&
895 	    netif_tx_queue_stopped(nq)) {
896 		netif_tx_wake_queue(nq);
897 	}
898 
899 	__netif_tx_unlock(nq);
900 
901 	return budget != 0;
902 }
903 
904 static bool tsnep_tx_pending(struct tsnep_tx *tx)
905 {
906 	struct tsnep_tx_entry *entry;
907 	struct netdev_queue *nq;
908 	bool pending = false;
909 
910 	nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
911 	__netif_tx_lock(nq, smp_processor_id());
912 
913 	if (tx->read != tx->write) {
914 		entry = &tx->entry[tx->read];
915 		if ((__le32_to_cpu(entry->desc_wb->properties) &
916 		     TSNEP_TX_DESC_OWNER_MASK) ==
917 		    (entry->properties & TSNEP_TX_DESC_OWNER_MASK))
918 			pending = true;
919 	}
920 
921 	__netif_tx_unlock(nq);
922 
923 	return pending;
924 }
925 
926 static int tsnep_tx_open(struct tsnep_tx *tx)
927 {
928 	int retval;
929 
930 	retval = tsnep_tx_ring_create(tx);
931 	if (retval)
932 		return retval;
933 
934 	tsnep_tx_init(tx);
935 
936 	return 0;
937 }
938 
939 static void tsnep_tx_close(struct tsnep_tx *tx)
940 {
941 	tsnep_tx_ring_cleanup(tx);
942 }
943 
944 static void tsnep_rx_ring_cleanup(struct tsnep_rx *rx)
945 {
946 	struct device *dmadev = rx->adapter->dmadev;
947 	struct tsnep_rx_entry *entry;
948 	int i;
949 
950 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
951 		entry = &rx->entry[i];
952 		if (!rx->xsk_pool && entry->page)
953 			page_pool_put_full_page(rx->page_pool, entry->page,
954 						false);
955 		if (rx->xsk_pool && entry->xdp)
956 			xsk_buff_free(entry->xdp);
957 		/* xdp is union with page */
958 		entry->page = NULL;
959 	}
960 
961 	if (rx->page_pool)
962 		page_pool_destroy(rx->page_pool);
963 
964 	memset(rx->entry, 0, sizeof(rx->entry));
965 
966 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
967 		if (rx->page[i]) {
968 			dma_free_coherent(dmadev, PAGE_SIZE, rx->page[i],
969 					  rx->page_dma[i]);
970 			rx->page[i] = NULL;
971 			rx->page_dma[i] = 0;
972 		}
973 	}
974 }
975 
976 static int tsnep_rx_ring_create(struct tsnep_rx *rx)
977 {
978 	struct device *dmadev = rx->adapter->dmadev;
979 	struct tsnep_rx_entry *entry;
980 	struct page_pool_params pp_params = { 0 };
981 	struct tsnep_rx_entry *next_entry;
982 	int i, j;
983 	int retval;
984 
985 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
986 		rx->page[i] =
987 			dma_alloc_coherent(dmadev, PAGE_SIZE, &rx->page_dma[i],
988 					   GFP_KERNEL);
989 		if (!rx->page[i]) {
990 			retval = -ENOMEM;
991 			goto failed;
992 		}
993 		for (j = 0; j < TSNEP_RING_ENTRIES_PER_PAGE; j++) {
994 			entry = &rx->entry[TSNEP_RING_ENTRIES_PER_PAGE * i + j];
995 			entry->desc_wb = (struct tsnep_rx_desc_wb *)
996 				(((u8 *)rx->page[i]) + TSNEP_DESC_SIZE * j);
997 			entry->desc = (struct tsnep_rx_desc *)
998 				(((u8 *)entry->desc_wb) + TSNEP_DESC_OFFSET);
999 			entry->desc_dma = rx->page_dma[i] + TSNEP_DESC_SIZE * j;
1000 		}
1001 	}
1002 
1003 	pp_params.flags = PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV;
1004 	pp_params.order = 0;
1005 	pp_params.pool_size = TSNEP_RING_SIZE;
1006 	pp_params.nid = dev_to_node(dmadev);
1007 	pp_params.dev = dmadev;
1008 	pp_params.dma_dir = DMA_BIDIRECTIONAL;
1009 	pp_params.max_len = TSNEP_MAX_RX_BUF_SIZE;
1010 	pp_params.offset = TSNEP_RX_OFFSET;
1011 	rx->page_pool = page_pool_create(&pp_params);
1012 	if (IS_ERR(rx->page_pool)) {
1013 		retval = PTR_ERR(rx->page_pool);
1014 		rx->page_pool = NULL;
1015 		goto failed;
1016 	}
1017 
1018 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
1019 		entry = &rx->entry[i];
1020 		next_entry = &rx->entry[(i + 1) & TSNEP_RING_MASK];
1021 		entry->desc->next = __cpu_to_le64(next_entry->desc_dma);
1022 	}
1023 
1024 	return 0;
1025 
1026 failed:
1027 	tsnep_rx_ring_cleanup(rx);
1028 	return retval;
1029 }
1030 
1031 static void tsnep_rx_init(struct tsnep_rx *rx)
1032 {
1033 	dma_addr_t dma;
1034 
1035 	dma = rx->entry[0].desc_dma | TSNEP_RESET_OWNER_COUNTER;
1036 	iowrite32(DMA_ADDR_LOW(dma), rx->addr + TSNEP_RX_DESC_ADDR_LOW);
1037 	iowrite32(DMA_ADDR_HIGH(dma), rx->addr + TSNEP_RX_DESC_ADDR_HIGH);
1038 	rx->write = 0;
1039 	rx->read = 0;
1040 	rx->owner_counter = 1;
1041 	rx->increment_owner_counter = TSNEP_RING_SIZE - 1;
1042 }
1043 
1044 static void tsnep_rx_enable(struct tsnep_rx *rx)
1045 {
1046 	/* descriptor properties shall be valid before hardware is notified */
1047 	dma_wmb();
1048 
1049 	iowrite32(TSNEP_CONTROL_RX_ENABLE, rx->addr + TSNEP_CONTROL);
1050 }
1051 
1052 static void tsnep_rx_disable(struct tsnep_rx *rx)
1053 {
1054 	u32 val;
1055 
1056 	iowrite32(TSNEP_CONTROL_RX_DISABLE, rx->addr + TSNEP_CONTROL);
1057 	readx_poll_timeout(ioread32, rx->addr + TSNEP_CONTROL, val,
1058 			   ((val & TSNEP_CONTROL_RX_ENABLE) == 0), 10000,
1059 			   1000000);
1060 }
1061 
1062 static int tsnep_rx_desc_available(struct tsnep_rx *rx)
1063 {
1064 	if (rx->read <= rx->write)
1065 		return TSNEP_RING_SIZE - rx->write + rx->read - 1;
1066 	else
1067 		return rx->read - rx->write - 1;
1068 }
1069 
1070 static void tsnep_rx_free_page_buffer(struct tsnep_rx *rx)
1071 {
1072 	struct page **page;
1073 
1074 	/* last entry of page_buffer is always zero, because ring cannot be
1075 	 * filled completely
1076 	 */
1077 	page = rx->page_buffer;
1078 	while (*page) {
1079 		page_pool_put_full_page(rx->page_pool, *page, false);
1080 		*page = NULL;
1081 		page++;
1082 	}
1083 }
1084 
1085 static int tsnep_rx_alloc_page_buffer(struct tsnep_rx *rx)
1086 {
1087 	int i;
1088 
1089 	/* alloc for all ring entries except the last one, because ring cannot
1090 	 * be filled completely
1091 	 */
1092 	for (i = 0; i < TSNEP_RING_SIZE - 1; i++) {
1093 		rx->page_buffer[i] = page_pool_dev_alloc_pages(rx->page_pool);
1094 		if (!rx->page_buffer[i]) {
1095 			tsnep_rx_free_page_buffer(rx);
1096 
1097 			return -ENOMEM;
1098 		}
1099 	}
1100 
1101 	return 0;
1102 }
1103 
1104 static void tsnep_rx_set_page(struct tsnep_rx *rx, struct tsnep_rx_entry *entry,
1105 			      struct page *page)
1106 {
1107 	entry->page = page;
1108 	entry->len = TSNEP_MAX_RX_BUF_SIZE;
1109 	entry->dma = page_pool_get_dma_addr(entry->page);
1110 	entry->desc->rx = __cpu_to_le64(entry->dma + TSNEP_RX_OFFSET);
1111 }
1112 
1113 static int tsnep_rx_alloc_buffer(struct tsnep_rx *rx, int index)
1114 {
1115 	struct tsnep_rx_entry *entry = &rx->entry[index];
1116 	struct page *page;
1117 
1118 	page = page_pool_dev_alloc_pages(rx->page_pool);
1119 	if (unlikely(!page))
1120 		return -ENOMEM;
1121 	tsnep_rx_set_page(rx, entry, page);
1122 
1123 	return 0;
1124 }
1125 
1126 static void tsnep_rx_reuse_buffer(struct tsnep_rx *rx, int index)
1127 {
1128 	struct tsnep_rx_entry *entry = &rx->entry[index];
