1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Driver for Marvell Discovery (MV643XX) and Marvell Orion ethernet ports
4  * Copyright (C) 2002 Matthew Dharm <mdharm@momenco.com>
5  *
6  * Based on the 64360 driver from:
7  * Copyright (C) 2002 Rabeeh Khoury <rabeeh@galileo.co.il>
8  *		      Rabeeh Khoury <rabeeh@marvell.com>
9  *
10  * Copyright (C) 2003 PMC-Sierra, Inc.,
11  *	written by Manish Lachwani
12  *
13  * Copyright (C) 2003 Ralf Baechle <ralf@linux-mips.org>
14  *
15  * Copyright (C) 2004-2006 MontaVista Software, Inc.
16  *			   Dale Farnsworth <dale@farnsworth.org>
17  *
18  * Copyright (C) 2004 Steven J. Hill <sjhill1@rockwellcollins.com>
19  *				     <sjhill@realitydiluted.com>
20  *
21  * Copyright (C) 2007-2008 Marvell Semiconductor
22  *			   Lennert Buytenhek <buytenh@marvell.com>
23  *
24  * Copyright (C) 2013 Michael Stapelberg <michael@stapelberg.de>
25  */
26 
27 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
28 
29 #include <linux/init.h>
30 #include <linux/dma-mapping.h>
31 #include <linux/in.h>
32 #include <linux/ip.h>
33 #include <net/tso.h>
34 #include <linux/tcp.h>
35 #include <linux/udp.h>
36 #include <linux/etherdevice.h>
37 #include <linux/delay.h>
38 #include <linux/ethtool.h>
39 #include <linux/platform_device.h>
40 #include <linux/module.h>
41 #include <linux/kernel.h>
42 #include <linux/spinlock.h>
43 #include <linux/workqueue.h>
44 #include <linux/phy.h>
45 #include <linux/mv643xx_eth.h>
46 #include <linux/io.h>
47 #include <linux/interrupt.h>
48 #include <linux/types.h>
49 #include <linux/slab.h>
50 #include <linux/clk.h>
51 #include <linux/of.h>
52 #include <linux/of_irq.h>
53 #include <linux/of_net.h>
54 #include <linux/of_mdio.h>
55 
56 static char mv643xx_eth_driver_name[] = "mv643xx_eth";
57 static char mv643xx_eth_driver_version[] = "1.4";
58 
59 
60 /*
61  * Registers shared between all ports.
62  */
63 #define PHY_ADDR			0x0000
64 #define WINDOW_BASE(w)			(0x0200 + ((w) << 3))
65 #define WINDOW_SIZE(w)			(0x0204 + ((w) << 3))
66 #define WINDOW_REMAP_HIGH(w)		(0x0280 + ((w) << 2))
67 #define WINDOW_BAR_ENABLE		0x0290
68 #define WINDOW_PROTECT(w)		(0x0294 + ((w) << 4))
69 
70 /*
71  * Main per-port registers.  These live at offset 0x0400 for
72  * port #0, 0x0800 for port #1, and 0x0c00 for port #2.
73  */
74 #define PORT_CONFIG			0x0000
75 #define  UNICAST_PROMISCUOUS_MODE	0x00000001
76 #define PORT_CONFIG_EXT			0x0004
77 #define MAC_ADDR_LOW			0x0014
78 #define MAC_ADDR_HIGH			0x0018
79 #define SDMA_CONFIG			0x001c
80 #define  TX_BURST_SIZE_16_64BIT		0x01000000
81 #define  TX_BURST_SIZE_4_64BIT		0x00800000
82 #define  BLM_TX_NO_SWAP			0x00000020
83 #define  BLM_RX_NO_SWAP			0x00000010
84 #define  RX_BURST_SIZE_16_64BIT		0x00000008
85 #define  RX_BURST_SIZE_4_64BIT		0x00000004
86 #define PORT_SERIAL_CONTROL		0x003c
87 #define  SET_MII_SPEED_TO_100		0x01000000
88 #define  SET_GMII_SPEED_TO_1000		0x00800000
89 #define  SET_FULL_DUPLEX_MODE		0x00200000
90 #define  MAX_RX_PACKET_9700BYTE		0x000a0000
91 #define  DISABLE_AUTO_NEG_SPEED_GMII	0x00002000
92 #define  DO_NOT_FORCE_LINK_FAIL		0x00000400
93 #define  SERIAL_PORT_CONTROL_RESERVED	0x00000200
94 #define  DISABLE_AUTO_NEG_FOR_FLOW_CTRL	0x00000008
95 #define  DISABLE_AUTO_NEG_FOR_DUPLEX	0x00000004
96 #define  FORCE_LINK_PASS		0x00000002
97 #define  SERIAL_PORT_ENABLE		0x00000001
98 #define PORT_STATUS			0x0044
99 #define  TX_FIFO_EMPTY			0x00000400
100 #define  TX_IN_PROGRESS			0x00000080
101 #define  PORT_SPEED_MASK		0x00000030
102 #define  PORT_SPEED_1000		0x00000010
103 #define  PORT_SPEED_100			0x00000020
104 #define  PORT_SPEED_10			0x00000000
105 #define  FLOW_CONTROL_ENABLED		0x00000008
106 #define  FULL_DUPLEX			0x00000004
107 #define  LINK_UP			0x00000002
108 #define TXQ_COMMAND			0x0048
109 #define TXQ_FIX_PRIO_CONF		0x004c
110 #define PORT_SERIAL_CONTROL1		0x004c
111 #define  CLK125_BYPASS_EN		0x00000010
112 #define TX_BW_RATE			0x0050
113 #define TX_BW_MTU			0x0058
114 #define TX_BW_BURST			0x005c
115 #define INT_CAUSE			0x0060
116 #define  INT_TX_END			0x07f80000
117 #define  INT_TX_END_0			0x00080000
118 #define  INT_RX				0x000003fc
119 #define  INT_RX_0			0x00000004
120 #define  INT_EXT			0x00000002
121 #define INT_CAUSE_EXT			0x0064
122 #define  INT_EXT_LINK_PHY		0x00110000
123 #define  INT_EXT_TX			0x000000ff
124 #define INT_MASK			0x0068
125 #define INT_MASK_EXT			0x006c
126 #define TX_FIFO_URGENT_THRESHOLD	0x0074
127 #define RX_DISCARD_FRAME_CNT		0x0084
128 #define RX_OVERRUN_FRAME_CNT		0x0088
129 #define TXQ_FIX_PRIO_CONF_MOVED		0x00dc
130 #define TX_BW_RATE_MOVED		0x00e0
131 #define TX_BW_MTU_MOVED			0x00e8
132 #define TX_BW_BURST_MOVED		0x00ec
133 #define RXQ_CURRENT_DESC_PTR(q)		(0x020c + ((q) << 4))
134 #define RXQ_COMMAND			0x0280
135 #define TXQ_CURRENT_DESC_PTR(q)		(0x02c0 + ((q) << 2))
136 #define TXQ_BW_TOKENS(q)		(0x0300 + ((q) << 4))
137 #define TXQ_BW_CONF(q)			(0x0304 + ((q) << 4))
138 #define TXQ_BW_WRR_CONF(q)		(0x0308 + ((q) << 4))
139 
140 /*
141  * Misc per-port registers.
142  */
143 #define MIB_COUNTERS(p)			(0x1000 + ((p) << 7))
144 #define SPECIAL_MCAST_TABLE(p)		(0x1400 + ((p) << 10))
145 #define OTHER_MCAST_TABLE(p)		(0x1500 + ((p) << 10))
146 #define UNICAST_TABLE(p)		(0x1600 + ((p) << 10))
147 
148 
149 /*
150  * SDMA configuration register default value.
151  */
152 #if defined(__BIG_ENDIAN)
153 #define PORT_SDMA_CONFIG_DEFAULT_VALUE		\
154 		(RX_BURST_SIZE_4_64BIT	|	\
155 		 TX_BURST_SIZE_4_64BIT)
156 #elif defined(__LITTLE_ENDIAN)
157 #define PORT_SDMA_CONFIG_DEFAULT_VALUE		\
158 		(RX_BURST_SIZE_4_64BIT	|	\
159 		 BLM_RX_NO_SWAP		|	\
160 		 BLM_TX_NO_SWAP		|	\
161 		 TX_BURST_SIZE_4_64BIT)
162 #else
163 #error One of __BIG_ENDIAN or __LITTLE_ENDIAN must be defined
164 #endif
165 
166 
167 /*
168  * Misc definitions.
169  */
170 #define DEFAULT_RX_QUEUE_SIZE	128
171 #define DEFAULT_TX_QUEUE_SIZE	512
172 #define SKB_DMA_REALIGN		((PAGE_SIZE - NET_SKB_PAD) % SMP_CACHE_BYTES)
173 
174 /* Max number of allowed TCP segments for software TSO */
175 #define MV643XX_MAX_TSO_SEGS 100
176 #define MV643XX_MAX_SKB_DESCS (MV643XX_MAX_TSO_SEGS * 2 + MAX_SKB_FRAGS)
177 
178 #define IS_TSO_HEADER(txq, addr) \
179 	((addr >= txq->tso_hdrs_dma) && \
180 	 (addr < txq->tso_hdrs_dma + txq->tx_ring_size * TSO_HEADER_SIZE))
181 
182 #define DESC_DMA_MAP_SINGLE 0
183 #define DESC_DMA_MAP_PAGE 1
184 
185 /*
186  * RX/TX descriptors.
187  */
188 #if defined(__BIG_ENDIAN)
189 struct rx_desc {
190 	u16 byte_cnt;		/* Descriptor buffer byte count		*/
191 	u16 buf_size;		/* Buffer size				*/
192 	u32 cmd_sts;		/* Descriptor command status		*/
193 	u32 next_desc_ptr;	/* Next descriptor pointer		*/
194 	u32 buf_ptr;		/* Descriptor buffer pointer		*/
195 };
196 
197 struct tx_desc {
198 	u16 byte_cnt;		/* buffer byte count			*/
199 	u16 l4i_chk;		/* CPU provided TCP checksum		*/
200 	u32 cmd_sts;		/* Command/status field			*/
201 	u32 next_desc_ptr;	/* Pointer to next descriptor		*/
202 	u32 buf_ptr;		/* pointer to buffer for this descriptor*/
203 };
204 #elif defined(__LITTLE_ENDIAN)
205 struct rx_desc {
206 	u32 cmd_sts;		/* Descriptor command status		*/
207 	u16 buf_size;		/* Buffer size				*/
208 	u16 byte_cnt;		/* Descriptor buffer byte count		*/
209 	u32 buf_ptr;		/* Descriptor buffer pointer		*/
210 	u32 next_desc_ptr;	/* Next descriptor pointer		*/
211 };
212 
213 struct tx_desc {
214 	u32 cmd_sts;		/* Command/status field			*/
215 	u16 l4i_chk;		/* CPU provided TCP checksum		*/
216 	u16 byte_cnt;		/* buffer byte count			*/
217 	u32 buf_ptr;		/* pointer to buffer for this descriptor*/
218 	u32 next_desc_ptr;	/* Pointer to next descriptor		*/
219 };
220 #else
221 #error One of __BIG_ENDIAN or __LITTLE_ENDIAN must be defined
222 #endif
223 
224 /* RX & TX descriptor command */
225 #define BUFFER_OWNED_BY_DMA		0x80000000
226 
227 /* RX & TX descriptor status */
228 #define ERROR_SUMMARY			0x00000001
229 
230 /* RX descriptor status */
231 #define LAYER_4_CHECKSUM_OK		0x40000000
232 #define RX_ENABLE_INTERRUPT		0x20000000
233 #define RX_FIRST_DESC			0x08000000
234 #define RX_LAST_DESC			0x04000000
235 #define RX_IP_HDR_OK			0x02000000
236 #define RX_PKT_IS_IPV4			0x01000000
237 #define RX_PKT_IS_ETHERNETV2		0x00800000
238 #define RX_PKT_LAYER4_TYPE_MASK		0x00600000
239 #define RX_PKT_LAYER4_TYPE_TCP_IPV4	0x00000000
240 #define RX_PKT_IS_VLAN_TAGGED		0x00080000
241 
242 /* TX descriptor command */
243 #define TX_ENABLE_INTERRUPT		0x00800000
244 #define GEN_CRC				0x00400000
245 #define TX_FIRST_DESC			0x00200000
246 #define TX_LAST_DESC			0x00100000
247 #define ZERO_PADDING			0x00080000
248 #define GEN_IP_V4_CHECKSUM		0x00040000
249 #define GEN_TCP_UDP_CHECKSUM		0x00020000
250 #define UDP_FRAME			0x00010000
251 #define MAC_HDR_EXTRA_4_BYTES		0x00008000
252 #define GEN_TCP_UDP_CHK_FULL		0x00000400
253 #define MAC_HDR_EXTRA_8_BYTES		0x00000200
254 
255 #define TX_IHL_SHIFT			11
256 
257 
258 /* global *******************************************************************/
259 struct mv643xx_eth_shared_private {
260 	/*
261 	 * Ethernet controller base address.
262 	 */
263 	void __iomem *base;
264 
265 	/*
266 	 * Per-port MBUS window access register value.
267 	 */
268 	u32 win_protect;
269 
270 	/*
271 	 * Hardware-specific parameters.
272 	 */
273 	int extended_rx_coal_limit;
274 	int tx_bw_control;
275 	int tx_csum_limit;
276 	struct clk *clk;
277 };
278 
279 #define TX_BW_CONTROL_ABSENT		0
280 #define TX_BW_CONTROL_OLD_LAYOUT	1
281 #define TX_BW_CONTROL_NEW_LAYOUT	2
282 
283 static int mv643xx_eth_open(struct net_device *dev);
284 static int mv643xx_eth_stop(struct net_device *dev);
285 
286 
287 /* per-port *****************************************************************/
288 struct mib_counters {
289 	u64 good_octets_received;
290 	u32 bad_octets_received;
291 	u32 internal_mac_transmit_err;
292 	u32 good_frames_received;
293 	u32 bad_frames_received;
294 	u32 broadcast_frames_received;
295 	u32 multicast_frames_received;
296 	u32 frames_64_octets;
297 	u32 frames_65_to_127_octets;
298 	u32 frames_128_to_255_octets;
299 	u32 frames_256_to_511_octets;
300 	u32 frames_512_to_1023_octets;
301 	u32 frames_1024_to_max_octets;
302 	u64 good_octets_sent;
303 	u32 good_frames_sent;
304 	u32 excessive_collision;
305 	u32 multicast_frames_sent;
306 	u32 broadcast_frames_sent;
307 	u32 unrec_mac_control_received;
308 	u32 fc_sent;
309 	u32 good_fc_received;
310 	u32 bad_fc_received;
311 	u32 undersize_received;
312 	u32 fragments_received;
313 	u32 oversize_received;
314 	u32 jabber_received;
315 	u32 mac_receive_error;
316 	u32 bad_crc_event;
317 	u32 collision;
318 	u32 late_collision;
319 	/* Non MIB hardware counters */
320 	u32 rx_discard;
321 	u32 rx_overrun;
322 };
323 
324 struct rx_queue {
325 	int index;
326 
327 	int rx_ring_size;
328 
329 	int rx_desc_count;
330 	int rx_curr_desc;
331 	int rx_used_desc;
332 
333 	struct rx_desc *rx_desc_area;
334 	dma_addr_t rx_desc_dma;
335 	int rx_desc_area_size;
336 	struct sk_buff **rx_skb;
337 };
338 
339 struct tx_queue {
340 	int index;
341 
342 	int tx_ring_size;
343 
344 	int tx_desc_count;
345 	int tx_curr_desc;
346 	int tx_used_desc;
347 
348 	int tx_stop_threshold;
349 	int tx_wake_threshold;
350 
351 	char *tso_hdrs;
352 	dma_addr_t tso_hdrs_dma;
353 
354 	struct tx_desc *tx_desc_area;
355 	char *tx_desc_mapping; /* array to track the type of the dma mapping */
356 	dma_addr_t tx_desc_dma;
357 	int tx_desc_area_size;
358 
359 	struct sk_buff_head tx_skb;
360 
361 	unsigned long tx_packets;
362 	unsigned long tx_bytes;
363 	unsigned long tx_dropped;
364 };
365 
366 struct mv643xx_eth_private {
367 	struct mv643xx_eth_shared_private *shared;
368 	void __iomem *base;
369 	int port_num;
370 
371 	struct net_device *dev;
372 
373 	struct timer_list mib_counters_timer;
374 	spinlock_t mib_counters_lock;
375 	struct mib_counters mib_counters;
376 
377 	struct work_struct tx_timeout_task;
378 
379 	struct napi_struct napi;
380 	u32 int_mask;
381 	u8 oom;
382 	u8 work_link;
383 	u8 work_tx;
384 	u8 work_tx_end;
385 	u8 work_rx;
386 	u8 work_rx_refill;
387 
388 	int skb_size;
389 
390 	/*
391 	 * RX state.
