xref: /openbmc/linux/drivers/net/ethernet/ibm/emac/core.c (revision 93d90ad7)
1 /*
2  * drivers/net/ethernet/ibm/emac/core.c
3  *
4  * Driver for PowerPC 4xx on-chip ethernet controller.
5  *
6  * Copyright 2007 Benjamin Herrenschmidt, IBM Corp.
7  *                <benh@kernel.crashing.org>
8  *
9  * Based on the arch/ppc version of the driver:
10  *
11  * Copyright (c) 2004, 2005 Zultys Technologies.
12  * Eugene Surovegin <eugene.surovegin@zultys.com> or <ebs@ebshome.net>
13  *
14  * Based on original work by
15  * 	Matt Porter <mporter@kernel.crashing.org>
16  *	(c) 2003 Benjamin Herrenschmidt <benh@kernel.crashing.org>
17  *      Armin Kuster <akuster@mvista.com>
18  * 	Johnnie Peters <jpeters@mvista.com>
19  *
20  * This program is free software; you can redistribute  it and/or modify it
21  * under  the terms of  the GNU General  Public License as published by the
22  * Free Software Foundation;  either version 2 of the  License, or (at your
23  * option) any later version.
24  *
25  */
26 
27 #include <linux/module.h>
28 #include <linux/sched.h>
29 #include <linux/string.h>
30 #include <linux/errno.h>
31 #include <linux/delay.h>
32 #include <linux/types.h>
33 #include <linux/pci.h>
34 #include <linux/etherdevice.h>
35 #include <linux/skbuff.h>
36 #include <linux/crc32.h>
37 #include <linux/ethtool.h>
38 #include <linux/mii.h>
39 #include <linux/bitops.h>
40 #include <linux/workqueue.h>
41 #include <linux/of.h>
42 #include <linux/of_address.h>
43 #include <linux/of_irq.h>
44 #include <linux/of_net.h>
45 #include <linux/slab.h>
46 
47 #include <asm/processor.h>
48 #include <asm/io.h>
49 #include <asm/dma.h>
50 #include <asm/uaccess.h>
51 #include <asm/dcr.h>
52 #include <asm/dcr-regs.h>
53 
54 #include "core.h"
55 
56 /*
57  * Lack of dma_unmap_???? calls is intentional.
58  *
59  * API-correct usage requires additional support state information to be
60  * maintained for every RX and TX buffer descriptor (BD). Unfortunately, due to
61  * EMAC design (e.g. TX buffer passed from network stack can be split into
62  * several BDs, dma_map_single/dma_map_page can be used to map particular BD),
63  * maintaining such information will add additional overhead.
64  * Current DMA API implementation for 4xx processors only ensures cache coherency
65  * and dma_unmap_???? routines are empty and are likely to stay this way.
66  * I decided to omit dma_unmap_??? calls because I don't want to add additional
67  * complexity just for the sake of following some abstract API, when it doesn't
68  * add any real benefit to the driver. I understand that this decision maybe
69  * controversial, but I really tried to make code API-correct and efficient
70  * at the same time and didn't come up with code I liked :(.                --ebs
71  */
72 
73 #define DRV_NAME        "emac"
74 #define DRV_VERSION     "3.54"
75 #define DRV_DESC        "PPC 4xx OCP EMAC driver"
76 
77 MODULE_DESCRIPTION(DRV_DESC);
78 MODULE_AUTHOR
79     ("Eugene Surovegin <eugene.surovegin@zultys.com> or <ebs@ebshome.net>");
80 MODULE_LICENSE("GPL");
81 
82 /*
83  * PPC64 doesn't (yet) have a cacheable_memcpy
84  */
85 #ifdef CONFIG_PPC64
86 #define cacheable_memcpy(d,s,n) memcpy((d),(s),(n))
87 #endif
88 
89 /* minimum number of free TX descriptors required to wake up TX process */
90 #define EMAC_TX_WAKEUP_THRESH		(NUM_TX_BUFF / 4)
91 
92 /* If packet size is less than this number, we allocate small skb and copy packet
93  * contents into it instead of just sending original big skb up
94  */
95 #define EMAC_RX_COPY_THRESH		CONFIG_IBM_EMAC_RX_COPY_THRESHOLD
96 
97 /* Since multiple EMACs share MDIO lines in various ways, we need
98  * to avoid re-using the same PHY ID in cases where the arch didn't
99  * setup precise phy_map entries
100  *
101  * XXX This is something that needs to be reworked as we can have multiple
102  * EMAC "sets" (multiple ASICs containing several EMACs) though we can
103  * probably require in that case to have explicit PHY IDs in the device-tree
104  */
105 static u32 busy_phy_map;
106 static DEFINE_MUTEX(emac_phy_map_lock);
107 
108 /* This is the wait queue used to wait on any event related to probe, that
109  * is discovery of MALs, other EMACs, ZMII/RGMIIs, etc...
110  */
111 static DECLARE_WAIT_QUEUE_HEAD(emac_probe_wait);
112 
113 /* Having stable interface names is a doomed idea. However, it would be nice
114  * if we didn't have completely random interface names at boot too :-) It's
115  * just a matter of making everybody's life easier. Since we are doing
116  * threaded probing, it's a bit harder though. The base idea here is that
117  * we make up a list of all emacs in the device-tree before we register the
118  * driver. Every emac will then wait for the previous one in the list to
119  * initialize before itself. We should also keep that list ordered by
120  * cell_index.
121  * That list is only 4 entries long, meaning that additional EMACs don't
122  * get ordering guarantees unless EMAC_BOOT_LIST_SIZE is increased.
123  */
124 
125 #define EMAC_BOOT_LIST_SIZE	4
126 static struct device_node *emac_boot_list[EMAC_BOOT_LIST_SIZE];
127 
128 /* How long should I wait for dependent devices ? */
129 #define EMAC_PROBE_DEP_TIMEOUT	(HZ * 5)
130 
131 /* I don't want to litter system log with timeout errors
132  * when we have brain-damaged PHY.
133  */
134 static inline void emac_report_timeout_error(struct emac_instance *dev,
135 					     const char *error)
136 {
137 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX |
138 				  EMAC_FTR_460EX_PHY_CLK_FIX |
139 				  EMAC_FTR_440EP_PHY_CLK_FIX))
140 		DBG(dev, "%s" NL, error);
141 	else if (net_ratelimit())
142 		printk(KERN_ERR "%s: %s\n", dev->ofdev->dev.of_node->full_name,
143 			error);
144 }
145 
146 /* EMAC PHY clock workaround:
147  * 440EP/440GR has more sane SDR0_MFR register implementation than 440GX,
148  * which allows controlling each EMAC clock
149  */
150 static inline void emac_rx_clk_tx(struct emac_instance *dev)
151 {
152 #ifdef CONFIG_PPC_DCR_NATIVE
153 	if (emac_has_feature(dev, EMAC_FTR_440EP_PHY_CLK_FIX))
154 		dcri_clrset(SDR0, SDR0_MFR,
155 			    0, SDR0_MFR_ECS >> dev->cell_index);
156 #endif
157 }
158 
159 static inline void emac_rx_clk_default(struct emac_instance *dev)
160 {
161 #ifdef CONFIG_PPC_DCR_NATIVE
162 	if (emac_has_feature(dev, EMAC_FTR_440EP_PHY_CLK_FIX))
163 		dcri_clrset(SDR0, SDR0_MFR,
164 			    SDR0_MFR_ECS >> dev->cell_index, 0);
165 #endif
166 }
167 
168 /* PHY polling intervals */
169 #define PHY_POLL_LINK_ON	HZ
170 #define PHY_POLL_LINK_OFF	(HZ / 5)
171 
172 /* Graceful stop timeouts in us.
173  * We should allow up to 1 frame time (full-duplex, ignoring collisions)
174  */
175 #define STOP_TIMEOUT_10		1230
176 #define STOP_TIMEOUT_100	124
177 #define STOP_TIMEOUT_1000	13
178 #define STOP_TIMEOUT_1000_JUMBO	73
179 
180 static unsigned char default_mcast_addr[] = {
181 	0x01, 0x80, 0xC2, 0x00, 0x00, 0x01
182 };
183 
184 /* Please, keep in sync with struct ibm_emac_stats/ibm_emac_error_stats */
185 static const char emac_stats_keys[EMAC_ETHTOOL_STATS_COUNT][ETH_GSTRING_LEN] = {
186 	"rx_packets", "rx_bytes", "tx_packets", "tx_bytes", "rx_packets_csum",
187 	"tx_packets_csum", "tx_undo", "rx_dropped_stack", "rx_dropped_oom",
188 	"rx_dropped_error", "rx_dropped_resize", "rx_dropped_mtu",
189 	"rx_stopped", "rx_bd_errors", "rx_bd_overrun", "rx_bd_bad_packet",
190 	"rx_bd_runt_packet", "rx_bd_short_event", "rx_bd_alignment_error",
191 	"rx_bd_bad_fcs", "rx_bd_packet_too_long", "rx_bd_out_of_range",
192 	"rx_bd_in_range", "rx_parity", "rx_fifo_overrun", "rx_overrun",
193 	"rx_bad_packet", "rx_runt_packet", "rx_short_event",
194 	"rx_alignment_error", "rx_bad_fcs", "rx_packet_too_long",
195 	"rx_out_of_range", "rx_in_range", "tx_dropped", "tx_bd_errors",
196 	"tx_bd_bad_fcs", "tx_bd_carrier_loss", "tx_bd_excessive_deferral",
197 	"tx_bd_excessive_collisions", "tx_bd_late_collision",
198 	"tx_bd_multple_collisions", "tx_bd_single_collision",
199 	"tx_bd_underrun", "tx_bd_sqe", "tx_parity", "tx_underrun", "tx_sqe",
200 	"tx_errors"
201 };
202 
203 static irqreturn_t emac_irq(int irq, void *dev_instance);
204 static void emac_clean_tx_ring(struct emac_instance *dev);
205 static void __emac_set_multicast_list(struct emac_instance *dev);
206 
207 static inline int emac_phy_supports_gige(int phy_mode)
208 {
209 	return  phy_mode == PHY_MODE_GMII ||
210 		phy_mode == PHY_MODE_RGMII ||
211 		phy_mode == PHY_MODE_SGMII ||
212 		phy_mode == PHY_MODE_TBI ||
213 		phy_mode == PHY_MODE_RTBI;
214 }
215 
216 static inline int emac_phy_gpcs(int phy_mode)
217 {
218 	return  phy_mode == PHY_MODE_SGMII ||
219 		phy_mode == PHY_MODE_TBI ||
220 		phy_mode == PHY_MODE_RTBI;
221 }
222 
223 static inline void emac_tx_enable(struct emac_instance *dev)
224 {
225 	struct emac_regs __iomem *p = dev->emacp;
226 	u32 r;
227 
228 	DBG(dev, "tx_enable" NL);
229 
230 	r = in_be32(&p->mr0);
231 	if (!(r & EMAC_MR0_TXE))
232 		out_be32(&p->mr0, r | EMAC_MR0_TXE);
233 }
234 
235 static void emac_tx_disable(struct emac_instance *dev)
236 {
237 	struct emac_regs __iomem *p = dev->emacp;
238 	u32 r;
239 
240 	DBG(dev, "tx_disable" NL);
241 
242 	r = in_be32(&p->mr0);
243 	if (r & EMAC_MR0_TXE) {
244 		int n = dev->stop_timeout;
245 		out_be32(&p->mr0, r & ~EMAC_MR0_TXE);
246 		while (!(in_be32(&p->mr0) & EMAC_MR0_TXI) && n) {
247 			udelay(1);
248 			--n;
249 		}
250 		if (unlikely(!n))
251 			emac_report_timeout_error(dev, "TX disable timeout");
252 	}
253 }
254 
255 static void emac_rx_enable(struct emac_instance *dev)
256 {
257 	struct emac_regs __iomem *p = dev->emacp;
258 	u32 r;
259 
260 	if (unlikely(test_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags)))
261 		goto out;
262 
263 	DBG(dev, "rx_enable" NL);
264 
265 	r = in_be32(&p->mr0);
266 	if (!(r & EMAC_MR0_RXE)) {
267 		if (unlikely(!(r & EMAC_MR0_RXI))) {
268 			/* Wait if previous async disable is still in progress */
269 			int n = dev->stop_timeout;
270 			while (!(r = in_be32(&p->mr0) & EMAC_MR0_RXI) && n) {
271 				udelay(1);
272 				--n;
273 			}
274 			if (unlikely(!n))
275 				emac_report_timeout_error(dev,
276 							  "RX disable timeout");
277 		}
278 		out_be32(&p->mr0, r | EMAC_MR0_RXE);
279 	}
280  out:
281 	;
282 }
283 
284 static void emac_rx_disable(struct emac_instance *dev)
285 {
286 	struct emac_regs __iomem *p = dev->emacp;
287 	u32 r;
288 
289 	DBG(dev, "rx_disable" NL);
290 
291 	r = in_be32(&p->mr0);
292 	if (r & EMAC_MR0_RXE) {
293 		int n = dev->stop_timeout;
294 		out_be32(&p->mr0, r & ~EMAC_MR0_RXE);
295 		while (!(in_be32(&p->mr0) & EMAC_MR0_RXI) && n) {
296 			udelay(1);
297 			--n;
298 		}
299 		if (unlikely(!n))
300 			emac_report_timeout_error(dev, "RX disable timeout");
301 	}
302 }
303 
304 static inline void emac_netif_stop(struct emac_instance *dev)
305 {
306 	netif_tx_lock_bh(dev->ndev);
307 	netif_addr_lock(dev->ndev);
308 	dev->no_mcast = 1;
309 	netif_addr_unlock(dev->ndev);
310 	netif_tx_unlock_bh(dev->ndev);
311 	dev->ndev->trans_start = jiffies;	/* prevent tx timeout */
312 	mal_poll_disable(dev->mal, &dev->commac);
313 	netif_tx_disable(dev->ndev);
314 }
315 
316 static inline void emac_netif_start(struct emac_instance *dev)
317 {
318 	netif_tx_lock_bh(dev->ndev);
319 	netif_addr_lock(dev->ndev);
320 	dev->no_mcast = 0;
321 	if (dev->mcast_pending && netif_running(dev->ndev))
322 		__emac_set_multicast_list(dev);
323 	netif_addr_unlock(dev->ndev);
324 	netif_tx_unlock_bh(dev->ndev);
325 
326 	netif_wake_queue(dev->ndev);
327 
328 	/* NOTE: unconditional netif_wake_queue is only appropriate
329 	 * so long as all callers are assured to have free tx slots
330 	 * (taken from tg3... though the case where that is wrong is
331 	 *  not terribly harmful)
332 	 */
333 	mal_poll_enable(dev->mal, &dev->commac);
334 }
335 
336 static inline void emac_rx_disable_async(struct emac_instance *dev)
337 {
338 	struct emac_regs __iomem *p = dev->emacp;
339 	u32 r;
340 
341 	DBG(dev, "rx_disable_async" NL);
342 
343 	r = in_be32(&p->mr0);
344 	if (r & EMAC_MR0_RXE)
345 		out_be32(&p->mr0, r & ~EMAC_MR0_RXE);
346 }
347 
348 static int emac_reset(struct emac_instance *dev)
349 {
350 	struct emac_regs __iomem *p = dev->emacp;
351 	int n = 20;
352 
353 	DBG(dev, "reset" NL);
354 
355 	if (!dev->reset_failed) {
356 		/* 40x erratum suggests stopping RX channel before reset,
357 		 * we stop TX as well
358 		 */
359 		emac_rx_disable(dev);
360 		emac_tx_disable(dev);
361 	}
362 
363 #ifdef CONFIG_PPC_DCR_NATIVE
364 	/*
365 	 * PPC460EX/GT Embedded Processor Advanced User's Manual
366 	 * section 28.10.1 Mode Register 0 (EMACx_MR0) states:
367 	 * Note: The PHY must provide a TX Clk in order to perform a soft reset
368 	 * of the EMAC. If none is present, select the internal clock
369 	 * (SDR0_ETH_CFG[EMACx_PHY_CLK] = 1).
370 	 * After a soft reset, select the external clock.
