xref: /openbmc/linux/drivers/net/ethernet/ibm/emac/core.c (revision 1eb4c977)
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_net.h>
43 #include <linux/slab.h>
44 
45 #include <asm/processor.h>
46 #include <asm/io.h>
47 #include <asm/dma.h>
48 #include <asm/uaccess.h>
49 #include <asm/dcr.h>
50 #include <asm/dcr-regs.h>
51 
52 #include "core.h"
53 
54 /*
55  * Lack of dma_unmap_???? calls is intentional.
56  *
57  * API-correct usage requires additional support state information to be
58  * maintained for every RX and TX buffer descriptor (BD). Unfortunately, due to
59  * EMAC design (e.g. TX buffer passed from network stack can be split into
60  * several BDs, dma_map_single/dma_map_page can be used to map particular BD),
61  * maintaining such information will add additional overhead.
62  * Current DMA API implementation for 4xx processors only ensures cache coherency
63  * and dma_unmap_???? routines are empty and are likely to stay this way.
64  * I decided to omit dma_unmap_??? calls because I don't want to add additional
65  * complexity just for the sake of following some abstract API, when it doesn't
66  * add any real benefit to the driver. I understand that this decision maybe
67  * controversial, but I really tried to make code API-correct and efficient
68  * at the same time and didn't come up with code I liked :(.                --ebs
69  */
70 
71 #define DRV_NAME        "emac"
72 #define DRV_VERSION     "3.54"
73 #define DRV_DESC        "PPC 4xx OCP EMAC driver"
74 
75 MODULE_DESCRIPTION(DRV_DESC);
76 MODULE_AUTHOR
77     ("Eugene Surovegin <eugene.surovegin@zultys.com> or <ebs@ebshome.net>");
78 MODULE_LICENSE("GPL");
79 
80 /*
81  * PPC64 doesn't (yet) have a cacheable_memcpy
82  */
83 #ifdef CONFIG_PPC64
84 #define cacheable_memcpy(d,s,n) memcpy((d),(s),(n))
85 #endif
86 
87 /* minimum number of free TX descriptors required to wake up TX process */
88 #define EMAC_TX_WAKEUP_THRESH		(NUM_TX_BUFF / 4)
89 
90 /* If packet size is less than this number, we allocate small skb and copy packet
91  * contents into it instead of just sending original big skb up
92  */
93 #define EMAC_RX_COPY_THRESH		CONFIG_IBM_EMAC_RX_COPY_THRESHOLD
94 
95 /* Since multiple EMACs share MDIO lines in various ways, we need
96  * to avoid re-using the same PHY ID in cases where the arch didn't
97  * setup precise phy_map entries
98  *
99  * XXX This is something that needs to be reworked as we can have multiple
100  * EMAC "sets" (multiple ASICs containing several EMACs) though we can
101  * probably require in that case to have explicit PHY IDs in the device-tree
102  */
103 static u32 busy_phy_map;
104 static DEFINE_MUTEX(emac_phy_map_lock);
105 
106 /* This is the wait queue used to wait on any event related to probe, that
107  * is discovery of MALs, other EMACs, ZMII/RGMIIs, etc...
108  */
109 static DECLARE_WAIT_QUEUE_HEAD(emac_probe_wait);
110 
111 /* Having stable interface names is a doomed idea. However, it would be nice
112  * if we didn't have completely random interface names at boot too :-) It's
113  * just a matter of making everybody's life easier. Since we are doing
114  * threaded probing, it's a bit harder though. The base idea here is that
115  * we make up a list of all emacs in the device-tree before we register the
116  * driver. Every emac will then wait for the previous one in the list to
117  * initialize before itself. We should also keep that list ordered by
118  * cell_index.
119  * That list is only 4 entries long, meaning that additional EMACs don't
120  * get ordering guarantees unless EMAC_BOOT_LIST_SIZE is increased.
121  */
122 
123 #define EMAC_BOOT_LIST_SIZE	4
124 static struct device_node *emac_boot_list[EMAC_BOOT_LIST_SIZE];
125 
126 /* How long should I wait for dependent devices ? */
127 #define EMAC_PROBE_DEP_TIMEOUT	(HZ * 5)
128 
129 /* I don't want to litter system log with timeout errors
130  * when we have brain-damaged PHY.
131  */
132 static inline void emac_report_timeout_error(struct emac_instance *dev,
133 					     const char *error)
134 {
135 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX |
136 				  EMAC_FTR_460EX_PHY_CLK_FIX |
137 				  EMAC_FTR_440EP_PHY_CLK_FIX))
138 		DBG(dev, "%s" NL, error);
139 	else if (net_ratelimit())
140 		printk(KERN_ERR "%s: %s\n", dev->ofdev->dev.of_node->full_name,
141 			error);
142 }
143 
144 /* EMAC PHY clock workaround:
145  * 440EP/440GR has more sane SDR0_MFR register implementation than 440GX,
146  * which allows controlling each EMAC clock
147  */
148 static inline void emac_rx_clk_tx(struct emac_instance *dev)
149 {
150 #ifdef CONFIG_PPC_DCR_NATIVE
151 	if (emac_has_feature(dev, EMAC_FTR_440EP_PHY_CLK_FIX))
152 		dcri_clrset(SDR0, SDR0_MFR,
153 			    0, SDR0_MFR_ECS >> dev->cell_index);
154 #endif
155 }
156 
157 static inline void emac_rx_clk_default(struct emac_instance *dev)
158 {
159 #ifdef CONFIG_PPC_DCR_NATIVE
160 	if (emac_has_feature(dev, EMAC_FTR_440EP_PHY_CLK_FIX))
161 		dcri_clrset(SDR0, SDR0_MFR,
162 			    SDR0_MFR_ECS >> dev->cell_index, 0);
163 #endif
164 }
165 
166 /* PHY polling intervals */
167 #define PHY_POLL_LINK_ON	HZ
168 #define PHY_POLL_LINK_OFF	(HZ / 5)
169 
170 /* Graceful stop timeouts in us.
171  * We should allow up to 1 frame time (full-duplex, ignoring collisions)
172  */
173 #define STOP_TIMEOUT_10		1230
174 #define STOP_TIMEOUT_100	124
175 #define STOP_TIMEOUT_1000	13
176 #define STOP_TIMEOUT_1000_JUMBO	73
177 
178 static unsigned char default_mcast_addr[] = {
179 	0x01, 0x80, 0xC2, 0x00, 0x00, 0x01
180 };
181 
182 /* Please, keep in sync with struct ibm_emac_stats/ibm_emac_error_stats */
183 static const char emac_stats_keys[EMAC_ETHTOOL_STATS_COUNT][ETH_GSTRING_LEN] = {
184 	"rx_packets", "rx_bytes", "tx_packets", "tx_bytes", "rx_packets_csum",
185 	"tx_packets_csum", "tx_undo", "rx_dropped_stack", "rx_dropped_oom",
186 	"rx_dropped_error", "rx_dropped_resize", "rx_dropped_mtu",
187 	"rx_stopped", "rx_bd_errors", "rx_bd_overrun", "rx_bd_bad_packet",
188 	"rx_bd_runt_packet", "rx_bd_short_event", "rx_bd_alignment_error",
189 	"rx_bd_bad_fcs", "rx_bd_packet_too_long", "rx_bd_out_of_range",
190 	"rx_bd_in_range", "rx_parity", "rx_fifo_overrun", "rx_overrun",
191 	"rx_bad_packet", "rx_runt_packet", "rx_short_event",
192 	"rx_alignment_error", "rx_bad_fcs", "rx_packet_too_long",
193 	"rx_out_of_range", "rx_in_range", "tx_dropped", "tx_bd_errors",
194 	"tx_bd_bad_fcs", "tx_bd_carrier_loss", "tx_bd_excessive_deferral",
195 	"tx_bd_excessive_collisions", "tx_bd_late_collision",
196 	"tx_bd_multple_collisions", "tx_bd_single_collision",
197 	"tx_bd_underrun", "tx_bd_sqe", "tx_parity", "tx_underrun", "tx_sqe",
198 	"tx_errors"
199 };
200 
201 static irqreturn_t emac_irq(int irq, void *dev_instance);
202 static void emac_clean_tx_ring(struct emac_instance *dev);
203 static void __emac_set_multicast_list(struct emac_instance *dev);
204 
205 static inline int emac_phy_supports_gige(int phy_mode)
206 {
207 	return  phy_mode == PHY_MODE_GMII ||
208 		phy_mode == PHY_MODE_RGMII ||
209 		phy_mode == PHY_MODE_SGMII ||
210 		phy_mode == PHY_MODE_TBI ||
211 		phy_mode == PHY_MODE_RTBI;
212 }
213 
214 static inline int emac_phy_gpcs(int phy_mode)
215 {
216 	return  phy_mode == PHY_MODE_SGMII ||
217 		phy_mode == PHY_MODE_TBI ||
218 		phy_mode == PHY_MODE_RTBI;
219 }
220 
221 static inline void emac_tx_enable(struct emac_instance *dev)
222 {
223 	struct emac_regs __iomem *p = dev->emacp;
224 	u32 r;
225 
226 	DBG(dev, "tx_enable" NL);
227 
228 	r = in_be32(&p->mr0);
229 	if (!(r & EMAC_MR0_TXE))
230 		out_be32(&p->mr0, r | EMAC_MR0_TXE);
231 }
232 
233 static void emac_tx_disable(struct emac_instance *dev)
234 {
235 	struct emac_regs __iomem *p = dev->emacp;
236 	u32 r;
237 
238 	DBG(dev, "tx_disable" NL);
239 
240 	r = in_be32(&p->mr0);
241 	if (r & EMAC_MR0_TXE) {
242 		int n = dev->stop_timeout;
243 		out_be32(&p->mr0, r & ~EMAC_MR0_TXE);
244 		while (!(in_be32(&p->mr0) & EMAC_MR0_TXI) && n) {
245 			udelay(1);
246 			--n;
247 		}
248 		if (unlikely(!n))
249 			emac_report_timeout_error(dev, "TX disable timeout");
250 	}
251 }
252 
253 static void emac_rx_enable(struct emac_instance *dev)
254 {
255 	struct emac_regs __iomem *p = dev->emacp;
256 	u32 r;
257 
258 	if (unlikely(test_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags)))
259 		goto out;
260 
261 	DBG(dev, "rx_enable" NL);
262 
263 	r = in_be32(&p->mr0);
264 	if (!(r & EMAC_MR0_RXE)) {
265 		if (unlikely(!(r & EMAC_MR0_RXI))) {
266 			/* Wait if previous async disable is still in progress */
267 			int n = dev->stop_timeout;
268 			while (!(r = in_be32(&p->mr0) & EMAC_MR0_RXI) && n) {
269 				udelay(1);
270 				--n;
271 			}
272 			if (unlikely(!n))
273 				emac_report_timeout_error(dev,
274 							  "RX disable timeout");
275 		}
276 		out_be32(&p->mr0, r | EMAC_MR0_RXE);
277 	}
278  out:
279 	;
280 }
281 
282 static void emac_rx_disable(struct emac_instance *dev)
283 {
284 	struct emac_regs __iomem *p = dev->emacp;
285 	u32 r;
286 
287 	DBG(dev, "rx_disable" NL);
288 
289 	r = in_be32(&p->mr0);
290 	if (r & EMAC_MR0_RXE) {
291 		int n = dev->stop_timeout;
292 		out_be32(&p->mr0, r & ~EMAC_MR0_RXE);
293 		while (!(in_be32(&p->mr0) & EMAC_MR0_RXI) && n) {
294 			udelay(1);
295 			--n;
296 		}
297 		if (unlikely(!n))
298 			emac_report_timeout_error(dev, "RX disable timeout");
299 	}
300 }
301 
302 static inline void emac_netif_stop(struct emac_instance *dev)
303 {
304 	netif_tx_lock_bh(dev->ndev);
305 	netif_addr_lock(dev->ndev);
306 	dev->no_mcast = 1;
307 	netif_addr_unlock(dev->ndev);
308 	netif_tx_unlock_bh(dev->ndev);
309 	dev->ndev->trans_start = jiffies;	/* prevent tx timeout */
310 	mal_poll_disable(dev->mal, &dev->commac);
311 	netif_tx_disable(dev->ndev);
312 }
313 
314 static inline void emac_netif_start(struct emac_instance *dev)
315 {
316 	netif_tx_lock_bh(dev->ndev);
317 	netif_addr_lock(dev->ndev);
318 	dev->no_mcast = 0;
319 	if (dev->mcast_pending && netif_running(dev->ndev))
320 		__emac_set_multicast_list(dev);
321 	netif_addr_unlock(dev->ndev);
322 	netif_tx_unlock_bh(dev->ndev);
323 
324 	netif_wake_queue(dev->ndev);
325 
326 	/* NOTE: unconditional netif_wake_queue is only appropriate
327 	 * so long as all callers are assured to have free tx slots
328 	 * (taken from tg3... though the case where that is wrong is
329 	 *  not terribly harmful)
330 	 */
331 	mal_poll_enable(dev->mal, &dev->commac);
332 }
333 
334 static inline void emac_rx_disable_async(struct emac_instance *dev)
335 {
336 	struct emac_regs __iomem *p = dev->emacp;
337 	u32 r;
338 
339 	DBG(dev, "rx_disable_async" NL);
340 
341 	r = in_be32(&p->mr0);
342 	if (r & EMAC_MR0_RXE)
343 		out_be32(&p->mr0, r & ~EMAC_MR0_RXE);
344 }
345 
346 static int emac_reset(struct emac_instance *dev)
347 {
348 	struct emac_regs __iomem *p = dev->emacp;
349 	int n = 20;
350 
351 	DBG(dev, "reset" NL);
352 
353 	if (!dev->reset_failed) {
354 		/* 40x erratum suggests stopping RX channel before reset,
355 		 * we stop TX as well
356 		 */
357 		emac_rx_disable(dev);
358 		emac_tx_disable(dev);
359 	}
360 
361 #ifdef CONFIG_PPC_DCR_NATIVE
362 	/* Enable internal clock source */
363 	if (emac_has_feature(dev, EMAC_FTR_460EX_PHY_CLK_FIX))
364 		dcri_clrset(SDR0, SDR0_ETH_CFG,
365 			    0, SDR0_ETH_CFG_ECS << dev->cell_index);
366 #endif
367 
368 	out_be32(&p->mr0, EMAC_MR0_SRST);
369 	while ((in_be32(&p->mr0) & EMAC_MR0_SRST) && n)
370 		--n;
371 
372 #ifdef CONFIG_PPC_DCR_NATIVE
373 	 /* Enable external clock source */
374 	if (emac_has_feature(dev, EMAC_FTR_460EX_PHY_CLK_FIX))
375 		dcri_clrset(SDR0, SDR0_ETH_CFG,
376 			    SDR0_ETH_CFG_ECS << dev->cell_index, 0);
377 #endif
378 
379 	if (n) {
380 		dev->reset_failed = 0;
381 		return 0;
382 	} else {
383 		emac_report_timeout_error(dev, "reset timeout");
384 		dev->reset_failed = 1;
385 		return -ETIMEDOUT;
386 	}
387 }
388 
389 static void emac_hash_mc(struct emac_instance *dev)
390 {
391 	const int regs = EMAC_XAHT_REGS(dev);
392 	u32 *gaht_base = emac_gaht_base(dev);
393 	u32 gaht_temp[regs];
394 	struct netdev_hw_addr *ha;
395 	int i;
396 
397 	DBG(dev, "hash_mc %d" NL, netdev_mc_count(dev->ndev));
398 
399 	memset(gaht_temp, 0, sizeof (gaht_temp));
400 
401 	netdev_for_each_mc_addr(ha, dev->ndev) {
402 		int slot, reg, mask;
403 		DBG2(dev, "mc %pM" NL, ha->addr);
404 
405 		slot = EMAC_XAHT_CRC_TO_SLOT(dev,
406 					     ether_crc(ETH_ALEN, ha->addr));
407 		reg = EMAC_XAHT_SLOT_TO_REG(dev, slot);
408 		mask = EMAC_XAHT_SLOT_TO_MASK(dev, slot);
409 
410 		gaht_temp[reg] |= mask;
411 	}
412 
413 	for (i = 0; i < regs; i++)
414 		out_be32(gaht_base + i, gaht_temp[i]);
415 }
416 
417 static inline u32 emac_iff2rmr(struct net_device *ndev)
418 {
419 	struct emac_instance *dev = netdev_priv(ndev);
420 	u32 r;
421 
422 	r = EMAC_RMR_SP | EMAC_RMR_SFCS | EMAC_RMR_IAE | EMAC_RMR_BAE;
423 
424 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
425 	    r |= EMAC4_RMR_BASE;
426 	else
427 	    r |= EMAC_RMR_BASE;
428 
429 	if (ndev->flags & IFF_PROMISC)
430 		r |= EMAC_RMR_PME;
431 	else if (ndev->flags & IFF_ALLMULTI ||
432 			 (netdev_mc_count(ndev) > EMAC_XAHT_SLOTS(dev)))
433 		r |= EMAC_RMR_PMME;
434 	else if (!netdev_mc_empty(ndev))
435 		r |= EMAC_RMR_MAE;
436 
437 	if (emac_has_feature(dev, EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE)) {
438 		r &= ~EMAC4_RMR_MJS_MASK;
439 		r |= EMAC4_RMR_MJS(ndev->mtu);
440 	}
441 
442 	return r;
443 }
444 
445 static u32 __emac_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
446 {
447 	u32 ret = EMAC_MR1_VLE | EMAC_MR1_IST | EMAC_MR1_TR0_MULT;
448 
449 	DBG2(dev, "__emac_calc_base_mr1" NL);
450 
451 	switch(tx_size) {
452 	case 2048:
453 		ret |= EMAC_MR1_TFS_2K;
454 		break;
455 	default:
456 		printk(KERN_WARNING "%s: Unknown Tx FIFO size %d\n",
457 		       dev->ndev->name, tx_size);
458 	}
459 
460 	switch(rx_size) {
461 	case 16384:
462 		ret |= EMAC_MR1_RFS_16K;
463 		break;
464 	case 4096:
465 		ret |= EMAC_MR1_RFS_4K;
466 		break;
467 	default:
468 		printk(KERN_WARNING "%s: Unknown Rx FIFO size %d\n",
469 		       dev->ndev->name, rx_size);
470 	}
471 
472 	return ret;
473 }
474 
475 static u32 __emac4_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
476 {
477 	u32 ret = EMAC_MR1_VLE | EMAC_MR1_IST | EMAC4_MR1_TR |
478 		EMAC4_MR1_OBCI(dev->opb_bus_freq / 1000000);
479 
480 	DBG2(dev, "__emac4_calc_base_mr1" NL);
481 
482 	switch(tx_size) {
483 	case 16384:
484 		ret |= EMAC4_MR1_TFS_16K;
485 		break;
486 	case 4096:
487 		ret |= EMAC4_MR1_TFS_4K;
488 		break;
489 	case 2048:
490 		ret |= EMAC4_MR1_TFS_2K;
491 		break;
492 	default:
493 		printk(KERN_WARNING "%s: Unknown Tx FIFO size %d\n",
494 		       dev->ndev->name, tx_size);
495 	}
496 
497 	switch(rx_size) {
498 	case 16384:
499 		ret |= EMAC4_MR1_RFS_16K;
500 		break;
501 	case 4096:
502 		ret |= EMAC4_MR1_RFS_4K;
503 		break;
504 	case 2048:
505 		ret |= EMAC4_MR1_RFS_2K;
506 		break;
507 	default:
508 		printk(KERN_WARNING "%s: Unknown Rx FIFO size %d\n",
509 		       dev->ndev->name, rx_size);
510 	}
511 
512 	return ret;
513 }
514 
515 static u32 emac_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
516 {
517 	return emac_has_feature(dev, EMAC_FTR_EMAC4) ?