1129 	struct tsnep_rx_entry *read = &rx->entry[rx->read];
1130 
1131 	tsnep_rx_set_page(rx, entry, read->page);
1132 	read->page = NULL;
1133 }
1134 
1135 static void tsnep_rx_activate(struct tsnep_rx *rx, int index)
1136 {
1137 	struct tsnep_rx_entry *entry = &rx->entry[index];
1138 
1139 	/* TSNEP_MAX_RX_BUF_SIZE and TSNEP_XSK_RX_BUF_SIZE are multiple of 4 */
1140 	entry->properties = entry->len & TSNEP_DESC_LENGTH_MASK;
1141 	entry->properties |= TSNEP_DESC_INTERRUPT_FLAG;
1142 	if (index == rx->increment_owner_counter) {
1143 		rx->owner_counter++;
1144 		if (rx->owner_counter == 4)
1145 			rx->owner_counter = 1;
1146 		rx->increment_owner_counter--;
1147 		if (rx->increment_owner_counter < 0)
1148 			rx->increment_owner_counter = TSNEP_RING_SIZE - 1;
1149 	}
1150 	entry->properties |=
1151 		(rx->owner_counter << TSNEP_DESC_OWNER_COUNTER_SHIFT) &
1152 		TSNEP_DESC_OWNER_COUNTER_MASK;
1153 
1154 	/* descriptor properties shall be written last, because valid data is
1155 	 * signaled there
1156 	 */
1157 	dma_wmb();
1158 
1159 	entry->desc->properties = __cpu_to_le32(entry->properties);
1160 }
1161 
1162 static int tsnep_rx_alloc(struct tsnep_rx *rx, int count, bool reuse)
1163 {
1164 	bool alloc_failed = false;
1165 	int i, index;
1166 
1167 	for (i = 0; i < count && !alloc_failed; i++) {
1168 		index = (rx->write + i) & TSNEP_RING_MASK;
1169 
1170 		if (unlikely(tsnep_rx_alloc_buffer(rx, index))) {
1171 			rx->alloc_failed++;
1172 			alloc_failed = true;
1173 
1174 			/* reuse only if no other allocation was successful */
1175 			if (i == 0 && reuse)
1176 				tsnep_rx_reuse_buffer(rx, index);
1177 			else
1178 				break;
1179 		}
1180 
1181 		tsnep_rx_activate(rx, index);
1182 	}
1183 
1184 	if (i)
1185 		rx->write = (rx->write + i) & TSNEP_RING_MASK;
1186 
1187 	return i;
1188 }
1189 
1190 static int tsnep_rx_refill(struct tsnep_rx *rx, int count, bool reuse)
1191 {
1192 	int desc_refilled;
1193 
1194 	desc_refilled = tsnep_rx_alloc(rx, count, reuse);
1195 	if (desc_refilled)
1196 		tsnep_rx_enable(rx);
1197 
1198 	return desc_refilled;
1199 }
1200 
1201 static void tsnep_rx_set_xdp(struct tsnep_rx *rx, struct tsnep_rx_entry *entry,
1202 			     struct xdp_buff *xdp)
1203 {
1204 	entry->xdp = xdp;
1205 	entry->len = TSNEP_XSK_RX_BUF_SIZE;
1206 	entry->dma = xsk_buff_xdp_get_dma(entry->xdp);
1207 	entry->desc->rx = __cpu_to_le64(entry->dma);
1208 }
1209 
1210 static void tsnep_rx_reuse_buffer_zc(struct tsnep_rx *rx, int index)
1211 {
1212 	struct tsnep_rx_entry *entry = &rx->entry[index];
1213 	struct tsnep_rx_entry *read = &rx->entry[rx->read];
1214 
1215 	tsnep_rx_set_xdp(rx, entry, read->xdp);
1216 	read->xdp = NULL;
1217 }
1218 
1219 static int tsnep_rx_alloc_zc(struct tsnep_rx *rx, int count, bool reuse)
1220 {
1221 	u32 allocated;
1222 	int i;
1223 
1224 	allocated = xsk_buff_alloc_batch(rx->xsk_pool, rx->xdp_batch, count);
1225 	for (i = 0; i < allocated; i++) {
1226 		int index = (rx->write + i) & TSNEP_RING_MASK;
1227 		struct tsnep_rx_entry *entry = &rx->entry[index];
1228 
1229 		tsnep_rx_set_xdp(rx, entry, rx->xdp_batch[i]);
1230 		tsnep_rx_activate(rx, index);
1231 	}
1232 	if (i == 0) {
1233 		rx->alloc_failed++;
1234 
1235 		if (reuse) {
1236 			tsnep_rx_reuse_buffer_zc(rx, rx->write);
1237 			tsnep_rx_activate(rx, rx->write);
1238 		}
1239 	}
1240 
1241 	if (i)
1242 		rx->write = (rx->write + i) & TSNEP_RING_MASK;
1243 
1244 	return i;
1245 }
1246 
1247 static void tsnep_rx_free_zc(struct tsnep_rx *rx)
1248 {
1249 	int i;
1250 
1251 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
1252 		struct tsnep_rx_entry *entry = &rx->entry[i];
1253 
1254 		if (entry->xdp)
1255 			xsk_buff_free(entry->xdp);
1256 		entry->xdp = NULL;
1257 	}
1258 }
1259 
1260 static int tsnep_rx_refill_zc(struct tsnep_rx *rx, int count, bool reuse)
1261 {
1262 	int desc_refilled;
1263 
1264 	desc_refilled = tsnep_rx_alloc_zc(rx, count, reuse);
1265 	if (desc_refilled)
1266 		tsnep_rx_enable(rx);
1267 
1268 	return desc_refilled;
1269 }
1270 
1271 static void tsnep_xsk_rx_need_wakeup(struct tsnep_rx *rx, int desc_available)
1272 {
1273 	if (desc_available)
1274 		xsk_set_rx_need_wakeup(rx->xsk_pool);
1275 	else
1276 		xsk_clear_rx_need_wakeup(rx->xsk_pool);
1277 }
1278 
1279 static bool tsnep_xdp_run_prog(struct tsnep_rx *rx, struct bpf_prog *prog,
1280 			       struct xdp_buff *xdp, int *status,
1281 			       struct netdev_queue *tx_nq, struct tsnep_tx *tx)
1282 {
1283 	unsigned int length;
1284 	unsigned int sync;
1285 	u32 act;
1286 
1287 	length = xdp->data_end - xdp->data_hard_start - XDP_PACKET_HEADROOM;
1288 
1289 	act = bpf_prog_run_xdp(prog, xdp);
1290 	switch (act) {
1291 	case XDP_PASS:
1292 		return false;
1293 	case XDP_TX:
1294 		if (!tsnep_xdp_xmit_back(rx->adapter, xdp, tx_nq, tx, false))
1295 			goto out_failure;
1296 		*status |= TSNEP_XDP_TX;
1297 		return true;
1298 	case XDP_REDIRECT:
1299 		if (xdp_do_redirect(rx->adapter->netdev, xdp, prog) < 0)
1300 			goto out_failure;
1301 		*status |= TSNEP_XDP_REDIRECT;
1302 		return true;
1303 	default:
1304 		bpf_warn_invalid_xdp_action(rx->adapter->netdev, prog, act);
1305 		fallthrough;
1306 	case XDP_ABORTED:
1307 out_failure:
1308 		trace_xdp_exception(rx->adapter->netdev, prog, act);
1309 		fallthrough;
1310 	case XDP_DROP:
1311 		/* Due xdp_adjust_tail: DMA sync for_device cover max len CPU
1312 		 * touch
1313 		 */
1314 		sync = xdp->data_end - xdp->data_hard_start -
1315 		       XDP_PACKET_HEADROOM;
1316 		sync = max(sync, length);
1317 		page_pool_put_page(rx->page_pool, virt_to_head_page(xdp->data),
1318 				   sync, true);
1319 		return true;
1320 	}
1321 }
1322 
1323 static bool tsnep_xdp_run_prog_zc(struct tsnep_rx *rx, struct bpf_prog *prog,
1324 				  struct xdp_buff *xdp, int *status,
1325 				  struct netdev_queue *tx_nq,
1326 				  struct tsnep_tx *tx)
1327 {
1328 	u32 act;
1329 
1330 	act = bpf_prog_run_xdp(prog, xdp);
1331 
1332 	/* XDP_REDIRECT is the main action for zero-copy */
1333 	if (likely(act == XDP_REDIRECT)) {
1334 		if (xdp_do_redirect(rx->adapter->netdev, xdp, prog) < 0)
1335 			goto out_failure;
1336 		*status |= TSNEP_XDP_REDIRECT;
1337 		return true;
1338 	}
1339 
1340 	switch (act) {
1341 	case XDP_PASS:
1342 		return false;
1343 	case XDP_TX:
1344 		if (!tsnep_xdp_xmit_back(rx->adapter, xdp, tx_nq, tx, true))
1345 			goto out_failure;
1346 		*status |= TSNEP_XDP_TX;
1347 		return true;
1348 	default:
1349 		bpf_warn_invalid_xdp_action(rx->adapter->netdev, prog, act);
1350 		fallthrough;
1351 	case XDP_ABORTED:
1352 out_failure:
1353 		trace_xdp_exception(rx->adapter->netdev, prog, act);
1354 		fallthrough;
1355 	case XDP_DROP:
1356 		xsk_buff_free(xdp);
1357 		return true;
1358 	}
1359 }
1360 
1361 static void tsnep_finalize_xdp(struct tsnep_adapter *adapter, int status,
1362 			       struct netdev_queue *tx_nq, struct tsnep_tx *tx)
1363 {
1364 	if (status & TSNEP_XDP_TX) {
1365 		__netif_tx_lock(tx_nq, smp_processor_id());
1366 		tsnep_xdp_xmit_flush(tx);
1367 		__netif_tx_unlock(tx_nq);
1368 	}
1369 
1370 	if (status & TSNEP_XDP_REDIRECT)
1371 		xdp_do_flush();
1372 }
1373 
1374 static struct sk_buff *tsnep_build_skb(struct tsnep_rx *rx, struct page *page,
1375 				       int length)
1376 {
1377 	struct sk_buff *skb;
1378 
1379 	skb = napi_build_skb(page_address(page), PAGE_SIZE);
1380 	if (unlikely(!skb))
1381 		return NULL;
1382 
1383 	/* update pointers within the skb to store the data */
1384 	skb_reserve(skb, TSNEP_RX_OFFSET + TSNEP_RX_INLINE_METADATA_SIZE);
1385 	__skb_put(skb, length - ETH_FCS_LEN);
1386 
1387 	if (rx->adapter->hwtstamp_config.rx_filter == HWTSTAMP_FILTER_ALL) {
1388 		struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
1389 		struct tsnep_rx_inline *rx_inline =
1390 			(struct tsnep_rx_inline *)(page_address(page) +
1391 						   TSNEP_RX_OFFSET);
1392 
1393 		skb_shinfo(skb)->tx_flags |=
1394 			SKBTX_HW_TSTAMP_NETDEV;
1395 		memset(hwtstamps, 0, sizeof(*hwtstamps));
1396 		hwtstamps->netdev_data = rx_inline;
1397 	}
1398 
1399 	skb_record_rx_queue(skb, rx->queue_index);
1400 	skb->protocol = eth_type_trans(skb, rx->adapter->netdev);
1401 
1402 	return skb;
1403 }
1404 
1405 static void tsnep_rx_page(struct tsnep_rx *rx, struct napi_struct *napi,
1406 			  struct page *page, int length)
1407 {
1408 	struct sk_buff *skb;
1409 
1410 	skb = tsnep_build_skb(rx, page, length);
1411 	if (skb) {
1412 		skb_mark_for_recycle(skb);
1413 
1414 		rx->packets++;
1415 		rx->bytes += length;
1416 		if (skb->pkt_type == PACKET_MULTICAST)
1417 			rx->multicast++;
1418 
1419 		napi_gro_receive(napi, skb);
1420 	} else {
1421 		page_pool_recycle_direct(rx->page_pool, page);
1422 
1423 		rx->dropped++;
1424 	}
1425 }
1426 
1427 static int tsnep_rx_poll(struct tsnep_rx *rx, struct napi_struct *napi,
1428 			 int budget)
1429 {
1430 	struct device *dmadev = rx->adapter->dmadev;
1431 	enum dma_data_direction dma_dir;
1432 	struct tsnep_rx_entry *entry;
1433 	struct netdev_queue *tx_nq;
1434 	struct bpf_prog *prog;
1435 	struct xdp_buff xdp;
1436 	struct tsnep_tx *tx;
1437 	int desc_available;
1438 	int xdp_status = 0;
1439 	int done = 0;
1440 	int length;
1441 
1442 	desc_available = tsnep_rx_desc_available(rx);
1443 	dma_dir = page_pool_get_dma_dir(rx->page_pool);
1444 	prog = READ_ONCE(rx->adapter->xdp_prog);
1445 	if (prog) {
1446 		tx_nq = netdev_get_tx_queue(rx->adapter->netdev,
1447 					    rx->tx_queue_index);
1448 		tx = &rx->adapter->tx[rx->tx_queue_index];
1449 
1450 		xdp_init_buff(&xdp, PAGE_SIZE, &rx->xdp_rxq);
1451 	}
1452 
1453 	while (likely(done < budget) && (rx->read != rx->write)) {
1454 		entry = &rx->entry[rx->read];
1455 		if ((__le32_to_cpu(entry->desc_wb->properties) &
1456 		     TSNEP_DESC_OWNER_COUNTER_MASK) !=
1457 		    (entry->properties & TSNEP_DESC_OWNER_COUNTER_MASK))
1458 			break;
1459 		done++;
1460 
1461 		if (desc_available >= TSNEP_RING_RX_REFILL) {
1462 			bool reuse = desc_available >= TSNEP_RING_RX_REUSE;
1463 
1464 			desc_available -= tsnep_rx_refill(rx, desc_available,
1465 							  reuse);
1466 			if (!entry->page) {
1467 				/* buffer has been reused for refill to prevent
1468 				 * empty RX ring, thus buffer cannot be used for
1469 				 * RX processing
1470 				 */
1471 				rx->read = (rx->read + 1) & TSNEP_RING_MASK;
1472 				desc_available++;
1473 
1474 				rx->dropped++;
1475 
1476 				continue;
1477 			}
1478 		}
1479 
1480 		/* descriptor properties shall be read first, because valid data
1481 		 * is signaled there
1482 		 */
1483 		dma_rmb();
1484 
1485 		prefetch(page_address(entry->page) + TSNEP_RX_OFFSET);
1486 		length = __le32_to_cpu(entry->desc_wb->properties) &
1487 			 TSNEP_DESC_LENGTH_MASK;
1488 		dma_sync_single_range_for_cpu(dmadev, entry->dma,
1489 					      TSNEP_RX_OFFSET, length, dma_dir);
1490 
1491 		/* RX metadata with timestamps is in front of actual data,
1492 		 * subtract metadata size to get length of actual data and
1493 		 * consider metadata size as offset of actual data during RX
1494 		 * processing
1495 		 */
1496 		length -= TSNEP_RX_INLINE_METADATA_SIZE;
1497 
1498 		rx->read = (rx->read + 1) & TSNEP_RING_MASK;
1499 		desc_available++;
1500 
1501 		if (prog) {
1502 			bool consume;
1503 
1504 			xdp_prepare_buff(&xdp, page_address(entry->page),
1505 					 XDP_PACKET_HEADROOM + TSNEP_RX_INLINE_METADATA_SIZE,
1506 					 length - ETH_FCS_LEN, false);
1507 
1508 			consume = tsnep_xdp_run_prog(rx, prog, &xdp,
1509 						     &xdp_status, tx_nq, tx);
1510 			if (consume) {
1511 				rx->packets++;
1512 				rx->bytes += length;
1513 
1514 				entry->page = NULL;
1515 
1516 				continue;
1517 			}
1518 		}
1519 
1520 		tsnep_rx_page(rx, napi, entry->page, length);
1521 		entry->page = NULL;
1522 	}
1523 
1524 	if (xdp_status)
1525 		tsnep_finalize_xdp(rx->adapter, xdp_status, tx_nq, tx);
1526 
1527 	if (desc_available)
1528 		tsnep_rx_refill(rx, desc_available, false);
1529 
1530 	return done;
1531 }
1532 
1533 static int tsnep_rx_poll_zc(struct tsnep_rx *rx, struct napi_struct *napi,
1534 			    int budget)
1535 {
1536 	struct tsnep_rx_entry *entry;
1537 	struct netdev_queue *tx_nq;
1538 	struct bpf_prog *prog;
1539 	struct tsnep_tx *tx;
1540 	int desc_available;
1541 	int xdp_status = 0;
1542 	struct page *page;
1543 	int done = 0;
1544 	int length;
1545 
1546 	desc_available = tsnep_rx_desc_available(rx);
1547 	prog = READ_ONCE(rx->adapter->xdp_prog);
1548 	if (prog) {
1549 		tx_nq = netdev_get_tx_queue(rx->adapter->netdev,
1550 					    rx->tx_queue_index);
1551 		tx = &rx->adapter->tx[rx->tx_queue_index];
1552 	}
1553 
1554 	while (likely(done < budget) && (rx->read != rx->write)) {
1555 		entry = &rx->entry[rx->read];
1556 		if ((__le32_to_cpu(entry->desc_wb->properties) &
1557 		     TSNEP_DESC_OWNER_COUNTER_MASK) !=
1558 		    (entry->properties & TSNEP_DESC_OWNER_COUNTER_MASK))
1559 			break;
1560 		done++;
1561 
1562 		if (desc_available >= TSNEP_RING_RX_REFILL) {
1563 			bool reuse = desc_available >= TSNEP_RING_RX_REUSE;
1564 
1565 			desc_available -= tsnep_rx_refill_zc(rx, desc_available,
1566 							     reuse);
1567 			if (!entry->xdp) {
1568 				/* buffer has been reused for refill to prevent
1569 				 * empty RX ring, thus buffer cannot be used for
1570 				 * RX processing
1571 				 */
1572 				rx->read = (rx->read + 1) & TSNEP_RING_MASK;
