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 		switch (ip_hdr(skb)->protocol) {
705 		case IPPROTO_UDP:
706 			cmd |= UDP_FRAME;
707 			*l4i_chk = 0;
708 			break;
709 		case IPPROTO_TCP:
710 			*l4i_chk = 0;
711 			break;
712 		default:
713 			WARN(1, "protocol not supported");
714 		}
715 	} else {
716 no_csum:
717 		/* Errata BTS #50, IHL must be 5 if no HW checksum */
718 		cmd |= 5 << TX_IHL_SHIFT;
719 	}
720 	*command = cmd;
721 	return 0;
722 }
723 
724 static inline int
725 txq_put_data_tso(struct net_device *dev, struct tx_queue *txq,
726 		 struct sk_buff *skb, char *data, int length,
727 		 bool last_tcp, bool is_last)
728 {
729 	int tx_index;
730 	u32 cmd_sts;
731 	struct tx_desc *desc;
732 
733 	tx_index = txq->tx_curr_desc++;
734 	if (txq->tx_curr_desc == txq->tx_ring_size)
735 		txq->tx_curr_desc = 0;
736 	desc = &txq->tx_desc_area[tx_index];
737 	txq->tx_desc_mapping[tx_index] = DESC_DMA_MAP_SINGLE;
738 
739 	desc->l4i_chk = 0;
740 	desc->byte_cnt = length;
741 
742 	if (length <= 8 && (uintptr_t)data & 0x7) {
743 		/* Copy unaligned small data fragment to TSO header data area */
744 		memcpy(txq->tso_hdrs + tx_index * TSO_HEADER_SIZE,
745 		       data, length);
746 		desc->buf_ptr = txq->tso_hdrs_dma
747 			+ tx_index * TSO_HEADER_SIZE;
748 	} else {
749 		/* Alignment is okay, map buffer and hand off to hardware */
750 		txq->tx_desc_mapping[tx_index] = DESC_DMA_MAP_SINGLE;
751 		desc->buf_ptr = dma_map_single(dev->dev.parent, data,
752 			length, DMA_TO_DEVICE);
753 		if (unlikely(dma_mapping_error(dev->dev.parent,
754 					       desc->buf_ptr))) {
755 			WARN(1, "dma_map_single failed!\n");
756 			return -ENOMEM;
757 		}
758 	}
759 
760 	cmd_sts = BUFFER_OWNED_BY_DMA;
761 	if (last_tcp) {
762 		/* last descriptor in the TCP packet */
763 		cmd_sts |= ZERO_PADDING | TX_LAST_DESC;
764 		/* last descriptor in SKB */
765 		if (is_last)
766 			cmd_sts |= TX_ENABLE_INTERRUPT;
767 	}
768 	desc->cmd_sts = cmd_sts;
769 	return 0;
770 }
771 
772 static inline void
773 txq_put_hdr_tso(struct sk_buff *skb, struct tx_queue *txq, int length,
774 		u32 *first_cmd_sts, bool first_desc)
775 {
776 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
777 	int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
778 	int tx_index;
779 	struct tx_desc *desc;
780 	int ret;
781 	u32 cmd_csum = 0;
782 	u16 l4i_chk = 0;
783 	u32 cmd_sts;
784 
785 	tx_index = txq->tx_curr_desc;
786 	desc = &txq->tx_desc_area[tx_index];
787 
788 	ret = skb_tx_csum(mp, skb, &l4i_chk, &cmd_csum, length);
789 	if (ret)
790 		WARN(1, "failed to prepare checksum!");
791 
792 	/* Should we set this? Can't use the value from skb_tx_csum()
793 	 * as it's not the correct initial L4 checksum to use. */
794 	desc->l4i_chk = 0;
795 
796 	desc->byte_cnt = hdr_len;
797 	desc->buf_ptr = txq->tso_hdrs_dma +
798 			txq->tx_curr_desc * TSO_HEADER_SIZE;
799 	cmd_sts = cmd_csum | BUFFER_OWNED_BY_DMA  | TX_FIRST_DESC |
800 				   GEN_CRC;
801 
802 	/* Defer updating the first command descriptor until all
803 	 * following descriptors have been written.
804 	 */
805 	if (first_desc)
806 		*first_cmd_sts = cmd_sts;
807 	else
808 		desc->cmd_sts = cmd_sts;
809 
810 	txq->tx_curr_desc++;
811 	if (txq->tx_curr_desc == txq->tx_ring_size)
812 		txq->tx_curr_desc = 0;
813 }
814 
815 static int txq_submit_tso(struct tx_queue *txq, struct sk_buff *skb,
816 			  struct net_device *dev)
817 {
818 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
819 	int hdr_len, total_len, data_left, ret;
820 	int desc_count = 0;
821 	struct tso_t tso;
822 	struct tx_desc *first_tx_desc;
823 	u32 first_cmd_sts = 0;
824 
825 	/* Count needed descriptors */
826 	if ((txq->tx_desc_count + tso_count_descs(skb)) >= txq->tx_ring_size) {
827 		netdev_dbg(dev, "not enough descriptors for TSO!\n");
828 		return -EBUSY;
829 	}
830 
831 	first_tx_desc = &txq->tx_desc_area[txq->tx_curr_desc];
832 
833 	/* Initialize the TSO handler, and prepare the first payload */
834 	hdr_len = tso_start(skb, &tso);
835 
836 	total_len = skb->len - hdr_len;
837 	while (total_len > 0) {
838 		bool first_desc = (desc_count == 0);
839 		char *hdr;
840 
841 		data_left = min_t(int, skb_shinfo(skb)->gso_size, total_len);
842 		total_len -= data_left;
843 		desc_count++;
844 
845 		/* prepare packet headers: MAC + IP + TCP */
846 		hdr = txq->tso_hdrs + txq->tx_curr_desc * TSO_HEADER_SIZE;
847 		tso_build_hdr(skb, hdr, &tso, data_left, total_len == 0);
848 		txq_put_hdr_tso(skb, txq, data_left, &first_cmd_sts,
849 				first_desc);
850 
851 		while (data_left > 0) {
852 			int size;
853 			desc_count++;
854 
855 			size = min_t(int, tso.size, data_left);
856 			ret = txq_put_data_tso(dev, txq, skb, tso.data, size,
857 					       size == data_left,
858 					       total_len == 0);
859 			if (ret)
860 				goto err_release;
861 			data_left -= size;
862 			tso_build_data(skb, &tso, size);
863 		}
864 	}
865 
866 	__skb_queue_tail(&txq->tx_skb, skb);
867 	skb_tx_timestamp(skb);
868 
869 	/* ensure all other descriptors are written before first cmd_sts */
870 	wmb();
871 	first_tx_desc->cmd_sts = first_cmd_sts;
872 
873 	/* clear TX_END status */
874 	mp->work_tx_end &= ~(1 << txq->index);
875 
876 	/* ensure all descriptors are written before poking hardware */
877 	wmb();
878 	txq_enable(txq);
879 	txq->tx_desc_count += desc_count;
880 	return 0;
881 err_release:
882 	/* TODO: Release all used data descriptors; header descriptors must not
883 	 * be DMA-unmapped.
884 	 */
885 	return ret;
886 }
887 
888 static void txq_submit_frag_skb(struct tx_queue *txq, struct sk_buff *skb)
889 {
890 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
891 	int nr_frags = skb_shinfo(skb)->nr_frags;
892 	int frag;
893 
894 	for (frag = 0; frag < nr_frags; frag++) {
895 		skb_frag_t *this_frag;
896 		int tx_index;
897 		struct tx_desc *desc;
898 
899 		this_frag = &skb_shinfo(skb)->frags[frag];
900 		tx_index = txq->tx_curr_desc++;
901 		if (txq->tx_curr_desc == txq->tx_ring_size)
902 			txq->tx_curr_desc = 0;
903 		desc = &txq->tx_desc_area[tx_index];
904 		txq->tx_desc_mapping[tx_index] = DESC_DMA_MAP_PAGE;
905 
906 		/*
907 		 * The last fragment will generate an interrupt
908 		 * which will free the skb on TX completion.
909 		 */
910 		if (frag == nr_frags - 1) {
911 			desc->cmd_sts = BUFFER_OWNED_BY_DMA |
912 					ZERO_PADDING | TX_LAST_DESC |
913 					TX_ENABLE_INTERRUPT;
914 		} else {
915 			desc->cmd_sts = BUFFER_OWNED_BY_DMA;
916 		}
917 
918 		desc->l4i_chk = 0;
919 		desc->byte_cnt = skb_frag_size(this_frag);
920 		desc->buf_ptr = skb_frag_dma_map(mp->dev->dev.parent,
921 						 this_frag, 0, desc->byte_cnt,
922 						 DMA_TO_DEVICE);
923 	}
924 }
925 
926 static int txq_submit_skb(struct tx_queue *txq, struct sk_buff *skb,
927 			  struct net_device *dev)
928 {
929 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
930 	int nr_frags = skb_shinfo(skb)->nr_frags;
931 	int tx_index;
932 	struct tx_desc *desc;
933 	u32 cmd_sts;
934 	u16 l4i_chk;
935 	int length, ret;
936 
937 	cmd_sts = 0;
938 	l4i_chk = 0;
939 
940 	if (txq->tx_ring_size - txq->tx_desc_count < MAX_SKB_FRAGS + 1) {
941 		if (net_ratelimit())
942 			netdev_err(dev, "tx queue full?!\n");
943 		return -EBUSY;
944 	}
945 
946 	ret = skb_tx_csum(mp, skb, &l4i_chk, &cmd_sts, skb->len);
947 	if (ret)
948 		return ret;
949 	cmd_sts |= TX_FIRST_DESC | GEN_CRC | BUFFER_OWNED_BY_DMA;
950 
951 	tx_index = txq->tx_curr_desc++;
952 	if (txq->tx_curr_desc == txq->tx_ring_size)
953 		txq->tx_curr_desc = 0;
954 	desc = &txq->tx_desc_area[tx_index];
955 	txq->tx_desc_mapping[tx_index] = DESC_DMA_MAP_SINGLE;
956 
957 	if (nr_frags) {
958 		txq_submit_frag_skb(txq, skb);
959 		length = skb_headlen(skb);
960 	} else {
961 		cmd_sts |= ZERO_PADDING | TX_LAST_DESC | TX_ENABLE_INTERRUPT;
962 		length = skb->len;
963 	}
964 
965 	desc->l4i_chk = l4i_chk;
966 	desc->byte_cnt = length;
967 	desc->buf_ptr = dma_map_single(mp->dev->dev.parent, skb->data,
968 				       length, DMA_TO_DEVICE);
969 
970 	__skb_queue_tail(&txq->tx_skb, skb);
971 
972 	skb_tx_timestamp(skb);
973 
974 	/* ensure all other descriptors are written before first cmd_sts */
975 	wmb();
976 	desc->cmd_sts = cmd_sts;
977 
978 	/* clear TX_END status */
979 	mp->work_tx_end &= ~(1 << txq->index);
980 
981 	/* ensure all descriptors are written before poking hardware */
982 	wmb();
983 	txq_enable(txq);
984 
985 	txq->tx_desc_count += nr_frags + 1;
986 
987 	return 0;
988 }
989 
990 static netdev_tx_t mv643xx_eth_xmit(struct sk_buff *skb, struct net_device *dev)
991 {
992 	struct mv643xx_eth_private *mp = netdev_priv(dev);
993 	int length, queue, ret;
994 	struct tx_queue *txq;
995 	struct netdev_queue *nq;
996 
997 	queue = skb_get_queue_mapping(skb);
998 	txq = mp->txq + queue;
999 	nq = netdev_get_tx_queue(dev, queue);
1000 
1001 	if (has_tiny_unaligned_frags(skb) && __skb_linearize(skb)) {
1002 		netdev_printk(KERN_DEBUG, dev,
1003 			      "failed to linearize skb with tiny unaligned fragment\n");
1004 		return NETDEV_TX_BUSY;
1005 	}
1006 
1007 	length = skb->len;
1008 
1009 	if (skb_is_gso(skb))
1010 		ret = txq_submit_tso(txq, skb, dev);
1011 	else
1012 		ret = txq_submit_skb(txq, skb, dev);
1013 	if (!ret) {
1014 		txq->tx_bytes += length;
1015 		txq->tx_packets++;
1016 
1017 		if (txq->tx_desc_count >= txq->tx_stop_threshold)
1018 			netif_tx_stop_queue(nq);
1019 	} else {
1020 		txq->tx_dropped++;
1021 		dev_kfree_skb_any(skb);
1022 	}
1023 
1024 	return NETDEV_TX_OK;
1025 }
1026 
1027 
1028 /* tx napi ******************************************************************/
1029 static void txq_kick(struct tx_queue *txq)
1030 {
1031 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
1032 	struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
1033 	u32 hw_desc_ptr;
1034 	u32 expected_ptr;
1035 
1036 	__netif_tx_lock(nq, smp_processor_id());
1037 
1038 	if (rdlp(mp, TXQ_COMMAND) & (1 << txq->index))
1039 		goto out;
1040 
1041 	hw_desc_ptr = rdlp(mp, TXQ_CURRENT_DESC_PTR(txq->index));
1042 	expected_ptr = (u32)txq->tx_desc_dma +
1043 				txq->tx_curr_desc * sizeof(struct tx_desc);
1044 
1045 	if (hw_desc_ptr != expected_ptr)
1046 		txq_enable(txq);
1047 
1048 out:
1049 	__netif_tx_unlock(nq);
1050 
1051 	mp->work_tx_end &= ~(1 << txq->index);
1052 }
1053 
1054 static int txq_reclaim(struct tx_queue *txq, int budget, int force)
1055 {
1056 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
1057 	struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
1058 	int reclaimed;
1059 
1060 	__netif_tx_lock_bh(nq);
1061 
1062 	reclaimed = 0;
1063 	while (reclaimed < budget && txq->tx_desc_count > 0) {
1064 		int tx_index;
1065 		struct tx_desc *desc;
1066 		u32 cmd_sts;
1067 		char desc_dma_map;
1068 
1069 		tx_index = txq->tx_used_desc;
1070 		desc = &txq->tx_desc_area[tx_index];
1071 		desc_dma_map = txq->tx_desc_mapping[tx_index];
1072 
1073 		cmd_sts = desc->cmd_sts;
1074 
1075 		if (cmd_sts & BUFFER_OWNED_BY_DMA) {
1076 			if (!force)
1077 				break;
1078 			desc->cmd_sts = cmd_sts & ~BUFFER_OWNED_BY_DMA;
1079 		}
1080 
1081 		txq->tx_used_desc = tx_index + 1;
1082 		if (txq->tx_used_desc == txq->tx_ring_size)
1083 			txq->tx_used_desc = 0;
1084 
1085 		reclaimed++;
1086 		txq->tx_desc_count--;
1087 
1088 		if (!IS_TSO_HEADER(txq, desc->buf_ptr)) {
1089 
1090 			if (desc_dma_map == DESC_DMA_MAP_PAGE)
1091 				dma_unmap_page(mp->dev->dev.parent,
1092 					       desc->buf_ptr,
1093 					       desc->byte_cnt,
1094 					       DMA_TO_DEVICE);
1095 			else
1096 				dma_unmap_single(mp->dev->dev.parent,
1097 						 desc->buf_ptr,
1098 						 desc->byte_cnt,
1099 						 DMA_TO_DEVICE);
1100 		}
1101 
1102 		if (cmd_sts & TX_ENABLE_INTERRUPT) {
1103 			struct sk_buff *skb = __skb_dequeue(&txq->tx_skb);
1104 
1105 			if (!WARN_ON(!skb))
1106 				dev_consume_skb_any(skb);
1107 		}
1108 
1109 		if (cmd_sts & ERROR_SUMMARY) {
1110 			netdev_info(mp->dev, "tx error\n");
1111 			mp->dev->stats.tx_errors++;
1112 		}
1113 
1114 	}
1115 
1116 	__netif_tx_unlock_bh(nq);
1117 
1118 	if (reclaimed < budget)
1119 		mp->work_tx &= ~(1 << txq->index);
1120 
1121 	return reclaimed;
1122 }
1123 
1124 
1125 /* tx rate control **********************************************************/
1126 /*
1127  * Set total maximum TX rate (shared by all TX queues for this port)
1128  * to 'rate' bits per second, with a maximum burst of 'burst' bytes.
