xref: /linux/drivers/net/ethernet/marvell/mv643xx_eth.c (revision b43ab901d671e3e3cad425ea5e9a3c74e266dcdd)
1 /*
2  * Driver for Marvell Discovery (MV643XX) and Marvell Orion ethernet ports
3  * Copyright (C) 2002 Matthew Dharm <mdharm@momenco.com>
4  *
5  * Based on the 64360 driver from:
6  * Copyright (C) 2002 Rabeeh Khoury <rabeeh@galileo.co.il>
7  *		      Rabeeh Khoury <rabeeh@marvell.com>
8  *
9  * Copyright (C) 2003 PMC-Sierra, Inc.,
10  *	written by Manish Lachwani
11  *
12  * Copyright (C) 2003 Ralf Baechle <ralf@linux-mips.org>
13  *
14  * Copyright (C) 2004-2006 MontaVista Software, Inc.
15  *			   Dale Farnsworth <dale@farnsworth.org>
16  *
17  * Copyright (C) 2004 Steven J. Hill <sjhill1@rockwellcollins.com>
18  *				     <sjhill@realitydiluted.com>
19  *
20  * Copyright (C) 2007-2008 Marvell Semiconductor
21  *			   Lennert Buytenhek <buytenh@marvell.com>
22  *
23  * This program is free software; you can redistribute it and/or
24  * modify it under the terms of the GNU General Public License
25  * as published by the Free Software Foundation; either version 2
26  * of the License, or (at your option) any later version.
27  *
28  * This program is distributed in the hope that it will be useful,
29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31  * GNU General Public License for more details.
32  *
33  * You should have received a copy of the GNU General Public License
34  * along with this program; if not, write to the Free Software
35  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
36  */
37 
38 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 
40 #include <linux/init.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/in.h>
43 #include <linux/ip.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/etherdevice.h>
47 #include <linux/delay.h>
48 #include <linux/ethtool.h>
49 #include <linux/platform_device.h>
50 #include <linux/module.h>
51 #include <linux/kernel.h>
52 #include <linux/spinlock.h>
53 #include <linux/workqueue.h>
54 #include <linux/phy.h>
55 #include <linux/mv643xx_eth.h>
56 #include <linux/io.h>
57 #include <linux/types.h>
58 #include <linux/inet_lro.h>
59 #include <linux/slab.h>
60 #include <asm/system.h>
61 
62 static char mv643xx_eth_driver_name[] = "mv643xx_eth";
63 static char mv643xx_eth_driver_version[] = "1.4";
64 
65 
66 /*
67  * Registers shared between all ports.
68  */
69 #define PHY_ADDR			0x0000
70 #define SMI_REG				0x0004
71 #define  SMI_BUSY			0x10000000
72 #define  SMI_READ_VALID			0x08000000
73 #define  SMI_OPCODE_READ		0x04000000
74 #define  SMI_OPCODE_WRITE		0x00000000
75 #define ERR_INT_CAUSE			0x0080
76 #define  ERR_INT_SMI_DONE		0x00000010
77 #define ERR_INT_MASK			0x0084
78 #define WINDOW_BASE(w)			(0x0200 + ((w) << 3))
79 #define WINDOW_SIZE(w)			(0x0204 + ((w) << 3))
80 #define WINDOW_REMAP_HIGH(w)		(0x0280 + ((w) << 2))
81 #define WINDOW_BAR_ENABLE		0x0290
82 #define WINDOW_PROTECT(w)		(0x0294 + ((w) << 4))
83 
84 /*
85  * Main per-port registers.  These live at offset 0x0400 for
86  * port #0, 0x0800 for port #1, and 0x0c00 for port #2.
87  */
88 #define PORT_CONFIG			0x0000
89 #define  UNICAST_PROMISCUOUS_MODE	0x00000001
90 #define PORT_CONFIG_EXT			0x0004
91 #define MAC_ADDR_LOW			0x0014
92 #define MAC_ADDR_HIGH			0x0018
93 #define SDMA_CONFIG			0x001c
94 #define  TX_BURST_SIZE_16_64BIT		0x01000000
95 #define  TX_BURST_SIZE_4_64BIT		0x00800000
96 #define  BLM_TX_NO_SWAP			0x00000020
97 #define  BLM_RX_NO_SWAP			0x00000010
98 #define  RX_BURST_SIZE_16_64BIT		0x00000008
99 #define  RX_BURST_SIZE_4_64BIT		0x00000004
100 #define PORT_SERIAL_CONTROL		0x003c
101 #define  SET_MII_SPEED_TO_100		0x01000000
102 #define  SET_GMII_SPEED_TO_1000		0x00800000
103 #define  SET_FULL_DUPLEX_MODE		0x00200000
104 #define  MAX_RX_PACKET_9700BYTE		0x000a0000
105 #define  DISABLE_AUTO_NEG_SPEED_GMII	0x00002000
106 #define  DO_NOT_FORCE_LINK_FAIL		0x00000400
107 #define  SERIAL_PORT_CONTROL_RESERVED	0x00000200
108 #define  DISABLE_AUTO_NEG_FOR_FLOW_CTRL	0x00000008
109 #define  DISABLE_AUTO_NEG_FOR_DUPLEX	0x00000004
110 #define  FORCE_LINK_PASS		0x00000002
111 #define  SERIAL_PORT_ENABLE		0x00000001
112 #define PORT_STATUS			0x0044
113 #define  TX_FIFO_EMPTY			0x00000400
114 #define  TX_IN_PROGRESS			0x00000080
115 #define  PORT_SPEED_MASK		0x00000030
116 #define  PORT_SPEED_1000		0x00000010
117 #define  PORT_SPEED_100			0x00000020
118 #define  PORT_SPEED_10			0x00000000
119 #define  FLOW_CONTROL_ENABLED		0x00000008
120 #define  FULL_DUPLEX			0x00000004
121 #define  LINK_UP			0x00000002
122 #define TXQ_COMMAND			0x0048
123 #define TXQ_FIX_PRIO_CONF		0x004c
124 #define TX_BW_RATE			0x0050
125 #define TX_BW_MTU			0x0058
126 #define TX_BW_BURST			0x005c
127 #define INT_CAUSE			0x0060
128 #define  INT_TX_END			0x07f80000
129 #define  INT_TX_END_0			0x00080000
130 #define  INT_RX				0x000003fc
131 #define  INT_RX_0			0x00000004
132 #define  INT_EXT			0x00000002
133 #define INT_CAUSE_EXT			0x0064
134 #define  INT_EXT_LINK_PHY		0x00110000
135 #define  INT_EXT_TX			0x000000ff
136 #define INT_MASK			0x0068
137 #define INT_MASK_EXT			0x006c
138 #define TX_FIFO_URGENT_THRESHOLD	0x0074
139 #define TXQ_FIX_PRIO_CONF_MOVED		0x00dc
140 #define TX_BW_RATE_MOVED		0x00e0
141 #define TX_BW_MTU_MOVED			0x00e8
142 #define TX_BW_BURST_MOVED		0x00ec
143 #define RXQ_CURRENT_DESC_PTR(q)		(0x020c + ((q) << 4))
144 #define RXQ_COMMAND			0x0280
145 #define TXQ_CURRENT_DESC_PTR(q)		(0x02c0 + ((q) << 2))
146 #define TXQ_BW_TOKENS(q)		(0x0300 + ((q) << 4))
147 #define TXQ_BW_CONF(q)			(0x0304 + ((q) << 4))
148 #define TXQ_BW_WRR_CONF(q)		(0x0308 + ((q) << 4))
149 
150 /*
151  * Misc per-port registers.
152  */
153 #define MIB_COUNTERS(p)			(0x1000 + ((p) << 7))
154 #define SPECIAL_MCAST_TABLE(p)		(0x1400 + ((p) << 10))
155 #define OTHER_MCAST_TABLE(p)		(0x1500 + ((p) << 10))
156 #define UNICAST_TABLE(p)		(0x1600 + ((p) << 10))
157 
158 
159 /*
160  * SDMA configuration register default value.
161  */
162 #if defined(__BIG_ENDIAN)
163 #define PORT_SDMA_CONFIG_DEFAULT_VALUE		\
164 		(RX_BURST_SIZE_4_64BIT	|	\
165 		 TX_BURST_SIZE_4_64BIT)
166 #elif defined(__LITTLE_ENDIAN)
167 #define PORT_SDMA_CONFIG_DEFAULT_VALUE		\
168 		(RX_BURST_SIZE_4_64BIT	|	\
169 		 BLM_RX_NO_SWAP		|	\
170 		 BLM_TX_NO_SWAP		|	\
171 		 TX_BURST_SIZE_4_64BIT)
172 #else
173 #error One of __BIG_ENDIAN or __LITTLE_ENDIAN must be defined
174 #endif
175 
176 
177 /*
178  * Misc definitions.
179  */
180 #define DEFAULT_RX_QUEUE_SIZE	128
181 #define DEFAULT_TX_QUEUE_SIZE	256
182 #define SKB_DMA_REALIGN		((PAGE_SIZE - NET_SKB_PAD) % SMP_CACHE_BYTES)
183 
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 MAC_HDR_EXTRA_8_BYTES		0x00000200
253 
254 #define TX_IHL_SHIFT			11
255 
256 
257 /* global *******************************************************************/
258 struct mv643xx_eth_shared_private {
259 	/*
260 	 * Ethernet controller base address.
261 	 */
262 	void __iomem *base;
263 
264 	/*
265 	 * Points at the right SMI instance to use.
266 	 */
267 	struct mv643xx_eth_shared_private *smi;
268 
269 	/*
270 	 * Provides access to local SMI interface.
271 	 */
272 	struct mii_bus *smi_bus;
273 
274 	/*
275 	 * If we have access to the error interrupt pin (which is
276 	 * somewhat misnamed as it not only reflects internal errors
277 	 * but also reflects SMI completion), use that to wait for
278 	 * SMI access completion instead of polling the SMI busy bit.
279 	 */
280 	int err_interrupt;
281 	wait_queue_head_t smi_busy_wait;
282 
283 	/*
284 	 * Per-port MBUS window access register value.
285 	 */
286 	u32 win_protect;
287 
288 	/*
289 	 * Hardware-specific parameters.
290 	 */
291 	unsigned int t_clk;
292 	int extended_rx_coal_limit;
293 	int tx_bw_control;
294 	int tx_csum_limit;
295 };
296 
297 #define TX_BW_CONTROL_ABSENT		0
298 #define TX_BW_CONTROL_OLD_LAYOUT	1
299 #define TX_BW_CONTROL_NEW_LAYOUT	2
300 
301 static int mv643xx_eth_open(struct net_device *dev);
302 static int mv643xx_eth_stop(struct net_device *dev);
303 
304 
305 /* per-port *****************************************************************/
306 struct mib_counters {
307 	u64 good_octets_received;
308 	u32 bad_octets_received;
309 	u32 internal_mac_transmit_err;
310 	u32 good_frames_received;
311 	u32 bad_frames_received;
312 	u32 broadcast_frames_received;
313 	u32 multicast_frames_received;
314 	u32 frames_64_octets;
315 	u32 frames_65_to_127_octets;
316 	u32 frames_128_to_255_octets;
317 	u32 frames_256_to_511_octets;
318 	u32 frames_512_to_1023_octets;
319 	u32 frames_1024_to_max_octets;
320 	u64 good_octets_sent;
321 	u32 good_frames_sent;
322 	u32 excessive_collision;
323 	u32 multicast_frames_sent;
324 	u32 broadcast_frames_sent;
325 	u32 unrec_mac_control_received;
326 	u32 fc_sent;
327 	u32 good_fc_received;
328 	u32 bad_fc_received;
329 	u32 undersize_received;
330 	u32 fragments_received;
331 	u32 oversize_received;
332 	u32 jabber_received;
333 	u32 mac_receive_error;
334 	u32 bad_crc_event;
335 	u32 collision;
336 	u32 late_collision;
337 };
338 
339 struct lro_counters {
340 	u32 lro_aggregated;
341 	u32 lro_flushed;
342 	u32 lro_no_desc;
343 };
344 
345 struct rx_queue {
346 	int index;
347 
348 	int rx_ring_size;
349 
350 	int rx_desc_count;
351 	int rx_curr_desc;
352 	int rx_used_desc;
353 
354 	struct rx_desc *rx_desc_area;
355 	dma_addr_t rx_desc_dma;
356 	int rx_desc_area_size;
357 	struct sk_buff **rx_skb;
358 
359 	struct net_lro_mgr lro_mgr;
360 	struct net_lro_desc lro_arr[8];
361 };
362 
363 struct tx_queue {
364 	int index;
365 
366 	int tx_ring_size;
367 
368 	int tx_desc_count;
369 	int tx_curr_desc;
370 	int tx_used_desc;
371 
372 	struct tx_desc *tx_desc_area;
373 	dma_addr_t tx_desc_dma;
374 	int tx_desc_area_size;
375 
376 	struct sk_buff_head tx_skb;
377 
378 	unsigned long tx_packets;
379 	unsigned long tx_bytes;
380 	unsigned long tx_dropped;
381 };
382 
383 struct mv643xx_eth_private {
384 	struct mv643xx_eth_shared_private *shared;
385 	void __iomem *base;
386 	int port_num;
387 
388 	struct net_device *dev;
389 
390 	struct phy_device *phy;
391 
392 	struct timer_list mib_counters_timer;
393 	spinlock_t mib_counters_lock;
394 	struct mib_counters mib_counters;
395 
396 	struct lro_counters lro_counters;
397 
398 	struct work_struct tx_timeout_task;
399 
400 	struct napi_struct napi;
401 	u32 int_mask;
402 	u8 oom;
403 	u8 work_link;
404 	u8 work_tx;
405 	u8 work_tx_end;
406 	u8 work_rx;
407 	u8 work_rx_refill;
408 
409 	int skb_size;
410 	struct sk_buff_head rx_recycle;
411 
412 	/*
413 	 * RX state.
414 	 */
415 	int rx_ring_size;
416 	unsigned long rx_desc_sram_addr;
417 	int rx_desc_sram_size;
418 	int rxq_count;
419 	struct timer_list rx_oom;
420 	struct rx_queue rxq[8];
421 
422 	/*
423 	 * TX state.
