xref: /openbmc/linux/drivers/net/ethernet/ibm/emac/core.c (revision bc5aa3a0)
1 /*
2  * drivers/net/ethernet/ibm/emac/core.c
3  *
4  * Driver for PowerPC 4xx on-chip ethernet controller.
5  *
6  * Copyright 2007 Benjamin Herrenschmidt, IBM Corp.
7  *                <benh@kernel.crashing.org>
8  *
9  * Based on the arch/ppc version of the driver:
10  *
11  * Copyright (c) 2004, 2005 Zultys Technologies.
12  * Eugene Surovegin <eugene.surovegin@zultys.com> or <ebs@ebshome.net>
13  *
14  * Based on original work by
15  * 	Matt Porter <mporter@kernel.crashing.org>
16  *	(c) 2003 Benjamin Herrenschmidt <benh@kernel.crashing.org>
17  *      Armin Kuster <akuster@mvista.com>
18  * 	Johnnie Peters <jpeters@mvista.com>
19  *
20  * This program is free software; you can redistribute  it and/or modify it
21  * under  the terms of  the GNU General  Public License as published by the
22  * Free Software Foundation;  either version 2 of the  License, or (at your
23  * option) any later version.
24  *
25  */
26 
27 #include <linux/module.h>
28 #include <linux/sched.h>
29 #include <linux/string.h>
30 #include <linux/errno.h>
31 #include <linux/delay.h>
32 #include <linux/types.h>
33 #include <linux/pci.h>
34 #include <linux/etherdevice.h>
35 #include <linux/skbuff.h>
36 #include <linux/crc32.h>
37 #include <linux/ethtool.h>
38 #include <linux/mii.h>
39 #include <linux/bitops.h>
40 #include <linux/workqueue.h>
41 #include <linux/of.h>
42 #include <linux/of_address.h>
43 #include <linux/of_irq.h>
44 #include <linux/of_net.h>
45 #include <linux/slab.h>
46 
47 #include <asm/processor.h>
48 #include <asm/io.h>
49 #include <asm/dma.h>
50 #include <asm/uaccess.h>
51 #include <asm/dcr.h>
52 #include <asm/dcr-regs.h>
53 
54 #include "core.h"
55 
56 /*
57  * Lack of dma_unmap_???? calls is intentional.
58  *
59  * API-correct usage requires additional support state information to be
60  * maintained for every RX and TX buffer descriptor (BD). Unfortunately, due to
61  * EMAC design (e.g. TX buffer passed from network stack can be split into
62  * several BDs, dma_map_single/dma_map_page can be used to map particular BD),
63  * maintaining such information will add additional overhead.
64  * Current DMA API implementation for 4xx processors only ensures cache coherency
65  * and dma_unmap_???? routines are empty and are likely to stay this way.
66  * I decided to omit dma_unmap_??? calls because I don't want to add additional
67  * complexity just for the sake of following some abstract API, when it doesn't
68  * add any real benefit to the driver. I understand that this decision maybe
69  * controversial, but I really tried to make code API-correct and efficient
70  * at the same time and didn't come up with code I liked :(.                --ebs
71  */
72 
73 #define DRV_NAME        "emac"
74 #define DRV_VERSION     "3.54"
75 #define DRV_DESC        "PPC 4xx OCP EMAC driver"
76 
77 MODULE_DESCRIPTION(DRV_DESC);
78 MODULE_AUTHOR
79     ("Eugene Surovegin <eugene.surovegin@zultys.com> or <ebs@ebshome.net>");
80 MODULE_LICENSE("GPL");
81 
82 /* minimum number of free TX descriptors required to wake up TX process */
83 #define EMAC_TX_WAKEUP_THRESH		(NUM_TX_BUFF / 4)
84 
85 /* If packet size is less than this number, we allocate small skb and copy packet
86  * contents into it instead of just sending original big skb up
87  */
88 #define EMAC_RX_COPY_THRESH		CONFIG_IBM_EMAC_RX_COPY_THRESHOLD
89 
90 /* Since multiple EMACs share MDIO lines in various ways, we need
91  * to avoid re-using the same PHY ID in cases where the arch didn't
92  * setup precise phy_map entries
93  *
94  * XXX This is something that needs to be reworked as we can have multiple
95  * EMAC "sets" (multiple ASICs containing several EMACs) though we can
96  * probably require in that case to have explicit PHY IDs in the device-tree
97  */
98 static u32 busy_phy_map;
99 static DEFINE_MUTEX(emac_phy_map_lock);
100 
101 /* This is the wait queue used to wait on any event related to probe, that
102  * is discovery of MALs, other EMACs, ZMII/RGMIIs, etc...
103  */
104 static DECLARE_WAIT_QUEUE_HEAD(emac_probe_wait);
105 
106 /* Having stable interface names is a doomed idea. However, it would be nice
107  * if we didn't have completely random interface names at boot too :-) It's
108  * just a matter of making everybody's life easier. Since we are doing
109  * threaded probing, it's a bit harder though. The base idea here is that
110  * we make up a list of all emacs in the device-tree before we register the
111  * driver. Every emac will then wait for the previous one in the list to
112  * initialize before itself. We should also keep that list ordered by
113  * cell_index.
114  * That list is only 4 entries long, meaning that additional EMACs don't
115  * get ordering guarantees unless EMAC_BOOT_LIST_SIZE is increased.
116  */
117 
118 #define EMAC_BOOT_LIST_SIZE	4
119 static struct device_node *emac_boot_list[EMAC_BOOT_LIST_SIZE];
120 
121 /* How long should I wait for dependent devices ? */
122 #define EMAC_PROBE_DEP_TIMEOUT	(HZ * 5)
123 
124 /* I don't want to litter system log with timeout errors
125  * when we have brain-damaged PHY.
126  */
127 static inline void emac_report_timeout_error(struct emac_instance *dev,
128 					     const char *error)
129 {
130 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX |
131 				  EMAC_FTR_460EX_PHY_CLK_FIX |
132 				  EMAC_FTR_440EP_PHY_CLK_FIX))
133 		DBG(dev, "%s" NL, error);
134 	else if (net_ratelimit())
135 		printk(KERN_ERR "%s: %s\n", dev->ofdev->dev.of_node->full_name,
136 			error);
137 }
138 
139 /* EMAC PHY clock workaround:
140  * 440EP/440GR has more sane SDR0_MFR register implementation than 440GX,
141  * which allows controlling each EMAC clock
142  */
143 static inline void emac_rx_clk_tx(struct emac_instance *dev)
144 {
145 #ifdef CONFIG_PPC_DCR_NATIVE
146 	if (emac_has_feature(dev, EMAC_FTR_440EP_PHY_CLK_FIX))
147 		dcri_clrset(SDR0, SDR0_MFR,
148 			    0, SDR0_MFR_ECS >> dev->cell_index);
149 #endif
150 }
151 
152 static inline void emac_rx_clk_default(struct emac_instance *dev)
153 {
154 #ifdef CONFIG_PPC_DCR_NATIVE
155 	if (emac_has_feature(dev, EMAC_FTR_440EP_PHY_CLK_FIX))
156 		dcri_clrset(SDR0, SDR0_MFR,
157 			    SDR0_MFR_ECS >> dev->cell_index, 0);
158 #endif
159 }
160 
161 /* PHY polling intervals */
162 #define PHY_POLL_LINK_ON	HZ
163 #define PHY_POLL_LINK_OFF	(HZ / 5)
164 
165 /* Graceful stop timeouts in us.
166  * We should allow up to 1 frame time (full-duplex, ignoring collisions)
167  */
168 #define STOP_TIMEOUT_10		1230
169 #define STOP_TIMEOUT_100	124
170 #define STOP_TIMEOUT_1000	13
171 #define STOP_TIMEOUT_1000_JUMBO	73
172 
173 static unsigned char default_mcast_addr[] = {
174 	0x01, 0x80, 0xC2, 0x00, 0x00, 0x01
175 };
176 
177 /* Please, keep in sync with struct ibm_emac_stats/ibm_emac_error_stats */
178 static const char emac_stats_keys[EMAC_ETHTOOL_STATS_COUNT][ETH_GSTRING_LEN] = {
179 	"rx_packets", "rx_bytes", "tx_packets", "tx_bytes", "rx_packets_csum",
180 	"tx_packets_csum", "tx_undo", "rx_dropped_stack", "rx_dropped_oom",
181 	"rx_dropped_error", "rx_dropped_resize", "rx_dropped_mtu",
182 	"rx_stopped", "rx_bd_errors", "rx_bd_overrun", "rx_bd_bad_packet",
183 	"rx_bd_runt_packet", "rx_bd_short_event", "rx_bd_alignment_error",
184 	"rx_bd_bad_fcs", "rx_bd_packet_too_long", "rx_bd_out_of_range",
185 	"rx_bd_in_range", "rx_parity", "rx_fifo_overrun", "rx_overrun",
186 	"rx_bad_packet", "rx_runt_packet", "rx_short_event",
187 	"rx_alignment_error", "rx_bad_fcs", "rx_packet_too_long",
188 	"rx_out_of_range", "rx_in_range", "tx_dropped", "tx_bd_errors",
189 	"tx_bd_bad_fcs", "tx_bd_carrier_loss", "tx_bd_excessive_deferral",
190 	"tx_bd_excessive_collisions", "tx_bd_late_collision",
191 	"tx_bd_multple_collisions", "tx_bd_single_collision",
192 	"tx_bd_underrun", "tx_bd_sqe", "tx_parity", "tx_underrun", "tx_sqe",
193 	"tx_errors"
194 };
195 
196 static irqreturn_t emac_irq(int irq, void *dev_instance);
197 static void emac_clean_tx_ring(struct emac_instance *dev);
198 static void __emac_set_multicast_list(struct emac_instance *dev);
199 
200 static inline int emac_phy_supports_gige(int phy_mode)
201 {
202 	return  phy_mode == PHY_MODE_GMII ||
203 		phy_mode == PHY_MODE_RGMII ||
204 		phy_mode == PHY_MODE_SGMII ||
205 		phy_mode == PHY_MODE_TBI ||
206 		phy_mode == PHY_MODE_RTBI;
207 }
208 
209 static inline int emac_phy_gpcs(int phy_mode)
210 {
211 	return  phy_mode == PHY_MODE_SGMII ||
212 		phy_mode == PHY_MODE_TBI ||
213 		phy_mode == PHY_MODE_RTBI;
214 }
215 
216 static inline void emac_tx_enable(struct emac_instance *dev)