1573 				desc_available++;
1574 
1575 				rx->dropped++;
1576 
1577 				continue;
1578 			}
1579 		}
1580 
1581 		/* descriptor properties shall be read first, because valid data
1582 		 * is signaled there
1583 		 */
1584 		dma_rmb();
1585 
1586 		prefetch(entry->xdp->data);
1587 		length = __le32_to_cpu(entry->desc_wb->properties) &
1588 			 TSNEP_DESC_LENGTH_MASK;
1589 		xsk_buff_set_size(entry->xdp, length - ETH_FCS_LEN);
1590 		xsk_buff_dma_sync_for_cpu(entry->xdp, rx->xsk_pool);
1591 
1592 		/* RX metadata with timestamps is in front of actual data,
1593 		 * subtract metadata size to get length of actual data and
1594 		 * consider metadata size as offset of actual data during RX
1595 		 * processing
1596 		 */
1597 		length -= TSNEP_RX_INLINE_METADATA_SIZE;
1598 
1599 		rx->read = (rx->read + 1) & TSNEP_RING_MASK;
1600 		desc_available++;
1601 
1602 		if (prog) {
1603 			bool consume;
1604 
1605 			entry->xdp->data += TSNEP_RX_INLINE_METADATA_SIZE;
1606 			entry->xdp->data_meta += TSNEP_RX_INLINE_METADATA_SIZE;
1607 
1608 			consume = tsnep_xdp_run_prog_zc(rx, prog, entry->xdp,
1609 							&xdp_status, tx_nq, tx);
1610 			if (consume) {
1611 				rx->packets++;
1612 				rx->bytes += length;
1613 
1614 				entry->xdp = NULL;
1615 
1616 				continue;
1617 			}
1618 		}
1619 
1620 		page = page_pool_dev_alloc_pages(rx->page_pool);
1621 		if (page) {
1622 			memcpy(page_address(page) + TSNEP_RX_OFFSET,
1623 			       entry->xdp->data - TSNEP_RX_INLINE_METADATA_SIZE,
1624 			       length + TSNEP_RX_INLINE_METADATA_SIZE);
1625 			tsnep_rx_page(rx, napi, page, length);
1626 		} else {
1627 			rx->dropped++;
1628 		}
1629 		xsk_buff_free(entry->xdp);
1630 		entry->xdp = NULL;
1631 	}
1632 
1633 	if (xdp_status)
1634 		tsnep_finalize_xdp(rx->adapter, xdp_status, tx_nq, tx);
1635 
1636 	if (desc_available)
1637 		desc_available -= tsnep_rx_refill_zc(rx, desc_available, false);
1638 
1639 	if (xsk_uses_need_wakeup(rx->xsk_pool)) {
1640 		tsnep_xsk_rx_need_wakeup(rx, desc_available);
1641 
1642 		return done;
1643 	}
1644 
1645 	return desc_available ? budget : done;
1646 }
1647 
1648 static bool tsnep_rx_pending(struct tsnep_rx *rx)
1649 {
1650 	struct tsnep_rx_entry *entry;
1651 
1652 	if (rx->read != rx->write) {
1653 		entry = &rx->entry[rx->read];
1654 		if ((__le32_to_cpu(entry->desc_wb->properties) &
1655 		     TSNEP_DESC_OWNER_COUNTER_MASK) ==
1656 		    (entry->properties & TSNEP_DESC_OWNER_COUNTER_MASK))
1657 			return true;
1658 	}
1659 
1660 	return false;
1661 }
1662 
1663 static int tsnep_rx_open(struct tsnep_rx *rx)
1664 {
1665 	int desc_available;
1666 	int retval;
1667 
1668 	retval = tsnep_rx_ring_create(rx);
1669 	if (retval)
1670 		return retval;
1671 
1672 	tsnep_rx_init(rx);
1673 
1674 	desc_available = tsnep_rx_desc_available(rx);
1675 	if (rx->xsk_pool)
1676 		retval = tsnep_rx_alloc_zc(rx, desc_available, false);
1677 	else
1678 		retval = tsnep_rx_alloc(rx, desc_available, false);
1679 	if (retval != desc_available) {
1680 		retval = -ENOMEM;
1681 
1682 		goto alloc_failed;
1683 	}
1684 
1685 	/* prealloc pages to prevent allocation failures when XSK pool is
1686 	 * disabled at runtime
1687 	 */
1688 	if (rx->xsk_pool) {
1689 		retval = tsnep_rx_alloc_page_buffer(rx);
1690 		if (retval)
1691 			goto alloc_failed;
1692 	}
1693 
1694 	return 0;
1695 
1696 alloc_failed:
1697 	tsnep_rx_ring_cleanup(rx);
1698 	return retval;
1699 }
1700 
1701 static void tsnep_rx_close(struct tsnep_rx *rx)
1702 {
1703 	if (rx->xsk_pool)
1704 		tsnep_rx_free_page_buffer(rx);
1705 
1706 	tsnep_rx_ring_cleanup(rx);
1707 }
1708 
1709 static void tsnep_rx_reopen(struct tsnep_rx *rx)
1710 {
1711 	struct page **page = rx->page_buffer;
1712 	int i;
1713 
1714 	tsnep_rx_init(rx);
1715 
1716 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
1717 		struct tsnep_rx_entry *entry = &rx->entry[i];
1718 
1719 		/* defined initial values for properties are required for
1720 		 * correct owner counter checking
1721 		 */
1722 		entry->desc->properties = 0;
1723 		entry->desc_wb->properties = 0;
1724 
1725 		/* prevent allocation failures by reusing kept pages */
1726 		if (*page) {
1727 			tsnep_rx_set_page(rx, entry, *page);
1728 			tsnep_rx_activate(rx, rx->write);
1729 			rx->write++;
1730 
1731 			*page = NULL;
1732 			page++;
1733 		}
1734 	}
1735 }
1736 
1737 static void tsnep_rx_reopen_xsk(struct tsnep_rx *rx)
1738 {
1739 	struct page **page = rx->page_buffer;
1740 	u32 allocated;
1741 	int i;
1742 
1743 	tsnep_rx_init(rx);
1744 
1745 	/* alloc all ring entries except the last one, because ring cannot be
1746 	 * filled completely, as many buffers as possible is enough as wakeup is
1747 	 * done if new buffers are available
1748 	 */
1749 	allocated = xsk_buff_alloc_batch(rx->xsk_pool, rx->xdp_batch,
1750 					 TSNEP_RING_SIZE - 1);
1751 
1752 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
1753 		struct tsnep_rx_entry *entry = &rx->entry[i];
1754 
1755 		/* keep pages to prevent allocation failures when xsk is
1756 		 * disabled
1757 		 */
1758 		if (entry->page) {
1759 			*page = entry->page;
1760 			entry->page = NULL;
1761 
1762 			page++;
1763 		}
1764 
1765 		/* defined initial values for properties are required for
1766 		 * correct owner counter checking
1767 		 */
1768 		entry->desc->properties = 0;
1769 		entry->desc_wb->properties = 0;
1770 
1771 		if (allocated) {
1772 			tsnep_rx_set_xdp(rx, entry,
1773 					 rx->xdp_batch[allocated - 1]);
1774 			tsnep_rx_activate(rx, rx->write);
1775 			rx->write++;
1776 
1777 			allocated--;
1778 		}
1779 	}
1780 
1781 	/* set need wakeup flag immediately if ring is not filled completely,
1782 	 * first polling would be too late as need wakeup signalisation would
1783 	 * be delayed for an indefinite time
1784 	 */
1785 	if (xsk_uses_need_wakeup(rx->xsk_pool))
1786 		tsnep_xsk_rx_need_wakeup(rx, tsnep_rx_desc_available(rx));
1787 }
1788 
1789 static bool tsnep_pending(struct tsnep_queue *queue)
1790 {
1791 	if (queue->tx && tsnep_tx_pending(queue->tx))
1792 		return true;
1793 
1794 	if (queue->rx && tsnep_rx_pending(queue->rx))
1795 		return true;
1796 
1797 	return false;
1798 }
1799 
1800 static int tsnep_poll(struct napi_struct *napi, int budget)
1801 {
1802 	struct tsnep_queue *queue = container_of(napi, struct tsnep_queue,
1803 						 napi);
1804 	bool complete = true;
1805 	int done = 0;
1806 
1807 	if (queue->tx)
1808 		complete = tsnep_tx_poll(queue->tx, budget);
1809 
1810 	/* handle case where we are called by netpoll with a budget of 0 */
1811 	if (unlikely(budget <= 0))
1812 		return budget;
1813 
1814 	if (queue->rx) {
1815 		done = queue->rx->xsk_pool ?