518 		__emac4_calc_base_mr1(dev, tx_size, rx_size) :
519 		__emac_calc_base_mr1(dev, tx_size, rx_size);
520 }
521 
522 static inline u32 emac_calc_trtr(struct emac_instance *dev, unsigned int size)
523 {
524 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
525 		return ((size >> 6) - 1) << EMAC_TRTR_SHIFT_EMAC4;
526 	else
527 		return ((size >> 6) - 1) << EMAC_TRTR_SHIFT;
528 }
529 
530 static inline u32 emac_calc_rwmr(struct emac_instance *dev,
531 				 unsigned int low, unsigned int high)
532 {
533 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
534 		return (low << 22) | ( (high & 0x3ff) << 6);
535 	else
536 		return (low << 23) | ( (high & 0x1ff) << 7);
537 }
538 
539 static int emac_configure(struct emac_instance *dev)
540 {
541 	struct emac_regs __iomem *p = dev->emacp;
542 	struct net_device *ndev = dev->ndev;
543 	int tx_size, rx_size, link = netif_carrier_ok(dev->ndev);
544 	u32 r, mr1 = 0;
545 
546 	DBG(dev, "configure" NL);
547 
548 	if (!link) {
549 		out_be32(&p->mr1, in_be32(&p->mr1)
550 			 | EMAC_MR1_FDE | EMAC_MR1_ILE);
551 		udelay(100);
552 	} else if (emac_reset(dev) < 0)
553 		return -ETIMEDOUT;
554 
555 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
556 		tah_reset(dev->tah_dev);
557 
558 	DBG(dev, " link = %d duplex = %d, pause = %d, asym_pause = %d\n",
559 	    link, dev->phy.duplex, dev->phy.pause, dev->phy.asym_pause);
560 
561 	/* Default fifo sizes */
562 	tx_size = dev->tx_fifo_size;
563 	rx_size = dev->rx_fifo_size;
564 
565 	/* No link, force loopback */
566 	if (!link)
567 		mr1 = EMAC_MR1_FDE | EMAC_MR1_ILE;
568 
569 	/* Check for full duplex */
570 	else if (dev->phy.duplex == DUPLEX_FULL)
571 		mr1 |= EMAC_MR1_FDE | EMAC_MR1_MWSW_001;
572 
573 	/* Adjust fifo sizes, mr1 and timeouts based on link speed */
574 	dev->stop_timeout = STOP_TIMEOUT_10;
575 	switch (dev->phy.speed) {
576 	case SPEED_1000:
577 		if (emac_phy_gpcs(dev->phy.mode)) {
578 			mr1 |= EMAC_MR1_MF_1000GPCS | EMAC_MR1_MF_IPPA(
579 				(dev->phy.gpcs_address != 0xffffffff) ?
580 				 dev->phy.gpcs_address : dev->phy.address);
581 
582 			/* Put some arbitrary OUI, Manuf & Rev IDs so we can
583 			 * identify this GPCS PHY later.
584 			 */
585 			out_be32(&p->u1.emac4.ipcr, 0xdeadbeef);
586 		} else
587 			mr1 |= EMAC_MR1_MF_1000;
588 
589 		/* Extended fifo sizes */
590 		tx_size = dev->tx_fifo_size_gige;
591 		rx_size = dev->rx_fifo_size_gige;
592 
593 		if (dev->ndev->mtu > ETH_DATA_LEN) {
594 			if (emac_has_feature(dev, EMAC_FTR_EMAC4))
595 				mr1 |= EMAC4_MR1_JPSM;
596 			else
597 				mr1 |= EMAC_MR1_JPSM;
598 			dev->stop_timeout = STOP_TIMEOUT_1000_JUMBO;
599 		} else
600 			dev->stop_timeout = STOP_TIMEOUT_1000;
601 		break;
602 	case SPEED_100:
603 		mr1 |= EMAC_MR1_MF_100;
604 		dev->stop_timeout = STOP_TIMEOUT_100;
605 		break;
606 	default: /* make gcc happy */
607 		break;
608 	}
609 
610 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
611 		rgmii_set_speed(dev->rgmii_dev, dev->rgmii_port,
612 				dev->phy.speed);
613 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
614 		zmii_set_speed(dev->zmii_dev, dev->zmii_port, dev->phy.speed);
615 
616 	/* on 40x erratum forces us to NOT use integrated flow control,
617 	 * let's hope it works on 44x ;)
618 	 */
619 	if (!emac_has_feature(dev, EMAC_FTR_NO_FLOW_CONTROL_40x) &&
620 	    dev->phy.duplex == DUPLEX_FULL) {
621 		if (dev->phy.pause)
622 			mr1 |= EMAC_MR1_EIFC | EMAC_MR1_APP;
623 		else if (dev->phy.asym_pause)
624 			mr1 |= EMAC_MR1_APP;
625 	}
626 
627 	/* Add base settings & fifo sizes & program MR1 */
628 	mr1 |= emac_calc_base_mr1(dev, tx_size, rx_size);
629 	out_be32(&p->mr1, mr1);
630 
631 	/* Set individual MAC address */
632 	out_be32(&p->iahr, (ndev->dev_addr[0] << 8) | ndev->dev_addr[1]);
633 	out_be32(&p->ialr, (ndev->dev_addr[2] << 24) |
634 		 (ndev->dev_addr[3] << 16) | (ndev->dev_addr[4] << 8) |
635 		 ndev->dev_addr[5]);
636 
637 	/* VLAN Tag Protocol ID */
638 	out_be32(&p->vtpid, 0x8100);
639 
640 	/* Receive mode register */
641 	r = emac_iff2rmr(ndev);
642 	if (r & EMAC_RMR_MAE)
643 		emac_hash_mc(dev);
644 	out_be32(&p->rmr, r);
645 
646 	/* FIFOs thresholds */
647 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
648 		r = EMAC4_TMR1((dev->mal_burst_size / dev->fifo_entry_size) + 1,
649 			       tx_size / 2 / dev->fifo_entry_size);
650 	else
651 		r = EMAC_TMR1((dev->mal_burst_size / dev->fifo_entry_size) + 1,
652 			      tx_size / 2 / dev->fifo_entry_size);
653 	out_be32(&p->tmr1, r);
654 	out_be32(&p->trtr, emac_calc_trtr(dev, tx_size / 2));
655 
656 	/* PAUSE frame is sent when RX FIFO reaches its high-water mark,
657 	   there should be still enough space in FIFO to allow the our link
658 	   partner time to process this frame and also time to send PAUSE
659 	   frame itself.
660 
661 	   Here is the worst case scenario for the RX FIFO "headroom"
662 	   (from "The Switch Book") (100Mbps, without preamble, inter-frame gap):
663 
664 	   1) One maximum-length frame on TX                    1522 bytes
665 	   2) One PAUSE frame time                                64 bytes
666 	   3) PAUSE frame decode time allowance                   64 bytes
667 	   4) One maximum-length frame on RX                    1522 bytes
668 	   5) Round-trip propagation delay of the link (100Mb)    15 bytes
669 	   ----------
670 	   3187 bytes
671 
672 	   I chose to set high-water mark to RX_FIFO_SIZE / 4 (1024 bytes)
673 	   low-water mark  to RX_FIFO_SIZE / 8 (512 bytes)
674 	 */
675 	r = emac_calc_rwmr(dev, rx_size / 8 / dev->fifo_entry_size,
676 			   rx_size / 4 / dev->fifo_entry_size);
677 	out_be32(&p->rwmr, r);
678 
679 	/* Set PAUSE timer to the maximum */
680 	out_be32(&p->ptr, 0xffff);
681 
682 	/* IRQ sources */
683 	r = EMAC_ISR_OVR | EMAC_ISR_BP | EMAC_ISR_SE |
684 		EMAC_ISR_ALE | EMAC_ISR_BFCS | EMAC_ISR_PTLE | EMAC_ISR_ORE |
685 		EMAC_ISR_IRE | EMAC_ISR_TE;
686 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
687 	    r |= EMAC4_ISR_TXPE | EMAC4_ISR_RXPE /* | EMAC4_ISR_TXUE |
688 						  EMAC4_ISR_RXOE | */;
689 	out_be32(&p->iser,  r);
690 
691 	/* We need to take GPCS PHY out of isolate mode after EMAC reset */
692 	if (emac_phy_gpcs(dev->phy.mode)) {
693 		if (dev->phy.gpcs_address != 0xffffffff)
694 			emac_mii_reset_gpcs(&dev->phy);
695 		else
696 			emac_mii_reset_phy(&dev->phy);
697 	}
698 
699 	return 0;
700 }
701 
702 static void emac_reinitialize(struct emac_instance *dev)
703 {
704 	DBG(dev, "reinitialize" NL);
705 
706 	emac_netif_stop(dev);
707 	if (!emac_configure(dev)) {
708 		emac_tx_enable(dev);
709 		emac_rx_enable(dev);
710 	}
711 	emac_netif_start(dev);
712 }
713 
714 static void emac_full_tx_reset(struct emac_instance *dev)
715 {
716 	DBG(dev, "full_tx_reset" NL);
717 
718 	emac_tx_disable(dev);
719 	mal_disable_tx_channel(dev->mal, dev->mal_tx_chan);
720 	emac_clean_tx_ring(dev);
721 	dev->tx_cnt = dev->tx_slot = dev->ack_slot = 0;
722 
723 	emac_configure(dev);
724 
725 	mal_enable_tx_channel(dev->mal, dev->mal_tx_chan);
726 	emac_tx_enable(dev);
727 	emac_rx_enable(dev);
728 }
729 
730 static void emac_reset_work(struct work_struct *work)
731 {
732 	struct emac_instance *dev = container_of(work, struct emac_instance, reset_work);
733 
734 	DBG(dev, "reset_work" NL);
735 
736 	mutex_lock(&dev->link_lock);
737 	if (dev->opened) {
738 		emac_netif_stop(dev);
739 		emac_full_tx_reset(dev);
740 		emac_netif_start(dev);
741 	}
742 	mutex_unlock(&dev->link_lock);
743 }
744 
745 static void emac_tx_timeout(struct net_device *ndev)
746 {
747 	struct emac_instance *dev = netdev_priv(ndev);
748 
749 	DBG(dev, "tx_timeout" NL);
750 
751 	schedule_work(&dev->reset_work);
752 }
753 
754 
755 static inline int emac_phy_done(struct emac_instance *dev, u32 stacr)
756 {
757 	int done = !!(stacr & EMAC_STACR_OC);
758 
759 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
760 		done = !done;
761 
762 	return done;
763 };
764 
765 static int __emac_mdio_read(struct emac_instance *dev, u8 id, u8 reg)
766 {
767 	struct emac_regs __iomem *p = dev->emacp;
768 	u32 r = 0;
769 	int n, err = -ETIMEDOUT;
770 
771 	mutex_lock(&dev->mdio_lock);
772 
773 	DBG2(dev, "mdio_read(%02x,%02x)" NL, id, reg);
774 
775 	/* Enable proper MDIO port */
776 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
777 		zmii_get_mdio(dev->zmii_dev, dev->zmii_port);
778 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
779 		rgmii_get_mdio(dev->rgmii_dev, dev->rgmii_port);
780 
781 	/* Wait for management interface to become idle */
782 	n = 20;
783 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
784 		udelay(1);
785 		if (!--n) {
786 			DBG2(dev, " -> timeout wait idle\n");
787 			goto bail;
788 		}
789 	}
790 
791 	/* Issue read command */
792 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
793 		r = EMAC4_STACR_BASE(dev->opb_bus_freq);
794 	else
795 		r = EMAC_STACR_BASE(dev->opb_bus_freq);
796 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
797 		r |= EMAC_STACR_OC;
798 	if (emac_has_feature(dev, EMAC_FTR_HAS_NEW_STACR))
799 		r |= EMACX_STACR_STAC_READ;
800 	else
801 		r |= EMAC_STACR_STAC_READ;
802 	r |= (reg & EMAC_STACR_PRA_MASK)
803 		| ((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT);
804 	out_be32(&p->stacr, r);
805 
806 	/* Wait for read to complete */
807 	n = 200;
808 	while (!emac_phy_done(dev, (r = in_be32(&p->stacr)))) {
809 		udelay(1);
810 		if (!--n) {
811 			DBG2(dev, " -> timeout wait complete\n");
812 			goto bail;
813 		}
814 	}
815 
816 	if (unlikely(r & EMAC_STACR_PHYE)) {
817 		DBG(dev, "mdio_read(%02x, %02x) failed" NL, id, reg);
818 		err = -EREMOTEIO;
819 		goto bail;
820 	}
821 
822 	r = ((r >> EMAC_STACR_PHYD_SHIFT) & EMAC_STACR_PHYD_MASK);
823 
824 	DBG2(dev, "mdio_read -> %04x" NL, r);
825 	err = 0;
826  bail:
827 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
828 		rgmii_put_mdio(dev->rgmii_dev, dev->rgmii_port);
829 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
830 		zmii_put_mdio(dev->zmii_dev, dev->zmii_port);
831 	mutex_unlock(&dev->mdio_lock);
832 
833 	return err == 0 ? r : err;
834 }
835 
836 static void __emac_mdio_write(struct emac_instance *dev, u8 id, u8 reg,
837 			      u16 val)
838 {
839 	struct emac_regs __iomem *p = dev->emacp;
840 	u32 r = 0;
841 	int n, err = -ETIMEDOUT;
842 
843 	mutex_lock(&dev->mdio_lock);
844 
845 	DBG2(dev, "mdio_write(%02x,%02x,%04x)" NL, id, reg, val);
846 
847 	/* Enable proper MDIO port */
848 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
849 		zmii_get_mdio(dev->zmii_dev, dev->zmii_port);
850 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
851 		rgmii_get_mdio(dev->rgmii_dev, dev->rgmii_port);
852 
853 	/* Wait for management interface to be idle */
854 	n = 20;
855 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
856 		udelay(1);
857 		if (!--n) {
858 			DBG2(dev, " -> timeout wait idle\n");
859 			goto bail;
860 		}
861 	}
862 
863 	/* Issue write command */
864 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
865 		r = EMAC4_STACR_BASE(dev->opb_bus_freq);
866 	else
867 		r = EMAC_STACR_BASE(dev->opb_bus_freq);
868 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
869 		r |= EMAC_STACR_OC;
870 	if (emac_has_feature(dev, EMAC_FTR_HAS_NEW_STACR))
871 		r |= EMACX_STACR_STAC_WRITE;
872 	else
873 		r |= EMAC_STACR_STAC_WRITE;
874 	r |= (reg & EMAC_STACR_PRA_MASK) |
875 		((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT) |
876 		(val << EMAC_STACR_PHYD_SHIFT);
877 	out_be32(&p->stacr, r);
878 
879 	/* Wait for write to complete */
880 	n = 200;
881 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
882 		udelay(1);
883 		if (!--n) {
884 			DBG2(dev, " -> timeout wait complete\n");
885 			goto bail;
886 		}
887 	}
888 	err = 0;
889  bail:
890 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
891 		rgmii_put_mdio(dev->rgmii_dev, dev->rgmii_port);
892 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
893 		zmii_put_mdio(dev->zmii_dev, dev->zmii_port);
894 	mutex_unlock(&dev->mdio_lock);
895 }
896 
897 static int emac_mdio_read(struct net_device *ndev, int id, int reg)
898 {
899 	struct emac_instance *dev = netdev_priv(ndev);
900 	int res;
901 
902 	res = __emac_mdio_read((dev->mdio_instance &&
903 				dev->phy.gpcs_address != id) ?