424 	 */
425 	int tx_ring_size;
426 	unsigned long tx_desc_sram_addr;
427 	int tx_desc_sram_size;
428 	int txq_count;
429 	struct tx_queue txq[8];
430 };
431 
432 
433 /* port register accessors **************************************************/
434 static inline u32 rdl(struct mv643xx_eth_private *mp, int offset)
435 {
436 	return readl(mp->shared->base + offset);
437 }
438 
439 static inline u32 rdlp(struct mv643xx_eth_private *mp, int offset)
440 {
441 	return readl(mp->base + offset);
442 }
443 
444 static inline void wrl(struct mv643xx_eth_private *mp, int offset, u32 data)
445 {
446 	writel(data, mp->shared->base + offset);
447 }
448 
449 static inline void wrlp(struct mv643xx_eth_private *mp, int offset, u32 data)
450 {
451 	writel(data, mp->base + offset);
452 }
453 
454 
455 /* rxq/txq helper functions *************************************************/
456 static struct mv643xx_eth_private *rxq_to_mp(struct rx_queue *rxq)
457 {
458 	return container_of(rxq, struct mv643xx_eth_private, rxq[rxq->index]);
459 }
460 
461 static struct mv643xx_eth_private *txq_to_mp(struct tx_queue *txq)
462 {
463 	return container_of(txq, struct mv643xx_eth_private, txq[txq->index]);
464 }
465 
466 static void rxq_enable(struct rx_queue *rxq)
467 {
468 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
469 	wrlp(mp, RXQ_COMMAND, 1 << rxq->index);
470 }
471 
472 static void rxq_disable(struct rx_queue *rxq)
473 {
474 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
475 	u8 mask = 1 << rxq->index;
476 
477 	wrlp(mp, RXQ_COMMAND, mask << 8);
478 	while (rdlp(mp, RXQ_COMMAND) & mask)
479 		udelay(10);
480 }
481 
482 static void txq_reset_hw_ptr(struct tx_queue *txq)
483 {
484 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
485 	u32 addr;
486 
487 	addr = (u32)txq->tx_desc_dma;
488 	addr += txq->tx_curr_desc * sizeof(struct tx_desc);
489 	wrlp(mp, TXQ_CURRENT_DESC_PTR(txq->index), addr);
490 }
491 
492 static void txq_enable(struct tx_queue *txq)
493 {
494 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
495 	wrlp(mp, TXQ_COMMAND, 1 << txq->index);
496 }
497 
498 static void txq_disable(struct tx_queue *txq)
499 {
500 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
501 	u8 mask = 1 << txq->index;
502 
503 	wrlp(mp, TXQ_COMMAND, mask << 8);
504 	while (rdlp(mp, TXQ_COMMAND) & mask)
505 		udelay(10);
506 }
507 
508 static void txq_maybe_wake(struct tx_queue *txq)
509 {
510 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
511 	struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
512 
513 	if (netif_tx_queue_stopped(nq)) {
514 		__netif_tx_lock(nq, smp_processor_id());
515 		if (txq->tx_ring_size - txq->tx_desc_count >= MAX_SKB_FRAGS + 1)
516 			netif_tx_wake_queue(nq);
517 		__netif_tx_unlock(nq);
518 	}
519 }
520 
521 
522 /* rx napi ******************************************************************/
523 static int
524 mv643xx_get_skb_header(struct sk_buff *skb, void **iphdr, void **tcph,
525 		       u64 *hdr_flags, void *priv)
526 {
527 	unsigned long cmd_sts = (unsigned long)priv;
528 
529 	/*
530 	 * Make sure that this packet is Ethernet II, is not VLAN
531 	 * tagged, is IPv4, has a valid IP header, and is TCP.
532 	 */
533 	if ((cmd_sts & (RX_IP_HDR_OK | RX_PKT_IS_IPV4 |
534 		       RX_PKT_IS_ETHERNETV2 | RX_PKT_LAYER4_TYPE_MASK |
535 		       RX_PKT_IS_VLAN_TAGGED)) !=
536 	    (RX_IP_HDR_OK | RX_PKT_IS_IPV4 |
537 	     RX_PKT_IS_ETHERNETV2 | RX_PKT_LAYER4_TYPE_TCP_IPV4))
538 		return -1;
539 
540 	skb_reset_network_header(skb);
541 	skb_set_transport_header(skb, ip_hdrlen(skb));
542 	*iphdr = ip_hdr(skb);
543 	*tcph = tcp_hdr(skb);
544 	*hdr_flags = LRO_IPV4 | LRO_TCP;
545 
546 	return 0;
547 }
548 
549 static int rxq_process(struct rx_queue *rxq, int budget)
550 {
551 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
552 	struct net_device_stats *stats = &mp->dev->stats;
553 	int lro_flush_needed;
554 	int rx;
555 
556 	lro_flush_needed = 0;
557 	rx = 0;
558 	while (rx < budget && rxq->rx_desc_count) {
559 		struct rx_desc *rx_desc;
560 		unsigned int cmd_sts;
561 		struct sk_buff *skb;
562 		u16 byte_cnt;
563 
564 		rx_desc = &rxq->rx_desc_area[rxq->rx_curr_desc];
565 
566 		cmd_sts = rx_desc->cmd_sts;
567 		if (cmd_sts & BUFFER_OWNED_BY_DMA)
568 			break;
569 		rmb();
570 
571 		skb = rxq->rx_skb[rxq->rx_curr_desc];
572 		rxq->rx_skb[rxq->rx_curr_desc] = NULL;
573 
574 		rxq->rx_curr_desc++;
575 		if (rxq->rx_curr_desc == rxq->rx_ring_size)
576 			rxq->rx_curr_desc = 0;
577 
578 		dma_unmap_single(mp->dev->dev.parent, rx_desc->buf_ptr,
579 				 rx_desc->buf_size, DMA_FROM_DEVICE);
580 		rxq->rx_desc_count--;
581 		rx++;
582 
583 		mp->work_rx_refill |= 1 << rxq->index;
584 
585 		byte_cnt = rx_desc->byte_cnt;
586 
587 		/*
588 		 * Update statistics.
589 		 *
590 		 * Note that the descriptor byte count includes 2 dummy
591 		 * bytes automatically inserted by the hardware at the
592 		 * start of the packet (which we don't count), and a 4
593 		 * byte CRC at the end of the packet (which we do count).
594 		 */
595 		stats->rx_packets++;
596 		stats->rx_bytes += byte_cnt - 2;
597 
598 		/*
599 		 * In case we received a packet without first / last bits
600 		 * on, or the error summary bit is set, the packet needs
601 		 * to be dropped.
602 		 */
603 		if ((cmd_sts & (RX_FIRST_DESC | RX_LAST_DESC | ERROR_SUMMARY))
604 			!= (RX_FIRST_DESC | RX_LAST_DESC))
605 			goto err;
606 
607 		/*
608 		 * The -4 is for the CRC in the trailer of the
609 		 * received packet
610 		 */
611 		skb_put(skb, byte_cnt - 2 - 4);
612 
613 		if (cmd_sts & LAYER_4_CHECKSUM_OK)
614 			skb->ip_summed = CHECKSUM_UNNECESSARY;
615 		skb->protocol = eth_type_trans(skb, mp->dev);
616 
617 		if (skb->dev->features & NETIF_F_LRO &&
618 		    skb->ip_summed == CHECKSUM_UNNECESSARY) {
619 			lro_receive_skb(&rxq->lro_mgr, skb, (void *)cmd_sts);
620 			lro_flush_needed = 1;
621 		} else
622 			netif_receive_skb(skb);
623 
624 		continue;
625 
626 err:
627 		stats->rx_dropped++;
628 
629 		if ((cmd_sts & (RX_FIRST_DESC | RX_LAST_DESC)) !=
630 			(RX_FIRST_DESC | RX_LAST_DESC)) {
631 			if (net_ratelimit())
632 				netdev_err(mp->dev,
633 					   "received packet spanning multiple descriptors\n");
634 		}
635 
636 		if (cmd_sts & ERROR_SUMMARY)
637 			stats->rx_errors++;
638 
639 		dev_kfree_skb(skb);
640 	}
641 
642 	if (lro_flush_needed)
643 		lro_flush_all(&rxq->lro_mgr);
644 
645 	if (rx < budget)
646 		mp->work_rx &= ~(1 << rxq->index);
647 
648 	return rx;
649 }
650 
651 static int rxq_refill(struct rx_queue *rxq, int budget)
652 {
653 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
654 	int refilled;
655 
656 	refilled = 0;
657 	while (refilled < budget && rxq->rx_desc_count < rxq->rx_ring_size) {
658 		struct sk_buff *skb;
659 		int rx;
660 		struct rx_desc *rx_desc;
661 		int size;
662 
663 		skb = __skb_dequeue(&mp->rx_recycle);
664 		if (skb == NULL)
665 			skb = dev_alloc_skb(mp->skb_size);
666 
667 		if (skb == NULL) {
668 			mp->oom = 1;
669 			goto oom;
670 		}
671 
672 		if (SKB_DMA_REALIGN)
673 			skb_reserve(skb, SKB_DMA_REALIGN);
674 
675 		refilled++;
676 		rxq->rx_desc_count++;
677 
678 		rx = rxq->rx_used_desc++;
679 		if (rxq->rx_used_desc == rxq->rx_ring_size)
680 			rxq->rx_used_desc = 0;
681 
682 		rx_desc = rxq->rx_desc_area + rx;
683 
684 		size = skb->end - skb->data;
685 		rx_desc->buf_ptr = dma_map_single(mp->dev->dev.parent,
686 						  skb->data, size,
687 						  DMA_FROM_DEVICE);
688 		rx_desc->buf_size = size;
689 		rxq->rx_skb[rx] = skb;
690 		wmb();
691 		rx_desc->cmd_sts = BUFFER_OWNED_BY_DMA | RX_ENABLE_INTERRUPT;
692 		wmb();
693 
694 		/*
695 		 * The hardware automatically prepends 2 bytes of
696 		 * dummy data to each received packet, so that the
697 		 * IP header ends up 16-byte aligned.
698 		 */
699 		skb_reserve(skb, 2);
700 	}
701 
702 	if (refilled < budget)
703 		mp->work_rx_refill &= ~(1 << rxq->index);
704 
705 oom:
706 	return refilled;
707 }
708 
709 
710 /* tx ***********************************************************************/
711 static inline unsigned int has_tiny_unaligned_frags(struct sk_buff *skb)
712 {
713 	int frag;
714 
715 	for (frag = 0; frag < skb_shinfo(skb)->nr_frags; frag++) {
716 		const skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];
717 
718 		if (skb_frag_size(fragp) <= 8 && fragp->page_offset & 7)
719 			return 1;
720 	}
721 
722 	return 0;
723 }
724 
725 static void txq_submit_frag_skb(struct tx_queue *txq, struct sk_buff *skb)
726 {
727 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
728 	int nr_frags = skb_shinfo(skb)->nr_frags;
729 	int frag;
730 
731 	for (frag = 0; frag < nr_frags; frag++) {
732 		skb_frag_t *this_frag;
733 		int tx_index;
734 		struct tx_desc *desc;
735 
736 		this_frag = &skb_shinfo(skb)->frags[frag];
737 		tx_index = txq->tx_curr_desc++;
738 		if (txq->tx_curr_desc == txq->tx_ring_size)
739 			txq->tx_curr_desc = 0;
740 		desc = &txq->tx_desc_area[tx_index];
741 
742 		/*
743 		 * The last fragment will generate an interrupt
744 		 * which will free the skb on TX completion.