392 	 */
393 	int rx_ring_size;
394 	unsigned long rx_desc_sram_addr;
395 	int rx_desc_sram_size;
396 	int rxq_count;
397 	struct timer_list rx_oom;
398 	struct rx_queue rxq[8];
399 
400 	/*
401 	 * TX state.
402 	 */
403 	int tx_ring_size;
404 	unsigned long tx_desc_sram_addr;
405 	int tx_desc_sram_size;
406 	int txq_count;
407 	struct tx_queue txq[8];
408 
409 	/*
410 	 * Hardware-specific parameters.
411 	 */
412 	struct clk *clk;
413 	unsigned int t_clk;
414 };
415 
416 
417 /* port register accessors **************************************************/
418 static inline u32 rdl(struct mv643xx_eth_private *mp, int offset)
419 {
420 	return readl(mp->shared->base + offset);
421 }
422 
423 static inline u32 rdlp(struct mv643xx_eth_private *mp, int offset)
424 {
425 	return readl(mp->base + offset);
426 }
427 
428 static inline void wrl(struct mv643xx_eth_private *mp, int offset, u32 data)
429 {
430 	writel(data, mp->shared->base + offset);
431 }
432 
433 static inline void wrlp(struct mv643xx_eth_private *mp, int offset, u32 data)
434 {
435 	writel(data, mp->base + offset);
436 }
437 
438 
439 /* rxq/txq helper functions *************************************************/
440 static struct mv643xx_eth_private *rxq_to_mp(struct rx_queue *rxq)
441 {
442 	return container_of(rxq, struct mv643xx_eth_private, rxq[rxq->index]);
443 }
444 
445 static struct mv643xx_eth_private *txq_to_mp(struct tx_queue *txq)
446 {
447 	return container_of(txq, struct mv643xx_eth_private, txq[txq->index]);
448 }
449 
450 static void rxq_enable(struct rx_queue *rxq)
451 {
452 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
453 	wrlp(mp, RXQ_COMMAND, 1 << rxq->index);
454 }
455 
456 static void rxq_disable(struct rx_queue *rxq)
457 {
458 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
459 	u8 mask = 1 << rxq->index;
460 
461 	wrlp(mp, RXQ_COMMAND, mask << 8);
462 	while (rdlp(mp, RXQ_COMMAND) & mask)
463 		udelay(10);
464 }
465 
466 static void txq_reset_hw_ptr(struct tx_queue *txq)
467 {
468 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
469 	u32 addr;
470 
471 	addr = (u32)txq->tx_desc_dma;
472 	addr += txq->tx_curr_desc * sizeof(struct tx_desc);
473 	wrlp(mp, TXQ_CURRENT_DESC_PTR(txq->index), addr);
474 }
475 
476 static void txq_enable(struct tx_queue *txq)
477 {
478 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
479 	wrlp(mp, TXQ_COMMAND, 1 << txq->index);
480 }
481 
482 static void txq_disable(struct tx_queue *txq)
483 {
484 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
485 	u8 mask = 1 << txq->index;
486 
487 	wrlp(mp, TXQ_COMMAND, mask << 8);
488 	while (rdlp(mp, TXQ_COMMAND) & mask)
489 		udelay(10);
490 }
491 
492 static void txq_maybe_wake(struct tx_queue *txq)
493 {
494 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
495 	struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
496 
497 	if (netif_tx_queue_stopped(nq)) {
498 		__netif_tx_lock(nq, smp_processor_id());
499 		if (txq->tx_desc_count <= txq->tx_wake_threshold)
500 			netif_tx_wake_queue(nq);
501 		__netif_tx_unlock(nq);
502 	}
503 }
504 
505 static int rxq_process(struct rx_queue *rxq, int budget)
506 {
507 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
508 	struct net_device_stats *stats = &mp->dev->stats;
509 	int rx;
510 
511 	rx = 0;
512 	while (rx < budget && rxq->rx_desc_count) {
513 		struct rx_desc *rx_desc;
514 		unsigned int cmd_sts;
515 		struct sk_buff *skb;
516 		u16 byte_cnt;
517 
518 		rx_desc = &rxq->rx_desc_area[rxq->rx_curr_desc];
519 
520 		cmd_sts = rx_desc->cmd_sts;
521 		if (cmd_sts & BUFFER_OWNED_BY_DMA)
522 			break;
523 		rmb();
524 
525 		skb = rxq->rx_skb[rxq->rx_curr_desc];
526 		rxq->rx_skb[rxq->rx_curr_desc] = NULL;
527 
528 		rxq->rx_curr_desc++;
529 		if (rxq->rx_curr_desc == rxq->rx_ring_size)
530 			rxq->rx_curr_desc = 0;
531 
532 		dma_unmap_single(mp->dev->dev.parent, rx_desc->buf_ptr,
533 				 rx_desc->buf_size, DMA_FROM_DEVICE);
534 		rxq->rx_desc_count--;
535 		rx++;
536 
537 		mp->work_rx_refill |= 1 << rxq->index;
538 
539 		byte_cnt = rx_desc->byte_cnt;
540 
541 		/*
542 		 * Update statistics.
543 		 *
544 		 * Note that the descriptor byte count includes 2 dummy
545 		 * bytes automatically inserted by the hardware at the
546 		 * start of the packet (which we don't count), and a 4
547 		 * byte CRC at the end of the packet (which we do count).
548 		 */
549 		stats->rx_packets++;
550 		stats->rx_bytes += byte_cnt - 2;
551 
552 		/*
553 		 * In case we received a packet without first / last bits
554 		 * on, or the error summary bit is set, the packet needs
555 		 * to be dropped.
556 		 */
557 		if ((cmd_sts & (RX_FIRST_DESC | RX_LAST_DESC | ERROR_SUMMARY))
558 			!= (RX_FIRST_DESC | RX_LAST_DESC))
559 			goto err;
560 
561 		/*
562 		 * The -4 is for the CRC in the trailer of the
563 		 * received packet
564 		 */
565 		skb_put(skb, byte_cnt - 2 - 4);
566 
567 		if (cmd_sts & LAYER_4_CHECKSUM_OK)
568 			skb->ip_summed = CHECKSUM_UNNECESSARY;
569 		skb->protocol = eth_type_trans(skb, mp->dev);
570 
571 		napi_gro_receive(&mp->napi, skb);
572 
573 		continue;
574 
575 err:
576 		stats->rx_dropped++;
577 
578 		if ((cmd_sts & (RX_FIRST_DESC | RX_LAST_DESC)) !=
579 			(RX_FIRST_DESC | RX_LAST_DESC)) {
580 			if (net_ratelimit())
581 				netdev_err(mp->dev,
582 					   "received packet spanning multiple descriptors\n");
583 		}
584 
585 		if (cmd_sts & ERROR_SUMMARY)
586 			stats->rx_errors++;
587 
588 		dev_kfree_skb(skb);
589 	}
590 
591 	if (rx < budget)
592 		mp->work_rx &= ~(1 << rxq->index);
593 
594 	return rx;
595 }
596 
597 static int rxq_refill(struct rx_queue *rxq, int budget)
598 {
599 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
600 	int refilled;
601 
602 	refilled = 0;
603 	while (refilled < budget && rxq->rx_desc_count < rxq->rx_ring_size) {
604 		struct sk_buff *skb;
605 		int rx;
606 		struct rx_desc *rx_desc;
607 		int size;
608 
609 		skb = netdev_alloc_skb(mp->dev, mp->skb_size);
610 
611 		if (skb == NULL) {
612 			mp->oom = 1;
613 			goto oom;
614 		}
615 
616 		if (SKB_DMA_REALIGN)
617 			skb_reserve(skb, SKB_DMA_REALIGN);
618 
619 		refilled++;
620 		rxq->rx_desc_count++;
621 
622 		rx = rxq->rx_used_desc++;
623 		if (rxq->rx_used_desc == rxq->rx_ring_size)
624 			rxq->rx_used_desc = 0;
625 
626 		rx_desc = rxq->rx_desc_area + rx;
627 
628 		size = skb_end_pointer(skb) - skb->data;
629 		rx_desc->buf_ptr = dma_map_single(mp->dev->dev.parent,
630 						  skb->data, size,
631 						  DMA_FROM_DEVICE);
632 		rx_desc->buf_size = size;
633 		rxq->rx_skb[rx] = skb;
634 		wmb();
635 		rx_desc->cmd_sts = BUFFER_OWNED_BY_DMA | RX_ENABLE_INTERRUPT;
636 		wmb();
637 
638 		/*
639 		 * The hardware automatically prepends 2 bytes of
640 		 * dummy data to each received packet, so that the
641 		 * IP header ends up 16-byte aligned.
642 		 */
643 		skb_reserve(skb, 2);
644 	}
645 
646 	if (refilled < budget)
647 		mp->work_rx_refill &= ~(1 << rxq->index);
648 
649 oom:
650 	return refilled;
651 }
652 
653 
654 /* tx ***********************************************************************/
655 static inline unsigned int has_tiny_unaligned_frags(struct sk_buff *skb)
656 {
657 	int frag;
658 
659 	for (frag = 0; frag < skb_shinfo(skb)->nr_frags; frag++) {
660 		const skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];
661 
662 		if (skb_frag_size(fragp) <= 8 && skb_frag_off(fragp) & 7)
663 			return 1;
664 	}
665 
666 	return 0;
667 }
668 
669 static int skb_tx_csum(struct mv643xx_eth_private *mp, struct sk_buff *skb,
670 		       u16 *l4i_chk, u32 *command, int length)
671 {
672 	int ret;
673 	u32 cmd = 0;
674 
675 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
676 		int hdr_len;
677 		int tag_bytes;
678 
679 		BUG_ON(skb->protocol != htons(ETH_P_IP) &&
680 		       skb->protocol != htons(ETH_P_8021Q));
681 
682 		hdr_len = (void *)ip_hdr(skb) - (void *)skb->data;
683 		tag_bytes = hdr_len - ETH_HLEN;
684 
685 		if (length - hdr_len > mp->shared->tx_csum_limit ||
686 		    unlikely(tag_bytes & ~12)) {
687 			ret = skb_checksum_help(skb);
688 			if (!ret)
689 				goto no_csum;
690 			return ret;
691 		}
692 
693 		if (tag_bytes & 4)
694 			cmd |= MAC_HDR_EXTRA_4_BYTES;
695 		if (tag_bytes & 8)
696 			cmd |= MAC_HDR_EXTRA_8_BYTES;
697 
698 		cmd |= GEN_TCP_UDP_CHECKSUM | GEN_TCP_UDP_CHK_FULL |
699 			   GEN_IP_V4_CHECKSUM   |
700 			   ip_hdr(skb)->ihl << TX_IHL_SHIFT;
701 
702 		/* TODO: Revisit this. With the usage of GEN_TCP_UDP_CHK_FULL
703 		 * it seems we don't need to pass the initial checksum.
704 		 */
705 		switch (ip_hdr(skb)->protocol) {
706 		case IPPROTO_UDP:
707 			cmd |= UDP_FRAME;
708 			*l4i_chk = 0;
709 			break;
710 		case IPPROTO_TCP:
711 			*l4i_chk = 0;
712 			break;
713 		default:
714 			WARN(1, "protocol not supported");
715 		}
716 	} else {
717 no_csum:
718 		/* Errata BTS #50, IHL must be 5 if no HW checksum */
719 		cmd |= 5 << TX_IHL_SHIFT;
720 	}
721 	*command = cmd;
722 	return 0;
723 }
724 
725 static inline int
726 txq_put_data_tso(struct net_device *dev, struct tx_queue *txq,
727 		 struct sk_buff *skb, char *data, int length,
728 		 bool last_tcp, bool is_last)
729 {
730 	int tx_index;
731 	u32 cmd_sts;
732 	struct tx_desc *desc;
733 
734 	tx_index = txq->tx_curr_desc++;
735 	if (txq->tx_curr_desc == txq->tx_ring_size)
736 		txq->tx_curr_desc = 0;
737 	desc = &txq->tx_desc_area[tx_index];
738 	txq->tx_desc_mapping[tx_index] = DESC_DMA_MAP_SINGLE;
739 
740 	desc->l4i_chk = 0;
741 	desc->byte_cnt = length;
742 
743 	if (length <= 8 && (uintptr_t)data & 0x7) {
744 		/* Copy unaligned small data fragment to TSO header data area */
745 		memcpy(txq->tso_hdrs + tx_index * TSO_HEADER_SIZE,
746 		       data, length);
747 		desc->buf_ptr = txq->tso_hdrs_dma
748 			+ tx_index * TSO_HEADER_SIZE;
749 	} else {
750 		/* Alignment is okay, map buffer and hand off to hardware */
751 		txq->tx_desc_mapping[tx_index] = DESC_DMA_MAP_SINGLE;
752 		desc->buf_ptr = dma_map_single(dev->dev.parent, data,
753 			length, DMA_TO_DEVICE);
754 		if (unlikely(dma_mapping_error(dev->dev.parent,
755 					       desc->buf_ptr))) {
756 			WARN(1, "dma_map_single failed!\n");
757 			return -ENOMEM;
758 		}
759 	}
760 
761 	cmd_sts = BUFFER_OWNED_BY_DMA;
762 	if (last_tcp) {
763 		/* last descriptor in the TCP packet */
764 		cmd_sts |= ZERO_PADDING | TX_LAST_DESC;
765 		/* last descriptor in SKB */
766 		if (is_last)
767 			cmd_sts |= TX_ENABLE_INTERRUPT;
768 	}
769 	desc->cmd_sts = cmd_sts;
770 	return 0;
771 }
772 
773 static inline void
774 txq_put_hdr_tso(struct sk_buff *skb, struct tx_queue *txq, int length,
775 		u32 *first_cmd_sts, bool first_desc)
776 {
777 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
778 	int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
779 	int tx_index;
780 	struct tx_desc *desc;
781 	int ret;
782 	u32 cmd_csum = 0;
783 	u16 l4i_chk = 0;
784 	u32 cmd_sts;
785 
786 	tx_index = txq->tx_curr_desc;
787 	desc = &txq->tx_desc_area[tx_index];
788 
789 	ret = skb_tx_csum(mp, skb, &l4i_chk, &cmd_csum, length);
790 	if (ret)
791 		WARN(1, "failed to prepare checksum!");
792 
793 	/* Should we set this? Can't use the value from skb_tx_csum()
794 	 * as it's not the correct initial L4 checksum to use.
795 	 */
796 	desc->l4i_chk = 0;
797 
798 	desc->byte_cnt = hdr_len;
799 	desc->buf_ptr = txq->tso_hdrs_dma +
800 			txq->tx_curr_desc * TSO_HEADER_SIZE;
801 	cmd_sts = cmd_csum | BUFFER_OWNED_BY_DMA  | TX_FIRST_DESC |
802 				   GEN_CRC;
803 
804 	/* Defer updating the first command descriptor until all
805 	 * following descriptors have been written.
