xref: /linux/drivers/net/ethernet/engleder/tsnep_main.c (revision 119ff04864a24470b1e531bb53e5c141aa8fefb0)
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 		tsnep_set_link_mode(adapter);
234 
235 	return retval;
236 }
237 
238 static int tsnep_phy_open(struct tsnep_adapter *adapter)
239 {
240 	struct phy_device *phydev;
241 	struct ethtool_eee ethtool_eee;
242 	int retval;
243 
244 	retval = phy_connect_direct(adapter->netdev, adapter->phydev,
245 				    tsnep_phy_link_status_change,
246 				    adapter->phy_mode);
247 	if (retval)
248 		return retval;
249 	phydev = adapter->netdev->phydev;
250 
251 	/* MAC supports only 100Mbps|1000Mbps full duplex
252 	 * SPE (Single Pair Ethernet) is also an option but not implemented yet
253 	 */
254 	phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_10baseT_Half_BIT);
255 	phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_10baseT_Full_BIT);
256 	phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_100baseT_Half_BIT);
257 	phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_1000baseT_Half_BIT);
258 
259 	/* disable EEE autoneg, EEE not supported by TSNEP */
260 	memset(&ethtool_eee, 0, sizeof(ethtool_eee));
261 	phy_ethtool_set_eee(adapter->phydev, &ethtool_eee);
262 
263 	adapter->phydev->irq = PHY_MAC_INTERRUPT;
264 	phy_start(adapter->phydev);
265 
266 	return 0;
267 }
268 
269 static void tsnep_phy_close(struct tsnep_adapter *adapter)
270 {
271 	phy_stop(adapter->netdev->phydev);
272 	phy_disconnect(adapter->netdev->phydev);
273 }
274 
275 static void tsnep_tx_ring_cleanup(struct tsnep_tx *tx)
276 {
277 	struct device *dmadev = tx->adapter->dmadev;
278 	int i;
279 
280 	memset(tx->entry, 0, sizeof(tx->entry));
281 
282 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
283 		if (tx->page[i]) {
284 			dma_free_coherent(dmadev, PAGE_SIZE, tx->page[i],
285 					  tx->page_dma[i]);
286 			tx->page[i] = NULL;
287 			tx->page_dma[i] = 0;
288 		}
289 	}
290 }
291 
292 static int tsnep_tx_ring_create(struct tsnep_tx *tx)
293 {
294 	struct device *dmadev = tx->adapter->dmadev;
295 	struct tsnep_tx_entry *entry;
296 	struct tsnep_tx_entry *next_entry;
297 	int i, j;
298 	int retval;
299 
300 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
301 		tx->page[i] =
302 			dma_alloc_coherent(dmadev, PAGE_SIZE, &tx->page_dma[i],
303 					   GFP_KERNEL);
304 		if (!tx->page[i]) {
305 			retval = -ENOMEM;
306 			goto alloc_failed;
307 		}
308 		for (j = 0; j < TSNEP_RING_ENTRIES_PER_PAGE; j++) {
309 			entry = &tx->entry[TSNEP_RING_ENTRIES_PER_PAGE * i + j];
310 			entry->desc_wb = (struct tsnep_tx_desc_wb *)
311 				(((u8 *)tx->page[i]) + TSNEP_DESC_SIZE * j);
312 			entry->desc = (struct tsnep_tx_desc *)
313 				(((u8 *)entry->desc_wb) + TSNEP_DESC_OFFSET);
314 			entry->desc_dma = tx->page_dma[i] + TSNEP_DESC_SIZE * j;
315 			entry->owner_user_flag = false;
316 		}
317 	}
318 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
319 		entry = &tx->entry[i];
320 		next_entry = &tx->entry[(i + 1) & TSNEP_RING_MASK];
321 		entry->desc->next = __cpu_to_le64(next_entry->desc_dma);
322 	}
323 
324 	return 0;
325 
326 alloc_failed:
327 	tsnep_tx_ring_cleanup(tx);
328 	return retval;
329 }
330 
331 static void tsnep_tx_init(struct tsnep_tx *tx)
332 {
333 	dma_addr_t dma;
334 
335 	dma = tx->entry[0].desc_dma | TSNEP_RESET_OWNER_COUNTER;
336 	iowrite32(DMA_ADDR_LOW(dma), tx->addr + TSNEP_TX_DESC_ADDR_LOW);
337 	iowrite32(DMA_ADDR_HIGH(dma), tx->addr + TSNEP_TX_DESC_ADDR_HIGH);
338 	tx->write = 0;
339 	tx->read = 0;
340 	tx->owner_counter = 1;
341 	tx->increment_owner_counter = TSNEP_RING_SIZE - 1;
342 }
343 
344 static void tsnep_tx_enable(struct tsnep_tx *tx)
345 {
346 	struct netdev_queue *nq;
347 
348 	nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
349 
350 	__netif_tx_lock_bh(nq);
351 	netif_tx_wake_queue(nq);
352 	__netif_tx_unlock_bh(nq);
353 }
354 
355 static void tsnep_tx_disable(struct tsnep_tx *tx, struct napi_struct *napi)
356 {
357 	struct netdev_queue *nq;
358 	u32 val;
359 
360 	nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
361 
362 	__netif_tx_lock_bh(nq);
363 	netif_tx_stop_queue(nq);
364 	__netif_tx_unlock_bh(nq);
365 
366 	/* wait until TX is done in hardware */
367 	readx_poll_timeout(ioread32, tx->addr + TSNEP_CONTROL, val,
368 			   ((val & TSNEP_CONTROL_TX_ENABLE) == 0), 10000,
369 			   1000000);
370 
371 	/* wait until TX is also done in software */
372 	while (READ_ONCE(tx->read) != tx->write) {
373 		napi_schedule(napi);
374 		napi_synchronize(napi);
375 	}
376 }
377 
378 static void tsnep_tx_activate(struct tsnep_tx *tx, int index, int length,
379 			      bool last)
380 {
381 	struct tsnep_tx_entry *entry = &tx->entry[index];
382 
383 	entry->properties = 0;
384 	/* xdpf and zc are union with skb */
385 	if (entry->skb) {
386 		entry->properties = length & TSNEP_DESC_LENGTH_MASK;
387 		entry->properties |= TSNEP_DESC_INTERRUPT_FLAG;
388 		if ((entry->type & TSNEP_TX_TYPE_SKB) &&
389 		    (skb_shinfo(entry->skb)->tx_flags & SKBTX_IN_PROGRESS))
390 			entry->properties |= TSNEP_DESC_EXTENDED_WRITEBACK_FLAG;
391 
392 		/* toggle user flag to prevent false acknowledge
393 		 *
394 		 * Only the first fragment is acknowledged. For all other
395 		 * fragments no acknowledge is done and the last written owner
396 		 * counter stays in the writeback descriptor. Therefore, it is
397 		 * possible that the last written owner counter is identical to
398 		 * the new incremented owner counter and a false acknowledge is
399 		 * detected before the real acknowledge has been done by
400 		 * hardware.
401 		 *
402 		 * The user flag is used to prevent this situation. The user
403 		 * flag is copied to the writeback descriptor by the hardware
404 		 * and is used as additional acknowledge data. By toggeling the
405 		 * user flag only for the first fragment (which is
406 		 * acknowledged), it is guaranteed that the last acknowledge
407 		 * done for this descriptor has used a different user flag and
408 		 * cannot be detected as false acknowledge.
409 		 */
410 		entry->owner_user_flag = !entry->owner_user_flag;
411 	}
412 	if (last)
413 		entry->properties |= TSNEP_TX_DESC_LAST_FRAGMENT_FLAG;
414 	if (index == tx->increment_owner_counter) {
415 		tx->owner_counter++;
416 		if (tx->owner_counter == 4)
417 			tx->owner_counter = 1;
418 		tx->increment_owner_counter--;
419 		if (tx->increment_owner_counter < 0)
420 			tx->increment_owner_counter = TSNEP_RING_SIZE - 1;
421 	}
422 	entry->properties |=
423 		(tx->owner_counter << TSNEP_DESC_OWNER_COUNTER_SHIFT) &
424 		TSNEP_DESC_OWNER_COUNTER_MASK;
425 	if (entry->owner_user_flag)
426 		entry->properties |= TSNEP_TX_DESC_OWNER_USER_FLAG;
427 	entry->desc->more_properties =
428 		__cpu_to_le32(entry->len & TSNEP_DESC_LENGTH_MASK);
429 	if (entry->type & TSNEP_TX_TYPE_INLINE)
430 		entry->properties |= TSNEP_TX_DESC_DATA_AFTER_DESC_FLAG;
431 
432 	/* descriptor properties shall be written last, because valid data is
433 	 * signaled there
434 	 */
435 	dma_wmb();
436 
437 	entry->desc->properties = __cpu_to_le32(entry->properties);
438 }
439 
440 static int tsnep_tx_desc_available(struct tsnep_tx *tx)
441 {
442 	if (tx->read <= tx->write)
443 		return TSNEP_RING_SIZE - tx->write + tx->read - 1;
444 	else
445 		return tx->read - tx->write - 1;
446 }
447 
448 static int tsnep_tx_map_frag(skb_frag_t *frag, struct tsnep_tx_entry *entry,
449 			     struct device *dmadev, dma_addr_t *dma)
450 {
451 	unsigned int len;
452 	int mapped;
453 
454 	len = skb_frag_size(frag);
455 	if (likely(len > TSNEP_DESC_SIZE_DATA_AFTER_INLINE)) {
456 		*dma = skb_frag_dma_map(dmadev, frag, 0, len, DMA_TO_DEVICE);
457 		if (dma_mapping_error(dmadev, *dma))
458 			return -ENOMEM;
459 		entry->type = TSNEP_TX_TYPE_SKB_FRAG_MAP_PAGE;
460 		mapped = 1;
461 	} else {
462 		void *fragdata = skb_frag_address_safe(frag);
463 
464 		if (likely(fragdata)) {
465 			memcpy(&entry->desc->tx, fragdata, len);
466 		} else {
467 			struct page *page = skb_frag_page(frag);
468 
469 			fragdata = kmap_local_page(page);
470 			memcpy(&entry->desc->tx, fragdata + skb_frag_off(frag),
471 			       len);
472 			kunmap_local(fragdata);
473 		}
474 		entry->type = TSNEP_TX_TYPE_SKB_FRAG_INLINE;
475 		mapped = 0;
476 	}
477 
478 	return mapped;
479 }
480 
481 static int tsnep_tx_map(struct sk_buff *skb, struct tsnep_tx *tx, int count)
482 {
483 	struct device *dmadev = tx->adapter->dmadev;
484 	struct tsnep_tx_entry *entry;
485 	unsigned int len;
486 	int map_len = 0;
487 	dma_addr_t dma;
488 	int i, mapped;
489 
490 	for (i = 0; i < count; i++) {
491 		entry = &tx->entry[(tx->write + i) & TSNEP_RING_MASK];
492 
493 		if (!i) {
494 			len = skb_headlen(skb);
495 			if (likely(len > TSNEP_DESC_SIZE_DATA_AFTER_INLINE)) {
496 				dma = dma_map_single(dmadev, skb->data, len,
497 						     DMA_TO_DEVICE);
498 				if (dma_mapping_error(dmadev, dma))
499 					return -ENOMEM;
500 				entry->type = TSNEP_TX_TYPE_SKB_MAP;
501 				mapped = 1;
502 			} else {
503 				memcpy(&entry->desc->tx, skb->data, len);
504 				entry->type = TSNEP_TX_TYPE_SKB_INLINE;
505 				mapped = 0;
506 			}
507 		} else {
508 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
509 
510 			len = skb_frag_size(frag);
511 			mapped = tsnep_tx_map_frag(frag, entry, dmadev, &dma);
512 			if (mapped < 0)