745 		 */
746 		if (frag == nr_frags - 1) {
747 			desc->cmd_sts = BUFFER_OWNED_BY_DMA |
748 					ZERO_PADDING | TX_LAST_DESC |
749 					TX_ENABLE_INTERRUPT;
750 		} else {
751 			desc->cmd_sts = BUFFER_OWNED_BY_DMA;
752 		}
753 
754 		desc->l4i_chk = 0;
755 		desc->byte_cnt = skb_frag_size(this_frag);
756 		desc->buf_ptr = skb_frag_dma_map(mp->dev->dev.parent,
757 						 this_frag, 0,
758 						 skb_frag_size(this_frag),
759 						 DMA_TO_DEVICE);
760 	}
761 }
762 
763 static inline __be16 sum16_as_be(__sum16 sum)
764 {
765 	return (__force __be16)sum;
766 }
767 
768 static int txq_submit_skb(struct tx_queue *txq, struct sk_buff *skb)
769 {
770 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
771 	int nr_frags = skb_shinfo(skb)->nr_frags;
772 	int tx_index;
773 	struct tx_desc *desc;
774 	u32 cmd_sts;
775 	u16 l4i_chk;
776 	int length;
777 
778 	cmd_sts = TX_FIRST_DESC | GEN_CRC | BUFFER_OWNED_BY_DMA;
779 	l4i_chk = 0;
780 
781 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
782 		int hdr_len;
783 		int tag_bytes;
784 
785 		BUG_ON(skb->protocol != htons(ETH_P_IP) &&
786 		       skb->protocol != htons(ETH_P_8021Q));
787 
788 		hdr_len = (void *)ip_hdr(skb) - (void *)skb->data;
789 		tag_bytes = hdr_len - ETH_HLEN;
790 		if (skb->len - hdr_len > mp->shared->tx_csum_limit ||
791 		    unlikely(tag_bytes & ~12)) {
792 			if (skb_checksum_help(skb) == 0)
793 				goto no_csum;
794 			kfree_skb(skb);
795 			return 1;
796 		}
797 
798 		if (tag_bytes & 4)
799 			cmd_sts |= MAC_HDR_EXTRA_4_BYTES;
800 		if (tag_bytes & 8)
801 			cmd_sts |= MAC_HDR_EXTRA_8_BYTES;
802 
803 		cmd_sts |= GEN_TCP_UDP_CHECKSUM |
804 			   GEN_IP_V4_CHECKSUM   |
805 			   ip_hdr(skb)->ihl << TX_IHL_SHIFT;
806 
807 		switch (ip_hdr(skb)->protocol) {
808 		case IPPROTO_UDP:
809 			cmd_sts |= UDP_FRAME;
810 			l4i_chk = ntohs(sum16_as_be(udp_hdr(skb)->check));
811 			break;
812 		case IPPROTO_TCP:
813 			l4i_chk = ntohs(sum16_as_be(tcp_hdr(skb)->check));
814 			break;
815 		default:
816 			BUG();
817 		}
818 	} else {
819 no_csum:
820 		/* Errata BTS #50, IHL must be 5 if no HW checksum */
821 		cmd_sts |= 5 << TX_IHL_SHIFT;
822 	}
823 
824 	tx_index = txq->tx_curr_desc++;
825 	if (txq->tx_curr_desc == txq->tx_ring_size)
826 		txq->tx_curr_desc = 0;
827 	desc = &txq->tx_desc_area[tx_index];
828 
829 	if (nr_frags) {
830 		txq_submit_frag_skb(txq, skb);
831 		length = skb_headlen(skb);
832 	} else {
833 		cmd_sts |= ZERO_PADDING | TX_LAST_DESC | TX_ENABLE_INTERRUPT;
834 		length = skb->len;
835 	}
836 
837 	desc->l4i_chk = l4i_chk;
838 	desc->byte_cnt = length;
839 	desc->buf_ptr = dma_map_single(mp->dev->dev.parent, skb->data,
840 				       length, DMA_TO_DEVICE);
841 
842 	__skb_queue_tail(&txq->tx_skb, skb);
843 
844 	skb_tx_timestamp(skb);
845 
846 	/* ensure all other descriptors are written before first cmd_sts */
847 	wmb();
848 	desc->cmd_sts = cmd_sts;
849 
850 	/* clear TX_END status */
851 	mp->work_tx_end &= ~(1 << txq->index);
852 
853 	/* ensure all descriptors are written before poking hardware */
854 	wmb();
855 	txq_enable(txq);
856 
857 	txq->tx_desc_count += nr_frags + 1;
858 
859 	return 0;
860 }
861 
862 static netdev_tx_t mv643xx_eth_xmit(struct sk_buff *skb, struct net_device *dev)
863 {
864 	struct mv643xx_eth_private *mp = netdev_priv(dev);
865 	int length, queue;
866 	struct tx_queue *txq;
867 	struct netdev_queue *nq;
868 
869 	queue = skb_get_queue_mapping(skb);
870 	txq = mp->txq + queue;
871 	nq = netdev_get_tx_queue(dev, queue);
872 
873 	if (has_tiny_unaligned_frags(skb) && __skb_linearize(skb)) {
874 		txq->tx_dropped++;
875 		netdev_printk(KERN_DEBUG, dev,
876 			      "failed to linearize skb with tiny unaligned fragment\n");
877 		return NETDEV_TX_BUSY;
878 	}
879 
880 	if (txq->tx_ring_size - txq->tx_desc_count < MAX_SKB_FRAGS + 1) {
881 		if (net_ratelimit())
882 			netdev_err(dev, "tx queue full?!\n");
883 		kfree_skb(skb);
884 		return NETDEV_TX_OK;
885 	}
886 
887 	length = skb->len;
888 
889 	if (!txq_submit_skb(txq, skb)) {
890 		int entries_left;
891 
892 		txq->tx_bytes += length;
893 		txq->tx_packets++;
894 
895 		entries_left = txq->tx_ring_size - txq->tx_desc_count;
896 		if (entries_left < MAX_SKB_FRAGS + 1)
897 			netif_tx_stop_queue(nq);
898 	}
899 
900 	return NETDEV_TX_OK;
901 }
902 
903 
904 /* tx napi ******************************************************************/
905 static void txq_kick(struct tx_queue *txq)
906 {
907 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
908 	struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
909 	u32 hw_desc_ptr;
910 	u32 expected_ptr;
911 
912 	__netif_tx_lock(nq, smp_processor_id());
913 
914 	if (rdlp(mp, TXQ_COMMAND) & (1 << txq->index))
915 		goto out;
916 
917 	hw_desc_ptr = rdlp(mp, TXQ_CURRENT_DESC_PTR(txq->index));
918 	expected_ptr = (u32)txq->tx_desc_dma +
919 				txq->tx_curr_desc * sizeof(struct tx_desc);
920 
921 	if (hw_desc_ptr != expected_ptr)
922 		txq_enable(txq);
923 
924 out:
925 	__netif_tx_unlock(nq);
926 
927 	mp->work_tx_end &= ~(1 << txq->index);
928 }
929 
930 static int txq_reclaim(struct tx_queue *txq, int budget, int force)
931 {
932 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
933 	struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
934 	int reclaimed;
935 
936 	__netif_tx_lock(nq, smp_processor_id());
937 
938 	reclaimed = 0;
939 	while (reclaimed < budget && txq->tx_desc_count > 0) {
940 		int tx_index;
941 		struct tx_desc *desc;
942 		u32 cmd_sts;
943 		struct sk_buff *skb;
944 
945 		tx_index = txq->tx_used_desc;
946 		desc = &txq->tx_desc_area[tx_index];
947 		cmd_sts = desc->cmd_sts;
948 
949 		if (cmd_sts & BUFFER_OWNED_BY_DMA) {
950 			if (!force)
951 				break;
952 			desc->cmd_sts = cmd_sts & ~BUFFER_OWNED_BY_DMA;
953 		}
954 
955 		txq->tx_used_desc = tx_index + 1;
956 		if (txq->tx_used_desc == txq->tx_ring_size)
957 			txq->tx_used_desc = 0;
958 
959 		reclaimed++;
960 		txq->tx_desc_count--;
961 
962 		skb = NULL;
963 		if (cmd_sts & TX_LAST_DESC)
964 			skb = __skb_dequeue(&txq->tx_skb);
965 
966 		if (cmd_sts & ERROR_SUMMARY) {
967 			netdev_info(mp->dev, "tx error\n");
968 			mp->dev->stats.tx_errors++;
969 		}
970 
971 		if (cmd_sts & TX_FIRST_DESC) {
972 			dma_unmap_single(mp->dev->dev.parent, desc->buf_ptr,
973 					 desc->byte_cnt, DMA_TO_DEVICE);
974 		} else {
975 			dma_unmap_page(mp->dev->dev.parent, desc->buf_ptr,
976 				       desc->byte_cnt, DMA_TO_DEVICE);
977 		}
978 
979 		if (skb != NULL) {
980 			if (skb_queue_len(&mp->rx_recycle) <
981 					mp->rx_ring_size &&
982 			    skb_recycle_check(skb, mp->skb_size))
983 				__skb_queue_head(&mp->rx_recycle, skb);
984 			else
985 				dev_kfree_skb(skb);
986 		}
987 	}
988 
989 	__netif_tx_unlock(nq);
990 
991 	if (reclaimed < budget)
992 		mp->work_tx &= ~(1 << txq->index);
993 
994 	return reclaimed;
995 }
996 
997 
998 /* tx rate control **********************************************************/
999 /*
1000  * Set total maximum TX rate (shared by all TX queues for this port)
1001  * to 'rate' bits per second, with a maximum burst of 'burst' bytes.
1002  */
1003 static void tx_set_rate(struct mv643xx_eth_private *mp, int rate, int burst)
1004 {
1005 	int token_rate;
1006 	int mtu;
1007 	int bucket_size;
1008 
1009 	token_rate = ((rate / 1000) * 64) / (mp->shared->t_clk / 1000);
1010 	if (token_rate > 1023)
1011 		token_rate = 1023;
1012 
1013 	mtu = (mp->dev->mtu + 255) >> 8;
1014 	if (mtu > 63)
1015 		mtu = 63;
1016 
1017 	bucket_size = (burst + 255) >> 8;
1018 	if (bucket_size > 65535)
1019 		bucket_size = 65535;
1020 
1021 	switch (mp->shared->tx_bw_control) {
1022 	case TX_BW_CONTROL_OLD_LAYOUT:
1023 		wrlp(mp, TX_BW_RATE, token_rate);
1024 		wrlp(mp, TX_BW_MTU, mtu);
1025 		wrlp(mp, TX_BW_BURST, bucket_size);
1026 		break;
1027 	case TX_BW_CONTROL_NEW_LAYOUT:
1028 		wrlp(mp, TX_BW_RATE_MOVED, token_rate);
1029 		wrlp(mp, TX_BW_MTU_MOVED, mtu);
1030 		wrlp(mp, TX_BW_BURST_MOVED, bucket_size);
1031 		break;
1032 	}
1033 }
1034 
1035 static void txq_set_rate(struct tx_queue *txq, int rate, int burst)
1036 {
1037 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
1038 	int token_rate;
1039 	int bucket_size;
1040 
1041 	token_rate = ((rate / 1000) * 64) / (mp->shared->t_clk / 1000);
1042 	if (token_rate > 1023)
1043 		token_rate = 1023;
1044 
1045 	bucket_size = (burst + 255) >> 8;
1046 	if (bucket_size > 65535)
1047 		bucket_size = 65535;
1048 
1049 	wrlp(mp, TXQ_BW_TOKENS(txq->index), token_rate << 14);
1050 	wrlp(mp, TXQ_BW_CONF(txq->index), (bucket_size << 10) | token_rate);
1051 }
1052 
1053 static void txq_set_fixed_prio_mode(struct tx_queue *txq)
1054 {
1055 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
1056 	int off;
1057 	u32 val;
1058 
1059 	/*
1060 	 * Turn on fixed priority mode.
1061 	 */
1062 	off = 0;
1063 	switch (mp->shared->tx_bw_control) {
1064 	case TX_BW_CONTROL_OLD_LAYOUT:
1065 		off = TXQ_FIX_PRIO_CONF;
1066 		break;
1067 	case TX_BW_CONTROL_NEW_LAYOUT:
1068 		off = TXQ_FIX_PRIO_CONF_MOVED;
1069 		break;
1070 	}
1071 
1072 	if (off) {
1073 		val = rdlp(mp, off);
1074 		val |= 1 << txq->index;
1075 		wrlp(mp, off, val);
1076 	}
1077 }
1078 
1079 
1080 /* mii management interface *************************************************/
1081 static irqreturn_t mv643xx_eth_err_irq(int irq, void *dev_id)
1082 {
1083 	struct mv643xx_eth_shared_private *msp = dev_id;
1084 
1085 	if (readl(msp->base + ERR_INT_CAUSE) & ERR_INT_SMI_DONE) {
1086 		writel(~ERR_INT_SMI_DONE, msp->base + ERR_INT_CAUSE);
1087 		wake_up(&msp->smi_busy_wait);
1088 		return IRQ_HANDLED;
1089 	}
1090 
1091 	return IRQ_NONE;
1092 }
1093 
1094 static int smi_is_done(struct mv643xx_eth_shared_private *msp)
1095 {
1096 	return !(readl(msp->base + SMI_REG) & SMI_BUSY);
1097 }
1098 
1099 static int smi_wait_ready(struct mv643xx_eth_shared_private *msp)
1100 {
1101 	if (msp->err_interrupt == NO_IRQ) {
1102 		int i;
1103 
1104 		for (i = 0; !smi_is_done(msp); i++) {
1105 			if (i == 10)
1106 				return -ETIMEDOUT;
1107 			msleep(10);
1108 		}
1109 
1110 		return 0;
1111 	}
1112 
1113 	if (!smi_is_done(msp)) {
1114 		wait_event_timeout(msp->smi_busy_wait, smi_is_done(msp),
1115 				   msecs_to_jiffies(100));
1116 		if (!smi_is_done(msp))
1117 			return -ETIMEDOUT;
1118 	}
1119 
1120 	return 0;
1121 }
1122 
1123 static int smi_bus_read(struct mii_bus *bus, int addr, int reg)
1124 {
1125 	struct mv643xx_eth_shared_private *msp = bus->priv;
1126 	void __iomem *smi_reg = msp->base + SMI_REG;
1127 	int ret;
1128 
1129 	if (smi_wait_ready(msp)) {
1130 		pr_warn("SMI bus busy timeout\n");
1131 		return -ETIMEDOUT;
1132 	}
1133 
1134 	writel(SMI_OPCODE_READ | (reg << 21) | (addr << 16), smi_reg);
1135 
1136 	if (smi_wait_ready(msp)) {
1137 		pr_warn("SMI bus busy timeout\n");
1138 		return -ETIMEDOUT;
1139 	}
1140 
1141 	ret = readl(smi_reg);
1142 	if (!(ret & SMI_READ_VALID)) {
1143 		pr_warn("SMI bus read not valid\n");
1144 		return -ENODEV;
1145 	}
1146 
1147 	return ret & 0xffff;
1148 }
1149 
1150 static int smi_bus_write(struct mii_bus *bus, int addr, int reg, u16 val)
1151 {
1152 	struct mv643xx_eth_shared_private *msp = bus->priv;
1153 	void __iomem *smi_reg = msp->base + SMI_REG;
1154 
1155 	if (smi_wait_ready(msp)) {
1156 		pr_warn("SMI bus busy timeout\n");
1157 		return -ETIMEDOUT;
1158 	}
1159 
1160 	writel(SMI_OPCODE_WRITE | (reg << 21) |
1161 		(addr << 16) | (val & 0xffff), smi_reg);
1162 
1163 	if (smi_wait_ready(msp)) {
1164 		pr_warn("SMI bus busy timeout\n");
1165 		return -ETIMEDOUT;
1166 	}
1167 
1168 	return 0;
1169 }
1170 
1171 
1172 /* statistics ***************************************************************/
1173 static struct net_device_stats *mv643xx_eth_get_stats(struct net_device *dev)
1174 {
1175 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1176 	struct net_device_stats *stats = &dev->stats;
1177 	unsigned long tx_packets = 0;
1178 	unsigned long tx_bytes = 0;
1179 	unsigned long tx_dropped = 0;
1180 	int i;
1181 
1182 	for (i = 0; i < mp->txq_count; i++) {
1183 		struct tx_queue *txq = mp->txq + i;
1184 
1185 		tx_packets += txq->tx_packets;
1186 		tx_bytes += txq->tx_bytes;
1187 		tx_dropped += txq->tx_dropped;
1188 	}
1189 
1190 	stats->tx_packets = tx_packets;
1191 	stats->tx_bytes = tx_bytes;
1192 	stats->tx_dropped = tx_dropped;
1193 
1194 	return stats;
1195 }
1196 
1197 static void mv643xx_eth_grab_lro_stats(struct mv643xx_eth_private *mp)
1198 {
1199 	u32 lro_aggregated = 0;
1200 	u32 lro_flushed = 0;
1201 	u32 lro_no_desc = 0;
1202 	int i;
1203 
1204 	for (i = 0; i < mp->rxq_count; i++) {
1205 		struct rx_queue *rxq = mp->rxq + i;
1206 
1207 		lro_aggregated += rxq->lro_mgr.stats.aggregated;
1208 		lro_flushed += rxq->lro_mgr.stats.flushed;
1209 		lro_no_desc += rxq->lro_mgr.stats.no_desc;
1210 	}
1211 
1212 	mp->lro_counters.lro_aggregated = lro_aggregated;
1213 	mp->lro_counters.lro_flushed = lro_flushed;
1214 	mp->lro_counters.lro_no_desc = lro_no_desc;
1215 }
1216 
1217 static inline u32 mib_read(struct mv643xx_eth_private *mp, int offset)
1218 {
1219 	return rdl(mp, MIB_COUNTERS(mp->port_num) + offset);
1220 }
1221 
1222 static void mib_counters_clear(struct mv643xx_eth_private *mp)
1223 {
1224 	int i;
1225 
1226 	for (i = 0; i < 0x80; i += 4)
1227 		mib_read(mp, i);
1228 }
1229 
1230 static void mib_counters_update(struct mv643xx_eth_private *mp)
1231 {
1232 	struct mib_counters *p = &mp->mib_counters;
1233 
1234 	spin_lock_bh(&mp->mib_counters_lock);
1235 	p->good_octets_received += mib_read(mp, 0x00);
1236 	p->bad_octets_received += mib_read(mp, 0x08);
1237 	p->internal_mac_transmit_err += mib_read(mp, 0x0c);
1238 	p->good_frames_received += mib_read(mp, 0x10);
1239 	p->bad_frames_received += mib_read(mp, 0x14);
1240 	p->broadcast_frames_received += mib_read(mp, 0x18);
1241 	p->multicast_frames_received += mib_read(mp, 0x1c);
1242 	p->frames_64_octets += mib_read(mp, 0x20);
1243 	p->frames_65_to_127_octets += mib_read(mp, 0x24);
1244 	p->frames_128_to_255_octets += mib_read(mp, 0x28);
1245 	p->frames_256_to_511_octets += mib_read(mp, 0x2c);
1246 	p->frames_512_to_1023_octets += mib_read(mp, 0x30);
1247 	p->frames_1024_to_max_octets += mib_read(mp, 0x34);
1248 	p->good_octets_sent += mib_read(mp, 0x38);
1249 	p->good_frames_sent += mib_read(mp, 0x40);
1250 	p->excessive_collision += mib_read(mp, 0x44);
1251 	p->multicast_frames_sent += mib_read(mp, 0x48);
1252 	p->broadcast_frames_sent += mib_read(mp, 0x4c);
1253 	p->unrec_mac_control_received += mib_read(mp, 0x50);
1254 	p->fc_sent += mib_read(mp, 0x54);
1255 	p->good_fc_received += mib_read(mp, 0x58);
1256 	p->bad_fc_received += mib_read(mp, 0x5c);
1257 	p->undersize_received += mib_read(mp, 0x60);
1258 	p->fragments_received += mib_read(mp, 0x64);
1259 	p->oversize_received += mib_read(mp, 0x68);
1260 	p->jabber_received += mib_read(mp, 0x6c);
1261 	p->mac_receive_error += mib_read(mp, 0x70);
1262 	p->bad_crc_event += mib_read(mp, 0x74);
1263 	p->collision += mib_read(mp, 0x78);
1264 	p->late_collision += mib_read(mp, 0x7c);
1265 	spin_unlock_bh(&mp->mib_counters_lock);
1266 
1267 	mod_timer(&mp->mib_counters_timer, jiffies + 30 * HZ);
1268 }
1269 
1270 static void mib_counters_timer_wrapper(unsigned long _mp)
1271 {
1272 	struct mv643xx_eth_private *mp = (void *)_mp;
1273 
1274 	mib_counters_update(mp);
1275 }
1276 
1277 
1278 /* interrupt coalescing *****************************************************/
1279 /*
1280  * Hardware coalescing parameters are set in units of 64 t_clk
1281  * cycles.  I.e.:
1282  *
1283  *	coal_delay_in_usec = 64000000 * register_value / t_clk_rate
1284  *
1285  *	register_value = coal_delay_in_usec * t_clk_rate / 64000000
1286  *
1287  * In the ->set*() methods, we round the computed register value
1288  * to the nearest integer.