1816 		       tsnep_rx_poll_zc(queue->rx, napi, budget) :
1817 		       tsnep_rx_poll(queue->rx, napi, budget);
1818 		if (done >= budget)
1819 			complete = false;
1820 	}
1821 
1822 	/* if all work not completed, return budget and keep polling */
1823 	if (!complete)
1824 		return budget;
1825 
1826 	if (likely(napi_complete_done(napi, done))) {
1827 		tsnep_enable_irq(queue->adapter, queue->irq_mask);
1828 
1829 		/* reschedule if work is already pending, prevent rotten packets
1830 		 * which are transmitted or received after polling but before
1831 		 * interrupt enable
1832 		 */
1833 		if (tsnep_pending(queue)) {
1834 			tsnep_disable_irq(queue->adapter, queue->irq_mask);
1835 			napi_schedule(napi);
1836 		}
1837 	}
1838 
1839 	return min(done, budget - 1);
1840 }
1841 
1842 static int tsnep_request_irq(struct tsnep_queue *queue, bool first)
1843 {
1844 	const char *name = netdev_name(queue->adapter->netdev);
1845 	irq_handler_t handler;
1846 	void *dev;
1847 	int retval;
1848 
1849 	if (first) {
1850 		sprintf(queue->name, "%s-mac", name);
1851 		handler = tsnep_irq;
1852 		dev = queue->adapter;
1853 	} else {
1854 		if (queue->tx && queue->rx)
1855 			snprintf(queue->name, sizeof(queue->name), "%s-txrx-%d",
1856 				 name, queue->rx->queue_index);
1857 		else if (queue->tx)
1858 			snprintf(queue->name, sizeof(queue->name), "%s-tx-%d",
1859 				 name, queue->tx->queue_index);
1860 		else
1861 			snprintf(queue->name, sizeof(queue->name), "%s-rx-%d",
1862 				 name, queue->rx->queue_index);
1863 		handler = tsnep_irq_txrx;
1864 		dev = queue;
1865 	}
1866 
1867 	retval = request_irq(queue->irq, handler, 0, queue->name, dev);
1868 	if (retval) {
1869 		/* if name is empty, then interrupt won't be freed */
1870 		memset(queue->name, 0, sizeof(queue->name));
1871 	}
1872 
1873 	return retval;
1874 }
1875 
1876 static void tsnep_free_irq(struct tsnep_queue *queue, bool first)
1877 {
1878 	void *dev;
1879 
1880 	if (!strlen(queue->name))
1881 		return;
1882 
1883 	if (first)
1884 		dev = queue->adapter;
1885 	else
1886 		dev = queue;
1887 
1888 	free_irq(queue->irq, dev);
1889 	memset(queue->name, 0, sizeof(queue->name));
1890 }
1891 
1892 static void tsnep_queue_close(struct tsnep_queue *queue, bool first)
1893 {
1894 	struct tsnep_rx *rx = queue->rx;
1895 
1896 	tsnep_free_irq(queue, first);
1897 
1898 	if (rx) {
1899 		if (xdp_rxq_info_is_reg(&rx->xdp_rxq))
1900 			xdp_rxq_info_unreg(&rx->xdp_rxq);
1901 		if (xdp_rxq_info_is_reg(&rx->xdp_rxq_zc))
1902 			xdp_rxq_info_unreg(&rx->xdp_rxq_zc);
1903 	}
1904 
1905 	netif_napi_del(&queue->napi);
1906 }
1907 
1908 static int tsnep_queue_open(struct tsnep_adapter *adapter,
1909 			    struct tsnep_queue *queue, bool first)
1910 {
1911 	struct tsnep_rx *rx = queue->rx;
1912 	struct tsnep_tx *tx = queue->tx;
1913 	int retval;
1914 
1915 	netif_napi_add(adapter->netdev, &queue->napi, tsnep_poll);
1916 
1917 	if (rx) {
1918 		/* choose TX queue for XDP_TX */
1919 		if (tx)
1920 			rx->tx_queue_index = tx->queue_index;
1921 		else if (rx->queue_index < adapter->num_tx_queues)
1922 			rx->tx_queue_index = rx->queue_index;
1923 		else
1924 			rx->tx_queue_index = 0;
1925 
1926 		/* prepare both memory models to eliminate possible registration
1927 		 * errors when memory model is switched between page pool and
1928 		 * XSK pool during runtime
1929 		 */
1930 		retval = xdp_rxq_info_reg(&rx->xdp_rxq, adapter->netdev,
1931 					  rx->queue_index, queue->napi.napi_id);
1932 		if (retval)
1933 			goto failed;
1934 		retval = xdp_rxq_info_reg_mem_model(&rx->xdp_rxq,
1935 						    MEM_TYPE_PAGE_POOL,
1936 						    rx->page_pool);
1937 		if (retval)
1938 			goto failed;
1939 		retval = xdp_rxq_info_reg(&rx->xdp_rxq_zc, adapter->netdev,
1940 					  rx->queue_index, queue->napi.napi_id);
1941 		if (retval)
1942 			goto failed;
1943 		retval = xdp_rxq_info_reg_mem_model(&rx->xdp_rxq_zc,
1944 						    MEM_TYPE_XSK_BUFF_POOL,
1945 						    NULL);
1946 		if (retval)
1947 			goto failed;
1948 		if (rx->xsk_pool)
1949 			xsk_pool_set_rxq_info(rx->xsk_pool, &rx->xdp_rxq_zc);
1950 	}
1951 
1952 	retval = tsnep_request_irq(queue, first);
1953 	if (retval) {
1954 		netif_err(adapter, drv, adapter->netdev,
1955 			  "can't get assigned irq %d.\n", queue->irq);
1956 		goto failed;
1957 	}
1958 
1959 	return 0;
1960 
1961 failed:
1962 	tsnep_queue_close(queue, first);
1963 
1964 	return retval;
1965 }
1966 
1967 static void tsnep_queue_enable(struct tsnep_queue *queue)
1968 {
1969 	napi_enable(&queue->napi);
1970 	tsnep_enable_irq(queue->adapter, queue->irq_mask);
1971 
1972 	if (queue->tx)
1973 		tsnep_tx_enable(queue->tx);
1974 
1975 	if (queue->rx)
1976 		tsnep_rx_enable(queue->rx);
1977 }
1978 
1979 static void tsnep_queue_disable(struct tsnep_queue *queue)
1980 {
1981 	if (queue->tx)
1982 		tsnep_tx_disable(queue->tx, &queue->napi);
1983 
1984 	napi_disable(&queue->napi);
1985 	tsnep_disable_irq(queue->adapter, queue->irq_mask);
1986 
1987 	/* disable RX after NAPI polling has been disabled, because RX can be
1988 	 * enabled during NAPI polling
1989 	 */
1990 	if (queue->rx)
1991 		tsnep_rx_disable(queue->rx);
1992 }
1993 
1994 static int tsnep_netdev_open(struct net_device *netdev)
1995 {
1996 	struct tsnep_adapter *adapter = netdev_priv(netdev);
1997 	int i, retval;
1998 
1999 	for (i = 0; i < adapter->num_queues; i++) {
2000 		if (adapter->queue[i].tx) {
2001 			retval = tsnep_tx_open(adapter->queue[i].tx);
2002 			if (retval)
2003 				goto failed;
2004 		}
2005 		if (adapter->queue[i].rx) {
2006 			retval = tsnep_rx_open(adapter->queue[i].rx);
2007 			if (retval)
2008 				goto failed;
2009 		}
2010 
2011 		retval = tsnep_queue_open(adapter, &adapter->queue[i], i == 0);
2012 		if (retval)
2013 			goto failed;
2014 	}
2015 
2016 	retval = netif_set_real_num_tx_queues(adapter->netdev,
2017 					      adapter->num_tx_queues);
2018 	if (retval)
2019 		goto failed;
2020 	retval = netif_set_real_num_rx_queues(adapter->netdev,
2021 					      adapter->num_rx_queues);
2022 	if (retval)
2023 		goto failed;
2024 
2025 	tsnep_enable_irq(adapter, ECM_INT_LINK);
2026 	retval = tsnep_phy_open(adapter);
2027 	if (retval)
2028 		goto phy_failed;
2029 
2030 	for (i = 0; i < adapter->num_queues; i++)
2031 		tsnep_queue_enable(&adapter->queue[i]);
2032 
2033 	return 0;
2034 
2035 phy_failed:
2036 	tsnep_disable_irq(adapter, ECM_INT_LINK);
2037 failed:
2038 	for (i = 0; i < adapter->num_queues; i++) {
2039 		tsnep_queue_close(&adapter->queue[i], i == 0);
2040 
2041 		if (adapter->queue[i].rx)
2042 			tsnep_rx_close(adapter->queue[i].rx);
2043 		if (adapter->queue[i].tx)
2044 			tsnep_tx_close(adapter->queue[i].tx);
2045 	}
2046 	return retval;
2047 }
2048 
2049 static int tsnep_netdev_close(struct net_device *netdev)
2050 {
2051 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2052 	int i;