513 				return mapped;
514 		}
515 
516 		entry->len = len;
517 		if (likely(mapped)) {
518 			dma_unmap_addr_set(entry, dma, dma);
519 			entry->desc->tx = __cpu_to_le64(dma);
520 		}
521 
522 		map_len += len;
523 	}
524 
525 	return map_len;
526 }
527 
528 static int tsnep_tx_unmap(struct tsnep_tx *tx, int index, int count)
529 {
530 	struct device *dmadev = tx->adapter->dmadev;
531 	struct tsnep_tx_entry *entry;
532 	int map_len = 0;
533 	int i;
534 
535 	for (i = 0; i < count; i++) {
536 		entry = &tx->entry[(index + i) & TSNEP_RING_MASK];
537 
538 		if (entry->len) {
539 			if (entry->type & TSNEP_TX_TYPE_MAP)
540 				dma_unmap_single(dmadev,
541 						 dma_unmap_addr(entry, dma),
542 						 dma_unmap_len(entry, len),
543 						 DMA_TO_DEVICE);
544 			else if (entry->type & TSNEP_TX_TYPE_MAP_PAGE)
545 				dma_unmap_page(dmadev,
546 					       dma_unmap_addr(entry, dma),
547 					       dma_unmap_len(entry, len),
548 					       DMA_TO_DEVICE);
549 			map_len += entry->len;
550 			entry->len = 0;
551 		}
552 	}
553 
554 	return map_len;
555 }
556 
557 static netdev_tx_t tsnep_xmit_frame_ring(struct sk_buff *skb,
558 					 struct tsnep_tx *tx)
559 {
560 	int count = 1;
561 	struct tsnep_tx_entry *entry;
562 	int length;
563 	int i;
564 	int retval;
565 
566 	if (skb_shinfo(skb)->nr_frags > 0)
567 		count += skb_shinfo(skb)->nr_frags;
568 
569 	if (tsnep_tx_desc_available(tx) < count) {
570 		/* ring full, shall not happen because queue is stopped if full
571 		 * below
572 		 */
573 		netif_stop_subqueue(tx->adapter->netdev, tx->queue_index);
574 
575 		return NETDEV_TX_BUSY;
576 	}
577 
578 	entry = &tx->entry[tx->write];
579 	entry->skb = skb;
580 
581 	retval = tsnep_tx_map(skb, tx, count);
582 	if (retval < 0) {
583 		tsnep_tx_unmap(tx, tx->write, count);
584 		dev_kfree_skb_any(entry->skb);
585 		entry->skb = NULL;
586 
587 		tx->dropped++;
588 
589 		return NETDEV_TX_OK;
590 	}
591 	length = retval;
592 
593 	if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)
594 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
595 
596 	for (i = 0; i < count; i++)
597 		tsnep_tx_activate(tx, (tx->write + i) & TSNEP_RING_MASK, length,
598 				  i == count - 1);
599 	tx->write = (tx->write + count) & TSNEP_RING_MASK;
600 
601 	skb_tx_timestamp(skb);
602 
603 	/* descriptor properties shall be valid before hardware is notified */
604 	dma_wmb();
605 
606 	iowrite32(TSNEP_CONTROL_TX_ENABLE, tx->addr + TSNEP_CONTROL);
607 
608 	if (tsnep_tx_desc_available(tx) < (MAX_SKB_FRAGS + 1)) {
609 		/* ring can get full with next frame */
610 		netif_stop_subqueue(tx->adapter->netdev, tx->queue_index);
611 	}
612 
613 	return NETDEV_TX_OK;
614 }
615 
616 static int tsnep_xdp_tx_map(struct xdp_frame *xdpf, struct tsnep_tx *tx,
617 			    struct skb_shared_info *shinfo, int count, u32 type)
618 {
619 	struct device *dmadev = tx->adapter->dmadev;
620 	struct tsnep_tx_entry *entry;
621 	struct page *page;
622 	skb_frag_t *frag;
623 	unsigned int len;
624 	int map_len = 0;
625 	dma_addr_t dma;
626 	void *data;
627 	int i;
628 
629 	frag = NULL;
630 	len = xdpf->len;
631 	for (i = 0; i < count; i++) {
632 		entry = &tx->entry[(tx->write + i) & TSNEP_RING_MASK];
633 		if (type & TSNEP_TX_TYPE_XDP_NDO) {
634 			data = unlikely(frag) ? skb_frag_address(frag) :
635 						xdpf->data;
636 			dma = dma_map_single(dmadev, data, len, DMA_TO_DEVICE);
637 			if (dma_mapping_error(dmadev, dma))
638 				return -ENOMEM;
639 
640 			entry->type = TSNEP_TX_TYPE_XDP_NDO_MAP_PAGE;
641 		} else {
642 			page = unlikely(frag) ? skb_frag_page(frag) :
643 						virt_to_page(xdpf->data);
644 			dma = page_pool_get_dma_addr(page);
645 			if (unlikely(frag))
646 				dma += skb_frag_off(frag);
647 			else
648 				dma += sizeof(*xdpf) + xdpf->headroom;
649 			dma_sync_single_for_device(dmadev, dma, len,
650 						   DMA_BIDIRECTIONAL);
651 
652 			entry->type = TSNEP_TX_TYPE_XDP_TX;
653 		}
654 
655 		entry->len = len;
656 		dma_unmap_addr_set(entry, dma, dma);
657 
658 		entry->desc->tx = __cpu_to_le64(dma);
659 
660 		map_len += len;
661 
662 		if (i + 1 < count) {
663 			frag = &shinfo->frags[i];
664 			len = skb_frag_size(frag);
665 		}
666 	}
667 
668 	return map_len;
669 }
670 
671 /* This function requires __netif_tx_lock is held by the caller. */
672 static bool tsnep_xdp_xmit_frame_ring(struct xdp_frame *xdpf,
673 				      struct tsnep_tx *tx, u32 type)
674 {
675 	struct skb_shared_info *shinfo = xdp_get_shared_info_from_frame(xdpf);
676 	struct tsnep_tx_entry *entry;
677 	int count, length, retval, i;
678 
679 	count = 1;
680 	if (unlikely(xdp_frame_has_frags(xdpf)))
681 		count += shinfo->nr_frags;
682 
683 	/* ensure that TX ring is not filled up by XDP, always MAX_SKB_FRAGS
684 	 * will be available for normal TX path and queue is stopped there if
685 	 * necessary
686 	 */
687 	if (tsnep_tx_desc_available(tx) < (MAX_SKB_FRAGS + 1 + count))
688 		return false;
689 
690 	entry = &tx->entry[tx->write];
691 	entry->xdpf = xdpf;
692 
693 	retval = tsnep_xdp_tx_map(xdpf, tx, shinfo, count, type);
694 	if (retval < 0) {
695 		tsnep_tx_unmap(tx, tx->write, count);
696 		entry->xdpf = NULL;
697 
698 		tx->dropped++;
699 
700 		return false;
701 	}
702 	length = retval;
703 
704 	for (i = 0; i < count; i++)
705 		tsnep_tx_activate(tx, (tx->write + i) & TSNEP_RING_MASK, length,
706 				  i == count - 1);
707 	tx->write = (tx->write + count) & TSNEP_RING_MASK;
708 
709 	/* descriptor properties shall be valid before hardware is notified */
710 	dma_wmb();
711 
712 	return true;
713 }
714 
715 static void tsnep_xdp_xmit_flush(struct tsnep_tx *tx)
716 {
717 	iowrite32(TSNEP_CONTROL_TX_ENABLE, tx->addr + TSNEP_CONTROL);
718 }
719 
720 static bool tsnep_xdp_xmit_back(struct tsnep_adapter *adapter,
721 				struct xdp_buff *xdp,
722 				struct netdev_queue *tx_nq, struct tsnep_tx *tx,
723 				bool zc)
724 {
725 	struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
726 	bool xmit;
727 	u32 type;
728 
729 	if (unlikely(!xdpf))
730 		return false;
731 
732 	/* no page pool for zero copy */
733 	if (zc)
734 		type = TSNEP_TX_TYPE_XDP_NDO;
735 	else
736 		type = TSNEP_TX_TYPE_XDP_TX;
737 
738 	__netif_tx_lock(tx_nq, smp_processor_id());
739 
740 	xmit = tsnep_xdp_xmit_frame_ring(xdpf, tx, type);
741 
742 	/* Avoid transmit queue timeout since we share it with the slow path */
743 	if (xmit)
744 		txq_trans_cond_update(tx_nq);
745 
746 	__netif_tx_unlock(tx_nq);
747 
748 	return xmit;
749 }
750 
751 static int tsnep_xdp_tx_map_zc(struct xdp_desc *xdpd, struct tsnep_tx *tx)
752 {
753 	struct tsnep_tx_entry *entry;
754 	dma_addr_t dma;
755 
756 	entry = &tx->entry[tx->write];
757 	entry->zc = true;
758 
759 	dma = xsk_buff_raw_get_dma(tx->xsk_pool, xdpd->addr);
760 	xsk_buff_raw_dma_sync_for_device(tx->xsk_pool, dma, xdpd->len);
761 
762 	entry->type = TSNEP_TX_TYPE_XSK;
763 	entry->len = xdpd->len;
764 
765 	entry->desc->tx = __cpu_to_le64(dma);
766 
767 	return xdpd->len;
768 }
769 
770 static void tsnep_xdp_xmit_frame_ring_zc(struct xdp_desc *xdpd,
771 					 struct tsnep_tx *tx)
772 {
773 	int length;
774 
775 	length = tsnep_xdp_tx_map_zc(xdpd, tx);
776 
777 	tsnep_tx_activate(tx, tx->write, length, true);
778 	tx->write = (tx->write + 1) & TSNEP_RING_MASK;
779 }
780 
781 static void tsnep_xdp_xmit_zc(struct tsnep_tx *tx)
782 {
783 	int desc_available = tsnep_tx_desc_available(tx);
784 	struct xdp_desc *descs = tx->xsk_pool->tx_descs;
785 	int batch, i;
786 
787 	/* ensure that TX ring is not filled up by XDP, always MAX_SKB_FRAGS
788 	 * will be available for normal TX path and queue is stopped there if
789 	 * necessary
790 	 */
791 	if (desc_available <= (MAX_SKB_FRAGS + 1))
792 		return;
793 	desc_available -= MAX_SKB_FRAGS + 1;
794 
795 	batch = xsk_tx_peek_release_desc_batch(tx->xsk_pool, desc_available);
796 	for (i = 0; i < batch; i++)
797 		tsnep_xdp_xmit_frame_ring_zc(&descs[i], tx);
798 
799 	if (batch) {
800 		/* descriptor properties shall be valid before hardware is
801 		 * notified
802 		 */
803 		dma_wmb();
804 
805 		tsnep_xdp_xmit_flush(tx);
806 	}
807 }
808 
809 static bool tsnep_tx_poll(struct tsnep_tx *tx, int napi_budget)
810 {
811 	struct tsnep_tx_entry *entry;
812 	struct netdev_queue *nq;
813 	int xsk_frames = 0;
814 	int budget = 128;
815 	int length;
816 	int count;
817 
818 	nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
819 	__netif_tx_lock(nq, smp_processor_id());
820 
821 	do {
822 		if (tx->read == tx->write)
823 			break;
824 
825 		entry = &tx->entry[tx->read];
826 		if ((__le32_to_cpu(entry->desc_wb->properties) &
827 		     TSNEP_TX_DESC_OWNER_MASK) !=
828 		    (entry->properties & TSNEP_TX_DESC_OWNER_MASK))
829 			break;
830 
831 		/* descriptor properties shall be read first, because valid data
832 		 * is signaled there
833 		 */
834 		dma_rmb();
835 
836 		count = 1;
837 		if ((entry->type & TSNEP_TX_TYPE_SKB) &&
838 		    skb_shinfo(entry->skb)->nr_frags > 0)
839 			count += skb_shinfo(entry->skb)->nr_frags;
840 		else if ((entry->type & TSNEP_TX_TYPE_XDP) &&
841 			 xdp_frame_has_frags(entry->xdpf))
842 			count += xdp_get_shared_info_from_frame(entry->xdpf)->nr_frags;
843 
844 		length = tsnep_tx_unmap(tx, tx->read, count);
845 
846 		if ((entry->type & TSNEP_TX_TYPE_SKB) &&
847 		    (skb_shinfo(entry->skb)->tx_flags & SKBTX_IN_PROGRESS) &&