904 				dev->mdio_instance : dev,
905 			       (u8) id, (u8) reg);
906 	return res;
907 }
908 
909 static void emac_mdio_write(struct net_device *ndev, int id, int reg, int val)
910 {
911 	struct emac_instance *dev = netdev_priv(ndev);
912 
913 	__emac_mdio_write((dev->mdio_instance &&
914 			   dev->phy.gpcs_address != id) ?
915 			   dev->mdio_instance : dev,
916 			  (u8) id, (u8) reg, (u16) val);
917 }
918 
919 /* Tx lock BH */
920 static void __emac_set_multicast_list(struct emac_instance *dev)
921 {
922 	struct emac_regs __iomem *p = dev->emacp;
923 	u32 rmr = emac_iff2rmr(dev->ndev);
924 
925 	DBG(dev, "__multicast %08x" NL, rmr);
926 
927 	/* I decided to relax register access rules here to avoid
928 	 * full EMAC reset.
929 	 *
930 	 * There is a real problem with EMAC4 core if we use MWSW_001 bit
931 	 * in MR1 register and do a full EMAC reset.
932 	 * One TX BD status update is delayed and, after EMAC reset, it
933 	 * never happens, resulting in TX hung (it'll be recovered by TX
934 	 * timeout handler eventually, but this is just gross).
935 	 * So we either have to do full TX reset or try to cheat here :)
936 	 *
937 	 * The only required change is to RX mode register, so I *think* all
938 	 * we need is just to stop RX channel. This seems to work on all
939 	 * tested SoCs.                                                --ebs
940 	 *
941 	 * If we need the full reset, we might just trigger the workqueue
942 	 * and do it async... a bit nasty but should work --BenH
943 	 */
944 	dev->mcast_pending = 0;
945 	emac_rx_disable(dev);
946 	if (rmr & EMAC_RMR_MAE)
947 		emac_hash_mc(dev);
948 	out_be32(&p->rmr, rmr);
949 	emac_rx_enable(dev);
950 }
951 
952 /* Tx lock BH */
953 static void emac_set_multicast_list(struct net_device *ndev)
954 {
955 	struct emac_instance *dev = netdev_priv(ndev);
956 
957 	DBG(dev, "multicast" NL);
958 
959 	BUG_ON(!netif_running(dev->ndev));
960 
961 	if (dev->no_mcast) {
962 		dev->mcast_pending = 1;
963 		return;
964 	}
965 	__emac_set_multicast_list(dev);
966 }
967 
968 static int emac_resize_rx_ring(struct emac_instance *dev, int new_mtu)
969 {
970 	int rx_sync_size = emac_rx_sync_size(new_mtu);
971 	int rx_skb_size = emac_rx_skb_size(new_mtu);
972 	int i, ret = 0;
973 	int mr1_jumbo_bit_change = 0;
974 
975 	mutex_lock(&dev->link_lock);
976 	emac_netif_stop(dev);
977 	emac_rx_disable(dev);
978 	mal_disable_rx_channel(dev->mal, dev->mal_rx_chan);
979 
980 	if (dev->rx_sg_skb) {
981 		++dev->estats.rx_dropped_resize;
982 		dev_kfree_skb(dev->rx_sg_skb);
983 		dev->rx_sg_skb = NULL;
984 	}
985 
986 	/* Make a first pass over RX ring and mark BDs ready, dropping
987 	 * non-processed packets on the way. We need this as a separate pass
988 	 * to simplify error recovery in the case of allocation failure later.
989 	 */
990 	for (i = 0; i < NUM_RX_BUFF; ++i) {
991 		if (dev->rx_desc[i].ctrl & MAL_RX_CTRL_FIRST)
992 			++dev->estats.rx_dropped_resize;
993 
994 		dev->rx_desc[i].data_len = 0;
995 		dev->rx_desc[i].ctrl = MAL_RX_CTRL_EMPTY |
996 		    (i == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
997 	}
998 
999 	/* Reallocate RX ring only if bigger skb buffers are required */
1000 	if (rx_skb_size <= dev->rx_skb_size)
1001 		goto skip;
1002 
1003 	/* Second pass, allocate new skbs */
1004 	for (i = 0; i < NUM_RX_BUFF; ++i) {
1005 		struct sk_buff *skb = alloc_skb(rx_skb_size, GFP_ATOMIC);
1006 		if (!skb) {
1007 			ret = -ENOMEM;
1008 			goto oom;
1009 		}
1010 
1011 		BUG_ON(!dev->rx_skb[i]);
1012 		dev_kfree_skb(dev->rx_skb[i]);
1013 
1014 		skb_reserve(skb, EMAC_RX_SKB_HEADROOM + 2);
1015 		dev->rx_desc[i].data_ptr =
1016 		    dma_map_single(&dev->ofdev->dev, skb->data - 2, rx_sync_size,
1017 				   DMA_FROM_DEVICE) + 2;
1018 		dev->rx_skb[i] = skb;
1019 	}
1020  skip:
1021 	/* Check if we need to change "Jumbo" bit in MR1 */
1022 	if (emac_has_feature(dev, EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE)) {
1023 		mr1_jumbo_bit_change = (new_mtu > ETH_DATA_LEN) ||
1024 				(dev->ndev->mtu > ETH_DATA_LEN);
1025 	} else {
1026 		mr1_jumbo_bit_change = (new_mtu > ETH_DATA_LEN) ^
1027 				(dev->ndev->mtu > ETH_DATA_LEN);
1028 	}
1029 
1030 	if (mr1_jumbo_bit_change) {
1031 		/* This is to prevent starting RX channel in emac_rx_enable() */
1032 		set_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1033 
1034 		dev->ndev->mtu = new_mtu;
1035 		emac_full_tx_reset(dev);
1036 	}
1037 
1038 	mal_set_rcbs(dev->mal, dev->mal_rx_chan, emac_rx_size(new_mtu));
1039  oom:
1040 	/* Restart RX */
1041 	clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1042 	dev->rx_slot = 0;
1043 	mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1044 	emac_rx_enable(dev);
1045 	emac_netif_start(dev);
1046 	mutex_unlock(&dev->link_lock);
1047 
1048 	return ret;
1049 }
1050 
1051 /* Process ctx, rtnl_lock semaphore */
1052 static int emac_change_mtu(struct net_device *ndev, int new_mtu)
1053 {
1054 	struct emac_instance *dev = netdev_priv(ndev);
1055 	int ret = 0;
1056 
1057 	if (new_mtu < EMAC_MIN_MTU || new_mtu > dev->max_mtu)
1058 		return -EINVAL;
1059 
1060 	DBG(dev, "change_mtu(%d)" NL, new_mtu);
1061 
1062 	if (netif_running(ndev)) {
1063 		/* Check if we really need to reinitialize RX ring */
1064 		if (emac_rx_skb_size(ndev->mtu) != emac_rx_skb_size(new_mtu))
1065 			ret = emac_resize_rx_ring(dev, new_mtu);
1066 	}
1067 
1068 	if (!ret) {
1069 		ndev->mtu = new_mtu;
1070 		dev->rx_skb_size = emac_rx_skb_size(new_mtu);
1071 		dev->rx_sync_size = emac_rx_sync_size(new_mtu);
1072 	}
1073 
1074 	return ret;
1075 }
1076 
1077 static void emac_clean_tx_ring(struct emac_instance *dev)
1078 {
1079 	int i;
1080 
1081 	for (i = 0; i < NUM_TX_BUFF; ++i) {
1082 		if (dev->tx_skb[i]) {
1083 			dev_kfree_skb(dev->tx_skb[i]);
1084 			dev->tx_skb[i] = NULL;
1085 			if (dev->tx_desc[i].ctrl & MAL_TX_CTRL_READY)
1086 				++dev->estats.tx_dropped;
1087 		}
1088 		dev->tx_desc[i].ctrl = 0;
1089 		dev->tx_desc[i].data_ptr = 0;
1090 	}
1091 }
1092 
1093 static void emac_clean_rx_ring(struct emac_instance *dev)
1094 {
1095 	int i;
1096 
1097 	for (i = 0; i < NUM_RX_BUFF; ++i)
1098 		if (dev->rx_skb[i]) {
1099 			dev->rx_desc[i].ctrl = 0;
1100 			dev_kfree_skb(dev->rx_skb[i]);
1101 			dev->rx_skb[i] = NULL;
1102 			dev->rx_desc[i].data_ptr = 0;
1103 		}
1104 
1105 	if (dev->rx_sg_skb) {
1106 		dev_kfree_skb(dev->rx_sg_skb);
1107 		dev->rx_sg_skb = NULL;
1108 	}
1109 }
1110 
1111 static inline int emac_alloc_rx_skb(struct emac_instance *dev, int slot,
1112 				    gfp_t flags)
1113 {
1114 	struct sk_buff *skb = alloc_skb(dev->rx_skb_size, flags);
1115 	if (unlikely(!skb))
1116 		return -ENOMEM;
1117 
1118 	dev->rx_skb[slot] = skb;
1119 	dev->rx_desc[slot].data_len = 0;
1120 
1121 	skb_reserve(skb, EMAC_RX_SKB_HEADROOM + 2);
1122 	dev->rx_desc[slot].data_ptr =
1123 	    dma_map_single(&dev->ofdev->dev, skb->data - 2, dev->rx_sync_size,
1124 			   DMA_FROM_DEVICE) + 2;
1125 	wmb();
1126 	dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1127 	    (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1128 
1129 	return 0;
1130 }
1131 
1132 static void emac_print_link_status(struct emac_instance *dev)
1133 {
1134 	if (netif_carrier_ok(dev->ndev))
1135 		printk(KERN_INFO "%s: link is up, %d %s%s\n",
1136 		       dev->ndev->name, dev->phy.speed,
1137 		       dev->phy.duplex == DUPLEX_FULL ? "FDX" : "HDX",
1138 		       dev->phy.pause ? ", pause enabled" :
1139 		       dev->phy.asym_pause ? ", asymmetric pause enabled" : "");
1140 	else
1141 		printk(KERN_INFO "%s: link is down\n", dev->ndev->name);
1142 }
1143 
1144 /* Process ctx, rtnl_lock semaphore */
1145 static int emac_open(struct net_device *ndev)
1146 {
1147 	struct emac_instance *dev = netdev_priv(ndev);
1148 	int err, i;
1149 
1150 	DBG(dev, "open" NL);
1151 
1152 	/* Setup error IRQ handler */
1153 	err = request_irq(dev->emac_irq, emac_irq, 0, "EMAC", dev);
1154 	if (err) {
1155 		printk(KERN_ERR "%s: failed to request IRQ %d\n",
1156 		       ndev->name, dev->emac_irq);
1157 		return err;
1158 	}
1159 
1160 	/* Allocate RX ring */
1161 	for (i = 0; i < NUM_RX_BUFF; ++i)
1162 		if (emac_alloc_rx_skb(dev, i, GFP_KERNEL)) {
1163 			printk(KERN_ERR "%s: failed to allocate RX ring\n",
1164 			       ndev->name);
1165 			goto oom;
1166 		}
1167 
1168 	dev->tx_cnt = dev->tx_slot = dev->ack_slot = dev->rx_slot = 0;
1169 	clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1170 	dev->rx_sg_skb = NULL;
1171 
1172 	mutex_lock(&dev->link_lock);
1173 	dev->opened = 1;
1174 
1175 	/* Start PHY polling now.