2053 
2054 	tsnep_disable_irq(adapter, ECM_INT_LINK);
2055 	tsnep_phy_close(adapter);
2056 
2057 	for (i = 0; i < adapter->num_queues; i++) {
2058 		tsnep_queue_disable(&adapter->queue[i]);
2059 
2060 		tsnep_queue_close(&adapter->queue[i], i == 0);
2061 
2062 		if (adapter->queue[i].rx)
2063 			tsnep_rx_close(adapter->queue[i].rx);
2064 		if (adapter->queue[i].tx)
2065 			tsnep_tx_close(adapter->queue[i].tx);
2066 	}
2067 
2068 	return 0;
2069 }
2070 
2071 int tsnep_enable_xsk(struct tsnep_queue *queue, struct xsk_buff_pool *pool)
2072 {
2073 	bool running = netif_running(queue->adapter->netdev);
2074 	u32 frame_size;
2075 
2076 	frame_size = xsk_pool_get_rx_frame_size(pool);
2077 	if (frame_size < TSNEP_XSK_RX_BUF_SIZE)
2078 		return -EOPNOTSUPP;
2079 
2080 	queue->rx->page_buffer = kcalloc(TSNEP_RING_SIZE,
2081 					 sizeof(*queue->rx->page_buffer),
2082 					 GFP_KERNEL);
2083 	if (!queue->rx->page_buffer)
2084 		return -ENOMEM;
2085 	queue->rx->xdp_batch = kcalloc(TSNEP_RING_SIZE,
2086 				       sizeof(*queue->rx->xdp_batch),
2087 				       GFP_KERNEL);
2088 	if (!queue->rx->xdp_batch) {
2089 		kfree(queue->rx->page_buffer);
2090 		queue->rx->page_buffer = NULL;
2091 
2092 		return -ENOMEM;
2093 	}
2094 
2095 	xsk_pool_set_rxq_info(pool, &queue->rx->xdp_rxq_zc);
2096 
2097 	if (running)
2098 		tsnep_queue_disable(queue);
2099 
2100 	queue->tx->xsk_pool = pool;
2101 	queue->rx->xsk_pool = pool;
2102 
2103 	if (running) {
2104 		tsnep_rx_reopen_xsk(queue->rx);
2105 		tsnep_queue_enable(queue);
2106 	}
2107 
2108 	return 0;
2109 }
2110 
2111 void tsnep_disable_xsk(struct tsnep_queue *queue)
2112 {
2113 	bool running = netif_running(queue->adapter->netdev);
2114 
2115 	if (running)
2116 		tsnep_queue_disable(queue);
2117 
2118 	tsnep_rx_free_zc(queue->rx);
2119 
2120 	queue->rx->xsk_pool = NULL;
2121 	queue->tx->xsk_pool = NULL;
2122 
2123 	if (running) {
2124 		tsnep_rx_reopen(queue->rx);
2125 		tsnep_queue_enable(queue);
2126 	}
2127 
2128 	kfree(queue->rx->xdp_batch);
2129 	queue->rx->xdp_batch = NULL;
2130 	kfree(queue->rx->page_buffer);
2131 	queue->rx->page_buffer = NULL;
2132 }
2133 
2134 static netdev_tx_t tsnep_netdev_xmit_frame(struct sk_buff *skb,
2135 					   struct net_device *netdev)
2136 {
2137 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2138 	u16 queue_mapping = skb_get_queue_mapping(skb);
2139 
2140 	if (queue_mapping >= adapter->num_tx_queues)
2141 		queue_mapping = 0;
2142 
2143 	return tsnep_xmit_frame_ring(skb, &adapter->tx[queue_mapping]);
2144 }
2145 
2146 static int tsnep_netdev_ioctl(struct net_device *netdev, struct ifreq *ifr,
2147 			      int cmd)
2148 {
2149 	if (!netif_running(netdev))
2150 		return -EINVAL;
2151 	if (cmd == SIOCSHWTSTAMP || cmd == SIOCGHWTSTAMP)
2152 		return tsnep_ptp_ioctl(netdev, ifr, cmd);
2153 	return phy_mii_ioctl(netdev->phydev, ifr, cmd);
2154 }
2155 
2156 static void tsnep_netdev_set_multicast(struct net_device *netdev)
2157 {
2158 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2159 
2160 	u16 rx_filter = 0;
2161 
2162 	/* configured MAC address and broadcasts are never filtered */
2163 	if (netdev->flags & IFF_PROMISC) {
2164 		rx_filter |= TSNEP_RX_FILTER_ACCEPT_ALL_MULTICASTS;
2165 		rx_filter |= TSNEP_RX_FILTER_ACCEPT_ALL_UNICASTS;
2166 	} else if (!netdev_mc_empty(netdev) || (netdev->flags & IFF_ALLMULTI)) {
2167 		rx_filter |= TSNEP_RX_FILTER_ACCEPT_ALL_MULTICASTS;
2168 	}
2169 	iowrite16(rx_filter, adapter->addr + TSNEP_RX_FILTER);
2170 }
2171 
2172 static void tsnep_netdev_get_stats64(struct net_device *netdev,
2173 				     struct rtnl_link_stats64 *stats)
2174 {
2175 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2176 	u32 reg;
2177 	u32 val;
2178 	int i;
2179 
2180 	for (i = 0; i < adapter->num_tx_queues; i++) {
2181 		stats->tx_packets += adapter->tx[i].packets;
2182 		stats->tx_bytes += adapter->tx[i].bytes;
2183 		stats->tx_dropped += adapter->tx[i].dropped;
2184 	}
2185 	for (i = 0; i < adapter->num_rx_queues; i++) {
2186 		stats->rx_packets += adapter->rx[i].packets;
2187 		stats->rx_bytes += adapter->rx[i].bytes;
2188 		stats->rx_dropped += adapter->rx[i].dropped;
2189 		stats->multicast += adapter->rx[i].multicast;
2190 
2191 		reg = ioread32(adapter->addr + TSNEP_QUEUE(i) +
2192 			       TSNEP_RX_STATISTIC);
2193 		val = (reg & TSNEP_RX_STATISTIC_NO_DESC_MASK) >>
2194 		      TSNEP_RX_STATISTIC_NO_DESC_SHIFT;
2195 		stats->rx_dropped += val;
2196 		val = (reg & TSNEP_RX_STATISTIC_BUFFER_TOO_SMALL_MASK) >>
2197 		      TSNEP_RX_STATISTIC_BUFFER_TOO_SMALL_SHIFT;
2198 		stats->rx_dropped += val;
2199 		val = (reg & TSNEP_RX_STATISTIC_FIFO_OVERFLOW_MASK) >>
2200 		      TSNEP_RX_STATISTIC_FIFO_OVERFLOW_SHIFT;
2201 		stats->rx_errors += val;
2202 		stats->rx_fifo_errors += val;
2203 		val = (reg & TSNEP_RX_STATISTIC_INVALID_FRAME_MASK) >>
2204 		      TSNEP_RX_STATISTIC_INVALID_FRAME_SHIFT;
2205 		stats->rx_errors += val;
2206 		stats->rx_frame_errors += val;
2207 	}
2208 
2209 	reg = ioread32(adapter->addr + ECM_STAT);
2210 	val = (reg & ECM_STAT_RX_ERR_MASK) >> ECM_STAT_RX_ERR_SHIFT;
2211 	stats->rx_errors += val;
2212 	val = (reg & ECM_STAT_INV_FRM_MASK) >> ECM_STAT_INV_FRM_SHIFT;
2213 	stats->rx_errors += val;
2214 	stats->rx_crc_errors += val;
2215 	val = (reg & ECM_STAT_FWD_RX_ERR_MASK) >> ECM_STAT_FWD_RX_ERR_SHIFT;
2216 	stats->rx_errors += val;
2217 }
2218 
2219 static void tsnep_mac_set_address(struct tsnep_adapter *adapter, u8 *addr)
2220 {
2221 	iowrite32(*(u32 *)addr, adapter->addr + TSNEP_MAC_ADDRESS_LOW);
2222 	iowrite16(*(u16 *)(addr + sizeof(u32)),
2223 		  adapter->addr + TSNEP_MAC_ADDRESS_HIGH);
2224 
2225 	ether_addr_copy(adapter->mac_address, addr);
2226 	netif_info(adapter, drv, adapter->netdev, "MAC address set to %pM\n",
2227 		   addr);
2228 }
2229 
2230 static int tsnep_netdev_set_mac_address(struct net_device *netdev, void *addr)
2231 {
2232 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2233 	struct sockaddr *sock_addr = addr;
2234 	int retval;
2235 
2236 	retval = eth_prepare_mac_addr_change(netdev, sock_addr);
2237 	if (retval)
2238 		return retval;
2239 	eth_hw_addr_set(netdev, sock_addr->sa_data);
2240 	tsnep_mac_set_address(adapter, sock_addr->sa_data);
2241 
2242 	return 0;
2243 }
2244 
2245 static int tsnep_netdev_set_features(struct net_device *netdev,
2246 				     netdev_features_t features)
2247 {
2248 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2249 	netdev_features_t changed = netdev->features ^ features;
2250 	bool enable;
2251 	int retval = 0;
2252 
2253 	if (changed & NETIF_F_LOOPBACK) {
2254 		enable = !!(features & NETIF_F_LOOPBACK);
2255 		retval = tsnep_phy_loopback(adapter, enable);
2256 	}
2257 
2258 	return retval;
2259 }
2260 
2261 static ktime_t tsnep_netdev_get_tstamp(struct net_device *netdev,
2262 				       const struct skb_shared_hwtstamps *hwtstamps,