848 		    (__le32_to_cpu(entry->desc_wb->properties) &
849 		     TSNEP_DESC_EXTENDED_WRITEBACK_FLAG)) {
850 			struct skb_shared_hwtstamps hwtstamps;
851 			u64 timestamp;
852 
853 			if (skb_shinfo(entry->skb)->tx_flags &
854 			    SKBTX_HW_TSTAMP_USE_CYCLES)
855 				timestamp =
856 					__le64_to_cpu(entry->desc_wb->counter);
857 			else
858 				timestamp =
859 					__le64_to_cpu(entry->desc_wb->timestamp);
860 
861 			memset(&hwtstamps, 0, sizeof(hwtstamps));
862 			hwtstamps.hwtstamp = ns_to_ktime(timestamp);
863 
864 			skb_tstamp_tx(entry->skb, &hwtstamps);
865 		}
866 
867 		if (entry->type & TSNEP_TX_TYPE_SKB)
868 			napi_consume_skb(entry->skb, napi_budget);
869 		else if (entry->type & TSNEP_TX_TYPE_XDP)
870 			xdp_return_frame_rx_napi(entry->xdpf);
871 		else
872 			xsk_frames++;
873 		/* xdpf and zc are union with skb */
874 		entry->skb = NULL;
875 
876 		tx->read = (tx->read + count) & TSNEP_RING_MASK;
877 
878 		tx->packets++;
879 		tx->bytes += length + ETH_FCS_LEN;
880 
881 		budget--;
882 	} while (likely(budget));
883 
884 	if (tx->xsk_pool) {
885 		if (xsk_frames)
886 			xsk_tx_completed(tx->xsk_pool, xsk_frames);
887 		if (xsk_uses_need_wakeup(tx->xsk_pool))
888 			xsk_set_tx_need_wakeup(tx->xsk_pool);
889 		tsnep_xdp_xmit_zc(tx);
890 	}
891 
892 	if ((tsnep_tx_desc_available(tx) >= ((MAX_SKB_FRAGS + 1) * 2)) &&
893 	    netif_tx_queue_stopped(nq)) {
894 		netif_tx_wake_queue(nq);
895 	}
896 
897 	__netif_tx_unlock(nq);
898 
899 	return budget != 0;
900 }
901 
902 static bool tsnep_tx_pending(struct tsnep_tx *tx)
903 {
904 	struct tsnep_tx_entry *entry;
905 	struct netdev_queue *nq;
906 	bool pending = false;
907 
908 	nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
909 	__netif_tx_lock(nq, smp_processor_id());
910 
911 	if (tx->read != tx->write) {
912 		entry = &tx->entry[tx->read];
913 		if ((__le32_to_cpu(entry->desc_wb->properties) &
914 		     TSNEP_TX_DESC_OWNER_MASK) ==
915 		    (entry->properties & TSNEP_TX_DESC_OWNER_MASK))
916 			pending = true;
917 	}
918 
919 	__netif_tx_unlock(nq);
920 
921 	return pending;
922 }
923 
924 static int tsnep_tx_open(struct tsnep_tx *tx)
925 {
926 	int retval;
927 
928 	retval = tsnep_tx_ring_create(tx);
929 	if (retval)
930 		return retval;
931 
932 	tsnep_tx_init(tx);
933 
934 	return 0;
935 }
936 
937 static void tsnep_tx_close(struct tsnep_tx *tx)
938 {
939 	tsnep_tx_ring_cleanup(tx);
940 }
941 
942 static void tsnep_rx_ring_cleanup(struct tsnep_rx *rx)
943 {
944 	struct device *dmadev = rx->adapter->dmadev;
945 	struct tsnep_rx_entry *entry;
946 	int i;
947 
948 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
949 		entry = &rx->entry[i];
950 		if (!rx->xsk_pool && entry->page)
951 			page_pool_put_full_page(rx->page_pool, entry->page,
952 						false);
953 		if (rx->xsk_pool && entry->xdp)
954 			xsk_buff_free(entry->xdp);
955 		/* xdp is union with page */
956 		entry->page = NULL;
957 	}
958 
959 	if (rx->page_pool)
960 		page_pool_destroy(rx->page_pool);
961 
962 	memset(rx->entry, 0, sizeof(rx->entry));
963 
964 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
965 		if (rx->page[i]) {
966 			dma_free_coherent(dmadev, PAGE_SIZE, rx->page[i],
967 					  rx->page_dma[i]);
968 			rx->page[i] = NULL;
969 			rx->page_dma[i] = 0;
970 		}
971 	}
972 }
973 
974 static int tsnep_rx_ring_create(struct tsnep_rx *rx)
975 {
976 	struct device *dmadev = rx->adapter->dmadev;
977 	struct tsnep_rx_entry *entry;
978 	struct page_pool_params pp_params = { 0 };
979 	struct tsnep_rx_entry *next_entry;
980 	int i, j;
981 	int retval;
982 
983 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
984 		rx->page[i] =
985 			dma_alloc_coherent(dmadev, PAGE_SIZE, &rx->page_dma[i],
986 					   GFP_KERNEL);
987 		if (!rx->page[i]) {
988 			retval = -ENOMEM;
989 			goto failed;
990 		}
991 		for (j = 0; j < TSNEP_RING_ENTRIES_PER_PAGE; j++) {
992 			entry = &rx->entry[TSNEP_RING_ENTRIES_PER_PAGE * i + j];
993 			entry->desc_wb = (struct tsnep_rx_desc_wb *)
994 				(((u8 *)rx->page[i]) + TSNEP_DESC_SIZE * j);
995 			entry->desc = (struct tsnep_rx_desc *)
996 				(((u8 *)entry->desc_wb) + TSNEP_DESC_OFFSET);
997 			entry->desc_dma = rx->page_dma[i] + TSNEP_DESC_SIZE * j;
998 		}
999 	}
1000 
1001 	pp_params.flags = PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV;
1002 	pp_params.order = 0;
1003 	pp_params.pool_size = TSNEP_RING_SIZE;
1004 	pp_params.nid = dev_to_node(dmadev);
1005 	pp_params.dev = dmadev;
1006 	pp_params.dma_dir = DMA_BIDIRECTIONAL;
1007 	pp_params.max_len = TSNEP_MAX_RX_BUF_SIZE;
1008 	pp_params.offset = TSNEP_RX_OFFSET;
1009 	rx->page_pool = page_pool_create(&pp_params);
1010 	if (IS_ERR(rx->page_pool)) {
1011 		retval = PTR_ERR(rx->page_pool);
1012 		rx->page_pool = NULL;
1013 		goto failed;
1014 	}
1015 
1016 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
1017 		entry = &rx->entry[i];
1018 		next_entry = &rx->entry[(i + 1) & TSNEP_RING_MASK];
1019 		entry->desc->next = __cpu_to_le64(next_entry->desc_dma);
1020 	}
1021 
1022 	return 0;
1023 
1024 failed:
1025 	tsnep_rx_ring_cleanup(rx);
1026 	return retval;
1027 }
1028 
1029 static void tsnep_rx_init(struct tsnep_rx *rx)
1030 {
1031 	dma_addr_t dma;
1032 
1033 	dma = rx->entry[0].desc_dma | TSNEP_RESET_OWNER_COUNTER;
1034 	iowrite32(DMA_ADDR_LOW(dma), rx->addr + TSNEP_RX_DESC_ADDR_LOW);
1035 	iowrite32(DMA_ADDR_HIGH(dma), rx->addr + TSNEP_RX_DESC_ADDR_HIGH);
1036 	rx->write = 0;
1037 	rx->read = 0;
1038 	rx->owner_counter = 1;
1039 	rx->increment_owner_counter = TSNEP_RING_SIZE - 1;
1040 }
1041 
1042 static void tsnep_rx_enable(struct tsnep_rx *rx)
1043 {
1044 	/* descriptor properties shall be valid before hardware is notified */
1045 	dma_wmb();
1046 
1047 	iowrite32(TSNEP_CONTROL_RX_ENABLE, rx->addr + TSNEP_CONTROL);
1048 }
1049 
1050 static void tsnep_rx_disable(struct tsnep_rx *rx)
1051 {
1052 	u32 val;
1053 
1054 	iowrite32(TSNEP_CONTROL_RX_DISABLE, rx->addr + TSNEP_CONTROL);
1055 	readx_poll_timeout(ioread32, rx->addr + TSNEP_CONTROL, val,
1056 			   ((val & TSNEP_CONTROL_RX_ENABLE) == 0), 10000,
1057 			   1000000);
1058 }
1059 
1060 static int tsnep_rx_desc_available(struct tsnep_rx *rx)
1061 {
1062 	if (rx->read <= rx->write)
1063 		return TSNEP_RING_SIZE - rx->write + rx->read - 1;
1064 	else
1065 		return rx->read - rx->write - 1;
1066 }
1067 
1068 static void tsnep_rx_free_page_buffer(struct tsnep_rx *rx)
1069 {
1070 	struct page **page;
1071 
1072 	/* last entry of page_buffer is always zero, because ring cannot be
1073 	 * filled completely
1074 	 */
1075 	page = rx->page_buffer;
1076 	while (*page) {
1077 		page_pool_put_full_page(rx->page_pool, *page, false);
1078 		*page = NULL;
1079 		page++;
1080 	}
1081 }
1082 
1083 static int tsnep_rx_alloc_page_buffer(struct tsnep_rx *rx)
1084 {
1085 	int i;
1086 
1087 	/* alloc for all ring entries except the last one, because ring cannot
1088 	 * be filled completely
1089 	 */
1090 	for (i = 0; i < TSNEP_RING_SIZE - 1; i++) {
1091 		rx->page_buffer[i] = page_pool_dev_alloc_pages(rx->page_pool);
1092 		if (!rx->page_buffer[i]) {
1093 			tsnep_rx_free_page_buffer(rx);
1094 
1095 			return -ENOMEM;
1096 		}
1097 	}
1098 
1099 	return 0;
1100 }
1101 
1102 static void tsnep_rx_set_page(struct tsnep_rx *rx, struct tsnep_rx_entry *entry,
1103 			      struct page *page)
1104 {
1105 	entry->page = page;
1106 	entry->len = TSNEP_MAX_RX_BUF_SIZE;
1107 	entry->dma = page_pool_get_dma_addr(entry->page);
1108 	entry->desc->rx = __cpu_to_le64(entry->dma + TSNEP_RX_OFFSET);
1109 }
1110 
1111 static int tsnep_rx_alloc_buffer(struct tsnep_rx *rx, int index)
1112 {
1113 	struct tsnep_rx_entry *entry = &rx->entry[index];
1114 	struct page *page;
1115 
1116 	page = page_pool_dev_alloc_pages(rx->page_pool);
1117 	if (unlikely(!page))
1118 		return -ENOMEM;
1119 	tsnep_rx_set_page(rx, entry, page);
1120 
1121 	return 0;
1122 }
1123 
1124 static void tsnep_rx_reuse_buffer(struct tsnep_rx *rx, int index)
1125 {
1126 	struct tsnep_rx_entry *entry = &rx->entry[index];
1127 	struct tsnep_rx_entry *read = &rx->entry[rx->read];
1128 
1129 	tsnep_rx_set_page(rx, entry, read->page);
1130 	read->page = NULL;
1131 }
1132 
1133 static void tsnep_rx_activate(struct tsnep_rx *rx, int index)
1134 {
1135 	struct tsnep_rx_entry *entry = &rx->entry[index];
1136 
1137 	/* TSNEP_MAX_RX_BUF_SIZE and TSNEP_XSK_RX_BUF_SIZE are multiple of 4 */
1138 	entry->properties = entry->len & TSNEP_DESC_LENGTH_MASK;
1139 	entry->properties |= TSNEP_DESC_INTERRUPT_FLAG;
1140 	if (index == rx->increment_owner_counter) {
1141 		rx->owner_counter++;
1142 		if (rx->owner_counter == 4)
1143 			rx->owner_counter = 1;
1144 		rx->increment_owner_counter--;
1145 		if (rx->increment_owner_counter < 0)
1146 			rx->increment_owner_counter = TSNEP_RING_SIZE - 1;
1147 	}
1148 	entry->properties |=
1149 		(rx->owner_counter << TSNEP_DESC_OWNER_COUNTER_SHIFT) &
1150 		TSNEP_DESC_OWNER_COUNTER_MASK;
1151 
1152 	/* descriptor properties shall be written last, because valid data is
1153 	 * signaled there
1154 	 */
1155 	dma_wmb();
1156 
1157 	entry->desc->properties = __cpu_to_le32(entry->properties);
1158 }
1159 
1160 static int tsnep_rx_alloc(struct tsnep_rx *rx, int count, bool reuse)
1161 {
1162 	bool alloc_failed = false;
1163 	int i, index;
1164 
1165 	for (i = 0; i < count && !alloc_failed; i++) {