217 {
218 	struct emac_regs __iomem *p = dev->emacp;
219 	u32 r;
220 
221 	DBG(dev, "tx_enable" NL);
222 
223 	r = in_be32(&p->mr0);
224 	if (!(r & EMAC_MR0_TXE))
225 		out_be32(&p->mr0, r | EMAC_MR0_TXE);
226 }
227 
228 static void emac_tx_disable(struct emac_instance *dev)
229 {
230 	struct emac_regs __iomem *p = dev->emacp;
231 	u32 r;
232 
233 	DBG(dev, "tx_disable" NL);
234 
235 	r = in_be32(&p->mr0);
236 	if (r & EMAC_MR0_TXE) {
237 		int n = dev->stop_timeout;
238 		out_be32(&p->mr0, r & ~EMAC_MR0_TXE);
239 		while (!(in_be32(&p->mr0) & EMAC_MR0_TXI) && n) {
240 			udelay(1);
241 			--n;
242 		}
243 		if (unlikely(!n))
244 			emac_report_timeout_error(dev, "TX disable timeout");
245 	}
246 }
247 
248 static void emac_rx_enable(struct emac_instance *dev)
249 {
250 	struct emac_regs __iomem *p = dev->emacp;
251 	u32 r;
252 
253 	if (unlikely(test_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags)))
254 		goto out;
255 
256 	DBG(dev, "rx_enable" NL);
257 
258 	r = in_be32(&p->mr0);
259 	if (!(r & EMAC_MR0_RXE)) {
260 		if (unlikely(!(r & EMAC_MR0_RXI))) {
261 			/* Wait if previous async disable is still in progress */
262 			int n = dev->stop_timeout;
263 			while (!(r = in_be32(&p->mr0) & EMAC_MR0_RXI) && n) {
264 				udelay(1);
265 				--n;
266 			}
267 			if (unlikely(!n))
268 				emac_report_timeout_error(dev,
269 							  "RX disable timeout");
270 		}
271 		out_be32(&p->mr0, r | EMAC_MR0_RXE);
272 	}
273  out:
274 	;
275 }
276 
277 static void emac_rx_disable(struct emac_instance *dev)
278 {
279 	struct emac_regs __iomem *p = dev->emacp;
280 	u32 r;
281 
282 	DBG(dev, "rx_disable" NL);
283 
284 	r = in_be32(&p->mr0);
285 	if (r & EMAC_MR0_RXE) {
286 		int n = dev->stop_timeout;
287 		out_be32(&p->mr0, r & ~EMAC_MR0_RXE);
288 		while (!(in_be32(&p->mr0) & EMAC_MR0_RXI) && n) {
289 			udelay(1);
290 			--n;
291 		}
292 		if (unlikely(!n))
293 			emac_report_timeout_error(dev, "RX disable timeout");
294 	}
295 }
296 
297 static inline void emac_netif_stop(struct emac_instance *dev)
298 {
299 	netif_tx_lock_bh(dev->ndev);
300 	netif_addr_lock(dev->ndev);
301 	dev->no_mcast = 1;
302 	netif_addr_unlock(dev->ndev);
303 	netif_tx_unlock_bh(dev->ndev);
304 	netif_trans_update(dev->ndev);	/* prevent tx timeout */
305 	mal_poll_disable(dev->mal, &dev->commac);
306 	netif_tx_disable(dev->ndev);
307 }
308 
309 static inline void emac_netif_start(struct emac_instance *dev)
310 {
311 	netif_tx_lock_bh(dev->ndev);
312 	netif_addr_lock(dev->ndev);
313 	dev->no_mcast = 0;
314 	if (dev->mcast_pending && netif_running(dev->ndev))
315 		__emac_set_multicast_list(dev);
316 	netif_addr_unlock(dev->ndev);
317 	netif_tx_unlock_bh(dev->ndev);
318 
319 	netif_wake_queue(dev->ndev);
320 
321 	/* NOTE: unconditional netif_wake_queue is only appropriate
322 	 * so long as all callers are assured to have free tx slots
323 	 * (taken from tg3... though the case where that is wrong is
324 	 *  not terribly harmful)
325 	 */
326 	mal_poll_enable(dev->mal, &dev->commac);
327 }
328 
329 static inline void emac_rx_disable_async(struct emac_instance *dev)
330 {
331 	struct emac_regs __iomem *p = dev->emacp;
332 	u32 r;
333 
334 	DBG(dev, "rx_disable_async" NL);
335 
336 	r = in_be32(&p->mr0);
337 	if (r & EMAC_MR0_RXE)
338 		out_be32(&p->mr0, r & ~EMAC_MR0_RXE);
339 }
340 
341 static int emac_reset(struct emac_instance *dev)
342 {
343 	struct emac_regs __iomem *p = dev->emacp;
344 	int n = 20;
345 
346 	DBG(dev, "reset" NL);
347 
348 	if (!dev->reset_failed) {
349 		/* 40x erratum suggests stopping RX channel before reset,
350 		 * we stop TX as well
351 		 */
352 		emac_rx_disable(dev);
353 		emac_tx_disable(dev);
354 	}
355 
356 #ifdef CONFIG_PPC_DCR_NATIVE
357 	/*
358 	 * PPC460EX/GT Embedded Processor Advanced User's Manual
359 	 * section 28.10.1 Mode Register 0 (EMACx_MR0) states:
360 	 * Note: The PHY must provide a TX Clk in order to perform a soft reset
361 	 * of the EMAC. If none is present, select the internal clock
362 	 * (SDR0_ETH_CFG[EMACx_PHY_CLK] = 1).
363 	 * After a soft reset, select the external clock.
364 	 */
365 	if (emac_has_feature(dev, EMAC_FTR_460EX_PHY_CLK_FIX)) {
366 		if (dev->phy_address == 0xffffffff &&
367 		    dev->phy_map == 0xffffffff) {
368 			/* No PHY: select internal loop clock before reset */
369 			dcri_clrset(SDR0, SDR0_ETH_CFG,
370 				    0, SDR0_ETH_CFG_ECS << dev->cell_index);
371 		} else {
372 			/* PHY present: select external clock before reset */
373 			dcri_clrset(SDR0, SDR0_ETH_CFG,
374 				    SDR0_ETH_CFG_ECS << dev->cell_index, 0);
375 		}
376 	}
377 #endif
378 
379 	out_be32(&p->mr0, EMAC_MR0_SRST);
380 	while ((in_be32(&p->mr0) & EMAC_MR0_SRST) && n)
381 		--n;
382 
383 #ifdef CONFIG_PPC_DCR_NATIVE
384 	if (emac_has_feature(dev, EMAC_FTR_460EX_PHY_CLK_FIX)) {
385 		if (dev->phy_address == 0xffffffff &&
386 		    dev->phy_map == 0xffffffff) {
387 			/* No PHY: restore external clock source after reset */
388 			dcri_clrset(SDR0, SDR0_ETH_CFG,
389 				    SDR0_ETH_CFG_ECS << dev->cell_index, 0);
390 		}
391 	}
392 #endif
393 
394 	if (n) {
395 		dev->reset_failed = 0;
396 		return 0;
397 	} else {
398 		emac_report_timeout_error(dev, "reset timeout");
399 		dev->reset_failed = 1;
400 		return -ETIMEDOUT;
401 	}
402 }
403 
404 static void emac_hash_mc(struct emac_instance *dev)
405 {
406 	const int regs = EMAC_XAHT_REGS(dev);
407 	u32 *gaht_base = emac_gaht_base(dev);
408 	u32 gaht_temp[regs];
409 	struct netdev_hw_addr *ha;
410 	int i;
411 
412 	DBG(dev, "hash_mc %d" NL, netdev_mc_count(dev->ndev));
413 
414 	memset(gaht_temp, 0, sizeof (gaht_temp));
415 
416 	netdev_for_each_mc_addr(ha, dev->ndev) {
417 		int slot, reg, mask;
418 		DBG2(dev, "mc %pM" NL, ha->addr);
419 
420 		slot = EMAC_XAHT_CRC_TO_SLOT(dev,
421 					     ether_crc(ETH_ALEN, ha->addr));
422 		reg = EMAC_XAHT_SLOT_TO_REG(dev, slot);
423 		mask = EMAC_XAHT_SLOT_TO_MASK(dev, slot);
424 
425 		gaht_temp[reg] |= mask;
426 	}
427 
428 	for (i = 0; i < regs; i++)
429 		out_be32(gaht_base + i, gaht_temp[i]);
430 }
431 
432 static inline u32 emac_iff2rmr(struct net_device *ndev)
433 {
434 	struct emac_instance *dev = netdev_priv(ndev);
435 	u32 r;
436 
437 	r = EMAC_RMR_SP | EMAC_RMR_SFCS | EMAC_RMR_IAE | EMAC_RMR_BAE;
438 
439 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
440 	    r |= EMAC4_RMR_BASE;
441 	else
442 	    r |= EMAC_RMR_BASE;
443 
444 	if (ndev->flags & IFF_PROMISC)
445 		r |= EMAC_RMR_PME;
446 	else if (ndev->flags & IFF_ALLMULTI ||
447 			 (netdev_mc_count(ndev) > EMAC_XAHT_SLOTS(dev)))
448 		r |= EMAC_RMR_PMME;
449 	else if (!netdev_mc_empty(ndev))
450 		r |= EMAC_RMR_MAE;
451 
452 	if (emac_has_feature(dev, EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE)) {
453 		r &= ~EMAC4_RMR_MJS_MASK;
454 		r |= EMAC4_RMR_MJS(ndev->mtu);
455 	}
456 
457 	return r;
458 }
459 
460 static u32 __emac_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
461 {
462 	u32 ret = EMAC_MR1_VLE | EMAC_MR1_IST | EMAC_MR1_TR0_MULT;
463 
464 	DBG2(dev, "__emac_calc_base_mr1" NL);
465 
466 	switch(tx_size) {
467 	case 2048:
468 		ret |= EMAC_MR1_TFS_2K;
469 		break;
470 	default:
471 		printk(KERN_WARNING "%s: Unknown Tx FIFO size %d\n",
472 		       dev->ndev->name, tx_size);
473 	}
474 
475 	switch(rx_size) {
476 	case 16384:
477 		ret |= EMAC_MR1_RFS_16K;
478 		break;
479 	case 4096:
480 		ret |= EMAC_MR1_RFS_4K;
481 		break;
482 	default:
483 		printk(KERN_WARNING "%s: Unknown Rx FIFO size %d\n",
484 		       dev->ndev->name, rx_size);
485 	}
486 
487 	return ret;
488 }
489 
490 static u32 __emac4_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
491 {
492 	u32 ret = EMAC_MR1_VLE | EMAC_MR1_IST | EMAC4_MR1_TR |
493 		EMAC4_MR1_OBCI(dev->opb_bus_freq / 1000000);
494 
495 	DBG2(dev, "__emac4_calc_base_mr1" NL);
496 
497 	switch(tx_size) {
498 	case 16384:
499 		ret |= EMAC4_MR1_TFS_16K;
500 		break;
501 	case 4096:
502 		ret |= EMAC4_MR1_TFS_4K;
503 		break;
504 	case 2048:
505 		ret |= EMAC4_MR1_TFS_2K;
506 		break;
507 	default:
508 		printk(KERN_WARNING "%s: Unknown Tx FIFO size %d\n",
509 		       dev->ndev->name, tx_size);
510 	}
511 
512 	switch(rx_size) {
513 	case 16384:
514 		ret |= EMAC4_MR1_RFS_16K;
515 		break;
516 	case 4096:
517 		ret |= EMAC4_MR1_RFS_4K;
518 		break;
519 	case 2048:
520 		ret |= EMAC4_MR1_RFS_2K;
521 		break;
522 	default:
523 		printk(KERN_WARNING "%s: Unknown Rx FIFO size %d\n",
524 		       dev->ndev->name, rx_size);
525 	}
526 
527 	return ret;
528 }
529 
530 static u32 emac_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
531 {
532 	return emac_has_feature(dev, EMAC_FTR_EMAC4) ?