1129  */
1130 static void tx_set_rate(struct mv643xx_eth_private *mp, int rate, int burst)
1131 {
1132 	int token_rate;
1133 	int mtu;
1134 	int bucket_size;
1135 
1136 	token_rate = ((rate / 1000) * 64) / (mp->t_clk / 1000);
1137 	if (token_rate > 1023)
1138 		token_rate = 1023;
1139 
1140 	mtu = (mp->dev->mtu + 255) >> 8;
1141 	if (mtu > 63)
1142 		mtu = 63;
1143 
1144 	bucket_size = (burst + 255) >> 8;
1145 	if (bucket_size > 65535)
1146 		bucket_size = 65535;
1147 
1148 	switch (mp->shared->tx_bw_control) {
1149 	case TX_BW_CONTROL_OLD_LAYOUT:
1150 		wrlp(mp, TX_BW_RATE, token_rate);
1151 		wrlp(mp, TX_BW_MTU, mtu);
1152 		wrlp(mp, TX_BW_BURST, bucket_size);
1153 		break;
1154 	case TX_BW_CONTROL_NEW_LAYOUT:
1155 		wrlp(mp, TX_BW_RATE_MOVED, token_rate);
1156 		wrlp(mp, TX_BW_MTU_MOVED, mtu);
1157 		wrlp(mp, TX_BW_BURST_MOVED, bucket_size);
1158 		break;
1159 	}
1160 }
1161 
1162 static void txq_set_rate(struct tx_queue *txq, int rate, int burst)
1163 {
1164 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
1165 	int token_rate;
1166 	int bucket_size;
1167 
1168 	token_rate = ((rate / 1000) * 64) / (mp->t_clk / 1000);
1169 	if (token_rate > 1023)
1170 		token_rate = 1023;
1171 
1172 	bucket_size = (burst + 255) >> 8;
1173 	if (bucket_size > 65535)
1174 		bucket_size = 65535;
1175 
1176 	wrlp(mp, TXQ_BW_TOKENS(txq->index), token_rate << 14);
1177 	wrlp(mp, TXQ_BW_CONF(txq->index), (bucket_size << 10) | token_rate);
1178 }
1179 
1180 static void txq_set_fixed_prio_mode(struct tx_queue *txq)
1181 {
1182 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
1183 	int off;
1184 	u32 val;
1185 
1186 	/*
1187 	 * Turn on fixed priority mode.
1188 	 */
1189 	off = 0;
1190 	switch (mp->shared->tx_bw_control) {
1191 	case TX_BW_CONTROL_OLD_LAYOUT:
1192 		off = TXQ_FIX_PRIO_CONF;
1193 		break;
1194 	case TX_BW_CONTROL_NEW_LAYOUT:
1195 		off = TXQ_FIX_PRIO_CONF_MOVED;
1196 		break;
1197 	}
1198 
1199 	if (off) {
1200 		val = rdlp(mp, off);
1201 		val |= 1 << txq->index;
1202 		wrlp(mp, off, val);
1203 	}
1204 }
1205 
1206 
1207 /* mii management interface *************************************************/
1208 static void mv643xx_eth_adjust_link(struct net_device *dev)
1209 {
1210 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1211 	u32 pscr = rdlp(mp, PORT_SERIAL_CONTROL);
1212 	u32 autoneg_disable = FORCE_LINK_PASS |
1213 	             DISABLE_AUTO_NEG_SPEED_GMII |
1214 		     DISABLE_AUTO_NEG_FOR_FLOW_CTRL |
1215 		     DISABLE_AUTO_NEG_FOR_DUPLEX;
1216 
1217 	if (dev->phydev->autoneg == AUTONEG_ENABLE) {
1218 		/* enable auto negotiation */
1219 		pscr &= ~autoneg_disable;
1220 		goto out_write;
1221 	}
1222 
1223 	pscr |= autoneg_disable;
1224 
1225 	if (dev->phydev->speed == SPEED_1000) {
1226 		/* force gigabit, half duplex not supported */
1227 		pscr |= SET_GMII_SPEED_TO_1000;
1228 		pscr |= SET_FULL_DUPLEX_MODE;
1229 		goto out_write;
1230 	}
1231 
1232 	pscr &= ~SET_GMII_SPEED_TO_1000;
1233 
1234 	if (dev->phydev->speed == SPEED_100)
1235 		pscr |= SET_MII_SPEED_TO_100;
1236 	else
1237 		pscr &= ~SET_MII_SPEED_TO_100;
1238 
1239 	if (dev->phydev->duplex == DUPLEX_FULL)
1240 		pscr |= SET_FULL_DUPLEX_MODE;
1241 	else
1242 		pscr &= ~SET_FULL_DUPLEX_MODE;
1243 
1244 out_write:
1245 	wrlp(mp, PORT_SERIAL_CONTROL, pscr);
1246 }
1247 
1248 /* statistics ***************************************************************/
1249 static struct net_device_stats *mv643xx_eth_get_stats(struct net_device *dev)
1250 {
1251 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1252 	struct net_device_stats *stats = &dev->stats;
1253 	unsigned long tx_packets = 0;
1254 	unsigned long tx_bytes = 0;
1255 	unsigned long tx_dropped = 0;
1256 	int i;
1257 
1258 	for (i = 0; i < mp->txq_count; i++) {
1259 		struct tx_queue *txq = mp->txq + i;
1260 
1261 		tx_packets += txq->tx_packets;
1262 		tx_bytes += txq->tx_bytes;
1263 		tx_dropped += txq->tx_dropped;
1264 	}
1265 
1266 	stats->tx_packets = tx_packets;
1267 	stats->tx_bytes = tx_bytes;
1268 	stats->tx_dropped = tx_dropped;
1269 
1270 	return stats;
1271 }
1272 
1273 static inline u32 mib_read(struct mv643xx_eth_private *mp, int offset)
1274 {
1275 	return rdl(mp, MIB_COUNTERS(mp->port_num) + offset);
1276 }
1277 
1278 static void mib_counters_clear(struct mv643xx_eth_private *mp)
1279 {
1280 	int i;
1281 
1282 	for (i = 0; i < 0x80; i += 4)
1283 		mib_read(mp, i);
1284 
1285 	/* Clear non MIB hw counters also */
1286 	rdlp(mp, RX_DISCARD_FRAME_CNT);
1287 	rdlp(mp, RX_OVERRUN_FRAME_CNT);
1288 }
1289 
1290 static void mib_counters_update(struct mv643xx_eth_private *mp)
1291 {
1292 	struct mib_counters *p = &mp->mib_counters;
1293 
1294 	spin_lock_bh(&mp->mib_counters_lock);
1295 	p->good_octets_received += mib_read(mp, 0x00);
1296 	p->bad_octets_received += mib_read(mp, 0x08);
1297 	p->internal_mac_transmit_err += mib_read(mp, 0x0c);
1298 	p->good_frames_received += mib_read(mp, 0x10);
1299 	p->bad_frames_received += mib_read(mp, 0x14);
1300 	p->broadcast_frames_received += mib_read(mp, 0x18);
1301 	p->multicast_frames_received += mib_read(mp, 0x1c);
1302 	p->frames_64_octets += mib_read(mp, 0x20);
1303 	p->frames_65_to_127_octets += mib_read(mp, 0x24);
1304 	p->frames_128_to_255_octets += mib_read(mp, 0x28);
1305 	p->frames_256_to_511_octets += mib_read(mp, 0x2c);
1306 	p->frames_512_to_1023_octets += mib_read(mp, 0x30);
1307 	p->frames_1024_to_max_octets += mib_read(mp, 0x34);
1308 	p->good_octets_sent += mib_read(mp, 0x38);
1309 	p->good_frames_sent += mib_read(mp, 0x40);
1310 	p->excessive_collision += mib_read(mp, 0x44);
1311 	p->multicast_frames_sent += mib_read(mp, 0x48);
1312 	p->broadcast_frames_sent += mib_read(mp, 0x4c);
1313 	p->unrec_mac_control_received += mib_read(mp, 0x50);
1314 	p->fc_sent += mib_read(mp, 0x54);
1315 	p->good_fc_received += mib_read(mp, 0x58);
1316 	p->bad_fc_received += mib_read(mp, 0x5c);
1317 	p->undersize_received += mib_read(mp, 0x60);
1318 	p->fragments_received += mib_read(mp, 0x64);
1319 	p->oversize_received += mib_read(mp, 0x68);
1320 	p->jabber_received += mib_read(mp, 0x6c);
1321 	p->mac_receive_error += mib_read(mp, 0x70);
1322 	p->bad_crc_event += mib_read(mp, 0x74);
1323 	p->collision += mib_read(mp, 0x78);
1324 	p->late_collision += mib_read(mp, 0x7c);
1325 	/* Non MIB hardware counters */
1326 	p->rx_discard += rdlp(mp, RX_DISCARD_FRAME_CNT);
1327 	p->rx_overrun += rdlp(mp, RX_OVERRUN_FRAME_CNT);
1328 	spin_unlock_bh(&mp->mib_counters_lock);
1329 }
1330 
1331 static void mib_counters_timer_wrapper(struct timer_list *t)
1332 {
1333 	struct mv643xx_eth_private *mp = from_timer(mp, t, mib_counters_timer);
1334 	mib_counters_update(mp);
1335 	mod_timer(&mp->mib_counters_timer, jiffies + 30 * HZ);
1336 }
1337 
1338 
1339 /* interrupt coalescing *****************************************************/
1340 /*
1341  * Hardware coalescing parameters are set in units of 64 t_clk
1342  * cycles.  I.e.:
1343  *
1344  *	coal_delay_in_usec = 64000000 * register_value / t_clk_rate
1345  *
1346  *	register_value = coal_delay_in_usec * t_clk_rate / 64000000
1347  *
1348  * In the ->set*() methods, we round the computed register value
1349  * to the nearest integer.