1166 		index = (rx->write + i) & TSNEP_RING_MASK;
1167 
1168 		if (unlikely(tsnep_rx_alloc_buffer(rx, index))) {
1169 			rx->alloc_failed++;
1170 			alloc_failed = true;
1171 
1172 			/* reuse only if no other allocation was successful */
1173 			if (i == 0 && reuse)
1174 				tsnep_rx_reuse_buffer(rx, index);
1175 			else
1176 				break;
1177 		}
1178 
1179 		tsnep_rx_activate(rx, index);
1180 	}
1181 
1182 	if (i)
1183 		rx->write = (rx->write + i) & TSNEP_RING_MASK;
1184 
1185 	return i;
1186 }
1187 
1188 static int tsnep_rx_refill(struct tsnep_rx *rx, int count, bool reuse)
1189 {
1190 	int desc_refilled;
1191 
1192 	desc_refilled = tsnep_rx_alloc(rx, count, reuse);
1193 	if (desc_refilled)
1194 		tsnep_rx_enable(rx);
1195 
1196 	return desc_refilled;
1197 }
1198 
1199 static void tsnep_rx_set_xdp(struct tsnep_rx *rx, struct tsnep_rx_entry *entry,
1200 			     struct xdp_buff *xdp)
1201 {
1202 	entry->xdp = xdp;
1203 	entry->len = TSNEP_XSK_RX_BUF_SIZE;
1204 	entry->dma = xsk_buff_xdp_get_dma(entry->xdp);
1205 	entry->desc->rx = __cpu_to_le64(entry->dma);
1206 }
1207 
1208 static void tsnep_rx_reuse_buffer_zc(struct tsnep_rx *rx, int index)
1209 {
1210 	struct tsnep_rx_entry *entry = &rx->entry[index];
1211 	struct tsnep_rx_entry *read = &rx->entry[rx->read];
1212 
1213 	tsnep_rx_set_xdp(rx, entry, read->xdp);
1214 	read->xdp = NULL;
1215 }
1216 
1217 static int tsnep_rx_alloc_zc(struct tsnep_rx *rx, int count, bool reuse)
1218 {
1219 	u32 allocated;
1220 	int i;
1221 
1222 	allocated = xsk_buff_alloc_batch(rx->xsk_pool, rx->xdp_batch, count);
1223 	for (i = 0; i < allocated; i++) {
1224 		int index = (rx->write + i) & TSNEP_RING_MASK;
1225 		struct tsnep_rx_entry *entry = &rx->entry[index];
1226 
1227 		tsnep_rx_set_xdp(rx, entry, rx->xdp_batch[i]);
1228 		tsnep_rx_activate(rx, index);
1229 	}
1230 	if (i == 0) {
1231 		rx->alloc_failed++;
1232 
1233 		if (reuse) {
1234 			tsnep_rx_reuse_buffer_zc(rx, rx->write);
1235 			tsnep_rx_activate(rx, rx->write);
1236 		}
1237 	}
1238 
1239 	if (i)
1240 		rx->write = (rx->write + i) & TSNEP_RING_MASK;
1241 
1242 	return i;
1243 }
1244 
1245 static void tsnep_rx_free_zc(struct tsnep_rx *rx)
1246 {
1247 	int i;
1248 
1249 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
1250 		struct tsnep_rx_entry *entry = &rx->entry[i];
1251 
1252 		if (entry->xdp)
1253 			xsk_buff_free(entry->xdp);
1254 		entry->xdp = NULL;
1255 	}
1256 }
1257 
1258 static int tsnep_rx_refill_zc(struct tsnep_rx *rx, int count, bool reuse)
1259 {
1260 	int desc_refilled;
1261 
1262 	desc_refilled = tsnep_rx_alloc_zc(rx, count, reuse);
1263 	if (desc_refilled)
1264 		tsnep_rx_enable(rx);
1265 
1266 	return desc_refilled;
1267 }
1268 
1269 static bool tsnep_xdp_run_prog(struct tsnep_rx *rx, struct bpf_prog *prog,
1270 			       struct xdp_buff *xdp, int *status,
1271 			       struct netdev_queue *tx_nq, struct tsnep_tx *tx)
1272 {
1273 	unsigned int length;
1274 	unsigned int sync;
1275 	u32 act;
1276 
1277 	length = xdp->data_end - xdp->data_hard_start - XDP_PACKET_HEADROOM;
1278 
1279 	act = bpf_prog_run_xdp(prog, xdp);
1280 	switch (act) {
1281 	case XDP_PASS:
1282 		return false;
1283 	case XDP_TX:
1284 		if (!tsnep_xdp_xmit_back(rx->adapter, xdp, tx_nq, tx, false))
1285 			goto out_failure;
1286 		*status |= TSNEP_XDP_TX;
1287 		return true;
1288 	case XDP_REDIRECT:
1289 		if (xdp_do_redirect(rx->adapter->netdev, xdp, prog) < 0)
1290 			goto out_failure;
1291 		*status |= TSNEP_XDP_REDIRECT;
1292 		return true;
1293 	default:
1294 		bpf_warn_invalid_xdp_action(rx->adapter->netdev, prog, act);
1295 		fallthrough;
1296 	case XDP_ABORTED:
1297 out_failure:
1298 		trace_xdp_exception(rx->adapter->netdev, prog, act);
1299 		fallthrough;
1300 	case XDP_DROP:
1301 		/* Due xdp_adjust_tail: DMA sync for_device cover max len CPU
1302 		 * touch
1303 		 */
1304 		sync = xdp->data_end - xdp->data_hard_start -
1305 		       XDP_PACKET_HEADROOM;
1306 		sync = max(sync, length);
1307 		page_pool_put_page(rx->page_pool, virt_to_head_page(xdp->data),
1308 				   sync, true);
1309 		return true;
1310 	}
1311 }
1312 
1313 static bool tsnep_xdp_run_prog_zc(struct tsnep_rx *rx, struct bpf_prog *prog,
1314 				  struct xdp_buff *xdp, int *status,
1315 				  struct netdev_queue *tx_nq,
1316 				  struct tsnep_tx *tx)
1317 {
1318 	u32 act;
1319 
1320 	act = bpf_prog_run_xdp(prog, xdp);
1321 
1322 	/* XDP_REDIRECT is the main action for zero-copy */
1323 	if (likely(act == XDP_REDIRECT)) {
1324 		if (xdp_do_redirect(rx->adapter->netdev, xdp, prog) < 0)
1325 			goto out_failure;
1326 		*status |= TSNEP_XDP_REDIRECT;
1327 		return true;
1328 	}
1329 
1330 	switch (act) {
1331 	case XDP_PASS:
1332 		return false;
1333 	case XDP_TX:
1334 		if (!tsnep_xdp_xmit_back(rx->adapter, xdp, tx_nq, tx, true))
1335 			goto out_failure;
1336 		*status |= TSNEP_XDP_TX;
1337 		return true;
1338 	default:
1339 		bpf_warn_invalid_xdp_action(rx->adapter->netdev, prog, act);
1340 		fallthrough;
1341 	case XDP_ABORTED:
1342 out_failure:
1343 		trace_xdp_exception(rx->adapter->netdev, prog, act);
1344 		fallthrough;
1345 	case XDP_DROP:
1346 		xsk_buff_free(xdp);
1347 		return true;
1348 	}
1349 }
1350 
1351 static void tsnep_finalize_xdp(struct tsnep_adapter *adapter, int status,
1352 			       struct netdev_queue *tx_nq, struct tsnep_tx *tx)
1353 {
1354 	if (status & TSNEP_XDP_TX) {
1355 		__netif_tx_lock(tx_nq, smp_processor_id());
1356 		tsnep_xdp_xmit_flush(tx);
1357 		__netif_tx_unlock(tx_nq);
1358 	}
1359 
1360 	if (status & TSNEP_XDP_REDIRECT)
1361 		xdp_do_flush();
1362 }
1363 
1364 static struct sk_buff *tsnep_build_skb(struct tsnep_rx *rx, struct page *page,
1365 				       int length)
1366 {
1367 	struct sk_buff *skb;
1368 
1369 	skb = napi_build_skb(page_address(page), PAGE_SIZE);
1370 	if (unlikely(!skb))
1371 		return NULL;
1372 
1373 	/* update pointers within the skb to store the data */
1374 	skb_reserve(skb, TSNEP_RX_OFFSET + TSNEP_RX_INLINE_METADATA_SIZE);
1375 	__skb_put(skb, length - ETH_FCS_LEN);
1376 
1377 	if (rx->adapter->hwtstamp_config.rx_filter == HWTSTAMP_FILTER_ALL) {
1378 		struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
1379 		struct tsnep_rx_inline *rx_inline =
1380 			(struct tsnep_rx_inline *)(page_address(page) +
1381 						   TSNEP_RX_OFFSET);
1382 
1383 		skb_shinfo(skb)->tx_flags |=
1384 			SKBTX_HW_TSTAMP_NETDEV;
1385 		memset(hwtstamps, 0, sizeof(*hwtstamps));
1386 		hwtstamps->netdev_data = rx_inline;
1387 	}
1388 
1389 	skb_record_rx_queue(skb, rx->queue_index);
1390 	skb->protocol = eth_type_trans(skb, rx->adapter->netdev);
1391 
1392 	return skb;
1393 }
1394 
1395 static void tsnep_rx_page(struct tsnep_rx *rx, struct napi_struct *napi,
1396 			  struct page *page, int length)
1397 {
1398 	struct sk_buff *skb;
1399 
1400 	skb = tsnep_build_skb(rx, page, length);
1401 	if (skb) {
1402 		skb_mark_for_recycle(skb);
1403 
1404 		rx->packets++;
1405 		rx->bytes += length;
1406 		if (skb->pkt_type == PACKET_MULTICAST)
1407 			rx->multicast++;
1408 
1409 		napi_gro_receive(napi, skb);
1410 	} else {
1411 		page_pool_recycle_direct(rx->page_pool, page);
1412 
1413 		rx->dropped++;
1414 	}
1415 }
1416 
1417 static int tsnep_rx_poll(struct tsnep_rx *rx, struct napi_struct *napi,
1418 			 int budget)
1419 {
1420 	struct device *dmadev = rx->adapter->dmadev;
1421 	enum dma_data_direction dma_dir;
1422 	struct tsnep_rx_entry *entry;
1423 	struct netdev_queue *tx_nq;
1424 	struct bpf_prog *prog;
1425 	struct xdp_buff xdp;
1426 	struct tsnep_tx *tx;
1427 	int desc_available;
1428 	int xdp_status = 0;
1429 	int done = 0;
1430 	int length;
1431 
1432 	desc_available = tsnep_rx_desc_available(rx);
1433 	dma_dir = page_pool_get_dma_dir(rx->page_pool);
1434 	prog = READ_ONCE(rx->adapter->xdp_prog);
1435 	if (prog) {
1436 		tx_nq = netdev_get_tx_queue(rx->adapter->netdev,
1437 					    rx->tx_queue_index);
1438 		tx = &rx->adapter->tx[rx->tx_queue_index];
1439 
1440 		xdp_init_buff(&xdp, PAGE_SIZE, &rx->xdp_rxq);
1441 	}
1442 
1443 	while (likely(done < budget) && (rx->read != rx->write)) {
1444 		entry = &rx->entry[rx->read];
1445 		if ((__le32_to_cpu(entry->desc_wb->properties) &
1446 		     TSNEP_DESC_OWNER_COUNTER_MASK) !=
1447 		    (entry->properties & TSNEP_DESC_OWNER_COUNTER_MASK))
1448 			break;
1449 		done++;
1450 
1451 		if (desc_available >= TSNEP_RING_RX_REFILL) {
1452 			bool reuse = desc_available >= TSNEP_RING_RX_REUSE;
1453 
1454 			desc_available -= tsnep_rx_refill(rx, desc_available,
1455 							  reuse);
1456 			if (!entry->page) {
1457 				/* buffer has been reused for refill to prevent
1458 				 * empty RX ring, thus buffer cannot be used for
1459 				 * RX processing
1460 				 */
1461 				rx->read = (rx->read + 1) & TSNEP_RING_MASK;
1462 				desc_available++;
1463 
1464 				rx->dropped++;
1465 
1466 				continue;
1467 			}
1468 		}
1469 
1470 		/* descriptor properties shall be read first, because valid data
1471 		 * is signaled there
1472 		 */
1473 		dma_rmb();
1474 
1475 		prefetch(page_address(entry->page) + TSNEP_RX_OFFSET);
1476 		length = __le32_to_cpu(entry->desc_wb->properties) &
1477 			 TSNEP_DESC_LENGTH_MASK;
1478 		dma_sync_single_range_for_cpu(dmadev, entry->dma,
1479 					      TSNEP_RX_OFFSET, length, dma_dir);
1480 
1481 		/* RX metadata with timestamps is in front of actual data,