1289  */
1290 static unsigned int get_rx_coal(struct mv643xx_eth_private *mp)
1291 {
1292 	u32 val = rdlp(mp, SDMA_CONFIG);
1293 	u64 temp;
1294 
1295 	if (mp->shared->extended_rx_coal_limit)
1296 		temp = ((val & 0x02000000) >> 10) | ((val & 0x003fff80) >> 7);
1297 	else
1298 		temp = (val & 0x003fff00) >> 8;
1299 
1300 	temp *= 64000000;
1301 	do_div(temp, mp->shared->t_clk);
1302 
1303 	return (unsigned int)temp;
1304 }
1305 
1306 static void set_rx_coal(struct mv643xx_eth_private *mp, unsigned int usec)
1307 {
1308 	u64 temp;
1309 	u32 val;
1310 
1311 	temp = (u64)usec * mp->shared->t_clk;
1312 	temp += 31999999;
1313 	do_div(temp, 64000000);
1314 
1315 	val = rdlp(mp, SDMA_CONFIG);
1316 	if (mp->shared->extended_rx_coal_limit) {
1317 		if (temp > 0xffff)
1318 			temp = 0xffff;
1319 		val &= ~0x023fff80;
1320 		val |= (temp & 0x8000) << 10;
1321 		val |= (temp & 0x7fff) << 7;
1322 	} else {
1323 		if (temp > 0x3fff)
1324 			temp = 0x3fff;
1325 		val &= ~0x003fff00;
1326 		val |= (temp & 0x3fff) << 8;
1327 	}
1328 	wrlp(mp, SDMA_CONFIG, val);
1329 }
1330 
1331 static unsigned int get_tx_coal(struct mv643xx_eth_private *mp)
1332 {
1333 	u64 temp;
1334 
1335 	temp = (rdlp(mp, TX_FIFO_URGENT_THRESHOLD) & 0x3fff0) >> 4;
1336 	temp *= 64000000;
1337 	do_div(temp, mp->shared->t_clk);
1338 
1339 	return (unsigned int)temp;
1340 }
1341 
1342 static void set_tx_coal(struct mv643xx_eth_private *mp, unsigned int usec)
1343 {
1344 	u64 temp;
1345 
1346 	temp = (u64)usec * mp->shared->t_clk;
1347 	temp += 31999999;
1348 	do_div(temp, 64000000);
1349 
1350 	if (temp > 0x3fff)
1351 		temp = 0x3fff;
1352 
1353 	wrlp(mp, TX_FIFO_URGENT_THRESHOLD, temp << 4);
1354 }
1355 
1356 
1357 /* ethtool ******************************************************************/
1358 struct mv643xx_eth_stats {
1359 	char stat_string[ETH_GSTRING_LEN];
1360 	int sizeof_stat;
1361 	int netdev_off;
1362 	int mp_off;
1363 };
1364 
1365 #define SSTAT(m)						\
1366 	{ #m, FIELD_SIZEOF(struct net_device_stats, m),		\
1367 	  offsetof(struct net_device, stats.m), -1 }
1368 
1369 #define MIBSTAT(m)						\
1370 	{ #m, FIELD_SIZEOF(struct mib_counters, m),		\
1371 	  -1, offsetof(struct mv643xx_eth_private, mib_counters.m) }
1372 
1373 #define LROSTAT(m)						\
1374 	{ #m, FIELD_SIZEOF(struct lro_counters, m),		\
1375 	  -1, offsetof(struct mv643xx_eth_private, lro_counters.m) }
1376 
1377 static const struct mv643xx_eth_stats mv643xx_eth_stats[] = {
1378 	SSTAT(rx_packets),
1379 	SSTAT(tx_packets),
1380 	SSTAT(rx_bytes),
1381 	SSTAT(tx_bytes),
1382 	SSTAT(rx_errors),
1383 	SSTAT(tx_errors),
1384 	SSTAT(rx_dropped),
1385 	SSTAT(tx_dropped),
1386 	MIBSTAT(good_octets_received),
1387 	MIBSTAT(bad_octets_received),
1388 	MIBSTAT(internal_mac_transmit_err),
1389 	MIBSTAT(good_frames_received),
1390 	MIBSTAT(bad_frames_received),
1391 	MIBSTAT(broadcast_frames_received),
1392 	MIBSTAT(multicast_frames_received),
1393 	MIBSTAT(frames_64_octets),
1394 	MIBSTAT(frames_65_to_127_octets),
1395 	MIBSTAT(frames_128_to_255_octets),
1396 	MIBSTAT(frames_256_to_511_octets),
1397 	MIBSTAT(frames_512_to_1023_octets),
1398 	MIBSTAT(frames_1024_to_max_octets),
1399 	MIBSTAT(good_octets_sent),
1400 	MIBSTAT(good_frames_sent),
1401 	MIBSTAT(excessive_collision),
1402 	MIBSTAT(multicast_frames_sent),
1403 	MIBSTAT(broadcast_frames_sent),
1404 	MIBSTAT(unrec_mac_control_received),
1405 	MIBSTAT(fc_sent),
1406 	MIBSTAT(good_fc_received),
1407 	MIBSTAT(bad_fc_received),
1408 	MIBSTAT(undersize_received),
1409 	MIBSTAT(fragments_received),
1410 	MIBSTAT(oversize_received),
1411 	MIBSTAT(jabber_received),
1412 	MIBSTAT(mac_receive_error),
1413 	MIBSTAT(bad_crc_event),
1414 	MIBSTAT(collision),
1415 	MIBSTAT(late_collision),
1416 	LROSTAT(lro_aggregated),
1417 	LROSTAT(lro_flushed),
1418 	LROSTAT(lro_no_desc),
1419 };
1420 
1421 static int
1422 mv643xx_eth_get_settings_phy(struct mv643xx_eth_private *mp,
1423 			     struct ethtool_cmd *cmd)
1424 {
1425 	int err;
1426 
1427 	err = phy_read_status(mp->phy);
1428 	if (err == 0)
1429 		err = phy_ethtool_gset(mp->phy, cmd);
1430 
1431 	/*
1432 	 * The MAC does not support 1000baseT_Half.
1433 	 */
1434 	cmd->supported &= ~SUPPORTED_1000baseT_Half;
1435 	cmd->advertising &= ~ADVERTISED_1000baseT_Half;
1436 
1437 	return err;
1438 }
1439 
1440 static int
1441 mv643xx_eth_get_settings_phyless(struct mv643xx_eth_private *mp,
1442 				 struct ethtool_cmd *cmd)
1443 {
1444 	u32 port_status;
1445 
1446 	port_status = rdlp(mp, PORT_STATUS);
1447 
1448 	cmd->supported = SUPPORTED_MII;
1449 	cmd->advertising = ADVERTISED_MII;
1450 	switch (port_status & PORT_SPEED_MASK) {
1451 	case PORT_SPEED_10:
1452 		ethtool_cmd_speed_set(cmd, SPEED_10);
1453 		break;
1454 	case PORT_SPEED_100:
1455 		ethtool_cmd_speed_set(cmd, SPEED_100);
1456 		break;
1457 	case PORT_SPEED_1000:
1458 		ethtool_cmd_speed_set(cmd, SPEED_1000);
1459 		break;
1460 	default:
1461 		cmd->speed = -1;
1462 		break;
1463 	}
1464 	cmd->duplex = (port_status & FULL_DUPLEX) ? DUPLEX_FULL : DUPLEX_HALF;
1465 	cmd->port = PORT_MII;
1466 	cmd->phy_address = 0;
1467 	cmd->transceiver = XCVR_INTERNAL;
1468 	cmd->autoneg = AUTONEG_DISABLE;
1469 	cmd->maxtxpkt = 1;
1470 	cmd->maxrxpkt = 1;
1471 
1472 	return 0;
1473 }
1474 
1475 static int
1476 mv643xx_eth_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1477 {
1478 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1479 
1480 	if (mp->phy != NULL)
1481 		return mv643xx_eth_get_settings_phy(mp, cmd);
1482 	else
1483 		return mv643xx_eth_get_settings_phyless(mp, cmd);
1484 }
1485 
1486 static int
1487 mv643xx_eth_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1488 {
1489 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1490 
1491 	if (mp->phy == NULL)
1492 		return -EINVAL;
1493 
1494 	/*
1495 	 * The MAC does not support 1000baseT_Half.