371 	 */
372 	if (emac_has_feature(dev, EMAC_FTR_460EX_PHY_CLK_FIX)) {
373 		if (dev->phy_address == 0xffffffff &&
374 		    dev->phy_map == 0xffffffff) {
375 			/* No PHY: select internal loop clock before reset */
376 			dcri_clrset(SDR0, SDR0_ETH_CFG,
377 				    0, SDR0_ETH_CFG_ECS << dev->cell_index);
378 		} else {
379 			/* PHY present: select external clock before reset */
380 			dcri_clrset(SDR0, SDR0_ETH_CFG,
381 				    SDR0_ETH_CFG_ECS << dev->cell_index, 0);
382 		}
383 	}
384 #endif
385 
386 	out_be32(&p->mr0, EMAC_MR0_SRST);
387 	while ((in_be32(&p->mr0) & EMAC_MR0_SRST) && n)
388 		--n;
389 
390 #ifdef CONFIG_PPC_DCR_NATIVE
391 	if (emac_has_feature(dev, EMAC_FTR_460EX_PHY_CLK_FIX)) {
392 		if (dev->phy_address == 0xffffffff &&
393 		    dev->phy_map == 0xffffffff) {
394 			/* No PHY: restore external clock source after reset */
395 			dcri_clrset(SDR0, SDR0_ETH_CFG,
396 				    SDR0_ETH_CFG_ECS << dev->cell_index, 0);
397 		}
398 	}
399 #endif
400 
401 	if (n) {
402 		dev->reset_failed = 0;
403 		return 0;
404 	} else {
405 		emac_report_timeout_error(dev, "reset timeout");
406 		dev->reset_failed = 1;
407 		return -ETIMEDOUT;
408 	}
409 }
410 
411 static void emac_hash_mc(struct emac_instance *dev)
412 {
413 	const int regs = EMAC_XAHT_REGS(dev);
414 	u32 *gaht_base = emac_gaht_base(dev);
415 	u32 gaht_temp[regs];
416 	struct netdev_hw_addr *ha;
417 	int i;
418 
419 	DBG(dev, "hash_mc %d" NL, netdev_mc_count(dev->ndev));
420 
421 	memset(gaht_temp, 0, sizeof (gaht_temp));
422 
423 	netdev_for_each_mc_addr(ha, dev->ndev) {
424 		int slot, reg, mask;
425 		DBG2(dev, "mc %pM" NL, ha->addr);
426 
427 		slot = EMAC_XAHT_CRC_TO_SLOT(dev,
428 					     ether_crc(ETH_ALEN, ha->addr));
429 		reg = EMAC_XAHT_SLOT_TO_REG(dev, slot);
430 		mask = EMAC_XAHT_SLOT_TO_MASK(dev, slot);
431 
432 		gaht_temp[reg] |= mask;
433 	}
434 
435 	for (i = 0; i < regs; i++)
436 		out_be32(gaht_base + i, gaht_temp[i]);
437 }
438 
439 static inline u32 emac_iff2rmr(struct net_device *ndev)
440 {
441 	struct emac_instance *dev = netdev_priv(ndev);
442 	u32 r;
443 
444 	r = EMAC_RMR_SP | EMAC_RMR_SFCS | EMAC_RMR_IAE | EMAC_RMR_BAE;
445 
446 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
447 	    r |= EMAC4_RMR_BASE;
448 	else
449 	    r |= EMAC_RMR_BASE;
450 
451 	if (ndev->flags & IFF_PROMISC)
452 		r |= EMAC_RMR_PME;
453 	else if (ndev->flags & IFF_ALLMULTI ||
454 			 (netdev_mc_count(ndev) > EMAC_XAHT_SLOTS(dev)))
455 		r |= EMAC_RMR_PMME;
456 	else if (!netdev_mc_empty(ndev))
457 		r |= EMAC_RMR_MAE;
458 
459 	if (emac_has_feature(dev, EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE)) {
460 		r &= ~EMAC4_RMR_MJS_MASK;
461 		r |= EMAC4_RMR_MJS(ndev->mtu);
462 	}
463 
464 	return r;
465 }
466 
467 static u32 __emac_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
468 {
469 	u32 ret = EMAC_MR1_VLE | EMAC_MR1_IST | EMAC_MR1_TR0_MULT;
470 
471 	DBG2(dev, "__emac_calc_base_mr1" NL);
472 
473 	switch(tx_size) {
474 	case 2048:
475 		ret |= EMAC_MR1_TFS_2K;
476 		break;
477 	default:
478 		printk(KERN_WARNING "%s: Unknown Tx FIFO size %d\n",
479 		       dev->ndev->name, tx_size);
480 	}
481 
482 	switch(rx_size) {
483 	case 16384:
484 		ret |= EMAC_MR1_RFS_16K;
485 		break;
486 	case 4096:
487 		ret |= EMAC_MR1_RFS_4K;
488 		break;
489 	default:
490 		printk(KERN_WARNING "%s: Unknown Rx FIFO size %d\n",
491 		       dev->ndev->name, rx_size);
492 	}
493 
494 	return ret;
495 }
496 
497 static u32 __emac4_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
498 {
499 	u32 ret = EMAC_MR1_VLE | EMAC_MR1_IST | EMAC4_MR1_TR |
500 		EMAC4_MR1_OBCI(dev->opb_bus_freq / 1000000);
501 
502 	DBG2(dev, "__emac4_calc_base_mr1" NL);
503 
504 	switch(tx_size) {
505 	case 16384:
506 		ret |= EMAC4_MR1_TFS_16K;
507 		break;
508 	case 4096:
509 		ret |= EMAC4_MR1_TFS_4K;
510 		break;
511 	case 2048:
512 		ret |= EMAC4_MR1_TFS_2K;
513 		break;
514 	default:
515 		printk(KERN_WARNING "%s: Unknown Tx FIFO size %d\n",
516 		       dev->ndev->name, tx_size);
517 	}
518 
519 	switch(rx_size) {
520 	case 16384:
521 		ret |= EMAC4_MR1_RFS_16K;
522 		break;
523 	case 4096:
524 		ret |= EMAC4_MR1_RFS_4K;
525 		break;
526 	case 2048:
527 		ret |= EMAC4_MR1_RFS_2K;
528 		break;
529 	default:
530 		printk(KERN_WARNING "%s: Unknown Rx FIFO size %d\n",
531 		       dev->ndev->name, rx_size);
532 	}
533 
534 	return ret;
535 }
536 
537 static u32 emac_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
538 {
539 	return emac_has_feature(dev, EMAC_FTR_EMAC4) ?
540 		__emac4_calc_base_mr1(dev, tx_size, rx_size) :
541 		__emac_calc_base_mr1(dev, tx_size, rx_size);
542 }
543 
544 static inline u32 emac_calc_trtr(struct emac_instance *dev, unsigned int size)
545 {
546 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
547 		return ((size >> 6) - 1) << EMAC_TRTR_SHIFT_EMAC4;
548 	else
549 		return ((size >> 6) - 1) << EMAC_TRTR_SHIFT;
550 }
551 
552 static inline u32 emac_calc_rwmr(struct emac_instance *dev,
553 				 unsigned int low, unsigned int high)
554 {
555 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
556 		return (low << 22) | ( (high & 0x3ff) << 6);
557 	else
558 		return (low << 23) | ( (high & 0x1ff) << 7);
559 }
560 
561 static int emac_configure(struct emac_instance *dev)
562 {
563 	struct emac_regs __iomem *p = dev->emacp;
564 	struct net_device *ndev = dev->ndev;
565 	int tx_size, rx_size, link = netif_carrier_ok(dev->ndev);
566 	u32 r, mr1 = 0;
567 
568 	DBG(dev, "configure" NL);
569 
570 	if (!link) {
571 		out_be32(&p->mr1, in_be32(&p->mr1)
572 			 | EMAC_MR1_FDE | EMAC_MR1_ILE);
573 		udelay(100);
574 	} else if (emac_reset(dev) < 0)
575 		return -ETIMEDOUT;
576 
577 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
578 		tah_reset(dev->tah_dev);
579 
580 	DBG(dev, " link = %d duplex = %d, pause = %d, asym_pause = %d\n",
581 	    link, dev->phy.duplex, dev->phy.pause, dev->phy.asym_pause);
582 
583 	/* Default fifo sizes */
584 	tx_size = dev->tx_fifo_size;
585 	rx_size = dev->rx_fifo_size;
586 
587 	/* No link, force loopback */
588 	if (!link)
589 		mr1 = EMAC_MR1_FDE | EMAC_MR1_ILE;
590 
591 	/* Check for full duplex */
592 	else if (dev->phy.duplex == DUPLEX_FULL)
593 		mr1 |= EMAC_MR1_FDE | EMAC_MR1_MWSW_001;
594 
595 	/* Adjust fifo sizes, mr1 and timeouts based on link speed */
596 	dev->stop_timeout = STOP_TIMEOUT_10;
597 	switch (dev->phy.speed) {
598 	case SPEED_1000:
599 		if (emac_phy_gpcs(dev->phy.mode)) {
600 			mr1 |= EMAC_MR1_MF_1000GPCS | EMAC_MR1_MF_IPPA(
601 				(dev->phy.gpcs_address != 0xffffffff) ?
602 				 dev->phy.gpcs_address : dev->phy.address);
603 
604 			/* Put some arbitrary OUI, Manuf & Rev IDs so we can
605 			 * identify this GPCS PHY later.
606 			 */
607 			out_be32(&p->u1.emac4.ipcr, 0xdeadbeef);
608 		} else
609 			mr1 |= EMAC_MR1_MF_1000;
610 
611 		/* Extended fifo sizes */
612 		tx_size = dev->tx_fifo_size_gige;
613 		rx_size = dev->rx_fifo_size_gige;
614 
615 		if (dev->ndev->mtu > ETH_DATA_LEN) {
616 			if (emac_has_feature(dev, EMAC_FTR_EMAC4))
617 				mr1 |= EMAC4_MR1_JPSM;
618 			else
619 				mr1 |= EMAC_MR1_JPSM;
620 			dev->stop_timeout = STOP_TIMEOUT_1000_JUMBO;
621 		} else
622 			dev->stop_timeout = STOP_TIMEOUT_1000;
623 		break;
624 	case SPEED_100:
625 		mr1 |= EMAC_MR1_MF_100;
626 		dev->stop_timeout = STOP_TIMEOUT_100;
627 		break;
628 	default: /* make gcc happy */
629 		break;
630 	}
631 
632 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
633 		rgmii_set_speed(dev->rgmii_dev, dev->rgmii_port,
634 				dev->phy.speed);
635 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
636 		zmii_set_speed(dev->zmii_dev, dev->zmii_port, dev->phy.speed);
637 
638 	/* on 40x erratum forces us to NOT use integrated flow control,
639 	 * let's hope it works on 44x ;)
640 	 */
641 	if (!emac_has_feature(dev, EMAC_FTR_NO_FLOW_CONTROL_40x) &&
642 	    dev->phy.duplex == DUPLEX_FULL) {
643 		if (dev->phy.pause)
644 			mr1 |= EMAC_MR1_EIFC | EMAC_MR1_APP;
645 		else if (dev->phy.asym_pause)
646 			mr1 |= EMAC_MR1_APP;
647 	}
648 
649 	/* Add base settings & fifo sizes & program MR1 */
650 	mr1 |= emac_calc_base_mr1(dev, tx_size, rx_size);
651 	out_be32(&p->mr1, mr1);
652 
653 	/* Set individual MAC address */
654 	out_be32(&p->iahr, (ndev->dev_addr[0] << 8) | ndev->dev_addr[1]);
655 	out_be32(&p->ialr, (ndev->dev_addr[2] << 24) |
656 		 (ndev->dev_addr[3] << 16) | (ndev->dev_addr[4] << 8) |
657 		 ndev->dev_addr[5]);
658 
659 	/* VLAN Tag Protocol ID */
660 	out_be32(&p->vtpid, 0x8100);
661 
662 	/* Receive mode register */
663 	r = emac_iff2rmr(ndev);
664 	if (r & EMAC_RMR_MAE)
665 		emac_hash_mc(dev);
666 	out_be32(&p->rmr, r);
667 
668 	/* FIFOs thresholds */
669 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
670 		r = EMAC4_TMR1((dev->mal_burst_size / dev->fifo_entry_size) + 1,
671 			       tx_size / 2 / dev->fifo_entry_size);
672 	else
673 		r = EMAC_TMR1((dev->mal_burst_size / dev->fifo_entry_size) + 1,
674 			      tx_size / 2 / dev->fifo_entry_size);
675 	out_be32(&p->tmr1, r);
676 	out_be32(&p->trtr, emac_calc_trtr(dev, tx_size / 2));
677 
678 	/* PAUSE frame is sent when RX FIFO reaches its high-water mark,
679 	   there should be still enough space in FIFO to allow the our link
680 	   partner time to process this frame and also time to send PAUSE
681 	   frame itself.