806 	 */
807 	if (first_desc)
808 		*first_cmd_sts = cmd_sts;
809 	else
810 		desc->cmd_sts = cmd_sts;
811 
812 	txq->tx_curr_desc++;
813 	if (txq->tx_curr_desc == txq->tx_ring_size)
814 		txq->tx_curr_desc = 0;
815 }
816 
817 static int txq_submit_tso(struct tx_queue *txq, struct sk_buff *skb,
818 			  struct net_device *dev)
819 {
820 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
821 	int hdr_len, total_len, data_left, ret;
822 	int desc_count = 0;
823 	struct tso_t tso;
824 	struct tx_desc *first_tx_desc;
825 	u32 first_cmd_sts = 0;
826 
827 	/* Count needed descriptors */
828 	if ((txq->tx_desc_count + tso_count_descs(skb)) >= txq->tx_ring_size) {
829 		netdev_dbg(dev, "not enough descriptors for TSO!\n");
830 		return -EBUSY;
831 	}
832 
833 	first_tx_desc = &txq->tx_desc_area[txq->tx_curr_desc];
834 
835 	/* Initialize the TSO handler, and prepare the first payload */
836 	hdr_len = tso_start(skb, &tso);
837 
838 	total_len = skb->len - hdr_len;
839 	while (total_len > 0) {
840 		bool first_desc = (desc_count == 0);
841 		char *hdr;
842 
843 		data_left = min_t(int, skb_shinfo(skb)->gso_size, total_len);
844 		total_len -= data_left;
845 		desc_count++;
846 
847 		/* prepare packet headers: MAC + IP + TCP */
848 		hdr = txq->tso_hdrs + txq->tx_curr_desc * TSO_HEADER_SIZE;
849 		tso_build_hdr(skb, hdr, &tso, data_left, total_len == 0);
850 		txq_put_hdr_tso(skb, txq, data_left, &first_cmd_sts,
851 				first_desc);
852 
853 		while (data_left > 0) {
854 			int size;
855 			desc_count++;
856 
857 			size = min_t(int, tso.size, data_left);
858 			ret = txq_put_data_tso(dev, txq, skb, tso.data, size,
859 					       size == data_left,
860 					       total_len == 0);
861 			if (ret)
862 				goto err_release;
863 			data_left -= size;
864 			tso_build_data(skb, &tso, size);
865 		}
866 	}
867 
868 	__skb_queue_tail(&txq->tx_skb, skb);
869 	skb_tx_timestamp(skb);
870 
871 	/* ensure all other descriptors are written before first cmd_sts */
872 	wmb();
873 	first_tx_desc->cmd_sts = first_cmd_sts;
874 
875 	/* clear TX_END status */
876 	mp->work_tx_end &= ~(1 << txq->index);
877 
878 	/* ensure all descriptors are written before poking hardware */
879 	wmb();
880 	txq_enable(txq);
881 	txq->tx_desc_count += desc_count;
882 	return 0;
883 err_release:
884 	/* TODO: Release all used data descriptors; header descriptors must not
885 	 * be DMA-unmapped.
886 	 */
887 	return ret;
888 }
889 
890 static void txq_submit_frag_skb(struct tx_queue *txq, struct sk_buff *skb)
891 {
892 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
893 	int nr_frags = skb_shinfo(skb)->nr_frags;
894 	int frag;
895 
896 	for (frag = 0; frag < nr_frags; frag++) {
897 		skb_frag_t *this_frag;
898 		int tx_index;
899 		struct tx_desc *desc;
900 
901 		this_frag = &skb_shinfo(skb)->frags[frag];
902 		tx_index = txq->tx_curr_desc++;
903 		if (txq->tx_curr_desc == txq->tx_ring_size)
904 			txq->tx_curr_desc = 0;
905 		desc = &txq->tx_desc_area[tx_index];
906 		txq->tx_desc_mapping[tx_index] = DESC_DMA_MAP_PAGE;
907 
908 		/*
909 		 * The last fragment will generate an interrupt
910 		 * which will free the skb on TX completion.
911 		 */
912 		if (frag == nr_frags - 1) {
913 			desc->cmd_sts = BUFFER_OWNED_BY_DMA |
914 					ZERO_PADDING | TX_LAST_DESC |
915 					TX_ENABLE_INTERRUPT;
916 		} else {
917 			desc->cmd_sts = BUFFER_OWNED_BY_DMA;
918 		}
919 
920 		desc->l4i_chk = 0;
921 		desc->byte_cnt = skb_frag_size(this_frag);
922 		desc->buf_ptr = skb_frag_dma_map(mp->dev->dev.parent,
923 						 this_frag, 0, desc->byte_cnt,
924 						 DMA_TO_DEVICE);
925 	}
926 }
927 
928 static int txq_submit_skb(struct tx_queue *txq, struct sk_buff *skb,
929 			  struct net_device *dev)
930 {
931 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
932 	int nr_frags = skb_shinfo(skb)->nr_frags;
933 	int tx_index;
934 	struct tx_desc *desc;
935 	u32 cmd_sts;
936 	u16 l4i_chk;
937 	int length, ret;
938 
939 	cmd_sts = 0;
940 	l4i_chk = 0;
941 
942 	if (txq->tx_ring_size - txq->tx_desc_count < MAX_SKB_FRAGS + 1) {
943 		if (net_ratelimit())
944 			netdev_err(dev, "tx queue full?!\n");
945 		return -EBUSY;
946 	}
947 
948 	ret = skb_tx_csum(mp, skb, &l4i_chk, &cmd_sts, skb->len);
949 	if (ret)
950 		return ret;
951 	cmd_sts |= TX_FIRST_DESC | GEN_CRC | BUFFER_OWNED_BY_DMA;
952 
953 	tx_index = txq->tx_curr_desc++;
954 	if (txq->tx_curr_desc == txq->tx_ring_size)
955 		txq->tx_curr_desc = 0;
956 	desc = &txq->tx_desc_area[tx_index];
957 	txq->tx_desc_mapping[tx_index] = DESC_DMA_MAP_SINGLE;
958 
959 	if (nr_frags) {
960 		txq_submit_frag_skb(txq, skb);
961 		length = skb_headlen(skb);
962 	} else {
963 		cmd_sts |= ZERO_PADDING | TX_LAST_DESC | TX_ENABLE_INTERRUPT;
964 		length = skb->len;
965 	}
966 
967 	desc->l4i_chk = l4i_chk;
968 	desc->byte_cnt = length;
969 	desc->buf_ptr = dma_map_single(mp->dev->dev.parent, skb->data,
970 				       length, DMA_TO_DEVICE);
971 
972 	__skb_queue_tail(&txq->tx_skb, skb);
973 
974 	skb_tx_timestamp(skb);
975 
976 	/* ensure all other descriptors are written before first cmd_sts */
977 	wmb();
978 	desc->cmd_sts = cmd_sts;
979 
980 	/* clear TX_END status */
981 	mp->work_tx_end &= ~(1 << txq->index);
982 
983 	/* ensure all descriptors are written before poking hardware */
984 	wmb();
985 	txq_enable(txq);
986 
987 	txq->tx_desc_count += nr_frags + 1;
988 
989 	return 0;
990 }
991 
992 static netdev_tx_t mv643xx_eth_xmit(struct sk_buff *skb, struct net_device *dev)
993 {
994 	struct mv643xx_eth_private *mp = netdev_priv(dev);
995 	int length, queue, ret;
996 	struct tx_queue *txq;
997 	struct netdev_queue *nq;
998 
999 	queue = skb_get_queue_mapping(skb);
1000 	txq = mp->txq + queue;
1001 	nq = netdev_get_tx_queue(dev, queue);
1002 
1003 	if (has_tiny_unaligned_frags(skb) && __skb_linearize(skb)) {
1004 		netdev_printk(KERN_DEBUG, dev,
1005 			      "failed to linearize skb with tiny unaligned fragment\n");
1006 		return NETDEV_TX_BUSY;
1007 	}
1008 
1009 	length = skb->len;
1010 
1011 	if (skb_is_gso(skb))
1012 		ret = txq_submit_tso(txq, skb, dev);
1013 	else
1014 		ret = txq_submit_skb(txq, skb, dev);
1015 	if (!ret) {
1016 		txq->tx_bytes += length;
1017 		txq->tx_packets++;
1018 
1019 		if (txq->tx_desc_count >= txq->tx_stop_threshold)
1020 			netif_tx_stop_queue(nq);
1021 	} else {
1022 		txq->tx_dropped++;
1023 		dev_kfree_skb_any(skb);
1024 	}
1025 
1026 	return NETDEV_TX_OK;
1027 }
1028 
1029 
1030 /* tx napi ******************************************************************/
1031 static void txq_kick(struct tx_queue *txq)
1032 {
1033 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
1034 	struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
1035 	u32 hw_desc_ptr;
1036 	u32 expected_ptr;
1037 
1038 	__netif_tx_lock(nq, smp_processor_id());
1039 
1040 	if (rdlp(mp, TXQ_COMMAND) & (1 << txq->index))
1041 		goto out;
1042 
1043 	hw_desc_ptr = rdlp(mp, TXQ_CURRENT_DESC_PTR(txq->index));
1044 	expected_ptr = (u32)txq->tx_desc_dma +
1045 				txq->tx_curr_desc * sizeof(struct tx_desc);
1046 
1047 	if (hw_desc_ptr != expected_ptr)
1048 		txq_enable(txq);
1049 
1050 out:
1051 	__netif_tx_unlock(nq);
1052 
1053 	mp->work_tx_end &= ~(1 << txq->index);
1054 }
1055 
1056 static int txq_reclaim(struct tx_queue *txq, int budget, int force)
1057 {
1058 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
1059 	struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
1060 	int reclaimed;
1061 
1062 	__netif_tx_lock_bh(nq);
1063 
1064 	reclaimed = 0;
1065 	while (reclaimed < budget && txq->tx_desc_count > 0) {
1066 		int tx_index;
1067 		struct tx_desc *desc;
1068 		u32 cmd_sts;
1069 		char desc_dma_map;
1070 
1071 		tx_index = txq->tx_used_desc;
1072 		desc = &txq->tx_desc_area[tx_index];
1073 		desc_dma_map = txq->tx_desc_mapping[tx_index];
1074 
1075 		cmd_sts = desc->cmd_sts;
1076 
1077 		if (cmd_sts & BUFFER_OWNED_BY_DMA) {
1078 			if (!force)
1079 				break;
1080 			desc->cmd_sts = cmd_sts & ~BUFFER_OWNED_BY_DMA;
1081 		}
1082 
1083 		txq->tx_used_desc = tx_index + 1;
1084 		if (txq->tx_used_desc == txq->tx_ring_size)
1085 			txq->tx_used_desc = 0;
1086 
1087 		reclaimed++;
1088 		txq->tx_desc_count--;
1089 
1090 		if (!IS_TSO_HEADER(txq, desc->buf_ptr)) {
1091 
1092 			if (desc_dma_map == DESC_DMA_MAP_PAGE)
1093 				dma_unmap_page(mp->dev->dev.parent,
1094 					       desc->buf_ptr,
1095 					       desc->byte_cnt,
1096 					       DMA_TO_DEVICE);
1097 			else
1098 				dma_unmap_single(mp->dev->dev.parent,
1099 						 desc->buf_ptr,
1100 						 desc->byte_cnt,
1101 						 DMA_TO_DEVICE);
1102 		}
1103 
1104 		if (cmd_sts & TX_ENABLE_INTERRUPT) {
1105 			struct sk_buff *skb = __skb_dequeue(&txq->tx_skb);
1106 
1107 			if (!WARN_ON(!skb))
1108 				dev_consume_skb_any(skb);
1109 		}
1110 
1111 		if (cmd_sts & ERROR_SUMMARY) {
1112 			netdev_info(mp->dev, "tx error\n");
1113 			mp->dev->stats.tx_errors++;
1114 		}
1115 
1116 	}
1117 
1118 	__netif_tx_unlock_bh(nq);
1119 
1120 	if (reclaimed < budget)
1121 		mp->work_tx &= ~(1 << txq->index);
1122 
1123 	return reclaimed;
1124 }
1125 
1126 
1127 /* tx rate control **********************************************************/
1128 /*
1129  * Set total maximum TX rate (shared by all TX queues for this port)
1130  * to 'rate' bits per second, with a maximum burst of 'burst' bytes.
1131  */
1132 static void tx_set_rate(struct mv643xx_eth_private *mp, int rate, int burst)
1133 {
1134 	int token_rate;
1135 	int mtu;
1136 	int bucket_size;
1137 
1138 	token_rate = ((rate / 1000) * 64) / (mp->t_clk / 1000);
1139 	if (token_rate > 1023)
1140 		token_rate = 1023;
1141 
1142 	mtu = (mp->dev->mtu + 255) >> 8;
1143 	if (mtu > 63)
1144 		mtu = 63;
1145 
1146 	bucket_size = (burst + 255) >> 8;
1147 	if (bucket_size > 65535)
1148 		bucket_size = 65535;
1149 
1150 	switch (mp->shared->tx_bw_control) {
1151 	case TX_BW_CONTROL_OLD_LAYOUT:
1152 		wrlp(mp, TX_BW_RATE, token_rate);
1153 		wrlp(mp, TX_BW_MTU, mtu);
1154 		wrlp(mp, TX_BW_BURST, bucket_size);
1155 		break;
1156 	case TX_BW_CONTROL_NEW_LAYOUT:
1157 		wrlp(mp, TX_BW_RATE_MOVED, token_rate);
1158 		wrlp(mp, TX_BW_MTU_MOVED, mtu);
1159 		wrlp(mp, TX_BW_BURST_MOVED, bucket_size);
1160 		break;
1161 	}
1162 }
1163 
1164 static void txq_set_rate(struct tx_queue *txq, int rate, int burst)
1165 {
1166 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
1167 	int token_rate;
1168 	int bucket_size;
1169 
1170 	token_rate = ((rate / 1000) * 64) / (mp->t_clk / 1000);
1171 	if (token_rate > 1023)
1172 		token_rate = 1023;
1173 
1174 	bucket_size = (burst + 255) >> 8;
1175 	if (bucket_size > 65535)
1176 		bucket_size = 65535;
1177 
1178 	wrlp(mp, TXQ_BW_TOKENS(txq->index), token_rate << 14);
1179 	wrlp(mp, TXQ_BW_CONF(txq->index), (bucket_size << 10) | token_rate);
1180 }
1181 
1182 static void txq_set_fixed_prio_mode(struct tx_queue *txq)
1183 {
1184 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
1185 	int off;
1186 	u32 val;
1187 
1188 	/*
1189 	 * Turn on fixed priority mode.