1350  */
1351 static unsigned int get_rx_coal(struct mv643xx_eth_private *mp)
1352 {
1353 	u32 val = rdlp(mp, SDMA_CONFIG);
1354 	u64 temp;
1355 
1356 	if (mp->shared->extended_rx_coal_limit)
1357 		temp = ((val & 0x02000000) >> 10) | ((val & 0x003fff80) >> 7);
1358 	else
1359 		temp = (val & 0x003fff00) >> 8;
1360 
1361 	temp *= 64000000;
1362 	temp += mp->t_clk / 2;
1363 	do_div(temp, mp->t_clk);
1364 
1365 	return (unsigned int)temp;
1366 }
1367 
1368 static void set_rx_coal(struct mv643xx_eth_private *mp, unsigned int usec)
1369 {
1370 	u64 temp;
1371 	u32 val;
1372 
1373 	temp = (u64)usec * mp->t_clk;
1374 	temp += 31999999;
1375 	do_div(temp, 64000000);
1376 
1377 	val = rdlp(mp, SDMA_CONFIG);
1378 	if (mp->shared->extended_rx_coal_limit) {
1379 		if (temp > 0xffff)
1380 			temp = 0xffff;
1381 		val &= ~0x023fff80;
1382 		val |= (temp & 0x8000) << 10;
1383 		val |= (temp & 0x7fff) << 7;
1384 	} else {
1385 		if (temp > 0x3fff)
1386 			temp = 0x3fff;
1387 		val &= ~0x003fff00;
1388 		val |= (temp & 0x3fff) << 8;
1389 	}
1390 	wrlp(mp, SDMA_CONFIG, val);
1391 }
1392 
1393 static unsigned int get_tx_coal(struct mv643xx_eth_private *mp)
1394 {
1395 	u64 temp;
1396 
1397 	temp = (rdlp(mp, TX_FIFO_URGENT_THRESHOLD) & 0x3fff0) >> 4;
1398 	temp *= 64000000;
1399 	temp += mp->t_clk / 2;
1400 	do_div(temp, mp->t_clk);
1401 
1402 	return (unsigned int)temp;
1403 }
1404 
1405 static void set_tx_coal(struct mv643xx_eth_private *mp, unsigned int usec)
1406 {
1407 	u64 temp;
1408 
1409 	temp = (u64)usec * mp->t_clk;
1410 	temp += 31999999;
1411 	do_div(temp, 64000000);
1412 
1413 	if (temp > 0x3fff)
1414 		temp = 0x3fff;
1415 
1416 	wrlp(mp, TX_FIFO_URGENT_THRESHOLD, temp << 4);
1417 }
1418 
1419 
1420 /* ethtool ******************************************************************/
1421 struct mv643xx_eth_stats {
1422 	char stat_string[ETH_GSTRING_LEN];
1423 	int sizeof_stat;
1424 	int netdev_off;
1425 	int mp_off;
1426 };
1427 
1428 #define SSTAT(m)						\
1429 	{ #m, sizeof_field(struct net_device_stats, m),		\
1430 	  offsetof(struct net_device, stats.m), -1 }
1431 
1432 #define MIBSTAT(m)						\
1433 	{ #m, sizeof_field(struct mib_counters, m),		\
1434 	  -1, offsetof(struct mv643xx_eth_private, mib_counters.m) }
1435 
1436 static const struct mv643xx_eth_stats mv643xx_eth_stats[] = {
1437 	SSTAT(rx_packets),
1438 	SSTAT(tx_packets),
1439 	SSTAT(rx_bytes),
1440 	SSTAT(tx_bytes),
1441 	SSTAT(rx_errors),
1442 	SSTAT(tx_errors),
1443 	SSTAT(rx_dropped),
1444 	SSTAT(tx_dropped),
1445 	MIBSTAT(good_octets_received),
1446 	MIBSTAT(bad_octets_received),
1447 	MIBSTAT(internal_mac_transmit_err),
1448 	MIBSTAT(good_frames_received),
1449 	MIBSTAT(bad_frames_received),
1450 	MIBSTAT(broadcast_frames_received),
1451 	MIBSTAT(multicast_frames_received),
1452 	MIBSTAT(frames_64_octets),
1453 	MIBSTAT(frames_65_to_127_octets),
1454 	MIBSTAT(frames_128_to_255_octets),
1455 	MIBSTAT(frames_256_to_511_octets),
1456 	MIBSTAT(frames_512_to_1023_octets),
1457 	MIBSTAT(frames_1024_to_max_octets),
1458 	MIBSTAT(good_octets_sent),
1459 	MIBSTAT(good_frames_sent),
1460 	MIBSTAT(excessive_collision),
1461 	MIBSTAT(multicast_frames_sent),
1462 	MIBSTAT(broadcast_frames_sent),
1463 	MIBSTAT(unrec_mac_control_received),
1464 	MIBSTAT(fc_sent),
1465 	MIBSTAT(good_fc_received),
1466 	MIBSTAT(bad_fc_received),
1467 	MIBSTAT(undersize_received),
1468 	MIBSTAT(fragments_received),
1469 	MIBSTAT(oversize_received),
1470 	MIBSTAT(jabber_received),
1471 	MIBSTAT(mac_receive_error),
1472 	MIBSTAT(bad_crc_event),
1473 	MIBSTAT(collision),
1474 	MIBSTAT(late_collision),
1475 	MIBSTAT(rx_discard),
1476 	MIBSTAT(rx_overrun),
1477 };
1478 
1479 static int
1480 mv643xx_eth_get_link_ksettings_phy(struct mv643xx_eth_private *mp,
1481 				   struct ethtool_link_ksettings *cmd)
1482 {
1483 	struct net_device *dev = mp->dev;
1484 
1485 	phy_ethtool_ksettings_get(dev->phydev, cmd);
1486 
1487 	/*
1488 	 * The MAC does not support 1000baseT_Half.
1489 	 */
1490 	linkmode_clear_bit(ETHTOOL_LINK_MODE_1000baseT_Half_BIT,
1491 			   cmd->link_modes.supported);
1492 	linkmode_clear_bit(ETHTOOL_LINK_MODE_1000baseT_Half_BIT,
1493 			   cmd->link_modes.advertising);
1494 
1495 	return 0;
1496 }
1497 
1498 static int
1499 mv643xx_eth_get_link_ksettings_phyless(struct mv643xx_eth_private *mp,
1500 				       struct ethtool_link_ksettings *cmd)
1501 {
1502 	u32 port_status;
1503 	u32 supported, advertising;
1504 
1505 	port_status = rdlp(mp, PORT_STATUS);
1506 
1507 	supported = SUPPORTED_MII;
1508 	advertising = ADVERTISED_MII;
1509 	switch (port_status & PORT_SPEED_MASK) {
1510 	case PORT_SPEED_10:
1511 		cmd->base.speed = SPEED_10;
1512 		break;
1513 	case PORT_SPEED_100:
1514 		cmd->base.speed = SPEED_100;
1515 		break;
1516 	case PORT_SPEED_1000:
1517 		cmd->base.speed = SPEED_1000;
1518 		break;
1519 	default:
1520 		cmd->base.speed = -1;
1521 		break;
1522 	}
1523 	cmd->base.duplex = (port_status & FULL_DUPLEX) ?
1524 		DUPLEX_FULL : DUPLEX_HALF;
1525 	cmd->base.port = PORT_MII;
1526 	cmd->base.phy_address = 0;
1527 	cmd->base.autoneg = AUTONEG_DISABLE;
1528 
1529 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
1530 						supported);
1531 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
1532 						advertising);
1533 
1534 	return 0;
1535 }
1536 
1537 static void
1538 mv643xx_eth_get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1539 {
1540 	wol->supported = 0;
1541 	wol->wolopts = 0;
1542 	if (dev->phydev)
1543 		phy_ethtool_get_wol(dev->phydev, wol);
1544 }
1545 
1546 static int
1547 mv643xx_eth_set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1548 {
1549 	int err;
1550 
1551 	if (!dev->phydev)
1552 		return -EOPNOTSUPP;
1553 
1554 	err = phy_ethtool_set_wol(dev->phydev, wol);
1555 	/* Given that mv643xx_eth works without the marvell-specific PHY driver,
1556 	 * this debugging hint is useful to have.
1557 	 */
1558 	if (err == -EOPNOTSUPP)
1559 		netdev_info(dev, "The PHY does not support set_wol, was CONFIG_MARVELL_PHY enabled?\n");
1560 	return err;
1561 }
1562 
1563 static int
1564 mv643xx_eth_get_link_ksettings(struct net_device *dev,
1565 			       struct ethtool_link_ksettings *cmd)
1566 {
1567 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1568 
1569 	if (dev->phydev)
1570 		return mv643xx_eth_get_link_ksettings_phy(mp, cmd);
1571 	else
1572 		return mv643xx_eth_get_link_ksettings_phyless(mp, cmd);
1573 }
1574 
1575 static int
1576 mv643xx_eth_set_link_ksettings(struct net_device *dev,
1577 			       const struct ethtool_link_ksettings *cmd)
1578 {
1579 	struct ethtool_link_ksettings c = *cmd;
1580 	u32 advertising;
1581 	int ret;
1582 
1583 	if (!dev->phydev)
1584 		return -EINVAL;
1585 
1586 	/*
1587 	 * The MAC does not support 1000baseT_Half.
1588 	 */
1589 	ethtool_convert_link_mode_to_legacy_u32(&advertising,
1590 						c.link_modes.advertising);
1591 	advertising &= ~ADVERTISED_1000baseT_Half;
1592 	ethtool_convert_legacy_u32_to_link_mode(c.link_modes.advertising,
1593 						advertising);
1594 
1595 	ret = phy_ethtool_ksettings_set(dev->phydev, &c);
1596 	if (!ret)
1597 		mv643xx_eth_adjust_link(dev);
1598 	return ret;
1599 }
1600 
1601 static void mv643xx_eth_get_drvinfo(struct net_device *dev,
1602 				    struct ethtool_drvinfo *drvinfo)
1603 {
1604 	strlcpy(drvinfo->driver, mv643xx_eth_driver_name,
1605 		sizeof(drvinfo->driver));
1606 	strlcpy(drvinfo->version, mv643xx_eth_driver_version,
1607 		sizeof(drvinfo->version));
1608 	strlcpy(drvinfo->fw_version, "N/A", sizeof(drvinfo->fw_version));
1609 	strlcpy(drvinfo->bus_info, "platform", sizeof(drvinfo->bus_info));
1610 }
1611 
1612 static int
1613 mv643xx_eth_get_coalesce(struct net_device *dev, struct ethtool_coalesce *ec)
1614 {
1615 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1616 
1617 	ec->rx_coalesce_usecs = get_rx_coal(mp);
1618 	ec->tx_coalesce_usecs = get_tx_coal(mp);
1619 
1620 	return 0;
1621 }
1622 
1623 static int
1624 mv643xx_eth_set_coalesce(struct net_device *dev, struct ethtool_coalesce *ec)
1625 {
1626 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1627 
1628 	set_rx_coal(mp, ec->rx_coalesce_usecs);
1629 	set_tx_coal(mp, ec->tx_coalesce_usecs);
1630 
1631 	return 0;
1632 }
1633 
1634 static void
1635 mv643xx_eth_get_ringparam(struct net_device *dev, struct ethtool_ringparam *er)
1636 {
1637 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1638 
1639 	er->rx_max_pending = 4096;
1640 	er->tx_max_pending = 4096;
1641 
1642 	er->rx_pending = mp->rx_ring_size;
1643 	er->tx_pending = mp->tx_ring_size;
1644 }
1645 
1646 static int
1647 mv643xx_eth_set_ringparam(struct net_device *dev, struct ethtool_ringparam *er)
1648 {
1649 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1650 
1651 	if (er->rx_mini_pending || er->rx_jumbo_pending)
1652 		return -EINVAL;
1653 
1654 	mp->rx_ring_size = er->rx_pending < 4096 ? er->rx_pending : 4096;
1655 	mp->tx_ring_size = clamp_t(unsigned int, er->tx_pending,
1656 				   MV643XX_MAX_SKB_DESCS * 2, 4096);
1657 	if (mp->tx_ring_size != er->tx_pending)
1658 		netdev_warn(dev, "TX queue size set to %u (requested %u)\n",
1659 			    mp->tx_ring_size, er->tx_pending);
1660 
1661 	if (netif_running(dev)) {
1662 		mv643xx_eth_stop(dev);
1663 		if (mv643xx_eth_open(dev)) {
1664 			netdev_err(dev,
1665 				   "fatal error on re-opening device after ring param change\n");
1666 			return -ENOMEM;
1667 		}
1668 	}
1669 
1670 	return 0;
1671 }
1672 
1673 
1674 static int
1675 mv643xx_eth_set_features(struct net_device *dev, netdev_features_t features)
1676 {
1677 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1678 	bool rx_csum = features & NETIF_F_RXCSUM;
1679 
1680 	wrlp(mp, PORT_CONFIG, rx_csum ? 0x02000000 : 0x00000000);
1681 
1682 	return 0;
1683 }
1684 
1685 static void mv643xx_eth_get_strings(struct net_device *dev,
1686 				    uint32_t stringset, uint8_t *data)
1687 {
1688 	int i;
1689 
1690 	if (stringset == ETH_SS_STATS) {
1691 		for (i = 0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
1692 			memcpy(data + i * ETH_GSTRING_LEN,
1693 				mv643xx_eth_stats[i].stat_string,
1694 				ETH_GSTRING_LEN);
1695 		}
1696 	}
1697 }
1698 
1699 static void mv643xx_eth_get_ethtool_stats(struct net_device *dev,
1700 					  struct ethtool_stats *stats,
1701 					  uint64_t *data)
1702 {
1703 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1704 	int i;
1705 
1706 	mv643xx_eth_get_stats(dev);
1707 	mib_counters_update(mp);
1708 
1709 	for (i = 0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
1710 		const struct mv643xx_eth_stats *stat;
1711 		void *p;
1712 
1713 		stat = mv643xx_eth_stats + i;
1714 
1715 		if (stat->netdev_off >= 0)
1716 			p = ((void *)mp->dev) + stat->netdev_off;
1717 		else
1718 			p = ((void *)mp) + stat->mp_off;
1719 
1720 		data[i] = (stat->sizeof_stat == 8) ?