1496 	 */
1497 	cmd->advertising &= ~ADVERTISED_1000baseT_Half;
1498 
1499 	return phy_ethtool_sset(mp->phy, cmd);
1500 }
1501 
1502 static void mv643xx_eth_get_drvinfo(struct net_device *dev,
1503 				    struct ethtool_drvinfo *drvinfo)
1504 {
1505 	strlcpy(drvinfo->driver, mv643xx_eth_driver_name,
1506 		sizeof(drvinfo->driver));
1507 	strlcpy(drvinfo->version, mv643xx_eth_driver_version,
1508 		sizeof(drvinfo->version));
1509 	strlcpy(drvinfo->fw_version, "N/A", sizeof(drvinfo->fw_version));
1510 	strlcpy(drvinfo->bus_info, "platform", sizeof(drvinfo->bus_info));
1511 	drvinfo->n_stats = ARRAY_SIZE(mv643xx_eth_stats);
1512 }
1513 
1514 static int mv643xx_eth_nway_reset(struct net_device *dev)
1515 {
1516 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1517 
1518 	if (mp->phy == NULL)
1519 		return -EINVAL;
1520 
1521 	return genphy_restart_aneg(mp->phy);
1522 }
1523 
1524 static int
1525 mv643xx_eth_get_coalesce(struct net_device *dev, struct ethtool_coalesce *ec)
1526 {
1527 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1528 
1529 	ec->rx_coalesce_usecs = get_rx_coal(mp);
1530 	ec->tx_coalesce_usecs = get_tx_coal(mp);
1531 
1532 	return 0;
1533 }
1534 
1535 static int
1536 mv643xx_eth_set_coalesce(struct net_device *dev, struct ethtool_coalesce *ec)
1537 {
1538 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1539 
1540 	set_rx_coal(mp, ec->rx_coalesce_usecs);
1541 	set_tx_coal(mp, ec->tx_coalesce_usecs);
1542 
1543 	return 0;
1544 }
1545 
1546 static void
1547 mv643xx_eth_get_ringparam(struct net_device *dev, struct ethtool_ringparam *er)
1548 {
1549 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1550 
1551 	er->rx_max_pending = 4096;
1552 	er->tx_max_pending = 4096;
1553 
1554 	er->rx_pending = mp->rx_ring_size;
1555 	er->tx_pending = mp->tx_ring_size;
1556 }
1557 
1558 static int
1559 mv643xx_eth_set_ringparam(struct net_device *dev, struct ethtool_ringparam *er)
1560 {
1561 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1562 
1563 	if (er->rx_mini_pending || er->rx_jumbo_pending)
1564 		return -EINVAL;
1565 
1566 	mp->rx_ring_size = er->rx_pending < 4096 ? er->rx_pending : 4096;
1567 	mp->tx_ring_size = er->tx_pending < 4096 ? er->tx_pending : 4096;
1568 
1569 	if (netif_running(dev)) {
1570 		mv643xx_eth_stop(dev);
1571 		if (mv643xx_eth_open(dev)) {
1572 			netdev_err(dev,
1573 				   "fatal error on re-opening device after ring param change\n");
1574 			return -ENOMEM;
1575 		}
1576 	}
1577 
1578 	return 0;
1579 }
1580 
1581 
1582 static int
1583 mv643xx_eth_set_features(struct net_device *dev, netdev_features_t features)
1584 {
1585 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1586 	bool rx_csum = features & NETIF_F_RXCSUM;
1587 
1588 	wrlp(mp, PORT_CONFIG, rx_csum ? 0x02000000 : 0x00000000);
1589 
1590 	return 0;
1591 }
1592 
1593 static void mv643xx_eth_get_strings(struct net_device *dev,
1594 				    uint32_t stringset, uint8_t *data)
1595 {
1596 	int i;
1597 
1598 	if (stringset == ETH_SS_STATS) {
1599 		for (i = 0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
1600 			memcpy(data + i * ETH_GSTRING_LEN,
1601 				mv643xx_eth_stats[i].stat_string,
1602 				ETH_GSTRING_LEN);
1603 		}
1604 	}
1605 }
1606 
1607 static void mv643xx_eth_get_ethtool_stats(struct net_device *dev,
1608 					  struct ethtool_stats *stats,
1609 					  uint64_t *data)
1610 {
1611 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1612 	int i;
1613 
1614 	mv643xx_eth_get_stats(dev);
1615 	mib_counters_update(mp);
1616 	mv643xx_eth_grab_lro_stats(mp);
1617 
1618 	for (i = 0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
1619 		const struct mv643xx_eth_stats *stat;
1620 		void *p;
1621 
1622 		stat = mv643xx_eth_stats + i;
1623 
1624 		if (stat->netdev_off >= 0)
1625 			p = ((void *)mp->dev) + stat->netdev_off;
1626 		else
1627 			p = ((void *)mp) + stat->mp_off;
1628 
1629 		data[i] = (stat->sizeof_stat == 8) ?
1630 				*(uint64_t *)p : *(uint32_t *)p;
1631 	}
1632 }
1633 
1634 static int mv643xx_eth_get_sset_count(struct net_device *dev, int sset)
1635 {
1636 	if (sset == ETH_SS_STATS)
1637 		return ARRAY_SIZE(mv643xx_eth_stats);
1638 
1639 	return -EOPNOTSUPP;
1640 }
1641 
1642 static const struct ethtool_ops mv643xx_eth_ethtool_ops = {
1643 	.get_settings		= mv643xx_eth_get_settings,
1644 	.set_settings		= mv643xx_eth_set_settings,
1645 	.get_drvinfo		= mv643xx_eth_get_drvinfo,
1646 	.nway_reset		= mv643xx_eth_nway_reset,
1647 	.get_link		= ethtool_op_get_link,
1648 	.get_coalesce		= mv643xx_eth_get_coalesce,
1649 	.set_coalesce		= mv643xx_eth_set_coalesce,
1650 	.get_ringparam		= mv643xx_eth_get_ringparam,
1651 	.set_ringparam		= mv643xx_eth_set_ringparam,
1652 	.get_strings		= mv643xx_eth_get_strings,
1653 	.get_ethtool_stats	= mv643xx_eth_get_ethtool_stats,
1654 	.get_sset_count		= mv643xx_eth_get_sset_count,
1655 };
1656 
1657 
1658 /* address handling *********************************************************/
1659 static void uc_addr_get(struct mv643xx_eth_private *mp, unsigned char *addr)
1660 {
1661 	unsigned int mac_h = rdlp(mp, MAC_ADDR_HIGH);
1662 	unsigned int mac_l = rdlp(mp, MAC_ADDR_LOW);
1663 
1664 	addr[0] = (mac_h >> 24) & 0xff;
1665 	addr[1] = (mac_h >> 16) & 0xff;
1666 	addr[2] = (mac_h >> 8) & 0xff;
1667 	addr[3] = mac_h & 0xff;
1668 	addr[4] = (mac_l >> 8) & 0xff;
1669 	addr[5] = mac_l & 0xff;
1670 }
1671 
1672 static void uc_addr_set(struct mv643xx_eth_private *mp, unsigned char *addr)
1673 {
1674 	wrlp(mp, MAC_ADDR_HIGH,
1675 		(addr[0] << 24) | (addr[1] << 16) | (addr[2] << 8) | addr[3]);
1676 	wrlp(mp, MAC_ADDR_LOW, (addr[4] << 8) | addr[5]);
1677 }
1678 
1679 static u32 uc_addr_filter_mask(struct net_device *dev)
1680 {
1681 	struct netdev_hw_addr *ha;
1682 	u32 nibbles;
1683 
1684 	if (dev->flags & IFF_PROMISC)
1685 		return 0;
1686 
1687 	nibbles = 1 << (dev->dev_addr[5] & 0x0f);
1688 	netdev_for_each_uc_addr(ha, dev) {
1689 		if (memcmp(dev->dev_addr, ha->addr, 5))
1690 			return 0;
1691 		if ((dev->dev_addr[5] ^ ha->addr[5]) & 0xf0)
1692 			return 0;
1693 
1694 		nibbles |= 1 << (ha->addr[5] & 0x0f);
1695 	}
1696 
1697 	return nibbles;
1698 }
1699 
1700 static void mv643xx_eth_program_unicast_filter(struct net_device *dev)
1701 {
1702 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1703 	u32 port_config;
1704 	u32 nibbles;
1705 	int i;
1706 
1707 	uc_addr_set(mp, dev->dev_addr);
1708 
1709 	port_config = rdlp(mp, PORT_CONFIG) & ~UNICAST_PROMISCUOUS_MODE;
1710 
1711 	nibbles = uc_addr_filter_mask(dev);
1712 	if (!nibbles) {
1713 		port_config |= UNICAST_PROMISCUOUS_MODE;
1714 		nibbles = 0xffff;
1715 	}
1716 
1717 	for (i = 0; i < 16; i += 4) {
1718 		int off = UNICAST_TABLE(mp->port_num) + i;
1719 		u32 v;
1720 
1721 		v = 0;
1722 		if (nibbles & 1)
1723 			v |= 0x00000001;
1724 		if (nibbles & 2)
1725 			v |= 0x00000100;
1726 		if (nibbles & 4)
1727 			v |= 0x00010000;
1728 		if (nibbles & 8)
1729 			v |= 0x01000000;
1730 		nibbles >>= 4;
1731 
1732 		wrl(mp, off, v);
1733 	}
1734 
1735 	wrlp(mp, PORT_CONFIG, port_config);
1736 }
1737 
1738 static int addr_crc(unsigned char *addr)
1739 {
1740 	int crc = 0;
1741 	int i;
1742 
1743 	for (i = 0; i < 6; i++) {
1744 		int j;
1745 
1746 		crc = (crc ^ addr[i]) << 8;
1747 		for (j = 7; j >= 0; j--) {
1748 			if (crc & (0x100 << j))
1749 				crc ^= 0x107 << j;
1750 		}
1751 	}
1752 
1753 	return crc;
1754 }
1755 
1756 static void mv643xx_eth_program_multicast_filter(struct net_device *dev)
1757 {
1758 	struct mv643xx_eth_private *mp = netdev_priv(dev);
1759 	u32 *mc_spec;
1760 	u32 *mc_other;
1761 	struct netdev_hw_addr *ha;
1762 	int i;
1763 
1764 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
1765 		int port_num;
1766 		u32 accept;
1767 
1768 oom:
1769 		port_num = mp->port_num;
1770 		accept = 0x01010101;
1771 		for (i = 0; i < 0x100; i += 4) {
1772 			wrl(mp, SPECIAL_MCAST_TABLE(port_num) + i, accept);
1773 			wrl(mp, OTHER_MCAST_TABLE(port_num) + i, accept);
1774 		}
1775 		return;
1776 	}
1777 
1778 	mc_spec = kmalloc(0x200, GFP_ATOMIC);
1779 	if (mc_spec == NULL)
1780 		goto oom;
1781 	mc_other = mc_spec + (0x100 >> 2);
1782 
1783 	memset(mc_spec, 0, 0x100);
1784 	memset(mc_other, 0, 0x100);
1785 
1786 	netdev_for_each_mc_addr(ha, dev) {
1787 		u8 *a = ha->addr;
1788 		u32 *table;
1789 		int entry;
1790 
1791 		if (memcmp(a, "\x01\x00\x5e\x00\x00", 5) == 0) {
1792 			table = mc_spec;
1793 			entry = a[5];
1794 		} else {
1795 			table = mc_other;
1796 			entry = addr_crc(a);
1797 		}
1798 
1799 		table[entry >> 2] |= 1 << (8 * (entry & 3));
1800 	}
1801 
1802 	for (i = 0; i < 0x100; i += 4) {
1803 		wrl(mp, SPECIAL_MCAST_TABLE(mp->port_num) + i, mc_spec[i >> 2]);
1804 		wrl(mp, OTHER_MCAST_TABLE(mp->port_num) + i, mc_other[i >> 2]);
1805 	}
1806 
1807 	kfree(mc_spec);
1808 }
1809 
1810 static void mv643xx_eth_set_rx_mode(struct net_device *dev)
1811 {
1812 	mv643xx_eth_program_unicast_filter(dev);
1813 	mv643xx_eth_program_multicast_filter(dev);
1814 }
1815 
1816 static int mv643xx_eth_set_mac_address(struct net_device *dev, void *addr)
1817 {
1818 	struct sockaddr *sa = addr;
1819 
1820 	if (!is_valid_ether_addr(sa->sa_data))
1821 		return -EINVAL;
1822 
1823 	memcpy(dev->dev_addr, sa->sa_data, ETH_ALEN);
1824 
1825 	netif_addr_lock_bh(dev);
1826 	mv643xx_eth_program_unicast_filter(dev);
1827 	netif_addr_unlock_bh(dev);
1828 
1829 	return 0;
1830 }
1831 
1832 
1833 /* rx/tx queue initialisation ***********************************************/
1834 static int rxq_init(struct mv643xx_eth_private *mp, int index)
1835 {
1836 	struct rx_queue *rxq = mp->rxq + index;
1837 	struct rx_desc *rx_desc;
1838 	int size;
1839 	int i;
1840 
1841 	rxq->index = index;
1842 
1843 	rxq->rx_ring_size = mp->rx_ring_size;
1844 
1845 	rxq->rx_desc_count = 0;
1846 	rxq->rx_curr_desc = 0;
1847 	rxq->rx_used_desc = 0;
1848 
1849 	size = rxq->rx_ring_size * sizeof(struct rx_desc);
1850 
1851 	if (index == 0 && size <= mp->rx_desc_sram_size) {
1852 		rxq->rx_desc_area = ioremap(mp->rx_desc_sram_addr,
1853 						mp->rx_desc_sram_size);
1854 		rxq->rx_desc_dma = mp->rx_desc_sram_addr;
1855 	} else {
1856 		rxq->rx_desc_area = dma_alloc_coherent(mp->dev->dev.parent,
1857 						       size, &rxq->rx_desc_dma,
1858 						       GFP_KERNEL);
1859 	}
1860 
1861 	if (rxq->rx_desc_area == NULL) {
1862 		netdev_err(mp->dev,
1863 			   "can't allocate rx ring (%d bytes)\n", size);
1864 		goto out;
1865 	}
1866 	memset(rxq->rx_desc_area, 0, size);
1867 
1868 	rxq->rx_desc_area_size = size;
1869 	rxq->rx_skb = kmalloc(rxq->rx_ring_size * sizeof(*rxq->rx_skb),
1870 								GFP_KERNEL);
1871 	if (rxq->rx_skb == NULL) {
1872 		netdev_err(mp->dev, "can't allocate rx skb ring\n");
1873 		goto out_free;
1874 	}
1875 
1876 	rx_desc = (struct rx_desc *)rxq->rx_desc_area;
1877 	for (i = 0; i < rxq->rx_ring_size; i++) {
1878 		int nexti;
1879 
1880 		nexti = i + 1;
1881 		if (nexti == rxq->rx_ring_size)
1882 			nexti = 0;
1883 
1884 		rx_desc[i].next_desc_ptr = rxq->rx_desc_dma +
1885 					nexti * sizeof(struct rx_desc);
1886 	}
1887 
1888 	rxq->lro_mgr.dev = mp->dev;
1889 	memset(&rxq->lro_mgr.stats, 0, sizeof(rxq->lro_mgr.stats));