1482 		 * subtract metadata size to get length of actual data and
1483 		 * consider metadata size as offset of actual data during RX
1484 		 * processing
1485 		 */
1486 		length -= TSNEP_RX_INLINE_METADATA_SIZE;
1487 
1488 		rx->read = (rx->read + 1) & TSNEP_RING_MASK;
1489 		desc_available++;
1490 
1491 		if (prog) {
1492 			bool consume;
1493 
1494 			xdp_prepare_buff(&xdp, page_address(entry->page),
1495 					 XDP_PACKET_HEADROOM + TSNEP_RX_INLINE_METADATA_SIZE,
1496 					 length - ETH_FCS_LEN, false);
1497 
1498 			consume = tsnep_xdp_run_prog(rx, prog, &xdp,
1499 						     &xdp_status, tx_nq, tx);
1500 			if (consume) {
1501 				rx->packets++;
1502 				rx->bytes += length;
1503 
1504 				entry->page = NULL;
1505 
1506 				continue;
1507 			}
1508 		}
1509 
1510 		tsnep_rx_page(rx, napi, entry->page, length);
1511 		entry->page = NULL;
1512 	}
1513 
1514 	if (xdp_status)
1515 		tsnep_finalize_xdp(rx->adapter, xdp_status, tx_nq, tx);
1516 
1517 	if (desc_available)
1518 		tsnep_rx_refill(rx, desc_available, false);
1519 
1520 	return done;
1521 }
1522 
1523 static int tsnep_rx_poll_zc(struct tsnep_rx *rx, struct napi_struct *napi,
1524 			    int budget)
1525 {
1526 	struct tsnep_rx_entry *entry;
1527 	struct netdev_queue *tx_nq;
1528 	struct bpf_prog *prog;
1529 	struct tsnep_tx *tx;
1530 	int desc_available;
1531 	int xdp_status = 0;
1532 	struct page *page;
1533 	int done = 0;
1534 	int length;
1535 
1536 	desc_available = tsnep_rx_desc_available(rx);
1537 	prog = READ_ONCE(rx->adapter->xdp_prog);
1538 	if (prog) {
1539 		tx_nq = netdev_get_tx_queue(rx->adapter->netdev,
1540 					    rx->tx_queue_index);
1541 		tx = &rx->adapter->tx[rx->tx_queue_index];
1542 	}
1543 
1544 	while (likely(done < budget) && (rx->read != rx->write)) {
1545 		entry = &rx->entry[rx->read];
1546 		if ((__le32_to_cpu(entry->desc_wb->properties) &
1547 		     TSNEP_DESC_OWNER_COUNTER_MASK) !=
1548 		    (entry->properties & TSNEP_DESC_OWNER_COUNTER_MASK))
1549 			break;
1550 		done++;
1551 
1552 		if (desc_available >= TSNEP_RING_RX_REFILL) {
1553 			bool reuse = desc_available >= TSNEP_RING_RX_REUSE;
1554 
1555 			desc_available -= tsnep_rx_refill_zc(rx, desc_available,
1556 							     reuse);
1557 			if (!entry->xdp) {
1558 				/* buffer has been reused for refill to prevent
1559 				 * empty RX ring, thus buffer cannot be used for
1560 				 * RX processing
1561 				 */
1562 				rx->read = (rx->read + 1) & TSNEP_RING_MASK;
1563 				desc_available++;
1564 
1565 				rx->dropped++;
1566 
1567 				continue;
1568 			}
1569 		}
1570 
1571 		/* descriptor properties shall be read first, because valid data
1572 		 * is signaled there
1573 		 */
1574 		dma_rmb();
1575 
1576 		prefetch(entry->xdp->data);
1577 		length = __le32_to_cpu(entry->desc_wb->properties) &
1578 			 TSNEP_DESC_LENGTH_MASK;
1579 		xsk_buff_set_size(entry->xdp, length - ETH_FCS_LEN);
1580 		xsk_buff_dma_sync_for_cpu(entry->xdp, rx->xsk_pool);
1581 
1582 		/* RX metadata with timestamps is in front of actual data,
1583 		 * subtract metadata size to get length of actual data and
1584 		 * consider metadata size as offset of actual data during RX
1585 		 * processing
1586 		 */
1587 		length -= TSNEP_RX_INLINE_METADATA_SIZE;
1588 
1589 		rx->read = (rx->read + 1) & TSNEP_RING_MASK;
1590 		desc_available++;
1591 
1592 		if (prog) {
1593 			bool consume;
1594 
1595 			entry->xdp->data += TSNEP_RX_INLINE_METADATA_SIZE;
1596 			entry->xdp->data_meta += TSNEP_RX_INLINE_METADATA_SIZE;
1597 
1598 			consume = tsnep_xdp_run_prog_zc(rx, prog, entry->xdp,
1599 							&xdp_status, tx_nq, tx);
1600 			if (consume) {
1601 				rx->packets++;
1602 				rx->bytes += length;
1603 
1604 				entry->xdp = NULL;
1605 
1606 				continue;
1607 			}
1608 		}
1609 
1610 		page = page_pool_dev_alloc_pages(rx->page_pool);
1611 		if (page) {
1612 			memcpy(page_address(page) + TSNEP_RX_OFFSET,
1613 			       entry->xdp->data - TSNEP_RX_INLINE_METADATA_SIZE,
1614 			       length + TSNEP_RX_INLINE_METADATA_SIZE);
1615 			tsnep_rx_page(rx, napi, page, length);
1616 		} else {
1617 			rx->dropped++;
1618 		}
1619 		xsk_buff_free(entry->xdp);
1620 		entry->xdp = NULL;
1621 	}
1622 
1623 	if (xdp_status)
1624 		tsnep_finalize_xdp(rx->adapter, xdp_status, tx_nq, tx);
1625 
1626 	if (desc_available)
1627 		desc_available -= tsnep_rx_refill_zc(rx, desc_available, false);
1628 
1629 	if (xsk_uses_need_wakeup(rx->xsk_pool)) {
1630 		if (desc_available)
1631 			xsk_set_rx_need_wakeup(rx->xsk_pool);
1632 		else
1633 			xsk_clear_rx_need_wakeup(rx->xsk_pool);
1634 
1635 		return done;
1636 	}
1637 
1638 	return desc_available ? budget : done;
1639 }
1640 
1641 static bool tsnep_rx_pending(struct tsnep_rx *rx)
1642 {
1643 	struct tsnep_rx_entry *entry;
1644 
1645 	if (rx->read != rx->write) {
1646 		entry = &rx->entry[rx->read];
1647 		if ((__le32_to_cpu(entry->desc_wb->properties) &
1648 		     TSNEP_DESC_OWNER_COUNTER_MASK) ==
1649 		    (entry->properties & TSNEP_DESC_OWNER_COUNTER_MASK))
1650 			return true;
1651 	}
1652 
1653 	return false;
1654 }
1655 
1656 static int tsnep_rx_open(struct tsnep_rx *rx)
1657 {
1658 	int desc_available;
1659 	int retval;
1660 
1661 	retval = tsnep_rx_ring_create(rx);
1662 	if (retval)
1663 		return retval;
1664 
1665 	tsnep_rx_init(rx);
1666 
1667 	desc_available = tsnep_rx_desc_available(rx);
1668 	if (rx->xsk_pool)
1669 		retval = tsnep_rx_alloc_zc(rx, desc_available, false);
1670 	else
1671 		retval = tsnep_rx_alloc(rx, desc_available, false);
1672 	if (retval != desc_available) {
1673 		retval = -ENOMEM;
1674 
1675 		goto alloc_failed;
1676 	}
1677 
1678 	/* prealloc pages to prevent allocation failures when XSK pool is
1679 	 * disabled at runtime
1680 	 */
1681 	if (rx->xsk_pool) {
1682 		retval = tsnep_rx_alloc_page_buffer(rx);
1683 		if (retval)
1684 			goto alloc_failed;
1685 	}
1686 
1687 	return 0;
1688 
1689 alloc_failed:
1690 	tsnep_rx_ring_cleanup(rx);
1691 	return retval;
1692 }
1693 
1694 static void tsnep_rx_close(struct tsnep_rx *rx)
1695 {
1696 	if (rx->xsk_pool)
1697 		tsnep_rx_free_page_buffer(rx);
1698 
1699 	tsnep_rx_ring_cleanup(rx);
1700 }
1701 
1702 static void tsnep_rx_reopen(struct tsnep_rx *rx)
1703 {
1704 	struct page **page = rx->page_buffer;
1705 	int i;
1706 
1707 	tsnep_rx_init(rx);
1708 
1709 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
1710 		struct tsnep_rx_entry *entry = &rx->entry[i];
1711 
1712 		/* defined initial values for properties are required for
1713 		 * correct owner counter checking
1714 		 */
1715 		entry->desc->properties = 0;
1716 		entry->desc_wb->properties = 0;
1717 
1718 		/* prevent allocation failures by reusing kept pages */
1719 		if (*page) {
1720 			tsnep_rx_set_page(rx, entry, *page);
1721 			tsnep_rx_activate(rx, rx->write);
1722 			rx->write++;
1723 
1724 			*page = NULL;
1725 			page++;
1726 		}
1727 	}
1728 }
1729 
1730 static void tsnep_rx_reopen_xsk(struct tsnep_rx *rx)
1731 {
1732 	struct page **page = rx->page_buffer;
1733 	u32 allocated;
1734 	int i;
1735 
1736 	tsnep_rx_init(rx);
1737 
1738 	/* alloc all ring entries except the last one, because ring cannot be
1739 	 * filled completely, as many buffers as possible is enough as wakeup is
1740 	 * done if new buffers are available
1741 	 */
1742 	allocated = xsk_buff_alloc_batch(rx->xsk_pool, rx->xdp_batch,
1743 					 TSNEP_RING_SIZE - 1);
1744 
1745 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
1746 		struct tsnep_rx_entry *entry = &rx->entry[i];
1747 
1748 		/* keep pages to prevent allocation failures when xsk is
1749 		 * disabled
1750 		 */
1751 		if (entry->page) {
1752 			*page = entry->page;
1753 			entry->page = NULL;
1754 
1755 			page++;
1756 		}
1757 
1758 		/* defined initial values for properties are required for
1759 		 * correct owner counter checking
1760 		 */
1761 		entry->desc->properties = 0;
1762 		entry->desc_wb->properties = 0;
1763 
1764 		if (allocated) {
1765 			tsnep_rx_set_xdp(rx, entry,
1766 					 rx->xdp_batch[allocated - 1]);
1767 			tsnep_rx_activate(rx, rx->write);
1768 			rx->write++;
1769 
1770 			allocated--;
1771 		}
1772 	}
1773 
1774 	/* set need wakeup flag immediately if ring is not filled completely,
1775 	 * first polling would be too late as need wakeup signalisation would
1776 	 * be delayed for an indefinite time
1777 	 */
1778 	if (xsk_uses_need_wakeup(rx->xsk_pool)) {
1779 		int desc_available = tsnep_rx_desc_available(rx);
1780 
1781 		if (desc_available)
1782 			xsk_set_rx_need_wakeup(rx->xsk_pool);
1783 		else
1784 			xsk_clear_rx_need_wakeup(rx->xsk_pool);
1785 	}
1786 }
1787 
1788 static bool tsnep_pending(struct tsnep_queue *queue)
1789 {
1790 	if (queue->tx && tsnep_tx_pending(queue->tx))
1791 		return true;
1792 
1793 	if (queue->rx && tsnep_rx_pending(queue->rx))
1794 		return true;
1795 
1796 	return false;
1797 }
1798 
1799 static int tsnep_poll(struct napi_struct *napi, int budget)
1800 {
1801 	struct tsnep_queue *queue = container_of(napi, struct tsnep_queue,
1802 						 napi);
1803 	bool complete = true;
1804 	int done = 0;
1805 
1806 	if (queue->tx)
1807 		complete = tsnep_tx_poll(queue->tx, budget);
1808 
1809 	/* handle case where we are called by netpoll with a budget of 0 */
1810 	if (unlikely(budget <= 0))
1811 		return budget;
1812 
1813 	if (queue->rx) {
1814 		done = queue->rx->xsk_pool ?