682 
683 	   Here is the worst case scenario for the RX FIFO "headroom"
684 	   (from "The Switch Book") (100Mbps, without preamble, inter-frame gap):
685 
686 	   1) One maximum-length frame on TX                    1522 bytes
687 	   2) One PAUSE frame time                                64 bytes
688 	   3) PAUSE frame decode time allowance                   64 bytes
689 	   4) One maximum-length frame on RX                    1522 bytes
690 	   5) Round-trip propagation delay of the link (100Mb)    15 bytes
691 	   ----------
692 	   3187 bytes
693 
694 	   I chose to set high-water mark to RX_FIFO_SIZE / 4 (1024 bytes)
695 	   low-water mark  to RX_FIFO_SIZE / 8 (512 bytes)
696 	 */
697 	r = emac_calc_rwmr(dev, rx_size / 8 / dev->fifo_entry_size,
698 			   rx_size / 4 / dev->fifo_entry_size);
699 	out_be32(&p->rwmr, r);
700 
701 	/* Set PAUSE timer to the maximum */
702 	out_be32(&p->ptr, 0xffff);
703 
704 	/* IRQ sources */
705 	r = EMAC_ISR_OVR | EMAC_ISR_BP | EMAC_ISR_SE |
706 		EMAC_ISR_ALE | EMAC_ISR_BFCS | EMAC_ISR_PTLE | EMAC_ISR_ORE |
707 		EMAC_ISR_IRE | EMAC_ISR_TE;
708 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
709 	    r |= EMAC4_ISR_TXPE | EMAC4_ISR_RXPE /* | EMAC4_ISR_TXUE |
710 						  EMAC4_ISR_RXOE | */;
711 	out_be32(&p->iser,  r);
712 
713 	/* We need to take GPCS PHY out of isolate mode after EMAC reset */
714 	if (emac_phy_gpcs(dev->phy.mode)) {
715 		if (dev->phy.gpcs_address != 0xffffffff)
716 			emac_mii_reset_gpcs(&dev->phy);
717 		else
718 			emac_mii_reset_phy(&dev->phy);
719 	}
720 
721 	return 0;
722 }
723 
724 static void emac_reinitialize(struct emac_instance *dev)
725 {
726 	DBG(dev, "reinitialize" NL);
727 
728 	emac_netif_stop(dev);
729 	if (!emac_configure(dev)) {
730 		emac_tx_enable(dev);
731 		emac_rx_enable(dev);
732 	}
733 	emac_netif_start(dev);
734 }
735 
736 static void emac_full_tx_reset(struct emac_instance *dev)
737 {
738 	DBG(dev, "full_tx_reset" NL);
739 
740 	emac_tx_disable(dev);
741 	mal_disable_tx_channel(dev->mal, dev->mal_tx_chan);
742 	emac_clean_tx_ring(dev);
743 	dev->tx_cnt = dev->tx_slot = dev->ack_slot = 0;
744 
745 	emac_configure(dev);
746 
747 	mal_enable_tx_channel(dev->mal, dev->mal_tx_chan);
748 	emac_tx_enable(dev);
749 	emac_rx_enable(dev);
750 }
751 
752 static void emac_reset_work(struct work_struct *work)
753 {
754 	struct emac_instance *dev = container_of(work, struct emac_instance, reset_work);
755 
756 	DBG(dev, "reset_work" NL);
757 
758 	mutex_lock(&dev->link_lock);
759 	if (dev->opened) {
760 		emac_netif_stop(dev);
761 		emac_full_tx_reset(dev);
762 		emac_netif_start(dev);
763 	}
764 	mutex_unlock(&dev->link_lock);
765 }
766 
767 static void emac_tx_timeout(struct net_device *ndev)
768 {
769 	struct emac_instance *dev = netdev_priv(ndev);
770 
771 	DBG(dev, "tx_timeout" NL);
772 
773 	schedule_work(&dev->reset_work);
774 }
775 
776 
777 static inline int emac_phy_done(struct emac_instance *dev, u32 stacr)
778 {
779 	int done = !!(stacr & EMAC_STACR_OC);
780 
781 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
782 		done = !done;
783 
784 	return done;
785 };
786 
787 static int __emac_mdio_read(struct emac_instance *dev, u8 id, u8 reg)
788 {
789 	struct emac_regs __iomem *p = dev->emacp;
790 	u32 r = 0;
791 	int n, err = -ETIMEDOUT;
792 
793 	mutex_lock(&dev->mdio_lock);
794 
795 	DBG2(dev, "mdio_read(%02x,%02x)" NL, id, reg);
796 
797 	/* Enable proper MDIO port */
798 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
799 		zmii_get_mdio(dev->zmii_dev, dev->zmii_port);
800 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
801 		rgmii_get_mdio(dev->rgmii_dev, dev->rgmii_port);
802 
803 	/* Wait for management interface to become idle */
804 	n = 20;
805 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
806 		udelay(1);
807 		if (!--n) {
808 			DBG2(dev, " -> timeout wait idle\n");
809 			goto bail;
810 		}
811 	}
812 
813 	/* Issue read command */
814 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
815 		r = EMAC4_STACR_BASE(dev->opb_bus_freq);
816 	else
817 		r = EMAC_STACR_BASE(dev->opb_bus_freq);
818 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
819 		r |= EMAC_STACR_OC;
820 	if (emac_has_feature(dev, EMAC_FTR_HAS_NEW_STACR))
821 		r |= EMACX_STACR_STAC_READ;
822 	else
823 		r |= EMAC_STACR_STAC_READ;
824 	r |= (reg & EMAC_STACR_PRA_MASK)
825 		| ((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT);
826 	out_be32(&p->stacr, r);
827 
828 	/* Wait for read to complete */
829 	n = 200;
830 	while (!emac_phy_done(dev, (r = in_be32(&p->stacr)))) {
831 		udelay(1);
832 		if (!--n) {
833 			DBG2(dev, " -> timeout wait complete\n");
834 			goto bail;
835 		}
836 	}
837 
838 	if (unlikely(r & EMAC_STACR_PHYE)) {
839 		DBG(dev, "mdio_read(%02x, %02x) failed" NL, id, reg);
840 		err = -EREMOTEIO;
841 		goto bail;
842 	}
843 
844 	r = ((r >> EMAC_STACR_PHYD_SHIFT) & EMAC_STACR_PHYD_MASK);
845 
846 	DBG2(dev, "mdio_read -> %04x" NL, r);
847 	err = 0;
848  bail:
849 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
850 		rgmii_put_mdio(dev->rgmii_dev, dev->rgmii_port);
851 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
852 		zmii_put_mdio(dev->zmii_dev, dev->zmii_port);
853 	mutex_unlock(&dev->mdio_lock);
854 
855 	return err == 0 ? r : err;
856 }
857 
858 static void __emac_mdio_write(struct emac_instance *dev, u8 id, u8 reg,
859 			      u16 val)
860 {
861 	struct emac_regs __iomem *p = dev->emacp;
862 	u32 r = 0;
863 	int n, err = -ETIMEDOUT;
864 
865 	mutex_lock(&dev->mdio_lock);
866 
867 	DBG2(dev, "mdio_write(%02x,%02x,%04x)" NL, id, reg, val);
868 
869 	/* Enable proper MDIO port */
870 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
871 		zmii_get_mdio(dev->zmii_dev, dev->zmii_port);
872 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
873 		rgmii_get_mdio(dev->rgmii_dev, dev->rgmii_port);
874 
875 	/* Wait for management interface to be idle */
876 	n = 20;
877 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
878 		udelay(1);
879 		if (!--n) {
880 			DBG2(dev, " -> timeout wait idle\n");
881 			goto bail;
882 		}
883 	}
884 
885 	/* Issue write command */
886 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
887 		r = EMAC4_STACR_BASE(dev->opb_bus_freq);
888 	else
889 		r = EMAC_STACR_BASE(dev->opb_bus_freq);
890 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
891 		r |= EMAC_STACR_OC;
892 	if (emac_has_feature(dev, EMAC_FTR_HAS_NEW_STACR))
893 		r |= EMACX_STACR_STAC_WRITE;
894 	else
895 		r |= EMAC_STACR_STAC_WRITE;
896 	r |= (reg & EMAC_STACR_PRA_MASK) |
897 		((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT) |
898 		(val << EMAC_STACR_PHYD_SHIFT);
899 	out_be32(&p->stacr, r);
900 
901 	/* Wait for write to complete */
902 	n = 200;
903 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
904 		udelay(1);
905 		if (!--n) {
906 			DBG2(dev, " -> timeout wait complete\n");
907 			goto bail;
908 		}
909 	}
910 	err = 0;
911  bail:
912 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
913 		rgmii_put_mdio(dev->rgmii_dev, dev->rgmii_port);
914 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
915 		zmii_put_mdio(dev->zmii_dev, dev->zmii_port);
916 	mutex_unlock(&dev->mdio_lock);
917 }
918 
919 static int emac_mdio_read(struct net_device *ndev, int id, int reg)
920 {
921 	struct emac_instance *dev = netdev_priv(ndev);
922 	int res;
923 
924 	res = __emac_mdio_read((dev->mdio_instance &&
925 				dev->phy.gpcs_address != id) ?
926 				dev->mdio_instance : dev,
927 			       (u8) id, (u8) reg);
928 	return res;
929 }
930 
931 static void emac_mdio_write(struct net_device *ndev, int id, int reg, int val)
932 {
933 	struct emac_instance *dev = netdev_priv(ndev);
934 
935 	__emac_mdio_write((dev->mdio_instance &&
936 			   dev->phy.gpcs_address != id) ?
937 			   dev->mdio_instance : dev,
938 			  (u8) id, (u8) reg, (u16) val);
939 }
940 
941 /* Tx lock BH */
942 static void __emac_set_multicast_list(struct emac_instance *dev)
943 {
944 	struct emac_regs __iomem *p = dev->emacp;
945 	u32 rmr = emac_iff2rmr(dev->ndev);
946 
947 	DBG(dev, "__multicast %08x" NL, rmr);
948 
949 	/* I decided to relax register access rules here to avoid
950 	 * full EMAC reset.
951 	 *
952 	 * There is a real problem with EMAC4 core if we use MWSW_001 bit
953 	 * in MR1 register and do a full EMAC reset.
954 	 * One TX BD status update is delayed and, after EMAC reset, it
955 	 * never happens, resulting in TX hung (it'll be recovered by TX
956 	 * timeout handler eventually, but this is just gross).
957 	 * So we either have to do full TX reset or try to cheat here :)
958 	 *
959 	 * The only required change is to RX mode register, so I *think* all
960 	 * we need is just to stop RX channel. This seems to work on all
961 	 * tested SoCs.                                                --ebs
962 	 *
963 	 * If we need the full reset, we might just trigger the workqueue
964 	 * and do it async... a bit nasty but should work --BenH
965 	 */
966 	dev->mcast_pending = 0;
967 	emac_rx_disable(dev);
968 	if (rmr & EMAC_RMR_MAE)
969 		emac_hash_mc(dev);
970 	out_be32(&p->rmr, rmr);
971 	emac_rx_enable(dev);
972 }
973 
974 /* Tx lock BH */
975 static void emac_set_multicast_list(struct net_device *ndev)
976 {
977 	struct emac_instance *dev = netdev_priv(ndev);
978 
979 	DBG(dev, "multicast" NL);
980 
981 	BUG_ON(!netif_running(dev->ndev));
982 
983 	if (dev->no_mcast) {
984 		dev->mcast_pending = 1;
985 		return;
986 	}
987 	__emac_set_multicast_list(dev);
988 }
989 
990 static int emac_resize_rx_ring(struct emac_instance *dev, int new_mtu)
991 {
992 	int rx_sync_size = emac_rx_sync_size(new_mtu);
993 	int rx_skb_size = emac_rx_skb_size(new_mtu);
994 	int i, ret = 0;
995 	int mr1_jumbo_bit_change = 0;
996 
997 	mutex_lock(&dev->link_lock);
998 	emac_netif_stop(dev);
999 	emac_rx_disable(dev);
1000 	mal_disable_rx_channel(dev->mal, dev->mal_rx_chan);
1001 
1002 	if (dev->rx_sg_skb) {
1003 		++dev->estats.rx_dropped_resize;
1004 		dev_kfree_skb(dev->rx_sg_skb);
1005 		dev->rx_sg_skb = NULL;
1006 	}
1007 
1008 	/* Make a first pass over RX ring and mark BDs ready, dropping
1009 	 * non-processed packets on the way. We need this as a separate pass
1010 	 * to simplify error recovery in the case of allocation failure later.
1011 	 */
1012 	for (i = 0; i < NUM_RX_BUFF; ++i) {
1013 		if (dev->rx_desc[i].ctrl & MAL_RX_CTRL_FIRST)
1014 			++dev->estats.rx_dropped_resize;
1015 
1016 		dev->rx_desc[i].data_len = 0;
1017 		dev->rx_desc[i].ctrl = MAL_RX_CTRL_EMPTY |
1018 		    (i == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1019 	}
1020 
1021 	/* Reallocate RX ring only if bigger skb buffers are required */
1022 	if (rx_skb_size <= dev->rx_skb_size)
1023 		goto skip;
1024 
1025 	/* Second pass, allocate new skbs */
1026 	for (i = 0; i < NUM_RX_BUFF; ++i) {
1027 		struct sk_buff *skb = alloc_skb(rx_skb_size, GFP_ATOMIC);
1028 		if (!skb) {
1029 			ret = -ENOMEM;
1030 			goto oom;
1031 		}
1032 
1033 		BUG_ON(!dev->rx_skb[i]);
1034 		dev_kfree_skb(dev->rx_skb[i]);
1035 
1036 		skb_reserve(skb, EMAC_RX_SKB_HEADROOM + 2);
1037 		dev->rx_desc[i].data_ptr =
1038 		    dma_map_single(&dev->ofdev->dev, skb->data - 2, rx_sync_size,
1039 				   DMA_FROM_DEVICE) + 2;
1040 		dev->rx_skb[i] = skb;
1041 	}
1042  skip:
1043 	/* Check if we need to change "Jumbo" bit in MR1 */
1044 	if (emac_has_feature(dev, EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE)) {
1045 		mr1_jumbo_bit_change = (new_mtu > ETH_DATA_LEN) ||
1046 				(dev->ndev->mtu > ETH_DATA_LEN);
1047 	} else {
1048 		mr1_jumbo_bit_change = (new_mtu > ETH_DATA_LEN) ^
1049 				(dev->ndev->mtu > ETH_DATA_LEN);
1050 	}
1051 
1052 	if (mr1_jumbo_bit_change) {
1053 		/* This is to prevent starting RX channel in emac_rx_enable() */
1054 		set_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1055 
1056 		dev->ndev->mtu = new_mtu;
1057 		emac_full_tx_reset(dev);
1058 	}
1059 
1060 	mal_set_rcbs(dev->mal, dev->mal_rx_chan, emac_rx_size(new_mtu));
1061  oom:
1062 	/* Restart RX */
1063 	clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1064 	dev->rx_slot = 0;
1065 	mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1066 	emac_rx_enable(dev);
1067 	emac_netif_start(dev);
1068 	mutex_unlock(&dev->link_lock);
1069 
1070 	return ret;
1071 }
1072 
1073 /* Process ctx, rtnl_lock semaphore */
1074 static int emac_change_mtu(struct net_device *ndev, int new_mtu)
1075 {
1076 	struct emac_instance *dev = netdev_priv(ndev);
1077 	int ret = 0;
1078 
1079 	if (new_mtu < EMAC_MIN_MTU || new_mtu > dev->max_mtu)
1080 		return -EINVAL;
1081 
1082 	DBG(dev, "change_mtu(%d)" NL, new_mtu);
1083 
1084 	if (netif_running(ndev)) {
1085 		/* Check if we really need to reinitialize RX ring */
1086 		if (emac_rx_skb_size(ndev->mtu) != emac_rx_skb_size(new_mtu))
1087 			ret = emac_resize_rx_ring(dev, new_mtu);
1088 	}
1089 
1090 	if (!ret) {
1091 		ndev->mtu = new_mtu;
1092 		dev->rx_skb_size = emac_rx_skb_size(new_mtu);
1093 		dev->rx_sync_size = emac_rx_sync_size(new_mtu);
1094 	}
1095 
1096 	return ret;
1097 }
1098 
1099 static void emac_clean_tx_ring(struct emac_instance *dev)
1100 {
1101 	int i;
1102 
1103 	for (i = 0; i < NUM_TX_BUFF; ++i) {
1104 		if (dev->tx_skb[i]) {
1105 			dev_kfree_skb(dev->tx_skb[i]);
1106 			dev->tx_skb[i] = NULL;
1107 			if (dev->tx_desc[i].ctrl & MAL_TX_CTRL_READY)
1108 				++dev->estats.tx_dropped;
1109 		}
1110 		dev->tx_desc[i].ctrl = 0;
1111 		dev->tx_desc[i].data_ptr = 0;
1112 	}
1113 }
1114 
1115 static void emac_clean_rx_ring(struct emac_instance *dev)
1116 {
1117 	int i;
1118 
1119 	for (i = 0; i < NUM_RX_BUFF; ++i)
1120 		if (dev->rx_skb[i]) {
1121 			dev->rx_desc[i].ctrl = 0;
1122 			dev_kfree_skb(dev->rx_skb[i]);
1123 			dev->rx_skb[i] = NULL;
1124 			dev->rx_desc[i].data_ptr = 0;
1125 		}
1126 
1127 	if (dev->rx_sg_skb) {
1128 		dev_kfree_skb(dev->rx_sg_skb);
1129 		dev->rx_sg_skb = NULL;
1130 	}
1131 }
1132 
1133 static inline int emac_alloc_rx_skb(struct emac_instance *dev, int slot,
1134 				    gfp_t flags)
1135 {
1136 	struct sk_buff *skb = alloc_skb(dev->rx_skb_size, flags);
1137 	if (unlikely(!skb))
1138 		return -ENOMEM;
1139 
1140 	dev->rx_skb[slot] = skb;
1141 	dev->rx_desc[slot].data_len = 0;
1142 
1143 	skb_reserve(skb, EMAC_RX_SKB_HEADROOM + 2);
1144 	dev->rx_desc[slot].data_ptr =
1145 	    dma_map_single(&dev->ofdev->dev, skb->data - 2, dev->rx_sync_size,
1146 			   DMA_FROM_DEVICE) + 2;
1147 	wmb();
1148 	dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1149 	    (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1150 
1151 	return 0;
1152 }
1153 
1154 static void emac_print_link_status(struct emac_instance *dev)
1155 {
1156 	if (netif_carrier_ok(dev->ndev))
1157 		printk(KERN_INFO "%s: link is up, %d %s%s\n",
1158 		       dev->ndev->name, dev->phy.speed,
1159 		       dev->phy.duplex == DUPLEX_FULL ? "FDX" : "HDX",
1160 		       dev->phy.pause ? ", pause enabled" :
1161 		       dev->phy.asym_pause ? ", asymmetric pause enabled" : "");
1162 	else
1163 		printk(KERN_INFO "%s: link is down\n", dev->ndev->name);
1164 }
1165 
1166 /* Process ctx, rtnl_lock semaphore */
1167 static int emac_open(struct net_device *ndev)
1168 {
1169 	struct emac_instance *dev = netdev_priv(ndev);
1170 	int err, i;
1171 
1172 	DBG(dev, "open" NL);
1173 
1174 	/* Setup error IRQ handler */
1175 	err = request_irq(dev->emac_irq, emac_irq, 0, "EMAC", dev);
1176 	if (err) {
1177 		printk(KERN_ERR "%s: failed to request IRQ %d\n",
1178 		       ndev->name, dev->emac_irq);
1179 		return err;
1180 	}
1181 
1182 	/* Allocate RX ring */
1183 	for (i = 0; i < NUM_RX_BUFF; ++i)
1184 		if (emac_alloc_rx_skb(dev, i, GFP_KERNEL)) {
1185 			printk(KERN_ERR "%s: failed to allocate RX ring\n",
1186 			       ndev->name);
1187 			goto oom;
1188 		}
1189 
1190 	dev->tx_cnt = dev->tx_slot = dev->ack_slot = dev->rx_slot = 0;
1191 	clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1192 	dev->rx_sg_skb = NULL;
1193 
1194 	mutex_lock(&dev->link_lock);
1195 	dev->opened = 1;
1196 
1197 	/* Start PHY polling now.