1190 	 */
1191 	off = 0;
1192 	switch (mp->shared->tx_bw_control) {
1193 	case TX_BW_CONTROL_OLD_LAYOUT:
1194 		off = TXQ_FIX_PRIO_CONF;
1195 		break;
1196 	case TX_BW_CONTROL_NEW_LAYOUT:
1197 		off = TXQ_FIX_PRIO_CONF_MOVED;
1198 		break;
1199 	}
1200 
1201 	if (off) {
1202 		val = rdlp(mp, off);
1203 		val |= 1 << txq->index;
1204 		wrlp(mp, off, val);
1205 	}
1206 }
1207 
1208 
1209 /* mii management interface *************************************************/
1210 static void mv643xx_eth_adjust_link(struct net_device *dev)
1211 {
1212 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1213 	u32 pscr = rdlp(mp, PORT_SERIAL_CONTROL);
1214 	u32 autoneg_disable = FORCE_LINK_PASS |
1215 	             DISABLE_AUTO_NEG_SPEED_GMII |
1216 		     DISABLE_AUTO_NEG_FOR_FLOW_CTRL |
1217 		     DISABLE_AUTO_NEG_FOR_DUPLEX;
1218 
1219 	if (dev->phydev->autoneg == AUTONEG_ENABLE) {
1220 		/* enable auto negotiation */
1221 		pscr &= ~autoneg_disable;
1222 		goto out_write;
1223 	}
1224 
1225 	pscr |= autoneg_disable;
1226 
1227 	if (dev->phydev->speed == SPEED_1000) {
1228 		/* force gigabit, half duplex not supported */
1229 		pscr |= SET_GMII_SPEED_TO_1000;
1230 		pscr |= SET_FULL_DUPLEX_MODE;
1231 		goto out_write;
1232 	}
1233 
1234 	pscr &= ~SET_GMII_SPEED_TO_1000;
1235 
1236 	if (dev->phydev->speed == SPEED_100)
1237 		pscr |= SET_MII_SPEED_TO_100;
1238 	else
1239 		pscr &= ~SET_MII_SPEED_TO_100;
1240 
1241 	if (dev->phydev->duplex == DUPLEX_FULL)
1242 		pscr |= SET_FULL_DUPLEX_MODE;
1243 	else
1244 		pscr &= ~SET_FULL_DUPLEX_MODE;
1245 
1246 out_write:
1247 	wrlp(mp, PORT_SERIAL_CONTROL, pscr);
1248 }
1249 
1250 /* statistics ***************************************************************/
1251 static struct net_device_stats *mv643xx_eth_get_stats(struct net_device *dev)
1252 {
1253 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1254 	struct net_device_stats *stats = &dev->stats;
1255 	unsigned long tx_packets = 0;
1256 	unsigned long tx_bytes = 0;
1257 	unsigned long tx_dropped = 0;
1258 	int i;
1259 
1260 	for (i = 0; i < mp->txq_count; i++) {
1261 		struct tx_queue *txq = mp->txq + i;
1262 
1263 		tx_packets += txq->tx_packets;
1264 		tx_bytes += txq->tx_bytes;
1265 		tx_dropped += txq->tx_dropped;
1266 	}
1267 
1268 	stats->tx_packets = tx_packets;
1269 	stats->tx_bytes = tx_bytes;
1270 	stats->tx_dropped = tx_dropped;
1271 
1272 	return stats;
1273 }
1274 
1275 static inline u32 mib_read(struct mv643xx_eth_private *mp, int offset)
1276 {
1277 	return rdl(mp, MIB_COUNTERS(mp->port_num) + offset);
1278 }
1279 
1280 static void mib_counters_clear(struct mv643xx_eth_private *mp)
1281 {
1282 	int i;
1283 
1284 	for (i = 0; i < 0x80; i += 4)
1285 		mib_read(mp, i);
1286 
1287 	/* Clear non MIB hw counters also */
1288 	rdlp(mp, RX_DISCARD_FRAME_CNT);
1289 	rdlp(mp, RX_OVERRUN_FRAME_CNT);
1290 }
1291 
1292 static void mib_counters_update(struct mv643xx_eth_private *mp)
1293 {
1294 	struct mib_counters *p = &mp->mib_counters;
1295 
1296 	spin_lock_bh(&mp->mib_counters_lock);
1297 	p->good_octets_received += mib_read(mp, 0x00);
1298 	p->bad_octets_received += mib_read(mp, 0x08);
1299 	p->internal_mac_transmit_err += mib_read(mp, 0x0c);
1300 	p->good_frames_received += mib_read(mp, 0x10);
1301 	p->bad_frames_received += mib_read(mp, 0x14);
1302 	p->broadcast_frames_received += mib_read(mp, 0x18);
1303 	p->multicast_frames_received += mib_read(mp, 0x1c);
1304 	p->frames_64_octets += mib_read(mp, 0x20);
1305 	p->frames_65_to_127_octets += mib_read(mp, 0x24);
1306 	p->frames_128_to_255_octets += mib_read(mp, 0x28);
1307 	p->frames_256_to_511_octets += mib_read(mp, 0x2c);
1308 	p->frames_512_to_1023_octets += mib_read(mp, 0x30);
1309 	p->frames_1024_to_max_octets += mib_read(mp, 0x34);
1310 	p->good_octets_sent += mib_read(mp, 0x38);
1311 	p->good_frames_sent += mib_read(mp, 0x40);
1312 	p->excessive_collision += mib_read(mp, 0x44);
1313 	p->multicast_frames_sent += mib_read(mp, 0x48);
1314 	p->broadcast_frames_sent += mib_read(mp, 0x4c);
1315 	p->unrec_mac_control_received += mib_read(mp, 0x50);
1316 	p->fc_sent += mib_read(mp, 0x54);
1317 	p->good_fc_received += mib_read(mp, 0x58);
1318 	p->bad_fc_received += mib_read(mp, 0x5c);
1319 	p->undersize_received += mib_read(mp, 0x60);
1320 	p->fragments_received += mib_read(mp, 0x64);
1321 	p->oversize_received += mib_read(mp, 0x68);
1322 	p->jabber_received += mib_read(mp, 0x6c);
1323 	p->mac_receive_error += mib_read(mp, 0x70);
1324 	p->bad_crc_event += mib_read(mp, 0x74);
1325 	p->collision += mib_read(mp, 0x78);
1326 	p->late_collision += mib_read(mp, 0x7c);
1327 	/* Non MIB hardware counters */
1328 	p->rx_discard += rdlp(mp, RX_DISCARD_FRAME_CNT);
1329 	p->rx_overrun += rdlp(mp, RX_OVERRUN_FRAME_CNT);
1330 	spin_unlock_bh(&mp->mib_counters_lock);
1331 }
1332 
1333 static void mib_counters_timer_wrapper(struct timer_list *t)
1334 {
1335 	struct mv643xx_eth_private *mp = from_timer(mp, t, mib_counters_timer);
1336 	mib_counters_update(mp);
1337 	mod_timer(&mp->mib_counters_timer, jiffies + 30 * HZ);
1338 }
1339 
1340 
1341 /* interrupt coalescing *****************************************************/
1342 /*
1343  * Hardware coalescing parameters are set in units of 64 t_clk
1344  * cycles.  I.e.:
1345  *
1346  *	coal_delay_in_usec = 64000000 * register_value / t_clk_rate
1347  *
1348  *	register_value = coal_delay_in_usec * t_clk_rate / 64000000
1349  *
1350  * In the ->set*() methods, we round the computed register value
1351  * to the nearest integer.
1352  */
1353 static unsigned int get_rx_coal(struct mv643xx_eth_private *mp)
1354 {
1355 	u32 val = rdlp(mp, SDMA_CONFIG);
1356 	u64 temp;
1357 
1358 	if (mp->shared->extended_rx_coal_limit)
1359 		temp = ((val & 0x02000000) >> 10) | ((val & 0x003fff80) >> 7);
1360 	else
1361 		temp = (val & 0x003fff00) >> 8;
1362 
1363 	temp *= 64000000;
1364 	temp += mp->t_clk / 2;
1365 	do_div(temp, mp->t_clk);
1366 
1367 	return (unsigned int)temp;
1368 }
1369 
1370 static void set_rx_coal(struct mv643xx_eth_private *mp, unsigned int usec)
1371 {
1372 	u64 temp;
1373 	u32 val;
1374 
1375 	temp = (u64)usec * mp->t_clk;
1376 	temp += 31999999;
1377 	do_div(temp, 64000000);
1378 
1379 	val = rdlp(mp, SDMA_CONFIG);
1380 	if (mp->shared->extended_rx_coal_limit) {
1381 		if (temp > 0xffff)
1382 			temp = 0xffff;
1383 		val &= ~0x023fff80;
1384 		val |= (temp & 0x8000) << 10;
1385 		val |= (temp & 0x7fff) << 7;
1386 	} else {
1387 		if (temp > 0x3fff)
1388 			temp = 0x3fff;
1389 		val &= ~0x003fff00;
1390 		val |= (temp & 0x3fff) << 8;
1391 	}
1392 	wrlp(mp, SDMA_CONFIG, val);
1393 }
1394 
1395 static unsigned int get_tx_coal(struct mv643xx_eth_private *mp)
1396 {
1397 	u64 temp;
1398 
1399 	temp = (rdlp(mp, TX_FIFO_URGENT_THRESHOLD) & 0x3fff0) >> 4;
1400 	temp *= 64000000;
1401 	temp += mp->t_clk / 2;
1402 	do_div(temp, mp->t_clk);
1403 
1404 	return (unsigned int)temp;
1405 }
1406 
1407 static void set_tx_coal(struct mv643xx_eth_private *mp, unsigned int usec)
1408 {
1409 	u64 temp;
1410 
1411 	temp = (u64)usec * mp->t_clk;
1412 	temp += 31999999;
1413 	do_div(temp, 64000000);
1414 
1415 	if (temp > 0x3fff)
1416 		temp = 0x3fff;
1417 
1418 	wrlp(mp, TX_FIFO_URGENT_THRESHOLD, temp << 4);
1419 }
1420 
1421 
1422 /* ethtool ******************************************************************/
1423 struct mv643xx_eth_stats {
1424 	char stat_string[ETH_GSTRING_LEN];
1425 	int sizeof_stat;
1426 	int netdev_off;
1427 	int mp_off;
1428 };
1429 
1430 #define SSTAT(m)						\
1431 	{ #m, sizeof_field(struct net_device_stats, m),		\
1432 	  offsetof(struct net_device, stats.m), -1 }
1433 
1434 #define MIBSTAT(m)						\
1435 	{ #m, sizeof_field(struct mib_counters, m),		\
1436 	  -1, offsetof(struct mv643xx_eth_private, mib_counters.m) }
1437 
1438 static const struct mv643xx_eth_stats mv643xx_eth_stats[] = {
1439 	SSTAT(rx_packets),
1440 	SSTAT(tx_packets),
1441 	SSTAT(rx_bytes),
1442 	SSTAT(tx_bytes),
1443 	SSTAT(rx_errors),
1444 	SSTAT(tx_errors),
1445 	SSTAT(rx_dropped),
1446 	SSTAT(tx_dropped),
1447 	MIBSTAT(good_octets_received),
1448 	MIBSTAT(bad_octets_received),
1449 	MIBSTAT(internal_mac_transmit_err),
1450 	MIBSTAT(good_frames_received),
1451 	MIBSTAT(bad_frames_received),
1452 	MIBSTAT(broadcast_frames_received),
1453 	MIBSTAT(multicast_frames_received),
1454 	MIBSTAT(frames_64_octets),
1455 	MIBSTAT(frames_65_to_127_octets),
1456 	MIBSTAT(frames_128_to_255_octets),
1457 	MIBSTAT(frames_256_to_511_octets),
1458 	MIBSTAT(frames_512_to_1023_octets),
1459 	MIBSTAT(frames_1024_to_max_octets),
1460 	MIBSTAT(good_octets_sent),
1461 	MIBSTAT(good_frames_sent),
1462 	MIBSTAT(excessive_collision),
1463 	MIBSTAT(multicast_frames_sent),
1464 	MIBSTAT(broadcast_frames_sent),
1465 	MIBSTAT(unrec_mac_control_received),
1466 	MIBSTAT(fc_sent),
1467 	MIBSTAT(good_fc_received),
1468 	MIBSTAT(bad_fc_received),
1469 	MIBSTAT(undersize_received),
1470 	MIBSTAT(fragments_received),
1471 	MIBSTAT(oversize_received),
1472 	MIBSTAT(jabber_received),
1473 	MIBSTAT(mac_receive_error),
1474 	MIBSTAT(bad_crc_event),
1475 	MIBSTAT(collision),
1476 	MIBSTAT(late_collision),
1477 	MIBSTAT(rx_discard),
1478 	MIBSTAT(rx_overrun),
1479 };
1480 
1481 static int
1482 mv643xx_eth_get_link_ksettings_phy(struct mv643xx_eth_private *mp,
1483 				   struct ethtool_link_ksettings *cmd)
1484 {
1485 	struct net_device *dev = mp->dev;
1486 
1487 	phy_ethtool_ksettings_get(dev->phydev, cmd);
1488 
1489 	/*
1490 	 * The MAC does not support 1000baseT_Half.
1491 	 */
1492 	linkmode_clear_bit(ETHTOOL_LINK_MODE_1000baseT_Half_BIT,
1493 			   cmd->link_modes.supported);
1494 	linkmode_clear_bit(ETHTOOL_LINK_MODE_1000baseT_Half_BIT,
1495 			   cmd->link_modes.advertising);
1496 
1497 	return 0;
1498 }
1499 
1500 static int
1501 mv643xx_eth_get_link_ksettings_phyless(struct mv643xx_eth_private *mp,
1502 				       struct ethtool_link_ksettings *cmd)
1503 {
1504 	u32 port_status;
1505 	u32 supported, advertising;
1506 
1507 	port_status = rdlp(mp, PORT_STATUS);
1508 
1509 	supported = SUPPORTED_MII;
1510 	advertising = ADVERTISED_MII;
1511 	switch (port_status & PORT_SPEED_MASK) {
1512 	case PORT_SPEED_10:
1513 		cmd->base.speed = SPEED_10;
1514 		break;
1515 	case PORT_SPEED_100:
1516 		cmd->base.speed = SPEED_100;
1517 		break;
1518 	case PORT_SPEED_1000:
1519 		cmd->base.speed = SPEED_1000;
1520 		break;
1521 	default:
1522 		cmd->base.speed = -1;
1523 		break;
1524 	}
1525 	cmd->base.duplex = (port_status & FULL_DUPLEX) ?
1526 		DUPLEX_FULL : DUPLEX_HALF;
1527 	cmd->base.port = PORT_MII;
1528 	cmd->base.phy_address = 0;
1529 	cmd->base.autoneg = AUTONEG_DISABLE;
1530 
1531 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
1532 						supported);
1533 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
1534 						advertising);
1535 
1536 	return 0;
1537 }
1538 
1539 static void
1540 mv643xx_eth_get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1541 {
1542 	wol->supported = 0;
1543 	wol->wolopts = 0;
1544 	if (dev->phydev)
1545 		phy_ethtool_get_wol(dev->phydev, wol);
1546 }
1547 
1548 static int
1549 mv643xx_eth_set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1550 {
1551 	int err;
1552 
1553 	if (!dev->phydev)
1554 		return -EOPNOTSUPP;
1555 
1556 	err = phy_ethtool_set_wol(dev->phydev, wol);
1557 	/* Given that mv643xx_eth works without the marvell-specific PHY driver,
1558 	 * this debugging hint is useful to have.
1559 	 */
1560 	if (err == -EOPNOTSUPP)
1561 		netdev_info(dev, "The PHY does not support set_wol, was CONFIG_MARVELL_PHY enabled?\n");
1562 	return err;
1563 }
1564 
1565 static int
1566 mv643xx_eth_get_link_ksettings(struct net_device *dev,
1567 			       struct ethtool_link_ksettings *cmd)
1568 {
1569 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1570 
1571 	if (dev->phydev)
1572 		return mv643xx_eth_get_link_ksettings_phy(mp, cmd);
1573 	else
1574 		return mv643xx_eth_get_link_ksettings_phyless(mp, cmd);
1575 }
1576 
1577 static int
1578 mv643xx_eth_set_link_ksettings(struct net_device *dev,
1579 			       const struct ethtool_link_ksettings *cmd)
1580 {
1581 	struct ethtool_link_ksettings c = *cmd;
1582 	u32 advertising;
1583 	int ret;
1584 
1585 	if (!dev->phydev)
1586 		return -EINVAL;
1587 
1588 	/*
1589 	 * The MAC does not support 1000baseT_Half.