1815 		       tsnep_rx_poll_zc(queue->rx, napi, budget) :
1816 		       tsnep_rx_poll(queue->rx, napi, budget);
1817 		if (done >= budget)
1818 			complete = false;
1819 	}
1820 
1821 	/* if all work not completed, return budget and keep polling */
1822 	if (!complete)
1823 		return budget;
1824 
1825 	if (likely(napi_complete_done(napi, done))) {
1826 		tsnep_enable_irq(queue->adapter, queue->irq_mask);
1827 
1828 		/* reschedule if work is already pending, prevent rotten packets
1829 		 * which are transmitted or received after polling but before
1830 		 * interrupt enable
1831 		 */
1832 		if (tsnep_pending(queue)) {
1833 			tsnep_disable_irq(queue->adapter, queue->irq_mask);
1834 			napi_schedule(napi);
1835 		}
1836 	}
1837 
1838 	return min(done, budget - 1);
1839 }
1840 
1841 static int tsnep_request_irq(struct tsnep_queue *queue, bool first)
1842 {
1843 	const char *name = netdev_name(queue->adapter->netdev);
1844 	irq_handler_t handler;
1845 	void *dev;
1846 	int retval;
1847 
1848 	if (first) {
1849 		sprintf(queue->name, "%s-mac", name);
1850 		handler = tsnep_irq;
1851 		dev = queue->adapter;
1852 	} else {
1853 		if (queue->tx && queue->rx)
1854 			snprintf(queue->name, sizeof(queue->name), "%s-txrx-%d",
1855 				 name, queue->rx->queue_index);
1856 		else if (queue->tx)
1857 			snprintf(queue->name, sizeof(queue->name), "%s-tx-%d",
1858 				 name, queue->tx->queue_index);
1859 		else
1860 			snprintf(queue->name, sizeof(queue->name), "%s-rx-%d",
1861 				 name, queue->rx->queue_index);
1862 		handler = tsnep_irq_txrx;
1863 		dev = queue;
1864 	}
1865 
1866 	retval = request_irq(queue->irq, handler, 0, queue->name, dev);
1867 	if (retval) {
1868 		/* if name is empty, then interrupt won't be freed */
1869 		memset(queue->name, 0, sizeof(queue->name));
1870 	}
1871 
1872 	return retval;
1873 }
1874 
1875 static void tsnep_free_irq(struct tsnep_queue *queue, bool first)
1876 {
1877 	void *dev;
1878 
1879 	if (!strlen(queue->name))
1880 		return;
1881 
1882 	if (first)
1883 		dev = queue->adapter;
1884 	else
1885 		dev = queue;
1886 
1887 	free_irq(queue->irq, dev);
1888 	memset(queue->name, 0, sizeof(queue->name));
1889 }
1890 
1891 static void tsnep_queue_close(struct tsnep_queue *queue, bool first)
1892 {
1893 	struct tsnep_rx *rx = queue->rx;
1894 
1895 	tsnep_free_irq(queue, first);
1896 
1897 	if (rx) {
1898 		if (xdp_rxq_info_is_reg(&rx->xdp_rxq))
1899 			xdp_rxq_info_unreg(&rx->xdp_rxq);
1900 		if (xdp_rxq_info_is_reg(&rx->xdp_rxq_zc))
1901 			xdp_rxq_info_unreg(&rx->xdp_rxq_zc);
1902 	}
1903 
1904 	netif_napi_del(&queue->napi);
1905 }
1906 
1907 static int tsnep_queue_open(struct tsnep_adapter *adapter,
1908 			    struct tsnep_queue *queue, bool first)
1909 {
1910 	struct tsnep_rx *rx = queue->rx;
1911 	struct tsnep_tx *tx = queue->tx;
1912 	int retval;
1913 
1914 	netif_napi_add(adapter->netdev, &queue->napi, tsnep_poll);
1915 
1916 	if (rx) {
1917 		/* choose TX queue for XDP_TX */
1918 		if (tx)
1919 			rx->tx_queue_index = tx->queue_index;
1920 		else if (rx->queue_index < adapter->num_tx_queues)
1921 			rx->tx_queue_index = rx->queue_index;
1922 		else
1923 			rx->tx_queue_index = 0;
1924 
1925 		/* prepare both memory models to eliminate possible registration
1926 		 * errors when memory model is switched between page pool and
1927 		 * XSK pool during runtime
1928 		 */
1929 		retval = xdp_rxq_info_reg(&rx->xdp_rxq, adapter->netdev,
1930 					  rx->queue_index, queue->napi.napi_id);
1931 		if (retval)
1932 			goto failed;
1933 		retval = xdp_rxq_info_reg_mem_model(&rx->xdp_rxq,
1934 						    MEM_TYPE_PAGE_POOL,
1935 						    rx->page_pool);
1936 		if (retval)
1937 			goto failed;
1938 		retval = xdp_rxq_info_reg(&rx->xdp_rxq_zc, adapter->netdev,
1939 					  rx->queue_index, queue->napi.napi_id);
1940 		if (retval)
1941 			goto failed;
1942 		retval = xdp_rxq_info_reg_mem_model(&rx->xdp_rxq_zc,
1943 						    MEM_TYPE_XSK_BUFF_POOL,
1944 						    NULL);
1945 		if (retval)
1946 			goto failed;
1947 		if (rx->xsk_pool)
1948 			xsk_pool_set_rxq_info(rx->xsk_pool, &rx->xdp_rxq_zc);
1949 	}
1950 
1951 	retval = tsnep_request_irq(queue, first);
1952 	if (retval) {
1953 		netif_err(adapter, drv, adapter->netdev,
1954 			  "can't get assigned irq %d.\n", queue->irq);
1955 		goto failed;
1956 	}
1957 
1958 	return 0;
1959 
1960 failed:
1961 	tsnep_queue_close(queue, first);
1962 
1963 	return retval;
1964 }
1965 
1966 static void tsnep_queue_enable(struct tsnep_queue *queue)
1967 {
1968 	napi_enable(&queue->napi);
1969 	tsnep_enable_irq(queue->adapter, queue->irq_mask);
1970 
1971 	if (queue->tx)
1972 		tsnep_tx_enable(queue->tx);
1973 
1974 	if (queue->rx)
1975 		tsnep_rx_enable(queue->rx);
1976 }
1977 
1978 static void tsnep_queue_disable(struct tsnep_queue *queue)
1979 {
1980 	if (queue->tx)
1981 		tsnep_tx_disable(queue->tx, &queue->napi);
1982 
1983 	napi_disable(&queue->napi);
1984 	tsnep_disable_irq(queue->adapter, queue->irq_mask);
1985 
1986 	/* disable RX after NAPI polling has been disabled, because RX can be
1987 	 * enabled during NAPI polling
1988 	 */
1989 	if (queue->rx)
1990 		tsnep_rx_disable(queue->rx);
1991 }
1992 
1993 static int tsnep_netdev_open(struct net_device *netdev)
1994 {
1995 	struct tsnep_adapter *adapter = netdev_priv(netdev);
1996 	int i, retval;
1997 
1998 	for (i = 0; i < adapter->num_queues; i++) {
1999 		if (adapter->queue[i].tx) {
2000 			retval = tsnep_tx_open(adapter->queue[i].tx);
2001 			if (retval)
2002 				goto failed;
2003 		}
2004 		if (adapter->queue[i].rx) {
2005 			retval = tsnep_rx_open(adapter->queue[i].rx);
2006 			if (retval)
2007 				goto failed;
2008 		}
2009 
2010 		retval = tsnep_queue_open(adapter, &adapter->queue[i], i == 0);
2011 		if (retval)
2012 			goto failed;
2013 	}
2014 
2015 	retval = netif_set_real_num_tx_queues(adapter->netdev,
2016 					      adapter->num_tx_queues);
2017 	if (retval)
2018 		goto failed;
2019 	retval = netif_set_real_num_rx_queues(adapter->netdev,
2020 					      adapter->num_rx_queues);
2021 	if (retval)
2022 		goto failed;
2023 
2024 	tsnep_enable_irq(adapter, ECM_INT_LINK);
2025 	retval = tsnep_phy_open(adapter);
2026 	if (retval)
2027 		goto phy_failed;
2028 
2029 	for (i = 0; i < adapter->num_queues; i++)
2030 		tsnep_queue_enable(&adapter->queue[i]);
2031 
2032 	return 0;
2033 
2034 phy_failed:
2035 	tsnep_disable_irq(adapter, ECM_INT_LINK);
2036 failed:
2037 	for (i = 0; i < adapter->num_queues; i++) {
2038 		tsnep_queue_close(&adapter->queue[i], i == 0);
2039 
2040 		if (adapter->queue[i].rx)
2041 			tsnep_rx_close(adapter->queue[i].rx);
2042 		if (adapter->queue[i].tx)
2043 			tsnep_tx_close(adapter->queue[i].tx);
2044 	}
2045 	return retval;
2046 }
2047 
2048 static int tsnep_netdev_close(struct net_device *netdev)
2049 {
2050 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2051 	int i;
2052 
2053 	tsnep_disable_irq(adapter, ECM_INT_LINK);
2054 	tsnep_phy_close(adapter);
2055 
2056 	for (i = 0; i < adapter->num_queues; i++) {
2057 		tsnep_queue_disable(&adapter->queue[i]);
2058 
2059 		tsnep_queue_close(&adapter->queue[i], i == 0);
2060 
2061 		if (adapter->queue[i].rx)
2062 			tsnep_rx_close(adapter->queue[i].rx);
2063 		if (adapter->queue[i].tx)
2064 			tsnep_tx_close(adapter->queue[i].tx);
2065 	}
2066 
2067 	return 0;
2068 }
2069 
2070 int tsnep_enable_xsk(struct tsnep_queue *queue, struct xsk_buff_pool *pool)
2071 {
2072 	bool running = netif_running(queue->adapter->netdev);
2073 	u32 frame_size;
2074 
2075 	frame_size = xsk_pool_get_rx_frame_size(pool);
2076 	if (frame_size < TSNEP_XSK_RX_BUF_SIZE)
2077 		return -EOPNOTSUPP;
2078 
2079 	queue->rx->page_buffer = kcalloc(TSNEP_RING_SIZE,
2080 					 sizeof(*queue->rx->page_buffer),
2081 					 GFP_KERNEL);
2082 	if (!queue->rx->page_buffer)
2083 		return -ENOMEM;
2084 	queue->rx->xdp_batch = kcalloc(TSNEP_RING_SIZE,
2085 				       sizeof(*queue->rx->xdp_batch),
2086 				       GFP_KERNEL);