1198 	 */
1199 	if (dev->phy.address >= 0) {
1200 		int link_poll_interval;
1201 		if (dev->phy.def->ops->poll_link(&dev->phy)) {
1202 			dev->phy.def->ops->read_link(&dev->phy);
1203 			emac_rx_clk_default(dev);
1204 			netif_carrier_on(dev->ndev);
1205 			link_poll_interval = PHY_POLL_LINK_ON;
1206 		} else {
1207 			emac_rx_clk_tx(dev);
1208 			netif_carrier_off(dev->ndev);
1209 			link_poll_interval = PHY_POLL_LINK_OFF;
1210 		}
1211 		dev->link_polling = 1;
1212 		wmb();
1213 		schedule_delayed_work(&dev->link_work, link_poll_interval);
1214 		emac_print_link_status(dev);
1215 	} else
1216 		netif_carrier_on(dev->ndev);
1217 
1218 	/* Required for Pause packet support in EMAC */
1219 	dev_mc_add_global(ndev, default_mcast_addr);
1220 
1221 	emac_configure(dev);
1222 	mal_poll_add(dev->mal, &dev->commac);
1223 	mal_enable_tx_channel(dev->mal, dev->mal_tx_chan);
1224 	mal_set_rcbs(dev->mal, dev->mal_rx_chan, emac_rx_size(ndev->mtu));
1225 	mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1226 	emac_tx_enable(dev);
1227 	emac_rx_enable(dev);
1228 	emac_netif_start(dev);
1229 
1230 	mutex_unlock(&dev->link_lock);
1231 
1232 	return 0;
1233  oom:
1234 	emac_clean_rx_ring(dev);
1235 	free_irq(dev->emac_irq, dev);
1236 
1237 	return -ENOMEM;
1238 }
1239 
1240 /* BHs disabled */
1241 #if 0
1242 static int emac_link_differs(struct emac_instance *dev)
1243 {
1244 	u32 r = in_be32(&dev->emacp->mr1);
1245 
1246 	int duplex = r & EMAC_MR1_FDE ? DUPLEX_FULL : DUPLEX_HALF;
1247 	int speed, pause, asym_pause;
1248 
1249 	if (r & EMAC_MR1_MF_1000)
1250 		speed = SPEED_1000;
1251 	else if (r & EMAC_MR1_MF_100)
1252 		speed = SPEED_100;
1253 	else
1254 		speed = SPEED_10;
1255 
1256 	switch (r & (EMAC_MR1_EIFC | EMAC_MR1_APP)) {
1257 	case (EMAC_MR1_EIFC | EMAC_MR1_APP):
1258 		pause = 1;
1259 		asym_pause = 0;
1260 		break;
1261 	case EMAC_MR1_APP:
1262 		pause = 0;
1263 		asym_pause = 1;
1264 		break;
1265 	default:
1266 		pause = asym_pause = 0;
1267 	}
1268 	return speed != dev->phy.speed || duplex != dev->phy.duplex ||
1269 	    pause != dev->phy.pause || asym_pause != dev->phy.asym_pause;
1270 }
1271 #endif
1272 
1273 static void emac_link_timer(struct work_struct *work)
1274 {
1275 	struct emac_instance *dev =
1276 		container_of(to_delayed_work(work),
1277 			     struct emac_instance, link_work);
1278 	int link_poll_interval;
1279 
1280 	mutex_lock(&dev->link_lock);
1281 	DBG2(dev, "link timer" NL);
1282 
1283 	if (!dev->opened)
1284 		goto bail;
1285 
1286 	if (dev->phy.def->ops->poll_link(&dev->phy)) {
1287 		if (!netif_carrier_ok(dev->ndev)) {
1288 			emac_rx_clk_default(dev);
1289 			/* Get new link parameters */
1290 			dev->phy.def->ops->read_link(&dev->phy);
1291 
1292 			netif_carrier_on(dev->ndev);
1293 			emac_netif_stop(dev);
1294 			emac_full_tx_reset(dev);
1295 			emac_netif_start(dev);
1296 			emac_print_link_status(dev);
1297 		}
1298 		link_poll_interval = PHY_POLL_LINK_ON;
1299 	} else {
1300 		if (netif_carrier_ok(dev->ndev)) {
1301 			emac_rx_clk_tx(dev);
1302 			netif_carrier_off(dev->ndev);
1303 			netif_tx_disable(dev->ndev);
1304 			emac_reinitialize(dev);
1305 			emac_print_link_status(dev);
1306 		}
1307 		link_poll_interval = PHY_POLL_LINK_OFF;
1308 	}
1309 	schedule_delayed_work(&dev->link_work, link_poll_interval);
1310  bail:
1311 	mutex_unlock(&dev->link_lock);
1312 }
1313 
1314 static void emac_force_link_update(struct emac_instance *dev)
1315 {
1316 	netif_carrier_off(dev->ndev);
1317 	smp_rmb();
1318 	if (dev->link_polling) {
1319 		cancel_delayed_work_sync(&dev->link_work);
1320 		if (dev->link_polling)
1321 			schedule_delayed_work(&dev->link_work,  PHY_POLL_LINK_OFF);
1322 	}
1323 }
1324 
1325 /* Process ctx, rtnl_lock semaphore */
1326 static int emac_close(struct net_device *ndev)
1327 {
1328 	struct emac_instance *dev = netdev_priv(ndev);
1329 
1330 	DBG(dev, "close" NL);
1331 
1332 	if (dev->phy.address >= 0) {
1333 		dev->link_polling = 0;
1334 		cancel_delayed_work_sync(&dev->link_work);
1335 	}
1336 	mutex_lock(&dev->link_lock);
1337 	emac_netif_stop(dev);
1338 	dev->opened = 0;
1339 	mutex_unlock(&dev->link_lock);
1340 
1341 	emac_rx_disable(dev);
1342 	emac_tx_disable(dev);
1343 	mal_disable_rx_channel(dev->mal, dev->mal_rx_chan);
1344 	mal_disable_tx_channel(dev->mal, dev->mal_tx_chan);
1345 	mal_poll_del(dev->mal, &dev->commac);
1346 
1347 	emac_clean_tx_ring(dev);
1348 	emac_clean_rx_ring(dev);
1349 
1350 	free_irq(dev->emac_irq, dev);
1351 
1352 	netif_carrier_off(ndev);
1353 
1354 	return 0;
1355 }
1356 
1357 static inline u16 emac_tx_csum(struct emac_instance *dev,
1358 			       struct sk_buff *skb)
1359 {
1360 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH) &&
1361 		(skb->ip_summed == CHECKSUM_PARTIAL)) {
1362 		++dev->stats.tx_packets_csum;
1363 		return EMAC_TX_CTRL_TAH_CSUM;
1364 	}
1365 	return 0;
1366 }
1367 
1368 static inline int emac_xmit_finish(struct emac_instance *dev, int len)
1369 {
1370 	struct emac_regs __iomem *p = dev->emacp;
1371 	struct net_device *ndev = dev->ndev;
1372 
1373 	/* Send the packet out. If the if makes a significant perf
1374 	 * difference, then we can store the TMR0 value in "dev"
1375 	 * instead
1376 	 */
1377 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
1378 		out_be32(&p->tmr0, EMAC4_TMR0_XMIT);
1379 	else
1380 		out_be32(&p->tmr0, EMAC_TMR0_XMIT);
1381 
1382 	if (unlikely(++dev->tx_cnt == NUM_TX_BUFF)) {
1383 		netif_stop_queue(ndev);
1384 		DBG2(dev, "stopped TX queue" NL);
1385 	}
1386 
1387 	ndev->trans_start = jiffies;
1388 	++dev->stats.tx_packets;
1389 	dev->stats.tx_bytes += len;
1390 
1391 	return NETDEV_TX_OK;
1392 }
1393 
1394 /* Tx lock BH */
1395 static int emac_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1396 {
1397 	struct emac_instance *dev = netdev_priv(ndev);
1398 	unsigned int len = skb->len;
1399 	int slot;
1400 
1401 	u16 ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1402 	    MAL_TX_CTRL_LAST | emac_tx_csum(dev, skb);
1403 
1404 	slot = dev->tx_slot++;
1405 	if (dev->tx_slot == NUM_TX_BUFF) {
1406 		dev->tx_slot = 0;
1407 		ctrl |= MAL_TX_CTRL_WRAP;
1408 	}
1409 
1410 	DBG2(dev, "xmit(%u) %d" NL, len, slot);
1411 
1412 	dev->tx_skb[slot] = skb;
1413 	dev->tx_desc[slot].data_ptr = dma_map_single(&dev->ofdev->dev,
1414 						     skb->data, len,
1415 						     DMA_TO_DEVICE);
1416 	dev->tx_desc[slot].data_len = (u16) len;
1417 	wmb();
1418 	dev->tx_desc[slot].ctrl = ctrl;
1419 
1420 	return emac_xmit_finish(dev, len);
1421 }
1422 
1423 static inline int emac_xmit_split(struct emac_instance *dev, int slot,
1424 				  u32 pd, int len, int last, u16 base_ctrl)
1425 {
1426 	while (1) {
1427 		u16 ctrl = base_ctrl;
1428 		int chunk = min(len, MAL_MAX_TX_SIZE);
1429 		len -= chunk;
1430 
1431 		slot = (slot + 1) % NUM_TX_BUFF;
1432 
1433 		if (last && !len)
1434 			ctrl |= MAL_TX_CTRL_LAST;
1435 		if (slot == NUM_TX_BUFF - 1)
1436 			ctrl |= MAL_TX_CTRL_WRAP;
1437 
1438 		dev->tx_skb[slot] = NULL;
1439 		dev->tx_desc[slot].data_ptr = pd;
1440 		dev->tx_desc[slot].data_len = (u16) chunk;
1441 		dev->tx_desc[slot].ctrl = ctrl;
1442 		++dev->tx_cnt;
1443 
1444 		if (!len)
1445 			break;
1446 
1447 		pd += chunk;
1448 	}
1449 	return slot;
1450 }
1451 
1452 /* Tx lock BH disabled (SG version for TAH equipped EMACs) */
1453 static int emac_start_xmit_sg(struct sk_buff *skb, struct net_device *ndev)
1454 {
1455 	struct emac_instance *dev = netdev_priv(ndev);
1456 	int nr_frags = skb_shinfo(skb)->nr_frags;
1457 	int len = skb->len, chunk;
1458 	int slot, i;
1459 	u16 ctrl;
1460 	u32 pd;
1461 
1462 	/* This is common "fast" path */
1463 	if (likely(!nr_frags && len <= MAL_MAX_TX_SIZE))
1464 		return emac_start_xmit(skb, ndev);
1465 
1466 	len -= skb->data_len;
1467 
1468 	/* Note, this is only an *estimation*, we can still run out of empty
1469 	 * slots because of the additional fragmentation into
1470 	 * MAL_MAX_TX_SIZE-sized chunks
1471 	 */
1472 	if (unlikely(dev->tx_cnt + nr_frags + mal_tx_chunks(len) > NUM_TX_BUFF))
1473 		goto stop_queue;
1474 
1475 	ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1476 	    emac_tx_csum(dev, skb);
1477 	slot = dev->tx_slot;
1478 
1479 	/* skb data */
1480 	dev->tx_skb[slot] = NULL;
1481 	chunk = min(len, MAL_MAX_TX_SIZE);
1482 	dev->tx_desc[slot].data_ptr = pd =
1483 	    dma_map_single(&dev->ofdev->dev, skb->data, len, DMA_TO_DEVICE);
1484 	dev->tx_desc[slot].data_len = (u16) chunk;
1485 	len -= chunk;
1486 	if (unlikely(len))
1487 		slot = emac_xmit_split(dev, slot, pd + chunk, len, !nr_frags,
1488 				       ctrl);
1489 	/* skb fragments */
1490 	for (i = 0; i < nr_frags; ++i) {
1491 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1492 		len = skb_frag_size(frag);
1493 
1494 		if (unlikely(dev->tx_cnt + mal_tx_chunks(len) >= NUM_TX_BUFF))
1495 			goto undo_frame;
1496 
1497 		pd = skb_frag_dma_map(&dev->ofdev->dev, frag, 0, len,
1498 				      DMA_TO_DEVICE);
1499 
1500 		slot = emac_xmit_split(dev, slot, pd, len, i == nr_frags - 1,
1501 				       ctrl);
1502 	}
1503 
1504 	DBG2(dev, "xmit_sg(%u) %d - %d" NL, skb->len, dev->tx_slot, slot);
1505 
1506 	/* Attach skb to the last slot so we don't release it too early */
1507 	dev->tx_skb[slot] = skb;
1508 
1509 	/* Send the packet out */
1510 	if (dev->tx_slot == NUM_TX_BUFF - 1)
1511 		ctrl |= MAL_TX_CTRL_WRAP;
1512 	wmb();
1513 	dev->tx_desc[dev->tx_slot].ctrl = ctrl;
1514 	dev->tx_slot = (slot + 1) % NUM_TX_BUFF;
1515 
1516 	return emac_xmit_finish(dev, skb->len);
1517 
1518  undo_frame:
1519 	/* Well, too bad. Our previous estimation was overly optimistic.
1520 	 * Undo everything.