533 		__emac4_calc_base_mr1(dev, tx_size, rx_size) :
534 		__emac_calc_base_mr1(dev, tx_size, rx_size);
535 }
536 
537 static inline u32 emac_calc_trtr(struct emac_instance *dev, unsigned int size)
538 {
539 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
540 		return ((size >> 6) - 1) << EMAC_TRTR_SHIFT_EMAC4;
541 	else
542 		return ((size >> 6) - 1) << EMAC_TRTR_SHIFT;
543 }
544 
545 static inline u32 emac_calc_rwmr(struct emac_instance *dev,
546 				 unsigned int low, unsigned int high)
547 {
548 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
549 		return (low << 22) | ( (high & 0x3ff) << 6);
550 	else
551 		return (low << 23) | ( (high & 0x1ff) << 7);
552 }
553 
554 static int emac_configure(struct emac_instance *dev)
555 {
556 	struct emac_regs __iomem *p = dev->emacp;
557 	struct net_device *ndev = dev->ndev;
558 	int tx_size, rx_size, link = netif_carrier_ok(dev->ndev);
559 	u32 r, mr1 = 0;
560 
561 	DBG(dev, "configure" NL);
562 
563 	if (!link) {
564 		out_be32(&p->mr1, in_be32(&p->mr1)
565 			 | EMAC_MR1_FDE | EMAC_MR1_ILE);
566 		udelay(100);
567 	} else if (emac_reset(dev) < 0)
568 		return -ETIMEDOUT;
569 
570 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
571 		tah_reset(dev->tah_dev);
572 
573 	DBG(dev, " link = %d duplex = %d, pause = %d, asym_pause = %d\n",
574 	    link, dev->phy.duplex, dev->phy.pause, dev->phy.asym_pause);
575 
576 	/* Default fifo sizes */
577 	tx_size = dev->tx_fifo_size;
578 	rx_size = dev->rx_fifo_size;
579 
580 	/* No link, force loopback */
581 	if (!link)
582 		mr1 = EMAC_MR1_FDE | EMAC_MR1_ILE;
583 
584 	/* Check for full duplex */
585 	else if (dev->phy.duplex == DUPLEX_FULL)
586 		mr1 |= EMAC_MR1_FDE | EMAC_MR1_MWSW_001;
587 
588 	/* Adjust fifo sizes, mr1 and timeouts based on link speed */
589 	dev->stop_timeout = STOP_TIMEOUT_10;
590 	switch (dev->phy.speed) {
591 	case SPEED_1000:
592 		if (emac_phy_gpcs(dev->phy.mode)) {
593 			mr1 |= EMAC_MR1_MF_1000GPCS | EMAC_MR1_MF_IPPA(
594 				(dev->phy.gpcs_address != 0xffffffff) ?
595 				 dev->phy.gpcs_address : dev->phy.address);
596 
597 			/* Put some arbitrary OUI, Manuf & Rev IDs so we can
598 			 * identify this GPCS PHY later.
599 			 */
600 			out_be32(&p->u1.emac4.ipcr, 0xdeadbeef);
601 		} else
602 			mr1 |= EMAC_MR1_MF_1000;
603 
604 		/* Extended fifo sizes */
605 		tx_size = dev->tx_fifo_size_gige;
606 		rx_size = dev->rx_fifo_size_gige;
607 
608 		if (dev->ndev->mtu > ETH_DATA_LEN) {
609 			if (emac_has_feature(dev, EMAC_FTR_EMAC4))
610 				mr1 |= EMAC4_MR1_JPSM;
611 			else
612 				mr1 |= EMAC_MR1_JPSM;
613 			dev->stop_timeout = STOP_TIMEOUT_1000_JUMBO;
614 		} else
615 			dev->stop_timeout = STOP_TIMEOUT_1000;
616 		break;
617 	case SPEED_100:
618 		mr1 |= EMAC_MR1_MF_100;
619 		dev->stop_timeout = STOP_TIMEOUT_100;
620 		break;
621 	default: /* make gcc happy */
622 		break;
623 	}
624 
625 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
626 		rgmii_set_speed(dev->rgmii_dev, dev->rgmii_port,
627 				dev->phy.speed);
628 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
629 		zmii_set_speed(dev->zmii_dev, dev->zmii_port, dev->phy.speed);
630 
631 	/* on 40x erratum forces us to NOT use integrated flow control,
632 	 * let's hope it works on 44x ;)
633 	 */
634 	if (!emac_has_feature(dev, EMAC_FTR_NO_FLOW_CONTROL_40x) &&
635 	    dev->phy.duplex == DUPLEX_FULL) {
636 		if (dev->phy.pause)
637 			mr1 |= EMAC_MR1_EIFC | EMAC_MR1_APP;
638 		else if (dev->phy.asym_pause)
639 			mr1 |= EMAC_MR1_APP;
640 	}
641 
642 	/* Add base settings & fifo sizes & program MR1 */
643 	mr1 |= emac_calc_base_mr1(dev, tx_size, rx_size);
644 	out_be32(&p->mr1, mr1);
645 
646 	/* Set individual MAC address */
647 	out_be32(&p->iahr, (ndev->dev_addr[0] << 8) | ndev->dev_addr[1]);
648 	out_be32(&p->ialr, (ndev->dev_addr[2] << 24) |
649 		 (ndev->dev_addr[3] << 16) | (ndev->dev_addr[4] << 8) |
650 		 ndev->dev_addr[5]);
651 
652 	/* VLAN Tag Protocol ID */
653 	out_be32(&p->vtpid, 0x8100);
654 
655 	/* Receive mode register */
656 	r = emac_iff2rmr(ndev);
657 	if (r & EMAC_RMR_MAE)
658 		emac_hash_mc(dev);
659 	out_be32(&p->rmr, r);
660 
661 	/* FIFOs thresholds */
662 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
663 		r = EMAC4_TMR1((dev->mal_burst_size / dev->fifo_entry_size) + 1,
664 			       tx_size / 2 / dev->fifo_entry_size);
665 	else
666 		r = EMAC_TMR1((dev->mal_burst_size / dev->fifo_entry_size) + 1,
667 			      tx_size / 2 / dev->fifo_entry_size);
668 	out_be32(&p->tmr1, r);
669 	out_be32(&p->trtr, emac_calc_trtr(dev, tx_size / 2));
670 
671 	/* PAUSE frame is sent when RX FIFO reaches its high-water mark,
672 	   there should be still enough space in FIFO to allow the our link
673 	   partner time to process this frame and also time to send PAUSE
674 	   frame itself.