1590 	 */
1591 	ethtool_convert_link_mode_to_legacy_u32(&advertising,
1592 						c.link_modes.advertising);
1593 	advertising &= ~ADVERTISED_1000baseT_Half;
1594 	ethtool_convert_legacy_u32_to_link_mode(c.link_modes.advertising,
1595 						advertising);
1596 
1597 	ret = phy_ethtool_ksettings_set(dev->phydev, &c);
1598 	if (!ret)
1599 		mv643xx_eth_adjust_link(dev);
1600 	return ret;
1601 }
1602 
1603 static void mv643xx_eth_get_drvinfo(struct net_device *dev,
1604 				    struct ethtool_drvinfo *drvinfo)
1605 {
1606 	strlcpy(drvinfo->driver, mv643xx_eth_driver_name,
1607 		sizeof(drvinfo->driver));
1608 	strlcpy(drvinfo->version, mv643xx_eth_driver_version,
1609 		sizeof(drvinfo->version));
1610 	strlcpy(drvinfo->fw_version, "N/A", sizeof(drvinfo->fw_version));
1611 	strlcpy(drvinfo->bus_info, "platform", sizeof(drvinfo->bus_info));
1612 }
1613 
1614 static int mv643xx_eth_get_coalesce(struct net_device *dev,
1615 				    struct ethtool_coalesce *ec,
1616 				    struct kernel_ethtool_coalesce *kernel_coal,
1617 				    struct netlink_ext_ack *extack)
1618 {
1619 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1620 
1621 	ec->rx_coalesce_usecs = get_rx_coal(mp);
1622 	ec->tx_coalesce_usecs = get_tx_coal(mp);
1623 
1624 	return 0;
1625 }
1626 
1627 static int mv643xx_eth_set_coalesce(struct net_device *dev,
1628 				    struct ethtool_coalesce *ec,
1629 				    struct kernel_ethtool_coalesce *kernel_coal,
1630 				    struct netlink_ext_ack *extack)
1631 {
1632 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1633 
1634 	set_rx_coal(mp, ec->rx_coalesce_usecs);
1635 	set_tx_coal(mp, ec->tx_coalesce_usecs);
1636 
1637 	return 0;
1638 }
1639 
1640 static void
1641 mv643xx_eth_get_ringparam(struct net_device *dev, struct ethtool_ringparam *er)
1642 {
1643 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1644 
1645 	er->rx_max_pending = 4096;
1646 	er->tx_max_pending = 4096;
1647 
1648 	er->rx_pending = mp->rx_ring_size;
1649 	er->tx_pending = mp->tx_ring_size;
1650 }
1651 
1652 static int
1653 mv643xx_eth_set_ringparam(struct net_device *dev, struct ethtool_ringparam *er)
1654 {
1655 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1656 
1657 	if (er->rx_mini_pending || er->rx_jumbo_pending)
1658 		return -EINVAL;
1659 
1660 	mp->rx_ring_size = er->rx_pending < 4096 ? er->rx_pending : 4096;
1661 	mp->tx_ring_size = clamp_t(unsigned int, er->tx_pending,
1662 				   MV643XX_MAX_SKB_DESCS * 2, 4096);
1663 	if (mp->tx_ring_size != er->tx_pending)
1664 		netdev_warn(dev, "TX queue size set to %u (requested %u)\n",
1665 			    mp->tx_ring_size, er->tx_pending);
1666 
1667 	if (netif_running(dev)) {
1668 		mv643xx_eth_stop(dev);
1669 		if (mv643xx_eth_open(dev)) {
1670 			netdev_err(dev,
1671 				   "fatal error on re-opening device after ring param change\n");
1672 			return -ENOMEM;
1673 		}
1674 	}
1675 
1676 	return 0;
1677 }
1678 
1679 
1680 static int
1681 mv643xx_eth_set_features(struct net_device *dev, netdev_features_t features)
1682 {
1683 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1684 	bool rx_csum = features & NETIF_F_RXCSUM;
1685 
1686 	wrlp(mp, PORT_CONFIG, rx_csum ? 0x02000000 : 0x00000000);
1687 
1688 	return 0;
1689 }
1690 
1691 static void mv643xx_eth_get_strings(struct net_device *dev,
1692 				    uint32_t stringset, uint8_t *data)
1693 {
1694 	int i;
1695 
1696 	if (stringset == ETH_SS_STATS) {
1697 		for (i = 0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
1698 			memcpy(data + i * ETH_GSTRING_LEN,
1699 				mv643xx_eth_stats[i].stat_string,
1700 				ETH_GSTRING_LEN);
1701 		}
1702 	}
1703 }
1704 
1705 static void mv643xx_eth_get_ethtool_stats(struct net_device *dev,
1706 					  struct ethtool_stats *stats,
1707 					  uint64_t *data)
1708 {
1709 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1710 	int i;
1711 
1712 	mv643xx_eth_get_stats(dev);
1713 	mib_counters_update(mp);
1714 
1715 	for (i = 0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
1716 		const struct mv643xx_eth_stats *stat;
1717 		void *p;
1718 
1719 		stat = mv643xx_eth_stats + i;
1720 
1721 		if (stat->netdev_off >= 0)
1722 			p = ((void *)mp->dev) + stat->netdev_off;
1723 		else
1724 			p = ((void *)mp) + stat->mp_off;
1725 
1726 		data[i] = (stat->sizeof_stat == 8) ?
1727 				*(uint64_t *)p : *(uint32_t *)p;
1728 	}
1729 }
1730 
1731 static int mv643xx_eth_get_sset_count(struct net_device *dev, int sset)
1732 {
1733 	if (sset == ETH_SS_STATS)
1734 		return ARRAY_SIZE(mv643xx_eth_stats);
1735 
1736 	return -EOPNOTSUPP;
1737 }
1738 
1739 static const struct ethtool_ops mv643xx_eth_ethtool_ops = {
1740 	.supported_coalesce_params = ETHTOOL_COALESCE_USECS,
1741 	.get_drvinfo		= mv643xx_eth_get_drvinfo,
1742 	.nway_reset		= phy_ethtool_nway_reset,
1743 	.get_link		= ethtool_op_get_link,
1744 	.get_coalesce		= mv643xx_eth_get_coalesce,
1745 	.set_coalesce		= mv643xx_eth_set_coalesce,
1746 	.get_ringparam		= mv643xx_eth_get_ringparam,
1747 	.set_ringparam		= mv643xx_eth_set_ringparam,
1748 	.get_strings		= mv643xx_eth_get_strings,
1749 	.get_ethtool_stats	= mv643xx_eth_get_ethtool_stats,
1750 	.get_sset_count		= mv643xx_eth_get_sset_count,
1751 	.get_ts_info		= ethtool_op_get_ts_info,
1752 	.get_wol                = mv643xx_eth_get_wol,
1753 	.set_wol                = mv643xx_eth_set_wol,
1754 	.get_link_ksettings	= mv643xx_eth_get_link_ksettings,
1755 	.set_link_ksettings	= mv643xx_eth_set_link_ksettings,
1756 };
1757 
1758 
1759 /* address handling *********************************************************/
1760 static void uc_addr_get(struct mv643xx_eth_private *mp, unsigned char *addr)
1761 {
1762 	unsigned int mac_h = rdlp(mp, MAC_ADDR_HIGH);
1763 	unsigned int mac_l = rdlp(mp, MAC_ADDR_LOW);
1764 
1765 	addr[0] = (mac_h >> 24) & 0xff;
1766 	addr[1] = (mac_h >> 16) & 0xff;
1767 	addr[2] = (mac_h >> 8) & 0xff;
1768 	addr[3] = mac_h & 0xff;
1769 	addr[4] = (mac_l >> 8) & 0xff;
1770 	addr[5] = mac_l & 0xff;
1771 }
1772 
1773 static void uc_addr_set(struct mv643xx_eth_private *mp, unsigned char *addr)
1774 {
1775 	wrlp(mp, MAC_ADDR_HIGH,
1776 		(addr[0] << 24) | (addr[1] << 16) | (addr[2] << 8) | addr[3]);
1777 	wrlp(mp, MAC_ADDR_LOW, (addr[4] << 8) | addr[5]);
1778 }
1779 
1780 static u32 uc_addr_filter_mask(struct net_device *dev)
1781 {
1782 	struct netdev_hw_addr *ha;
1783 	u32 nibbles;
1784 
1785 	if (dev->flags & IFF_PROMISC)
1786 		return 0;
1787 
1788 	nibbles = 1 << (dev->dev_addr[5] & 0x0f);
1789 	netdev_for_each_uc_addr(ha, dev) {
1790 		if (memcmp(dev->dev_addr, ha->addr, 5))
1791 			return 0;
1792 		if ((dev->dev_addr[5] ^ ha->addr[5]) & 0xf0)
1793 			return 0;
1794 
1795 		nibbles |= 1 << (ha->addr[5] & 0x0f);
1796 	}
1797 
1798 	return nibbles;
1799 }
1800 
1801 static void mv643xx_eth_program_unicast_filter(struct net_device *dev)
1802 {
1803 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1804 	u32 port_config;
1805 	u32 nibbles;
1806 	int i;
1807 
1808 	uc_addr_set(mp, dev->dev_addr);
1809 
1810 	port_config = rdlp(mp, PORT_CONFIG) & ~UNICAST_PROMISCUOUS_MODE;
1811 
1812 	nibbles = uc_addr_filter_mask(dev);
1813 	if (!nibbles) {
1814 		port_config |= UNICAST_PROMISCUOUS_MODE;
1815 		nibbles = 0xffff;
1816 	}
1817 
1818 	for (i = 0; i < 16; i += 4) {
1819 		int off = UNICAST_TABLE(mp->port_num) + i;
1820 		u32 v;
1821 
1822 		v = 0;
1823 		if (nibbles & 1)
1824 			v |= 0x00000001;
1825 		if (nibbles & 2)
1826 			v |= 0x00000100;
1827 		if (nibbles & 4)
1828 			v |= 0x00010000;
1829 		if (nibbles & 8)
1830 			v |= 0x01000000;
1831 		nibbles >>= 4;
1832 
1833 		wrl(mp, off, v);
1834 	}
1835 
1836 	wrlp(mp, PORT_CONFIG, port_config);
1837 }
1838 
1839 static int addr_crc(unsigned char *addr)
1840 {
1841 	int crc = 0;
1842 	int i;
1843 
1844 	for (i = 0; i < 6; i++) {
1845 		int j;
1846 
1847 		crc = (crc ^ addr[i]) << 8;
1848 		for (j = 7; j >= 0; j--) {
1849 			if (crc & (0x100 << j))
1850 				crc ^= 0x107 << j;
1851 		}
1852 	}
1853 
1854 	return crc;
1855 }
1856 
1857 static void mv643xx_eth_program_multicast_filter(struct net_device *dev)
1858 {
1859 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1860 	u32 *mc_spec;
1861 	u32 *mc_other;
1862 	struct netdev_hw_addr *ha;
1863 	int i;
1864 
1865 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI))
1866 		goto promiscuous;
1867 
1868 	/* Allocate both mc_spec and mc_other tables */
1869 	mc_spec = kcalloc(128, sizeof(u32), GFP_ATOMIC);
1870 	if (!mc_spec)
1871 		goto promiscuous;
1872 	mc_other = &mc_spec[64];
1873 
1874 	netdev_for_each_mc_addr(ha, dev) {
1875 		u8 *a = ha->addr;
1876 		u32 *table;
1877 		u8 entry;
1878 
1879 		if (memcmp(a, "\x01\x00\x5e\x00\x00", 5) == 0) {
1880 			table = mc_spec;
1881 			entry = a[5];
1882 		} else {
1883 			table = mc_other;
1884 			entry = addr_crc(a);
1885 		}
1886 
1887 		table[entry >> 2] |= 1 << (8 * (entry & 3));
1888 	}
1889 
1890 	for (i = 0; i < 64; i++) {
1891 		wrl(mp, SPECIAL_MCAST_TABLE(mp->port_num) + i * sizeof(u32),
1892 		    mc_spec[i]);
1893 		wrl(mp, OTHER_MCAST_TABLE(mp->port_num) + i * sizeof(u32),
1894 		    mc_other[i]);
1895 	}
1896 
1897 	kfree(mc_spec);
1898 	return;
1899 
1900 promiscuous:
1901 	for (i = 0; i < 64; i++) {
1902 		wrl(mp, SPECIAL_MCAST_TABLE(mp->port_num) + i * sizeof(u32),
1903 		    0x01010101u);
1904 		wrl(mp, OTHER_MCAST_TABLE(mp->port_num) + i * sizeof(u32),
1905 		    0x01010101u);
1906 	}
1907 }
1908 
1909 static void mv643xx_eth_set_rx_mode(struct net_device *dev)
1910 {
1911 	mv643xx_eth_program_unicast_filter(dev);
1912 	mv643xx_eth_program_multicast_filter(dev);
1913 }
1914 
1915 static int mv643xx_eth_set_mac_address(struct net_device *dev, void *addr)
1916 {
1917 	struct sockaddr *sa = addr;
1918 
1919 	if (!is_valid_ether_addr(sa->sa_data))
1920 		return -EADDRNOTAVAIL;
1921 
1922 	eth_hw_addr_set(dev, sa->sa_data);
1923 
1924 	netif_addr_lock_bh(dev);
1925 	mv643xx_eth_program_unicast_filter(dev);
1926 	netif_addr_unlock_bh(dev);
1927 
1928 	return 0;
1929 }
1930 
1931 
1932 /* rx/tx queue initialisation ***********************************************/
1933 static int rxq_init(struct mv643xx_eth_private *mp, int index)
1934 {
1935 	struct rx_queue *rxq = mp->rxq + index;
1936 	struct rx_desc *rx_desc;
1937 	int size;
1938 	int i;
1939 
1940 	rxq->index = index;
1941 
1942 	rxq->rx_ring_size = mp->rx_ring_size;
1943 
1944 	rxq->rx_desc_count = 0;
1945 	rxq->rx_curr_desc = 0;
1946 	rxq->rx_used_desc = 0;
1947 
1948 	size = rxq->rx_ring_size * sizeof(struct rx_desc);
1949 
1950 	if (index == 0 && size <= mp->rx_desc_sram_size) {
1951 		rxq->rx_desc_area = ioremap(mp->rx_desc_sram_addr,
1952 						mp->rx_desc_sram_size);
1953 		rxq->rx_desc_dma = mp->rx_desc_sram_addr;
1954 	} else {
1955 		rxq->rx_desc_area = dma_alloc_coherent(mp->dev->dev.parent,
1956 						       size, &rxq->rx_desc_dma,
1957 						       GFP_KERNEL);
1958 	}
1959 
1960 	if (rxq->rx_desc_area == NULL) {
1961 		netdev_err(mp->dev,
1962 			   "can't allocate rx ring (%d bytes)\n", size);
1963 		goto out;
1964 	}
1965 	memset(rxq->rx_desc_area, 0, size);
1966 
1967 	rxq->rx_desc_area_size = size;
1968 	rxq->rx_skb = kcalloc(rxq->rx_ring_size, sizeof(*rxq->rx_skb),
1969 				    GFP_KERNEL);
1970 	if (rxq->rx_skb == NULL)
1971 		goto out_free;
1972 
1973 	rx_desc = rxq->rx_desc_area;
1974 	for (i = 0; i < rxq->rx_ring_size; i++) {
1975 		int nexti;
1976 
1977 		nexti = i + 1;
1978 		if (nexti == rxq->rx_ring_size)
1979 			nexti = 0;
1980 
1981 		rx_desc[i].next_desc_ptr = rxq->rx_desc_dma +
1982 					nexti * sizeof(struct rx_desc);
1983 	}
1984 
1985 	return 0;
1986 
1987 
1988 out_free:
1989 	if (index == 0 && size <= mp->rx_desc_sram_size)
1990 		iounmap(rxq->rx_desc_area);
1991 	else
1992 		dma_free_coherent(mp->dev->dev.parent, size,
1993 				  rxq->rx_desc_area,
1994 				  rxq->rx_desc_dma);
1995 
1996 out:
1997 	return -ENOMEM;
1998 }
1999 
2000 static void rxq_deinit(struct rx_queue *rxq)
2001 {
2002 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
2003 	int i;
2004 
2005 	rxq_disable(rxq);
2006 
2007 	for (i = 0; i < rxq->rx_ring_size; i++) {
2008 		if (rxq->rx_skb[i]) {
2009 			dev_consume_skb_any(rxq->rx_skb[i]);
2010 			rxq->rx_desc_count--;
2011 		}
2012 	}
2013 
2014 	if (rxq->rx_desc_count) {
2015 		netdev_err(mp->dev, "error freeing rx ring -- %d skbs stuck\n",
2016 			   rxq->rx_desc_count);
2017 	}
2018 
2019 	if (rxq->index == 0 &&
2020 	    rxq->rx_desc_area_size <= mp->rx_desc_sram_size)
2021 		iounmap(rxq->rx_desc_area);
2022 	else
2023 		dma_free_coherent(mp->dev->dev.parent, rxq->rx_desc_area_size,
2024 				  rxq->rx_desc_area, rxq->rx_desc_dma);
2025 
2026 	kfree(rxq->rx_skb);
2027 }
2028 
2029 static int txq_init(struct mv643xx_eth_private *mp, int index)
2030 {
2031 	struct tx_queue *txq = mp->txq + index;
2032 	struct tx_desc *tx_desc;
2033 	int size;
2034 	int ret;
2035 	int i;
2036 
2037 	txq->index = index;
2038 
2039 	txq->tx_ring_size = mp->tx_ring_size;
2040 
2041 	/* A queue must always have room for at least one skb.
2042 	 * Therefore, stop the queue when the free entries reaches
2043 	 * the maximum number of descriptors per skb.