1721 				*(uint64_t *)p : *(uint32_t *)p;
1722 	}
1723 }
1724 
1725 static int mv643xx_eth_get_sset_count(struct net_device *dev, int sset)
1726 {
1727 	if (sset == ETH_SS_STATS)
1728 		return ARRAY_SIZE(mv643xx_eth_stats);
1729 
1730 	return -EOPNOTSUPP;
1731 }
1732 
1733 static const struct ethtool_ops mv643xx_eth_ethtool_ops = {
1734 	.supported_coalesce_params = ETHTOOL_COALESCE_USECS,
1735 	.get_drvinfo		= mv643xx_eth_get_drvinfo,
1736 	.nway_reset		= phy_ethtool_nway_reset,
1737 	.get_link		= ethtool_op_get_link,
1738 	.get_coalesce		= mv643xx_eth_get_coalesce,
1739 	.set_coalesce		= mv643xx_eth_set_coalesce,
1740 	.get_ringparam		= mv643xx_eth_get_ringparam,
1741 	.set_ringparam		= mv643xx_eth_set_ringparam,
1742 	.get_strings		= mv643xx_eth_get_strings,
1743 	.get_ethtool_stats	= mv643xx_eth_get_ethtool_stats,
1744 	.get_sset_count		= mv643xx_eth_get_sset_count,
1745 	.get_ts_info		= ethtool_op_get_ts_info,
1746 	.get_wol                = mv643xx_eth_get_wol,
1747 	.set_wol                = mv643xx_eth_set_wol,
1748 	.get_link_ksettings	= mv643xx_eth_get_link_ksettings,
1749 	.set_link_ksettings	= mv643xx_eth_set_link_ksettings,
1750 };
1751 
1752 
1753 /* address handling *********************************************************/
1754 static void uc_addr_get(struct mv643xx_eth_private *mp, unsigned char *addr)
1755 {
1756 	unsigned int mac_h = rdlp(mp, MAC_ADDR_HIGH);
1757 	unsigned int mac_l = rdlp(mp, MAC_ADDR_LOW);
1758 
1759 	addr[0] = (mac_h >> 24) & 0xff;
1760 	addr[1] = (mac_h >> 16) & 0xff;
1761 	addr[2] = (mac_h >> 8) & 0xff;
1762 	addr[3] = mac_h & 0xff;
1763 	addr[4] = (mac_l >> 8) & 0xff;
1764 	addr[5] = mac_l & 0xff;
1765 }
1766 
1767 static void uc_addr_set(struct mv643xx_eth_private *mp, unsigned char *addr)
1768 {
1769 	wrlp(mp, MAC_ADDR_HIGH,
1770 		(addr[0] << 24) | (addr[1] << 16) | (addr[2] << 8) | addr[3]);
1771 	wrlp(mp, MAC_ADDR_LOW, (addr[4] << 8) | addr[5]);
1772 }
1773 
1774 static u32 uc_addr_filter_mask(struct net_device *dev)
1775 {
1776 	struct netdev_hw_addr *ha;
1777 	u32 nibbles;
1778 
1779 	if (dev->flags & IFF_PROMISC)
1780 		return 0;
1781 
1782 	nibbles = 1 << (dev->dev_addr[5] & 0x0f);
1783 	netdev_for_each_uc_addr(ha, dev) {
1784 		if (memcmp(dev->dev_addr, ha->addr, 5))
1785 			return 0;
1786 		if ((dev->dev_addr[5] ^ ha->addr[5]) & 0xf0)
1787 			return 0;
1788 
1789 		nibbles |= 1 << (ha->addr[5] & 0x0f);
1790 	}
1791 
1792 	return nibbles;
1793 }
1794 
1795 static void mv643xx_eth_program_unicast_filter(struct net_device *dev)
1796 {
1797 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1798 	u32 port_config;
1799 	u32 nibbles;
1800 	int i;
1801 
1802 	uc_addr_set(mp, dev->dev_addr);
1803 
1804 	port_config = rdlp(mp, PORT_CONFIG) & ~UNICAST_PROMISCUOUS_MODE;
1805 
1806 	nibbles = uc_addr_filter_mask(dev);
1807 	if (!nibbles) {
1808 		port_config |= UNICAST_PROMISCUOUS_MODE;
1809 		nibbles = 0xffff;
1810 	}
1811 
1812 	for (i = 0; i < 16; i += 4) {
1813 		int off = UNICAST_TABLE(mp->port_num) + i;
1814 		u32 v;
1815 
1816 		v = 0;
1817 		if (nibbles & 1)
1818 			v |= 0x00000001;
1819 		if (nibbles & 2)
1820 			v |= 0x00000100;
1821 		if (nibbles & 4)
1822 			v |= 0x00010000;
1823 		if (nibbles & 8)
1824 			v |= 0x01000000;
1825 		nibbles >>= 4;
1826 
1827 		wrl(mp, off, v);
1828 	}
1829 
1830 	wrlp(mp, PORT_CONFIG, port_config);
1831 }
1832 
1833 static int addr_crc(unsigned char *addr)
1834 {
1835 	int crc = 0;
1836 	int i;
1837 
1838 	for (i = 0; i < 6; i++) {
1839 		int j;
1840 
1841 		crc = (crc ^ addr[i]) << 8;
1842 		for (j = 7; j >= 0; j--) {
1843 			if (crc & (0x100 << j))
1844 				crc ^= 0x107 << j;
1845 		}
1846 	}
1847 
1848 	return crc;
1849 }
1850 
1851 static void mv643xx_eth_program_multicast_filter(struct net_device *dev)
1852 {
1853 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1854 	u32 *mc_spec;
1855 	u32 *mc_other;
1856 	struct netdev_hw_addr *ha;
1857 	int i;
1858 
1859 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI))
1860 		goto promiscuous;
1861 
1862 	/* Allocate both mc_spec and mc_other tables */
1863 	mc_spec = kcalloc(128, sizeof(u32), GFP_ATOMIC);
1864 	if (!mc_spec)
1865 		goto promiscuous;
1866 	mc_other = &mc_spec[64];
1867 
1868 	netdev_for_each_mc_addr(ha, dev) {
1869 		u8 *a = ha->addr;
1870 		u32 *table;
1871 		u8 entry;
1872 
1873 		if (memcmp(a, "\x01\x00\x5e\x00\x00", 5) == 0) {
1874 			table = mc_spec;
1875 			entry = a[5];
1876 		} else {
1877 			table = mc_other;
1878 			entry = addr_crc(a);
1879 		}
1880 
1881 		table[entry >> 2] |= 1 << (8 * (entry & 3));
1882 	}
1883 
1884 	for (i = 0; i < 64; i++) {
1885 		wrl(mp, SPECIAL_MCAST_TABLE(mp->port_num) + i * sizeof(u32),
1886 		    mc_spec[i]);
1887 		wrl(mp, OTHER_MCAST_TABLE(mp->port_num) + i * sizeof(u32),
1888 		    mc_other[i]);
1889 	}
1890 
1891 	kfree(mc_spec);
1892 	return;
1893 
1894 promiscuous:
1895 	for (i = 0; i < 64; i++) {
1896 		wrl(mp, SPECIAL_MCAST_TABLE(mp->port_num) + i * sizeof(u32),
1897 		    0x01010101u);
1898 		wrl(mp, OTHER_MCAST_TABLE(mp->port_num) + i * sizeof(u32),
1899 		    0x01010101u);
1900 	}
1901 }
1902 
1903 static void mv643xx_eth_set_rx_mode(struct net_device *dev)
1904 {
1905 	mv643xx_eth_program_unicast_filter(dev);
1906 	mv643xx_eth_program_multicast_filter(dev);
1907 }
1908 
1909 static int mv643xx_eth_set_mac_address(struct net_device *dev, void *addr)
1910 {
1911 	struct sockaddr *sa = addr;
1912 
1913 	if (!is_valid_ether_addr(sa->sa_data))
1914 		return -EADDRNOTAVAIL;
1915 
1916 	memcpy(dev->dev_addr, sa->sa_data, ETH_ALEN);
1917 
1918 	netif_addr_lock_bh(dev);
1919 	mv643xx_eth_program_unicast_filter(dev);
1920 	netif_addr_unlock_bh(dev);
1921 
1922 	return 0;
1923 }
1924 
1925 
1926 /* rx/tx queue initialisation ***********************************************/
1927 static int rxq_init(struct mv643xx_eth_private *mp, int index)
1928 {
1929 	struct rx_queue *rxq = mp->rxq + index;
1930 	struct rx_desc *rx_desc;
1931 	int size;
1932 	int i;
1933 
1934 	rxq->index = index;
1935 
1936 	rxq->rx_ring_size = mp->rx_ring_size;
1937 
1938 	rxq->rx_desc_count = 0;
1939 	rxq->rx_curr_desc = 0;
1940 	rxq->rx_used_desc = 0;
1941 
1942 	size = rxq->rx_ring_size * sizeof(struct rx_desc);
1943 
1944 	if (index == 0 && size <= mp->rx_desc_sram_size) {
1945 		rxq->rx_desc_area = ioremap(mp->rx_desc_sram_addr,
1946 						mp->rx_desc_sram_size);
1947 		rxq->rx_desc_dma = mp->rx_desc_sram_addr;
1948 	} else {
1949 		rxq->rx_desc_area = dma_alloc_coherent(mp->dev->dev.parent,
1950 						       size, &rxq->rx_desc_dma,
1951 						       GFP_KERNEL);
1952 	}
1953 
1954 	if (rxq->rx_desc_area == NULL) {
1955 		netdev_err(mp->dev,
1956 			   "can't allocate rx ring (%d bytes)\n", size);
1957 		goto out;
1958 	}
1959 	memset(rxq->rx_desc_area, 0, size);
1960 
1961 	rxq->rx_desc_area_size = size;
1962 	rxq->rx_skb = kcalloc(rxq->rx_ring_size, sizeof(*rxq->rx_skb),
1963 				    GFP_KERNEL);
1964 	if (rxq->rx_skb == NULL)
1965 		goto out_free;
1966 
1967 	rx_desc = rxq->rx_desc_area;
1968 	for (i = 0; i < rxq->rx_ring_size; i++) {
1969 		int nexti;
1970 
1971 		nexti = i + 1;
1972 		if (nexti == rxq->rx_ring_size)
1973 			nexti = 0;
1974 
1975 		rx_desc[i].next_desc_ptr = rxq->rx_desc_dma +
1976 					nexti * sizeof(struct rx_desc);
1977 	}
1978 
1979 	return 0;
1980 
1981 
1982 out_free:
1983 	if (index == 0 && size <= mp->rx_desc_sram_size)
1984 		iounmap(rxq->rx_desc_area);
1985 	else
1986 		dma_free_coherent(mp->dev->dev.parent, size,
1987 				  rxq->rx_desc_area,
1988 				  rxq->rx_desc_dma);
1989 
1990 out:
1991 	return -ENOMEM;
1992 }
1993 
1994 static void rxq_deinit(struct rx_queue *rxq)
1995 {
1996 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
1997 	int i;
1998 
1999 	rxq_disable(rxq);
2000 
2001 	for (i = 0; i < rxq->rx_ring_size; i++) {
2002 		if (rxq->rx_skb[i]) {
2003 			dev_consume_skb_any(rxq->rx_skb[i]);
2004 			rxq->rx_desc_count--;
2005 		}
2006 	}
2007 
2008 	if (rxq->rx_desc_count) {
2009 		netdev_err(mp->dev, "error freeing rx ring -- %d skbs stuck\n",
2010 			   rxq->rx_desc_count);
2011 	}
2012 
2013 	if (rxq->index == 0 &&
2014 	    rxq->rx_desc_area_size <= mp->rx_desc_sram_size)
2015 		iounmap(rxq->rx_desc_area);
2016 	else
2017 		dma_free_coherent(mp->dev->dev.parent, rxq->rx_desc_area_size,
2018 				  rxq->rx_desc_area, rxq->rx_desc_dma);
2019 
2020 	kfree(rxq->rx_skb);
2021 }
2022 
2023 static int txq_init(struct mv643xx_eth_private *mp, int index)
2024 {
2025 	struct tx_queue *txq = mp->txq + index;
2026 	struct tx_desc *tx_desc;
2027 	int size;
2028 	int ret;
2029 	int i;
2030 
2031 	txq->index = index;
2032 
2033 	txq->tx_ring_size = mp->tx_ring_size;
2034 
2035 	/* A queue must always have room for at least one skb.