675 
676 	   Here is the worst case scenario for the RX FIFO "headroom"
677 	   (from "The Switch Book") (100Mbps, without preamble, inter-frame gap):
678 
679 	   1) One maximum-length frame on TX                    1522 bytes
680 	   2) One PAUSE frame time                                64 bytes
681 	   3) PAUSE frame decode time allowance                   64 bytes
682 	   4) One maximum-length frame on RX                    1522 bytes
683 	   5) Round-trip propagation delay of the link (100Mb)    15 bytes
684 	   ----------
685 	   3187 bytes
686 
687 	   I chose to set high-water mark to RX_FIFO_SIZE / 4 (1024 bytes)
688 	   low-water mark  to RX_FIFO_SIZE / 8 (512 bytes)
689 	 */
690 	r = emac_calc_rwmr(dev, rx_size / 8 / dev->fifo_entry_size,
691 			   rx_size / 4 / dev->fifo_entry_size);
692 	out_be32(&p->rwmr, r);
693 
694 	/* Set PAUSE timer to the maximum */
695 	out_be32(&p->ptr, 0xffff);
696 
697 	/* IRQ sources */
698 	r = EMAC_ISR_OVR | EMAC_ISR_BP | EMAC_ISR_SE |
699 		EMAC_ISR_ALE | EMAC_ISR_BFCS | EMAC_ISR_PTLE | EMAC_ISR_ORE |
700 		EMAC_ISR_IRE | EMAC_ISR_TE;
701 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
702 	    r |= EMAC4_ISR_TXPE | EMAC4_ISR_RXPE /* | EMAC4_ISR_TXUE |
703 						  EMAC4_ISR_RXOE | */;
704 	out_be32(&p->iser,  r);
705 
706 	/* We need to take GPCS PHY out of isolate mode after EMAC reset */
707 	if (emac_phy_gpcs(dev->phy.mode)) {
708 		if (dev->phy.gpcs_address != 0xffffffff)
709 			emac_mii_reset_gpcs(&dev->phy);
710 		else
711 			emac_mii_reset_phy(&dev->phy);
712 	}
713 
714 	return 0;
715 }
716 
717 static void emac_reinitialize(struct emac_instance *dev)
718 {
719 	DBG(dev, "reinitialize" NL);
720 
721 	emac_netif_stop(dev);
722 	if (!emac_configure(dev)) {
723 		emac_tx_enable(dev);
724 		emac_rx_enable(dev);
725 	}
726 	emac_netif_start(dev);
727 }
728 
729 static void emac_full_tx_reset(struct emac_instance *dev)
730 {
731 	DBG(dev, "full_tx_reset" NL);
732 
733 	emac_tx_disable(dev);
734 	mal_disable_tx_channel(dev->mal, dev->mal_tx_chan);
735 	emac_clean_tx_ring(dev);
736 	dev->tx_cnt = dev->tx_slot = dev->ack_slot = 0;
737 
738 	emac_configure(dev);
739 
740 	mal_enable_tx_channel(dev->mal, dev->mal_tx_chan);
741 	emac_tx_enable(dev);
742 	emac_rx_enable(dev);
743 }
744 
745 static void emac_reset_work(struct work_struct *work)
746 {
747 	struct emac_instance *dev = container_of(work, struct emac_instance, reset_work);
748 
749 	DBG(dev, "reset_work" NL);
750 
751 	mutex_lock(&dev->link_lock);
752 	if (dev->opened) {
753 		emac_netif_stop(dev);
754 		emac_full_tx_reset(dev);
755 		emac_netif_start(dev);
756 	}
757 	mutex_unlock(&dev->link_lock);
758 }
759 
760 static void emac_tx_timeout(struct net_device *ndev)
761 {
762 	struct emac_instance *dev = netdev_priv(ndev);
763 
764 	DBG(dev, "tx_timeout" NL);
765 
766 	schedule_work(&dev->reset_work);
767 }
768 
769 
770 static inline int emac_phy_done(struct emac_instance *dev, u32 stacr)
771 {
772 	int done = !!(stacr & EMAC_STACR_OC);
773 
774 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
775 		done = !done;
776 
777 	return done;
778 };
779 
780 static int __emac_mdio_read(struct emac_instance *dev, u8 id, u8 reg)
781 {
782 	struct emac_regs __iomem *p = dev->emacp;
783 	u32 r = 0;
784 	int n, err = -ETIMEDOUT;
785 
786 	mutex_lock(&dev->mdio_lock);
787 
788 	DBG2(dev, "mdio_read(%02x,%02x)" NL, id, reg);
789 
790 	/* Enable proper MDIO port */
791 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
792 		zmii_get_mdio(dev->zmii_dev, dev->zmii_port);
793 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
794 		rgmii_get_mdio(dev->rgmii_dev, dev->rgmii_port);
795 
796 	/* Wait for management interface to become idle */
797 	n = 20;
798 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
799 		udelay(1);
800 		if (!--n) {
801 			DBG2(dev, " -> timeout wait idle\n");
802 			goto bail;
803 		}
804 	}
805 
806 	/* Issue read command */
807 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
808 		r = EMAC4_STACR_BASE(dev->opb_bus_freq);
809 	else
810 		r = EMAC_STACR_BASE(dev->opb_bus_freq);
811 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
812 		r |= EMAC_STACR_OC;
813 	if (emac_has_feature(dev, EMAC_FTR_HAS_NEW_STACR))
814 		r |= EMACX_STACR_STAC_READ;
815 	else
816 		r |= EMAC_STACR_STAC_READ;
817 	r |= (reg & EMAC_STACR_PRA_MASK)
818 		| ((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT);
819 	out_be32(&p->stacr, r);
820 
821 	/* Wait for read to complete */
822 	n = 200;
823 	while (!emac_phy_done(dev, (r = in_be32(&p->stacr)))) {
824 		udelay(1);
825 		if (!--n) {
826 			DBG2(dev, " -> timeout wait complete\n");
827 			goto bail;
828 		}
829 	}
830 
831 	if (unlikely(r & EMAC_STACR_PHYE)) {
832 		DBG(dev, "mdio_read(%02x, %02x) failed" NL, id, reg);
833 		err = -EREMOTEIO;
834 		goto bail;
835 	}
836 
837 	r = ((r >> EMAC_STACR_PHYD_SHIFT) & EMAC_STACR_PHYD_MASK);
838 
839 	DBG2(dev, "mdio_read -> %04x" NL, r);
840 	err = 0;
841  bail:
842 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
843 		rgmii_put_mdio(dev->rgmii_dev, dev->rgmii_port);
844 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
845 		zmii_put_mdio(dev->zmii_dev, dev->zmii_port);
846 	mutex_unlock(&dev->mdio_lock);
847 
848 	return err == 0 ? r : err;
849 }
850 
851 static void __emac_mdio_write(struct emac_instance *dev, u8 id, u8 reg,
852 			      u16 val)
853 {
854 	struct emac_regs __iomem *p = dev->emacp;
855 	u32 r = 0;
856 	int n, err = -ETIMEDOUT;
857 
858 	mutex_lock(&dev->mdio_lock);
859 
860 	DBG2(dev, "mdio_write(%02x,%02x,%04x)" NL, id, reg, val);
861 
862 	/* Enable proper MDIO port */
863 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
864 		zmii_get_mdio(dev->zmii_dev, dev->zmii_port);
865 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
866 		rgmii_get_mdio(dev->rgmii_dev, dev->rgmii_port);
867 
868 	/* Wait for management interface to be idle */
869 	n = 20;
870 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
871 		udelay(1);
872 		if (!--n) {
873 			DBG2(dev, " -> timeout wait idle\n");
874 			goto bail;
875 		}
876 	}
877 
878 	/* Issue write command */
879 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
880 		r = EMAC4_STACR_BASE(dev->opb_bus_freq);
881 	else
882 		r = EMAC_STACR_BASE(dev->opb_bus_freq);
883 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
884 		r |= EMAC_STACR_OC;
885 	if (emac_has_feature(dev, EMAC_FTR_HAS_NEW_STACR))
886 		r |= EMACX_STACR_STAC_WRITE;
887 	else
888 		r |= EMAC_STACR_STAC_WRITE;
889 	r |= (reg & EMAC_STACR_PRA_MASK) |
890 		((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT) |
891 		(val << EMAC_STACR_PHYD_SHIFT);
892 	out_be32(&p->stacr, r);
893 
894 	/* Wait for write to complete */
895 	n = 200;
896 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
897 		udelay(1);
898 		if (!--n) {
899 			DBG2(dev, " -> timeout wait complete\n");
900 			goto bail;
901 		}
902 	}
903 	err = 0;
904  bail:
905 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
906 		rgmii_put_mdio(dev->rgmii_dev, dev->rgmii_port);
907 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
908 		zmii_put_mdio(dev->zmii_dev, dev->zmii_port);
909 	mutex_unlock(&dev->mdio_lock);
910 }
911 
912 static int emac_mdio_read(struct net_device *ndev, int id, int reg)
913 {
914 	struct emac_instance *dev = netdev_priv(ndev);
915 	int res;
916 
917 	res = __emac_mdio_read((dev->mdio_instance &&
918 				dev->phy.gpcs_address != id) ?
919 				dev->mdio_instance : dev,
920 			       (u8) id, (u8) reg);
921 	return res;
922 }
923 
924 static void emac_mdio_write(struct net_device *ndev, int id, int reg, int val)
925 {
926 	struct emac_instance *dev = netdev_priv(ndev);
927 
928 	__emac_mdio_write((dev->mdio_instance &&
929 			   dev->phy.gpcs_address != id) ?
930 			   dev->mdio_instance : dev,
931 			  (u8) id, (u8) reg, (u16) val);
932 }
933 
934 /* Tx lock BH */
935 static void __emac_set_multicast_list(struct emac_instance *dev)
936 {
937 	struct emac_regs __iomem *p = dev->emacp;
938 	u32 rmr = emac_iff2rmr(dev->ndev);
939 
940 	DBG(dev, "__multicast %08x" NL, rmr);
941 
942 	/* I decided to relax register access rules here to avoid
943 	 * full EMAC reset.
944 	 *
945 	 * There is a real problem with EMAC4 core if we use MWSW_001 bit
946 	 * in MR1 register and do a full EMAC reset.
947 	 * One TX BD status update is delayed and, after EMAC reset, it
948 	 * never happens, resulting in TX hung (it'll be recovered by TX
949 	 * timeout handler eventually, but this is just gross).
950 	 * So we either have to do full TX reset or try to cheat here :)
951 	 *
952 	 * The only required change is to RX mode register, so I *think* all
953 	 * we need is just to stop RX channel. This seems to work on all
954 	 * tested SoCs.                                                --ebs
955 	 *
956 	 * If we need the full reset, we might just trigger the workqueue
957 	 * and do it async... a bit nasty but should work --BenH
958 	 */
959 	dev->mcast_pending = 0;
960 	emac_rx_disable(dev);
961 	if (rmr & EMAC_RMR_MAE)
962 		emac_hash_mc(dev);
963 	out_be32(&p->rmr, rmr);
964 	emac_rx_enable(dev);
965 }
966 
967 /* Tx lock BH */
968 static void emac_set_multicast_list(struct net_device *ndev)
969 {
970 	struct emac_instance *dev = netdev_priv(ndev);
971 
972 	DBG(dev, "multicast" NL);
973 
974 	BUG_ON(!netif_running(dev->ndev));
975 
976 	if (dev->no_mcast) {
977 		dev->mcast_pending = 1;
978 		return;
979 	}
980 	__emac_set_multicast_list(dev);
981 }
982 
983 static int emac_resize_rx_ring(struct emac_instance *dev, int new_mtu)
984 {
985 	int rx_sync_size = emac_rx_sync_size(new_mtu);
986 	int rx_skb_size = emac_rx_skb_size(new_mtu);
987 	int i, ret = 0;
988 	int mr1_jumbo_bit_change = 0;
989 
990 	mutex_lock(&dev->link_lock);
991 	emac_netif_stop(dev);
992 	emac_rx_disable(dev);
993 	mal_disable_rx_channel(dev->mal, dev->mal_rx_chan);
994 
995 	if (dev->rx_sg_skb) {
996 		++dev->estats.rx_dropped_resize;
997 		dev_kfree_skb(dev->rx_sg_skb);
998 		dev->rx_sg_skb = NULL;
999 	}
1000 
1001 	/* Make a first pass over RX ring and mark BDs ready, dropping
1002 	 * non-processed packets on the way. We need this as a separate pass
1003 	 * to simplify error recovery in the case of allocation failure later.