1521 	 */
1522 	while (slot != dev->tx_slot) {
1523 		dev->tx_desc[slot].ctrl = 0;
1524 		--dev->tx_cnt;
1525 		if (--slot < 0)
1526 			slot = NUM_TX_BUFF - 1;
1527 	}
1528 	++dev->estats.tx_undo;
1529 
1530  stop_queue:
1531 	netif_stop_queue(ndev);
1532 	DBG2(dev, "stopped TX queue" NL);
1533 	return NETDEV_TX_BUSY;
1534 }
1535 
1536 /* Tx lock BHs */
1537 static void emac_parse_tx_error(struct emac_instance *dev, u16 ctrl)
1538 {
1539 	struct emac_error_stats *st = &dev->estats;
1540 
1541 	DBG(dev, "BD TX error %04x" NL, ctrl);
1542 
1543 	++st->tx_bd_errors;
1544 	if (ctrl & EMAC_TX_ST_BFCS)
1545 		++st->tx_bd_bad_fcs;
1546 	if (ctrl & EMAC_TX_ST_LCS)
1547 		++st->tx_bd_carrier_loss;
1548 	if (ctrl & EMAC_TX_ST_ED)
1549 		++st->tx_bd_excessive_deferral;
1550 	if (ctrl & EMAC_TX_ST_EC)
1551 		++st->tx_bd_excessive_collisions;
1552 	if (ctrl & EMAC_TX_ST_LC)
1553 		++st->tx_bd_late_collision;
1554 	if (ctrl & EMAC_TX_ST_MC)
1555 		++st->tx_bd_multple_collisions;
1556 	if (ctrl & EMAC_TX_ST_SC)
1557 		++st->tx_bd_single_collision;
1558 	if (ctrl & EMAC_TX_ST_UR)
1559 		++st->tx_bd_underrun;
1560 	if (ctrl & EMAC_TX_ST_SQE)
1561 		++st->tx_bd_sqe;
1562 }
1563 
1564 static void emac_poll_tx(void *param)
1565 {
1566 	struct emac_instance *dev = param;
1567 	u32 bad_mask;
1568 
1569 	DBG2(dev, "poll_tx, %d %d" NL, dev->tx_cnt, dev->ack_slot);
1570 
1571 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
1572 		bad_mask = EMAC_IS_BAD_TX_TAH;
1573 	else
1574 		bad_mask = EMAC_IS_BAD_TX;
1575 
1576 	netif_tx_lock_bh(dev->ndev);
1577 	if (dev->tx_cnt) {
1578 		u16 ctrl;
1579 		int slot = dev->ack_slot, n = 0;
1580 	again:
1581 		ctrl = dev->tx_desc[slot].ctrl;
1582 		if (!(ctrl & MAL_TX_CTRL_READY)) {
1583 			struct sk_buff *skb = dev->tx_skb[slot];
1584 			++n;
1585 
1586 			if (skb) {
1587 				dev_kfree_skb(skb);
1588 				dev->tx_skb[slot] = NULL;
1589 			}
1590 			slot = (slot + 1) % NUM_TX_BUFF;
1591 
1592 			if (unlikely(ctrl & bad_mask))
1593 				emac_parse_tx_error(dev, ctrl);
1594 
1595 			if (--dev->tx_cnt)
1596 				goto again;
1597 		}
1598 		if (n) {
1599 			dev->ack_slot = slot;
1600 			if (netif_queue_stopped(dev->ndev) &&
1601 			    dev->tx_cnt < EMAC_TX_WAKEUP_THRESH)
1602 				netif_wake_queue(dev->ndev);
1603 
1604 			DBG2(dev, "tx %d pkts" NL, n);
1605 		}
1606 	}
1607 	netif_tx_unlock_bh(dev->ndev);
1608 }
1609 
1610 static inline void emac_recycle_rx_skb(struct emac_instance *dev, int slot,
1611 				       int len)
1612 {
1613 	struct sk_buff *skb = dev->rx_skb[slot];
1614 
1615 	DBG2(dev, "recycle %d %d" NL, slot, len);
1616 
1617 	if (len)
1618 		dma_map_single(&dev->ofdev->dev, skb->data - 2,
1619 			       EMAC_DMA_ALIGN(len + 2), DMA_FROM_DEVICE);
1620 
1621 	dev->rx_desc[slot].data_len = 0;
1622 	wmb();
1623 	dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1624 	    (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1625 }
1626 
1627 static void emac_parse_rx_error(struct emac_instance *dev, u16 ctrl)
1628 {
1629 	struct emac_error_stats *st = &dev->estats;
1630 
1631 	DBG(dev, "BD RX error %04x" NL, ctrl);
1632 
1633 	++st->rx_bd_errors;
1634 	if (ctrl & EMAC_RX_ST_OE)
1635 		++st->rx_bd_overrun;
1636 	if (ctrl & EMAC_RX_ST_BP)
1637 		++st->rx_bd_bad_packet;
1638 	if (ctrl & EMAC_RX_ST_RP)
1639 		++st->rx_bd_runt_packet;
1640 	if (ctrl & EMAC_RX_ST_SE)
1641 		++st->rx_bd_short_event;
1642 	if (ctrl & EMAC_RX_ST_AE)
1643 		++st->rx_bd_alignment_error;
1644 	if (ctrl & EMAC_RX_ST_BFCS)
1645 		++st->rx_bd_bad_fcs;
1646 	if (ctrl & EMAC_RX_ST_PTL)
1647 		++st->rx_bd_packet_too_long;
1648 	if (ctrl & EMAC_RX_ST_ORE)
1649 		++st->rx_bd_out_of_range;
1650 	if (ctrl & EMAC_RX_ST_IRE)
1651 		++st->rx_bd_in_range;
1652 }
1653 
1654 static inline void emac_rx_csum(struct emac_instance *dev,
1655 				struct sk_buff *skb, u16 ctrl)
1656 {
1657 #ifdef CONFIG_IBM_EMAC_TAH
1658 	if (!ctrl && dev->tah_dev) {
1659 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1660 		++dev->stats.rx_packets_csum;
1661 	}
1662 #endif
1663 }
1664 
1665 static inline int emac_rx_sg_append(struct emac_instance *dev, int slot)
1666 {
1667 	if (likely(dev->rx_sg_skb != NULL)) {
1668 		int len = dev->rx_desc[slot].data_len;
1669 		int tot_len = dev->rx_sg_skb->len + len;
1670 
1671 		if (unlikely(tot_len + 2 > dev->rx_skb_size)) {
1672 			++dev->estats.rx_dropped_mtu;
1673 			dev_kfree_skb(dev->rx_sg_skb);
1674 			dev->rx_sg_skb = NULL;
1675 		} else {
1676 			cacheable_memcpy(skb_tail_pointer(dev->rx_sg_skb),
1677 					 dev->rx_skb[slot]->data, len);
1678 			skb_put(dev->rx_sg_skb, len);
1679 			emac_recycle_rx_skb(dev, slot, len);
1680 			return 0;
1681 		}
1682 	}
1683 	emac_recycle_rx_skb(dev, slot, 0);
1684 	return -1;
1685 }
1686 
1687 /* NAPI poll context */
1688 static int emac_poll_rx(void *param, int budget)
1689 {
1690 	struct emac_instance *dev = param;
1691 	int slot = dev->rx_slot, received = 0;
1692 
1693 	DBG2(dev, "poll_rx(%d)" NL, budget);
1694 
1695  again:
1696 	while (budget > 0) {
1697 		int len;
1698 		struct sk_buff *skb;
1699 		u16 ctrl = dev->rx_desc[slot].ctrl;
1700 
1701 		if (ctrl & MAL_RX_CTRL_EMPTY)
1702 			break;
1703 
1704 		skb = dev->rx_skb[slot];
1705 		mb();
1706 		len = dev->rx_desc[slot].data_len;
1707 
1708 		if (unlikely(!MAL_IS_SINGLE_RX(ctrl)))
1709 			goto sg;
1710 
1711 		ctrl &= EMAC_BAD_RX_MASK;
1712 		if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1713 			emac_parse_rx_error(dev, ctrl);
1714 			++dev->estats.rx_dropped_error;
1715 			emac_recycle_rx_skb(dev, slot, 0);
1716 			len = 0;
1717 			goto next;
1718 		}
1719 
1720 		if (len < ETH_HLEN) {
1721 			++dev->estats.rx_dropped_stack;
1722 			emac_recycle_rx_skb(dev, slot, len);
1723 			goto next;
1724 		}
1725 
1726 		if (len && len < EMAC_RX_COPY_THRESH) {
1727 			struct sk_buff *copy_skb =
1728 			    alloc_skb(len + EMAC_RX_SKB_HEADROOM + 2, GFP_ATOMIC);
1729 			if (unlikely(!copy_skb))
1730 				goto oom;
1731 
1732 			skb_reserve(copy_skb, EMAC_RX_SKB_HEADROOM + 2);
1733 			cacheable_memcpy(copy_skb->data - 2, skb->data - 2,
1734 					 len + 2);
1735 			emac_recycle_rx_skb(dev, slot, len);
1736 			skb = copy_skb;
1737 		} else if (unlikely(emac_alloc_rx_skb(dev, slot, GFP_ATOMIC)))
1738 			goto oom;
1739 
1740 		skb_put(skb, len);
1741 	push_packet:
1742 		skb->protocol = eth_type_trans(skb, dev->ndev);
1743 		emac_rx_csum(dev, skb, ctrl);
1744 
1745 		if (unlikely(netif_receive_skb(skb) == NET_RX_DROP))
1746 			++dev->estats.rx_dropped_stack;
1747 	next:
1748 		++dev->stats.rx_packets;
1749 	skip:
1750 		dev->stats.rx_bytes += len;
1751 		slot = (slot + 1) % NUM_RX_BUFF;
1752 		--budget;
1753 		++received;
1754 		continue;
1755 	sg:
1756 		if (ctrl & MAL_RX_CTRL_FIRST) {
1757 			BUG_ON(dev->rx_sg_skb);
1758 			if (unlikely(emac_alloc_rx_skb(dev, slot, GFP_ATOMIC))) {
1759 				DBG(dev, "rx OOM %d" NL, slot);
1760 				++dev->estats.rx_dropped_oom;
1761 				emac_recycle_rx_skb(dev, slot, 0);
1762 			} else {
1763 				dev->rx_sg_skb = skb;
1764 				skb_put(skb, len);
1765 			}
1766 		} else if (!emac_rx_sg_append(dev, slot) &&
1767 			   (ctrl & MAL_RX_CTRL_LAST)) {
1768 
1769 			skb = dev->rx_sg_skb;
1770 			dev->rx_sg_skb = NULL;
1771 
1772 			ctrl &= EMAC_BAD_RX_MASK;
1773 			if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1774 				emac_parse_rx_error(dev, ctrl);
1775 				++dev->estats.rx_dropped_error;
1776 				dev_kfree_skb(skb);
1777 				len = 0;
1778 			} else
1779 				goto push_packet;
1780 		}
1781 		goto skip;
1782 	oom:
1783 		DBG(dev, "rx OOM %d" NL, slot);
1784 		/* Drop the packet and recycle skb */
1785 		++dev->estats.rx_dropped_oom;
1786 		emac_recycle_rx_skb(dev, slot, 0);
1787 		goto next;
1788 	}
1789 
1790 	if (received) {
1791 		DBG2(dev, "rx %d BDs" NL, received);
1792 		dev->rx_slot = slot;
1793 	}
1794 
1795 	if (unlikely(budget && test_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags))) {
1796 		mb();
1797 		if (!(dev->rx_desc[slot].ctrl & MAL_RX_CTRL_EMPTY)) {
1798 			DBG2(dev, "rx restart" NL);
1799 			received = 0;
1800 			goto again;
1801 		}
1802 
1803 		if (dev->rx_sg_skb) {
1804 			DBG2(dev, "dropping partial rx packet" NL);
1805 			++dev->estats.rx_dropped_error;
1806 			dev_kfree_skb(dev->rx_sg_skb);
1807 			dev->rx_sg_skb = NULL;
1808 		}
1809 
1810 		clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1811 		mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1812 		emac_rx_enable(dev);
1813 		dev->rx_slot = 0;
1814 	}
1815 	return received;
1816 }
1817 
1818 /* NAPI poll context */
1819 static int emac_peek_rx(void *param)
1820 {
1821 	struct emac_instance *dev = param;
1822 
1823 	return !(dev->rx_desc[dev->rx_slot].ctrl & MAL_RX_CTRL_EMPTY);
1824 }
1825 
1826 /* NAPI poll context */
1827 static int emac_peek_rx_sg(void *param)
1828 {
1829 	struct emac_instance *dev = param;
1830 
1831 	int slot = dev->rx_slot;
1832 	while (1) {
1833 		u16 ctrl = dev->rx_desc[slot].ctrl;
1834 		if (ctrl & MAL_RX_CTRL_EMPTY)
1835 			return 0;
1836 		else if (ctrl & MAL_RX_CTRL_LAST)
1837 			return 1;
1838 
1839 		slot = (slot + 1) % NUM_RX_BUFF;
1840 
1841 		/* I'm just being paranoid here :) */
1842 		if (unlikely(slot == dev->rx_slot))
1843 			return 0;
1844 	}
1845 }
1846 
1847 /* Hard IRQ */
1848 static void emac_rxde(void *param)
1849 {
1850 	struct emac_instance *dev = param;
1851 
1852 	++dev->estats.rx_stopped;
1853 	emac_rx_disable_async(dev);
1854 }
1855 
1856 /* Hard IRQ */
1857 static irqreturn_t emac_irq(int irq, void *dev_instance)
1858 {
1859 	struct emac_instance *dev = dev_instance;
1860 	struct emac_regs __iomem *p = dev->emacp;
1861 	struct emac_error_stats *st = &dev->estats;
1862 	u32 isr;
1863 
1864 	spin_lock(&dev->lock);
1865 
1866 	isr = in_be32(&p->isr);
1867 	out_be32(&p->isr, isr);
1868 
1869 	DBG(dev, "isr = %08x" NL, isr);
1870 
1871 	if (isr & EMAC4_ISR_TXPE)
1872 		++st->tx_parity;
1873 	if (isr & EMAC4_ISR_RXPE)
1874 		++st->rx_parity;
1875 	if (isr & EMAC4_ISR_TXUE)
1876 		++st->tx_underrun;
1877 	if (isr & EMAC4_ISR_RXOE)
1878 		++st->rx_fifo_overrun;
1879 	if (isr & EMAC_ISR_OVR)
1880 		++st->rx_overrun;
1881 	if (isr & EMAC_ISR_BP)
1882 		++st->rx_bad_packet;
1883 	if (isr & EMAC_ISR_RP)
1884 		++st->rx_runt_packet;
1885 	if (isr & EMAC_ISR_SE)
1886 		++st->rx_short_event;
1887 	if (isr & EMAC_ISR_ALE)
1888 		++st->rx_alignment_error;
1889 	if (isr & EMAC_ISR_BFCS)
1890 		++st->rx_bad_fcs;
1891 	if (isr & EMAC_ISR_PTLE)
1892 		++st->rx_packet_too_long;
1893 	if (isr & EMAC_ISR_ORE)
1894 		++st->rx_out_of_range;
1895 	if (isr & EMAC_ISR_IRE)
1896 		++st->rx_in_range;
1897 	if (isr & EMAC_ISR_SQE)
1898 		++st->tx_sqe;
1899 	if (isr & EMAC_ISR_TE)
1900 		++st->tx_errors;
1901 
1902 	spin_unlock(&dev->lock);
1903 
1904 	return IRQ_HANDLED;
1905 }
1906 
1907 static struct net_device_stats *emac_stats(struct net_device *ndev)
1908 {
1909 	struct emac_instance *dev = netdev_priv(ndev);
1910 	struct emac_stats *st = &dev->stats;
1911 	struct emac_error_stats *est = &dev->estats;
1912 	struct net_device_stats *nst = &dev->nstats;
1913 	unsigned long flags;
1914 
1915 	DBG2(dev, "stats" NL);
1916 
1917 	/* Compute "legacy" statistics */
1918 	spin_lock_irqsave(&dev->lock, flags);
1919 	nst->rx_packets = (unsigned long)st->rx_packets;
1920 	nst->rx_bytes = (unsigned long)st->rx_bytes;
1921 	nst->tx_packets = (unsigned long)st->tx_packets;
1922 	nst->tx_bytes = (unsigned long)st->tx_bytes;
1923 	nst->rx_dropped = (unsigned long)(est->rx_dropped_oom +
1924 					  est->rx_dropped_error +
1925 					  est->rx_dropped_resize +
1926 					  est->rx_dropped_mtu);
1927 	nst->tx_dropped = (unsigned long)est->tx_dropped;
1928 
1929 	nst->rx_errors = (unsigned long)est->rx_bd_errors;
1930 	nst->rx_fifo_errors = (unsigned long)(est->rx_bd_overrun +
1931 					      est->rx_fifo_overrun +
1932 					      est->rx_overrun);
1933 	nst->rx_frame_errors = (unsigned long)(est->rx_bd_alignment_error +
1934 					       est->rx_alignment_error);
1935 	nst->rx_crc_errors = (unsigned long)(est->rx_bd_bad_fcs +
1936 					     est->rx_bad_fcs);
1937 	nst->rx_length_errors = (unsigned long)(est->rx_bd_runt_packet +
1938 						est->rx_bd_short_event +
1939 						est->rx_bd_packet_too_long +
1940 						est->rx_bd_out_of_range +
1941 						est->rx_bd_in_range +
1942 						est->rx_runt_packet +
1943 						est->rx_short_event +
1944 						est->rx_packet_too_long +
1945 						est->rx_out_of_range +
1946 						est->rx_in_range);
1947 
1948 	nst->tx_errors = (unsigned long)(est->tx_bd_errors + est->tx_errors);
1949 	nst->tx_fifo_errors = (unsigned long)(est->tx_bd_underrun +
1950 					      est->tx_underrun);
1951 	nst->tx_carrier_errors = (unsigned long)est->tx_bd_carrier_loss;
1952 	nst->collisions = (unsigned long)(est->tx_bd_excessive_deferral +
1953 					  est->tx_bd_excessive_collisions +
1954 					  est->tx_bd_late_collision +
1955 					  est->tx_bd_multple_collisions);
1956 	spin_unlock_irqrestore(&dev->lock, flags);
1957 	return nst;
1958 }
1959 
1960 static struct mal_commac_ops emac_commac_ops = {
1961 	.poll_tx = &emac_poll_tx,
1962 	.poll_rx = &emac_poll_rx,
1963 	.peek_rx = &emac_peek_rx,
1964 	.rxde = &emac_rxde,
1965 };
1966 
1967 static struct mal_commac_ops emac_commac_sg_ops = {
1968 	.poll_tx = &emac_poll_tx,
1969 	.poll_rx = &emac_poll_rx,
1970 	.peek_rx = &emac_peek_rx_sg,
1971 	.rxde = &emac_rxde,
1972 };
1973 
1974 /* Ethtool support */
1975 static int emac_ethtool_get_settings(struct net_device *ndev,
1976 				     struct ethtool_cmd *cmd)
1977 {
1978 	struct emac_instance *dev = netdev_priv(ndev);
1979 
1980 	cmd->supported = dev->phy.features;
1981 	cmd->port = PORT_MII;
1982 	cmd->phy_address = dev->phy.address;