2036 	 * Therefore, stop the queue when the free entries reaches
2037 	 * the maximum number of descriptors per skb.
2038 	 */
2039 	txq->tx_stop_threshold = txq->tx_ring_size - MV643XX_MAX_SKB_DESCS;
2040 	txq->tx_wake_threshold = txq->tx_stop_threshold / 2;
2041 
2042 	txq->tx_desc_count = 0;
2043 	txq->tx_curr_desc = 0;
2044 	txq->tx_used_desc = 0;
2045 
2046 	size = txq->tx_ring_size * sizeof(struct tx_desc);
2047 
2048 	if (index == 0 && size <= mp->tx_desc_sram_size) {
2049 		txq->tx_desc_area = ioremap(mp->tx_desc_sram_addr,
2050 						mp->tx_desc_sram_size);
2051 		txq->tx_desc_dma = mp->tx_desc_sram_addr;
2052 	} else {
2053 		txq->tx_desc_area = dma_alloc_coherent(mp->dev->dev.parent,
2054 						       size, &txq->tx_desc_dma,
2055 						       GFP_KERNEL);
2056 	}
2057 
2058 	if (txq->tx_desc_area == NULL) {
2059 		netdev_err(mp->dev,
2060 			   "can't allocate tx ring (%d bytes)\n", size);
2061 		return -ENOMEM;
2062 	}
2063 	memset(txq->tx_desc_area, 0, size);
2064 
2065 	txq->tx_desc_area_size = size;
2066 
2067 	tx_desc = txq->tx_desc_area;
2068 	for (i = 0; i < txq->tx_ring_size; i++) {
2069 		struct tx_desc *txd = tx_desc + i;
2070 		int nexti;
2071 
2072 		nexti = i + 1;
2073 		if (nexti == txq->tx_ring_size)
2074 			nexti = 0;
2075 
2076 		txd->cmd_sts = 0;
2077 		txd->next_desc_ptr = txq->tx_desc_dma +
2078 					nexti * sizeof(struct tx_desc);
2079 	}
2080 
2081 	txq->tx_desc_mapping = kcalloc(txq->tx_ring_size, sizeof(char),
2082 				       GFP_KERNEL);
2083 	if (!txq->tx_desc_mapping) {
2084 		ret = -ENOMEM;
2085 		goto err_free_desc_area;
2086 	}
2087 
2088 	/* Allocate DMA buffers for TSO MAC/IP/TCP headers */
2089 	txq->tso_hdrs = dma_alloc_coherent(mp->dev->dev.parent,
2090 					   txq->tx_ring_size * TSO_HEADER_SIZE,
2091 					   &txq->tso_hdrs_dma, GFP_KERNEL);
2092 	if (txq->tso_hdrs == NULL) {
2093 		ret = -ENOMEM;
2094 		goto err_free_desc_mapping;
2095 	}
2096 	skb_queue_head_init(&txq->tx_skb);
2097 
2098 	return 0;
2099 
2100 err_free_desc_mapping:
2101 	kfree(txq->tx_desc_mapping);
2102 err_free_desc_area:
2103 	if (index == 0 && size <= mp->tx_desc_sram_size)
2104 		iounmap(txq->tx_desc_area);
2105 	else
2106 		dma_free_coherent(mp->dev->dev.parent, txq->tx_desc_area_size,
2107 				  txq->tx_desc_area, txq->tx_desc_dma);
2108 	return ret;
2109 }
2110 
2111 static void txq_deinit(struct tx_queue *txq)
2112 {
2113 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
2114 
2115 	txq_disable(txq);
2116 	txq_reclaim(txq, txq->tx_ring_size, 1);
2117 
2118 	BUG_ON(txq->tx_used_desc != txq->tx_curr_desc);
2119 
2120 	if (txq->index == 0 &&
2121 	    txq->tx_desc_area_size <= mp->tx_desc_sram_size)
2122 		iounmap(txq->tx_desc_area);
2123 	else
2124 		dma_free_coherent(mp->dev->dev.parent, txq->tx_desc_area_size,
2125 				  txq->tx_desc_area, txq->tx_desc_dma);
2126 	kfree(txq->tx_desc_mapping);
2127 
2128 	if (txq->tso_hdrs)
2129 		dma_free_coherent(mp->dev->dev.parent,
2130 				  txq->tx_ring_size * TSO_HEADER_SIZE,
2131 				  txq->tso_hdrs, txq->tso_hdrs_dma);
2132 }
2133 
2134 
2135 /* netdev ops and related ***************************************************/
2136 static int mv643xx_eth_collect_events(struct mv643xx_eth_private *mp)
2137 {
2138 	u32 int_cause;
2139 	u32 int_cause_ext;
2140 
2141 	int_cause = rdlp(mp, INT_CAUSE) & mp->int_mask;
2142 	if (int_cause == 0)
2143 		return 0;
2144 
2145 	int_cause_ext = 0;
2146 	if (int_cause & INT_EXT) {
2147 		int_cause &= ~INT_EXT;
2148 		int_cause_ext = rdlp(mp, INT_CAUSE_EXT);
2149 	}
2150 
2151 	if (int_cause) {
2152 		wrlp(mp, INT_CAUSE, ~int_cause);
2153 		mp->work_tx_end |= ((int_cause & INT_TX_END) >> 19) &
2154 				~(rdlp(mp, TXQ_COMMAND) & 0xff);
2155 		mp->work_rx |= (int_cause & INT_RX) >> 2;
2156 	}
2157 
2158 	int_cause_ext &= INT_EXT_LINK_PHY | INT_EXT_TX;
2159 	if (int_cause_ext) {
2160 		wrlp(mp, INT_CAUSE_EXT, ~int_cause_ext);
2161 		if (int_cause_ext & INT_EXT_LINK_PHY)
2162 			mp->work_link = 1;
2163 		mp->work_tx |= int_cause_ext & INT_EXT_TX;
2164 	}
2165 
2166 	return 1;
2167 }
2168 
2169 static irqreturn_t mv643xx_eth_irq(int irq, void *dev_id)
2170 {
2171 	struct net_device *dev = (struct net_device *)dev_id;
2172 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2173 
2174 	if (unlikely(!mv643xx_eth_collect_events(mp)))
2175 		return IRQ_NONE;
2176 
2177 	wrlp(mp, INT_MASK, 0);
2178 	napi_schedule(&mp->napi);
2179 
2180 	return IRQ_HANDLED;
2181 }
2182 
2183 static void handle_link_event(struct mv643xx_eth_private *mp)
2184 {
2185 	struct net_device *dev = mp->dev;
2186 	u32 port_status;
2187 	int speed;
2188 	int duplex;
2189 	int fc;
2190 
2191 	port_status = rdlp(mp, PORT_STATUS);
2192 	if (!(port_status & LINK_UP)) {
2193 		if (netif_carrier_ok(dev)) {
2194 			int i;
2195 
2196 			netdev_info(dev, "link down\n");
2197 
2198 			netif_carrier_off(dev);
2199 
2200 			for (i = 0; i < mp->txq_count; i++) {
2201 				struct tx_queue *txq = mp->txq + i;
2202 
2203 				txq_reclaim(txq, txq->tx_ring_size, 1);
2204 				txq_reset_hw_ptr(txq);
2205 			}
2206 		}
2207 		return;
2208 	}
2209 
2210 	switch (port_status & PORT_SPEED_MASK) {
2211 	case PORT_SPEED_10:
2212 		speed = 10;
2213 		break;
2214 	case PORT_SPEED_100:
2215 		speed = 100;
2216 		break;
2217 	case PORT_SPEED_1000:
2218 		speed = 1000;
2219 		break;
2220 	default:
2221 		speed = -1;
2222 		break;
2223 	}
2224 	duplex = (port_status & FULL_DUPLEX) ? 1 : 0;
2225 	fc = (port_status & FLOW_CONTROL_ENABLED) ? 1 : 0;
2226 
2227 	netdev_info(dev, "link up, %d Mb/s, %s duplex, flow control %sabled\n",
2228 		    speed, duplex ? "full" : "half", fc ? "en" : "dis");
2229 
2230 	if (!netif_carrier_ok(dev))
2231 		netif_carrier_on(dev);
2232 }
2233 
2234 static int mv643xx_eth_poll(struct napi_struct *napi, int budget)
2235 {
2236 	struct mv643xx_eth_private *mp;
2237 	int work_done;
2238 
2239 	mp = container_of(napi, struct mv643xx_eth_private, napi);
2240 
2241 	if (unlikely(mp->oom)) {
2242 		mp->oom = 0;
2243 		del_timer(&mp->rx_oom);
2244 	}
2245 
2246 	work_done = 0;
2247 	while (work_done < budget) {
2248 		u8 queue_mask;
2249 		int queue;
2250 		int work_tbd;
2251 
2252 		if (mp->work_link) {
2253 			mp->work_link = 0;
2254 			handle_link_event(mp);
2255 			work_done++;
2256 			continue;
2257 		}
2258 
2259 		queue_mask = mp->work_tx | mp->work_tx_end | mp->work_rx;
2260 		if (likely(!mp->oom))
2261 			queue_mask |= mp->work_rx_refill;
2262 
2263 		if (!queue_mask) {
2264 			if (mv643xx_eth_collect_events(mp))
2265 				continue;
2266 			break;
2267 		}
2268 
2269 		queue = fls(queue_mask) - 1;
2270 		queue_mask = 1 << queue;
2271 
2272 		work_tbd = budget - work_done;
2273 		if (work_tbd > 16)
2274 			work_tbd = 16;
2275 
2276 		if (mp->work_tx_end & queue_mask) {
2277 			txq_kick(mp->txq + queue);
2278 		} else if (mp->work_tx & queue_mask) {
2279 			work_done += txq_reclaim(mp->txq + queue, work_tbd, 0);
2280 			txq_maybe_wake(mp->txq + queue);
2281 		} else if (mp->work_rx & queue_mask) {
2282 			work_done += rxq_process(mp->rxq + queue, work_tbd);
2283 		} else if (!mp->oom && (mp->work_rx_refill & queue_mask)) {
2284 			work_done += rxq_refill(mp->rxq + queue, work_tbd);
2285 		} else {
2286 			BUG();
2287 		}
2288 	}
2289 
2290 	if (work_done < budget) {
2291 		if (mp->oom)
2292 			mod_timer(&mp->rx_oom, jiffies + (HZ / 10));
2293 		napi_complete_done(napi, work_done);
2294 		wrlp(mp, INT_MASK, mp->int_mask);
2295 	}
2296 
2297 	return work_done;
2298 }
2299 
2300 static inline void oom_timer_wrapper(struct timer_list *t)
2301 {
2302 	struct mv643xx_eth_private *mp = from_timer(mp, t, rx_oom);
2303 
2304 	napi_schedule(&mp->napi);
2305 }
2306 
2307 static void port_start(struct mv643xx_eth_private *mp)
2308 {
2309 	struct net_device *dev = mp->dev;
2310 	u32 pscr;
2311 	int i;
2312 
2313 	/*
2314 	 * Perform PHY reset, if there is a PHY.
2315 	 */
2316 	if (dev->phydev) {
2317 		struct ethtool_link_ksettings cmd;
2318 
2319 		mv643xx_eth_get_link_ksettings(dev, &cmd);
2320 		phy_init_hw(dev->phydev);
2321 		mv643xx_eth_set_link_ksettings(
2322 			dev, (const struct ethtool_link_ksettings *)&cmd);
2323 		phy_start(dev->phydev);
2324 	}
2325 
2326 	/*
2327 	 * Configure basic link parameters.