1890 	rxq->lro_mgr.features = LRO_F_NAPI;
1891 	rxq->lro_mgr.ip_summed = CHECKSUM_UNNECESSARY;
1892 	rxq->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1893 	rxq->lro_mgr.max_desc = ARRAY_SIZE(rxq->lro_arr);
1894 	rxq->lro_mgr.max_aggr = 32;
1895 	rxq->lro_mgr.frag_align_pad = 0;
1896 	rxq->lro_mgr.lro_arr = rxq->lro_arr;
1897 	rxq->lro_mgr.get_skb_header = mv643xx_get_skb_header;
1898 
1899 	memset(&rxq->lro_arr, 0, sizeof(rxq->lro_arr));
1900 
1901 	return 0;
1902 
1903 
1904 out_free:
1905 	if (index == 0 && size <= mp->rx_desc_sram_size)
1906 		iounmap(rxq->rx_desc_area);
1907 	else
1908 		dma_free_coherent(mp->dev->dev.parent, size,
1909 				  rxq->rx_desc_area,
1910 				  rxq->rx_desc_dma);
1911 
1912 out:
1913 	return -ENOMEM;
1914 }
1915 
1916 static void rxq_deinit(struct rx_queue *rxq)
1917 {
1918 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
1919 	int i;
1920 
1921 	rxq_disable(rxq);
1922 
1923 	for (i = 0; i < rxq->rx_ring_size; i++) {
1924 		if (rxq->rx_skb[i]) {
1925 			dev_kfree_skb(rxq->rx_skb[i]);
1926 			rxq->rx_desc_count--;
1927 		}
1928 	}
1929 
1930 	if (rxq->rx_desc_count) {
1931 		netdev_err(mp->dev, "error freeing rx ring -- %d skbs stuck\n",
1932 			   rxq->rx_desc_count);
1933 	}
1934 
1935 	if (rxq->index == 0 &&
1936 	    rxq->rx_desc_area_size <= mp->rx_desc_sram_size)
1937 		iounmap(rxq->rx_desc_area);
1938 	else
1939 		dma_free_coherent(mp->dev->dev.parent, rxq->rx_desc_area_size,
1940 				  rxq->rx_desc_area, rxq->rx_desc_dma);
1941 
1942 	kfree(rxq->rx_skb);
1943 }
1944 
1945 static int txq_init(struct mv643xx_eth_private *mp, int index)
1946 {
1947 	struct tx_queue *txq = mp->txq + index;
1948 	struct tx_desc *tx_desc;
1949 	int size;
1950 	int i;
1951 
1952 	txq->index = index;
1953 
1954 	txq->tx_ring_size = mp->tx_ring_size;
1955 
1956 	txq->tx_desc_count = 0;
1957 	txq->tx_curr_desc = 0;
1958 	txq->tx_used_desc = 0;
1959 
1960 	size = txq->tx_ring_size * sizeof(struct tx_desc);
1961 
1962 	if (index == 0 && size <= mp->tx_desc_sram_size) {
1963 		txq->tx_desc_area = ioremap(mp->tx_desc_sram_addr,
1964 						mp->tx_desc_sram_size);
1965 		txq->tx_desc_dma = mp->tx_desc_sram_addr;
1966 	} else {
1967 		txq->tx_desc_area = dma_alloc_coherent(mp->dev->dev.parent,
1968 						       size, &txq->tx_desc_dma,
1969 						       GFP_KERNEL);
1970 	}
1971 
1972 	if (txq->tx_desc_area == NULL) {
1973 		netdev_err(mp->dev,
1974 			   "can't allocate tx ring (%d bytes)\n", size);
1975 		return -ENOMEM;
1976 	}
1977 	memset(txq->tx_desc_area, 0, size);
1978 
1979 	txq->tx_desc_area_size = size;
1980 
1981 	tx_desc = (struct tx_desc *)txq->tx_desc_area;
1982 	for (i = 0; i < txq->tx_ring_size; i++) {
1983 		struct tx_desc *txd = tx_desc + i;
1984 		int nexti;
1985 
1986 		nexti = i + 1;
1987 		if (nexti == txq->tx_ring_size)
1988 			nexti = 0;
1989 
1990 		txd->cmd_sts = 0;
1991 		txd->next_desc_ptr = txq->tx_desc_dma +
1992 					nexti * sizeof(struct tx_desc);
1993 	}
1994 
1995 	skb_queue_head_init(&txq->tx_skb);
1996 
1997 	return 0;
1998 }
1999 
2000 static void txq_deinit(struct tx_queue *txq)
2001 {
2002 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
2003 
2004 	txq_disable(txq);
2005 	txq_reclaim(txq, txq->tx_ring_size, 1);
2006 
2007 	BUG_ON(txq->tx_used_desc != txq->tx_curr_desc);
2008 
2009 	if (txq->index == 0 &&
2010 	    txq->tx_desc_area_size <= mp->tx_desc_sram_size)
2011 		iounmap(txq->tx_desc_area);
2012 	else
2013 		dma_free_coherent(mp->dev->dev.parent, txq->tx_desc_area_size,
2014 				  txq->tx_desc_area, txq->tx_desc_dma);
2015 }
2016 
2017 
2018 /* netdev ops and related ***************************************************/
2019 static int mv643xx_eth_collect_events(struct mv643xx_eth_private *mp)
2020 {
2021 	u32 int_cause;
2022 	u32 int_cause_ext;
2023 
2024 	int_cause = rdlp(mp, INT_CAUSE) & mp->int_mask;
2025 	if (int_cause == 0)
2026 		return 0;
2027 
2028 	int_cause_ext = 0;
2029 	if (int_cause & INT_EXT) {
2030 		int_cause &= ~INT_EXT;
2031 		int_cause_ext = rdlp(mp, INT_CAUSE_EXT);
2032 	}
2033 
2034 	if (int_cause) {
2035 		wrlp(mp, INT_CAUSE, ~int_cause);
2036 		mp->work_tx_end |= ((int_cause & INT_TX_END) >> 19) &
2037 				~(rdlp(mp, TXQ_COMMAND) & 0xff);
2038 		mp->work_rx |= (int_cause & INT_RX) >> 2;
2039 	}
2040 
2041 	int_cause_ext &= INT_EXT_LINK_PHY | INT_EXT_TX;
2042 	if (int_cause_ext) {
2043 		wrlp(mp, INT_CAUSE_EXT, ~int_cause_ext);
2044 		if (int_cause_ext & INT_EXT_LINK_PHY)
2045 			mp->work_link = 1;
2046 		mp->work_tx |= int_cause_ext & INT_EXT_TX;
2047 	}
2048 
2049 	return 1;
2050 }
2051 
2052 static irqreturn_t mv643xx_eth_irq(int irq, void *dev_id)
2053 {
2054 	struct net_device *dev = (struct net_device *)dev_id;
2055 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2056 
2057 	if (unlikely(!mv643xx_eth_collect_events(mp)))
2058 		return IRQ_NONE;
2059 
2060 	wrlp(mp, INT_MASK, 0);
2061 	napi_schedule(&mp->napi);
2062 
2063 	return IRQ_HANDLED;
2064 }
2065 
2066 static void handle_link_event(struct mv643xx_eth_private *mp)
2067 {
2068 	struct net_device *dev = mp->dev;
2069 	u32 port_status;
2070 	int speed;
2071 	int duplex;
2072 	int fc;
2073 
2074 	port_status = rdlp(mp, PORT_STATUS);
2075 	if (!(port_status & LINK_UP)) {
2076 		if (netif_carrier_ok(dev)) {
2077 			int i;
2078 
2079 			netdev_info(dev, "link down\n");
2080 
2081 			netif_carrier_off(dev);
2082 
2083 			for (i = 0; i < mp->txq_count; i++) {
2084 				struct tx_queue *txq = mp->txq + i;
2085 
2086 				txq_reclaim(txq, txq->tx_ring_size, 1);
2087 				txq_reset_hw_ptr(txq);
2088 			}
2089 		}
2090 		return;
2091 	}
2092 
2093 	switch (port_status & PORT_SPEED_MASK) {
2094 	case PORT_SPEED_10:
2095 		speed = 10;
2096 		break;
2097 	case PORT_SPEED_100:
2098 		speed = 100;
2099 		break;
2100 	case PORT_SPEED_1000:
2101 		speed = 1000;
2102 		break;
2103 	default:
2104 		speed = -1;
2105 		break;
2106 	}
2107 	duplex = (port_status & FULL_DUPLEX) ? 1 : 0;
2108 	fc = (port_status & FLOW_CONTROL_ENABLED) ? 1 : 0;
2109 
2110 	netdev_info(dev, "link up, %d Mb/s, %s duplex, flow control %sabled\n",
2111 		    speed, duplex ? "full" : "half", fc ? "en" : "dis");
2112 
2113 	if (!netif_carrier_ok(dev))
2114 		netif_carrier_on(dev);
2115 }
2116 
2117 static int mv643xx_eth_poll(struct napi_struct *napi, int budget)
2118 {
2119 	struct mv643xx_eth_private *mp;
2120 	int work_done;
2121 
2122 	mp = container_of(napi, struct mv643xx_eth_private, napi);
2123 
2124 	if (unlikely(mp->oom)) {
2125 		mp->oom = 0;
2126 		del_timer(&mp->rx_oom);
2127 	}
2128 
2129 	work_done = 0;
2130 	while (work_done < budget) {
2131 		u8 queue_mask;
2132 		int queue;
2133 		int work_tbd;
2134 
2135 		if (mp->work_link) {
2136 			mp->work_link = 0;
2137 			handle_link_event(mp);
2138 			work_done++;
2139 			continue;
2140 		}
2141 
2142 		queue_mask = mp->work_tx | mp->work_tx_end | mp->work_rx;
2143 		if (likely(!mp->oom))
2144 			queue_mask |= mp->work_rx_refill;
2145 
2146 		if (!queue_mask) {
2147 			if (mv643xx_eth_collect_events(mp))
2148 				continue;
2149 			break;
2150 		}
2151 
2152 		queue = fls(queue_mask) - 1;
2153 		queue_mask = 1 << queue;
2154 
2155 		work_tbd = budget - work_done;
2156 		if (work_tbd > 16)
2157 			work_tbd = 16;
2158 
2159 		if (mp->work_tx_end & queue_mask) {
2160 			txq_kick(mp->txq + queue);
2161 		} else if (mp->work_tx & queue_mask) {
2162 			work_done += txq_reclaim(mp->txq + queue, work_tbd, 0);
2163 			txq_maybe_wake(mp->txq + queue);
2164 		} else if (mp->work_rx & queue_mask) {
2165 			work_done += rxq_process(mp->rxq + queue, work_tbd);
2166 		} else if (!mp->oom && (mp->work_rx_refill & queue_mask)) {
2167 			work_done += rxq_refill(mp->rxq + queue, work_tbd);
2168 		} else {
2169 			BUG();
2170 		}
2171 	}
2172 
2173 	if (work_done < budget) {
2174 		if (mp->oom)
2175 			mod_timer(&mp->rx_oom, jiffies + (HZ / 10));
2176 		napi_complete(napi);
2177 		wrlp(mp, INT_MASK, mp->int_mask);
2178 	}
2179 
2180 	return work_done;
2181 }
2182 
2183 static inline void oom_timer_wrapper(unsigned long data)
2184 {
2185 	struct mv643xx_eth_private *mp = (void *)data;
2186 
2187 	napi_schedule(&mp->napi);
2188 }
2189 
2190 static void phy_reset(struct mv643xx_eth_private *mp)
2191 {
2192 	int data;
2193 
2194 	data = phy_read(mp->phy, MII_BMCR);
2195 	if (data < 0)
2196 		return;
2197 
2198 	data |= BMCR_RESET;
2199 	if (phy_write(mp->phy, MII_BMCR, data) < 0)
2200 		return;
2201 
2202 	do {
2203 		data = phy_read(mp->phy, MII_BMCR);
2204 	} while (data >= 0 && data & BMCR_RESET);
2205 }
2206 
2207 static void port_start(struct mv643xx_eth_private *mp)
2208 {
2209 	u32 pscr;
2210 	int i;
2211 
2212 	/*
2213 	 * Perform PHY reset, if there is a PHY.
2214 	 */
2215 	if (mp->phy != NULL) {
2216 		struct ethtool_cmd cmd;
2217 
2218 		mv643xx_eth_get_settings(mp->dev, &cmd);
2219 		phy_reset(mp);
2220 		mv643xx_eth_set_settings(mp->dev, &cmd);
2221 	}
2222 
2223 	/*
2224 	 * Configure basic link parameters.
2225 	 */
2226 	pscr = rdlp(mp, PORT_SERIAL_CONTROL);
2227 
2228 	pscr |= SERIAL_PORT_ENABLE;
2229 	wrlp(mp, PORT_SERIAL_CONTROL, pscr);
2230 
2231 	pscr |= DO_NOT_FORCE_LINK_FAIL;
2232 	if (mp->phy == NULL)
2233 		pscr |= FORCE_LINK_PASS;
2234 	wrlp(mp, PORT_SERIAL_CONTROL, pscr);
2235 
2236 	/*
2237 	 * Configure TX path and queues.
2238 	 */
2239 	tx_set_rate(mp, 1000000000, 16777216);
2240 	for (i = 0; i < mp->txq_count; i++) {
2241 		struct tx_queue *txq = mp->txq + i;
2242 
2243 		txq_reset_hw_ptr(txq);
2244 		txq_set_rate(txq, 1000000000, 16777216);
2245 		txq_set_fixed_prio_mode(txq);
2246 	}
2247 
2248 	/*
2249 	 * Receive all unmatched unicast, TCP, UDP, BPDU and broadcast
2250 	 * frames to RX queue #0, and include the pseudo-header when
2251 	 * calculating receive checksums.
2252 	 */
2253 	mv643xx_eth_set_features(mp->dev, mp->dev->features);
2254 
2255 	/*
2256 	 * Treat BPDUs as normal multicasts, and disable partition mode.
2257 	 */
2258 	wrlp(mp, PORT_CONFIG_EXT, 0x00000000);
2259 
2260 	/*
2261 	 * Add configured unicast addresses to address filter table.
2262 	 */
2263 	mv643xx_eth_program_unicast_filter(mp->dev);
2264 
2265 	/*
2266 	 * Enable the receive queues.
2267 	 */
2268 	for (i = 0; i < mp->rxq_count; i++) {
2269 		struct rx_queue *rxq = mp->rxq + i;
2270 		u32 addr;
2271 
2272 		addr = (u32)rxq->rx_desc_dma;
2273 		addr += rxq->rx_curr_desc * sizeof(struct rx_desc);
2274 		wrlp(mp, RXQ_CURRENT_DESC_PTR(i), addr);
2275 
2276 		rxq_enable(rxq);
2277 	}
2278 }
2279 
2280 static void mv643xx_eth_recalc_skb_size(struct mv643xx_eth_private *mp)
2281 {
2282 	int skb_size;
2283 
2284 	/*
2285 	 * Reserve 2+14 bytes for an ethernet header (the hardware
2286 	 * automatically prepends 2 bytes of dummy data to each
2287 	 * received packet), 16 bytes for up to four VLAN tags, and
2288 	 * 4 bytes for the trailing FCS -- 36 bytes total.
2289 	 */
2290 	skb_size = mp->dev->mtu + 36;
2291 
2292 	/*
2293 	 * Make sure that the skb size is a multiple of 8 bytes, as
2294 	 * the lower three bits of the receive descriptor's buffer
2295 	 * size field are ignored by the hardware.