2263 				       bool cycles)
2264 {
2265 	struct tsnep_rx_inline *rx_inline = hwtstamps->netdev_data;
2266 	u64 timestamp;
2267 
2268 	if (cycles)
2269 		timestamp = __le64_to_cpu(rx_inline->counter);
2270 	else
2271 		timestamp = __le64_to_cpu(rx_inline->timestamp);
2272 
2273 	return ns_to_ktime(timestamp);
2274 }
2275 
2276 static int tsnep_netdev_bpf(struct net_device *dev, struct netdev_bpf *bpf)
2277 {
2278 	struct tsnep_adapter *adapter = netdev_priv(dev);
2279 
2280 	switch (bpf->command) {
2281 	case XDP_SETUP_PROG:
2282 		return tsnep_xdp_setup_prog(adapter, bpf->prog, bpf->extack);
2283 	case XDP_SETUP_XSK_POOL:
2284 		return tsnep_xdp_setup_pool(adapter, bpf->xsk.pool,
2285 					    bpf->xsk.queue_id);
2286 	default:
2287 		return -EOPNOTSUPP;
2288 	}
2289 }
2290 
2291 static struct tsnep_tx *tsnep_xdp_get_tx(struct tsnep_adapter *adapter, u32 cpu)
2292 {
2293 	if (cpu >= TSNEP_MAX_QUEUES)
2294 		cpu &= TSNEP_MAX_QUEUES - 1;
2295 
2296 	while (cpu >= adapter->num_tx_queues)
2297 		cpu -= adapter->num_tx_queues;
2298 
2299 	return &adapter->tx[cpu];
2300 }
2301 
2302 static int tsnep_netdev_xdp_xmit(struct net_device *dev, int n,
2303 				 struct xdp_frame **xdp, u32 flags)
2304 {
2305 	struct tsnep_adapter *adapter = netdev_priv(dev);
2306 	u32 cpu = smp_processor_id();
2307 	struct netdev_queue *nq;
2308 	struct tsnep_tx *tx;
2309 	int nxmit;
2310 	bool xmit;
2311 
2312 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2313 		return -EINVAL;
2314 
2315 	tx = tsnep_xdp_get_tx(adapter, cpu);
2316 	nq = netdev_get_tx_queue(adapter->netdev, tx->queue_index);
2317 
2318 	__netif_tx_lock(nq, cpu);
2319 
2320 	for (nxmit = 0; nxmit < n; nxmit++) {
2321 		xmit = tsnep_xdp_xmit_frame_ring(xdp[nxmit], tx,
2322 						 TSNEP_TX_TYPE_XDP_NDO);
2323 		if (!xmit)
2324 			break;
2325 
2326 		/* avoid transmit queue timeout since we share it with the slow
2327 		 * path
2328 		 */
2329 		txq_trans_cond_update(nq);
2330 	}
2331 
2332 	if (flags & XDP_XMIT_FLUSH)
2333 		tsnep_xdp_xmit_flush(tx);
2334 
2335 	__netif_tx_unlock(nq);
2336 
2337 	return nxmit;
2338 }
2339 
2340 static int tsnep_netdev_xsk_wakeup(struct net_device *dev, u32 queue_id,
2341 				   u32 flags)
2342 {
2343 	struct tsnep_adapter *adapter = netdev_priv(dev);
2344 	struct tsnep_queue *queue;
2345 
2346 	if (queue_id >= adapter->num_rx_queues ||
2347 	    queue_id >= adapter->num_tx_queues)
2348 		return -EINVAL;
2349 
2350 	queue = &adapter->queue[queue_id];
2351 
2352 	if (!napi_if_scheduled_mark_missed(&queue->napi))
2353 		napi_schedule(&queue->napi);
2354 
2355 	return 0;
2356 }
2357 
2358 static const struct net_device_ops tsnep_netdev_ops = {
2359 	.ndo_open = tsnep_netdev_open,
2360 	.ndo_stop = tsnep_netdev_close,
2361 	.ndo_start_xmit = tsnep_netdev_xmit_frame,
2362 	.ndo_eth_ioctl = tsnep_netdev_ioctl,
2363 	.ndo_set_rx_mode = tsnep_netdev_set_multicast,
2364 	.ndo_get_stats64 = tsnep_netdev_get_stats64,
2365 	.ndo_set_mac_address = tsnep_netdev_set_mac_address,
2366 	.ndo_set_features = tsnep_netdev_set_features,
2367 	.ndo_get_tstamp = tsnep_netdev_get_tstamp,
2368 	.ndo_setup_tc = tsnep_tc_setup,
2369 	.ndo_bpf = tsnep_netdev_bpf,
2370 	.ndo_xdp_xmit = tsnep_netdev_xdp_xmit,
2371 	.ndo_xsk_wakeup = tsnep_netdev_xsk_wakeup,
2372 };
2373 
2374 static int tsnep_mac_init(struct tsnep_adapter *adapter)
2375 {
2376 	int retval;
2377 
2378 	/* initialize RX filtering, at least configured MAC address and
2379 	 * broadcast are not filtered
2380 	 */
2381 	iowrite16(0, adapter->addr + TSNEP_RX_FILTER);
2382 
2383 	/* try to get MAC address in the following order:
2384 	 * - device tree
2385 	 * - valid MAC address already set
2386 	 * - MAC address register if valid
2387 	 * - random MAC address
2388 	 */
2389 	retval = of_get_mac_address(adapter->pdev->dev.of_node,
2390 				    adapter->mac_address);
2391 	if (retval == -EPROBE_DEFER)
2392 		return retval;
2393 	if (retval && !is_valid_ether_addr(adapter->mac_address)) {
2394 		*(u32 *)adapter->mac_address =
2395 			ioread32(adapter->addr + TSNEP_MAC_ADDRESS_LOW);
2396 		*(u16 *)(adapter->mac_address + sizeof(u32)) =
2397 			ioread16(adapter->addr + TSNEP_MAC_ADDRESS_HIGH);
2398 		if (!is_valid_ether_addr(adapter->mac_address))
2399 			eth_random_addr(adapter->mac_address);
2400 	}
2401 
2402 	tsnep_mac_set_address(adapter, adapter->mac_address);
2403 	eth_hw_addr_set(adapter->netdev, adapter->mac_address);
2404 
2405 	return 0;
2406 }
2407 
2408 static int tsnep_mdio_init(struct tsnep_adapter *adapter)
2409 {
2410 	struct device_node *np = adapter->pdev->dev.of_node;
2411 	int retval;
2412 
2413 	if (np) {
2414 		np = of_get_child_by_name(np, "mdio");
2415 		if (!np)
2416 			return 0;
2417 
2418 		adapter->suppress_preamble =
2419 			of_property_read_bool(np, "suppress-preamble");
2420 	}
2421 
2422 	adapter->mdiobus = devm_mdiobus_alloc(&adapter->pdev->dev);
2423 	if (!adapter->mdiobus) {
2424 		retval = -ENOMEM;
2425 
2426 		goto out;
2427 	}
2428 
2429 	adapter->mdiobus->priv = (void *)adapter;
2430 	adapter->mdiobus->parent = &adapter->pdev->dev;
2431 	adapter->mdiobus->read = tsnep_mdiobus_read;
2432 	adapter->mdiobus->write = tsnep_mdiobus_write;
2433 	adapter->mdiobus->name = TSNEP "-mdiobus";
2434 	snprintf(adapter->mdiobus->id, MII_BUS_ID_SIZE, "%s",
2435 		 adapter->pdev->name);
2436 
2437 	/* do not scan broadcast address */
2438 	adapter->mdiobus->phy_mask = 0x0000001;
2439 
2440 	retval = of_mdiobus_register(adapter->mdiobus, np);
2441 
2442 out:
2443 	of_node_put(np);
2444 
2445 	return retval;
2446 }
2447 
2448 static int tsnep_phy_init(struct tsnep_adapter *adapter)
2449 {
2450 	struct device_node *phy_node;
2451 	int retval;
2452 
2453 	retval = of_get_phy_mode(adapter->pdev->dev.of_node,
2454 				 &adapter->phy_mode);
2455 	if (retval)
2456 		adapter->phy_mode = PHY_INTERFACE_MODE_GMII;
2457 
2458 	phy_node = of_parse_phandle(adapter->pdev->dev.of_node, "phy-handle",
2459 				    0);
2460 	adapter->phydev = of_phy_find_device(phy_node);
2461 	of_node_put(phy_node);
2462 	if (!adapter->phydev && adapter->mdiobus)
2463 		adapter->phydev = phy_find_first(adapter->mdiobus);
2464 	if (!adapter->phydev)
2465 		return -EIO;
2466 
2467 	return 0;
2468 }
2469 
2470 static int tsnep_queue_init(struct tsnep_adapter *adapter, int queue_count)
2471 {
2472 	u32 irq_mask = ECM_INT_TX_0 | ECM_INT_RX_0;
2473 	char name[8];
2474 	int i;
2475 	int retval;
2476 
2477 	/* one TX/RX queue pair for netdev is mandatory */
2478 	if (platform_irq_count(adapter->pdev) == 1)
2479 		retval = platform_get_irq(adapter->pdev, 0);
2480 	else
2481 		retval = platform_get_irq_byname(adapter->pdev, "mac");
2482 	if (retval < 0)
2483 		return retval;
2484 	adapter->num_tx_queues = 1;
2485 	adapter->num_rx_queues = 1;
2486 	adapter->num_queues = 1;
2487 	adapter->queue[0].adapter = adapter;