2044 	 */
2045 	txq->tx_stop_threshold = txq->tx_ring_size - MV643XX_MAX_SKB_DESCS;
2046 	txq->tx_wake_threshold = txq->tx_stop_threshold / 2;
2047 
2048 	txq->tx_desc_count = 0;
2049 	txq->tx_curr_desc = 0;
2050 	txq->tx_used_desc = 0;
2051 
2052 	size = txq->tx_ring_size * sizeof(struct tx_desc);
2053 
2054 	if (index == 0 && size <= mp->tx_desc_sram_size) {
2055 		txq->tx_desc_area = ioremap(mp->tx_desc_sram_addr,
2056 						mp->tx_desc_sram_size);
2057 		txq->tx_desc_dma = mp->tx_desc_sram_addr;
2058 	} else {
2059 		txq->tx_desc_area = dma_alloc_coherent(mp->dev->dev.parent,
2060 						       size, &txq->tx_desc_dma,
2061 						       GFP_KERNEL);
2062 	}
2063 
2064 	if (txq->tx_desc_area == NULL) {
2065 		netdev_err(mp->dev,
2066 			   "can't allocate tx ring (%d bytes)\n", size);
2067 		return -ENOMEM;
2068 	}
2069 	memset(txq->tx_desc_area, 0, size);
2070 
2071 	txq->tx_desc_area_size = size;
2072 
2073 	tx_desc = txq->tx_desc_area;
2074 	for (i = 0; i < txq->tx_ring_size; i++) {
2075 		struct tx_desc *txd = tx_desc + i;
2076 		int nexti;
2077 
2078 		nexti = i + 1;
2079 		if (nexti == txq->tx_ring_size)
2080 			nexti = 0;
2081 
2082 		txd->cmd_sts = 0;
2083 		txd->next_desc_ptr = txq->tx_desc_dma +
2084 					nexti * sizeof(struct tx_desc);
2085 	}
2086 
2087 	txq->tx_desc_mapping = kcalloc(txq->tx_ring_size, sizeof(char),
2088 				       GFP_KERNEL);
2089 	if (!txq->tx_desc_mapping) {
2090 		ret = -ENOMEM;
2091 		goto err_free_desc_area;
2092 	}
2093 
2094 	/* Allocate DMA buffers for TSO MAC/IP/TCP headers */
2095 	txq->tso_hdrs = dma_alloc_coherent(mp->dev->dev.parent,
2096 					   txq->tx_ring_size * TSO_HEADER_SIZE,
2097 					   &txq->tso_hdrs_dma, GFP_KERNEL);
2098 	if (txq->tso_hdrs == NULL) {
2099 		ret = -ENOMEM;
2100 		goto err_free_desc_mapping;
2101 	}
2102 	skb_queue_head_init(&txq->tx_skb);
2103 
2104 	return 0;
2105 
2106 err_free_desc_mapping:
2107 	kfree(txq->tx_desc_mapping);
2108 err_free_desc_area:
2109 	if (index == 0 && size <= mp->tx_desc_sram_size)
2110 		iounmap(txq->tx_desc_area);
2111 	else
2112 		dma_free_coherent(mp->dev->dev.parent, txq->tx_desc_area_size,
2113 				  txq->tx_desc_area, txq->tx_desc_dma);
2114 	return ret;
2115 }
2116 
2117 static void txq_deinit(struct tx_queue *txq)
2118 {
2119 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
2120 
2121 	txq_disable(txq);
2122 	txq_reclaim(txq, txq->tx_ring_size, 1);
2123 
2124 	BUG_ON(txq->tx_used_desc != txq->tx_curr_desc);
2125 
2126 	if (txq->index == 0 &&
2127 	    txq->tx_desc_area_size <= mp->tx_desc_sram_size)
2128 		iounmap(txq->tx_desc_area);
2129 	else
2130 		dma_free_coherent(mp->dev->dev.parent, txq->tx_desc_area_size,
2131 				  txq->tx_desc_area, txq->tx_desc_dma);
2132 	kfree(txq->tx_desc_mapping);
2133 
2134 	if (txq->tso_hdrs)
2135 		dma_free_coherent(mp->dev->dev.parent,
2136 				  txq->tx_ring_size * TSO_HEADER_SIZE,
2137 				  txq->tso_hdrs, txq->tso_hdrs_dma);
2138 }
2139 
2140 
2141 /* netdev ops and related ***************************************************/
2142 static int mv643xx_eth_collect_events(struct mv643xx_eth_private *mp)
2143 {
2144 	u32 int_cause;
2145 	u32 int_cause_ext;
2146 
2147 	int_cause = rdlp(mp, INT_CAUSE) & mp->int_mask;
2148 	if (int_cause == 0)
2149 		return 0;
2150 
2151 	int_cause_ext = 0;
2152 	if (int_cause & INT_EXT) {
2153 		int_cause &= ~INT_EXT;
2154 		int_cause_ext = rdlp(mp, INT_CAUSE_EXT);
2155 	}
2156 
2157 	if (int_cause) {
2158 		wrlp(mp, INT_CAUSE, ~int_cause);
2159 		mp->work_tx_end |= ((int_cause & INT_TX_END) >> 19) &
2160 				~(rdlp(mp, TXQ_COMMAND) & 0xff);
2161 		mp->work_rx |= (int_cause & INT_RX) >> 2;
2162 	}
2163 
2164 	int_cause_ext &= INT_EXT_LINK_PHY | INT_EXT_TX;
2165 	if (int_cause_ext) {
2166 		wrlp(mp, INT_CAUSE_EXT, ~int_cause_ext);
2167 		if (int_cause_ext & INT_EXT_LINK_PHY)
2168 			mp->work_link = 1;
2169 		mp->work_tx |= int_cause_ext & INT_EXT_TX;
2170 	}
2171 
2172 	return 1;
2173 }
2174 
2175 static irqreturn_t mv643xx_eth_irq(int irq, void *dev_id)
2176 {
2177 	struct net_device *dev = (struct net_device *)dev_id;
2178 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2179 
2180 	if (unlikely(!mv643xx_eth_collect_events(mp)))
2181 		return IRQ_NONE;
2182 
2183 	wrlp(mp, INT_MASK, 0);
2184 	napi_schedule(&mp->napi);
2185 
2186 	return IRQ_HANDLED;
2187 }
2188 
2189 static void handle_link_event(struct mv643xx_eth_private *mp)
2190 {
2191 	struct net_device *dev = mp->dev;
2192 	u32 port_status;
2193 	int speed;
2194 	int duplex;
2195 	int fc;
2196 
2197 	port_status = rdlp(mp, PORT_STATUS);
2198 	if (!(port_status & LINK_UP)) {
2199 		if (netif_carrier_ok(dev)) {
2200 			int i;
2201 
2202 			netdev_info(dev, "link down\n");
2203 
2204 			netif_carrier_off(dev);
2205 
2206 			for (i = 0; i < mp->txq_count; i++) {
2207 				struct tx_queue *txq = mp->txq + i;
2208 
2209 				txq_reclaim(txq, txq->tx_ring_size, 1);
2210 				txq_reset_hw_ptr(txq);
2211 			}
2212 		}
2213 		return;
2214 	}
2215 
2216 	switch (port_status & PORT_SPEED_MASK) {
2217 	case PORT_SPEED_10:
2218 		speed = 10;
2219 		break;
2220 	case PORT_SPEED_100:
2221 		speed = 100;
2222 		break;
2223 	case PORT_SPEED_1000:
2224 		speed = 1000;
2225 		break;
2226 	default:
2227 		speed = -1;
2228 		break;
2229 	}
2230 	duplex = (port_status & FULL_DUPLEX) ? 1 : 0;
2231 	fc = (port_status & FLOW_CONTROL_ENABLED) ? 1 : 0;
2232 
2233 	netdev_info(dev, "link up, %d Mb/s, %s duplex, flow control %sabled\n",
2234 		    speed, duplex ? "full" : "half", fc ? "en" : "dis");
2235 
2236 	if (!netif_carrier_ok(dev))
2237 		netif_carrier_on(dev);
2238 }
2239 
2240 static int mv643xx_eth_poll(struct napi_struct *napi, int budget)
2241 {
2242 	struct mv643xx_eth_private *mp;
2243 	int work_done;
2244 
2245 	mp = container_of(napi, struct mv643xx_eth_private, napi);
2246 
2247 	if (unlikely(mp->oom)) {
2248 		mp->oom = 0;
2249 		del_timer(&mp->rx_oom);
2250 	}
2251 
2252 	work_done = 0;
2253 	while (work_done < budget) {
2254 		u8 queue_mask;
2255 		int queue;
2256 		int work_tbd;
2257 
2258 		if (mp->work_link) {
2259 			mp->work_link = 0;
2260 			handle_link_event(mp);
2261 			work_done++;
2262 			continue;
2263 		}
2264 
2265 		queue_mask = mp->work_tx | mp->work_tx_end | mp->work_rx;
2266 		if (likely(!mp->oom))
2267 			queue_mask |= mp->work_rx_refill;
2268 
2269 		if (!queue_mask) {
2270 			if (mv643xx_eth_collect_events(mp))
2271 				continue;
2272 			break;
2273 		}
2274 
2275 		queue = fls(queue_mask) - 1;
2276 		queue_mask = 1 << queue;
2277 
2278 		work_tbd = budget - work_done;
2279 		if (work_tbd > 16)
2280 			work_tbd = 16;
2281 
2282 		if (mp->work_tx_end & queue_mask) {
2283 			txq_kick(mp->txq + queue);
2284 		} else if (mp->work_tx & queue_mask) {
2285 			work_done += txq_reclaim(mp->txq + queue, work_tbd, 0);
2286 			txq_maybe_wake(mp->txq + queue);
2287 		} else if (mp->work_rx & queue_mask) {
2288 			work_done += rxq_process(mp->rxq + queue, work_tbd);
2289 		} else if (!mp->oom && (mp->work_rx_refill & queue_mask)) {
2290 			work_done += rxq_refill(mp->rxq + queue, work_tbd);
2291 		} else {
2292 			BUG();
2293 		}
2294 	}
2295 
2296 	if (work_done < budget) {
2297 		if (mp->oom)
2298 			mod_timer(&mp->rx_oom, jiffies + (HZ / 10));
2299 		napi_complete_done(napi, work_done);
2300 		wrlp(mp, INT_MASK, mp->int_mask);
2301 	}
2302 
2303 	return work_done;
2304 }
2305 
2306 static inline void oom_timer_wrapper(struct timer_list *t)
2307 {
2308 	struct mv643xx_eth_private *mp = from_timer(mp, t, rx_oom);
2309 
2310 	napi_schedule(&mp->napi);
2311 }
2312 
2313 static void port_start(struct mv643xx_eth_private *mp)
2314 {
2315 	struct net_device *dev = mp->dev;
2316 	u32 pscr;
2317 	int i;
2318 
2319 	/*
2320 	 * Perform PHY reset, if there is a PHY.
2321 	 */
2322 	if (dev->phydev) {
2323 		struct ethtool_link_ksettings cmd;
2324 
2325 		mv643xx_eth_get_link_ksettings(dev, &cmd);
2326 		phy_init_hw(dev->phydev);
2327 		mv643xx_eth_set_link_ksettings(
2328 			dev, (const struct ethtool_link_ksettings *)&cmd);
2329 		phy_start(dev->phydev);
2330 	}
2331 
2332 	/*
2333 	 * Configure basic link parameters.
2334 	 */
2335 	pscr = rdlp(mp, PORT_SERIAL_CONTROL);
2336 
2337 	pscr |= SERIAL_PORT_ENABLE;
2338 	wrlp(mp, PORT_SERIAL_CONTROL, pscr);
2339 
2340 	pscr |= DO_NOT_FORCE_LINK_FAIL;
2341 	if (!dev->phydev)
2342 		pscr |= FORCE_LINK_PASS;
2343 	wrlp(mp, PORT_SERIAL_CONTROL, pscr);
2344 
2345 	/*
2346 	 * Configure TX path and queues.
2347 	 */
2348 	tx_set_rate(mp, 1000000000, 16777216);
2349 	for (i = 0; i < mp->txq_count; i++) {
2350 		struct tx_queue *txq = mp->txq + i;
2351 
2352 		txq_reset_hw_ptr(txq);
2353 		txq_set_rate(txq, 1000000000, 16777216);
2354 		txq_set_fixed_prio_mode(txq);
2355 	}
2356 
2357 	/*
2358 	 * Receive all unmatched unicast, TCP, UDP, BPDU and broadcast
2359 	 * frames to RX queue #0, and include the pseudo-header when
2360 	 * calculating receive checksums.
2361 	 */
2362 	mv643xx_eth_set_features(mp->dev, mp->dev->features);
2363 
2364 	/*
2365 	 * Treat BPDUs as normal multicasts, and disable partition mode.
2366 	 */
2367 	wrlp(mp, PORT_CONFIG_EXT, 0x00000000);
2368 
2369 	/*
2370 	 * Add configured unicast addresses to address filter table.
2371 	 */
2372 	mv643xx_eth_program_unicast_filter(mp->dev);
2373 
2374 	/*
2375 	 * Enable the receive queues.
2376 	 */
2377 	for (i = 0; i < mp->rxq_count; i++) {
2378 		struct rx_queue *rxq = mp->rxq + i;
2379 		u32 addr;
2380 
2381 		addr = (u32)rxq->rx_desc_dma;
2382 		addr += rxq->rx_curr_desc * sizeof(struct rx_desc);
2383 		wrlp(mp, RXQ_CURRENT_DESC_PTR(i), addr);
2384 
2385 		rxq_enable(rxq);
2386 	}
2387 }
2388 
2389 static void mv643xx_eth_recalc_skb_size(struct mv643xx_eth_private *mp)
2390 {
2391 	int skb_size;
2392 
2393 	/*
2394 	 * Reserve 2+14 bytes for an ethernet header (the hardware
2395 	 * automatically prepends 2 bytes of dummy data to each
2396 	 * received packet), 16 bytes for up to four VLAN tags, and
2397 	 * 4 bytes for the trailing FCS -- 36 bytes total.
2398 	 */
2399 	skb_size = mp->dev->mtu + 36;
2400 
2401 	/*
2402 	 * Make sure that the skb size is a multiple of 8 bytes, as
2403 	 * the lower three bits of the receive descriptor's buffer
2404 	 * size field are ignored by the hardware.
2405 	 */
2406 	mp->skb_size = (skb_size + 7) & ~7;
2407 
2408 	/*
2409 	 * If NET_SKB_PAD is smaller than a cache line,
2410 	 * netdev_alloc_skb() will cause skb->data to be misaligned
2411 	 * to a cache line boundary.  If this is the case, include
2412 	 * some extra space to allow re-aligning the data area.