2087 	if (!queue->rx->xdp_batch) {
2088 		kfree(queue->rx->page_buffer);
2089 		queue->rx->page_buffer = NULL;
2090 
2091 		return -ENOMEM;
2092 	}
2093 
2094 	xsk_pool_set_rxq_info(pool, &queue->rx->xdp_rxq_zc);
2095 
2096 	if (running)
2097 		tsnep_queue_disable(queue);
2098 
2099 	queue->tx->xsk_pool = pool;
2100 	queue->rx->xsk_pool = pool;
2101 
2102 	if (running) {
2103 		tsnep_rx_reopen_xsk(queue->rx);
2104 		tsnep_queue_enable(queue);
2105 	}
2106 
2107 	return 0;
2108 }
2109 
2110 void tsnep_disable_xsk(struct tsnep_queue *queue)
2111 {
2112 	bool running = netif_running(queue->adapter->netdev);
2113 
2114 	if (running)
2115 		tsnep_queue_disable(queue);
2116 
2117 	tsnep_rx_free_zc(queue->rx);
2118 
2119 	queue->rx->xsk_pool = NULL;
2120 	queue->tx->xsk_pool = NULL;
2121 
2122 	if (running) {
2123 		tsnep_rx_reopen(queue->rx);
2124 		tsnep_queue_enable(queue);
2125 	}
2126 
2127 	kfree(queue->rx->xdp_batch);
2128 	queue->rx->xdp_batch = NULL;
2129 	kfree(queue->rx->page_buffer);
2130 	queue->rx->page_buffer = NULL;
2131 }
2132 
2133 static netdev_tx_t tsnep_netdev_xmit_frame(struct sk_buff *skb,
2134 					   struct net_device *netdev)
2135 {
2136 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2137 	u16 queue_mapping = skb_get_queue_mapping(skb);
2138 
2139 	if (queue_mapping >= adapter->num_tx_queues)
2140 		queue_mapping = 0;
2141 
2142 	return tsnep_xmit_frame_ring(skb, &adapter->tx[queue_mapping]);
2143 }
2144 
2145 static int tsnep_netdev_ioctl(struct net_device *netdev, struct ifreq *ifr,
2146 			      int cmd)
2147 {
2148 	if (!netif_running(netdev))
2149 		return -EINVAL;
2150 	if (cmd == SIOCSHWTSTAMP || cmd == SIOCGHWTSTAMP)
2151 		return tsnep_ptp_ioctl(netdev, ifr, cmd);
2152 	return phy_mii_ioctl(netdev->phydev, ifr, cmd);
2153 }
2154 
2155 static void tsnep_netdev_set_multicast(struct net_device *netdev)
2156 {
2157 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2158 
2159 	u16 rx_filter = 0;
2160 
2161 	/* configured MAC address and broadcasts are never filtered */
2162 	if (netdev->flags & IFF_PROMISC) {
2163 		rx_filter |= TSNEP_RX_FILTER_ACCEPT_ALL_MULTICASTS;
2164 		rx_filter |= TSNEP_RX_FILTER_ACCEPT_ALL_UNICASTS;
2165 	} else if (!netdev_mc_empty(netdev) || (netdev->flags & IFF_ALLMULTI)) {
2166 		rx_filter |= TSNEP_RX_FILTER_ACCEPT_ALL_MULTICASTS;
2167 	}
2168 	iowrite16(rx_filter, adapter->addr + TSNEP_RX_FILTER);
2169 }
2170 
2171 static void tsnep_netdev_get_stats64(struct net_device *netdev,
2172 				     struct rtnl_link_stats64 *stats)
2173 {
2174 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2175 	u32 reg;
2176 	u32 val;
2177 	int i;
2178 
2179 	for (i = 0; i < adapter->num_tx_queues; i++) {
2180 		stats->tx_packets += adapter->tx[i].packets;
2181 		stats->tx_bytes += adapter->tx[i].bytes;
2182 		stats->tx_dropped += adapter->tx[i].dropped;
2183 	}
2184 	for (i = 0; i < adapter->num_rx_queues; i++) {
2185 		stats->rx_packets += adapter->rx[i].packets;
2186 		stats->rx_bytes += adapter->rx[i].bytes;
2187 		stats->rx_dropped += adapter->rx[i].dropped;
2188 		stats->multicast += adapter->rx[i].multicast;
2189 
2190 		reg = ioread32(adapter->addr + TSNEP_QUEUE(i) +
2191 			       TSNEP_RX_STATISTIC);
2192 		val = (reg & TSNEP_RX_STATISTIC_NO_DESC_MASK) >>
2193 		      TSNEP_RX_STATISTIC_NO_DESC_SHIFT;
2194 		stats->rx_dropped += val;
2195 		val = (reg & TSNEP_RX_STATISTIC_BUFFER_TOO_SMALL_MASK) >>
2196 		      TSNEP_RX_STATISTIC_BUFFER_TOO_SMALL_SHIFT;
2197 		stats->rx_dropped += val;
2198 		val = (reg & TSNEP_RX_STATISTIC_FIFO_OVERFLOW_MASK) >>
2199 		      TSNEP_RX_STATISTIC_FIFO_OVERFLOW_SHIFT;
2200 		stats->rx_errors += val;
2201 		stats->rx_fifo_errors += val;
2202 		val = (reg & TSNEP_RX_STATISTIC_INVALID_FRAME_MASK) >>
2203 		      TSNEP_RX_STATISTIC_INVALID_FRAME_SHIFT;
2204 		stats->rx_errors += val;
2205 		stats->rx_frame_errors += val;
2206 	}
2207 
2208 	reg = ioread32(adapter->addr + ECM_STAT);
2209 	val = (reg & ECM_STAT_RX_ERR_MASK) >> ECM_STAT_RX_ERR_SHIFT;
2210 	stats->rx_errors += val;
2211 	val = (reg & ECM_STAT_INV_FRM_MASK) >> ECM_STAT_INV_FRM_SHIFT;
2212 	stats->rx_errors += val;
2213 	stats->rx_crc_errors += val;
2214 	val = (reg & ECM_STAT_FWD_RX_ERR_MASK) >> ECM_STAT_FWD_RX_ERR_SHIFT;
2215 	stats->rx_errors += val;
2216 }
2217 
2218 static void tsnep_mac_set_address(struct tsnep_adapter *adapter, u8 *addr)
2219 {
2220 	iowrite32(*(u32 *)addr, adapter->addr + TSNEP_MAC_ADDRESS_LOW);
2221 	iowrite16(*(u16 *)(addr + sizeof(u32)),
2222 		  adapter->addr + TSNEP_MAC_ADDRESS_HIGH);
2223 
2224 	ether_addr_copy(adapter->mac_address, addr);
2225 	netif_info(adapter, drv, adapter->netdev, "MAC address set to %pM\n",
2226 		   addr);
2227 }
2228 
2229 static int tsnep_netdev_set_mac_address(struct net_device *netdev, void *addr)
2230 {
2231 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2232 	struct sockaddr *sock_addr = addr;
2233 	int retval;
2234 
2235 	retval = eth_prepare_mac_addr_change(netdev, sock_addr);
2236 	if (retval)
2237 		return retval;
2238 	eth_hw_addr_set(netdev, sock_addr->sa_data);
2239 	tsnep_mac_set_address(adapter, sock_addr->sa_data);
2240 
2241 	return 0;
2242 }
2243 
2244 static int tsnep_netdev_set_features(struct net_device *netdev,
2245 				     netdev_features_t features)
2246 {
2247 	struct tsnep_adapter *adapter = netdev_priv(netdev);
2248 	netdev_features_t changed = netdev->features ^ features;
2249 	bool enable;
2250 	int retval = 0;
2251 
2252 	if (changed & NETIF_F_LOOPBACK) {
2253 		enable = !!(features & NETIF_F_LOOPBACK);
2254 		retval = tsnep_phy_loopback(adapter, enable);
2255 	}
2256 
2257 	return retval;
2258 }
2259 
2260 static ktime_t tsnep_netdev_get_tstamp(struct net_device *netdev,
2261 				       const struct skb_shared_hwtstamps *hwtstamps,
2262 				       bool cycles)
2263 {
2264 	struct tsnep_rx_inline *rx_inline = hwtstamps->netdev_data;
2265 	u64 timestamp;
2266 
2267 	if (cycles)
2268 		timestamp = __le64_to_cpu(rx_inline->counter);
2269 	else
2270 		timestamp = __le64_to_cpu(rx_inline->timestamp);
2271 
2272 	return ns_to_ktime(timestamp);
2273 }
2274 
2275 static int tsnep_netdev_bpf(struct net_device *dev, struct netdev_bpf *bpf)
2276 {
2277 	struct tsnep_adapter *adapter = netdev_priv(dev);
2278 
2279 	switch (bpf->command) {
2280 	case XDP_SETUP_PROG:
2281 		return tsnep_xdp_setup_prog(adapter, bpf->prog, bpf->extack);
2282 	case XDP_SETUP_XSK_POOL:
2283 		return tsnep_xdp_setup_pool(adapter, bpf->xsk.pool,
2284 					    bpf->xsk.queue_id);
2285 	default:
2286 		return -EOPNOTSUPP;
2287 	}
2288 }
2289 
2290 static struct tsnep_tx *tsnep_xdp_get_tx(struct tsnep_adapter *adapter, u32 cpu)
2291 {
2292 	if (cpu >= TSNEP_MAX_QUEUES)
2293 		cpu &= TSNEP_MAX_QUEUES - 1;
2294 
2295 	while (cpu >= adapter->num_tx_queues)
2296 		cpu -= adapter->num_tx_queues;
2297 
2298 	return &adapter->tx[cpu];
2299 }
2300 
2301 static int tsnep_netdev_xdp_xmit(struct net_device *dev, int n,
2302 				 struct xdp_frame **xdp, u32 flags)
2303 {
2304 	struct tsnep_adapter *adapter = netdev_priv(dev);
2305 	u32 cpu = smp_processor_id();
2306 	struct netdev_queue *nq;
2307 	struct tsnep_tx *tx;
2308 	int nxmit;
2309 	bool xmit;
2310 
2311 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2312 		return -EINVAL;
2313 
2314 	tx = tsnep_xdp_get_tx(adapter, cpu);
2315 	nq = netdev_get_tx_queue(adapter->netdev, tx->queue_index);
2316 
2317 	__netif_tx_lock(nq, cpu);
2318 
2319 	for (nxmit = 0; nxmit < n; nxmit++) {
2320 		xmit = tsnep_xdp_xmit_frame_ring(xdp[nxmit], tx,
2321 						 TSNEP_TX_TYPE_XDP_NDO);
2322 		if (!xmit)
2323 			break;
2324 
2325 		/* avoid transmit queue timeout since we share it with the slow
2326 		 * path
2327 		 */
2328 		txq_trans_cond_update(nq);
2329 	}
2330 
2331 	if (flags & XDP_XMIT_FLUSH)
2332 		tsnep_xdp_xmit_flush(tx);
2333 
2334 	__netif_tx_unlock(nq);
2335 
2336 	return nxmit;
2337 }
2338 
2339 static int tsnep_netdev_xsk_wakeup(struct net_device *dev, u32 queue_id,