1983 	cmd->transceiver =
1984 	    dev->phy.address >= 0 ? XCVR_EXTERNAL : XCVR_INTERNAL;
1985 
1986 	mutex_lock(&dev->link_lock);
1987 	cmd->advertising = dev->phy.advertising;
1988 	cmd->autoneg = dev->phy.autoneg;
1989 	cmd->speed = dev->phy.speed;
1990 	cmd->duplex = dev->phy.duplex;
1991 	mutex_unlock(&dev->link_lock);
1992 
1993 	return 0;
1994 }
1995 
1996 static int emac_ethtool_set_settings(struct net_device *ndev,
1997 				     struct ethtool_cmd *cmd)
1998 {
1999 	struct emac_instance *dev = netdev_priv(ndev);
2000 	u32 f = dev->phy.features;
2001 
2002 	DBG(dev, "set_settings(%d, %d, %d, 0x%08x)" NL,
2003 	    cmd->autoneg, cmd->speed, cmd->duplex, cmd->advertising);
2004 
2005 	/* Basic sanity checks */
2006 	if (dev->phy.address < 0)
2007 		return -EOPNOTSUPP;
2008 	if (cmd->autoneg != AUTONEG_ENABLE && cmd->autoneg != AUTONEG_DISABLE)
2009 		return -EINVAL;
2010 	if (cmd->autoneg == AUTONEG_ENABLE && cmd->advertising == 0)
2011 		return -EINVAL;
2012 	if (cmd->duplex != DUPLEX_HALF && cmd->duplex != DUPLEX_FULL)
2013 		return -EINVAL;
2014 
2015 	if (cmd->autoneg == AUTONEG_DISABLE) {
2016 		switch (cmd->speed) {
2017 		case SPEED_10:
2018 			if (cmd->duplex == DUPLEX_HALF &&
2019 			    !(f & SUPPORTED_10baseT_Half))
2020 				return -EINVAL;
2021 			if (cmd->duplex == DUPLEX_FULL &&
2022 			    !(f & SUPPORTED_10baseT_Full))
2023 				return -EINVAL;
2024 			break;
2025 		case SPEED_100:
2026 			if (cmd->duplex == DUPLEX_HALF &&
2027 			    !(f & SUPPORTED_100baseT_Half))
2028 				return -EINVAL;
2029 			if (cmd->duplex == DUPLEX_FULL &&
2030 			    !(f & SUPPORTED_100baseT_Full))
2031 				return -EINVAL;
2032 			break;
2033 		case SPEED_1000:
2034 			if (cmd->duplex == DUPLEX_HALF &&
2035 			    !(f & SUPPORTED_1000baseT_Half))
2036 				return -EINVAL;
2037 			if (cmd->duplex == DUPLEX_FULL &&
2038 			    !(f & SUPPORTED_1000baseT_Full))
2039 				return -EINVAL;
2040 			break;
2041 		default:
2042 			return -EINVAL;
2043 		}
2044 
2045 		mutex_lock(&dev->link_lock);
2046 		dev->phy.def->ops->setup_forced(&dev->phy, cmd->speed,
2047 						cmd->duplex);
2048 		mutex_unlock(&dev->link_lock);
2049 
2050 	} else {
2051 		if (!(f & SUPPORTED_Autoneg))
2052 			return -EINVAL;
2053 
2054 		mutex_lock(&dev->link_lock);
2055 		dev->phy.def->ops->setup_aneg(&dev->phy,
2056 					      (cmd->advertising & f) |
2057 					      (dev->phy.advertising &
2058 					       (ADVERTISED_Pause |
2059 						ADVERTISED_Asym_Pause)));
2060 		mutex_unlock(&dev->link_lock);
2061 	}
2062 	emac_force_link_update(dev);
2063 
2064 	return 0;
2065 }
2066 
2067 static void emac_ethtool_get_ringparam(struct net_device *ndev,
2068 				       struct ethtool_ringparam *rp)
2069 {
2070 	rp->rx_max_pending = rp->rx_pending = NUM_RX_BUFF;
2071 	rp->tx_max_pending = rp->tx_pending = NUM_TX_BUFF;
2072 }
2073 
2074 static void emac_ethtool_get_pauseparam(struct net_device *ndev,
2075 					struct ethtool_pauseparam *pp)
2076 {
2077 	struct emac_instance *dev = netdev_priv(ndev);
2078 
2079 	mutex_lock(&dev->link_lock);
2080 	if ((dev->phy.features & SUPPORTED_Autoneg) &&
2081 	    (dev->phy.advertising & (ADVERTISED_Pause | ADVERTISED_Asym_Pause)))
2082 		pp->autoneg = 1;
2083 
2084 	if (dev->phy.duplex == DUPLEX_FULL) {
2085 		if (dev->phy.pause)
2086 			pp->rx_pause = pp->tx_pause = 1;
2087 		else if (dev->phy.asym_pause)
2088 			pp->tx_pause = 1;
2089 	}
2090 	mutex_unlock(&dev->link_lock);
2091 }
2092 
2093 static int emac_get_regs_len(struct emac_instance *dev)
2094 {
2095 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
2096 		return sizeof(struct emac_ethtool_regs_subhdr) +
2097 			EMAC4_ETHTOOL_REGS_SIZE(dev);
2098 	else
2099 		return sizeof(struct emac_ethtool_regs_subhdr) +
2100 			EMAC_ETHTOOL_REGS_SIZE(dev);
2101 }
2102 
2103 static int emac_ethtool_get_regs_len(struct net_device *ndev)
2104 {
2105 	struct emac_instance *dev = netdev_priv(ndev);
2106 	int size;
2107 
2108 	size = sizeof(struct emac_ethtool_regs_hdr) +
2109 		emac_get_regs_len(dev) + mal_get_regs_len(dev->mal);
2110 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
2111 		size += zmii_get_regs_len(dev->zmii_dev);
2112 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
2113 		size += rgmii_get_regs_len(dev->rgmii_dev);
2114 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
2115 		size += tah_get_regs_len(dev->tah_dev);
2116 
2117 	return size;
2118 }
2119 
2120 static void *emac_dump_regs(struct emac_instance *dev, void *buf)
2121 {
2122 	struct emac_ethtool_regs_subhdr *hdr = buf;
2123 
2124 	hdr->index = dev->cell_index;
2125 	if (emac_has_feature(dev, EMAC_FTR_EMAC4)) {
2126 		hdr->version = EMAC4_ETHTOOL_REGS_VER;
2127 		memcpy_fromio(hdr + 1, dev->emacp, EMAC4_ETHTOOL_REGS_SIZE(dev));
2128 		return (void *)(hdr + 1) + EMAC4_ETHTOOL_REGS_SIZE(dev);
2129 	} else {
2130 		hdr->version = EMAC_ETHTOOL_REGS_VER;
2131 		memcpy_fromio(hdr + 1, dev->emacp, EMAC_ETHTOOL_REGS_SIZE(dev));
2132 		return (void *)(hdr + 1) + EMAC_ETHTOOL_REGS_SIZE(dev);
2133 	}
2134 }
2135 
2136 static void emac_ethtool_get_regs(struct net_device *ndev,
2137 				  struct ethtool_regs *regs, void *buf)
2138 {
2139 	struct emac_instance *dev = netdev_priv(ndev);
2140 	struct emac_ethtool_regs_hdr *hdr = buf;
2141 
2142 	hdr->components = 0;
2143 	buf = hdr + 1;
2144 
2145 	buf = mal_dump_regs(dev->mal, buf);
2146 	buf = emac_dump_regs(dev, buf);
2147 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII)) {
2148 		hdr->components |= EMAC_ETHTOOL_REGS_ZMII;
2149 		buf = zmii_dump_regs(dev->zmii_dev, buf);
2150 	}
2151 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII)) {
2152 		hdr->components |= EMAC_ETHTOOL_REGS_RGMII;
2153 		buf = rgmii_dump_regs(dev->rgmii_dev, buf);
2154 	}
2155 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH)) {
2156 		hdr->components |= EMAC_ETHTOOL_REGS_TAH;
2157 		buf = tah_dump_regs(dev->tah_dev, buf);
2158 	}
2159 }
2160 
2161 static int emac_ethtool_nway_reset(struct net_device *ndev)
2162 {
2163 	struct emac_instance *dev = netdev_priv(ndev);
2164 	int res = 0;
2165 
2166 	DBG(dev, "nway_reset" NL);
2167 
2168 	if (dev->phy.address < 0)
2169 		return -EOPNOTSUPP;
2170 
2171 	mutex_lock(&dev->link_lock);
2172 	if (!dev->phy.autoneg) {
2173 		res = -EINVAL;
2174 		goto out;
2175 	}
2176 
2177 	dev->phy.def->ops->setup_aneg(&dev->phy, dev->phy.advertising);
2178  out:
2179 	mutex_unlock(&dev->link_lock);
2180 	emac_force_link_update(dev);
2181 	return res;
2182 }
2183 
2184 static int emac_ethtool_get_sset_count(struct net_device *ndev, int stringset)
2185 {
2186 	if (stringset == ETH_SS_STATS)
2187 		return EMAC_ETHTOOL_STATS_COUNT;
2188 	else
2189 		return -EINVAL;
2190 }
2191 
2192 static void emac_ethtool_get_strings(struct net_device *ndev, u32 stringset,
2193 				     u8 * buf)
2194 {
2195 	if (stringset == ETH_SS_STATS)
2196 		memcpy(buf, &emac_stats_keys, sizeof(emac_stats_keys));
2197 }
2198 
2199 static void emac_ethtool_get_ethtool_stats(struct net_device *ndev,
2200 					   struct ethtool_stats *estats,
2201 					   u64 * tmp_stats)
2202 {
2203 	struct emac_instance *dev = netdev_priv(ndev);
2204 
2205 	memcpy(tmp_stats, &dev->stats, sizeof(dev->stats));
2206 	tmp_stats += sizeof(dev->stats) / sizeof(u64);
2207 	memcpy(tmp_stats, &dev->estats, sizeof(dev->estats));
2208 }
2209 
2210 static void emac_ethtool_get_drvinfo(struct net_device *ndev,
2211 				     struct ethtool_drvinfo *info)
2212 {
2213 	struct emac_instance *dev = netdev_priv(ndev);
2214 
2215 	strlcpy(info->driver, "ibm_emac", sizeof(info->driver));
2216 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2217 	snprintf(info->bus_info, sizeof(info->bus_info), "PPC 4xx EMAC-%d %s",
2218 		 dev->cell_index, dev->ofdev->dev.of_node->full_name);
2219 	info->regdump_len = emac_ethtool_get_regs_len(ndev);
2220 }
2221 
2222 static const struct ethtool_ops emac_ethtool_ops = {
2223 	.get_settings = emac_ethtool_get_settings,
2224 	.set_settings = emac_ethtool_set_settings,
2225 	.get_drvinfo = emac_ethtool_get_drvinfo,
2226 
2227 	.get_regs_len = emac_ethtool_get_regs_len,
2228 	.get_regs = emac_ethtool_get_regs,
2229 
2230 	.nway_reset = emac_ethtool_nway_reset,
2231 
2232 	.get_ringparam = emac_ethtool_get_ringparam,
2233 	.get_pauseparam = emac_ethtool_get_pauseparam,
2234 
2235 	.get_strings = emac_ethtool_get_strings,
2236 	.get_sset_count = emac_ethtool_get_sset_count,
2237 	.get_ethtool_stats = emac_ethtool_get_ethtool_stats,
2238 
2239 	.get_link = ethtool_op_get_link,
2240 };
2241 
2242 static int emac_ioctl(struct net_device *ndev, struct ifreq *rq, int cmd)
2243 {
2244 	struct emac_instance *dev = netdev_priv(ndev);
2245 	struct mii_ioctl_data *data = if_mii(rq);
2246 
2247 	DBG(dev, "ioctl %08x" NL, cmd);
2248 
2249 	if (dev->phy.address < 0)
2250 		return -EOPNOTSUPP;
2251 
2252 	switch (cmd) {
2253 	case SIOCGMIIPHY:
2254 		data->phy_id = dev->phy.address;
2255 		/* Fall through */
2256 	case SIOCGMIIREG:
2257 		data->val_out = emac_mdio_read(ndev, dev->phy.address,
2258 					       data->reg_num);
2259 		return 0;
2260 
2261 	case SIOCSMIIREG:
2262 		emac_mdio_write(ndev, dev->phy.address, data->reg_num,
2263 				data->val_in);
2264 		return 0;
2265 	default:
2266 		return -EOPNOTSUPP;
2267 	}
2268 }
2269 
2270 struct emac_depentry {
2271 	u32			phandle;
2272 	struct device_node	*node;
2273 	struct platform_device	*ofdev;
2274 	void			*drvdata;
2275 };
2276 
2277 #define	EMAC_DEP_MAL_IDX	0
2278 #define	EMAC_DEP_ZMII_IDX	1
2279 #define	EMAC_DEP_RGMII_IDX	2
2280 #define	EMAC_DEP_TAH_IDX	3
2281 #define	EMAC_DEP_MDIO_IDX	4
2282 #define	EMAC_DEP_PREV_IDX	5
2283 #define	EMAC_DEP_COUNT		6
2284 
2285 static int emac_check_deps(struct emac_instance *dev,
2286 			   struct emac_depentry *deps)
2287 {
2288 	int i, there = 0;
2289 	struct device_node *np;
2290 
2291 	for (i = 0; i < EMAC_DEP_COUNT; i++) {
2292 		/* no dependency on that item, allright */
2293 		if (deps[i].phandle == 0) {
2294 			there++;
2295 			continue;
2296 		}
2297 		/* special case for blist as the dependency might go away */
2298 		if (i == EMAC_DEP_PREV_IDX) {
2299 			np = *(dev->blist - 1);
2300 			if (np == NULL) {
2301 				deps[i].phandle = 0;
2302 				there++;
2303 				continue;
2304 			}
2305 			if (deps[i].node == NULL)
2306 				deps[i].node = of_node_get(np);
2307 		}
2308 		if (deps[i].node == NULL)
2309 			deps[i].node = of_find_node_by_phandle(deps[i].phandle);
2310 		if (deps[i].node == NULL)
2311 			continue;
2312 		if (deps[i].ofdev == NULL)
2313 			deps[i].ofdev = of_find_device_by_node(deps[i].node);
2314 		if (deps[i].ofdev == NULL)
2315 			continue;
2316 		if (deps[i].drvdata == NULL)
2317 			deps[i].drvdata = platform_get_drvdata(deps[i].ofdev);
2318 		if (deps[i].drvdata != NULL)
2319 			there++;
2320 	}
2321 	return there == EMAC_DEP_COUNT;
2322 }
2323 
2324 static void emac_put_deps(struct emac_instance *dev)
2325 {
2326 	of_dev_put(dev->mal_dev);
2327 	of_dev_put(dev->zmii_dev);
2328 	of_dev_put(dev->rgmii_dev);
2329 	of_dev_put(dev->mdio_dev);
2330 	of_dev_put(dev->tah_dev);
2331 }
2332 
2333 static int emac_of_bus_notify(struct notifier_block *nb, unsigned long action,
2334 			      void *data)
2335 {
2336 	/* We are only intereted in device addition */
2337 	if (action == BUS_NOTIFY_BOUND_DRIVER)
2338 		wake_up_all(&emac_probe_wait);
2339 	return 0;
2340 }
2341 
2342 static struct notifier_block emac_of_bus_notifier = {
2343 	.notifier_call = emac_of_bus_notify
2344 };
2345 
2346 static int emac_wait_deps(struct emac_instance *dev)
2347 {
2348 	struct emac_depentry deps[EMAC_DEP_COUNT];
2349 	int i, err;
2350 
2351 	memset(&deps, 0, sizeof(deps));
2352 
2353 	deps[EMAC_DEP_MAL_IDX].phandle = dev->mal_ph;
2354 	deps[EMAC_DEP_ZMII_IDX].phandle = dev->zmii_ph;
2355 	deps[EMAC_DEP_RGMII_IDX].phandle = dev->rgmii_ph;
2356 	if (dev->tah_ph)
2357 		deps[EMAC_DEP_TAH_IDX].phandle = dev->tah_ph;
2358 	if (dev->mdio_ph)
2359 		deps[EMAC_DEP_MDIO_IDX].phandle = dev->mdio_ph;
2360 	if (dev->blist && dev->blist > emac_boot_list)
2361 		deps[EMAC_DEP_PREV_IDX].phandle = 0xffffffffu;
2362 	bus_register_notifier(&platform_bus_type, &emac_of_bus_notifier);
2363 	wait_event_timeout(emac_probe_wait,
2364 			   emac_check_deps(dev, deps),
2365 			   EMAC_PROBE_DEP_TIMEOUT);
2366 	bus_unregister_notifier(&platform_bus_type, &emac_of_bus_notifier);
2367 	err = emac_check_deps(dev, deps) ? 0 : -ENODEV;
2368 	for (i = 0; i < EMAC_DEP_COUNT; i++) {
2369 		of_node_put(deps[i].node);
2370 		if (err && deps[i].ofdev)
2371 			of_dev_put(deps[i].ofdev);
2372 	}
2373 	if (err == 0) {
2374 		dev->mal_dev = deps[EMAC_DEP_MAL_IDX].ofdev;
2375 		dev->zmii_dev = deps[EMAC_DEP_ZMII_IDX].ofdev;
2376 		dev->rgmii_dev = deps[EMAC_DEP_RGMII_IDX].ofdev;
2377 		dev->tah_dev = deps[EMAC_DEP_TAH_IDX].ofdev;
2378 		dev->mdio_dev = deps[EMAC_DEP_MDIO_IDX].ofdev;
2379 	}
2380 	of_dev_put(deps[EMAC_DEP_PREV_IDX].ofdev);
2381 	return err;
2382 }
2383 
2384 static int emac_read_uint_prop(struct device_node *np, const char *name,
2385 			       u32 *val, int fatal)
2386 {
2387 	int len;
2388 	const u32 *prop = of_get_property(np, name, &len);
2389 	if (prop == NULL || len < sizeof(u32)) {
2390 		if (fatal)
2391 			printk(KERN_ERR "%s: missing %s property\n",
2392 			       np->full_name, name);
2393 		return -ENODEV;
2394 	}
2395 	*val = *prop;
2396 	return 0;
2397 }
2398 
2399 static int emac_init_phy(struct emac_instance *dev)
2400 {
2401 	struct device_node *np = dev->ofdev->dev.of_node;
2402 	struct net_device *ndev = dev->ndev;
2403 	u32 phy_map, adv;
2404 	int i;
2405 
2406 	dev->phy.dev = ndev;
2407 	dev->phy.mode = dev->phy_mode;
2408 
2409 	/* PHY-less configuration.