2413 	 */
2414 	mp->skb_size += SKB_DMA_REALIGN;
2415 }
2416 
2417 static int mv643xx_eth_open(struct net_device *dev)
2418 {
2419 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2420 	int err;
2421 	int i;
2422 
2423 	wrlp(mp, INT_CAUSE, 0);
2424 	wrlp(mp, INT_CAUSE_EXT, 0);
2425 	rdlp(mp, INT_CAUSE_EXT);
2426 
2427 	err = request_irq(dev->irq, mv643xx_eth_irq,
2428 			  IRQF_SHARED, dev->name, dev);
2429 	if (err) {
2430 		netdev_err(dev, "can't assign irq\n");
2431 		return -EAGAIN;
2432 	}
2433 
2434 	mv643xx_eth_recalc_skb_size(mp);
2435 
2436 	napi_enable(&mp->napi);
2437 
2438 	mp->int_mask = INT_EXT;
2439 
2440 	for (i = 0; i < mp->rxq_count; i++) {
2441 		err = rxq_init(mp, i);
2442 		if (err) {
2443 			while (--i >= 0)
2444 				rxq_deinit(mp->rxq + i);
2445 			goto out;
2446 		}
2447 
2448 		rxq_refill(mp->rxq + i, INT_MAX);
2449 		mp->int_mask |= INT_RX_0 << i;
2450 	}
2451 
2452 	if (mp->oom) {
2453 		mp->rx_oom.expires = jiffies + (HZ / 10);
2454 		add_timer(&mp->rx_oom);
2455 	}
2456 
2457 	for (i = 0; i < mp->txq_count; i++) {
2458 		err = txq_init(mp, i);
2459 		if (err) {
2460 			while (--i >= 0)
2461 				txq_deinit(mp->txq + i);
2462 			goto out_free;
2463 		}
2464 		mp->int_mask |= INT_TX_END_0 << i;
2465 	}
2466 
2467 	add_timer(&mp->mib_counters_timer);
2468 	port_start(mp);
2469 
2470 	wrlp(mp, INT_MASK_EXT, INT_EXT_LINK_PHY | INT_EXT_TX);
2471 	wrlp(mp, INT_MASK, mp->int_mask);
2472 
2473 	return 0;
2474 
2475 
2476 out_free:
2477 	for (i = 0; i < mp->rxq_count; i++)
2478 		rxq_deinit(mp->rxq + i);
2479 out:
2480 	free_irq(dev->irq, dev);
2481 
2482 	return err;
2483 }
2484 
2485 static void port_reset(struct mv643xx_eth_private *mp)
2486 {
2487 	unsigned int data;
2488 	int i;
2489 
2490 	for (i = 0; i < mp->rxq_count; i++)
2491 		rxq_disable(mp->rxq + i);
2492 	for (i = 0; i < mp->txq_count; i++)
2493 		txq_disable(mp->txq + i);
2494 
2495 	while (1) {
2496 		u32 ps = rdlp(mp, PORT_STATUS);
2497 
2498 		if ((ps & (TX_IN_PROGRESS | TX_FIFO_EMPTY)) == TX_FIFO_EMPTY)
2499 			break;
2500 		udelay(10);
2501 	}
2502 
2503 	/* Reset the Enable bit in the Configuration Register */
2504 	data = rdlp(mp, PORT_SERIAL_CONTROL);
2505 	data &= ~(SERIAL_PORT_ENABLE		|
2506 		  DO_NOT_FORCE_LINK_FAIL	|
2507 		  FORCE_LINK_PASS);
2508 	wrlp(mp, PORT_SERIAL_CONTROL, data);
2509 }
2510 
2511 static int mv643xx_eth_stop(struct net_device *dev)
2512 {
2513 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2514 	int i;
2515 
2516 	wrlp(mp, INT_MASK_EXT, 0x00000000);
2517 	wrlp(mp, INT_MASK, 0x00000000);
2518 	rdlp(mp, INT_MASK);
2519 
2520 	napi_disable(&mp->napi);
2521 
2522 	del_timer_sync(&mp->rx_oom);
2523 
2524 	netif_carrier_off(dev);
2525 	if (dev->phydev)
2526 		phy_stop(dev->phydev);
2527 	free_irq(dev->irq, dev);
2528 
2529 	port_reset(mp);
2530 	mv643xx_eth_get_stats(dev);
2531 	mib_counters_update(mp);
2532 	del_timer_sync(&mp->mib_counters_timer);
2533 
2534 	for (i = 0; i < mp->rxq_count; i++)
2535 		rxq_deinit(mp->rxq + i);
2536 	for (i = 0; i < mp->txq_count; i++)
2537 		txq_deinit(mp->txq + i);
2538 
2539 	return 0;
2540 }
2541 
2542 static int mv643xx_eth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2543 {
2544 	int ret;
2545 
2546 	if (!dev->phydev)
2547 		return -ENOTSUPP;
2548 
2549 	ret = phy_mii_ioctl(dev->phydev, ifr, cmd);
2550 	if (!ret)
2551 		mv643xx_eth_adjust_link(dev);
2552 	return ret;
2553 }
2554 
2555 static int mv643xx_eth_change_mtu(struct net_device *dev, int new_mtu)
2556 {
2557 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2558 
2559 	dev->mtu = new_mtu;
2560 	mv643xx_eth_recalc_skb_size(mp);
2561 	tx_set_rate(mp, 1000000000, 16777216);
2562 
2563 	if (!netif_running(dev))
2564 		return 0;
2565 
2566 	/*
2567 	 * Stop and then re-open the interface. This will allocate RX
2568 	 * skbs of the new MTU.
2569 	 * There is a possible danger that the open will not succeed,
2570 	 * due to memory being full.
2571 	 */
2572 	mv643xx_eth_stop(dev);
2573 	if (mv643xx_eth_open(dev)) {
2574 		netdev_err(dev,
2575 			   "fatal error on re-opening device after MTU change\n");
2576 	}
2577 
2578 	return 0;
2579 }
2580 
2581 static void tx_timeout_task(struct work_struct *ugly)
2582 {
2583 	struct mv643xx_eth_private *mp;
2584 
2585 	mp = container_of(ugly, struct mv643xx_eth_private, tx_timeout_task);
2586 	if (netif_running(mp->dev)) {
2587 		netif_tx_stop_all_queues(mp->dev);
2588 		port_reset(mp);
2589 		port_start(mp);
2590 		netif_tx_wake_all_queues(mp->dev);
2591 	}
2592 }
2593 
2594 static void mv643xx_eth_tx_timeout(struct net_device *dev, unsigned int txqueue)
2595 {
2596 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2597 
2598 	netdev_info(dev, "tx timeout\n");
2599 
2600 	schedule_work(&mp->tx_timeout_task);
2601 }
2602 
2603 #ifdef CONFIG_NET_POLL_CONTROLLER
2604 static void mv643xx_eth_netpoll(struct net_device *dev)
2605 {
2606 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2607 
2608 	wrlp(mp, INT_MASK, 0x00000000);
2609 	rdlp(mp, INT_MASK);
2610 
2611 	mv643xx_eth_irq(dev->irq, dev);
2612 
2613 	wrlp(mp, INT_MASK, mp->int_mask);
2614 }
2615 #endif
2616 
2617 
2618 /* platform glue ************************************************************/
2619 static void
2620 mv643xx_eth_conf_mbus_windows(struct mv643xx_eth_shared_private *msp,
2621 			      const struct mbus_dram_target_info *dram)
2622 {
2623 	void __iomem *base = msp->base;
2624 	u32 win_enable;
2625 	u32 win_protect;
2626 	int i;
2627 
2628 	for (i = 0; i < 6; i++) {
2629 		writel(0, base + WINDOW_BASE(i));
2630 		writel(0, base + WINDOW_SIZE(i));
2631 		if (i < 4)
2632 			writel(0, base + WINDOW_REMAP_HIGH(i));
2633 	}
2634 
2635 	win_enable = 0x3f;
2636 	win_protect = 0;
2637 
2638 	for (i = 0; i < dram->num_cs; i++) {
2639 		const struct mbus_dram_window *cs = dram->cs + i;
2640 
2641 		writel((cs->base & 0xffff0000) |
2642 			(cs->mbus_attr << 8) |
2643 			dram->mbus_dram_target_id, base + WINDOW_BASE(i));
2644 		writel((cs->size - 1) & 0xffff0000, base + WINDOW_SIZE(i));
2645 
2646 		win_enable &= ~(1 << i);
2647 		win_protect |= 3 << (2 * i);
2648 	}
2649 
2650 	writel(win_enable, base + WINDOW_BAR_ENABLE);
2651 	msp->win_protect = win_protect;
2652 }
2653 
2654 static void infer_hw_params(struct mv643xx_eth_shared_private *msp)
2655 {
2656 	/*
2657 	 * Check whether we have a 14-bit coal limit field in bits
2658 	 * [21:8], or a 16-bit coal limit in bits [25,21:7] of the
2659 	 * SDMA config register.
2660 	 */
2661 	writel(0x02000000, msp->base + 0x0400 + SDMA_CONFIG);
2662 	if (readl(msp->base + 0x0400 + SDMA_CONFIG) & 0x02000000)
2663 		msp->extended_rx_coal_limit = 1;
2664 	else
2665 		msp->extended_rx_coal_limit = 0;
2666 
2667 	/*
2668 	 * Check whether the MAC supports TX rate control, and if
2669 	 * yes, whether its associated registers are in the old or
2670 	 * the new place.
2671 	 */
2672 	writel(1, msp->base + 0x0400 + TX_BW_MTU_MOVED);
2673 	if (readl(msp->base + 0x0400 + TX_BW_MTU_MOVED) & 1) {
2674 		msp->tx_bw_control = TX_BW_CONTROL_NEW_LAYOUT;
2675 	} else {
2676 		writel(7, msp->base + 0x0400 + TX_BW_RATE);
2677 		if (readl(msp->base + 0x0400 + TX_BW_RATE) & 7)
2678 			msp->tx_bw_control = TX_BW_CONTROL_OLD_LAYOUT;
2679 		else
2680 			msp->tx_bw_control = TX_BW_CONTROL_ABSENT;
2681 	}
2682 }
2683 
2684 #if defined(CONFIG_OF)
2685 static const struct of_device_id mv643xx_eth_shared_ids[] = {
2686 	{ .compatible = "marvell,orion-eth", },
2687 	{ .compatible = "marvell,kirkwood-eth", },
2688 	{ }
2689 };
2690 MODULE_DEVICE_TABLE(of, mv643xx_eth_shared_ids);
2691 #endif
2692 
2693 #ifdef CONFIG_OF_IRQ
2694 #define mv643xx_eth_property(_np, _name, _v)				\
2695 	do {								\
2696 		u32 tmp;						\
2697 		if (!of_property_read_u32(_np, "marvell," _name, &tmp))	\
2698 			_v = tmp;					\
2699 	} while (0)
2700 
2701 static struct platform_device *port_platdev[3];
2702 
2703 static int mv643xx_eth_shared_of_add_port(struct platform_device *pdev,
2704 					  struct device_node *pnp)
2705 {
2706 	struct platform_device *ppdev;
2707 	struct mv643xx_eth_platform_data ppd;
2708 	struct resource res;
2709 	int ret;
2710 	int dev_num = 0;
2711 
2712 	memset(&ppd, 0, sizeof(ppd));
2713 	ppd.shared = pdev;
2714 
2715 	memset(&res, 0, sizeof(res));
2716 	if (of_irq_to_resource(pnp, 0, &res) <= 0) {
2717 		dev_err(&pdev->dev, "missing interrupt on %pOFn\n", pnp);
2718 		return -EINVAL;
2719 	}
2720 
2721 	if (of_property_read_u32(pnp, "reg", &ppd.port_number)) {
2722 		dev_err(&pdev->dev, "missing reg property on %pOFn\n", pnp);
2723 		return -EINVAL;
2724 	}
2725 
2726 	if (ppd.port_number >= 3) {
2727 		dev_err(&pdev->dev, "invalid reg property on %pOFn\n", pnp);
2728 		return -EINVAL;
2729 	}
2730 
2731 	while (dev_num < 3 && port_platdev[dev_num])
2732 		dev_num++;
2733 
2734 	if (dev_num == 3) {
2735 		dev_err(&pdev->dev, "too many ports registered\n");
2736 		return -EINVAL;
2737 	}
2738 
2739 	of_get_mac_address(pnp, ppd.mac_addr);
2740 
2741 	mv643xx_eth_property(pnp, "tx-queue-size", ppd.tx_queue_size);
2742 	mv643xx_eth_property(pnp, "tx-sram-addr", ppd.tx_sram_addr);
2743 	mv643xx_eth_property(pnp, "tx-sram-size", ppd.tx_sram_size);
2744 	mv643xx_eth_property(pnp, "rx-queue-size", ppd.rx_queue_size);
2745 	mv643xx_eth_property(pnp, "rx-sram-addr", ppd.rx_sram_addr);
2746 	mv643xx_eth_property(pnp, "rx-sram-size", ppd.rx_sram_size);
2747 
2748 	ppd.phy_node = of_parse_phandle(pnp, "phy-handle", 0);
2749 	if (!ppd.phy_node) {
2750 		ppd.phy_addr = MV643XX_ETH_PHY_NONE;
2751 		of_property_read_u32(pnp, "speed", &ppd.speed);
2752 		of_property_read_u32(pnp, "duplex", &ppd.duplex);
2753 	}
2754 
2755 	ppdev = platform_device_alloc(MV643XX_ETH_NAME, dev_num);
2756 	if (!ppdev)
2757 		return -ENOMEM;
2758 	ppdev->dev.coherent_dma_mask = DMA_BIT_MASK(32);
2759 	ppdev->dev.of_node = pnp;
2760 
2761 	ret = platform_device_add_resources(ppdev, &res, 1);
2762 	if (ret)
2763 		goto port_err;
2764 
2765 	ret = platform_device_add_data(ppdev, &ppd, sizeof(ppd));
2766 	if (ret)
2767 		goto port_err;
2768 
2769 	ret = platform_device_add(ppdev);
2770 	if (ret)
2771 		goto port_err;
2772 
2773 	port_platdev[dev_num] = ppdev;
2774 
2775 	return 0;
2776 
2777 port_err:
2778 	platform_device_put(ppdev);
2779 	return ret;
2780 }
2781 
2782 static int mv643xx_eth_shared_of_probe(struct platform_device *pdev)
2783 {
2784 	struct mv643xx_eth_shared_platform_data *pd;
2785 	struct device_node *pnp, *np = pdev->dev.of_node;
2786 	int ret;
2787 
2788 	/* bail out if not registered from DT */
2789 	if (!np)
2790 		return 0;
2791 
2792 	pd = devm_kzalloc(&pdev->dev, sizeof(*pd), GFP_KERNEL);
2793 	if (!pd)
2794 		return -ENOMEM;
2795 	pdev->dev.platform_data = pd;
2796 
2797 	mv643xx_eth_property(np, "tx-checksum-limit", pd->tx_csum_limit);
2798 
2799 	for_each_available_child_of_node(np, pnp) {
2800 		ret = mv643xx_eth_shared_of_add_port(pdev, pnp);
2801 		if (ret) {
2802 			of_node_put(pnp);
2803 			return ret;
2804 		}
2805 	}
2806 	return 0;
2807 }
2808 
2809 static void mv643xx_eth_shared_of_remove(void)
2810 {
2811 	int n;
2812 
2813 	for (n = 0; n < 3; n++) {
2814 		platform_device_del(port_platdev[n]);
2815 		port_platdev[n] = NULL;
2816 	}
2817 }
2818 #else
2819 static inline int mv643xx_eth_shared_of_probe(struct platform_device *pdev)
2820 {
2821 	return 0;
2822 }
2823 
2824 static inline void mv643xx_eth_shared_of_remove(void)
2825 {
2826 }
2827 #endif
2828 
2829 static int mv643xx_eth_shared_probe(struct platform_device *pdev)
2830 {
2831 	static int mv643xx_eth_version_printed;
2832 	struct mv643xx_eth_shared_platform_data *pd;
2833 	struct mv643xx_eth_shared_private *msp;
2834 	const struct mbus_dram_target_info *dram;
2835 	struct resource *res;
2836 	int ret;
2837 
2838 	if (!mv643xx_eth_version_printed++)
2839 		pr_notice("MV-643xx 10/100/1000 ethernet driver version %s\n",
2840 			  mv643xx_eth_driver_version);
2841 
2842 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
2843 	if (res == NULL)
2844 		return -EINVAL;
2845 
2846 	msp = devm_kzalloc(&pdev->dev, sizeof(*msp), GFP_KERNEL);
2847 	if (msp == NULL)
2848 		return -ENOMEM;
2849 	platform_set_drvdata(pdev, msp);
2850 
2851 	msp->base = devm_ioremap(&pdev->dev, res->start, resource_size(res));
2852 	if (msp->base == NULL)
2853 		return -ENOMEM;
2854 
2855 	msp->clk = devm_clk_get(&pdev->dev, NULL);
2856 	if (!IS_ERR(msp->clk))
2857 		clk_prepare_enable(msp->clk);
2858 
2859 	/*
2860 	 * (Re-)program MBUS remapping windows if we are asked to.
2861 	 */
2862 	dram = mv_mbus_dram_info();
2863 	if (dram)
2864 		mv643xx_eth_conf_mbus_windows(msp, dram);
2865 
2866 	ret = mv643xx_eth_shared_of_probe(pdev);
2867 	if (ret)
2868 		goto err_put_clk;
2869 	pd = dev_get_platdata(&pdev->dev);
2870 
2871 	msp->tx_csum_limit = (pd != NULL && pd->tx_csum_limit) ?