2296 	 */
2297 	mp->skb_size = (skb_size + 7) & ~7;
2298 
2299 	/*
2300 	 * If NET_SKB_PAD is smaller than a cache line,
2301 	 * netdev_alloc_skb() will cause skb->data to be misaligned
2302 	 * to a cache line boundary.  If this is the case, include
2303 	 * some extra space to allow re-aligning the data area.
2304 	 */
2305 	mp->skb_size += SKB_DMA_REALIGN;
2306 }
2307 
2308 static int mv643xx_eth_open(struct net_device *dev)
2309 {
2310 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2311 	int err;
2312 	int i;
2313 
2314 	wrlp(mp, INT_CAUSE, 0);
2315 	wrlp(mp, INT_CAUSE_EXT, 0);
2316 	rdlp(mp, INT_CAUSE_EXT);
2317 
2318 	err = request_irq(dev->irq, mv643xx_eth_irq,
2319 			  IRQF_SHARED, dev->name, dev);
2320 	if (err) {
2321 		netdev_err(dev, "can't assign irq\n");
2322 		return -EAGAIN;
2323 	}
2324 
2325 	mv643xx_eth_recalc_skb_size(mp);
2326 
2327 	napi_enable(&mp->napi);
2328 
2329 	skb_queue_head_init(&mp->rx_recycle);
2330 
2331 	mp->int_mask = INT_EXT;
2332 
2333 	for (i = 0; i < mp->rxq_count; i++) {
2334 		err = rxq_init(mp, i);
2335 		if (err) {
2336 			while (--i >= 0)
2337 				rxq_deinit(mp->rxq + i);
2338 			goto out;
2339 		}
2340 
2341 		rxq_refill(mp->rxq + i, INT_MAX);
2342 		mp->int_mask |= INT_RX_0 << i;
2343 	}
2344 
2345 	if (mp->oom) {
2346 		mp->rx_oom.expires = jiffies + (HZ / 10);
2347 		add_timer(&mp->rx_oom);
2348 	}
2349 
2350 	for (i = 0; i < mp->txq_count; i++) {
2351 		err = txq_init(mp, i);
2352 		if (err) {
2353 			while (--i >= 0)
2354 				txq_deinit(mp->txq + i);
2355 			goto out_free;
2356 		}
2357 		mp->int_mask |= INT_TX_END_0 << i;
2358 	}
2359 
2360 	port_start(mp);
2361 
2362 	wrlp(mp, INT_MASK_EXT, INT_EXT_LINK_PHY | INT_EXT_TX);
2363 	wrlp(mp, INT_MASK, mp->int_mask);
2364 
2365 	return 0;
2366 
2367 
2368 out_free:
2369 	for (i = 0; i < mp->rxq_count; i++)
2370 		rxq_deinit(mp->rxq + i);
2371 out:
2372 	free_irq(dev->irq, dev);
2373 
2374 	return err;
2375 }
2376 
2377 static void port_reset(struct mv643xx_eth_private *mp)
2378 {
2379 	unsigned int data;
2380 	int i;
2381 
2382 	for (i = 0; i < mp->rxq_count; i++)
2383 		rxq_disable(mp->rxq + i);
2384 	for (i = 0; i < mp->txq_count; i++)
2385 		txq_disable(mp->txq + i);
2386 
2387 	while (1) {
2388 		u32 ps = rdlp(mp, PORT_STATUS);
2389 
2390 		if ((ps & (TX_IN_PROGRESS | TX_FIFO_EMPTY)) == TX_FIFO_EMPTY)
2391 			break;
2392 		udelay(10);
2393 	}
2394 
2395 	/* Reset the Enable bit in the Configuration Register */
2396 	data = rdlp(mp, PORT_SERIAL_CONTROL);
2397 	data &= ~(SERIAL_PORT_ENABLE		|
2398 		  DO_NOT_FORCE_LINK_FAIL	|
2399 		  FORCE_LINK_PASS);
2400 	wrlp(mp, PORT_SERIAL_CONTROL, data);
2401 }
2402 
2403 static int mv643xx_eth_stop(struct net_device *dev)
2404 {
2405 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2406 	int i;
2407 
2408 	wrlp(mp, INT_MASK_EXT, 0x00000000);
2409 	wrlp(mp, INT_MASK, 0x00000000);
2410 	rdlp(mp, INT_MASK);
2411 
2412 	napi_disable(&mp->napi);
2413 
2414 	del_timer_sync(&mp->rx_oom);
2415 
2416 	netif_carrier_off(dev);
2417 
2418 	free_irq(dev->irq, dev);
2419 
2420 	port_reset(mp);
2421 	mv643xx_eth_get_stats(dev);
2422 	mib_counters_update(mp);
2423 	del_timer_sync(&mp->mib_counters_timer);
2424 
2425 	skb_queue_purge(&mp->rx_recycle);
2426 
2427 	for (i = 0; i < mp->rxq_count; i++)
2428 		rxq_deinit(mp->rxq + i);
2429 	for (i = 0; i < mp->txq_count; i++)
2430 		txq_deinit(mp->txq + i);
2431 
2432 	return 0;
2433 }
2434 
2435 static int mv643xx_eth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2436 {
2437 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2438 
2439 	if (mp->phy != NULL)
2440 		return phy_mii_ioctl(mp->phy, ifr, cmd);
2441 
2442 	return -EOPNOTSUPP;
2443 }
2444 
2445 static int mv643xx_eth_change_mtu(struct net_device *dev, int new_mtu)
2446 {
2447 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2448 
2449 	if (new_mtu < 64 || new_mtu > 9500)
2450 		return -EINVAL;
2451 
2452 	dev->mtu = new_mtu;
2453 	mv643xx_eth_recalc_skb_size(mp);
2454 	tx_set_rate(mp, 1000000000, 16777216);
2455 
2456 	if (!netif_running(dev))
2457 		return 0;
2458 
2459 	/*
2460 	 * Stop and then re-open the interface. This will allocate RX
2461 	 * skbs of the new MTU.
2462 	 * There is a possible danger that the open will not succeed,
2463 	 * due to memory being full.
2464 	 */
2465 	mv643xx_eth_stop(dev);
2466 	if (mv643xx_eth_open(dev)) {
2467 		netdev_err(dev,
2468 			   "fatal error on re-opening device after MTU change\n");
2469 	}
2470 
2471 	return 0;
2472 }
2473 
2474 static void tx_timeout_task(struct work_struct *ugly)
2475 {
2476 	struct mv643xx_eth_private *mp;
2477 
2478 	mp = container_of(ugly, struct mv643xx_eth_private, tx_timeout_task);
2479 	if (netif_running(mp->dev)) {
2480 		netif_tx_stop_all_queues(mp->dev);
2481 		port_reset(mp);
2482 		port_start(mp);
2483 		netif_tx_wake_all_queues(mp->dev);
2484 	}
2485 }
2486 
2487 static void mv643xx_eth_tx_timeout(struct net_device *dev)
2488 {
2489 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2490 
2491 	netdev_info(dev, "tx timeout\n");
2492 
2493 	schedule_work(&mp->tx_timeout_task);
2494 }
2495 
2496 #ifdef CONFIG_NET_POLL_CONTROLLER
2497 static void mv643xx_eth_netpoll(struct net_device *dev)
2498 {
2499 	struct mv643xx_eth_private *mp = netdev_priv(dev);
2500 
2501 	wrlp(mp, INT_MASK, 0x00000000);
2502 	rdlp(mp, INT_MASK);
2503 
2504 	mv643xx_eth_irq(dev->irq, dev);
2505 
2506 	wrlp(mp, INT_MASK, mp->int_mask);
2507 }
2508 #endif
2509 
2510 
2511 /* platform glue ************************************************************/
2512 static void
2513 mv643xx_eth_conf_mbus_windows(struct mv643xx_eth_shared_private *msp,
2514 			      const struct mbus_dram_target_info *dram)
2515 {
2516 	void __iomem *base = msp->base;
2517 	u32 win_enable;
2518 	u32 win_protect;
2519 	int i;
2520 
2521 	for (i = 0; i < 6; i++) {
2522 		writel(0, base + WINDOW_BASE(i));
2523 		writel(0, base + WINDOW_SIZE(i));
2524 		if (i < 4)
2525 			writel(0, base + WINDOW_REMAP_HIGH(i));
2526 	}
2527 
2528 	win_enable = 0x3f;
2529 	win_protect = 0;
2530 
2531 	for (i = 0; i < dram->num_cs; i++) {
2532 		const struct mbus_dram_window *cs = dram->cs + i;
2533 
2534 		writel((cs->base & 0xffff0000) |
2535 			(cs->mbus_attr << 8) |
2536 			dram->mbus_dram_target_id, base + WINDOW_BASE(i));
2537 		writel((cs->size - 1) & 0xffff0000, base + WINDOW_SIZE(i));
2538 
2539 		win_enable &= ~(1 << i);
2540 		win_protect |= 3 << (2 * i);
2541 	}
2542 
2543 	writel(win_enable, base + WINDOW_BAR_ENABLE);
2544 	msp->win_protect = win_protect;
2545 }
2546 
2547 static void infer_hw_params(struct mv643xx_eth_shared_private *msp)
2548 {
2549 	/*
2550 	 * Check whether we have a 14-bit coal limit field in bits
2551 	 * [21:8], or a 16-bit coal limit in bits [25,21:7] of the
2552 	 * SDMA config register.
2553 	 */
2554 	writel(0x02000000, msp->base + 0x0400 + SDMA_CONFIG);
2555 	if (readl(msp->base + 0x0400 + SDMA_CONFIG) & 0x02000000)
2556 		msp->extended_rx_coal_limit = 1;
2557 	else
2558 		msp->extended_rx_coal_limit = 0;
2559 
2560 	/*
2561 	 * Check whether the MAC supports TX rate control, and if
2562 	 * yes, whether its associated registers are in the old or
2563 	 * the new place.
2564 	 */
2565 	writel(1, msp->base + 0x0400 + TX_BW_MTU_MOVED);
2566 	if (readl(msp->base + 0x0400 + TX_BW_MTU_MOVED) & 1) {
2567 		msp->tx_bw_control = TX_BW_CONTROL_NEW_LAYOUT;
2568 	} else {
2569 		writel(7, msp->base + 0x0400 + TX_BW_RATE);
2570 		if (readl(msp->base + 0x0400 + TX_BW_RATE) & 7)
2571 			msp->tx_bw_control = TX_BW_CONTROL_OLD_LAYOUT;
2572 		else
2573 			msp->tx_bw_control = TX_BW_CONTROL_ABSENT;
2574 	}
2575 }
2576 
2577 static int mv643xx_eth_shared_probe(struct platform_device *pdev)
2578 {
2579 	static int mv643xx_eth_version_printed;
2580 	struct mv643xx_eth_shared_platform_data *pd = pdev->dev.platform_data;
2581 	struct mv643xx_eth_shared_private *msp;
2582 	const struct mbus_dram_target_info *dram;
2583 	struct resource *res;
2584 	int ret;
2585 
2586 	if (!mv643xx_eth_version_printed++)
2587 		pr_notice("MV-643xx 10/100/1000 ethernet driver version %s\n",
2588 			  mv643xx_eth_driver_version);
2589 
2590 	ret = -EINVAL;
2591 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
2592 	if (res == NULL)
2593 		goto out;
2594 
2595 	ret = -ENOMEM;
2596 	msp = kzalloc(sizeof(*msp), GFP_KERNEL);
2597 	if (msp == NULL)
2598 		goto out;
2599 
2600 	msp->base = ioremap(res->start, resource_size(res));
2601 	if (msp->base == NULL)
2602 		goto out_free;
2603 
2604 	/*
2605 	 * Set up and register SMI bus.
2606 	 */
2607 	if (pd == NULL || pd->shared_smi == NULL) {
2608 		msp->smi_bus = mdiobus_alloc();
2609 		if (msp->smi_bus == NULL)
2610 			goto out_unmap;
2611 
2612 		msp->smi_bus->priv = msp;
2613 		msp->smi_bus->name = "mv643xx_eth smi";
2614 		msp->smi_bus->read = smi_bus_read;
2615 		msp->smi_bus->write = smi_bus_write,
2616 		snprintf(msp->smi_bus->id, MII_BUS_ID_SIZE, "%s-%d",
2617 			pdev->name, pdev->id);
2618 		msp->smi_bus->parent = &pdev->dev;
2619 		msp->smi_bus->phy_mask = 0xffffffff;
2620 		if (mdiobus_register(msp->smi_bus) < 0)
2621 			goto out_free_mii_bus;
2622 		msp->smi = msp;
2623 	} else {
2624 		msp->smi = platform_get_drvdata(pd->shared_smi);
2625 	}
2626 
2627 	msp->err_interrupt = NO_IRQ;
2628 	init_waitqueue_head(&msp->smi_busy_wait);
2629 
2630 	/*
2631 	 * Check whether the error interrupt is hooked up.
2632 	 */
2633 	res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
2634 	if (res != NULL) {
2635 		int err;
2636 
2637 		err = request_irq(res->start, mv643xx_eth_err_irq,
2638 				  IRQF_SHARED, "mv643xx_eth", msp);
2639 		if (!err) {
2640 			writel(ERR_INT_SMI_DONE, msp->base + ERR_INT_MASK);
2641 			msp->err_interrupt = res->start;
2642 		}
2643 	}
2644 
2645 	/*
2646 	 * (Re-)program MBUS remapping windows if we are asked to.
2647 	 */
2648 	dram = mv_mbus_dram_info();
2649 	if (dram)
2650 		mv643xx_eth_conf_mbus_windows(msp, dram);
2651 
2652 	/*
2653 	 * Detect hardware parameters.
2654 	 */
2655 	msp->t_clk = (pd != NULL && pd->t_clk != 0) ? pd->t_clk : 133000000;
2656 	msp->tx_csum_limit = (pd != NULL && pd->tx_csum_limit) ?