2410 	 * XXX I probably should move these settings to the dev tree
2411 	 */
2412 	if (dev->phy_address == 0xffffffff && dev->phy_map == 0xffffffff) {
2413 		emac_reset(dev);
2414 
2415 		/* PHY-less configuration.
2416 		 * XXX I probably should move these settings to the dev tree
2417 		 */
2418 		dev->phy.address = -1;
2419 		dev->phy.features = SUPPORTED_MII;
2420 		if (emac_phy_supports_gige(dev->phy_mode))
2421 			dev->phy.features |= SUPPORTED_1000baseT_Full;
2422 		else
2423 			dev->phy.features |= SUPPORTED_100baseT_Full;
2424 		dev->phy.pause = 1;
2425 
2426 		return 0;
2427 	}
2428 
2429 	mutex_lock(&emac_phy_map_lock);
2430 	phy_map = dev->phy_map | busy_phy_map;
2431 
2432 	DBG(dev, "PHY maps %08x %08x" NL, dev->phy_map, busy_phy_map);
2433 
2434 	dev->phy.mdio_read = emac_mdio_read;
2435 	dev->phy.mdio_write = emac_mdio_write;
2436 
2437 	/* Enable internal clock source */
2438 #ifdef CONFIG_PPC_DCR_NATIVE
2439 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2440 		dcri_clrset(SDR0, SDR0_MFR, 0, SDR0_MFR_ECS);
2441 #endif
2442 	/* PHY clock workaround */
2443 	emac_rx_clk_tx(dev);
2444 
2445 	/* Enable internal clock source on 440GX*/
2446 #ifdef CONFIG_PPC_DCR_NATIVE
2447 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2448 		dcri_clrset(SDR0, SDR0_MFR, 0, SDR0_MFR_ECS);
2449 #endif
2450 	/* Configure EMAC with defaults so we can at least use MDIO
2451 	 * This is needed mostly for 440GX
2452 	 */
2453 	if (emac_phy_gpcs(dev->phy.mode)) {
2454 		/* XXX
2455 		 * Make GPCS PHY address equal to EMAC index.
2456 		 * We probably should take into account busy_phy_map
2457 		 * and/or phy_map here.
2458 		 *
2459 		 * Note that the busy_phy_map is currently global
2460 		 * while it should probably be per-ASIC...
2461 		 */
2462 		dev->phy.gpcs_address = dev->gpcs_address;
2463 		if (dev->phy.gpcs_address == 0xffffffff)
2464 			dev->phy.address = dev->cell_index;
2465 	}
2466 
2467 	emac_configure(dev);
2468 
2469 	if (dev->phy_address != 0xffffffff)
2470 		phy_map = ~(1 << dev->phy_address);
2471 
2472 	for (i = 0; i < 0x20; phy_map >>= 1, ++i)
2473 		if (!(phy_map & 1)) {
2474 			int r;
2475 			busy_phy_map |= 1 << i;
2476 
2477 			/* Quick check if there is a PHY at the address */
2478 			r = emac_mdio_read(dev->ndev, i, MII_BMCR);
2479 			if (r == 0xffff || r < 0)
2480 				continue;
2481 			if (!emac_mii_phy_probe(&dev->phy, i))
2482 				break;
2483 		}
2484 
2485 	/* Enable external clock source */
2486 #ifdef CONFIG_PPC_DCR_NATIVE
2487 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2488 		dcri_clrset(SDR0, SDR0_MFR, SDR0_MFR_ECS, 0);
2489 #endif
2490 	mutex_unlock(&emac_phy_map_lock);
2491 	if (i == 0x20) {
2492 		printk(KERN_WARNING "%s: can't find PHY!\n", np->full_name);
2493 		return -ENXIO;
2494 	}
2495 
2496 	/* Init PHY */
2497 	if (dev->phy.def->ops->init)
2498 		dev->phy.def->ops->init(&dev->phy);
2499 
2500 	/* Disable any PHY features not supported by the platform */
2501 	dev->phy.def->features &= ~dev->phy_feat_exc;
2502 	dev->phy.features &= ~dev->phy_feat_exc;
2503 
2504 	/* Setup initial link parameters */
2505 	if (dev->phy.features & SUPPORTED_Autoneg) {
2506 		adv = dev->phy.features;
2507 		if (!emac_has_feature(dev, EMAC_FTR_NO_FLOW_CONTROL_40x))
2508 			adv |= ADVERTISED_Pause | ADVERTISED_Asym_Pause;
2509 		/* Restart autonegotiation */
2510 		dev->phy.def->ops->setup_aneg(&dev->phy, adv);
2511 	} else {
2512 		u32 f = dev->phy.def->features;
2513 		int speed = SPEED_10, fd = DUPLEX_HALF;
2514 
2515 		/* Select highest supported speed/duplex */
2516 		if (f & SUPPORTED_1000baseT_Full) {
2517 			speed = SPEED_1000;
2518 			fd = DUPLEX_FULL;
2519 		} else if (f & SUPPORTED_1000baseT_Half)
2520 			speed = SPEED_1000;
2521 		else if (f & SUPPORTED_100baseT_Full) {
2522 			speed = SPEED_100;
2523 			fd = DUPLEX_FULL;
2524 		} else if (f & SUPPORTED_100baseT_Half)
2525 			speed = SPEED_100;
2526 		else if (f & SUPPORTED_10baseT_Full)
2527 			fd = DUPLEX_FULL;
2528 
2529 		/* Force link parameters */
2530 		dev->phy.def->ops->setup_forced(&dev->phy, speed, fd);
2531 	}
2532 	return 0;
2533 }
2534 
2535 static int emac_init_config(struct emac_instance *dev)
2536 {
2537 	struct device_node *np = dev->ofdev->dev.of_node;
2538 	const void *p;
2539 
2540 	/* Read config from device-tree */
2541 	if (emac_read_uint_prop(np, "mal-device", &dev->mal_ph, 1))
2542 		return -ENXIO;
2543 	if (emac_read_uint_prop(np, "mal-tx-channel", &dev->mal_tx_chan, 1))
2544 		return -ENXIO;
2545 	if (emac_read_uint_prop(np, "mal-rx-channel", &dev->mal_rx_chan, 1))
2546 		return -ENXIO;
2547 	if (emac_read_uint_prop(np, "cell-index", &dev->cell_index, 1))
2548 		return -ENXIO;
2549 	if (emac_read_uint_prop(np, "max-frame-size", &dev->max_mtu, 0))
2550 		dev->max_mtu = 1500;
2551 	if (emac_read_uint_prop(np, "rx-fifo-size", &dev->rx_fifo_size, 0))
2552 		dev->rx_fifo_size = 2048;
2553 	if (emac_read_uint_prop(np, "tx-fifo-size", &dev->tx_fifo_size, 0))
2554 		dev->tx_fifo_size = 2048;
2555 	if (emac_read_uint_prop(np, "rx-fifo-size-gige", &dev->rx_fifo_size_gige, 0))
2556 		dev->rx_fifo_size_gige = dev->rx_fifo_size;
2557 	if (emac_read_uint_prop(np, "tx-fifo-size-gige", &dev->tx_fifo_size_gige, 0))
2558 		dev->tx_fifo_size_gige = dev->tx_fifo_size;
2559 	if (emac_read_uint_prop(np, "phy-address", &dev->phy_address, 0))
2560 		dev->phy_address = 0xffffffff;
2561 	if (emac_read_uint_prop(np, "phy-map", &dev->phy_map, 0))
2562 		dev->phy_map = 0xffffffff;
2563 	if (emac_read_uint_prop(np, "gpcs-address", &dev->gpcs_address, 0))
2564 		dev->gpcs_address = 0xffffffff;
2565 	if (emac_read_uint_prop(np->parent, "clock-frequency", &dev->opb_bus_freq, 1))
2566 		return -ENXIO;
2567 	if (emac_read_uint_prop(np, "tah-device", &dev->tah_ph, 0))
2568 		dev->tah_ph = 0;
2569 	if (emac_read_uint_prop(np, "tah-channel", &dev->tah_port, 0))
2570 		dev->tah_port = 0;
2571 	if (emac_read_uint_prop(np, "mdio-device", &dev->mdio_ph, 0))
2572 		dev->mdio_ph = 0;
2573 	if (emac_read_uint_prop(np, "zmii-device", &dev->zmii_ph, 0))
2574 		dev->zmii_ph = 0;
2575 	if (emac_read_uint_prop(np, "zmii-channel", &dev->zmii_port, 0))
2576 		dev->zmii_port = 0xffffffff;
2577 	if (emac_read_uint_prop(np, "rgmii-device", &dev->rgmii_ph, 0))
2578 		dev->rgmii_ph = 0;
2579 	if (emac_read_uint_prop(np, "rgmii-channel", &dev->rgmii_port, 0))
2580 		dev->rgmii_port = 0xffffffff;
2581 	if (emac_read_uint_prop(np, "fifo-entry-size", &dev->fifo_entry_size, 0))
2582 		dev->fifo_entry_size = 16;
2583 	if (emac_read_uint_prop(np, "mal-burst-size", &dev->mal_burst_size, 0))
2584 		dev->mal_burst_size = 256;
2585 
2586 	/* PHY mode needs some decoding */
2587 	dev->phy_mode = of_get_phy_mode(np);
2588 	if (dev->phy_mode < 0)
2589 		dev->phy_mode = PHY_MODE_NA;
2590 
2591 	/* Check EMAC version */
2592 	if (of_device_is_compatible(np, "ibm,emac4sync")) {
2593 		dev->features |= (EMAC_FTR_EMAC4 | EMAC_FTR_EMAC4SYNC);
2594 		if (of_device_is_compatible(np, "ibm,emac-460ex") ||
2595 		    of_device_is_compatible(np, "ibm,emac-460gt"))
2596 			dev->features |= EMAC_FTR_460EX_PHY_CLK_FIX;
2597 		if (of_device_is_compatible(np, "ibm,emac-405ex") ||
2598 		    of_device_is_compatible(np, "ibm,emac-405exr"))
2599 			dev->features |= EMAC_FTR_440EP_PHY_CLK_FIX;
2600 		if (of_device_is_compatible(np, "ibm,emac-apm821xx")) {
2601 			dev->features |= (EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE |
2602 					  EMAC_FTR_APM821XX_NO_HALF_DUPLEX |
2603 					  EMAC_FTR_460EX_PHY_CLK_FIX);
2604 		}
2605 	} else if (of_device_is_compatible(np, "ibm,emac4")) {
2606 		dev->features |= EMAC_FTR_EMAC4;
2607 		if (of_device_is_compatible(np, "ibm,emac-440gx"))
2608 			dev->features |= EMAC_FTR_440GX_PHY_CLK_FIX;
2609 	} else {
2610 		if (of_device_is_compatible(np, "ibm,emac-440ep") ||
2611 		    of_device_is_compatible(np, "ibm,emac-440gr"))
2612 			dev->features |= EMAC_FTR_440EP_PHY_CLK_FIX;
2613 		if (of_device_is_compatible(np, "ibm,emac-405ez")) {
2614 #ifdef CONFIG_IBM_EMAC_NO_FLOW_CTRL
2615 			dev->features |= EMAC_FTR_NO_FLOW_CONTROL_40x;
2616 #else
2617 			printk(KERN_ERR "%s: Flow control not disabled!\n",
2618 					np->full_name);
2619 			return -ENXIO;
2620 #endif
2621 		}
2622 
2623 	}
2624 
2625 	/* Fixup some feature bits based on the device tree */
2626 	if (of_get_property(np, "has-inverted-stacr-oc", NULL))
2627 		dev->features |= EMAC_FTR_STACR_OC_INVERT;
2628 	if (of_get_property(np, "has-new-stacr-staopc", NULL))
2629 		dev->features |= EMAC_FTR_HAS_NEW_STACR;
2630 
2631 	/* CAB lacks the appropriate properties */
2632 	if (of_device_is_compatible(np, "ibm,emac-axon"))
2633 		dev->features |= EMAC_FTR_HAS_NEW_STACR |
2634 			EMAC_FTR_STACR_OC_INVERT;
2635 
2636 	/* Enable TAH/ZMII/RGMII features as found */
2637 	if (dev->tah_ph != 0) {
2638 #ifdef CONFIG_IBM_EMAC_TAH
2639 		dev->features |= EMAC_FTR_HAS_TAH;
2640 #else
2641 		printk(KERN_ERR "%s: TAH support not enabled !\n",
2642 		       np->full_name);
2643 		return -ENXIO;
2644 #endif
2645 	}
2646 
2647 	if (dev->zmii_ph != 0) {
2648 #ifdef CONFIG_IBM_EMAC_ZMII
2649 		dev->features |= EMAC_FTR_HAS_ZMII;
2650 #else
2651 		printk(KERN_ERR "%s: ZMII support not enabled !\n",
2652 		       np->full_name);
2653 		return -ENXIO;
2654 #endif
2655 	}
2656 
2657 	if (dev->rgmii_ph != 0) {
2658 #ifdef CONFIG_IBM_EMAC_RGMII
2659 		dev->features |= EMAC_FTR_HAS_RGMII;
2660 #else
2661 		printk(KERN_ERR "%s: RGMII support not enabled !\n",
2662 		       np->full_name);
2663 		return -ENXIO;
2664 #endif
2665 	}
2666 
2667 	/* Read MAC-address */
2668 	p = of_get_property(np, "local-mac-address", NULL);
2669 	if (p == NULL) {
2670 		printk(KERN_ERR "%s: Can't find local-mac-address property\n",
2671 		       np->full_name);
2672 		return -ENXIO;
2673 	}
2674 	memcpy(dev->ndev->dev_addr, p, ETH_ALEN);
2675 
2676 	/* IAHT and GAHT filter parameterization */
2677 	if (emac_has_feature(dev, EMAC_FTR_EMAC4SYNC)) {
2678 		dev->xaht_slots_shift = EMAC4SYNC_XAHT_SLOTS_SHIFT;
2679 		dev->xaht_width_shift = EMAC4SYNC_XAHT_WIDTH_SHIFT;
2680 	} else {
2681 		dev->xaht_slots_shift = EMAC4_XAHT_SLOTS_SHIFT;
2682 		dev->xaht_width_shift = EMAC4_XAHT_WIDTH_SHIFT;
2683 	}
2684 
2685 	DBG(dev, "features     : 0x%08x / 0x%08x\n", dev->features, EMAC_FTRS_POSSIBLE);
2686 	DBG(dev, "tx_fifo_size : %d (%d gige)\n", dev->tx_fifo_size, dev->tx_fifo_size_gige);
2687 	DBG(dev, "rx_fifo_size : %d (%d gige)\n", dev->rx_fifo_size, dev->rx_fifo_size_gige);
2688 	DBG(dev, "max_mtu      : %d\n", dev->max_mtu);
2689 	DBG(dev, "OPB freq     : %d\n", dev->opb_bus_freq);
2690 
2691 	return 0;
2692 }
2693 
2694 static const struct net_device_ops emac_netdev_ops = {
2695 	.ndo_open		= emac_open,
2696 	.ndo_stop		= emac_close,
2697 	.ndo_get_stats		= emac_stats,
2698 	.ndo_set_rx_mode	= emac_set_multicast_list,
2699 	.ndo_do_ioctl		= emac_ioctl,
2700 	.ndo_tx_timeout		= emac_tx_timeout,
2701 	.ndo_validate_addr	= eth_validate_addr,
2702 	.ndo_set_mac_address	= eth_mac_addr,
2703 	.ndo_start_xmit		= emac_start_xmit,
2704 	.ndo_change_mtu		= eth_change_mtu,
2705 };
2706 
2707 static const struct net_device_ops emac_gige_netdev_ops = {
2708 	.ndo_open		= emac_open,
2709 	.ndo_stop		= emac_close,
2710 	.ndo_get_stats		= emac_stats,
2711 	.ndo_set_rx_mode	= emac_set_multicast_list,
2712 	.ndo_do_ioctl		= emac_ioctl,
2713 	.ndo_tx_timeout		= emac_tx_timeout,
2714 	.ndo_validate_addr	= eth_validate_addr,
2715 	.ndo_set_mac_address	= eth_mac_addr,
2716 	.ndo_start_xmit		= emac_start_xmit_sg,
2717 	.ndo_change_mtu		= emac_change_mtu,
2718 };
2719 
2720 static int emac_probe(struct platform_device *ofdev)
2721 {
2722 	struct net_device *ndev;
2723 	struct emac_instance *dev;
2724 	struct device_node *np = ofdev->dev.of_node;
2725 	struct device_node **blist = NULL;
2726 	int err, i;
2727 
2728 	/* Skip unused/unwired EMACS.  We leave the check for an unused
2729 	 * property here for now, but new flat device trees should set a
2730 	 * status property to "disabled" instead.