2340 				   u32 flags)
2341 {
2342 	struct tsnep_adapter *adapter = netdev_priv(dev);
2343 	struct tsnep_queue *queue;
2344 
2345 	if (queue_id >= adapter->num_rx_queues ||
2346 	    queue_id >= adapter->num_tx_queues)
2347 		return -EINVAL;
2348 
2349 	queue = &adapter->queue[queue_id];
2350 
2351 	if (!napi_if_scheduled_mark_missed(&queue->napi))
2352 		napi_schedule(&queue->napi);
2353 
2354 	return 0;
2355 }
2356 
2357 static const struct net_device_ops tsnep_netdev_ops = {
2358 	.ndo_open = tsnep_netdev_open,
2359 	.ndo_stop = tsnep_netdev_close,
2360 	.ndo_start_xmit = tsnep_netdev_xmit_frame,
2361 	.ndo_eth_ioctl = tsnep_netdev_ioctl,
2362 	.ndo_set_rx_mode = tsnep_netdev_set_multicast,
2363 	.ndo_get_stats64 = tsnep_netdev_get_stats64,
2364 	.ndo_set_mac_address = tsnep_netdev_set_mac_address,
2365 	.ndo_set_features = tsnep_netdev_set_features,
2366 	.ndo_get_tstamp = tsnep_netdev_get_tstamp,
2367 	.ndo_setup_tc = tsnep_tc_setup,
2368 	.ndo_bpf = tsnep_netdev_bpf,
2369 	.ndo_xdp_xmit = tsnep_netdev_xdp_xmit,
2370 	.ndo_xsk_wakeup = tsnep_netdev_xsk_wakeup,
2371 };
2372 
2373 static int tsnep_mac_init(struct tsnep_adapter *adapter)
2374 {
2375 	int retval;
2376 
2377 	/* initialize RX filtering, at least configured MAC address and
2378 	 * broadcast are not filtered
2379 	 */
2380 	iowrite16(0, adapter->addr + TSNEP_RX_FILTER);
2381 
2382 	/* try to get MAC address in the following order:
2383 	 * - device tree
2384 	 * - valid MAC address already set
2385 	 * - MAC address register if valid
2386 	 * - random MAC address
2387 	 */
2388 	retval = of_get_mac_address(adapter->pdev->dev.of_node,
2389 				    adapter->mac_address);
2390 	if (retval == -EPROBE_DEFER)
2391 		return retval;
2392 	if (retval && !is_valid_ether_addr(adapter->mac_address)) {
2393 		*(u32 *)adapter->mac_address =
2394 			ioread32(adapter->addr + TSNEP_MAC_ADDRESS_LOW);
2395 		*(u16 *)(adapter->mac_address + sizeof(u32)) =
2396 			ioread16(adapter->addr + TSNEP_MAC_ADDRESS_HIGH);
2397 		if (!is_valid_ether_addr(adapter->mac_address))
2398 			eth_random_addr(adapter->mac_address);
2399 	}
2400 
2401 	tsnep_mac_set_address(adapter, adapter->mac_address);
2402 	eth_hw_addr_set(adapter->netdev, adapter->mac_address);
2403 
2404 	return 0;
2405 }
2406 
2407 static int tsnep_mdio_init(struct tsnep_adapter *adapter)
2408 {
2409 	struct device_node *np = adapter->pdev->dev.of_node;
2410 	int retval;
2411 
2412 	if (np) {
2413 		np = of_get_child_by_name(np, "mdio");
2414 		if (!np)
2415 			return 0;
2416 
2417 		adapter->suppress_preamble =
2418 			of_property_read_bool(np, "suppress-preamble");
2419 	}
2420 
2421 	adapter->mdiobus = devm_mdiobus_alloc(&adapter->pdev->dev);
2422 	if (!adapter->mdiobus) {
2423 		retval = -ENOMEM;
2424 
2425 		goto out;
2426 	}
2427 
2428 	adapter->mdiobus->priv = (void *)adapter;
2429 	adapter->mdiobus->parent = &adapter->pdev->dev;
2430 	adapter->mdiobus->read = tsnep_mdiobus_read;
2431 	adapter->mdiobus->write = tsnep_mdiobus_write;
2432 	adapter->mdiobus->name = TSNEP "-mdiobus";
2433 	snprintf(adapter->mdiobus->id, MII_BUS_ID_SIZE, "%s",
2434 		 adapter->pdev->name);
2435 
2436 	/* do not scan broadcast address */
2437 	adapter->mdiobus->phy_mask = 0x0000001;
2438 
2439 	retval = of_mdiobus_register(adapter->mdiobus, np);
2440 
2441 out:
2442 	of_node_put(np);
2443 
2444 	return retval;
2445 }
2446 
2447 static int tsnep_phy_init(struct tsnep_adapter *adapter)
2448 {
2449 	struct device_node *phy_node;
2450 	int retval;
2451 
2452 	retval = of_get_phy_mode(adapter->pdev->dev.of_node,
2453 				 &adapter->phy_mode);
2454 	if (retval)
2455 		adapter->phy_mode = PHY_INTERFACE_MODE_GMII;
2456 
2457 	phy_node = of_parse_phandle(adapter->pdev->dev.of_node, "phy-handle",
2458 				    0);
2459 	adapter->phydev = of_phy_find_device(phy_node);
2460 	of_node_put(phy_node);
2461 	if (!adapter->phydev && adapter->mdiobus)
2462 		adapter->phydev = phy_find_first(adapter->mdiobus);
2463 	if (!adapter->phydev)
2464 		return -EIO;
2465 
2466 	return 0;
2467 }
2468 
2469 static int tsnep_queue_init(struct tsnep_adapter *adapter, int queue_count)
2470 {
2471 	u32 irq_mask = ECM_INT_TX_0 | ECM_INT_RX_0;
2472 	char name[8];
2473 	int i;
2474 	int retval;
2475 
2476 	/* one TX/RX queue pair for netdev is mandatory */
2477 	if (platform_irq_count(adapter->pdev) == 1)
2478 		retval = platform_get_irq(adapter->pdev, 0);
2479 	else
2480 		retval = platform_get_irq_byname(adapter->pdev, "mac");
2481 	if (retval < 0)
2482 		return retval;
2483 	adapter->num_tx_queues = 1;
2484 	adapter->num_rx_queues = 1;
2485 	adapter->num_queues = 1;
2486 	adapter->queue[0].adapter = adapter;
2487 	adapter->queue[0].irq = retval;
2488 	adapter->queue[0].tx = &adapter->tx[0];
2489 	adapter->queue[0].tx->adapter = adapter;
2490 	adapter->queue[0].tx->addr = adapter->addr + TSNEP_QUEUE(0);
2491 	adapter->queue[0].tx->queue_index = 0;
2492 	adapter->queue[0].rx = &adapter->rx[0];
2493 	adapter->queue[0].rx->adapter = adapter;
2494 	adapter->queue[0].rx->addr = adapter->addr + TSNEP_QUEUE(0);
2495 	adapter->queue[0].rx->queue_index = 0;
2496 	adapter->queue[0].irq_mask = irq_mask;
2497 	adapter->queue[0].irq_delay_addr = adapter->addr + ECM_INT_DELAY;
2498 	retval = tsnep_set_irq_coalesce(&adapter->queue[0],
2499 					TSNEP_COALESCE_USECS_DEFAULT);
2500 	if (retval < 0)
2501 		return retval;
2502 
2503 	adapter->netdev->irq = adapter->queue[0].irq;
2504 
2505 	/* add additional TX/RX queue pairs only if dedicated interrupt is
2506 	 * available
2507 	 */
2508 	for (i = 1; i < queue_count; i++) {
2509 		sprintf(name, "txrx-%d", i);
2510 		retval = platform_get_irq_byname_optional(adapter->pdev, name);
2511 		if (retval < 0)
2512 			break;
2513 
2514 		adapter->num_tx_queues++;
2515 		adapter->num_rx_queues++;
2516 		adapter->num_queues++;
2517 		adapter->queue[i].adapter = adapter;
2518 		adapter->queue[i].irq = retval;
2519 		adapter->queue[i].tx = &adapter->tx[i];
2520 		adapter->queue[i].tx->adapter = adapter;
2521 		adapter->queue[i].tx->addr = adapter->addr + TSNEP_QUEUE(i);
2522 		adapter->queue[i].tx->queue_index = i;
2523 		adapter->queue[i].rx = &adapter->rx[i];
2524 		adapter->queue[i].rx->adapter = adapter;
2525 		adapter->queue[i].rx->addr = adapter->addr + TSNEP_QUEUE(i);
2526 		adapter->queue[i].rx->queue_index = i;
2527 		adapter->queue[i].irq_mask =
2528 			irq_mask << (ECM_INT_TXRX_SHIFT * i);
2529 		adapter->queue[i].irq_delay_addr =
2530 			adapter->addr + ECM_INT_DELAY + ECM_INT_DELAY_OFFSET * i;
2531 		retval = tsnep_set_irq_coalesce(&adapter->queue[i],
2532 						TSNEP_COALESCE_USECS_DEFAULT);
2533 		if (retval < 0)
2534 			return retval;
2535 	}
2536 
2537 	return 0;
2538 }
2539 
2540 static int tsnep_probe(struct platform_device *pdev)
2541 {
2542 	struct tsnep_adapter *adapter;
2543 	struct net_device *netdev;
2544 	struct resource *io;
2545 	u32 type;
2546 	int revision;
2547 	int version;
2548 	int queue_count;
2549 	int retval;
2550 
2551 	netdev = devm_alloc_etherdev_mqs(&pdev->dev,
2552 					 sizeof(struct tsnep_adapter),
2553 					 TSNEP_MAX_QUEUES, TSNEP_MAX_QUEUES);
2554 	if (!netdev)
2555 		return -ENODEV;
2556 	SET_NETDEV_DEV(netdev, &pdev->dev);
2557 	adapter = netdev_priv(netdev);
2558 	platform_set_drvdata(pdev, adapter);
2559 	adapter->pdev = pdev;
2560 	adapter->dmadev = &pdev->dev;
2561 	adapter->netdev = netdev;
2562 	adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE |
2563 			      NETIF_MSG_LINK | NETIF_MSG_IFUP |
2564 			      NETIF_MSG_IFDOWN | NETIF_MSG_TX_QUEUED;
2565 
2566 	netdev->min_mtu = ETH_MIN_MTU;
2567 	netdev->max_mtu = TSNEP_MAX_FRAME_SIZE;
2568 
2569 	mutex_init(&adapter->gate_control_lock);
2570 	mutex_init(&adapter->rxnfc_lock);
2571 	INIT_LIST_HEAD(&adapter->rxnfc_rules);
2572 
2573 	io = platform_get_resource(pdev, IORESOURCE_MEM, 0);
2574 	adapter->addr = devm_ioremap_resource(&pdev->dev, 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