1004 	 */
1005 	for (i = 0; i < NUM_RX_BUFF; ++i) {
1006 		if (dev->rx_desc[i].ctrl & MAL_RX_CTRL_FIRST)
1007 			++dev->estats.rx_dropped_resize;
1008 
1009 		dev->rx_desc[i].data_len = 0;
1010 		dev->rx_desc[i].ctrl = MAL_RX_CTRL_EMPTY |
1011 		    (i == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1012 	}
1013 
1014 	/* Reallocate RX ring only if bigger skb buffers are required */
1015 	if (rx_skb_size <= dev->rx_skb_size)
1016 		goto skip;
1017 
1018 	/* Second pass, allocate new skbs */
1019 	for (i = 0; i < NUM_RX_BUFF; ++i) {
1020 		struct sk_buff *skb = alloc_skb(rx_skb_size, GFP_ATOMIC);
1021 		if (!skb) {
1022 			ret = -ENOMEM;
1023 			goto oom;
1024 		}
1025 
1026 		BUG_ON(!dev->rx_skb[i]);
1027 		dev_kfree_skb(dev->rx_skb[i]);
1028 
1029 		skb_reserve(skb, EMAC_RX_SKB_HEADROOM + 2);
1030 		dev->rx_desc[i].data_ptr =
1031 		    dma_map_single(&dev->ofdev->dev, skb->data - 2, rx_sync_size,
1032 				   DMA_FROM_DEVICE) + 2;
1033 		dev->rx_skb[i] = skb;
1034 	}
1035  skip:
1036 	/* Check if we need to change "Jumbo" bit in MR1 */
1037 	if (emac_has_feature(dev, EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE)) {
1038 		mr1_jumbo_bit_change = (new_mtu > ETH_DATA_LEN) ||
1039 				(dev->ndev->mtu > ETH_DATA_LEN);
1040 	} else {
1041 		mr1_jumbo_bit_change = (new_mtu > ETH_DATA_LEN) ^
1042 				(dev->ndev->mtu > ETH_DATA_LEN);
1043 	}
1044 
1045 	if (mr1_jumbo_bit_change) {
1046 		/* This is to prevent starting RX channel in emac_rx_enable() */
1047 		set_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1048 
1049 		dev->ndev->mtu = new_mtu;
1050 		emac_full_tx_reset(dev);
1051 	}
1052 
1053 	mal_set_rcbs(dev->mal, dev->mal_rx_chan, emac_rx_size(new_mtu));
1054  oom:
1055 	/* Restart RX */
1056 	clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1057 	dev->rx_slot = 0;
1058 	mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1059 	emac_rx_enable(dev);
1060 	emac_netif_start(dev);
1061 	mutex_unlock(&dev->link_lock);
1062 
1063 	return ret;
1064 }
1065 
1066 /* Process ctx, rtnl_lock semaphore */
1067 static int emac_change_mtu(struct net_device *ndev, int new_mtu)
1068 {
1069 	struct emac_instance *dev = netdev_priv(ndev);
1070 	int ret = 0;
1071 
1072 	if (new_mtu < EMAC_MIN_MTU || new_mtu > dev->max_mtu)
1073 		return -EINVAL;
1074 
1075 	DBG(dev, "change_mtu(%d)" NL, new_mtu);
1076 
1077 	if (netif_running(ndev)) {
1078 		/* Check if we really need to reinitialize RX ring */
1079 		if (emac_rx_skb_size(ndev->mtu) != emac_rx_skb_size(new_mtu))
1080 			ret = emac_resize_rx_ring(dev, new_mtu);
1081 	}
1082 
1083 	if (!ret) {
1084 		ndev->mtu = new_mtu;
1085 		dev->rx_skb_size = emac_rx_skb_size(new_mtu);
1086 		dev->rx_sync_size = emac_rx_sync_size(new_mtu);
1087 	}
1088 
1089 	return ret;
1090 }
1091 
1092 static void emac_clean_tx_ring(struct emac_instance *dev)
1093 {
1094 	int i;
1095 
1096 	for (i = 0; i < NUM_TX_BUFF; ++i) {
1097 		if (dev->tx_skb[i]) {
1098 			dev_kfree_skb(dev->tx_skb[i]);
1099 			dev->tx_skb[i] = NULL;
1100 			if (dev->tx_desc[i].ctrl & MAL_TX_CTRL_READY)
1101 				++dev->estats.tx_dropped;
1102 		}
1103 		dev->tx_desc[i].ctrl = 0;
1104 		dev->tx_desc[i].data_ptr = 0;
1105 	}
1106 }
1107 
1108 static void emac_clean_rx_ring(struct emac_instance *dev)
1109 {
1110 	int i;
1111 
1112 	for (i = 0; i < NUM_RX_BUFF; ++i)
1113 		if (dev->rx_skb[i]) {
1114 			dev->rx_desc[i].ctrl = 0;
1115 			dev_kfree_skb(dev->rx_skb[i]);
1116 			dev->rx_skb[i] = NULL;
1117 			dev->rx_desc[i].data_ptr = 0;
1118 		}
1119 
1120 	if (dev->rx_sg_skb) {
1121 		dev_kfree_skb(dev->rx_sg_skb);
1122 		dev->rx_sg_skb = NULL;
1123 	}
1124 }
1125 
1126 static inline int emac_alloc_rx_skb(struct emac_instance *dev, int slot,
1127 				    gfp_t flags)
1128 {
1129 	struct sk_buff *skb = alloc_skb(dev->rx_skb_size, flags);
1130 	if (unlikely(!skb))
1131 		return -ENOMEM;
1132 
1133 	dev->rx_skb[slot] = skb;
1134 	dev->rx_desc[slot].data_len = 0;
1135 
1136 	skb_reserve(skb, EMAC_RX_SKB_HEADROOM + 2);
1137 	dev->rx_desc[slot].data_ptr =
1138 	    dma_map_single(&dev->ofdev->dev, skb->data - 2, dev->rx_sync_size,
1139 			   DMA_FROM_DEVICE) + 2;
1140 	wmb();
1141 	dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1142 	    (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1143 
1144 	return 0;
1145 }
1146 
1147 static void emac_print_link_status(struct emac_instance *dev)
1148 {
1149 	if (netif_carrier_ok(dev->ndev))
1150 		printk(KERN_INFO "%s: link is up, %d %s%s\n",
1151 		       dev->ndev->name, dev->phy.speed,
1152 		       dev->phy.duplex == DUPLEX_FULL ? "FDX" : "HDX",
1153 		       dev->phy.pause ? ", pause enabled" :
1154 		       dev->phy.asym_pause ? ", asymmetric pause enabled" : "");
1155 	else
1156 		printk(KERN_INFO "%s: link is down\n", dev->ndev->name);
1157 }
1158 
1159 /* Process ctx, rtnl_lock semaphore */
1160 static int emac_open(struct net_device *ndev)
1161 {
1162 	struct emac_instance *dev = netdev_priv(ndev);
1163 	int err, i;
1164 
1165 	DBG(dev, "open" NL);
1166 
1167 	/* Setup error IRQ handler */
1168 	err = request_irq(dev->emac_irq, emac_irq, 0, "EMAC", dev);
1169 	if (err) {
1170 		printk(KERN_ERR "%s: failed to request IRQ %d\n",
1171 		       ndev->name, dev->emac_irq);
1172 		return err;
1173 	}
1174 
1175 	/* Allocate RX ring */
1176 	for (i = 0; i < NUM_RX_BUFF; ++i)
1177 		if (emac_alloc_rx_skb(dev, i, GFP_KERNEL)) {
1178 			printk(KERN_ERR "%s: failed to allocate RX ring\n",
1179 			       ndev->name);
1180 			goto oom;
1181 		}
1182 
1183 	dev->tx_cnt = dev->tx_slot = dev->ack_slot = dev->rx_slot = 0;
1184 	clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1185 	dev->rx_sg_skb = NULL;
1186 
1187 	mutex_lock(&dev->link_lock);
1188 	dev->opened = 1;
1189 
1190 	/* Start PHY polling now.
1191 	 */
1192 	if (dev->phy.address >= 0) {
1193 		int link_poll_interval;
1194 		if (dev->phy.def->ops->poll_link(&dev->phy)) {
1195 			dev->phy.def->ops->read_link(&dev->phy);
1196 			emac_rx_clk_default(dev);
1197 			netif_carrier_on(dev->ndev);
1198 			link_poll_interval = PHY_POLL_LINK_ON;
1199 		} else {
1200 			emac_rx_clk_tx(dev);
1201 			netif_carrier_off(dev->ndev);
1202 			link_poll_interval = PHY_POLL_LINK_OFF;
1203 		}
1204 		dev->link_polling = 1;
1205 		wmb();
1206 		schedule_delayed_work(&dev->link_work, link_poll_interval);
1207 		emac_print_link_status(dev);
1208 	} else
1209 		netif_carrier_on(dev->ndev);
1210 
1211 	/* Required for Pause packet support in EMAC */
1212 	dev_mc_add_global(ndev, default_mcast_addr);
1213 
1214 	emac_configure(dev);
1215 	mal_poll_add(dev->mal, &dev->commac);
1216 	mal_enable_tx_channel(dev->mal, dev->mal_tx_chan);
1217 	mal_set_rcbs(dev->mal, dev->mal_rx_chan, emac_rx_size(ndev->mtu));
1218 	mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1219 	emac_tx_enable(dev);
1220 	emac_rx_enable(dev);
1221 	emac_netif_start(dev);
1222 
1223 	mutex_unlock(&dev->link_lock);
1224 
1225 	return 0;
1226  oom:
1227 	emac_clean_rx_ring(dev);
1228 	free_irq(dev->emac_irq, dev);
1229 
1230 	return -ENOMEM;
1231 }
1232 
1233 /* BHs disabled */
1234 #if 0
1235 static int emac_link_differs(struct emac_instance *dev)
1236 {
1237 	u32 r = in_be32(&dev->emacp->mr1);
1238 
1239 	int duplex = r & EMAC_MR1_FDE ? DUPLEX_FULL : DUPLEX_HALF;
1240 	int speed, pause, asym_pause;
1241 
1242 	if (r & EMAC_MR1_MF_1000)
1243 		speed = SPEED_1000;
1244 	else if (r & EMAC_MR1_MF_100)
1245 		speed = SPEED_100;
1246 	else
1247 		speed = SPEED_10;
1248 
1249 	switch (r & (EMAC_MR1_EIFC | EMAC_MR1_APP)) {
1250 	case (EMAC_MR1_EIFC | EMAC_MR1_APP):
1251 		pause = 1;
1252 		asym_pause = 0;
1253 		break;
1254 	case EMAC_MR1_APP:
1255 		pause = 0;
1256 		asym_pause = 1;
1257 		break;