1176 	 */
1177 	if (dev->phy.address >= 0) {
1178 		int link_poll_interval;
1179 		if (dev->phy.def->ops->poll_link(&dev->phy)) {
1180 			dev->phy.def->ops->read_link(&dev->phy);
1181 			emac_rx_clk_default(dev);
1182 			netif_carrier_on(dev->ndev);
1183 			link_poll_interval = PHY_POLL_LINK_ON;
1184 		} else {
1185 			emac_rx_clk_tx(dev);
1186 			netif_carrier_off(dev->ndev);
1187 			link_poll_interval = PHY_POLL_LINK_OFF;
1188 		}
1189 		dev->link_polling = 1;
1190 		wmb();
1191 		schedule_delayed_work(&dev->link_work, link_poll_interval);
1192 		emac_print_link_status(dev);
1193 	} else
1194 		netif_carrier_on(dev->ndev);
1195 
1196 	/* Required for Pause packet support in EMAC */
1197 	dev_mc_add_global(ndev, default_mcast_addr);
1198 
1199 	emac_configure(dev);
1200 	mal_poll_add(dev->mal, &dev->commac);
1201 	mal_enable_tx_channel(dev->mal, dev->mal_tx_chan);
1202 	mal_set_rcbs(dev->mal, dev->mal_rx_chan, emac_rx_size(ndev->mtu));
1203 	mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1204 	emac_tx_enable(dev);
1205 	emac_rx_enable(dev);
1206 	emac_netif_start(dev);
1207 
1208 	mutex_unlock(&dev->link_lock);
1209 
1210 	return 0;
1211  oom:
1212 	emac_clean_rx_ring(dev);
1213 	free_irq(dev->emac_irq, dev);
1214 
1215 	return -ENOMEM;
1216 }
1217 
1218 /* BHs disabled */
1219 #if 0
1220 static int emac_link_differs(struct emac_instance *dev)
1221 {
1222 	u32 r = in_be32(&dev->emacp->mr1);
1223 
1224 	int duplex = r & EMAC_MR1_FDE ? DUPLEX_FULL : DUPLEX_HALF;
1225 	int speed, pause, asym_pause;
1226 
1227 	if (r & EMAC_MR1_MF_1000)
1228 		speed = SPEED_1000;
1229 	else if (r & EMAC_MR1_MF_100)
1230 		speed = SPEED_100;
1231 	else
1232 		speed = SPEED_10;
1233 
1234 	switch (r & (EMAC_MR1_EIFC | EMAC_MR1_APP)) {
1235 	case (EMAC_MR1_EIFC | EMAC_MR1_APP):
1236 		pause = 1;
1237 		asym_pause = 0;
1238 		break;
1239 	case EMAC_MR1_APP:
1240 		pause = 0;
1241 		asym_pause = 1;
1242 		break;
1243 	default:
1244 		pause = asym_pause = 0;
1245 	}
1246 	return speed != dev->phy.speed || duplex != dev->phy.duplex ||
1247 	    pause != dev->phy.pause || asym_pause != dev->phy.asym_pause;
1248 }
1249 #endif
1250 
1251 static void emac_link_timer(struct work_struct *work)
1252 {
1253 	struct emac_instance *dev =
1254 		container_of(to_delayed_work(work),
1255 			     struct emac_instance, link_work);
1256 	int link_poll_interval;
1257 
1258 	mutex_lock(&dev->link_lock);
1259 	DBG2(dev, "link timer" NL);
1260 
1261 	if (!dev->opened)
1262 		goto bail;
1263 
1264 	if (dev->phy.def->ops->poll_link(&dev->phy)) {
1265 		if (!netif_carrier_ok(dev->ndev)) {
1266 			emac_rx_clk_default(dev);
1267 			/* Get new link parameters */
1268 			dev->phy.def->ops->read_link(&dev->phy);
1269 
1270 			netif_carrier_on(dev->ndev);
1271 			emac_netif_stop(dev);
1272 			emac_full_tx_reset(dev);
1273 			emac_netif_start(dev);
1274 			emac_print_link_status(dev);
1275 		}
1276 		link_poll_interval = PHY_POLL_LINK_ON;
1277 	} else {
1278 		if (netif_carrier_ok(dev->ndev)) {
1279 			emac_rx_clk_tx(dev);
1280 			netif_carrier_off(dev->ndev);
1281 			netif_tx_disable(dev->ndev);
1282 			emac_reinitialize(dev);
1283 			emac_print_link_status(dev);
1284 		}
1285 		link_poll_interval = PHY_POLL_LINK_OFF;
1286 	}
1287 	schedule_delayed_work(&dev->link_work, link_poll_interval);
1288  bail:
1289 	mutex_unlock(&dev->link_lock);
1290 }
1291 
1292 static void emac_force_link_update(struct emac_instance *dev)
1293 {
1294 	netif_carrier_off(dev->ndev);
1295 	smp_rmb();
1296 	if (dev->link_polling) {
1297 		cancel_delayed_work_sync(&dev->link_work);
1298 		if (dev->link_polling)
1299 			schedule_delayed_work(&dev->link_work,  PHY_POLL_LINK_OFF);
1300 	}
1301 }
1302 
1303 /* Process ctx, rtnl_lock semaphore */
1304 static int emac_close(struct net_device *ndev)
1305 {
1306 	struct emac_instance *dev = netdev_priv(ndev);
1307 
1308 	DBG(dev, "close" NL);
1309 
1310 	if (dev->phy.address >= 0) {
1311 		dev->link_polling = 0;
1312 		cancel_delayed_work_sync(&dev->link_work);
1313 	}
1314 	mutex_lock(&dev->link_lock);
1315 	emac_netif_stop(dev);
1316 	dev->opened = 0;
1317 	mutex_unlock(&dev->link_lock);
1318 
1319 	emac_rx_disable(dev);
1320 	emac_tx_disable(dev);
1321 	mal_disable_rx_channel(dev->mal, dev->mal_rx_chan);
1322 	mal_disable_tx_channel(dev->mal, dev->mal_tx_chan);
1323 	mal_poll_del(dev->mal, &dev->commac);
1324 
1325 	emac_clean_tx_ring(dev);
1326 	emac_clean_rx_ring(dev);
1327 
1328 	free_irq(dev->emac_irq, dev);
1329 
1330 	netif_carrier_off(ndev);
1331 
1332 	return 0;
1333 }
1334 
1335 static inline u16 emac_tx_csum(struct emac_instance *dev,
1336 			       struct sk_buff *skb)
1337 {
1338 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH) &&
1339 		(skb->ip_summed == CHECKSUM_PARTIAL)) {
1340 		++dev->stats.tx_packets_csum;
1341 		return EMAC_TX_CTRL_TAH_CSUM;
1342 	}
1343 	return 0;
1344 }
1345 
1346 static inline int emac_xmit_finish(struct emac_instance *dev, int len)
1347 {
1348 	struct emac_regs __iomem *p = dev->emacp;
1349 	struct net_device *ndev = dev->ndev;
1350 
1351 	/* Send the packet out. If the if makes a significant perf
1352 	 * difference, then we can store the TMR0 value in "dev"
1353 	 * instead
1354 	 */
1355 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
1356 		out_be32(&p->tmr0, EMAC4_TMR0_XMIT);
1357 	else
1358 		out_be32(&p->tmr0, EMAC_TMR0_XMIT);
1359 
1360 	if (unlikely(++dev->tx_cnt == NUM_TX_BUFF)) {
1361 		netif_stop_queue(ndev);
1362 		DBG2(dev, "stopped TX queue" NL);
1363 	}
1364 
1365 	ndev->trans_start = jiffies;
1366 	++dev->stats.tx_packets;
1367 	dev->stats.tx_bytes += len;
1368 
1369 	return NETDEV_TX_OK;
1370 }
1371 
1372 /* Tx lock BH */
1373 static int emac_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1374 {
1375 	struct emac_instance *dev = netdev_priv(ndev);
1376 	unsigned int len = skb->len;
1377 	int slot;
1378 
1379 	u16 ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1380 	    MAL_TX_CTRL_LAST | emac_tx_csum(dev, skb);
1381 
1382 	slot = dev->tx_slot++;
1383 	if (dev->tx_slot == NUM_TX_BUFF) {
1384 		dev->tx_slot = 0;
1385 		ctrl |= MAL_TX_CTRL_WRAP;
1386 	}
1387 
1388 	DBG2(dev, "xmit(%u) %d" NL, len, slot);
1389 
1390 	dev->tx_skb[slot] = skb;
1391 	dev->tx_desc[slot].data_ptr = dma_map_single(&dev->ofdev->dev,
1392 						     skb->data, len,
1393 						     DMA_TO_DEVICE);
1394 	dev->tx_desc[slot].data_len = (u16) len;
1395 	wmb();
1396 	dev->tx_desc[slot].ctrl = ctrl;
1397 
1398 	return emac_xmit_finish(dev, len);
1399 }
1400 
1401 static inline int emac_xmit_split(struct emac_instance *dev, int slot,
1402 				  u32 pd, int len, int last, u16 base_ctrl)
1403 {
1404 	while (1) {
1405 		u16 ctrl = base_ctrl;
1406 		int chunk = min(len, MAL_MAX_TX_SIZE);
1407 		len -= chunk;
1408 
1409 		slot = (slot + 1) % NUM_TX_BUFF;
1410 
1411 		if (last && !len)
1412 			ctrl |= MAL_TX_CTRL_LAST;
1413 		if (slot == NUM_TX_BUFF - 1)
1414 			ctrl |= MAL_TX_CTRL_WRAP;
1415 
1416 		dev->tx_skb[slot] = NULL;
1417 		dev->tx_desc[slot].data_ptr = pd;
1418 		dev->tx_desc[slot].data_len = (u16) chunk;
1419 		dev->tx_desc[slot].ctrl = ctrl;
1420 		++dev->tx_cnt;
1421 
1422 		if (!len)
1423 			break;
1424 
1425 		pd += chunk;
1426 	}
1427 	return slot;
1428 }
1429 
1430 /* Tx lock BH disabled (SG version for TAH equipped EMACs) */
1431 static int emac_start_xmit_sg(struct sk_buff *skb, struct net_device *ndev)
1432 {
1433 	struct emac_instance *dev = netdev_priv(ndev);
1434 	int nr_frags = skb_shinfo(skb)->nr_frags;
1435 	int len = skb->len, chunk;
1436 	int slot, i;
1437 	u16 ctrl;
1438 	u32 pd;
1439 
1440 	/* This is common "fast" path */
1441 	if (likely(!nr_frags && len <= MAL_MAX_TX_SIZE))
1442 		return emac_start_xmit(skb, ndev);
1443 
1444 	len -= skb->data_len;
1445 
1446 	/* Note, this is only an *estimation*, we can still run out of empty
1447 	 * slots because of the additional fragmentation into
1448 	 * MAL_MAX_TX_SIZE-sized chunks
1449 	 */
1450 	if (unlikely(dev->tx_cnt + nr_frags + mal_tx_chunks(len) > NUM_TX_BUFF))
1451 		goto stop_queue;
1452 
1453 	ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1454 	    emac_tx_csum(dev, skb);
1455 	slot = dev->tx_slot;
1456 
1457 	/* skb data */
1458 	dev->tx_skb[slot] = NULL;
1459 	chunk = min(len, MAL_MAX_TX_SIZE);
1460 	dev->tx_desc[slot].data_ptr = pd =
1461 	    dma_map_single(&dev->ofdev->dev, skb->data, len, DMA_TO_DEVICE);
1462 	dev->tx_desc[slot].data_len = (u16) chunk;
1463 	len -= chunk;
1464 	if (unlikely(len))
1465 		slot = emac_xmit_split(dev, slot, pd + chunk, len, !nr_frags,
1466 				       ctrl);
1467 	/* skb fragments */
1468 	for (i = 0; i < nr_frags; ++i) {
1469 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1470 		len = skb_frag_size(frag);
1471 
1472 		if (unlikely(dev->tx_cnt + mal_tx_chunks(len) >= NUM_TX_BUFF))
1473 			goto undo_frame;
1474 
1475 		pd = skb_frag_dma_map(&dev->ofdev->dev, frag, 0, len,
1476 				      DMA_TO_DEVICE);
1477 
1478 		slot = emac_xmit_split(dev, slot, pd, len, i == nr_frags - 1,
1479 				       ctrl);
1480 	}
1481 
1482 	DBG2(dev, "xmit_sg(%u) %d - %d" NL, skb->len, dev->tx_slot, slot);
1483 
1484 	/* Attach skb to the last slot so we don't release it too early */
1485 	dev->tx_skb[slot] = skb;
1486 
1487 	/* Send the packet out */
1488 	if (dev->tx_slot == NUM_TX_BUFF - 1)
1489 		ctrl |= MAL_TX_CTRL_WRAP;
1490 	wmb();
1491 	dev->tx_desc[dev->tx_slot].ctrl = ctrl;
1492 	dev->tx_slot = (slot + 1) % NUM_TX_BUFF;
1493 
1494 	return emac_xmit_finish(dev, skb->len);
1495 
1496  undo_frame:
1497 	/* Well, too bad. Our previous estimation was overly optimistic.
1498 	 * Undo everything.
1499 	 */
1500 	while (slot != dev->tx_slot) {
1501 		dev->tx_desc[slot].ctrl = 0;
1502 		--dev->tx_cnt;
1503 		if (--slot < 0)
1504 			slot = NUM_TX_BUFF - 1;
1505 	}
1506 	++dev->estats.tx_undo;
1507 
1508  stop_queue:
1509 	netif_stop_queue(ndev);
1510 	DBG2(dev, "stopped TX queue" NL);
1511 	return NETDEV_TX_BUSY;
1512 }
1513 
1514 /* Tx lock BHs */
1515 static void emac_parse_tx_error(struct emac_instance *dev, u16 ctrl)
1516 {
1517 	struct emac_error_stats *st = &dev->estats;
1518 
1519 	DBG(dev, "BD TX error %04x" NL, ctrl);
1520 
1521 	++st->tx_bd_errors;
1522 	if (ctrl & EMAC_TX_ST_BFCS)
1523 		++st->tx_bd_bad_fcs;
1524 	if (ctrl & EMAC_TX_ST_LCS)
1525 		++st->tx_bd_carrier_loss;
1526 	if (ctrl & EMAC_TX_ST_ED)
1527 		++st->tx_bd_excessive_deferral;
1528 	if (ctrl & EMAC_TX_ST_EC)
1529 		++st->tx_bd_excessive_collisions;
1530 	if (ctrl & EMAC_TX_ST_LC)
1531 		++st->tx_bd_late_collision;
1532 	if (ctrl & EMAC_TX_ST_MC)
1533 		++st->tx_bd_multple_collisions;
1534 	if (ctrl & EMAC_TX_ST_SC)
1535 		++st->tx_bd_single_collision;
1536 	if (ctrl & EMAC_TX_ST_UR)
1537 		++st->tx_bd_underrun;
1538 	if (ctrl & EMAC_TX_ST_SQE)
1539 		++st->tx_bd_sqe;
1540 }
1541 
1542 static void emac_poll_tx(void *param)
1543 {
1544 	struct emac_instance *dev = param;
1545 	u32 bad_mask;
1546 
1547 	DBG2(dev, "poll_tx, %d %d" NL, dev->tx_cnt, dev->ack_slot);
1548 
1549 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
1550 		bad_mask = EMAC_IS_BAD_TX_TAH;
1551 	else
1552 		bad_mask = EMAC_IS_BAD_TX;
1553 
1554 	netif_tx_lock_bh(dev->ndev);
1555 	if (dev->tx_cnt) {
1556 		u16 ctrl;
1557 		int slot = dev->ack_slot, n = 0;
1558 	again:
1559 		ctrl = dev->tx_desc[slot].ctrl;
1560 		if (!(ctrl & MAL_TX_CTRL_READY)) {
1561 			struct sk_buff *skb = dev->tx_skb[slot];
1562 			++n;
1563 
1564 			if (skb) {
1565 				dev_kfree_skb(skb);
1566 				dev->tx_skb[slot] = NULL;
1567 			}
1568 			slot = (slot + 1) % NUM_TX_BUFF;
1569 
1570 			if (unlikely(ctrl & bad_mask))
1571 				emac_parse_tx_error(dev, ctrl);
1572 
1573 			if (--dev->tx_cnt)
1574 				goto again;
1575 		}
1576 		if (n) {
1577 			dev->ack_slot = slot;
1578 			if (netif_queue_stopped(dev->ndev) &&
1579 			    dev->tx_cnt < EMAC_TX_WAKEUP_THRESH)
1580 				netif_wake_queue(dev->ndev);
1581 
1582 			DBG2(dev, "tx %d pkts" NL, n);
1583 		}
1584 	}
1585 	netif_tx_unlock_bh(dev->ndev);
1586 }
1587 
1588 static inline void emac_recycle_rx_skb(struct emac_instance *dev, int slot,
1589 				       int len)
1590 {
1591 	struct sk_buff *skb = dev->rx_skb[slot];
1592 
1593 	DBG2(dev, "recycle %d %d" NL, slot, len);
1594 
1595 	if (len)
1596 		dma_map_single(&dev->ofdev->dev, skb->data - 2,
1597 			       EMAC_DMA_ALIGN(len + 2), DMA_FROM_DEVICE);
1598 
1599 	dev->rx_desc[slot].data_len = 0;