2328 	 */
2329 	pscr = rdlp(mp, PORT_SERIAL_CONTROL);
2330 
2331 	pscr |= SERIAL_PORT_ENABLE;
2332 	wrlp(mp, PORT_SERIAL_CONTROL, pscr);
2333 
2334 	pscr |= DO_NOT_FORCE_LINK_FAIL;
2335 	if (!dev->phydev)
2336 		pscr |= FORCE_LINK_PASS;
2337 	wrlp(mp, PORT_SERIAL_CONTROL, pscr);
2338 
2339 	/*
2340 	 * Configure TX path and queues.
2341 	 */
2342 	tx_set_rate(mp, 1000000000, 16777216);
2343 	for (i = 0; i < mp->txq_count; i++) {
2344 		struct tx_queue *txq = mp->txq + i;
2345 
2346 		txq_reset_hw_ptr(txq);
2347 		txq_set_rate(txq, 1000000000, 16777216);
2348 		txq_set_fixed_prio_mode(txq);
2349 	}
2350 
2351 	/*
2352 	 * Receive all unmatched unicast, TCP, UDP, BPDU and broadcast
2353 	 * frames to RX queue #0, and include the pseudo-header when
2354 	 * calculating receive checksums.
2355 	 */
2356 	mv643xx_eth_set_features(mp->dev, mp->dev->features);
2357 
2358 	/*
2359 	 * Treat BPDUs as normal multicasts, and disable partition mode.
2360 	 */
2361 	wrlp(mp, PORT_CONFIG_EXT, 0x00000000);
2362 
2363 	/*
2364 	 * Add configured unicast addresses to address filter table.
2365 	 */
2366 	mv643xx_eth_program_unicast_filter(mp->dev);
2367 
2368 	/*
2369 	 * Enable the receive queues.
2370 	 */
2371 	for (i = 0; i < mp->rxq_count; i++) {
2372 		struct rx_queue *rxq = mp->rxq + i;
2373 		u32 addr;
2374 
2375 		addr = (u32)rxq->rx_desc_dma;
2376 		addr += rxq->rx_curr_desc * sizeof(struct rx_desc);
2377 		wrlp(mp, RXQ_CURRENT_DESC_PTR(i), addr);
2378 
2379 		rxq_enable(rxq);
2380 	}
2381 }
2382 
2383 static void mv643xx_eth_recalc_skb_size(struct mv643xx_eth_private *mp)
2384 {
2385 	int skb_size;
2386 
2387 	/*
2388 	 * Reserve 2+14 bytes for an ethernet header (the hardware
2389 	 * automatically prepends 2 bytes of dummy data to each
2390 	 * received packet), 16 bytes for up to four VLAN tags, and
2391 	 * 4 bytes for the trailing FCS -- 36 bytes total.
2392 	 */
2393 	skb_size = mp->dev->mtu + 36;
2394 
2395 	/*
2396 	 * Make sure that the skb size is a multiple of 8 bytes, as
2397 	 * the lower three bits of the receive descriptor's buffer
2398 	 * size field are ignored by the hardware.
2399 	 */
2400 	mp->skb_size = (skb_size + 7) & ~7;
2401 
2402 	/*
2403 	 * If NET_SKB_PAD is smaller than a cache line,
2404 	 * netdev_alloc_skb() will cause skb->data to be misaligned
2405 	 * to a cache line boundary.  If this is the case, include
2406 	 * some extra space to allow re-aligning the data area.
2407 	 */
2408 	mp->skb_size += SKB_DMA_REALIGN;
2409 }
2410 
2411 static int mv643xx_eth_open(struct net_device *dev)
2412 {
2413 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2414 	int err;
2415 	int i;
2416 
2417 	wrlp(mp, INT_CAUSE, 0);
2418 	wrlp(mp, INT_CAUSE_EXT, 0);
2419 	rdlp(mp, INT_CAUSE_EXT);
2420 
2421 	err = request_irq(dev->irq, mv643xx_eth_irq,
2422 			  IRQF_SHARED, dev->name, dev);
2423 	if (err) {
2424 		netdev_err(dev, "can't assign irq\n");
2425 		return -EAGAIN;
2426 	}
2427 
2428 	mv643xx_eth_recalc_skb_size(mp);
2429 
2430 	napi_enable(&mp->napi);
2431 
2432 	mp->int_mask = INT_EXT;
2433 
2434 	for (i = 0; i < mp->rxq_count; i++) {
2435 		err = rxq_init(mp, i);
2436 		if (err) {
2437 			while (--i >= 0)
2438 				rxq_deinit(mp->rxq + i);
2439 			goto out;
2440 		}
2441 
2442 		rxq_refill(mp->rxq + i, INT_MAX);
2443 		mp->int_mask |= INT_RX_0 << i;
2444 	}
2445 
2446 	if (mp->oom) {
2447 		mp->rx_oom.expires = jiffies + (HZ / 10);
2448 		add_timer(&mp->rx_oom);
2449 	}
2450 
2451 	for (i = 0; i < mp->txq_count; i++) {
2452 		err = txq_init(mp, i);
2453 		if (err) {
2454 			while (--i >= 0)
2455 				txq_deinit(mp->txq + i);
2456 			goto out_free;
2457 		}
2458 		mp->int_mask |= INT_TX_END_0 << i;
2459 	}
2460 
2461 	add_timer(&mp->mib_counters_timer);
2462 	port_start(mp);
2463 
2464 	wrlp(mp, INT_MASK_EXT, INT_EXT_LINK_PHY | INT_EXT_TX);
2465 	wrlp(mp, INT_MASK, mp->int_mask);
2466 
2467 	return 0;
2468 
2469 
2470 out_free:
2471 	for (i = 0; i < mp->rxq_count; i++)
2472 		rxq_deinit(mp->rxq + i);
2473 out:
2474 	free_irq(dev->irq, dev);
2475 
2476 	return err;
2477 }
2478 
2479 static void port_reset(struct mv643xx_eth_private *mp)
2480 {
2481 	unsigned int data;
2482 	int i;
2483 
2484 	for (i = 0; i < mp->rxq_count; i++)
2485 		rxq_disable(mp->rxq + i);
2486 	for (i = 0; i < mp->txq_count; i++)
2487 		txq_disable(mp->txq + i);
2488 
2489 	while (1) {
2490 		u32 ps = rdlp(mp, PORT_STATUS);
2491 
2492 		if ((ps & (TX_IN_PROGRESS | TX_FIFO_EMPTY)) == TX_FIFO_EMPTY)
2493 			break;
2494 		udelay(10);
2495 	}
2496 
2497 	/* Reset the Enable bit in the Configuration Register */
2498 	data = rdlp(mp, PORT_SERIAL_CONTROL);
2499 	data &= ~(SERIAL_PORT_ENABLE		|
2500 		  DO_NOT_FORCE_LINK_FAIL	|
2501 		  FORCE_LINK_PASS);
2502 	wrlp(mp, PORT_SERIAL_CONTROL, data);
2503 }
2504 
2505 static int mv643xx_eth_stop(struct net_device *dev)
2506 {
2507 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2508 	int i;
2509 
2510 	wrlp(mp, INT_MASK_EXT, 0x00000000);
2511 	wrlp(mp, INT_MASK, 0x00000000);
2512 	rdlp(mp, INT_MASK);
2513 
2514 	napi_disable(&mp->napi);
2515 
2516 	del_timer_sync(&mp->rx_oom);
2517 
2518 	netif_carrier_off(dev);
2519 	if (dev->phydev)
2520 		phy_stop(dev->phydev);
2521 	free_irq(dev->irq, dev);
2522 
2523 	port_reset(mp);
2524 	mv643xx_eth_get_stats(dev);
2525 	mib_counters_update(mp);
2526 	del_timer_sync(&mp->mib_counters_timer);
2527 
2528 	for (i = 0; i < mp->rxq_count; i++)
2529 		rxq_deinit(mp->rxq + i);
2530 	for (i = 0; i < mp->txq_count; i++)
2531 		txq_deinit(mp->txq + i);
2532 
2533 	return 0;
2534 }
2535 
2536 static int mv643xx_eth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2537 {
2538 	int ret;
2539 
2540 	if (!dev->phydev)
2541 		return -ENOTSUPP;
2542 
2543 	ret = phy_mii_ioctl(dev->phydev, ifr, cmd);
2544 	if (!ret)
2545 		mv643xx_eth_adjust_link(dev);
2546 	return ret;
2547 }
2548 
2549 static int mv643xx_eth_change_mtu(struct net_device *dev, int new_mtu)
2550 {
2551 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2552 
2553 	dev->mtu = new_mtu;
2554 	mv643xx_eth_recalc_skb_size(mp);
2555 	tx_set_rate(mp, 1000000000, 16777216);
2556 
2557 	if (!netif_running(dev))
2558 		return 0;
2559 
2560 	/*
2561 	 * Stop and then re-open the interface. This will allocate RX
2562 	 * skbs of the new MTU.
2563 	 * There is a possible danger that the open will not succeed,
2564 	 * due to memory being full.
2565 	 */
2566 	mv643xx_eth_stop(dev);
2567 	if (mv643xx_eth_open(dev)) {
2568 		netdev_err(dev,
2569 			   "fatal error on re-opening device after MTU change\n");
2570 	}
2571 
2572 	return 0;
2573 }
2574 
2575 static void tx_timeout_task(struct work_struct *ugly)
2576 {
2577 	struct mv643xx_eth_private *mp;
2578 
2579 	mp = container_of(ugly, struct mv643xx_eth_private, tx_timeout_task);
2580 	if (netif_running(mp->dev)) {
2581 		netif_tx_stop_all_queues(mp->dev);
2582 		port_reset(mp);
2583 		port_start(mp);
2584 		netif_tx_wake_all_queues(mp->dev);
2585 	}
2586 }
2587 
2588 static void mv643xx_eth_tx_timeout(struct net_device *dev, unsigned int txqueue)
2589 {
2590 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2591 
2592 	netdev_info(dev, "tx timeout\n");
2593 
2594 	schedule_work(&mp->tx_timeout_task);
2595 }
2596 
2597 #ifdef CONFIG_NET_POLL_CONTROLLER
2598 static void mv643xx_eth_netpoll(struct net_device *dev)
2599 {
2600 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2601 
2602 	wrlp(mp, INT_MASK, 0x00000000);
2603 	rdlp(mp, INT_MASK);
2604 
2605 	mv643xx_eth_irq(dev->irq, dev);
2606 
2607 	wrlp(mp, INT_MASK, mp->int_mask);
2608 }
2609 #endif
2610 
2611 
2612 /* platform glue ************************************************************/
2613 static void
2614 mv643xx_eth_conf_mbus_windows(struct mv643xx_eth_shared_private *msp,
2615 			      const struct mbus_dram_target_info *dram)
2616 {
2617 	void __iomem *base = msp->base;
2618 	u32 win_enable;
2619 	u32 win_protect;
2620 	int i;
2621 
2622 	for (i = 0; i < 6; i++) {
2623 		writel(0, base + WINDOW_BASE(i));
2624 		writel(0, base + WINDOW_SIZE(i));
2625 		if (i < 4)
2626 			writel(0, base + WINDOW_REMAP_HIGH(i));
2627 	}
2628 
2629 	win_enable = 0x3f;
2630 	win_protect = 0;
2631 
2632 	for (i = 0; i < dram->num_cs; i++) {
2633 		const struct mbus_dram_window *cs = dram->cs + i;
2634 
2635 		writel((cs->base & 0xffff0000) |
2636 			(cs->mbus_attr << 8) |
2637 			dram->mbus_dram_target_id, base + WINDOW_BASE(i));
2638 		writel((cs->size - 1) & 0xffff0000, base + WINDOW_SIZE(i));
2639 
2640 		win_enable &= ~(1 << i);
2641 		win_protect |= 3 << (2 * i);
2642 	}
2643 
2644 	writel(win_enable, base + WINDOW_BAR_ENABLE);
2645 	msp->win_protect = win_protect;
2646 }
2647 
2648 static void infer_hw_params(struct mv643xx_eth_shared_private *msp)
2649 {
2650 	/*
2651 	 * Check whether we have a 14-bit coal limit field in bits
2652 	 * [21:8], or a 16-bit coal limit in bits [25,21:7] of the
2653 	 * SDMA config register.
2654 	 */
2655 	writel(0x02000000, msp->base + 0x0400 + SDMA_CONFIG);
2656 	if (readl(msp->base + 0x0400 + SDMA_CONFIG) & 0x02000000)
2657 		msp->extended_rx_coal_limit = 1;
2658 	else
2659 		msp->extended_rx_coal_limit = 0;
2660 
2661 	/*
2662 	 * Check whether the MAC supports TX rate control, and if
2663 	 * yes, whether its associated registers are in the old or
2664 	 * the new place.