2488 	adapter->queue[0].irq = retval;
2489 	adapter->queue[0].tx = &adapter->tx[0];
2490 	adapter->queue[0].tx->adapter = adapter;
2491 	adapter->queue[0].tx->addr = adapter->addr + TSNEP_QUEUE(0);
2492 	adapter->queue[0].tx->queue_index = 0;
2493 	adapter->queue[0].rx = &adapter->rx[0];
2494 	adapter->queue[0].rx->adapter = adapter;
2495 	adapter->queue[0].rx->addr = adapter->addr + TSNEP_QUEUE(0);
2496 	adapter->queue[0].rx->queue_index = 0;
2497 	adapter->queue[0].irq_mask = irq_mask;
2498 	adapter->queue[0].irq_delay_addr = adapter->addr + ECM_INT_DELAY;
2499 	retval = tsnep_set_irq_coalesce(&adapter->queue[0],
2500 					TSNEP_COALESCE_USECS_DEFAULT);
2501 	if (retval < 0)
2502 		return retval;
2503 
2504 	adapter->netdev->irq = adapter->queue[0].irq;
2505 
2506 	/* add additional TX/RX queue pairs only if dedicated interrupt is
2507 	 * available
2508 	 */
2509 	for (i = 1; i < queue_count; i++) {
2510 		sprintf(name, "txrx-%d", i);
2511 		retval = platform_get_irq_byname_optional(adapter->pdev, name);
2512 		if (retval < 0)
2513 			break;
2514 
2515 		adapter->num_tx_queues++;
2516 		adapter->num_rx_queues++;
2517 		adapter->num_queues++;
2518 		adapter->queue[i].adapter = adapter;
2519 		adapter->queue[i].irq = retval;
2520 		adapter->queue[i].tx = &adapter->tx[i];
2521 		adapter->queue[i].tx->adapter = adapter;
2522 		adapter->queue[i].tx->addr = adapter->addr + TSNEP_QUEUE(i);
2523 		adapter->queue[i].tx->queue_index = i;
2524 		adapter->queue[i].rx = &adapter->rx[i];
2525 		adapter->queue[i].rx->adapter = adapter;
2526 		adapter->queue[i].rx->addr = adapter->addr + TSNEP_QUEUE(i);
2527 		adapter->queue[i].rx->queue_index = i;
2528 		adapter->queue[i].irq_mask =
2529 			irq_mask << (ECM_INT_TXRX_SHIFT * i);
2530 		adapter->queue[i].irq_delay_addr =
2531 			adapter->addr + ECM_INT_DELAY + ECM_INT_DELAY_OFFSET * i;
2532 		retval = tsnep_set_irq_coalesce(&adapter->queue[i],
2533 						TSNEP_COALESCE_USECS_DEFAULT);
2534 		if (retval < 0)
2535 			return retval;
2536 	}
2537 
2538 	return 0;
2539 }
2540 
2541 static int tsnep_probe(struct platform_device *pdev)
2542 {
2543 	struct tsnep_adapter *adapter;
2544 	struct net_device *netdev;
2545 	struct resource *io;
2546 	u32 type;
2547 	int revision;
2548 	int version;
2549 	int queue_count;
2550 	int retval;
2551 
2552 	netdev = devm_alloc_etherdev_mqs(&pdev->dev,
2553 					 sizeof(struct tsnep_adapter),
2554 					 TSNEP_MAX_QUEUES, TSNEP_MAX_QUEUES);
2555 	if (!netdev)
2556 		return -ENODEV;
2557 	SET_NETDEV_DEV(netdev, &pdev->dev);
2558 	adapter = netdev_priv(netdev);
2559 	platform_set_drvdata(pdev, adapter);
2560 	adapter->pdev = pdev;
2561 	adapter->dmadev = &pdev->dev;
2562 	adapter->netdev = netdev;
2563 	adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE |
2564 			      NETIF_MSG_LINK | NETIF_MSG_IFUP |
2565 			      NETIF_MSG_IFDOWN | NETIF_MSG_TX_QUEUED;
2566 
2567 	netdev->min_mtu = ETH_MIN_MTU;
2568 	netdev->max_mtu = TSNEP_MAX_FRAME_SIZE;
2569 
2570 	mutex_init(&adapter->gate_control_lock);
2571 	mutex_init(&adapter->rxnfc_lock);
2572 	INIT_LIST_HEAD(&adapter->rxnfc_rules);
2573 
2574 	adapter->addr = devm_platform_get_and_ioremap_resource(pdev, 0, &io);
2575 	if (IS_ERR(adapter->addr))
2576 		return PTR_ERR(adapter->addr);
2577 	netdev->mem_start = io->start;
2578 	netdev->mem_end = io->end;
2579 
2580 	type = ioread32(adapter->addr + ECM_TYPE);
2581 	revision = (type & ECM_REVISION_MASK) >> ECM_REVISION_SHIFT;
2582 	version = (type & ECM_VERSION_MASK) >> ECM_VERSION_SHIFT;
2583 	queue_count = (type & ECM_QUEUE_COUNT_MASK) >> ECM_QUEUE_COUNT_SHIFT;
2584 	adapter->gate_control = type & ECM_GATE_CONTROL;
2585 	adapter->rxnfc_max = TSNEP_RX_ASSIGN_ETHER_TYPE_COUNT;
2586 
2587 	tsnep_disable_irq(adapter, ECM_INT_ALL);
2588 
2589 	retval = tsnep_queue_init(adapter, queue_count);
2590 	if (retval)
2591 		return retval;
2592 
2593 	retval = dma_set_mask_and_coherent(&adapter->pdev->dev,
2594 					   DMA_BIT_MASK(64));
2595 	if (retval) {
2596 		dev_err(&adapter->pdev->dev, "no usable DMA configuration.\n");
2597 		return retval;
2598 	}
2599 
2600 	retval = tsnep_mac_init(adapter);
2601 	if (retval)
2602 		return retval;
2603 
2604 	retval = tsnep_mdio_init(adapter);
2605 	if (retval)
2606 		goto mdio_init_failed;
2607 
2608 	retval = tsnep_phy_init(adapter);
2609 	if (retval)
2610 		goto phy_init_failed;
2611 
2612 	retval = tsnep_ptp_init(adapter);
2613 	if (retval)
2614 		goto ptp_init_failed;
2615 
2616 	retval = tsnep_tc_init(adapter);
2617 	if (retval)
2618 		goto tc_init_failed;
2619 
2620 	retval = tsnep_rxnfc_init(adapter);
2621 	if (retval)
2622 		goto rxnfc_init_failed;
2623 
2624 	netdev->netdev_ops = &tsnep_netdev_ops;
2625 	netdev->ethtool_ops = &tsnep_ethtool_ops;
2626 	netdev->features = NETIF_F_SG;
2627 	netdev->hw_features = netdev->features | NETIF_F_LOOPBACK;
2628 
2629 	netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
2630 			       NETDEV_XDP_ACT_NDO_XMIT |
2631 			       NETDEV_XDP_ACT_NDO_XMIT_SG |
2632 			       NETDEV_XDP_ACT_XSK_ZEROCOPY;
2633 
2634 	/* carrier off reporting is important to ethtool even BEFORE open */
2635 	netif_carrier_off(netdev);
2636 
2637 	retval = register_netdev(netdev);
2638 	if (retval)
2639 		goto register_failed;
2640 
2641 	dev_info(&adapter->pdev->dev, "device version %d.%02d\n", version,
2642 		 revision);
2643 	if (adapter->gate_control)
2644 		dev_info(&adapter->pdev->dev, "gate control detected\n");
2645 
2646 	return 0;
2647 
2648 register_failed:
2649 	tsnep_rxnfc_cleanup(adapter);
2650 rxnfc_init_failed:
2651 	tsnep_tc_cleanup(adapter);
2652 tc_init_failed:
2653 	tsnep_ptp_cleanup(adapter);
2654 ptp_init_failed:
2655 phy_init_failed:
2656 	if (adapter->mdiobus)
2657 		mdiobus_unregister(adapter->mdiobus);
2658 mdio_init_failed:
2659 	return retval;
2660 }
2661 
2662 static void tsnep_remove(struct platform_device *pdev)
2663 {
2664 	struct tsnep_adapter *adapter = platform_get_drvdata(pdev);
2665 
2666 	unregister_netdev(adapter->netdev);
2667 
2668 	tsnep_rxnfc_cleanup(adapter);
2669 
2670 	tsnep_tc_cleanup(adapter);
2671 
2672 	tsnep_ptp_cleanup(adapter);
2673 
2674 	if (adapter->mdiobus)
2675 		mdiobus_unregister(adapter->mdiobus);
2676 
2677 	tsnep_disable_irq(adapter, ECM_INT_ALL);
2678 }
2679 
2680 static const struct of_device_id tsnep_of_match[] = {
2681 	{ .compatible = "engleder,tsnep", },
2682 { },
2683 };
2684 MODULE_DEVICE_TABLE(of, tsnep_of_match);
2685 
2686 static struct platform_driver tsnep_driver = {
2687 	.driver = {
2688 		.name = TSNEP,
2689 		.of_match_table = tsnep_of_match,
2690 	},
2691 	.probe = tsnep_probe,
2692 	.remove_new = tsnep_remove,
2693 };
2694 module_platform_driver(tsnep_driver);
2695 
2696 MODULE_AUTHOR("Gerhard Engleder <gerhard@engleder-embedded.com>");
2697 MODULE_DESCRIPTION("TSN endpoint Ethernet MAC driver");
2698 MODULE_LICENSE("GPL");
2699