1600 	wmb();
1601 	dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1602 	    (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1603 }
1604 
1605 static void emac_parse_rx_error(struct emac_instance *dev, u16 ctrl)
1606 {
1607 	struct emac_error_stats *st = &dev->estats;
1608 
1609 	DBG(dev, "BD RX error %04x" NL, ctrl);
1610 
1611 	++st->rx_bd_errors;
1612 	if (ctrl & EMAC_RX_ST_OE)
1613 		++st->rx_bd_overrun;
1614 	if (ctrl & EMAC_RX_ST_BP)
1615 		++st->rx_bd_bad_packet;
1616 	if (ctrl & EMAC_RX_ST_RP)
1617 		++st->rx_bd_runt_packet;
1618 	if (ctrl & EMAC_RX_ST_SE)
1619 		++st->rx_bd_short_event;
1620 	if (ctrl & EMAC_RX_ST_AE)
1621 		++st->rx_bd_alignment_error;
1622 	if (ctrl & EMAC_RX_ST_BFCS)
1623 		++st->rx_bd_bad_fcs;
1624 	if (ctrl & EMAC_RX_ST_PTL)
1625 		++st->rx_bd_packet_too_long;
1626 	if (ctrl & EMAC_RX_ST_ORE)
1627 		++st->rx_bd_out_of_range;
1628 	if (ctrl & EMAC_RX_ST_IRE)
1629 		++st->rx_bd_in_range;
1630 }
1631 
1632 static inline void emac_rx_csum(struct emac_instance *dev,
1633 				struct sk_buff *skb, u16 ctrl)
1634 {
1635 #ifdef CONFIG_IBM_EMAC_TAH
1636 	if (!ctrl && dev->tah_dev) {
1637 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1638 		++dev->stats.rx_packets_csum;
1639 	}
1640 #endif
1641 }
1642 
1643 static inline int emac_rx_sg_append(struct emac_instance *dev, int slot)
1644 {
1645 	if (likely(dev->rx_sg_skb != NULL)) {
1646 		int len = dev->rx_desc[slot].data_len;
1647 		int tot_len = dev->rx_sg_skb->len + len;
1648 
1649 		if (unlikely(tot_len + 2 > dev->rx_skb_size)) {
1650 			++dev->estats.rx_dropped_mtu;
1651 			dev_kfree_skb(dev->rx_sg_skb);
1652 			dev->rx_sg_skb = NULL;
1653 		} else {
1654 			cacheable_memcpy(skb_tail_pointer(dev->rx_sg_skb),
1655 					 dev->rx_skb[slot]->data, len);
1656 			skb_put(dev->rx_sg_skb, len);
1657 			emac_recycle_rx_skb(dev, slot, len);
1658 			return 0;
1659 		}
1660 	}
1661 	emac_recycle_rx_skb(dev, slot, 0);
1662 	return -1;
1663 }
1664 
1665 /* NAPI poll context */
1666 static int emac_poll_rx(void *param, int budget)
1667 {
1668 	struct emac_instance *dev = param;
1669 	int slot = dev->rx_slot, received = 0;
1670 
1671 	DBG2(dev, "poll_rx(%d)" NL, budget);
1672 
1673  again:
1674 	while (budget > 0) {
1675 		int len;
1676 		struct sk_buff *skb;
1677 		u16 ctrl = dev->rx_desc[slot].ctrl;
1678 
1679 		if (ctrl & MAL_RX_CTRL_EMPTY)
1680 			break;
1681 
1682 		skb = dev->rx_skb[slot];
1683 		mb();
1684 		len = dev->rx_desc[slot].data_len;
1685 
1686 		if (unlikely(!MAL_IS_SINGLE_RX(ctrl)))
1687 			goto sg;
1688 
1689 		ctrl &= EMAC_BAD_RX_MASK;
1690 		if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1691 			emac_parse_rx_error(dev, ctrl);
1692 			++dev->estats.rx_dropped_error;
1693 			emac_recycle_rx_skb(dev, slot, 0);
1694 			len = 0;
1695 			goto next;
1696 		}
1697 
1698 		if (len < ETH_HLEN) {
1699 			++dev->estats.rx_dropped_stack;
1700 			emac_recycle_rx_skb(dev, slot, len);
1701 			goto next;
1702 		}
1703 
1704 		if (len && len < EMAC_RX_COPY_THRESH) {
1705 			struct sk_buff *copy_skb =
1706 			    alloc_skb(len + EMAC_RX_SKB_HEADROOM + 2, GFP_ATOMIC);
1707 			if (unlikely(!copy_skb))
1708 				goto oom;
1709 
1710 			skb_reserve(copy_skb, EMAC_RX_SKB_HEADROOM + 2);
1711 			cacheable_memcpy(copy_skb->data - 2, skb->data - 2,
1712 					 len + 2);
1713 			emac_recycle_rx_skb(dev, slot, len);
1714 			skb = copy_skb;
1715 		} else if (unlikely(emac_alloc_rx_skb(dev, slot, GFP_ATOMIC)))
1716 			goto oom;
1717 
1718 		skb_put(skb, len);
1719 	push_packet:
1720 		skb->protocol = eth_type_trans(skb, dev->ndev);
1721 		emac_rx_csum(dev, skb, ctrl);
1722 
1723 		if (unlikely(netif_receive_skb(skb) == NET_RX_DROP))
1724 			++dev->estats.rx_dropped_stack;
1725 	next:
1726 		++dev->stats.rx_packets;
1727 	skip:
1728 		dev->stats.rx_bytes += len;
1729 		slot = (slot + 1) % NUM_RX_BUFF;
1730 		--budget;
1731 		++received;
1732 		continue;
1733 	sg:
1734 		if (ctrl & MAL_RX_CTRL_FIRST) {
1735 			BUG_ON(dev->rx_sg_skb);
1736 			if (unlikely(emac_alloc_rx_skb(dev, slot, GFP_ATOMIC))) {
1737 				DBG(dev, "rx OOM %d" NL, slot);
1738 				++dev->estats.rx_dropped_oom;
1739 				emac_recycle_rx_skb(dev, slot, 0);
1740 			} else {
1741 				dev->rx_sg_skb = skb;
1742 				skb_put(skb, len);
1743 			}
1744 		} else if (!emac_rx_sg_append(dev, slot) &&
1745 			   (ctrl & MAL_RX_CTRL_LAST)) {
1746 
1747 			skb = dev->rx_sg_skb;
1748 			dev->rx_sg_skb = NULL;
1749 
1750 			ctrl &= EMAC_BAD_RX_MASK;
1751 			if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1752 				emac_parse_rx_error(dev, ctrl);
1753 				++dev->estats.rx_dropped_error;
1754 				dev_kfree_skb(skb);
1755 				len = 0;
1756 			} else
1757 				goto push_packet;
1758 		}
1759 		goto skip;
1760 	oom:
1761 		DBG(dev, "rx OOM %d" NL, slot);
1762 		/* Drop the packet and recycle skb */
1763 		++dev->estats.rx_dropped_oom;
1764 		emac_recycle_rx_skb(dev, slot, 0);
1765 		goto next;
1766 	}
1767 
1768 	if (received) {
1769 		DBG2(dev, "rx %d BDs" NL, received);
1770 		dev->rx_slot = slot;
1771 	}
1772 
1773 	if (unlikely(budget && test_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags))) {
1774 		mb();
1775 		if (!(dev->rx_desc[slot].ctrl & MAL_RX_CTRL_EMPTY)) {
1776 			DBG2(dev, "rx restart" NL);
1777 			received = 0;
1778 			goto again;
1779 		}
1780 
1781 		if (dev->rx_sg_skb) {
1782 			DBG2(dev, "dropping partial rx packet" NL);
1783 			++dev->estats.rx_dropped_error;
1784 			dev_kfree_skb(dev->rx_sg_skb);
1785 			dev->rx_sg_skb = NULL;
1786 		}
1787 
1788 		clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1789 		mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1790 		emac_rx_enable(dev);
1791 		dev->rx_slot = 0;
1792 	}
1793 	return received;
1794 }
1795 
1796 /* NAPI poll context */
1797 static int emac_peek_rx(void *param)
1798 {
1799 	struct emac_instance *dev = param;
1800 
1801 	return !(dev->rx_desc[dev->rx_slot].ctrl & MAL_RX_CTRL_EMPTY);
1802 }
1803 
1804 /* NAPI poll context */
1805 static int emac_peek_rx_sg(void *param)
1806 {
1807 	struct emac_instance *dev = param;
1808 
1809 	int slot = dev->rx_slot;
1810 	while (1) {
1811 		u16 ctrl = dev->rx_desc[slot].ctrl;
1812 		if (ctrl & MAL_RX_CTRL_EMPTY)
1813 			return 0;
1814 		else if (ctrl & MAL_RX_CTRL_LAST)
1815 			return 1;
1816 
1817 		slot = (slot + 1) % NUM_RX_BUFF;
1818 
1819 		/* I'm just being paranoid here :) */
1820 		if (unlikely(slot == dev->rx_slot))
1821 			return 0;
1822 	}
1823 }
1824 
1825 /* Hard IRQ */
1826 static void emac_rxde(void *param)
1827 {
1828 	struct emac_instance *dev = param;
1829 
1830 	++dev->estats.rx_stopped;
1831 	emac_rx_disable_async(dev);
1832 }
1833 
1834 /* Hard IRQ */
1835 static irqreturn_t emac_irq(int irq, void *dev_instance)
1836 {
1837 	struct emac_instance *dev = dev_instance;
1838 	struct emac_regs __iomem *p = dev->emacp;
1839 	struct emac_error_stats *st = &dev->estats;
1840 	u32 isr;
1841 
1842 	spin_lock(&dev->lock);
1843 
1844 	isr = in_be32(&p->isr);
1845 	out_be32(&p->isr, isr);
1846 
1847 	DBG(dev, "isr = %08x" NL, isr);
1848 
1849 	if (isr & EMAC4_ISR_TXPE)
1850 		++st->tx_parity;
1851 	if (isr & EMAC4_ISR_RXPE)
1852 		++st->rx_parity;
1853 	if (isr & EMAC4_ISR_TXUE)
1854 		++st->tx_underrun;
1855 	if (isr & EMAC4_ISR_RXOE)
1856 		++st->rx_fifo_overrun;
1857 	if (isr & EMAC_ISR_OVR)
1858 		++st->rx_overrun;
1859 	if (isr & EMAC_ISR_BP)
1860 		++st->rx_bad_packet;
1861 	if (isr & EMAC_ISR_RP)
1862 		++st->rx_runt_packet;
1863 	if (isr & EMAC_ISR_SE)
1864 		++st->rx_short_event;
1865 	if (isr & EMAC_ISR_ALE)
1866 		++st->rx_alignment_error;
1867 	if (isr & EMAC_ISR_BFCS)
1868 		++st->rx_bad_fcs;
1869 	if (isr & EMAC_ISR_PTLE)
1870 		++st->rx_packet_too_long;
1871 	if (isr & EMAC_ISR_ORE)
1872 		++st->rx_out_of_range;
1873 	if (isr & EMAC_ISR_IRE)
1874 		++st->rx_in_range;
1875 	if (isr & EMAC_ISR_SQE)
1876 		++st->tx_sqe;
1877 	if (isr & EMAC_ISR_TE)
1878 		++st->tx_errors;
1879 
1880 	spin_unlock(&dev->lock);
1881 
1882 	return IRQ_HANDLED;
1883 }
1884 
1885 static struct net_device_stats *emac_stats(struct net_device *ndev)
1886 {
1887 	struct emac_instance *dev = netdev_priv(ndev);
1888 	struct emac_stats *st = &dev->stats;
1889 	struct emac_error_stats *est = &dev->estats;
1890 	struct net_device_stats *nst = &dev->nstats;
1891 	unsigned long flags;
1892 
1893 	DBG2(dev, "stats" NL);
1894 
1895 	/* Compute "legacy" statistics */
1896 	spin_lock_irqsave(&dev->lock, flags);
1897 	nst->rx_packets = (unsigned long)st->rx_packets;
1898 	nst->rx_bytes = (unsigned long)st->rx_bytes;
1899 	nst->tx_packets = (unsigned long)st->tx_packets;
1900 	nst->tx_bytes = (unsigned long)st->tx_bytes;
1901 	nst->rx_dropped = (unsigned long)(est->rx_dropped_oom +
1902 					  est->rx_dropped_error +
1903 					  est->rx_dropped_resize +
1904 					  est->rx_dropped_mtu);
1905 	nst->tx_dropped = (unsigned long)est->tx_dropped;
1906 
1907 	nst->rx_errors = (unsigned long)est->rx_bd_errors;
1908 	nst->rx_fifo_errors = (unsigned long)(est->rx_bd_overrun +
1909 					      est->rx_fifo_overrun +
1910 					      est->rx_overrun);
1911 	nst->rx_frame_errors = (unsigned long)(est->rx_bd_alignment_error +
1912 					       est->rx_alignment_error);
1913 	nst->rx_crc_errors = (unsigned long)(est->rx_bd_bad_fcs +
1914 					     est->rx_bad_fcs);
1915 	nst->rx_length_errors = (unsigned long)(est->rx_bd_runt_packet +
1916 						est->rx_bd_short_event +
1917 						est->rx_bd_packet_too_long +
1918 						est->rx_bd_out_of_range +
1919 						est->rx_bd_in_range +
1920 						est->rx_runt_packet +
1921 						est->rx_short_event +
1922 						est->rx_packet_too_long +
1923 						est->rx_out_of_range +
1924 						est->rx_in_range);
1925 
1926 	nst->tx_errors = (unsigned long)(est->tx_bd_errors + est->tx_errors);
1927 	nst->tx_fifo_errors = (unsigned long)(est->tx_bd_underrun +
1928 					      est->tx_underrun);
1929 	nst->tx_carrier_errors = (unsigned long)est->tx_bd_carrier_loss;
1930 	nst->collisions = (unsigned long)(est->tx_bd_excessive_deferral +
1931 					  est->tx_bd_excessive_collisions +
1932 					  est->tx_bd_late_collision +
1933 					  est->tx_bd_multple_collisions);
1934 	spin_unlock_irqrestore(&dev->lock, flags);
1935 	return nst;
1936 }
1937 
1938 static struct mal_commac_ops emac_commac_ops = {
1939 	.poll_tx = &emac_poll_tx,
1940 	.poll_rx = &emac_poll_rx,
1941 	.peek_rx = &emac_peek_rx,
1942 	.rxde = &emac_rxde,
1943 };
1944 
1945 static struct mal_commac_ops emac_commac_sg_ops = {
1946 	.poll_tx = &emac_poll_tx,
1947 	.poll_rx = &emac_poll_rx,
1948 	.peek_rx = &emac_peek_rx_sg,
1949 	.rxde = &emac_rxde,
1950 };
1951 
1952 /* Ethtool support */
1953 static int emac_ethtool_get_settings(struct net_device *ndev,
1954 				     struct ethtool_cmd *cmd)
1955 {
1956 	struct emac_instance *dev = netdev_priv(ndev);
1957 
1958 	cmd->supported = dev->phy.features;
1959 	cmd->port = PORT_MII;
1960 	cmd->phy_address = dev->phy.address;
1961 	cmd->transceiver =
1962 	    dev->phy.address >= 0 ? XCVR_EXTERNAL : XCVR_INTERNAL;
1963 
1964 	mutex_lock(&dev->link_lock);
1965 	cmd->advertising = dev->phy.advertising;
1966 	cmd->autoneg = dev->phy.autoneg;
1967 	cmd->speed = dev->phy.speed;
1968 	cmd->duplex = dev->phy.duplex;
1969 	mutex_unlock(&dev->link_lock);
1970 
1971 	return 0;
1972 }
1973 
1974 static int emac_ethtool_set_settings(struct net_device *ndev,
1975 				     struct ethtool_cmd *cmd)
1976 {
1977 	struct emac_instance *dev = netdev_priv(ndev);
1978 	u32 f = dev->phy.features;
1979 
1980 	DBG(dev, "set_settings(%d, %d, %d, 0x%08x)" NL,
1981 	    cmd->autoneg, cmd->speed, cmd->duplex, cmd->advertising);
1982 
1983 	/* Basic sanity checks */
1984 	if (dev->phy.address < 0)
1985 		return -EOPNOTSUPP;
1986 	if (cmd->autoneg != AUTONEG_ENABLE && cmd->autoneg != AUTONEG_DISABLE)
1987 		return -EINVAL;
1988 	if (cmd->autoneg == AUTONEG_ENABLE && cmd->advertising == 0)
1989 		return -EINVAL;
1990 	if (cmd->duplex != DUPLEX_HALF && cmd->duplex != DUPLEX_FULL)
1991 		return -EINVAL;
1992 
1993 	if (cmd->autoneg == AUTONEG_DISABLE) {
1994 		switch (cmd->speed) {
1995 		case SPEED_10:
1996 			if (cmd->duplex == DUPLEX_HALF &&
1997 			    !(f & SUPPORTED_10baseT_Half))
1998 				return -EINVAL;
1999 			if (cmd->duplex == DUPLEX_FULL &&
2000 			    !(f & SUPPORTED_10baseT_Full))
2001 				return -EINVAL;
2002 			break;
2003 		case SPEED_100:
2004 			if (cmd->duplex == DUPLEX_HALF &&
2005 			    !(f & SUPPORTED_100baseT_Half))