2665 	 */
2666 	writel(1, msp->base + 0x0400 + TX_BW_MTU_MOVED);
2667 	if (readl(msp->base + 0x0400 + TX_BW_MTU_MOVED) & 1) {
2668 		msp->tx_bw_control = TX_BW_CONTROL_NEW_LAYOUT;
2669 	} else {
2670 		writel(7, msp->base + 0x0400 + TX_BW_RATE);
2671 		if (readl(msp->base + 0x0400 + TX_BW_RATE) & 7)
2672 			msp->tx_bw_control = TX_BW_CONTROL_OLD_LAYOUT;
2673 		else
2674 			msp->tx_bw_control = TX_BW_CONTROL_ABSENT;
2675 	}
2676 }
2677 
2678 #if defined(CONFIG_OF)
2679 static const struct of_device_id mv643xx_eth_shared_ids[] = {
2680 	{ .compatible = "marvell,orion-eth", },
2681 	{ .compatible = "marvell,kirkwood-eth", },
2682 	{ }
2683 };
2684 MODULE_DEVICE_TABLE(of, mv643xx_eth_shared_ids);
2685 #endif
2686 
2687 #if defined(CONFIG_OF_IRQ) && !defined(CONFIG_MV64X60)
2688 #define mv643xx_eth_property(_np, _name, _v)				\
2689 	do {								\
2690 		u32 tmp;						\
2691 		if (!of_property_read_u32(_np, "marvell," _name, &tmp))	\
2692 			_v = tmp;					\
2693 	} while (0)
2694 
2695 static struct platform_device *port_platdev[3];
2696 
2697 static int mv643xx_eth_shared_of_add_port(struct platform_device *pdev,
2698 					  struct device_node *pnp)
2699 {
2700 	struct platform_device *ppdev;
2701 	struct mv643xx_eth_platform_data ppd;
2702 	struct resource res;
2703 	const char *mac_addr;
2704 	int ret;
2705 	int dev_num = 0;
2706 
2707 	memset(&ppd, 0, sizeof(ppd));
2708 	ppd.shared = pdev;
2709 
2710 	memset(&res, 0, sizeof(res));
2711 	if (of_irq_to_resource(pnp, 0, &res) <= 0) {
2712 		dev_err(&pdev->dev, "missing interrupt on %pOFn\n", pnp);
2713 		return -EINVAL;
2714 	}
2715 
2716 	if (of_property_read_u32(pnp, "reg", &ppd.port_number)) {
2717 		dev_err(&pdev->dev, "missing reg property on %pOFn\n", pnp);
2718 		return -EINVAL;
2719 	}
2720 
2721 	if (ppd.port_number >= 3) {
2722 		dev_err(&pdev->dev, "invalid reg property on %pOFn\n", pnp);
2723 		return -EINVAL;
2724 	}
2725 
2726 	while (dev_num < 3 && port_platdev[dev_num])
2727 		dev_num++;
2728 
2729 	if (dev_num == 3) {
2730 		dev_err(&pdev->dev, "too many ports registered\n");
2731 		return -EINVAL;
2732 	}
2733 
2734 	mac_addr = of_get_mac_address(pnp);
2735 	if (!IS_ERR(mac_addr))
2736 		ether_addr_copy(ppd.mac_addr, mac_addr);
2737 
2738 	mv643xx_eth_property(pnp, "tx-queue-size", ppd.tx_queue_size);
2739 	mv643xx_eth_property(pnp, "tx-sram-addr", ppd.tx_sram_addr);
2740 	mv643xx_eth_property(pnp, "tx-sram-size", ppd.tx_sram_size);
2741 	mv643xx_eth_property(pnp, "rx-queue-size", ppd.rx_queue_size);
2742 	mv643xx_eth_property(pnp, "rx-sram-addr", ppd.rx_sram_addr);
2743 	mv643xx_eth_property(pnp, "rx-sram-size", ppd.rx_sram_size);
2744 
2745 	ppd.phy_node = of_parse_phandle(pnp, "phy-handle", 0);
2746 	if (!ppd.phy_node) {
2747 		ppd.phy_addr = MV643XX_ETH_PHY_NONE;
2748 		of_property_read_u32(pnp, "speed", &ppd.speed);
2749 		of_property_read_u32(pnp, "duplex", &ppd.duplex);
2750 	}
2751 
2752 	ppdev = platform_device_alloc(MV643XX_ETH_NAME, dev_num);
2753 	if (!ppdev)
2754 		return -ENOMEM;
2755 	ppdev->dev.coherent_dma_mask = DMA_BIT_MASK(32);
2756 	ppdev->dev.of_node = pnp;
2757 
2758 	ret = platform_device_add_resources(ppdev, &res, 1);
2759 	if (ret)
2760 		goto port_err;
2761 
2762 	ret = platform_device_add_data(ppdev, &ppd, sizeof(ppd));
2763 	if (ret)
2764 		goto port_err;
2765 
2766 	ret = platform_device_add(ppdev);
2767 	if (ret)
2768 		goto port_err;
2769 
2770 	port_platdev[dev_num] = ppdev;
2771 
2772 	return 0;
2773 
2774 port_err:
2775 	platform_device_put(ppdev);
2776 	return ret;
2777 }
2778 
2779 static int mv643xx_eth_shared_of_probe(struct platform_device *pdev)
2780 {
2781 	struct mv643xx_eth_shared_platform_data *pd;
2782 	struct device_node *pnp, *np = pdev->dev.of_node;
2783 	int ret;
2784 
2785 	/* bail out if not registered from DT */
2786 	if (!np)
2787 		return 0;
2788 
2789 	pd = devm_kzalloc(&pdev->dev, sizeof(*pd), GFP_KERNEL);
2790 	if (!pd)
2791 		return -ENOMEM;
2792 	pdev->dev.platform_data = pd;
2793 
2794 	mv643xx_eth_property(np, "tx-checksum-limit", pd->tx_csum_limit);
2795 
2796 	for_each_available_child_of_node(np, pnp) {
2797 		ret = mv643xx_eth_shared_of_add_port(pdev, pnp);
2798 		if (ret) {
2799 			of_node_put(pnp);
2800 			return ret;
2801 		}
2802 	}
2803 	return 0;
2804 }
2805 
2806 static void mv643xx_eth_shared_of_remove(void)
2807 {
2808 	int n;
2809 
2810 	for (n = 0; n < 3; n++) {
2811 		platform_device_del(port_platdev[n]);
2812 		port_platdev[n] = NULL;
2813 	}
2814 }
2815 #else
2816 static inline int mv643xx_eth_shared_of_probe(struct platform_device *pdev)
2817 {
2818 	return 0;
2819 }
2820 
2821 static inline void mv643xx_eth_shared_of_remove(void)
2822 {
2823 }
2824 #endif
2825 
2826 static int mv643xx_eth_shared_probe(struct platform_device *pdev)
2827 {
2828 	static int mv643xx_eth_version_printed;
2829 	struct mv643xx_eth_shared_platform_data *pd;
2830 	struct mv643xx_eth_shared_private *msp;
2831 	const struct mbus_dram_target_info *dram;
2832 	struct resource *res;
2833 	int ret;
2834 
2835 	if (!mv643xx_eth_version_printed++)
2836 		pr_notice("MV-643xx 10/100/1000 ethernet driver version %s\n",
2837 			  mv643xx_eth_driver_version);
2838 
2839 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
2840 	if (res == NULL)
2841 		return -EINVAL;
2842 
2843 	msp = devm_kzalloc(&pdev->dev, sizeof(*msp), GFP_KERNEL);
2844 	if (msp == NULL)
2845 		return -ENOMEM;
2846 	platform_set_drvdata(pdev, msp);
2847 
2848 	msp->base = devm_ioremap(&pdev->dev, res->start, resource_size(res));
2849 	if (msp->base == NULL)
2850 		return -ENOMEM;
2851 
2852 	msp->clk = devm_clk_get(&pdev->dev, NULL);
2853 	if (!IS_ERR(msp->clk))
2854 		clk_prepare_enable(msp->clk);
2855 
2856 	/*
2857 	 * (Re-)program MBUS remapping windows if we are asked to.
2858 	 */
2859 	dram = mv_mbus_dram_info();
2860 	if (dram)
2861 		mv643xx_eth_conf_mbus_windows(msp, dram);
2862 
2863 	ret = mv643xx_eth_shared_of_probe(pdev);
2864 	if (ret)
2865 		goto err_put_clk;
2866 	pd = dev_get_platdata(&pdev->dev);
2867 
2868 	msp->tx_csum_limit = (pd != NULL && pd->tx_csum_limit) ?
2869 					pd->tx_csum_limit : 9 * 1024;
2870 	infer_hw_params(msp);
2871 
2872 	return 0;
2873 
2874 err_put_clk:
2875 	if (!IS_ERR(msp->clk))
2876 		clk_disable_unprepare(msp->clk);
2877 	return ret;
2878 }
2879 
2880 static int mv643xx_eth_shared_remove(struct platform_device *pdev)
2881 {
2882 	struct mv643xx_eth_shared_private *msp = platform_get_drvdata(pdev);
2883 
2884 	mv643xx_eth_shared_of_remove();
2885 	if (!IS_ERR(msp->clk))
2886 		clk_disable_unprepare(msp->clk);
2887 	return 0;
2888 }
2889 
2890 static struct platform_driver mv643xx_eth_shared_driver = {
2891 	.probe		= mv643xx_eth_shared_probe,
2892 	.remove		= mv643xx_eth_shared_remove,
2893 	.driver = {
2894 		.name	= MV643XX_ETH_SHARED_NAME,
2895 		.of_match_table = of_match_ptr(mv643xx_eth_shared_ids),
2896 	},
2897 };
2898 
2899 static void phy_addr_set(struct mv643xx_eth_private *mp, int phy_addr)
2900 {
2901 	int addr_shift = 5 * mp->port_num;
2902 	u32 data;
2903 
2904 	data = rdl(mp, PHY_ADDR);
2905 	data &= ~(0x1f << addr_shift);
2906 	data |= (phy_addr & 0x1f) << addr_shift;
2907 	wrl(mp, PHY_ADDR, data);
2908 }
2909 
2910 static int phy_addr_get(struct mv643xx_eth_private *mp)
2911 {
2912 	unsigned int data;
2913 
2914 	data = rdl(mp, PHY_ADDR);
2915 
2916 	return (data >> (5 * mp->port_num)) & 0x1f;
2917 }
2918 
2919 static void set_params(struct mv643xx_eth_private *mp,
2920 		       struct mv643xx_eth_platform_data *pd)
2921 {
2922 	struct net_device *dev = mp->dev;
2923 	unsigned int tx_ring_size;
2924 
2925 	if (is_valid_ether_addr(pd->mac_addr))
2926 		memcpy(dev->dev_addr, pd->mac_addr, ETH_ALEN);
2927 	else
2928 		uc_addr_get(mp, dev->dev_addr);
2929 
2930 	mp->rx_ring_size = DEFAULT_RX_QUEUE_SIZE;
2931 	if (pd->rx_queue_size)
2932 		mp->rx_ring_size = pd->rx_queue_size;
2933 	mp->rx_desc_sram_addr = pd->rx_sram_addr;
2934 	mp->rx_desc_sram_size = pd->rx_sram_size;
2935 
2936 	mp->rxq_count = pd->rx_queue_count ? : 1;
2937 
2938 	tx_ring_size = DEFAULT_TX_QUEUE_SIZE;
2939 	if (pd->tx_queue_size)
2940 		tx_ring_size = pd->tx_queue_size;
2941 
2942 	mp->tx_ring_size = clamp_t(unsigned int, tx_ring_size,
2943 				   MV643XX_MAX_SKB_DESCS * 2, 4096);
2944 	if (mp->tx_ring_size != tx_ring_size)
2945 		netdev_warn(dev, "TX queue size set to %u (requested %u)\n",
2946 			    mp->tx_ring_size, tx_ring_size);
2947 
2948 	mp->tx_desc_sram_addr = pd->tx_sram_addr;
2949 	mp->tx_desc_sram_size = pd->tx_sram_size;
2950 
2951 	mp->txq_count = pd->tx_queue_count ? : 1;
2952 }
2953 
2954 static int get_phy_mode(struct mv643xx_eth_private *mp)
2955 {
2956 	struct device *dev = mp->dev->dev.parent;
2957 	phy_interface_t iface;
2958 	int err;
2959 
2960 	if (dev->of_node)
2961 		err = of_get_phy_mode(dev->of_node, &iface);
2962 
2963 	/* Historical default if unspecified. We could also read/write
2964 	 * the interface state in the PSC1
2965 	 */
2966 	if (!dev->of_node || err)
2967 		iface = PHY_INTERFACE_MODE_GMII;
2968 	return iface;
2969 }
2970 
2971 static struct phy_device *phy_scan(struct mv643xx_eth_private *mp,
2972 				   int phy_addr)
2973 {
2974 	struct phy_device *phydev;
2975 	int start;
2976 	int num;
2977 	int i;