2731 	 */
2732 	if (of_get_property(np, "unused", NULL) || !of_device_is_available(np))
2733 		return -ENODEV;
2734 
2735 	/* Find ourselves in the bootlist if we are there */
2736 	for (i = 0; i < EMAC_BOOT_LIST_SIZE; i++)
2737 		if (emac_boot_list[i] == np)
2738 			blist = &emac_boot_list[i];
2739 
2740 	/* Allocate our net_device structure */
2741 	err = -ENOMEM;
2742 	ndev = alloc_etherdev(sizeof(struct emac_instance));
2743 	if (!ndev)
2744 		goto err_gone;
2745 
2746 	dev = netdev_priv(ndev);
2747 	dev->ndev = ndev;
2748 	dev->ofdev = ofdev;
2749 	dev->blist = blist;
2750 	SET_NETDEV_DEV(ndev, &ofdev->dev);
2751 
2752 	/* Initialize some embedded data structures */
2753 	mutex_init(&dev->mdio_lock);
2754 	mutex_init(&dev->link_lock);
2755 	spin_lock_init(&dev->lock);
2756 	INIT_WORK(&dev->reset_work, emac_reset_work);
2757 
2758 	/* Init various config data based on device-tree */
2759 	err = emac_init_config(dev);
2760 	if (err != 0)
2761 		goto err_free;
2762 
2763 	/* Get interrupts. EMAC irq is mandatory, WOL irq is optional */
2764 	dev->emac_irq = irq_of_parse_and_map(np, 0);
2765 	dev->wol_irq = irq_of_parse_and_map(np, 1);
2766 	if (dev->emac_irq == NO_IRQ) {
2767 		printk(KERN_ERR "%s: Can't map main interrupt\n", np->full_name);
2768 		goto err_free;
2769 	}
2770 	ndev->irq = dev->emac_irq;
2771 
2772 	/* Map EMAC regs */
2773 	if (of_address_to_resource(np, 0, &dev->rsrc_regs)) {
2774 		printk(KERN_ERR "%s: Can't get registers address\n",
2775 		       np->full_name);
2776 		goto err_irq_unmap;
2777 	}
2778 	// TODO : request_mem_region
2779 	dev->emacp = ioremap(dev->rsrc_regs.start,
2780 			     resource_size(&dev->rsrc_regs));
2781 	if (dev->emacp == NULL) {
2782 		printk(KERN_ERR "%s: Can't map device registers!\n",
2783 		       np->full_name);
2784 		err = -ENOMEM;
2785 		goto err_irq_unmap;
2786 	}
2787 
2788 	/* Wait for dependent devices */
2789 	err = emac_wait_deps(dev);
2790 	if (err) {
2791 		printk(KERN_ERR
2792 		       "%s: Timeout waiting for dependent devices\n",
2793 		       np->full_name);
2794 		/*  display more info about what's missing ? */
2795 		goto err_reg_unmap;
2796 	}
2797 	dev->mal = platform_get_drvdata(dev->mal_dev);
2798 	if (dev->mdio_dev != NULL)
2799 		dev->mdio_instance = platform_get_drvdata(dev->mdio_dev);
2800 
2801 	/* Register with MAL */
2802 	dev->commac.ops = &emac_commac_ops;
2803 	dev->commac.dev = dev;
2804 	dev->commac.tx_chan_mask = MAL_CHAN_MASK(dev->mal_tx_chan);
2805 	dev->commac.rx_chan_mask = MAL_CHAN_MASK(dev->mal_rx_chan);
2806 	err = mal_register_commac(dev->mal, &dev->commac);
2807 	if (err) {
2808 		printk(KERN_ERR "%s: failed to register with mal %s!\n",
2809 		       np->full_name, dev->mal_dev->dev.of_node->full_name);
2810 		goto err_rel_deps;
2811 	}
2812 	dev->rx_skb_size = emac_rx_skb_size(ndev->mtu);
2813 	dev->rx_sync_size = emac_rx_sync_size(ndev->mtu);
2814 
2815 	/* Get pointers to BD rings */
2816 	dev->tx_desc =
2817 	    dev->mal->bd_virt + mal_tx_bd_offset(dev->mal, dev->mal_tx_chan);
2818 	dev->rx_desc =
2819 	    dev->mal->bd_virt + mal_rx_bd_offset(dev->mal, dev->mal_rx_chan);
2820 
2821 	DBG(dev, "tx_desc %p" NL, dev->tx_desc);
2822 	DBG(dev, "rx_desc %p" NL, dev->rx_desc);
2823 
2824 	/* Clean rings */
2825 	memset(dev->tx_desc, 0, NUM_TX_BUFF * sizeof(struct mal_descriptor));
2826 	memset(dev->rx_desc, 0, NUM_RX_BUFF * sizeof(struct mal_descriptor));
2827 	memset(dev->tx_skb, 0, NUM_TX_BUFF * sizeof(struct sk_buff *));
2828 	memset(dev->rx_skb, 0, NUM_RX_BUFF * sizeof(struct sk_buff *));
2829 
2830 	/* Attach to ZMII, if needed */
2831 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII) &&
2832 	    (err = zmii_attach(dev->zmii_dev, dev->zmii_port, &dev->phy_mode)) != 0)
2833 		goto err_unreg_commac;
2834 
2835 	/* Attach to RGMII, if needed */
2836 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII) &&
2837 	    (err = rgmii_attach(dev->rgmii_dev, dev->rgmii_port, dev->phy_mode)) != 0)
2838 		goto err_detach_zmii;
2839 
2840 	/* Attach to TAH, if needed */
2841 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH) &&
2842 	    (err = tah_attach(dev->tah_dev, dev->tah_port)) != 0)
2843 		goto err_detach_rgmii;
2844 
2845 	/* Set some link defaults before we can find out real parameters */
2846 	dev->phy.speed = SPEED_100;
2847 	dev->phy.duplex = DUPLEX_FULL;
2848 	dev->phy.autoneg = AUTONEG_DISABLE;
2849 	dev->phy.pause = dev->phy.asym_pause = 0;
2850 	dev->stop_timeout = STOP_TIMEOUT_100;
2851 	INIT_DELAYED_WORK(&dev->link_work, emac_link_timer);
2852 
2853 	/* Some SoCs like APM821xx does not support Half Duplex mode. */
2854 	if (emac_has_feature(dev, EMAC_FTR_APM821XX_NO_HALF_DUPLEX)) {
2855 		dev->phy_feat_exc = (SUPPORTED_1000baseT_Half |
2856 				     SUPPORTED_100baseT_Half |
2857 				     SUPPORTED_10baseT_Half);
2858 	}
2859 
2860 	/* Find PHY if any */
2861 	err = emac_init_phy(dev);
2862 	if (err != 0)
2863 		goto err_detach_tah;
2864 
2865 	if (dev->tah_dev) {
2866 		ndev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG;
2867 		ndev->features |= ndev->hw_features | NETIF_F_RXCSUM;
2868 	}
2869 	ndev->watchdog_timeo = 5 * HZ;
2870 	if (emac_phy_supports_gige(dev->phy_mode)) {
2871 		ndev->netdev_ops = &emac_gige_netdev_ops;
2872 		dev->commac.ops = &emac_commac_sg_ops;
2873 	} else
2874 		ndev->netdev_ops = &emac_netdev_ops;
2875 	ndev->ethtool_ops = &emac_ethtool_ops;
2876 
2877 	netif_carrier_off(ndev);
2878 
2879 	err = register_netdev(ndev);
2880 	if (err) {
2881 		printk(KERN_ERR "%s: failed to register net device (%d)!\n",
2882 		       np->full_name, err);
2883 		goto err_detach_tah;
2884 	}
2885 
2886 	/* Set our drvdata last as we don't want them visible until we are
2887 	 * fully initialized
2888 	 */
2889 	wmb();
2890 	platform_set_drvdata(ofdev, dev);
2891 
2892 	/* There's a new kid in town ! Let's tell everybody */
2893 	wake_up_all(&emac_probe_wait);
2894 
2895 
2896 	printk(KERN_INFO "%s: EMAC-%d %s, MAC %pM\n",
2897 	       ndev->name, dev->cell_index, np->full_name, ndev->dev_addr);
2898 
2899 	if (dev->phy_mode == PHY_MODE_SGMII)
2900 		printk(KERN_NOTICE "%s: in SGMII mode\n", ndev->name);
2901 
2902 	if (dev->phy.address >= 0)
2903 		printk("%s: found %s PHY (0x%02x)\n", ndev->name,
2904 		       dev->phy.def->name, dev->phy.address);
2905 
2906 	emac_dbg_register(dev);
2907 
2908 	/* Life is good */
2909 	return 0;
2910 
2911 	/* I have a bad feeling about this ... */
2912 
2913  err_detach_tah:
2914 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
2915 		tah_detach(dev->tah_dev, dev->tah_port);
2916  err_detach_rgmii:
2917 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
2918 		rgmii_detach(dev->rgmii_dev, dev->rgmii_port);
2919  err_detach_zmii:
2920 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
2921 		zmii_detach(dev->zmii_dev, dev->zmii_port);
2922  err_unreg_commac:
2923 	mal_unregister_commac(dev->mal, &dev->commac);
2924  err_rel_deps:
2925 	emac_put_deps(dev);
2926  err_reg_unmap:
2927 	iounmap(dev->emacp);
2928  err_irq_unmap:
2929 	if (dev->wol_irq != NO_IRQ)
2930 		irq_dispose_mapping(dev->wol_irq);
2931 	if (dev->emac_irq != NO_IRQ)
2932 		irq_dispose_mapping(dev->emac_irq);
2933  err_free:
2934 	free_netdev(ndev);
2935  err_gone:
2936 	/* if we were on the bootlist, remove us as we won't show up and
2937 	 * wake up all waiters to notify them in case they were waiting
2938 	 * on us
2939 	 */
2940 	if (blist) {
2941 		*blist = NULL;
2942 		wake_up_all(&emac_probe_wait);
2943 	}
2944 	return err;
2945 }
2946 
2947 static int emac_remove(struct platform_device *ofdev)
2948 {
2949 	struct emac_instance *dev = platform_get_drvdata(ofdev);
2950 
2951 	DBG(dev, "remove" NL);
2952 
2953 	unregister_netdev(dev->ndev);
2954 
2955 	cancel_work_sync(&dev->reset_work);
2956 
2957 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
2958 		tah_detach(dev->tah_dev, dev->tah_port);
2959 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
2960 		rgmii_detach(dev->rgmii_dev, dev->rgmii_port);
2961 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
2962 		zmii_detach(dev->zmii_dev, dev->zmii_port);
2963 
2964 	busy_phy_map &= ~(1 << dev->phy.address);
2965 	DBG(dev, "busy_phy_map now %#x" NL, busy_phy_map);
2966 
2967 	mal_unregister_commac(dev->mal, &dev->commac);
2968 	emac_put_deps(dev);
2969 
2970 	emac_dbg_unregister(dev);
2971 	iounmap(dev->emacp);
2972 
2973 	if (dev->wol_irq != NO_IRQ)
2974 		irq_dispose_mapping(dev->wol_irq);
2975 	if (dev->emac_irq != NO_IRQ)
2976 		irq_dispose_mapping(dev->emac_irq);
2977 
2978 	free_netdev(dev->ndev);
2979 
2980 	return 0;
2981 }
2982 
2983 /* XXX Features in here should be replaced by properties... */
2984 static struct of_device_id emac_match[] =
2985 {
2986 	{
2987 		.type		= "network",
2988 		.compatible	= "ibm,emac",
2989 	},
2990 	{
2991 		.type		= "network",
2992 		.compatible	= "ibm,emac4",
2993 	},
2994 	{
2995 		.type		= "network",
2996 		.compatible	= "ibm,emac4sync",
2997 	},
2998 	{},
2999 };
3000 MODULE_DEVICE_TABLE(of, emac_match);
3001 
3002 static struct platform_driver emac_driver = {
3003 	.driver = {
3004 		.name = "emac",
3005 		.of_match_table = emac_match,
3006 	},
3007 	.probe = emac_probe,
3008 	.remove = emac_remove,
3009 };
3010 
3011 static void __init emac_make_bootlist(void)
3012 {
3013 	struct device_node *np = NULL;
3014 	int j, max, i = 0, k;
3015 	int cell_indices[EMAC_BOOT_LIST_SIZE];
3016 
3017 	/* Collect EMACs */
3018 	while((np = of_find_all_nodes(np)) != NULL) {
3019 		const u32 *idx;
3020 
3021 		if (of_match_node(emac_match, np) == NULL)
3022 			continue;
3023 		if (of_get_property(np, "unused", NULL))
3024 			continue;
3025 		idx = of_get_property(np, "cell-index", NULL);
3026 		if (idx == NULL)
3027 			continue;
3028 		cell_indices[i] = *idx;
3029 		emac_boot_list[i++] = of_node_get(np);
3030 		if (i >= EMAC_BOOT_LIST_SIZE) {
3031 			of_node_put(np);
3032 			break;
3033 		}
3034 	}
3035 	max = i;
3036 
3037 	/* Bubble sort them (doh, what a creative algorithm :-) */
3038 	for (i = 0; max > 1 && (i < (max - 1)); i++)
3039 		for (j = i; j < max; j++) {
3040 			if (cell_indices[i] > cell_indices[j]) {
3041 				np = emac_boot_list[i];
3042 				emac_boot_list[i] = emac_boot_list[j];
3043 				emac_boot_list[j] = np;
3044 				k = cell_indices[i];
3045 				cell_indices[i] = cell_indices[j];
3046 				cell_indices[j] = k;
3047 			}
3048 		}
3049 }
3050 
3051 static int __init emac_init(void)
3052 {
3053 	int rc;
3054 
3055 	printk(KERN_INFO DRV_DESC ", version " DRV_VERSION "\n");
3056 
3057 	/* Init debug stuff */
3058 	emac_init_debug();
3059 
3060 	/* Build EMAC boot list */
3061 	emac_make_bootlist();
3062 
3063 	/* Init submodules */
3064 	rc = mal_init();
3065 	if (rc)
3066 		goto err;
3067 	rc = zmii_init();
3068 	if (rc)
3069 		goto err_mal;
3070 	rc = rgmii_init();
3071 	if (rc)
3072 		goto err_zmii;
3073 	rc = tah_init();
3074 	if (rc)
3075 		goto err_rgmii;
3076 	rc = platform_driver_register(&emac_driver);
3077 	if (rc)
3078 		goto err_tah;
3079 
3080 	return 0;
3081 
3082  err_tah:
3083 	tah_exit();
3084  err_rgmii:
3085 	rgmii_exit();
3086  err_zmii:
3087 	zmii_exit();
3088  err_mal:
3089 	mal_exit();
3090  err:
3091 	return rc;
3092 }
3093 
3094 static void __exit emac_exit(void)
3095 {
3096 	int i;
3097 
3098 	platform_driver_unregister(&emac_driver);
3099 
3100 	tah_exit();
3101 	rgmii_exit();
3102 	zmii_exit();
3103 	mal_exit();
3104 	emac_fini_debug();
3105 
3106 	/* Destroy EMAC boot list */
3107 	for (i = 0; i < EMAC_BOOT_LIST_SIZE; i++)
3108 		of_node_put(emac_boot_list[i]);
3109 }
3110 
3111 module_init(emac_init);
3112 module_exit(emac_exit);
3113