1258 	default:
1259 		pause = asym_pause = 0;
1260 	}
1261 	return speed != dev->phy.speed || duplex != dev->phy.duplex ||
1262 	    pause != dev->phy.pause || asym_pause != dev->phy.asym_pause;
1263 }
1264 #endif
1265 
1266 static void emac_link_timer(struct work_struct *work)
1267 {
1268 	struct emac_instance *dev =
1269 		container_of(to_delayed_work(work),
1270 			     struct emac_instance, link_work);
1271 	int link_poll_interval;
1272 
1273 	mutex_lock(&dev->link_lock);
1274 	DBG2(dev, "link timer" NL);
1275 
1276 	if (!dev->opened)
1277 		goto bail;
1278 
1279 	if (dev->phy.def->ops->poll_link(&dev->phy)) {
1280 		if (!netif_carrier_ok(dev->ndev)) {
1281 			emac_rx_clk_default(dev);
1282 			/* Get new link parameters */
1283 			dev->phy.def->ops->read_link(&dev->phy);
1284 
1285 			netif_carrier_on(dev->ndev);
1286 			emac_netif_stop(dev);
1287 			emac_full_tx_reset(dev);
1288 			emac_netif_start(dev);
1289 			emac_print_link_status(dev);
1290 		}
1291 		link_poll_interval = PHY_POLL_LINK_ON;
1292 	} else {
1293 		if (netif_carrier_ok(dev->ndev)) {
1294 			emac_rx_clk_tx(dev);
1295 			netif_carrier_off(dev->ndev);
1296 			netif_tx_disable(dev->ndev);
1297 			emac_reinitialize(dev);
1298 			emac_print_link_status(dev);
1299 		}
1300 		link_poll_interval = PHY_POLL_LINK_OFF;
1301 	}
1302 	schedule_delayed_work(&dev->link_work, link_poll_interval);
1303  bail:
1304 	mutex_unlock(&dev->link_lock);
1305 }
1306 
1307 static void emac_force_link_update(struct emac_instance *dev)
1308 {
1309 	netif_carrier_off(dev->ndev);
1310 	smp_rmb();
1311 	if (dev->link_polling) {
1312 		cancel_delayed_work_sync(&dev->link_work);
1313 		if (dev->link_polling)
1314 			schedule_delayed_work(&dev->link_work,  PHY_POLL_LINK_OFF);
1315 	}
1316 }
1317 
1318 /* Process ctx, rtnl_lock semaphore */
1319 static int emac_close(struct net_device *ndev)
1320 {
1321 	struct emac_instance *dev = netdev_priv(ndev);
1322 
1323 	DBG(dev, "close" NL);
1324 
1325 	if (dev->phy.address >= 0) {
1326 		dev->link_polling = 0;
1327 		cancel_delayed_work_sync(&dev->link_work);
1328 	}
1329 	mutex_lock(&dev->link_lock);
1330 	emac_netif_stop(dev);
1331 	dev->opened = 0;
1332 	mutex_unlock(&dev->link_lock);
1333 
1334 	emac_rx_disable(dev);
1335 	emac_tx_disable(dev);
1336 	mal_disable_rx_channel(dev->mal, dev->mal_rx_chan);
1337 	mal_disable_tx_channel(dev->mal, dev->mal_tx_chan);
1338 	mal_poll_del(dev->mal, &dev->commac);
1339 
1340 	emac_clean_tx_ring(dev);
1341 	emac_clean_rx_ring(dev);
1342 
1343 	free_irq(dev->emac_irq, dev);
1344 
1345 	netif_carrier_off(ndev);
1346 
1347 	return 0;
1348 }
1349 
1350 static inline u16 emac_tx_csum(struct emac_instance *dev,
1351 			       struct sk_buff *skb)
1352 {
1353 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH) &&
1354 		(skb->ip_summed == CHECKSUM_PARTIAL)) {
1355 		++dev->stats.tx_packets_csum;
1356 		return EMAC_TX_CTRL_TAH_CSUM;
1357 	}
1358 	return 0;
1359 }
1360 
1361 static inline int emac_xmit_finish(struct emac_instance *dev, int len)
1362 {
1363 	struct emac_regs __iomem *p = dev->emacp;
1364 	struct net_device *ndev = dev->ndev;
1365 
1366 	/* Send the packet out. If the if makes a significant perf
1367 	 * difference, then we can store the TMR0 value in "dev"
1368 	 * instead
1369 	 */
1370 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
1371 		out_be32(&p->tmr0, EMAC4_TMR0_XMIT);
1372 	else
1373 		out_be32(&p->tmr0, EMAC_TMR0_XMIT);
1374 
1375 	if (unlikely(++dev->tx_cnt == NUM_TX_BUFF)) {
1376 		netif_stop_queue(ndev);
1377 		DBG2(dev, "stopped TX queue" NL);
1378 	}
1379 
1380 	netif_trans_update(ndev);
1381 	++dev->stats.tx_packets;
1382 	dev->stats.tx_bytes += len;
1383 
1384 	return NETDEV_TX_OK;
1385 }
1386 
1387 /* Tx lock BH */
1388 static int emac_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1389 {
1390 	struct emac_instance *dev = netdev_priv(ndev);
1391 	unsigned int len = skb->len;
1392 	int slot;
1393 
1394 	u16 ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1395 	    MAL_TX_CTRL_LAST | emac_tx_csum(dev, skb);
1396 
1397 	slot = dev->tx_slot++;
1398 	if (dev->tx_slot == NUM_TX_BUFF) {
1399 		dev->tx_slot = 0;
1400 		ctrl |= MAL_TX_CTRL_WRAP;
1401 	}
1402 
1403 	DBG2(dev, "xmit(%u) %d" NL, len, slot);
1404 
1405 	dev->tx_skb[slot] = skb;
1406 	dev->tx_desc[slot].data_ptr = dma_map_single(&dev->ofdev->dev,
1407 						     skb->data, len,
1408 						     DMA_TO_DEVICE);
1409 	dev->tx_desc[slot].data_len = (u16) len;
1410 	wmb();
1411 	dev->tx_desc[slot].ctrl = ctrl;
1412 
1413 	return emac_xmit_finish(dev, len);
1414 }
1415 
1416 static inline int emac_xmit_split(struct emac_instance *dev, int slot,
1417 				  u32 pd, int len, int last, u16 base_ctrl)
1418 {
1419 	while (1) {
1420 		u16 ctrl = base_ctrl;
1421 		int chunk = min(len, MAL_MAX_TX_SIZE);
1422 		len -= chunk;
1423 
1424 		slot = (slot + 1) % NUM_TX_BUFF;
1425 
1426 		if (last && !len)
1427 			ctrl |= MAL_TX_CTRL_LAST;
1428 		if (slot == NUM_TX_BUFF - 1)
1429 			ctrl |= MAL_TX_CTRL_WRAP;
1430 
1431 		dev->tx_skb[slot] = NULL;
1432 		dev->tx_desc[slot].data_ptr = pd;
1433 		dev->tx_desc[slot].data_len = (u16) chunk;
1434 		dev->tx_desc[slot].ctrl = ctrl;
1435 		++dev->tx_cnt;
1436 
1437 		if (!len)
1438 			break;
1439 
1440 		pd += chunk;
1441 	}
1442 	return slot;
1443 }
1444 
1445 /* Tx lock BH disabled (SG version for TAH equipped EMACs) */
1446 static int emac_start_xmit_sg(struct sk_buff *skb, struct net_device *ndev)
1447 {
1448 	struct emac_instance *dev = netdev_priv(ndev);
1449 	int nr_frags = skb_shinfo(skb)->nr_frags;
1450 	int len = skb->len, chunk;
1451 	int slot, i;
1452 	u16 ctrl;
1453 	u32 pd;
1454 
1455 	/* This is common "fast" path */
1456 	if (likely(!nr_frags && len <= MAL_MAX_TX_SIZE))
1457 		return emac_start_xmit(skb, ndev);
1458 
1459 	len -= skb->data_len;
1460 
1461 	/* Note, this is only an *estimation*, we can still run out of empty
1462 	 * slots because of the additional fragmentation into
1463 	 * MAL_MAX_TX_SIZE-sized chunks
1464 	 */
1465 	if (unlikely(dev->tx_cnt + nr_frags + mal_tx_chunks(len) > NUM_TX_BUFF))
1466 		goto stop_queue;
1467 
1468 	ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1469 	    emac_tx_csum(dev, skb);
1470 	slot = dev->tx_slot;
1471 
1472 	/* skb data */
1473 	dev->tx_skb[slot] = NULL;
1474 	chunk = min(len, MAL_MAX_TX_SIZE);
1475 	dev->tx_desc[slot].data_ptr = pd =
1476 	    dma_map_single(&dev->ofdev->dev, skb->data, len, DMA_TO_DEVICE);
1477 	dev->tx_desc[slot].data_len = (u16) chunk;
1478 	len -= chunk;
1479 	if (unlikely(len))
1480 		slot = emac_xmit_split(dev, slot, pd + chunk, len, !nr_frags,
1481 				       ctrl);
1482 	/* skb fragments */
1483 	for (i = 0; i < nr_frags; ++i) {
1484 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1485 		len = skb_frag_size(frag);
1486 
1487 		if (unlikely(dev->tx_cnt + mal_tx_chunks(len) >= NUM_TX_BUFF))
1488 			goto undo_frame;
1489 
1490 		pd = skb_frag_dma_map(&dev->ofdev->dev, frag, 0, len,
1491 				      DMA_TO_DEVICE);
1492 
1493 		slot = emac_xmit_split(dev, slot, pd, len, i == nr_frags - 1,
1494 				       ctrl);
1495 	}
1496 
1497 	DBG2(dev, "xmit_sg(%u) %d - %d" NL, skb->len, dev->tx_slot, slot);
1498 
1499 	/* Attach skb to the last slot so we don't release it too early */
1500 	dev->tx_skb[slot] = skb;
1501 
1502 	/* Send the packet out */
1503 	if (dev->tx_slot == NUM_TX_BUFF - 1)
1504 		ctrl |= MAL_TX_CTRL_WRAP;
1505 	wmb();
1506 	dev->tx_desc[dev->tx_slot].ctrl = ctrl;
1507 	dev->tx_slot = (slot + 1) % NUM_TX_BUFF;
1508 
1509 	return emac_xmit_finish(dev, skb->len);
1510 
1511  undo_frame:
1512 	/* Well, too bad. Our previous estimation was overly optimistic.
1513 	 * Undo everything.