2978 	char phy_id[MII_BUS_ID_SIZE + 3];
2979 
2980 	if (phy_addr == MV643XX_ETH_PHY_ADDR_DEFAULT) {
2981 		start = phy_addr_get(mp) & 0x1f;
2982 		num = 32;
2983 	} else {
2984 		start = phy_addr & 0x1f;
2985 		num = 1;
2986 	}
2987 
2988 	/* Attempt to connect to the PHY using orion-mdio */
2989 	phydev = ERR_PTR(-ENODEV);
2990 	for (i = 0; i < num; i++) {
2991 		int addr = (start + i) & 0x1f;
2992 
2993 		snprintf(phy_id, sizeof(phy_id), PHY_ID_FMT,
2994 				"orion-mdio-mii", addr);
2995 
2996 		phydev = phy_connect(mp->dev, phy_id, mv643xx_eth_adjust_link,
2997 				     get_phy_mode(mp));
2998 		if (!IS_ERR(phydev)) {
2999 			phy_addr_set(mp, addr);
3000 			break;
3001 		}
3002 	}
3003 
3004 	return phydev;
3005 }
3006 
3007 static void phy_init(struct mv643xx_eth_private *mp, int speed, int duplex)
3008 {
3009 	struct net_device *dev = mp->dev;
3010 	struct phy_device *phy = dev->phydev;
3011 
3012 	if (speed == 0) {
3013 		phy->autoneg = AUTONEG_ENABLE;
3014 		phy->speed = 0;
3015 		phy->duplex = 0;
3016 		linkmode_copy(phy->advertising, phy->supported);
3017 		linkmode_set_bit(ETHTOOL_LINK_MODE_Autoneg_BIT,
3018 				 phy->advertising);
3019 	} else {
3020 		phy->autoneg = AUTONEG_DISABLE;
3021 		linkmode_zero(phy->advertising);
3022 		phy->speed = speed;
3023 		phy->duplex = duplex;
3024 	}
3025 	phy_start_aneg(phy);
3026 }
3027 
3028 static void init_pscr(struct mv643xx_eth_private *mp, int speed, int duplex)
3029 {
3030 	struct net_device *dev = mp->dev;
3031 	u32 pscr;
3032 
3033 	pscr = rdlp(mp, PORT_SERIAL_CONTROL);
3034 	if (pscr & SERIAL_PORT_ENABLE) {
3035 		pscr &= ~SERIAL_PORT_ENABLE;
3036 		wrlp(mp, PORT_SERIAL_CONTROL, pscr);
3037 	}
3038 
3039 	pscr = MAX_RX_PACKET_9700BYTE | SERIAL_PORT_CONTROL_RESERVED;
3040 	if (!dev->phydev) {
3041 		pscr |= DISABLE_AUTO_NEG_SPEED_GMII;
3042 		if (speed == SPEED_1000)
3043 			pscr |= SET_GMII_SPEED_TO_1000;
3044 		else if (speed == SPEED_100)
3045 			pscr |= SET_MII_SPEED_TO_100;
3046 
3047 		pscr |= DISABLE_AUTO_NEG_FOR_FLOW_CTRL;
3048 
3049 		pscr |= DISABLE_AUTO_NEG_FOR_DUPLEX;
3050 		if (duplex == DUPLEX_FULL)
3051 			pscr |= SET_FULL_DUPLEX_MODE;
3052 	}
3053 
3054 	wrlp(mp, PORT_SERIAL_CONTROL, pscr);
3055 }
3056 
3057 static const struct net_device_ops mv643xx_eth_netdev_ops = {
3058 	.ndo_open		= mv643xx_eth_open,
3059 	.ndo_stop		= mv643xx_eth_stop,
3060 	.ndo_start_xmit		= mv643xx_eth_xmit,
3061 	.ndo_set_rx_mode	= mv643xx_eth_set_rx_mode,
3062 	.ndo_set_mac_address	= mv643xx_eth_set_mac_address,
3063 	.ndo_validate_addr	= eth_validate_addr,
3064 	.ndo_do_ioctl		= mv643xx_eth_ioctl,
3065 	.ndo_change_mtu		= mv643xx_eth_change_mtu,
3066 	.ndo_set_features	= mv643xx_eth_set_features,
3067 	.ndo_tx_timeout		= mv643xx_eth_tx_timeout,
3068 	.ndo_get_stats		= mv643xx_eth_get_stats,
3069 #ifdef CONFIG_NET_POLL_CONTROLLER
3070 	.ndo_poll_controller	= mv643xx_eth_netpoll,
3071 #endif
3072 };
3073 
3074 static int mv643xx_eth_probe(struct platform_device *pdev)
3075 {
3076 	struct mv643xx_eth_platform_data *pd;
3077 	struct mv643xx_eth_private *mp;
3078 	struct net_device *dev;
3079 	struct phy_device *phydev = NULL;
3080 	struct resource *res;
3081 	int err;
3082 
3083 	pd = dev_get_platdata(&pdev->dev);
3084 	if (pd == NULL) {
3085 		dev_err(&pdev->dev, "no mv643xx_eth_platform_data\n");
3086 		return -ENODEV;
3087 	}
3088 
3089 	if (pd->shared == NULL) {
3090 		dev_err(&pdev->dev, "no mv643xx_eth_platform_data->shared\n");
3091 		return -ENODEV;
3092 	}
3093 
3094 	dev = alloc_etherdev_mq(sizeof(struct mv643xx_eth_private), 8);
3095 	if (!dev)
3096 		return -ENOMEM;
3097 
3098 	SET_NETDEV_DEV(dev, &pdev->dev);
3099 	mp = netdev_priv(dev);
3100 	platform_set_drvdata(pdev, mp);
3101 
3102 	mp->shared = platform_get_drvdata(pd->shared);
3103 	mp->base = mp->shared->base + 0x0400 + (pd->port_number << 10);
3104 	mp->port_num = pd->port_number;
3105 
3106 	mp->dev = dev;
3107 
3108 	/* Kirkwood resets some registers on gated clocks. Especially
3109 	 * CLK125_BYPASS_EN must be cleared but is not available on
3110 	 * all other SoCs/System Controllers using this driver.
3111 	 */
3112 	if (of_device_is_compatible(pdev->dev.of_node,
3113 				    "marvell,kirkwood-eth-port"))
3114 		wrlp(mp, PORT_SERIAL_CONTROL1,
3115 		     rdlp(mp, PORT_SERIAL_CONTROL1) & ~CLK125_BYPASS_EN);
3116 
3117 	/*
3118 	 * Start with a default rate, and if there is a clock, allow
3119 	 * it to override the default.
3120 	 */
3121 	mp->t_clk = 133000000;
3122 	mp->clk = devm_clk_get(&pdev->dev, NULL);
3123 	if (!IS_ERR(mp->clk)) {
3124 		clk_prepare_enable(mp->clk);
3125 		mp->t_clk = clk_get_rate(mp->clk);
3126 	} else if (!IS_ERR(mp->shared->clk)) {
3127 		mp->t_clk = clk_get_rate(mp->shared->clk);
3128 	}
3129 
3130 	set_params(mp, pd);
3131 	netif_set_real_num_tx_queues(dev, mp->txq_count);
3132 	netif_set_real_num_rx_queues(dev, mp->rxq_count);
3133 
3134 	err = 0;
3135 	if (pd->phy_node) {
3136 		phydev = of_phy_connect(mp->dev, pd->phy_node,
3137 					mv643xx_eth_adjust_link, 0,
3138 					get_phy_mode(mp));
3139 		if (!phydev)
3140 			err = -ENODEV;
3141 		else
3142 			phy_addr_set(mp, phydev->mdio.addr);
3143 	} else if (pd->phy_addr != MV643XX_ETH_PHY_NONE) {
3144 		phydev = phy_scan(mp, pd->phy_addr);
3145 
3146 		if (IS_ERR(phydev))
3147 			err = PTR_ERR(phydev);
3148 		else
3149 			phy_init(mp, pd->speed, pd->duplex);
3150 	}
3151 	if (err == -ENODEV) {
3152 		err = -EPROBE_DEFER;
3153 		goto out;
3154 	}
3155 	if (err)
3156 		goto out;
3157 
3158 	dev->ethtool_ops = &mv643xx_eth_ethtool_ops;
3159 
3160 	init_pscr(mp, pd->speed, pd->duplex);
3161 
3162 
3163 	mib_counters_clear(mp);
3164 
3165 	timer_setup(&mp->mib_counters_timer, mib_counters_timer_wrapper, 0);
3166 	mp->mib_counters_timer.expires = jiffies + 30 * HZ;
3167 
3168 	spin_lock_init(&mp->mib_counters_lock);
3169 
3170 	INIT_WORK(&mp->tx_timeout_task, tx_timeout_task);
3171 
3172 	netif_napi_add(dev, &mp->napi, mv643xx_eth_poll, NAPI_POLL_WEIGHT);
3173 
3174 	timer_setup(&mp->rx_oom, oom_timer_wrapper, 0);
3175 
3176 
3177 	res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
3178 	BUG_ON(!res);
3179 	dev->irq = res->start;
3180 
3181 	dev->netdev_ops = &mv643xx_eth_netdev_ops;
3182 
3183 	dev->watchdog_timeo = 2 * HZ;
3184 	dev->base_addr = 0;
3185 
3186 	dev->features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO;
3187 	dev->vlan_features = dev->features;
3188 
3189 	dev->features |= NETIF_F_RXCSUM;
3190 	dev->hw_features = dev->features;
3191 
3192 	dev->priv_flags |= IFF_UNICAST_FLT;
3193 	dev->gso_max_segs = MV643XX_MAX_TSO_SEGS;
3194 
3195 	/* MTU range: 64 - 9500 */
3196 	dev->min_mtu = 64;
3197 	dev->max_mtu = 9500;
3198 
3199 	if (mp->shared->win_protect)
3200 		wrl(mp, WINDOW_PROTECT(mp->port_num), mp->shared->win_protect);
3201 
3202 	netif_carrier_off(dev);
3203 
3204 	wrlp(mp, SDMA_CONFIG, PORT_SDMA_CONFIG_DEFAULT_VALUE);
3205 
3206 	set_rx_coal(mp, 250);
3207 	set_tx_coal(mp, 0);
3208 
3209 	err = register_netdev(dev);
3210 	if (err)
3211 		goto out;
3212 
3213 	netdev_notice(dev, "port %d with MAC address %pM\n",
3214 		      mp->port_num, dev->dev_addr);
3215 
3216 	if (mp->tx_desc_sram_size > 0)
3217 		netdev_notice(dev, "configured with sram\n");
3218 
3219 	return 0;
3220 
3221 out:
3222 	if (!IS_ERR(mp->clk))
3223 		clk_disable_unprepare(mp->clk);
3224 	free_netdev(dev);
3225 
3226 	return err;
3227 }
3228 
3229 static int mv643xx_eth_remove(struct platform_device *pdev)
3230 {
3231 	struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
3232 	struct net_device *dev = mp->dev;
3233 
3234 	unregister_netdev(mp->dev);
3235 	if (dev->phydev)
3236 		phy_disconnect(dev->phydev);
3237 	cancel_work_sync(&mp->tx_timeout_task);
3238 
3239 	if (!IS_ERR(mp->clk))
3240 		clk_disable_unprepare(mp->clk);
3241 
3242 	free_netdev(mp->dev);
3243 
3244 	return 0;
3245 }
3246 
3247 static void mv643xx_eth_shutdown(struct platform_device *pdev)
3248 {
3249 	struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
3250 
3251 	/* Mask all interrupts on ethernet port */
3252 	wrlp(mp, INT_MASK, 0);
3253 	rdlp(mp, INT_MASK);
3254 
3255 	if (netif_running(mp->dev))
3256 		port_reset(mp);
3257 }
3258 
3259 static struct platform_driver mv643xx_eth_driver = {
3260 	.probe		= mv643xx_eth_probe,
3261 	.remove		= mv643xx_eth_remove,
3262 	.shutdown	= mv643xx_eth_shutdown,
3263 	.driver = {
3264 		.name	= MV643XX_ETH_NAME,
3265 	},
3266 };
3267 
3268 static struct platform_driver * const drivers[] = {
3269 	&mv643xx_eth_shared_driver,
3270 	&mv643xx_eth_driver,
3271 };
3272 
3273 static int __init mv643xx_eth_init_module(void)
3274 {
3275 	return platform_register_drivers(drivers, ARRAY_SIZE(drivers));
3276 }
3277 module_init(mv643xx_eth_init_module);
3278 
3279 static void __exit mv643xx_eth_cleanup_module(void)
3280 {
3281 	platform_unregister_drivers(drivers, ARRAY_SIZE(drivers));
3282 }
3283 module_exit(mv643xx_eth_cleanup_module);
3284 
3285 MODULE_AUTHOR("Rabeeh Khoury, Assaf Hoffman, Matthew Dharm, "
3286 	      "Manish Lachwani, Dale Farnsworth and Lennert Buytenhek");
3287 MODULE_DESCRIPTION("Ethernet driver for Marvell MV643XX");
3288 MODULE_LICENSE("GPL");
3289 MODULE_ALIAS("platform:" MV643XX_ETH_SHARED_NAME);
3290 MODULE_ALIAS("platform:" MV643XX_ETH_NAME);
3291