2006 				return -EINVAL;
2007 			if (cmd->duplex == DUPLEX_FULL &&
2008 			    !(f & SUPPORTED_100baseT_Full))
2009 				return -EINVAL;
2010 			break;
2011 		case SPEED_1000:
2012 			if (cmd->duplex == DUPLEX_HALF &&
2013 			    !(f & SUPPORTED_1000baseT_Half))
2014 				return -EINVAL;
2015 			if (cmd->duplex == DUPLEX_FULL &&
2016 			    !(f & SUPPORTED_1000baseT_Full))
2017 				return -EINVAL;
2018 			break;
2019 		default:
2020 			return -EINVAL;
2021 		}
2022 
2023 		mutex_lock(&dev->link_lock);
2024 		dev->phy.def->ops->setup_forced(&dev->phy, cmd->speed,
2025 						cmd->duplex);
2026 		mutex_unlock(&dev->link_lock);
2027 
2028 	} else {
2029 		if (!(f & SUPPORTED_Autoneg))
2030 			return -EINVAL;
2031 
2032 		mutex_lock(&dev->link_lock);
2033 		dev->phy.def->ops->setup_aneg(&dev->phy,
2034 					      (cmd->advertising & f) |
2035 					      (dev->phy.advertising &
2036 					       (ADVERTISED_Pause |
2037 						ADVERTISED_Asym_Pause)));
2038 		mutex_unlock(&dev->link_lock);
2039 	}
2040 	emac_force_link_update(dev);
2041 
2042 	return 0;
2043 }
2044 
2045 static void emac_ethtool_get_ringparam(struct net_device *ndev,
2046 				       struct ethtool_ringparam *rp)
2047 {
2048 	rp->rx_max_pending = rp->rx_pending = NUM_RX_BUFF;
2049 	rp->tx_max_pending = rp->tx_pending = NUM_TX_BUFF;
2050 }
2051 
2052 static void emac_ethtool_get_pauseparam(struct net_device *ndev,
2053 					struct ethtool_pauseparam *pp)
2054 {
2055 	struct emac_instance *dev = netdev_priv(ndev);
2056 
2057 	mutex_lock(&dev->link_lock);
2058 	if ((dev->phy.features & SUPPORTED_Autoneg) &&
2059 	    (dev->phy.advertising & (ADVERTISED_Pause | ADVERTISED_Asym_Pause)))
2060 		pp->autoneg = 1;
2061 
2062 	if (dev->phy.duplex == DUPLEX_FULL) {
2063 		if (dev->phy.pause)
2064 			pp->rx_pause = pp->tx_pause = 1;
2065 		else if (dev->phy.asym_pause)
2066 			pp->tx_pause = 1;
2067 	}
2068 	mutex_unlock(&dev->link_lock);
2069 }
2070 
2071 static int emac_get_regs_len(struct emac_instance *dev)
2072 {
2073 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
2074 		return sizeof(struct emac_ethtool_regs_subhdr) +
2075 			EMAC4_ETHTOOL_REGS_SIZE(dev);
2076 	else
2077 		return sizeof(struct emac_ethtool_regs_subhdr) +
2078 			EMAC_ETHTOOL_REGS_SIZE(dev);
2079 }
2080 
2081 static int emac_ethtool_get_regs_len(struct net_device *ndev)
2082 {
2083 	struct emac_instance *dev = netdev_priv(ndev);
2084 	int size;
2085 
2086 	size = sizeof(struct emac_ethtool_regs_hdr) +
2087 		emac_get_regs_len(dev) + mal_get_regs_len(dev->mal);
2088 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
2089 		size += zmii_get_regs_len(dev->zmii_dev);
2090 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
2091 		size += rgmii_get_regs_len(dev->rgmii_dev);
2092 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
2093 		size += tah_get_regs_len(dev->tah_dev);
2094 
2095 	return size;
2096 }
2097 
2098 static void *emac_dump_regs(struct emac_instance *dev, void *buf)
2099 {
2100 	struct emac_ethtool_regs_subhdr *hdr = buf;
2101 
2102 	hdr->index = dev->cell_index;
2103 	if (emac_has_feature(dev, EMAC_FTR_EMAC4)) {
2104 		hdr->version = EMAC4_ETHTOOL_REGS_VER;
2105 		memcpy_fromio(hdr + 1, dev->emacp, EMAC4_ETHTOOL_REGS_SIZE(dev));
2106 		return (void *)(hdr + 1) + EMAC4_ETHTOOL_REGS_SIZE(dev);
2107 	} else {
2108 		hdr->version = EMAC_ETHTOOL_REGS_VER;
2109 		memcpy_fromio(hdr + 1, dev->emacp, EMAC_ETHTOOL_REGS_SIZE(dev));
2110 		return (void *)(hdr + 1) + EMAC_ETHTOOL_REGS_SIZE(dev);
2111 	}
2112 }
2113 
2114 static void emac_ethtool_get_regs(struct net_device *ndev,
2115 				  struct ethtool_regs *regs, void *buf)
2116 {
2117 	struct emac_instance *dev = netdev_priv(ndev);
2118 	struct emac_ethtool_regs_hdr *hdr = buf;
2119 
2120 	hdr->components = 0;
2121 	buf = hdr + 1;
2122 
2123 	buf = mal_dump_regs(dev->mal, buf);
2124 	buf = emac_dump_regs(dev, buf);
2125 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII)) {
2126 		hdr->components |= EMAC_ETHTOOL_REGS_ZMII;
2127 		buf = zmii_dump_regs(dev->zmii_dev, buf);
2128 	}
2129 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII)) {
2130 		hdr->components |= EMAC_ETHTOOL_REGS_RGMII;
2131 		buf = rgmii_dump_regs(dev->rgmii_dev, buf);
2132 	}
2133 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH)) {
2134 		hdr->components |= EMAC_ETHTOOL_REGS_TAH;
2135 		buf = tah_dump_regs(dev->tah_dev, buf);
2136 	}
2137 }
2138 
2139 static int emac_ethtool_nway_reset(struct net_device *ndev)
2140 {
2141 	struct emac_instance *dev = netdev_priv(ndev);
2142 	int res = 0;
2143 
2144 	DBG(dev, "nway_reset" NL);
2145 
2146 	if (dev->phy.address < 0)
2147 		return -EOPNOTSUPP;
2148 
2149 	mutex_lock(&dev->link_lock);
2150 	if (!dev->phy.autoneg) {
2151 		res = -EINVAL;
2152 		goto out;
2153 	}
2154 
2155 	dev->phy.def->ops->setup_aneg(&dev->phy, dev->phy.advertising);
2156  out:
2157 	mutex_unlock(&dev->link_lock);
2158 	emac_force_link_update(dev);
2159 	return res;
2160 }
2161 
2162 static int emac_ethtool_get_sset_count(struct net_device *ndev, int stringset)
2163 {
2164 	if (stringset == ETH_SS_STATS)
2165 		return EMAC_ETHTOOL_STATS_COUNT;
2166 	else
2167 		return -EINVAL;
2168 }
2169 
2170 static void emac_ethtool_get_strings(struct net_device *ndev, u32 stringset,
2171 				     u8 * buf)
2172 {
2173 	if (stringset == ETH_SS_STATS)
2174 		memcpy(buf, &emac_stats_keys, sizeof(emac_stats_keys));
2175 }
2176 
2177 static void emac_ethtool_get_ethtool_stats(struct net_device *ndev,
2178 					   struct ethtool_stats *estats,
2179 					   u64 * tmp_stats)
2180 {
2181 	struct emac_instance *dev = netdev_priv(ndev);
2182 
2183 	memcpy(tmp_stats, &dev->stats, sizeof(dev->stats));
2184 	tmp_stats += sizeof(dev->stats) / sizeof(u64);
2185 	memcpy(tmp_stats, &dev->estats, sizeof(dev->estats));
2186 }
2187 
2188 static void emac_ethtool_get_drvinfo(struct net_device *ndev,
2189 				     struct ethtool_drvinfo *info)
2190 {
2191 	struct emac_instance *dev = netdev_priv(ndev);
2192 
2193 	strcpy(info->driver, "ibm_emac");
2194 	strcpy(info->version, DRV_VERSION);
2195 	info->fw_version[0] = '\0';
2196 	sprintf(info->bus_info, "PPC 4xx EMAC-%d %s",
2197 		dev->cell_index, dev->ofdev->dev.of_node->full_name);
2198 	info->regdump_len = emac_ethtool_get_regs_len(ndev);
2199 }
2200 
2201 static const struct ethtool_ops emac_ethtool_ops = {
2202 	.get_settings = emac_ethtool_get_settings,
2203 	.set_settings = emac_ethtool_set_settings,
2204 	.get_drvinfo = emac_ethtool_get_drvinfo,
2205 
2206 	.get_regs_len = emac_ethtool_get_regs_len,
2207 	.get_regs = emac_ethtool_get_regs,
2208 
2209 	.nway_reset = emac_ethtool_nway_reset,
2210 
2211 	.get_ringparam = emac_ethtool_get_ringparam,
2212 	.get_pauseparam = emac_ethtool_get_pauseparam,
2213 
2214 	.get_strings = emac_ethtool_get_strings,
2215 	.get_sset_count = emac_ethtool_get_sset_count,
2216 	.get_ethtool_stats = emac_ethtool_get_ethtool_stats,
2217 
2218 	.get_link = ethtool_op_get_link,
2219 };
2220 
2221 static int emac_ioctl(struct net_device *ndev, struct ifreq *rq, int cmd)
2222 {
2223 	struct emac_instance *dev = netdev_priv(ndev);
2224 	struct mii_ioctl_data *data = if_mii(rq);
2225 
2226 	DBG(dev, "ioctl %08x" NL, cmd);
2227 
2228 	if (dev->phy.address < 0)
2229 		return -EOPNOTSUPP;
2230 
2231 	switch (cmd) {
2232 	case SIOCGMIIPHY:
2233 		data->phy_id = dev->phy.address;
2234 		/* Fall through */
2235 	case SIOCGMIIREG:
2236 		data->val_out = emac_mdio_read(ndev, dev->phy.address,
2237 					       data->reg_num);
2238 		return 0;
2239 
2240 	case SIOCSMIIREG:
2241 		emac_mdio_write(ndev, dev->phy.address, data->reg_num,
2242 				data->val_in);
2243 		return 0;
2244 	default:
2245 		return -EOPNOTSUPP;
2246 	}
2247 }
2248 
2249 struct emac_depentry {
2250 	u32			phandle;
2251 	struct device_node	*node;
2252 	struct platform_device	*ofdev;
2253 	void			*drvdata;
2254 };
2255 
2256 #define	EMAC_DEP_MAL_IDX	0
2257 #define	EMAC_DEP_ZMII_IDX	1
2258 #define	EMAC_DEP_RGMII_IDX	2
2259 #define	EMAC_DEP_TAH_IDX	3
2260 #define	EMAC_DEP_MDIO_IDX	4
2261 #define	EMAC_DEP_PREV_IDX	5
2262 #define	EMAC_DEP_COUNT		6
2263 
2264 static int __devinit emac_check_deps(struct emac_instance *dev,
2265 				     struct emac_depentry *deps)
2266 {
2267 	int i, there = 0;
2268 	struct device_node *np;
2269 
2270 	for (i = 0; i < EMAC_DEP_COUNT; i++) {
2271 		/* no dependency on that item, allright */
2272 		if (deps[i].phandle == 0) {
2273 			there++;
2274 			continue;
2275 		}
2276 		/* special case for blist as the dependency might go away */
2277 		if (i == EMAC_DEP_PREV_IDX) {
2278 			np = *(dev->blist - 1);
2279 			if (np == NULL) {
2280 				deps[i].phandle = 0;
2281 				there++;
2282 				continue;
2283 			}
2284 			if (deps[i].node == NULL)
2285 				deps[i].node = of_node_get(np);
2286 		}
2287 		if (deps[i].node == NULL)
2288 			deps[i].node = of_find_node_by_phandle(deps[i].phandle);
2289 		if (deps[i].node == NULL)
2290 			continue;
2291 		if (deps[i].ofdev == NULL)
2292 			deps[i].ofdev = of_find_device_by_node(deps[i].node);
2293 		if (deps[i].ofdev == NULL)
2294 			continue;
2295 		if (deps[i].drvdata == NULL)
2296 			deps[i].drvdata = dev_get_drvdata(&deps[i].ofdev->dev);
2297 		if (deps[i].drvdata != NULL)
2298 			there++;
2299 	}
2300 	return there == EMAC_DEP_COUNT;
2301 }
2302 
2303 static void emac_put_deps(struct emac_instance *dev)
2304 {
2305 	if (dev->mal_dev)
2306 		of_dev_put(dev->mal_dev);
2307 	if (dev->zmii_dev)
2308 		of_dev_put(dev->zmii_dev);
2309 	if (dev->rgmii_dev)
2310 		of_dev_put(dev->rgmii_dev);
2311 	if (dev->mdio_dev)
2312 		of_dev_put(dev->mdio_dev);
2313 	if (dev->tah_dev)
2314 		of_dev_put(dev->tah_dev);
2315 }
2316 
2317 static int __devinit emac_of_bus_notify(struct notifier_block *nb,
2318 					unsigned long action, void *data)
2319 {
2320 	/* We are only intereted in device addition */
2321 	if (action == BUS_NOTIFY_BOUND_DRIVER)
2322 		wake_up_all(&emac_probe_wait);
2323 	return 0;
2324 }
2325 
2326 static struct notifier_block emac_of_bus_notifier __devinitdata = {
2327 	.notifier_call = emac_of_bus_notify
2328 };
2329 
2330 static int __devinit emac_wait_deps(struct emac_instance *dev)
2331 {
2332 	struct emac_depentry deps[EMAC_DEP_COUNT];
2333 	int i, err;
2334 
2335 	memset(&deps, 0, sizeof(deps));
2336 
2337 	deps[EMAC_DEP_MAL_IDX].phandle = dev->mal_ph;
2338 	deps[EMAC_DEP_ZMII_IDX].phandle = dev->zmii_ph;
2339 	deps[EMAC_DEP_RGMII_IDX].phandle = dev->rgmii_ph;
2340 	if (dev->tah_ph)
2341 		deps[EMAC_DEP_TAH_IDX].phandle = dev->tah_ph;
2342 	if (dev->mdio_ph)
2343 		deps[EMAC_DEP_MDIO_IDX].phandle = dev->mdio_ph;
2344 	if (dev->blist && dev->blist > emac_boot_list)
2345 		deps[EMAC_DEP_PREV_IDX].phandle = 0xffffffffu;
2346 	bus_register_notifier(&platform_bus_type, &emac_of_bus_notifier);
2347 	wait_event_timeout(emac_probe_wait,
2348 			   emac_check_deps(dev, deps),
2349 			   EMAC_PROBE_DEP_TIMEOUT);
2350 	bus_unregister_notifier(&platform_bus_type, &emac_of_bus_notifier);
2351 	err = emac_check_deps(dev, deps) ? 0 : -ENODEV;
2352 	for (i = 0; i < EMAC_DEP_COUNT; i++) {
2353 		if (deps[i].node)
2354 			of_node_put(deps[i].node);
2355 		if (err && deps[i].ofdev)
2356 			of_dev_put(deps[i].ofdev);
2357 	}
2358 	if (err == 0) {
2359 		dev->mal_dev = deps[EMAC_DEP_MAL_IDX].ofdev;
2360 		dev->zmii_dev = deps[EMAC_DEP_ZMII_IDX].ofdev;
2361 		dev->rgmii_dev = deps[EMAC_DEP_RGMII_IDX].ofdev;
2362 		dev->tah_dev = deps[EMAC_DEP_TAH_IDX].ofdev;
2363 		dev->mdio_dev = deps[EMAC_DEP_MDIO_IDX].ofdev;
2364 	}
2365 	if (deps[EMAC_DEP_PREV_IDX].ofdev)
2366 		of_dev_put(deps[EMAC_DEP_PREV_IDX].ofdev);
2367 	return err;
2368 }
2369 
2370 static int __devinit emac_read_uint_prop(struct device_node *np, const char *name,
2371 					 u32 *val, int fatal)
2372 {
2373 	int len;
2374 	const u32 *prop = of_get_property(np, name, &len);
2375 	if (prop == NULL || len < sizeof(u32)) {
2376 		if (fatal)
2377 			printk(KERN_ERR "%s: missing %s property\n",
2378 			       np->full_name, name);
2379 		return -ENODEV;
2380 	}
2381 	*val = *prop;
2382 	return 0;
2383 }
2384 
2385 static int __devinit emac_init_phy(struct emac_instance *dev)
2386 {
2387 	struct device_node *np = dev->ofdev->dev.of_node;
2388 	struct net_device *ndev = dev->ndev;
2389 	u32 phy_map, adv;
2390 	int i;
2391 
2392 	dev->phy.dev = ndev;
2393 	dev->phy.mode = dev->phy_mode;
2394 
2395 	/* PHY-less configuration.