1514 	 */
1515 	while (slot != dev->tx_slot) {
1516 		dev->tx_desc[slot].ctrl = 0;
1517 		--dev->tx_cnt;
1518 		if (--slot < 0)
1519 			slot = NUM_TX_BUFF - 1;
1520 	}
1521 	++dev->estats.tx_undo;
1522 
1523  stop_queue:
1524 	netif_stop_queue(ndev);
1525 	DBG2(dev, "stopped TX queue" NL);
1526 	return NETDEV_TX_BUSY;
1527 }
1528 
1529 /* Tx lock BHs */
1530 static void emac_parse_tx_error(struct emac_instance *dev, u16 ctrl)
1531 {
1532 	struct emac_error_stats *st = &dev->estats;
1533 
1534 	DBG(dev, "BD TX error %04x" NL, ctrl);
1535 
1536 	++st->tx_bd_errors;
1537 	if (ctrl & EMAC_TX_ST_BFCS)
1538 		++st->tx_bd_bad_fcs;
1539 	if (ctrl & EMAC_TX_ST_LCS)
1540 		++st->tx_bd_carrier_loss;
1541 	if (ctrl & EMAC_TX_ST_ED)
1542 		++st->tx_bd_excessive_deferral;
1543 	if (ctrl & EMAC_TX_ST_EC)
1544 		++st->tx_bd_excessive_collisions;
1545 	if (ctrl & EMAC_TX_ST_LC)
1546 		++st->tx_bd_late_collision;
1547 	if (ctrl & EMAC_TX_ST_MC)
1548 		++st->tx_bd_multple_collisions;
1549 	if (ctrl & EMAC_TX_ST_SC)
1550 		++st->tx_bd_single_collision;
1551 	if (ctrl & EMAC_TX_ST_UR)
1552 		++st->tx_bd_underrun;
1553 	if (ctrl & EMAC_TX_ST_SQE)
1554 		++st->tx_bd_sqe;
1555 }
1556 
1557 static void emac_poll_tx(void *param)
1558 {
1559 	struct emac_instance *dev = param;
1560 	u32 bad_mask;
1561 
1562 	DBG2(dev, "poll_tx, %d %d" NL, dev->tx_cnt, dev->ack_slot);
1563 
1564 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
1565 		bad_mask = EMAC_IS_BAD_TX_TAH;
1566 	else
1567 		bad_mask = EMAC_IS_BAD_TX;
1568 
1569 	netif_tx_lock_bh(dev->ndev);
1570 	if (dev->tx_cnt) {
1571 		u16 ctrl;
1572 		int slot = dev->ack_slot, n = 0;
1573 	again:
1574 		ctrl = dev->tx_desc[slot].ctrl;
1575 		if (!(ctrl & MAL_TX_CTRL_READY)) {
1576 			struct sk_buff *skb = dev->tx_skb[slot];
1577 			++n;
1578 
1579 			if (skb) {
1580 				dev_kfree_skb(skb);
1581 				dev->tx_skb[slot] = NULL;
1582 			}
1583 			slot = (slot + 1) % NUM_TX_BUFF;
1584 
1585 			if (unlikely(ctrl & bad_mask))
1586 				emac_parse_tx_error(dev, ctrl);
1587 
1588 			if (--dev->tx_cnt)
1589 				goto again;
1590 		}
1591 		if (n) {
1592 			dev->ack_slot = slot;
1593 			if (netif_queue_stopped(dev->ndev) &&
1594 			    dev->tx_cnt < EMAC_TX_WAKEUP_THRESH)
1595 				netif_wake_queue(dev->ndev);
1596 
1597 			DBG2(dev, "tx %d pkts" NL, n);
1598 		}
1599 	}
1600 	netif_tx_unlock_bh(dev->ndev);
1601 }
1602 
1603 static inline void emac_recycle_rx_skb(struct emac_instance *dev, int slot,
1604 				       int len)
1605 {
1606 	struct sk_buff *skb = dev->rx_skb[slot];
1607 
1608 	DBG2(dev, "recycle %d %d" NL, slot, len);
1609 
1610 	if (len)
1611 		dma_map_single(&dev->ofdev->dev, skb->data - 2,
1612 			       EMAC_DMA_ALIGN(len + 2), DMA_FROM_DEVICE);
1613 
1614 	dev->rx_desc[slot].data_len = 0;
1615 	wmb();
1616 	dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1617 	    (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1618 }
1619 
1620 static void emac_parse_rx_error(struct emac_instance *dev, u16 ctrl)
1621 {
1622 	struct emac_error_stats *st = &dev->estats;
1623 
1624 	DBG(dev, "BD RX error %04x" NL, ctrl);
1625 
1626 	++st->rx_bd_errors;
1627 	if (ctrl & EMAC_RX_ST_OE)
1628 		++st->rx_bd_overrun;
1629 	if (ctrl & EMAC_RX_ST_BP)
1630 		++st->rx_bd_bad_packet;
1631 	if (ctrl & EMAC_RX_ST_RP)
1632 		++st->rx_bd_runt_packet;
1633 	if (ctrl & EMAC_RX_ST_SE)
1634 		++st->rx_bd_short_event;
1635 	if (ctrl & EMAC_RX_ST_AE)
1636 		++st->rx_bd_alignment_error;
1637 	if (ctrl & EMAC_RX_ST_BFCS)
1638 		++st->rx_bd_bad_fcs;
1639 	if (ctrl & EMAC_RX_ST_PTL)
1640 		++st->rx_bd_packet_too_long;
1641 	if (ctrl & EMAC_RX_ST_ORE)
1642 		++st->rx_bd_out_of_range;
1643 	if (ctrl & EMAC_RX_ST_IRE)
1644 		++st->rx_bd_in_range;
1645 }
1646 
1647 static inline void emac_rx_csum(struct emac_instance *dev,
1648 				struct sk_buff *skb, u16 ctrl)
1649 {
1650 #ifdef CONFIG_IBM_EMAC_TAH
1651 	if (!ctrl && dev->tah_dev) {
1652 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1653 		++dev->stats.rx_packets_csum;
1654 	}
1655 #endif
1656 }
1657 
1658 static inline int emac_rx_sg_append(struct emac_instance *dev, int slot)
1659 {
1660 	if (likely(dev->rx_sg_skb != NULL)) {
1661 		int len = dev->rx_desc[slot].data_len;
1662 		int tot_len = dev->rx_sg_skb->len + len;
1663 
1664 		if (unlikely(tot_len + 2 > dev->rx_skb_size)) {
1665 			++dev->estats.rx_dropped_mtu;
1666 			dev_kfree_skb(dev->rx_sg_skb);
1667 			dev->rx_sg_skb = NULL;
1668 		} else {
1669 			memcpy(skb_tail_pointer(dev->rx_sg_skb),
1670 					 dev->rx_skb[slot]->data, len);
1671 			skb_put(dev->rx_sg_skb, len);
1672 			emac_recycle_rx_skb(dev, slot, len);
1673 			return 0;
1674 		}
1675 	}
1676 	emac_recycle_rx_skb(dev, slot, 0);
1677 	return -1;
1678 }
1679 
1680 /* NAPI poll context */
1681 static int emac_poll_rx(void *param, int budget)
1682 {
1683 	struct emac_instance *dev = param;
1684 	int slot = dev->rx_slot, received = 0;
1685 
1686 	DBG2(dev, "poll_rx(%d)" NL, budget);
1687 
1688  again:
1689 	while (budget > 0) {
1690 		int len;
1691 		struct sk_buff *skb;
1692 		u16 ctrl = dev->rx_desc[slot].ctrl;
1693 
1694 		if (ctrl & MAL_RX_CTRL_EMPTY)
1695 			break;
1696 
1697 		skb = dev->rx_skb[slot];
1698 		mb();
1699 		len = dev->rx_desc[slot].data_len;
1700 
1701 		if (unlikely(!MAL_IS_SINGLE_RX(ctrl)))
1702 			goto sg;
1703 
1704 		ctrl &= EMAC_BAD_RX_MASK;
1705 		if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1706 			emac_parse_rx_error(dev, ctrl);
1707 			++dev->estats.rx_dropped_error;
1708 			emac_recycle_rx_skb(dev, slot, 0);
1709 			len = 0;
1710 			goto next;
1711 		}
1712 
1713 		if (len < ETH_HLEN) {
1714 			++dev->estats.rx_dropped_stack;
1715 			emac_recycle_rx_skb(dev, slot, len);
1716 			goto next;
1717 		}
1718 
1719 		if (len && len < EMAC_RX_COPY_THRESH) {
1720 			struct sk_buff *copy_skb =
1721 			    alloc_skb(len + EMAC_RX_SKB_HEADROOM + 2, GFP_ATOMIC);
1722 			if (unlikely(!copy_skb))
1723 				goto oom;
1724 
1725 			skb_reserve(copy_skb, EMAC_RX_SKB_HEADROOM + 2);
1726 			memcpy(copy_skb->data - 2, skb->data - 2, len + 2);
1727 			emac_recycle_rx_skb(dev, slot, len);
1728 			skb = copy_skb;
1729 		} else if (unlikely(emac_alloc_rx_skb(dev, slot, GFP_ATOMIC)))
1730 			goto oom;
1731 
1732 		skb_put(skb, len);
1733 	push_packet:
1734 		skb->protocol = eth_type_trans(skb, dev->ndev);
1735 		emac_rx_csum(dev, skb, ctrl);
1736 
1737 		if (unlikely(netif_receive_skb(skb) == NET_RX_DROP))
1738 			++dev->estats.rx_dropped_stack;
1739 	next:
1740 		++dev->stats.rx_packets;
1741 	skip:
1742 		dev->stats.rx_bytes += len;
1743 		slot = (slot + 1) % NUM_RX_BUFF;
1744 		--budget;
1745 		++received;
1746 		continue;
1747 	sg:
1748 		if (ctrl & MAL_RX_CTRL_FIRST) {
1749 			BUG_ON(dev->rx_sg_skb);
1750 			if (unlikely(emac_alloc_rx_skb(dev, slot, GFP_ATOMIC))) {
1751 				DBG(dev, "rx OOM %d" NL, slot);
1752 				++dev->estats.rx_dropped_oom;
1753 				emac_recycle_rx_skb(dev, slot, 0);
1754 			} else {
1755 				dev->rx_sg_skb = skb;
1756 				skb_put(skb, len);
1757 			}
1758 		} else if (!emac_rx_sg_append(dev, slot) &&
1759 			   (ctrl & MAL_RX_CTRL_LAST)) {
1760 
1761 			skb = dev->rx_sg_skb;
1762 			dev->rx_sg_skb = NULL;
1763 
1764 			ctrl &= EMAC_BAD_RX_MASK;
1765 			if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1766 				emac_parse_rx_error(dev, ctrl);
1767 				++dev->estats.rx_dropped_error;
1768 				dev_kfree_skb(skb);
1769 				len = 0;
1770 			} else
1771 				goto push_packet;
1772 		}
1773 		goto skip;
1774 	oom:
1775 		DBG(dev, "rx OOM %d" NL, slot);
1776 		/* Drop the packet and recycle skb */
1777 		++dev->estats.rx_dropped_oom;
1778 		emac_recycle_rx_skb(dev, slot, 0);
1779 		goto next;
1780 	}
1781 
1782 	if (received) {
1783 		DBG2(dev, "rx %d BDs" NL, received);
1784 		dev->rx_slot = slot;
1785 	}
1786 
1787 	if (unlikely(budget && test_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags))) {
1788 		mb();