2872 					pd->tx_csum_limit : 9 * 1024;
2873 	infer_hw_params(msp);
2874 
2875 	return 0;
2876 
2877 err_put_clk:
2878 	if (!IS_ERR(msp->clk))
2879 		clk_disable_unprepare(msp->clk);
2880 	return ret;
2881 }
2882 
2883 static int mv643xx_eth_shared_remove(struct platform_device *pdev)
2884 {
2885 	struct mv643xx_eth_shared_private *msp = platform_get_drvdata(pdev);
2886 
2887 	mv643xx_eth_shared_of_remove();
2888 	if (!IS_ERR(msp->clk))
2889 		clk_disable_unprepare(msp->clk);
2890 	return 0;
2891 }
2892 
2893 static struct platform_driver mv643xx_eth_shared_driver = {
2894 	.probe		= mv643xx_eth_shared_probe,
2895 	.remove		= mv643xx_eth_shared_remove,
2896 	.driver = {
2897 		.name	= MV643XX_ETH_SHARED_NAME,
2898 		.of_match_table = of_match_ptr(mv643xx_eth_shared_ids),
2899 	},
2900 };
2901 
2902 static void phy_addr_set(struct mv643xx_eth_private *mp, int phy_addr)
2903 {
2904 	int addr_shift = 5 * mp->port_num;
2905 	u32 data;
2906 
2907 	data = rdl(mp, PHY_ADDR);
2908 	data &= ~(0x1f << addr_shift);
2909 	data |= (phy_addr & 0x1f) << addr_shift;
2910 	wrl(mp, PHY_ADDR, data);
2911 }
2912 
2913 static int phy_addr_get(struct mv643xx_eth_private *mp)
2914 {
2915 	unsigned int data;
2916 
2917 	data = rdl(mp, PHY_ADDR);
2918 
2919 	return (data >> (5 * mp->port_num)) & 0x1f;
2920 }
2921 
2922 static void set_params(struct mv643xx_eth_private *mp,
2923 		       struct mv643xx_eth_platform_data *pd)
2924 {
2925 	struct net_device *dev = mp->dev;
2926 	unsigned int tx_ring_size;
2927 
2928 	if (is_valid_ether_addr(pd->mac_addr))
2929 		eth_hw_addr_set(dev, pd->mac_addr);
2930 	else
2931 		uc_addr_get(mp, dev->dev_addr);
2932 
2933 	mp->rx_ring_size = DEFAULT_RX_QUEUE_SIZE;
2934 	if (pd->rx_queue_size)
2935 		mp->rx_ring_size = pd->rx_queue_size;
2936 	mp->rx_desc_sram_addr = pd->rx_sram_addr;
2937 	mp->rx_desc_sram_size = pd->rx_sram_size;
2938 
2939 	mp->rxq_count = pd->rx_queue_count ? : 1;
2940 
2941 	tx_ring_size = DEFAULT_TX_QUEUE_SIZE;
2942 	if (pd->tx_queue_size)
2943 		tx_ring_size = pd->tx_queue_size;
2944 
2945 	mp->tx_ring_size = clamp_t(unsigned int, tx_ring_size,
2946 				   MV643XX_MAX_SKB_DESCS * 2, 4096);
2947 	if (mp->tx_ring_size != tx_ring_size)
2948 		netdev_warn(dev, "TX queue size set to %u (requested %u)\n",
2949 			    mp->tx_ring_size, tx_ring_size);
2950 
2951 	mp->tx_desc_sram_addr = pd->tx_sram_addr;
2952 	mp->tx_desc_sram_size = pd->tx_sram_size;
2953 
2954 	mp->txq_count = pd->tx_queue_count ? : 1;
2955 }
2956 
2957 static int get_phy_mode(struct mv643xx_eth_private *mp)
2958 {
2959 	struct device *dev = mp->dev->dev.parent;
2960 	phy_interface_t iface;
2961 	int err;
2962 
2963 	if (dev->of_node)
2964 		err = of_get_phy_mode(dev->of_node, &iface);
2965 
2966 	/* Historical default if unspecified. We could also read/write
2967 	 * the interface state in the PSC1
2968 	 */
2969 	if (!dev->of_node || err)
2970 		iface = PHY_INTERFACE_MODE_GMII;
2971 	return iface;
2972 }
2973 
2974 static struct phy_device *phy_scan(struct mv643xx_eth_private *mp,
2975 				   int phy_addr)
2976 {
2977 	struct phy_device *phydev;
2978 	int start;
2979 	int num;
2980 	int i;
2981 	char phy_id[MII_BUS_ID_SIZE + 3];
2982 
2983 	if (phy_addr == MV643XX_ETH_PHY_ADDR_DEFAULT) {
2984 		start = phy_addr_get(mp) & 0x1f;
2985 		num = 32;
2986 	} else {
2987 		start = phy_addr & 0x1f;
2988 		num = 1;
2989 	}
2990 
2991 	/* Attempt to connect to the PHY using orion-mdio */
2992 	phydev = ERR_PTR(-ENODEV);
2993 	for (i = 0; i < num; i++) {
2994 		int addr = (start + i) & 0x1f;
2995 
2996 		snprintf(phy_id, sizeof(phy_id), PHY_ID_FMT,
2997 				"orion-mdio-mii", addr);
2998 
2999 		phydev = phy_connect(mp->dev, phy_id, mv643xx_eth_adjust_link,
3000 				     get_phy_mode(mp));
3001 		if (!IS_ERR(phydev)) {
3002 			phy_addr_set(mp, addr);
3003 			break;
3004 		}
3005 	}
3006 
3007 	return phydev;
3008 }
3009 
3010 static void phy_init(struct mv643xx_eth_private *mp, int speed, int duplex)
3011 {
3012 	struct net_device *dev = mp->dev;
3013 	struct phy_device *phy = dev->phydev;
3014 
3015 	if (speed == 0) {
3016 		phy->autoneg = AUTONEG_ENABLE;
3017 		phy->speed = 0;
3018 		phy->duplex = 0;
3019 		linkmode_copy(phy->advertising, phy->supported);
3020 		linkmode_set_bit(ETHTOOL_LINK_MODE_Autoneg_BIT,
3021 				 phy->advertising);
3022 	} else {
3023 		phy->autoneg = AUTONEG_DISABLE;
3024 		linkmode_zero(phy->advertising);
3025 		phy->speed = speed;
3026 		phy->duplex = duplex;
3027 	}
3028 	phy_start_aneg(phy);
3029 }
3030 
3031 static void init_pscr(struct mv643xx_eth_private *mp, int speed, int duplex)
3032 {
3033 	struct net_device *dev = mp->dev;
3034 	u32 pscr;
3035 
3036 	pscr = rdlp(mp, PORT_SERIAL_CONTROL);
3037 	if (pscr & SERIAL_PORT_ENABLE) {
3038 		pscr &= ~SERIAL_PORT_ENABLE;
3039 		wrlp(mp, PORT_SERIAL_CONTROL, pscr);
3040 	}
3041 
3042 	pscr = MAX_RX_PACKET_9700BYTE | SERIAL_PORT_CONTROL_RESERVED;
3043 	if (!dev->phydev) {
3044 		pscr |= DISABLE_AUTO_NEG_SPEED_GMII;
3045 		if (speed == SPEED_1000)
3046 			pscr |= SET_GMII_SPEED_TO_1000;
3047 		else if (speed == SPEED_100)
3048 			pscr |= SET_MII_SPEED_TO_100;
3049 
3050 		pscr |= DISABLE_AUTO_NEG_FOR_FLOW_CTRL;
3051 
3052 		pscr |= DISABLE_AUTO_NEG_FOR_DUPLEX;
3053 		if (duplex == DUPLEX_FULL)
3054 			pscr |= SET_FULL_DUPLEX_MODE;
3055 	}
3056 
3057 	wrlp(mp, PORT_SERIAL_CONTROL, pscr);
3058 }
3059 
3060 static const struct net_device_ops mv643xx_eth_netdev_ops = {
3061 	.ndo_open		= mv643xx_eth_open,
3062 	.ndo_stop		= mv643xx_eth_stop,
3063 	.ndo_start_xmit		= mv643xx_eth_xmit,
3064 	.ndo_set_rx_mode	= mv643xx_eth_set_rx_mode,
3065 	.ndo_set_mac_address	= mv643xx_eth_set_mac_address,
3066 	.ndo_validate_addr	= eth_validate_addr,
3067 	.ndo_eth_ioctl		= mv643xx_eth_ioctl,
3068 	.ndo_change_mtu		= mv643xx_eth_change_mtu,
3069 	.ndo_set_features	= mv643xx_eth_set_features,
3070 	.ndo_tx_timeout		= mv643xx_eth_tx_timeout,
3071 	.ndo_get_stats		= mv643xx_eth_get_stats,
3072 #ifdef CONFIG_NET_POLL_CONTROLLER
3073 	.ndo_poll_controller	= mv643xx_eth_netpoll,
3074 #endif
3075 };
3076 
3077 static int mv643xx_eth_probe(struct platform_device *pdev)
3078 {
3079 	struct mv643xx_eth_platform_data *pd;
3080 	struct mv643xx_eth_private *mp;
3081 	struct net_device *dev;
3082 	struct phy_device *phydev = NULL;
3083 	struct resource *res;
3084 	int err;
3085 
3086 	pd = dev_get_platdata(&pdev->dev);
3087 	if (pd == NULL) {
3088 		dev_err(&pdev->dev, "no mv643xx_eth_platform_data\n");
3089 		return -ENODEV;
3090 	}
3091 
3092 	if (pd->shared == NULL) {
3093 		dev_err(&pdev->dev, "no mv643xx_eth_platform_data->shared\n");
3094 		return -ENODEV;
3095 	}
3096 
3097 	dev = alloc_etherdev_mq(sizeof(struct mv643xx_eth_private), 8);
3098 	if (!dev)
3099 		return -ENOMEM;
3100 
3101 	SET_NETDEV_DEV(dev, &pdev->dev);
3102 	mp = netdev_priv(dev);
3103 	platform_set_drvdata(pdev, mp);
3104 
3105 	mp->shared = platform_get_drvdata(pd->shared);
3106 	mp->base = mp->shared->base + 0x0400 + (pd->port_number << 10);
3107 	mp->port_num = pd->port_number;
3108 
3109 	mp->dev = dev;
3110 
3111 	/* Kirkwood resets some registers on gated clocks. Especially
3112 	 * CLK125_BYPASS_EN must be cleared but is not available on
3113 	 * all other SoCs/System Controllers using this driver.
3114 	 */
3115 	if (of_device_is_compatible(pdev->dev.of_node,
3116 				    "marvell,kirkwood-eth-port"))
3117 		wrlp(mp, PORT_SERIAL_CONTROL1,
3118 		     rdlp(mp, PORT_SERIAL_CONTROL1) & ~CLK125_BYPASS_EN);
3119 
3120 	/*
3121 	 * Start with a default rate, and if there is a clock, allow
3122 	 * it to override the default.
3123 	 */
3124 	mp->t_clk = 133000000;
3125 	mp->clk = devm_clk_get(&pdev->dev, NULL);
3126 	if (!IS_ERR(mp->clk)) {
3127 		clk_prepare_enable(mp->clk);
3128 		mp->t_clk = clk_get_rate(mp->clk);
3129 	} else if (!IS_ERR(mp->shared->clk)) {
3130 		mp->t_clk = clk_get_rate(mp->shared->clk);
3131 	}
3132 
3133 	set_params(mp, pd);
3134 	netif_set_real_num_tx_queues(dev, mp->txq_count);
3135 	netif_set_real_num_rx_queues(dev, mp->rxq_count);
3136 
3137 	err = 0;
3138 	if (pd->phy_node) {
3139 		phydev = of_phy_connect(mp->dev, pd->phy_node,
3140 					mv643xx_eth_adjust_link, 0,
3141 					get_phy_mode(mp));
3142 		if (!phydev)
3143 			err = -ENODEV;
3144 		else
3145 			phy_addr_set(mp, phydev->mdio.addr);
3146 	} else if (pd->phy_addr != MV643XX_ETH_PHY_NONE) {
3147 		phydev = phy_scan(mp, pd->phy_addr);
3148 
3149 		if (IS_ERR(phydev))
3150 			err = PTR_ERR(phydev);
3151 		else
3152 			phy_init(mp, pd->speed, pd->duplex);
3153 	}
3154 	if (err == -ENODEV) {
3155 		err = -EPROBE_DEFER;
3156 		goto out;
3157 	}
3158 	if (err)
3159 		goto out;
3160 
3161 	dev->ethtool_ops = &mv643xx_eth_ethtool_ops;
3162 
3163 	init_pscr(mp, pd->speed, pd->duplex);
3164 
3165 
3166 	mib_counters_clear(mp);
3167 
3168 	timer_setup(&mp->mib_counters_timer, mib_counters_timer_wrapper, 0);
3169 	mp->mib_counters_timer.expires = jiffies + 30 * HZ;
3170 
3171 	spin_lock_init(&mp->mib_counters_lock);
3172 
3173 	INIT_WORK(&mp->tx_timeout_task, tx_timeout_task);
3174 
3175 	netif_napi_add(dev, &mp->napi, mv643xx_eth_poll, NAPI_POLL_WEIGHT);
3176 
3177 	timer_setup(&mp->rx_oom, oom_timer_wrapper, 0);
3178 
3179 
3180 	res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
3181 	BUG_ON(!res);
3182 	dev->irq = res->start;
3183 
3184 	dev->netdev_ops = &mv643xx_eth_netdev_ops;
3185 
3186 	dev->watchdog_timeo = 2 * HZ;
3187 	dev->base_addr = 0;
3188 
3189 	dev->features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO;
3190 	dev->vlan_features = dev->features;
3191 
3192 	dev->features |= NETIF_F_RXCSUM;
3193 	dev->hw_features = dev->features;
3194 
3195 	dev->priv_flags |= IFF_UNICAST_FLT;
3196 	dev->gso_max_segs = MV643XX_MAX_TSO_SEGS;
3197 
3198 	/* MTU range: 64 - 9500 */
3199 	dev->min_mtu = 64;
3200 	dev->max_mtu = 9500;
3201 
3202 	if (mp->shared->win_protect)
3203 		wrl(mp, WINDOW_PROTECT(mp->port_num), mp->shared->win_protect);
3204 
3205 	netif_carrier_off(dev);
3206 
3207 	wrlp(mp, SDMA_CONFIG, PORT_SDMA_CONFIG_DEFAULT_VALUE);
3208 
3209 	set_rx_coal(mp, 250);
3210 	set_tx_coal(mp, 0);
3211 
3212 	err = register_netdev(dev);
3213 	if (err)
3214 		goto out;
3215 
3216 	netdev_notice(dev, "port %d with MAC address %pM\n",
3217 		      mp->port_num, dev->dev_addr);
3218 
3219 	if (mp->tx_desc_sram_size > 0)
3220 		netdev_notice(dev, "configured with sram\n");
3221 
3222 	return 0;
3223 
3224 out:
3225 	if (!IS_ERR(mp->clk))
3226 		clk_disable_unprepare(mp->clk);
3227 	free_netdev(dev);
3228 
3229 	return err;
3230 }
3231 
3232 static int mv643xx_eth_remove(struct platform_device *pdev)
3233 {
3234 	struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
3235 	struct net_device *dev = mp->dev;
3236 
3237 	unregister_netdev(mp->dev);
3238 	if (dev->phydev)
3239 		phy_disconnect(dev->phydev);
3240 	cancel_work_sync(&mp->tx_timeout_task);
3241 
3242 	if (!IS_ERR(mp->clk))
3243 		clk_disable_unprepare(mp->clk);
3244 
3245 	free_netdev(mp->dev);
3246 
3247 	return 0;
3248 }
3249 
3250 static void mv643xx_eth_shutdown(struct platform_device *pdev)
3251 {
3252 	struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
3253 
3254 	/* Mask all interrupts on ethernet port */
3255 	wrlp(mp, INT_MASK, 0);
3256 	rdlp(mp, INT_MASK);
3257 
3258 	if (netif_running(mp->dev))
3259 		port_reset(mp);
3260 }
3261 
3262 static struct platform_driver mv643xx_eth_driver = {
3263 	.probe		= mv643xx_eth_probe,
3264 	.remove		= mv643xx_eth_remove,
3265 	.shutdown	= mv643xx_eth_shutdown,
3266 	.driver = {
3267 		.name	= MV643XX_ETH_NAME,
3268 	},
3269 };
3270 
3271 static struct platform_driver * const drivers[] = {
3272 	&mv643xx_eth_shared_driver,
3273 	&mv643xx_eth_driver,
3274 };
3275 
3276 static int __init mv643xx_eth_init_module(void)
3277 {
3278 	return platform_register_drivers(drivers, ARRAY_SIZE(drivers));
3279 }
3280 module_init(mv643xx_eth_init_module);
3281 
3282 static void __exit mv643xx_eth_cleanup_module(void)
3283 {
3284 	platform_unregister_drivers(drivers, ARRAY_SIZE(drivers));
3285 }
3286 module_exit(mv643xx_eth_cleanup_module);
3287 
3288 MODULE_AUTHOR("Rabeeh Khoury, Assaf Hoffman, Matthew Dharm, "
3289 	      "Manish Lachwani, Dale Farnsworth and Lennert Buytenhek");
3290 MODULE_DESCRIPTION("Ethernet driver for Marvell MV643XX");
3291 MODULE_LICENSE("GPL");
3292 MODULE_ALIAS("platform:" MV643XX_ETH_SHARED_NAME);
3293 MODULE_ALIAS("platform:" MV643XX_ETH_NAME);
3294