2396 	 * XXX I probably should move these settings to the dev tree
2397 	 */
2398 	if (dev->phy_address == 0xffffffff && dev->phy_map == 0xffffffff) {
2399 		emac_reset(dev);
2400 
2401 		/* PHY-less configuration.
2402 		 * XXX I probably should move these settings to the dev tree
2403 		 */
2404 		dev->phy.address = -1;
2405 		dev->phy.features = SUPPORTED_MII;
2406 		if (emac_phy_supports_gige(dev->phy_mode))
2407 			dev->phy.features |= SUPPORTED_1000baseT_Full;
2408 		else
2409 			dev->phy.features |= SUPPORTED_100baseT_Full;
2410 		dev->phy.pause = 1;
2411 
2412 		return 0;
2413 	}
2414 
2415 	mutex_lock(&emac_phy_map_lock);
2416 	phy_map = dev->phy_map | busy_phy_map;
2417 
2418 	DBG(dev, "PHY maps %08x %08x" NL, dev->phy_map, busy_phy_map);
2419 
2420 	dev->phy.mdio_read = emac_mdio_read;
2421 	dev->phy.mdio_write = emac_mdio_write;
2422 
2423 	/* Enable internal clock source */
2424 #ifdef CONFIG_PPC_DCR_NATIVE
2425 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2426 		dcri_clrset(SDR0, SDR0_MFR, 0, SDR0_MFR_ECS);
2427 #endif
2428 	/* PHY clock workaround */
2429 	emac_rx_clk_tx(dev);
2430 
2431 	/* Enable internal clock source on 440GX*/
2432 #ifdef CONFIG_PPC_DCR_NATIVE
2433 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2434 		dcri_clrset(SDR0, SDR0_MFR, 0, SDR0_MFR_ECS);
2435 #endif
2436 	/* Configure EMAC with defaults so we can at least use MDIO
2437 	 * This is needed mostly for 440GX
2438 	 */
2439 	if (emac_phy_gpcs(dev->phy.mode)) {
2440 		/* XXX
2441 		 * Make GPCS PHY address equal to EMAC index.
2442 		 * We probably should take into account busy_phy_map
2443 		 * and/or phy_map here.
2444 		 *
2445 		 * Note that the busy_phy_map is currently global
2446 		 * while it should probably be per-ASIC...
2447 		 */
2448 		dev->phy.gpcs_address = dev->gpcs_address;
2449 		if (dev->phy.gpcs_address == 0xffffffff)
2450 			dev->phy.address = dev->cell_index;
2451 	}
2452 
2453 	emac_configure(dev);
2454 
2455 	if (dev->phy_address != 0xffffffff)
2456 		phy_map = ~(1 << dev->phy_address);
2457 
2458 	for (i = 0; i < 0x20; phy_map >>= 1, ++i)
2459 		if (!(phy_map & 1)) {
2460 			int r;
2461 			busy_phy_map |= 1 << i;
2462 
2463 			/* Quick check if there is a PHY at the address */
2464 			r = emac_mdio_read(dev->ndev, i, MII_BMCR);
2465 			if (r == 0xffff || r < 0)
2466 				continue;
2467 			if (!emac_mii_phy_probe(&dev->phy, i))
2468 				break;
2469 		}
2470 
2471 	/* Enable external clock source */
2472 #ifdef CONFIG_PPC_DCR_NATIVE
2473 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2474 		dcri_clrset(SDR0, SDR0_MFR, SDR0_MFR_ECS, 0);
2475 #endif
2476 	mutex_unlock(&emac_phy_map_lock);
2477 	if (i == 0x20) {
2478 		printk(KERN_WARNING "%s: can't find PHY!\n", np->full_name);
2479 		return -ENXIO;
2480 	}
2481 
2482 	/* Init PHY */
2483 	if (dev->phy.def->ops->init)
2484 		dev->phy.def->ops->init(&dev->phy);
2485 
2486 	/* Disable any PHY features not supported by the platform */
2487 	dev->phy.def->features &= ~dev->phy_feat_exc;
2488 	dev->phy.features &= ~dev->phy_feat_exc;
2489 
2490 	/* Setup initial link parameters */
2491 	if (dev->phy.features & SUPPORTED_Autoneg) {
2492 		adv = dev->phy.features;
2493 		if (!emac_has_feature(dev, EMAC_FTR_NO_FLOW_CONTROL_40x))
2494 			adv |= ADVERTISED_Pause | ADVERTISED_Asym_Pause;
2495 		/* Restart autonegotiation */
2496 		dev->phy.def->ops->setup_aneg(&dev->phy, adv);
2497 	} else {
2498 		u32 f = dev->phy.def->features;
2499 		int speed = SPEED_10, fd = DUPLEX_HALF;
2500 
2501 		/* Select highest supported speed/duplex */
2502 		if (f & SUPPORTED_1000baseT_Full) {
2503 			speed = SPEED_1000;
2504 			fd = DUPLEX_FULL;
2505 		} else if (f & SUPPORTED_1000baseT_Half)
2506 			speed = SPEED_1000;
2507 		else if (f & SUPPORTED_100baseT_Full) {
2508 			speed = SPEED_100;
2509 			fd = DUPLEX_FULL;
2510 		} else if (f & SUPPORTED_100baseT_Half)
2511 			speed = SPEED_100;
2512 		else if (f & SUPPORTED_10baseT_Full)
2513 			fd = DUPLEX_FULL;
2514 
2515 		/* Force link parameters */
2516 		dev->phy.def->ops->setup_forced(&dev->phy, speed, fd);
2517 	}
2518 	return 0;
2519 }
2520 
2521 static int __devinit emac_init_config(struct emac_instance *dev)
2522 {
2523 	struct device_node *np = dev->ofdev->dev.of_node;
2524 	const void *p;
2525 
2526 	/* Read config from device-tree */
2527 	if (emac_read_uint_prop(np, "mal-device", &dev->mal_ph, 1))
2528 		return -ENXIO;
2529 	if (emac_read_uint_prop(np, "mal-tx-channel", &dev->mal_tx_chan, 1))
2530 		return -ENXIO;
2531 	if (emac_read_uint_prop(np, "mal-rx-channel", &dev->mal_rx_chan, 1))
2532 		return -ENXIO;
2533 	if (emac_read_uint_prop(np, "cell-index", &dev->cell_index, 1))
2534 		return -ENXIO;
2535 	if (emac_read_uint_prop(np, "max-frame-size", &dev->max_mtu, 0))
2536 		dev->max_mtu = 1500;
2537 	if (emac_read_uint_prop(np, "rx-fifo-size", &dev->rx_fifo_size, 0))
2538 		dev->rx_fifo_size = 2048;
2539 	if (emac_read_uint_prop(np, "tx-fifo-size", &dev->tx_fifo_size, 0))
2540 		dev->tx_fifo_size = 2048;
2541 	if (emac_read_uint_prop(np, "rx-fifo-size-gige", &dev->rx_fifo_size_gige, 0))
2542 		dev->rx_fifo_size_gige = dev->rx_fifo_size;
2543 	if (emac_read_uint_prop(np, "tx-fifo-size-gige", &dev->tx_fifo_size_gige, 0))
2544 		dev->tx_fifo_size_gige = dev->tx_fifo_size;
2545 	if (emac_read_uint_prop(np, "phy-address", &dev->phy_address, 0))
2546 		dev->phy_address = 0xffffffff;
2547 	if (emac_read_uint_prop(np, "phy-map", &dev->phy_map, 0))
2548 		dev->phy_map = 0xffffffff;
2549 	if (emac_read_uint_prop(np, "gpcs-address", &dev->gpcs_address, 0))
2550 		dev->gpcs_address = 0xffffffff;
2551 	if (emac_read_uint_prop(np->parent, "clock-frequency", &dev->opb_bus_freq, 1))
2552 		return -ENXIO;
2553 	if (emac_read_uint_prop(np, "tah-device", &dev->tah_ph, 0))
2554 		dev->tah_ph = 0;
2555 	if (emac_read_uint_prop(np, "tah-channel", &dev->tah_port, 0))
2556 		dev->tah_port = 0;
2557 	if (emac_read_uint_prop(np, "mdio-device", &dev->mdio_ph, 0))
2558 		dev->mdio_ph = 0;
2559 	if (emac_read_uint_prop(np, "zmii-device", &dev->zmii_ph, 0))
2560 		dev->zmii_ph = 0;
2561 	if (emac_read_uint_prop(np, "zmii-channel", &dev->zmii_port, 0))
2562 		dev->zmii_port = 0xffffffff;
2563 	if (emac_read_uint_prop(np, "rgmii-device", &dev->rgmii_ph, 0))
2564 		dev->rgmii_ph = 0;
2565 	if (emac_read_uint_prop(np, "rgmii-channel", &dev->rgmii_port, 0))
2566 		dev->rgmii_port = 0xffffffff;
2567 	if (emac_read_uint_prop(np, "fifo-entry-size", &dev->fifo_entry_size, 0))
2568 		dev->fifo_entry_size = 16;
2569 	if (emac_read_uint_prop(np, "mal-burst-size", &dev->mal_burst_size, 0))
2570 		dev->mal_burst_size = 256;
2571 
2572 	/* PHY mode needs some decoding */
2573 	dev->phy_mode = of_get_phy_mode(np);
2574 	if (dev->phy_mode < 0)
2575 		dev->phy_mode = PHY_MODE_NA;
2576 
2577 	/* Check EMAC version */
2578 	if (of_device_is_compatible(np, "ibm,emac4sync")) {
2579 		dev->features |= (EMAC_FTR_EMAC4 | EMAC_FTR_EMAC4SYNC);
2580 		if (of_device_is_compatible(np, "ibm,emac-460ex") ||
2581 		    of_device_is_compatible(np, "ibm,emac-460gt"))
2582 			dev->features |= EMAC_FTR_460EX_PHY_CLK_FIX;
2583 		if (of_device_is_compatible(np, "ibm,emac-405ex") ||
2584 		    of_device_is_compatible(np, "ibm,emac-405exr"))
2585 			dev->features |= EMAC_FTR_440EP_PHY_CLK_FIX;
2586 		if (of_device_is_compatible(np, "ibm,emac-apm821xx")) {
2587 			dev->features |= (EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE |
2588 					  EMAC_FTR_APM821XX_NO_HALF_DUPLEX |
2589 					  EMAC_FTR_460EX_PHY_CLK_FIX);
2590 		}
2591 	} else if (of_device_is_compatible(np, "ibm,emac4")) {
2592 		dev->features |= EMAC_FTR_EMAC4;
2593 		if (of_device_is_compatible(np, "ibm,emac-440gx"))
2594 			dev->features |= EMAC_FTR_440GX_PHY_CLK_FIX;
2595 	} else {
2596 		if (of_device_is_compatible(np, "ibm,emac-440ep") ||
2597 		    of_device_is_compatible(np, "ibm,emac-440gr"))
2598 			dev->features |= EMAC_FTR_440EP_PHY_CLK_FIX;
2599 		if (of_device_is_compatible(np, "ibm,emac-405ez")) {
2600 #ifdef CONFIG_IBM_EMAC_NO_FLOW_CTRL
2601 			dev->features |= EMAC_FTR_NO_FLOW_CONTROL_40x;
2602 #else
2603 			printk(KERN_ERR "%s: Flow control not disabled!\n",
2604 					np->full_name);
2605 			return -ENXIO;
2606 #endif
2607 		}
2608 
2609 	}
2610 
2611 	/* Fixup some feature bits based on the device tree */
2612 	if (of_get_property(np, "has-inverted-stacr-oc", NULL))
2613 		dev->features |= EMAC_FTR_STACR_OC_INVERT;
2614 	if (of_get_property(np, "has-new-stacr-staopc", NULL))
2615 		dev->features |= EMAC_FTR_HAS_NEW_STACR;
2616 
2617 	/* CAB lacks the appropriate properties */
2618 	if (of_device_is_compatible(np, "ibm,emac-axon"))
2619 		dev->features |= EMAC_FTR_HAS_NEW_STACR |
2620 			EMAC_FTR_STACR_OC_INVERT;
2621 
2622 	/* Enable TAH/ZMII/RGMII features as found */
2623 	if (dev->tah_ph != 0) {
2624 #ifdef CONFIG_IBM_EMAC_TAH
2625 		dev->features |= EMAC_FTR_HAS_TAH;
2626 #else
2627 		printk(KERN_ERR "%s: TAH support not enabled !\n",
2628 		       np->full_name);
2629 		return -ENXIO;
2630 #endif
2631 	}
2632 
2633 	if (dev->zmii_ph != 0) {
2634 #ifdef CONFIG_IBM_EMAC_ZMII
2635 		dev->features |= EMAC_FTR_HAS_ZMII;
2636 #else
2637 		printk(KERN_ERR "%s: ZMII support not enabled !\n",
2638 		       np->full_name);
2639 		return -ENXIO;
2640 #endif
2641 	}
2642 
2643 	if (dev->rgmii_ph != 0) {
2644 #ifdef CONFIG_IBM_EMAC_RGMII
2645 		dev->features |= EMAC_FTR_HAS_RGMII;
2646 #else
2647 		printk(KERN_ERR "%s: RGMII support not enabled !\n",
2648 		       np->full_name);
2649 		return -ENXIO;
2650 #endif
2651 	}
2652 
2653 	/* Read MAC-address */
2654 	p = of_get_property(np, "local-mac-address", NULL);
2655 	if (p == NULL) {
2656 		printk(KERN_ERR "%s: Can't find local-mac-address property\n",
2657 		       np->full_name);
2658 		return -ENXIO;
2659 	}
2660 	memcpy(dev->ndev->dev_addr, p, 6);
2661 
2662 	/* IAHT and GAHT filter parameterization */
2663 	if (emac_has_feature(dev, EMAC_FTR_EMAC4SYNC)) {
2664 		dev->xaht_slots_shift = EMAC4SYNC_XAHT_SLOTS_SHIFT;
2665 		dev->xaht_width_shift = EMAC4SYNC_XAHT_WIDTH_SHIFT;
2666 	} else {
2667 		dev->xaht_slots_shift = EMAC4_XAHT_SLOTS_SHIFT;
2668 		dev->xaht_width_shift = EMAC4_XAHT_WIDTH_SHIFT;
2669 	}
2670 
2671 	DBG(dev, "features     : 0x%08x / 0x%08x\n", dev->features, EMAC_FTRS_POSSIBLE);
2672 	DBG(dev, "tx_fifo_size : %d (%d gige)\n", dev->tx_fifo_size, dev->tx_fifo_size_gige);
2673 	DBG(dev, "rx_fifo_size : %d (%d gige)\n", dev->rx_fifo_size, dev->rx_fifo_size_gige);
2674 	DBG(dev, "max_mtu      : %d\n", dev->max_mtu);
2675 	DBG(dev, "OPB freq     : %d\n", dev->opb_bus_freq);
2676 
2677 	return 0;
2678 }
2679 
2680 static const struct net_device_ops emac_netdev_ops = {
2681 	.ndo_open		= emac_open,
2682 	.ndo_stop		= emac_close,
2683 	.ndo_get_stats		= emac_stats,
2684 	.ndo_set_rx_mode	= emac_set_multicast_list,
2685 	.ndo_do_ioctl		= emac_ioctl,
2686 	.ndo_tx_timeout		= emac_tx_timeout,
2687 	.ndo_validate_addr	= eth_validate_addr,
2688 	.ndo_set_mac_address	= eth_mac_addr,
2689 	.ndo_start_xmit		= emac_start_xmit,
2690 	.ndo_change_mtu		= eth_change_mtu,
2691 };
2692 
2693 static const struct net_device_ops emac_gige_netdev_ops = {
2694 	.ndo_open		= emac_open,
2695 	.ndo_stop		= emac_close,
2696 	.ndo_get_stats		= emac_stats,
2697 	.ndo_set_rx_mode	= emac_set_multicast_list,
2698 	.ndo_do_ioctl		= emac_ioctl,
2699 	.ndo_tx_timeout		= emac_tx_timeout,
2700 	.ndo_validate_addr	= eth_validate_addr,
2701 	.ndo_set_mac_address	= eth_mac_addr,
2702 	.ndo_start_xmit		= emac_start_xmit_sg,
2703 	.ndo_change_mtu		= emac_change_mtu,
2704 };
2705 
2706 static int __devinit emac_probe(struct platform_device *ofdev)
2707 {
2708 	struct net_device *ndev;
2709 	struct emac_instance *dev;
2710 	struct device_node *np = ofdev->dev.of_node;
2711 	struct device_node **blist = NULL;
2712 	int err, i;
2713 
2714 	/* Skip unused/unwired EMACS.  We leave the check for an unused
2715 	 * property here for now, but new flat device trees should set a
2716 	 * status property to "disabled" instead.