2657 					pd->tx_csum_limit : 9 * 1024;
2658 	infer_hw_params(msp);
2659 
2660 	platform_set_drvdata(pdev, msp);
2661 
2662 	return 0;
2663 
2664 out_free_mii_bus:
2665 	mdiobus_free(msp->smi_bus);
2666 out_unmap:
2667 	iounmap(msp->base);
2668 out_free:
2669 	kfree(msp);
2670 out:
2671 	return ret;
2672 }
2673 
2674 static int mv643xx_eth_shared_remove(struct platform_device *pdev)
2675 {
2676 	struct mv643xx_eth_shared_private *msp = platform_get_drvdata(pdev);
2677 	struct mv643xx_eth_shared_platform_data *pd = pdev->dev.platform_data;
2678 
2679 	if (pd == NULL || pd->shared_smi == NULL) {
2680 		mdiobus_unregister(msp->smi_bus);
2681 		mdiobus_free(msp->smi_bus);
2682 	}
2683 	if (msp->err_interrupt != NO_IRQ)
2684 		free_irq(msp->err_interrupt, msp);
2685 	iounmap(msp->base);
2686 	kfree(msp);
2687 
2688 	return 0;
2689 }
2690 
2691 static struct platform_driver mv643xx_eth_shared_driver = {
2692 	.probe		= mv643xx_eth_shared_probe,
2693 	.remove		= mv643xx_eth_shared_remove,
2694 	.driver = {
2695 		.name	= MV643XX_ETH_SHARED_NAME,
2696 		.owner	= THIS_MODULE,
2697 	},
2698 };
2699 
2700 static void phy_addr_set(struct mv643xx_eth_private *mp, int phy_addr)
2701 {
2702 	int addr_shift = 5 * mp->port_num;
2703 	u32 data;
2704 
2705 	data = rdl(mp, PHY_ADDR);
2706 	data &= ~(0x1f << addr_shift);
2707 	data |= (phy_addr & 0x1f) << addr_shift;
2708 	wrl(mp, PHY_ADDR, data);
2709 }
2710 
2711 static int phy_addr_get(struct mv643xx_eth_private *mp)
2712 {
2713 	unsigned int data;
2714 
2715 	data = rdl(mp, PHY_ADDR);
2716 
2717 	return (data >> (5 * mp->port_num)) & 0x1f;
2718 }
2719 
2720 static void set_params(struct mv643xx_eth_private *mp,
2721 		       struct mv643xx_eth_platform_data *pd)
2722 {
2723 	struct net_device *dev = mp->dev;
2724 
2725 	if (is_valid_ether_addr(pd->mac_addr))
2726 		memcpy(dev->dev_addr, pd->mac_addr, 6);
2727 	else
2728 		uc_addr_get(mp, dev->dev_addr);
2729 
2730 	mp->rx_ring_size = DEFAULT_RX_QUEUE_SIZE;
2731 	if (pd->rx_queue_size)
2732 		mp->rx_ring_size = pd->rx_queue_size;
2733 	mp->rx_desc_sram_addr = pd->rx_sram_addr;
2734 	mp->rx_desc_sram_size = pd->rx_sram_size;
2735 
2736 	mp->rxq_count = pd->rx_queue_count ? : 1;
2737 
2738 	mp->tx_ring_size = DEFAULT_TX_QUEUE_SIZE;
2739 	if (pd->tx_queue_size)
2740 		mp->tx_ring_size = pd->tx_queue_size;
2741 	mp->tx_desc_sram_addr = pd->tx_sram_addr;
2742 	mp->tx_desc_sram_size = pd->tx_sram_size;
2743 
2744 	mp->txq_count = pd->tx_queue_count ? : 1;
2745 }
2746 
2747 static struct phy_device *phy_scan(struct mv643xx_eth_private *mp,
2748 				   int phy_addr)
2749 {
2750 	struct mii_bus *bus = mp->shared->smi->smi_bus;
2751 	struct phy_device *phydev;
2752 	int start;
2753 	int num;
2754 	int i;
2755 
2756 	if (phy_addr == MV643XX_ETH_PHY_ADDR_DEFAULT) {
2757 		start = phy_addr_get(mp) & 0x1f;
2758 		num = 32;
2759 	} else {
2760 		start = phy_addr & 0x1f;
2761 		num = 1;
2762 	}
2763 
2764 	phydev = NULL;
2765 	for (i = 0; i < num; i++) {
2766 		int addr = (start + i) & 0x1f;
2767 
2768 		if (bus->phy_map[addr] == NULL)
2769 			mdiobus_scan(bus, addr);
2770 
2771 		if (phydev == NULL) {
2772 			phydev = bus->phy_map[addr];
2773 			if (phydev != NULL)
2774 				phy_addr_set(mp, addr);
2775 		}
2776 	}
2777 
2778 	return phydev;
2779 }
2780 
2781 static void phy_init(struct mv643xx_eth_private *mp, int speed, int duplex)
2782 {
2783 	struct phy_device *phy = mp->phy;
2784 
2785 	phy_reset(mp);
2786 
2787 	phy_attach(mp->dev, dev_name(&phy->dev), 0, PHY_INTERFACE_MODE_GMII);
2788 
2789 	if (speed == 0) {
2790 		phy->autoneg = AUTONEG_ENABLE;
2791 		phy->speed = 0;
2792 		phy->duplex = 0;
2793 		phy->advertising = phy->supported | ADVERTISED_Autoneg;
2794 	} else {
2795 		phy->autoneg = AUTONEG_DISABLE;
2796 		phy->advertising = 0;
2797 		phy->speed = speed;
2798 		phy->duplex = duplex;
2799 	}
2800 	phy_start_aneg(phy);
2801 }
2802 
2803 static void init_pscr(struct mv643xx_eth_private *mp, int speed, int duplex)
2804 {
2805 	u32 pscr;
2806 
2807 	pscr = rdlp(mp, PORT_SERIAL_CONTROL);
2808 	if (pscr & SERIAL_PORT_ENABLE) {
2809 		pscr &= ~SERIAL_PORT_ENABLE;
2810 		wrlp(mp, PORT_SERIAL_CONTROL, pscr);
2811 	}
2812 
2813 	pscr = MAX_RX_PACKET_9700BYTE | SERIAL_PORT_CONTROL_RESERVED;
2814 	if (mp->phy == NULL) {
2815 		pscr |= DISABLE_AUTO_NEG_SPEED_GMII;
2816 		if (speed == SPEED_1000)
2817 			pscr |= SET_GMII_SPEED_TO_1000;
2818 		else if (speed == SPEED_100)
2819 			pscr |= SET_MII_SPEED_TO_100;
2820 
2821 		pscr |= DISABLE_AUTO_NEG_FOR_FLOW_CTRL;
2822 
2823 		pscr |= DISABLE_AUTO_NEG_FOR_DUPLEX;
2824 		if (duplex == DUPLEX_FULL)
2825 			pscr |= SET_FULL_DUPLEX_MODE;
2826 	}
2827 
2828 	wrlp(mp, PORT_SERIAL_CONTROL, pscr);
2829 }
2830 
2831 static const struct net_device_ops mv643xx_eth_netdev_ops = {
2832 	.ndo_open		= mv643xx_eth_open,
2833 	.ndo_stop		= mv643xx_eth_stop,
2834 	.ndo_start_xmit		= mv643xx_eth_xmit,
2835 	.ndo_set_rx_mode	= mv643xx_eth_set_rx_mode,
2836 	.ndo_set_mac_address	= mv643xx_eth_set_mac_address,
2837 	.ndo_validate_addr	= eth_validate_addr,
2838 	.ndo_do_ioctl		= mv643xx_eth_ioctl,
2839 	.ndo_change_mtu		= mv643xx_eth_change_mtu,
2840 	.ndo_set_features	= mv643xx_eth_set_features,
2841 	.ndo_tx_timeout		= mv643xx_eth_tx_timeout,
2842 	.ndo_get_stats		= mv643xx_eth_get_stats,
2843 #ifdef CONFIG_NET_POLL_CONTROLLER
2844 	.ndo_poll_controller	= mv643xx_eth_netpoll,
2845 #endif
2846 };
2847 
2848 static int mv643xx_eth_probe(struct platform_device *pdev)
2849 {
2850 	struct mv643xx_eth_platform_data *pd;
2851 	struct mv643xx_eth_private *mp;
2852 	struct net_device *dev;
2853 	struct resource *res;
2854 	int err;
2855 
2856 	pd = pdev->dev.platform_data;
2857 	if (pd == NULL) {
2858 		dev_err(&pdev->dev, "no mv643xx_eth_platform_data\n");
2859 		return -ENODEV;
2860 	}
2861 
2862 	if (pd->shared == NULL) {
2863 		dev_err(&pdev->dev, "no mv643xx_eth_platform_data->shared\n");
2864 		return -ENODEV;
2865 	}
2866 
2867 	dev = alloc_etherdev_mq(sizeof(struct mv643xx_eth_private), 8);
2868 	if (!dev)
2869 		return -ENOMEM;
2870 
2871 	mp = netdev_priv(dev);
2872 	platform_set_drvdata(pdev, mp);
2873 
2874 	mp->shared = platform_get_drvdata(pd->shared);
2875 	mp->base = mp->shared->base + 0x0400 + (pd->port_number << 10);
2876 	mp->port_num = pd->port_number;
2877 
2878 	mp->dev = dev;
2879 
2880 	set_params(mp, pd);
2881 	netif_set_real_num_tx_queues(dev, mp->txq_count);
2882 	netif_set_real_num_rx_queues(dev, mp->rxq_count);
2883 
2884 	if (pd->phy_addr != MV643XX_ETH_PHY_NONE)
2885 		mp->phy = phy_scan(mp, pd->phy_addr);
2886 
2887 	if (mp->phy != NULL)
2888 		phy_init(mp, pd->speed, pd->duplex);
2889 
2890 	SET_ETHTOOL_OPS(dev, &mv643xx_eth_ethtool_ops);
2891 
2892 	init_pscr(mp, pd->speed, pd->duplex);
2893 
2894 
2895 	mib_counters_clear(mp);
2896 
2897 	init_timer(&mp->mib_counters_timer);
2898 	mp->mib_counters_timer.data = (unsigned long)mp;
2899 	mp->mib_counters_timer.function = mib_counters_timer_wrapper;
2900 	mp->mib_counters_timer.expires = jiffies + 30 * HZ;
2901 	add_timer(&mp->mib_counters_timer);
2902 
2903 	spin_lock_init(&mp->mib_counters_lock);
2904 
2905 	INIT_WORK(&mp->tx_timeout_task, tx_timeout_task);
2906 
2907 	netif_napi_add(dev, &mp->napi, mv643xx_eth_poll, 128);
2908 
2909 	init_timer(&mp->rx_oom);
2910 	mp->rx_oom.data = (unsigned long)mp;
2911 	mp->rx_oom.function = oom_timer_wrapper;
2912 
2913 
2914 	res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
2915 	BUG_ON(!res);
2916 	dev->irq = res->start;
2917 
2918 	dev->netdev_ops = &mv643xx_eth_netdev_ops;
2919 
2920 	dev->watchdog_timeo = 2 * HZ;
2921 	dev->base_addr = 0;
2922 
2923 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM |
2924 		NETIF_F_RXCSUM | NETIF_F_LRO;
2925 	dev->features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_RXCSUM;
2926 	dev->vlan_features = NETIF_F_SG | NETIF_F_IP_CSUM;
2927 
2928 	dev->priv_flags |= IFF_UNICAST_FLT;
2929 
2930 	SET_NETDEV_DEV(dev, &pdev->dev);
2931 
2932 	if (mp->shared->win_protect)
2933 		wrl(mp, WINDOW_PROTECT(mp->port_num), mp->shared->win_protect);
2934 
2935 	netif_carrier_off(dev);
2936 
2937 	wrlp(mp, SDMA_CONFIG, PORT_SDMA_CONFIG_DEFAULT_VALUE);
2938 
2939 	set_rx_coal(mp, 250);
2940 	set_tx_coal(mp, 0);
2941 
2942 	err = register_netdev(dev);
2943 	if (err)
2944 		goto out;
2945 
2946 	netdev_notice(dev, "port %d with MAC address %pM\n",
2947 		      mp->port_num, dev->dev_addr);
2948 
2949 	if (mp->tx_desc_sram_size > 0)
2950 		netdev_notice(dev, "configured with sram\n");
2951 
2952 	return 0;
2953 
2954 out:
2955 	free_netdev(dev);
2956 
2957 	return err;
2958 }
2959 
2960 static int mv643xx_eth_remove(struct platform_device *pdev)
2961 {
2962 	struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
2963 
2964 	unregister_netdev(mp->dev);
2965 	if (mp->phy != NULL)
2966 		phy_detach(mp->phy);
2967 	cancel_work_sync(&mp->tx_timeout_task);
2968 	free_netdev(mp->dev);
2969 
2970 	platform_set_drvdata(pdev, NULL);
2971 
2972 	return 0;
2973 }
2974 
2975 static void mv643xx_eth_shutdown(struct platform_device *pdev)
2976 {
2977 	struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
2978 
2979 	/* Mask all interrupts on ethernet port */
2980 	wrlp(mp, INT_MASK, 0);
2981 	rdlp(mp, INT_MASK);
2982 
2983 	if (netif_running(mp->dev))
2984 		port_reset(mp);
2985 }
2986 
2987 static struct platform_driver mv643xx_eth_driver = {
2988 	.probe		= mv643xx_eth_probe,
2989 	.remove		= mv643xx_eth_remove,
2990 	.shutdown	= mv643xx_eth_shutdown,
2991 	.driver = {
2992 		.name	= MV643XX_ETH_NAME,
2993 		.owner	= THIS_MODULE,
2994 	},
2995 };
2996 
2997 static int __init mv643xx_eth_init_module(void)
2998 {
2999 	int rc;
3000 
3001 	rc = platform_driver_register(&mv643xx_eth_shared_driver);
3002 	if (!rc) {
3003 		rc = platform_driver_register(&mv643xx_eth_driver);
3004 		if (rc)
3005 			platform_driver_unregister(&mv643xx_eth_shared_driver);
3006 	}
3007 
3008 	return rc;
3009 }
3010 module_init(mv643xx_eth_init_module);
3011 
3012 static void __exit mv643xx_eth_cleanup_module(void)
3013 {
3014 	platform_driver_unregister(&mv643xx_eth_driver);
3015 	platform_driver_unregister(&mv643xx_eth_shared_driver);
3016 }
3017 module_exit(mv643xx_eth_cleanup_module);
3018 
3019 MODULE_AUTHOR("Rabeeh Khoury, Assaf Hoffman, Matthew Dharm, "
3020 	      "Manish Lachwani, Dale Farnsworth and Lennert Buytenhek");
3021 MODULE_DESCRIPTION("Ethernet driver for Marvell MV643XX");
3022 MODULE_LICENSE("GPL");
3023 MODULE_ALIAS("platform:" MV643XX_ETH_SHARED_NAME);
3024 MODULE_ALIAS("platform:" MV643XX_ETH_NAME);
3025