2717 	 */
2718 	if (of_get_property(np, "unused", NULL) || !of_device_is_available(np))
2719 		return -ENODEV;
2720 
2721 	/* Find ourselves in the bootlist if we are there */
2722 	for (i = 0; i < EMAC_BOOT_LIST_SIZE; i++)
2723 		if (emac_boot_list[i] == np)
2724 			blist = &emac_boot_list[i];
2725 
2726 	/* Allocate our net_device structure */
2727 	err = -ENOMEM;
2728 	ndev = alloc_etherdev(sizeof(struct emac_instance));
2729 	if (!ndev)
2730 		goto err_gone;
2731 
2732 	dev = netdev_priv(ndev);
2733 	dev->ndev = ndev;
2734 	dev->ofdev = ofdev;
2735 	dev->blist = blist;
2736 	SET_NETDEV_DEV(ndev, &ofdev->dev);
2737 
2738 	/* Initialize some embedded data structures */
2739 	mutex_init(&dev->mdio_lock);
2740 	mutex_init(&dev->link_lock);
2741 	spin_lock_init(&dev->lock);
2742 	INIT_WORK(&dev->reset_work, emac_reset_work);
2743 
2744 	/* Init various config data based on device-tree */
2745 	err = emac_init_config(dev);
2746 	if (err != 0)
2747 		goto err_free;
2748 
2749 	/* Get interrupts. EMAC irq is mandatory, WOL irq is optional */
2750 	dev->emac_irq = irq_of_parse_and_map(np, 0);
2751 	dev->wol_irq = irq_of_parse_and_map(np, 1);
2752 	if (dev->emac_irq == NO_IRQ) {
2753 		printk(KERN_ERR "%s: Can't map main interrupt\n", np->full_name);
2754 		goto err_free;
2755 	}
2756 	ndev->irq = dev->emac_irq;
2757 
2758 	/* Map EMAC regs */
2759 	if (of_address_to_resource(np, 0, &dev->rsrc_regs)) {
2760 		printk(KERN_ERR "%s: Can't get registers address\n",
2761 		       np->full_name);
2762 		goto err_irq_unmap;
2763 	}
2764 	// TODO : request_mem_region
2765 	dev->emacp = ioremap(dev->rsrc_regs.start,
2766 			     resource_size(&dev->rsrc_regs));
2767 	if (dev->emacp == NULL) {
2768 		printk(KERN_ERR "%s: Can't map device registers!\n",
2769 		       np->full_name);
2770 		err = -ENOMEM;
2771 		goto err_irq_unmap;
2772 	}
2773 
2774 	/* Wait for dependent devices */
2775 	err = emac_wait_deps(dev);
2776 	if (err) {
2777 		printk(KERN_ERR
2778 		       "%s: Timeout waiting for dependent devices\n",
2779 		       np->full_name);
2780 		/*  display more info about what's missing ? */
2781 		goto err_reg_unmap;
2782 	}
2783 	dev->mal = dev_get_drvdata(&dev->mal_dev->dev);
2784 	if (dev->mdio_dev != NULL)
2785 		dev->mdio_instance = dev_get_drvdata(&dev->mdio_dev->dev);
2786 
2787 	/* Register with MAL */
2788 	dev->commac.ops = &emac_commac_ops;
2789 	dev->commac.dev = dev;
2790 	dev->commac.tx_chan_mask = MAL_CHAN_MASK(dev->mal_tx_chan);
2791 	dev->commac.rx_chan_mask = MAL_CHAN_MASK(dev->mal_rx_chan);
2792 	err = mal_register_commac(dev->mal, &dev->commac);
2793 	if (err) {
2794 		printk(KERN_ERR "%s: failed to register with mal %s!\n",
2795 		       np->full_name, dev->mal_dev->dev.of_node->full_name);
2796 		goto err_rel_deps;
2797 	}
2798 	dev->rx_skb_size = emac_rx_skb_size(ndev->mtu);
2799 	dev->rx_sync_size = emac_rx_sync_size(ndev->mtu);
2800 
2801 	/* Get pointers to BD rings */
2802 	dev->tx_desc =
2803 	    dev->mal->bd_virt + mal_tx_bd_offset(dev->mal, dev->mal_tx_chan);
2804 	dev->rx_desc =
2805 	    dev->mal->bd_virt + mal_rx_bd_offset(dev->mal, dev->mal_rx_chan);
2806 
2807 	DBG(dev, "tx_desc %p" NL, dev->tx_desc);
2808 	DBG(dev, "rx_desc %p" NL, dev->rx_desc);
2809 
2810 	/* Clean rings */
2811 	memset(dev->tx_desc, 0, NUM_TX_BUFF * sizeof(struct mal_descriptor));
2812 	memset(dev->rx_desc, 0, NUM_RX_BUFF * sizeof(struct mal_descriptor));
2813 	memset(dev->tx_skb, 0, NUM_TX_BUFF * sizeof(struct sk_buff *));
2814 	memset(dev->rx_skb, 0, NUM_RX_BUFF * sizeof(struct sk_buff *));
2815 
2816 	/* Attach to ZMII, if needed */
2817 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII) &&
2818 	    (err = zmii_attach(dev->zmii_dev, dev->zmii_port, &dev->phy_mode)) != 0)
2819 		goto err_unreg_commac;
2820 
2821 	/* Attach to RGMII, if needed */
2822 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII) &&
2823 	    (err = rgmii_attach(dev->rgmii_dev, dev->rgmii_port, dev->phy_mode)) != 0)
2824 		goto err_detach_zmii;
2825 
2826 	/* Attach to TAH, if needed */
2827 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH) &&
2828 	    (err = tah_attach(dev->tah_dev, dev->tah_port)) != 0)
2829 		goto err_detach_rgmii;
2830 
2831 	/* Set some link defaults before we can find out real parameters */
2832 	dev->phy.speed = SPEED_100;
2833 	dev->phy.duplex = DUPLEX_FULL;
2834 	dev->phy.autoneg = AUTONEG_DISABLE;
2835 	dev->phy.pause = dev->phy.asym_pause = 0;
2836 	dev->stop_timeout = STOP_TIMEOUT_100;
2837 	INIT_DELAYED_WORK(&dev->link_work, emac_link_timer);
2838 
2839 	/* Some SoCs like APM821xx does not support Half Duplex mode. */
2840 	if (emac_has_feature(dev, EMAC_FTR_APM821XX_NO_HALF_DUPLEX)) {
2841 		dev->phy_feat_exc = (SUPPORTED_1000baseT_Half |
2842 				     SUPPORTED_100baseT_Half |
2843 				     SUPPORTED_10baseT_Half);
2844 	}
2845 
2846 	/* Find PHY if any */
2847 	err = emac_init_phy(dev);
2848 	if (err != 0)
2849 		goto err_detach_tah;
2850 
2851 	if (dev->tah_dev) {
2852 		ndev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG;
2853 		ndev->features |= ndev->hw_features | NETIF_F_RXCSUM;
2854 	}
2855 	ndev->watchdog_timeo = 5 * HZ;
2856 	if (emac_phy_supports_gige(dev->phy_mode)) {
2857 		ndev->netdev_ops = &emac_gige_netdev_ops;
2858 		dev->commac.ops = &emac_commac_sg_ops;
2859 	} else
2860 		ndev->netdev_ops = &emac_netdev_ops;
2861 	SET_ETHTOOL_OPS(ndev, &emac_ethtool_ops);
2862 
2863 	netif_carrier_off(ndev);
2864 
2865 	err = register_netdev(ndev);
2866 	if (err) {
2867 		printk(KERN_ERR "%s: failed to register net device (%d)!\n",
2868 		       np->full_name, err);
2869 		goto err_detach_tah;
2870 	}
2871 
2872 	/* Set our drvdata last as we don't want them visible until we are
2873 	 * fully initialized
2874 	 */
2875 	wmb();
2876 	dev_set_drvdata(&ofdev->dev, dev);
2877 
2878 	/* There's a new kid in town ! Let's tell everybody */
2879 	wake_up_all(&emac_probe_wait);
2880 
2881 
2882 	printk(KERN_INFO "%s: EMAC-%d %s, MAC %pM\n",
2883 	       ndev->name, dev->cell_index, np->full_name, ndev->dev_addr);
2884 
2885 	if (dev->phy_mode == PHY_MODE_SGMII)
2886 		printk(KERN_NOTICE "%s: in SGMII mode\n", ndev->name);
2887 
2888 	if (dev->phy.address >= 0)
2889 		printk("%s: found %s PHY (0x%02x)\n", ndev->name,
2890 		       dev->phy.def->name, dev->phy.address);
2891 
2892 	emac_dbg_register(dev);
2893 
2894 	/* Life is good */
2895 	return 0;
2896 
2897 	/* I have a bad feeling about this ... */
2898 
2899  err_detach_tah:
2900 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
2901 		tah_detach(dev->tah_dev, dev->tah_port);
2902  err_detach_rgmii:
2903 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
2904 		rgmii_detach(dev->rgmii_dev, dev->rgmii_port);
2905  err_detach_zmii:
2906 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
2907 		zmii_detach(dev->zmii_dev, dev->zmii_port);
2908  err_unreg_commac:
2909 	mal_unregister_commac(dev->mal, &dev->commac);
2910  err_rel_deps:
2911 	emac_put_deps(dev);
2912  err_reg_unmap:
2913 	iounmap(dev->emacp);
2914  err_irq_unmap:
2915 	if (dev->wol_irq != NO_IRQ)
2916 		irq_dispose_mapping(dev->wol_irq);
2917 	if (dev->emac_irq != NO_IRQ)
2918 		irq_dispose_mapping(dev->emac_irq);
2919  err_free:
2920 	free_netdev(ndev);
2921  err_gone:
2922 	/* if we were on the bootlist, remove us as we won't show up and
2923 	 * wake up all waiters to notify them in case they were waiting
2924 	 * on us
2925 	 */
2926 	if (blist) {
2927 		*blist = NULL;
2928 		wake_up_all(&emac_probe_wait);
2929 	}
2930 	return err;
2931 }
2932 
2933 static int __devexit emac_remove(struct platform_device *ofdev)
2934 {
2935 	struct emac_instance *dev = dev_get_drvdata(&ofdev->dev);
2936 
2937 	DBG(dev, "remove" NL);
2938 
2939 	dev_set_drvdata(&ofdev->dev, NULL);
2940 
2941 	unregister_netdev(dev->ndev);
2942 
2943 	cancel_work_sync(&dev->reset_work);
2944 
2945 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
2946 		tah_detach(dev->tah_dev, dev->tah_port);
2947 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
2948 		rgmii_detach(dev->rgmii_dev, dev->rgmii_port);
2949 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
2950 		zmii_detach(dev->zmii_dev, dev->zmii_port);
2951 
2952 	busy_phy_map &= ~(1 << dev->phy.address);
2953 	DBG(dev, "busy_phy_map now %#x" NL, busy_phy_map);
2954 
2955 	mal_unregister_commac(dev->mal, &dev->commac);
2956 	emac_put_deps(dev);
2957 
2958 	emac_dbg_unregister(dev);
2959 	iounmap(dev->emacp);
2960 
2961 	if (dev->wol_irq != NO_IRQ)
2962 		irq_dispose_mapping(dev->wol_irq);
2963 	if (dev->emac_irq != NO_IRQ)
2964 		irq_dispose_mapping(dev->emac_irq);
2965 
2966 	free_netdev(dev->ndev);
2967 
2968 	return 0;
2969 }
2970 
2971 /* XXX Features in here should be replaced by properties... */
2972 static struct of_device_id emac_match[] =
2973 {
2974 	{
2975 		.type		= "network",
2976 		.compatible	= "ibm,emac",
2977 	},
2978 	{
2979 		.type		= "network",
2980 		.compatible	= "ibm,emac4",
2981 	},
2982 	{
2983 		.type		= "network",
2984 		.compatible	= "ibm,emac4sync",
2985 	},
2986 	{},
2987 };
2988 MODULE_DEVICE_TABLE(of, emac_match);
2989 
2990 static struct platform_driver emac_driver = {
2991 	.driver = {
2992 		.name = "emac",
2993 		.owner = THIS_MODULE,
2994 		.of_match_table = emac_match,
2995 	},
2996 	.probe = emac_probe,
2997 	.remove = emac_remove,
2998 };
2999 
3000 static void __init emac_make_bootlist(void)
3001 {
3002 	struct device_node *np = NULL;
3003 	int j, max, i = 0, k;
3004 	int cell_indices[EMAC_BOOT_LIST_SIZE];
3005 
3006 	/* Collect EMACs */
3007 	while((np = of_find_all_nodes(np)) != NULL) {
3008 		const u32 *idx;
3009 
3010 		if (of_match_node(emac_match, np) == NULL)
3011 			continue;
3012 		if (of_get_property(np, "unused", NULL))
3013 			continue;
3014 		idx = of_get_property(np, "cell-index", NULL);
3015 		if (idx == NULL)
3016 			continue;
3017 		cell_indices[i] = *idx;
3018 		emac_boot_list[i++] = of_node_get(np);
3019 		if (i >= EMAC_BOOT_LIST_SIZE) {
3020 			of_node_put(np);
3021 			break;
3022 		}
3023 	}
3024 	max = i;
3025 
3026 	/* Bubble sort them (doh, what a creative algorithm :-) */
3027 	for (i = 0; max > 1 && (i < (max - 1)); i++)
3028 		for (j = i; j < max; j++) {
3029 			if (cell_indices[i] > cell_indices[j]) {
3030 				np = emac_boot_list[i];
3031 				emac_boot_list[i] = emac_boot_list[j];
3032 				emac_boot_list[j] = np;
3033 				k = cell_indices[i];
3034 				cell_indices[i] = cell_indices[j];
3035 				cell_indices[j] = k;
3036 			}
3037 		}
3038 }
3039 
3040 static int __init emac_init(void)
3041 {
3042 	int rc;
3043 
3044 	printk(KERN_INFO DRV_DESC ", version " DRV_VERSION "\n");
3045 
3046 	/* Init debug stuff */
3047 	emac_init_debug();
3048 
3049 	/* Build EMAC boot list */
3050 	emac_make_bootlist();
3051 
3052 	/* Init submodules */
3053 	rc = mal_init();
3054 	if (rc)
3055 		goto err;
3056 	rc = zmii_init();
3057 	if (rc)
3058 		goto err_mal;
3059 	rc = rgmii_init();
3060 	if (rc)
3061 		goto err_zmii;
3062 	rc = tah_init();
3063 	if (rc)
3064 		goto err_rgmii;
3065 	rc = platform_driver_register(&emac_driver);
3066 	if (rc)
3067 		goto err_tah;
3068 
3069 	return 0;
3070 
3071  err_tah:
3072 	tah_exit();
3073  err_rgmii:
3074 	rgmii_exit();
3075  err_zmii:
3076 	zmii_exit();
3077  err_mal:
3078 	mal_exit();
3079  err:
3080 	return rc;
3081 }
3082 
3083 static void __exit emac_exit(void)
3084 {
3085 	int i;
3086 
3087 	platform_driver_unregister(&emac_driver);
3088 
3089 	tah_exit();
3090 	rgmii_exit();
3091 	zmii_exit();
3092 	mal_exit();
3093 	emac_fini_debug();
3094 
3095 	/* Destroy EMAC boot list */
3096 	for (i = 0; i < EMAC_BOOT_LIST_SIZE; i++)
3097 		if (emac_boot_list[i])
3098 			of_node_put(emac_boot_list[i]);
3099 }
3100 
3101 module_init(emac_init);
3102 module_exit(emac_exit);
3103