1789 		if (!(dev->rx_desc[slot].ctrl & MAL_RX_CTRL_EMPTY)) {
1790 			DBG2(dev, "rx restart" NL);
1791 			received = 0;
1792 			goto again;
1793 		}
1794 
1795 		if (dev->rx_sg_skb) {
1796 			DBG2(dev, "dropping partial rx packet" NL);
1797 			++dev->estats.rx_dropped_error;
1798 			dev_kfree_skb(dev->rx_sg_skb);
1799 			dev->rx_sg_skb = NULL;
1800 		}
1801 
1802 		clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1803 		mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1804 		emac_rx_enable(dev);
1805 		dev->rx_slot = 0;
1806 	}
1807 	return received;
1808 }
1809 
1810 /* NAPI poll context */
1811 static int emac_peek_rx(void *param)
1812 {
1813 	struct emac_instance *dev = param;
1814 
1815 	return !(dev->rx_desc[dev->rx_slot].ctrl & MAL_RX_CTRL_EMPTY);
1816 }
1817 
1818 /* NAPI poll context */
1819 static int emac_peek_rx_sg(void *param)
1820 {
1821 	struct emac_instance *dev = param;
1822 
1823 	int slot = dev->rx_slot;
1824 	while (1) {
1825 		u16 ctrl = dev->rx_desc[slot].ctrl;
1826 		if (ctrl & MAL_RX_CTRL_EMPTY)
1827 			return 0;
1828 		else if (ctrl & MAL_RX_CTRL_LAST)
1829 			return 1;
1830 
1831 		slot = (slot + 1) % NUM_RX_BUFF;
1832 
1833 		/* I'm just being paranoid here :) */
1834 		if (unlikely(slot == dev->rx_slot))
1835 			return 0;
1836 	}
1837 }
1838 
1839 /* Hard IRQ */
1840 static void emac_rxde(void *param)
1841 {
1842 	struct emac_instance *dev = param;
1843 
1844 	++dev->estats.rx_stopped;
1845 	emac_rx_disable_async(dev);
1846 }
1847 
1848 /* Hard IRQ */
1849 static irqreturn_t emac_irq(int irq, void *dev_instance)
1850 {
1851 	struct emac_instance *dev = dev_instance;
1852 	struct emac_regs __iomem *p = dev->emacp;
1853 	struct emac_error_stats *st = &dev->estats;
1854 	u32 isr;
1855 
1856 	spin_lock(&dev->lock);
1857 
1858 	isr = in_be32(&p->isr);
1859 	out_be32(&p->isr, isr);
1860 
1861 	DBG(dev, "isr = %08x" NL, isr);
1862 
1863 	if (isr & EMAC4_ISR_TXPE)
1864 		++st->tx_parity;
1865 	if (isr & EMAC4_ISR_RXPE)
1866 		++st->rx_parity;
1867 	if (isr & EMAC4_ISR_TXUE)
1868 		++st->tx_underrun;
1869 	if (isr & EMAC4_ISR_RXOE)
1870 		++st->rx_fifo_overrun;
1871 	if (isr & EMAC_ISR_OVR)
1872 		++st->rx_overrun;
1873 	if (isr & EMAC_ISR_BP)
1874 		++st->rx_bad_packet;
1875 	if (isr & EMAC_ISR_RP)
1876 		++st->rx_runt_packet;
1877 	if (isr & EMAC_ISR_SE)
1878 		++st->rx_short_event;
1879 	if (isr & EMAC_ISR_ALE)
1880 		++st->rx_alignment_error;
1881 	if (isr & EMAC_ISR_BFCS)
1882 		++st->rx_bad_fcs;
1883 	if (isr & EMAC_ISR_PTLE)
1884 		++st->rx_packet_too_long;
1885 	if (isr & EMAC_ISR_ORE)
1886 		++st->rx_out_of_range;
1887 	if (isr & EMAC_ISR_IRE)
1888 		++st->rx_in_range;
1889 	if (isr & EMAC_ISR_SQE)
1890 		++st->tx_sqe;
1891 	if (isr & EMAC_ISR_TE)
1892 		++st->tx_errors;
1893 
1894 	spin_unlock(&dev->lock);
1895 
1896 	return IRQ_HANDLED;
1897 }
1898 
1899 static struct net_device_stats *emac_stats(struct net_device *ndev)
1900 {
1901 	struct emac_instance *dev = netdev_priv(ndev);
1902 	struct emac_stats *st = &dev->stats;
1903 	struct emac_error_stats *est = &dev->estats;
1904 	struct net_device_stats *nst = &dev->nstats;
1905 	unsigned long flags;
1906 
1907 	DBG2(dev, "stats" NL);
1908 
1909 	/* Compute "legacy" statistics */
1910 	spin_lock_irqsave(&dev->lock, flags);
1911 	nst->rx_packets = (unsigned long)st->rx_packets;
1912 	nst->rx_bytes = (unsigned long)st->rx_bytes;
1913 	nst->tx_packets = (unsigned long)st->tx_packets;
1914 	nst->tx_bytes = (unsigned long)st->tx_bytes;
1915 	nst->rx_dropped = (unsigned long)(est->rx_dropped_oom +
1916 					  est->rx_dropped_error +
1917 					  est->rx_dropped_resize +
1918 					  est->rx_dropped_mtu);
1919 	nst->tx_dropped = (unsigned long)est->tx_dropped;
1920 
1921 	nst->rx_errors = (unsigned long)est->rx_bd_errors;
1922 	nst->rx_fifo_errors = (unsigned long)(est->rx_bd_overrun +
1923 					      est->rx_fifo_overrun +
1924 					      est->rx_overrun);
1925 	nst->rx_frame_errors = (unsigned long)(est->rx_bd_alignment_error +
1926 					       est->rx_alignment_error);
1927 	nst->rx_crc_errors = (unsigned long)(est->rx_bd_bad_fcs +
1928 					     est->rx_bad_fcs);
1929 	nst->rx_length_errors = (unsigned long)(est->rx_bd_runt_packet +
1930 						est->rx_bd_short_event +
1931 						est->rx_bd_packet_too_long +
1932 						est->rx_bd_out_of_range +
1933 						est->rx_bd_in_range +
1934 						est->rx_runt_packet +
1935 						est->rx_short_event +
1936 						est->rx_packet_too_long +
1937 						est->rx_out_of_range +
1938 						est->rx_in_range);
1939 
1940 	nst->tx_errors = (unsigned long)(est->tx_bd_errors + est->tx_errors);
1941 	nst->tx_fifo_errors = (unsigned long)(est->tx_bd_underrun +
1942 					      est->tx_underrun);
1943 	nst->tx_carrier_errors = (unsigned long)est->tx_bd_carrier_loss;
1944 	nst->collisions = (unsigned long)(est->tx_bd_excessive_deferral +
1945 					  est->tx_bd_excessive_collisions +
1946 					  est->tx_bd_late_collision +
1947 					  est->tx_bd_multple_collisions);
1948 	spin_unlock_irqrestore(&dev->lock, flags);
1949 	return nst;
1950 }
1951 
1952 static struct mal_commac_ops emac_commac_ops = {
1953 	.poll_tx = &emac_poll_tx,
1954 	.poll_rx = &emac_poll_rx,
1955 	.peek_rx = &emac_peek_rx,
1956 	.rxde = &emac_rxde,
1957 };
1958 
1959 static struct mal_commac_ops emac_commac_sg_ops = {
1960 	.poll_tx = &emac_poll_tx,
1961 	.poll_rx = &emac_poll_rx,
1962 	.peek_rx = &emac_peek_rx_sg,
1963 	.rxde = &emac_rxde,
1964 };
1965 
1966 /* Ethtool support */
1967 static int emac_ethtool_get_settings(struct net_device *ndev,
1968 				     struct ethtool_cmd *cmd)
1969 {
1970 	struct emac_instance *dev = netdev_priv(ndev);
1971 
1972 	cmd->supported = dev->phy.features;
1973 	cmd->port = PORT_MII;
1974 	cmd->phy_address = dev->phy.address;
1975 	cmd->transceiver =
1976 	    dev->phy.address >= 0 ? XCVR_EXTERNAL : XCVR_INTERNAL;
1977 
1978 	mutex_lock(&dev->link_lock);
1979 	cmd->advertising = dev->phy.advertising;
1980 	cmd->autoneg = dev->phy.autoneg;
1981 	cmd->speed = dev->phy.speed;
1982 	cmd->duplex = dev->phy.duplex;
1983 	mutex_unlock(&dev->link_lock);
1984 
1985 	return 0;
1986 }
1987 
1988 static int emac_ethtool_set_settings(struct net_device *ndev,
1989 				     struct ethtool_cmd *cmd)
1990 {
1991 	struct emac_instance *dev = netdev_priv(ndev);
1992 	u32 f = dev->phy.features;
1993 
1994 	DBG(dev, "set_settings(%d, %d, %d, 0x%08x)" NL,
1995 	    cmd->autoneg, cmd->speed, cmd->duplex, cmd->advertising);
1996 
1997 	/* Basic sanity checks */
1998 	if (dev->phy.address < 0)
1999 		return -EOPNOTSUPP;
2000 	if (cmd->autoneg != AUTONEG_ENABLE && cmd->autoneg != AUTONEG_DISABLE)
2001 		return -EINVAL;
2002 	if (cmd->autoneg == AUTONEG_ENABLE && cmd->advertising == 0)
2003 		return -EINVAL;
2004 	if (cmd->duplex != DUPLEX_HALF && cmd->duplex != DUPLEX_FULL)
2005 		return -EINVAL;
2006 
2007 	if (cmd->autoneg == AUTONEG_DISABLE) {
2008 		switch (cmd->speed) {
2009 		case SPEED_10:
2010 			if (cmd->duplex == DUPLEX_HALF &&
2011 			    !(f & SUPPORTED_10baseT_Half))
2012 				return -EINVAL;
2013 			if (cmd->duplex == DUPLEX_FULL &&
2014 			    !(f & SUPPORTED_10baseT_Full))
2015 				return -EINVAL;
2016 			break;
2017 		case SPEED_100:
2018 			if (cmd->duplex == DUPLEX_HALF &&
2019 			    !(f & SUPPORTED_100baseT_Half))
2020 				return -EINVAL;
2021 			if (cmd->duplex == DUPLEX_FULL &&
2022 			    !(f & SUPPORTED_100baseT_Full))
2023 				return -EINVAL;
2024 			break;
2025 		case SPEED_1000:
2026 			if (cmd->duplex == DUPLEX_HALF &&
2027 			    !(f & SUPPORTED_1000baseT_Half))
2028 				return -EINVAL;
2029 			if (cmd->duplex == DUPLEX_FULL &&
2030 			    !(f & SUPPORTED_1000baseT_Full))
2031 				return -EINVAL;
2032 			break;
2033 		default:
2034 			return -EINVAL;
2035 		}
2036 
2037 		mutex_lock(&dev->link_lock);
2038 		dev->phy.def->ops->setup_forced(&dev->phy, cmd->speed,
2039 						cmd->duplex);
2040 		mutex_unlock(&dev->link_lock);
2041 
2042 	} else {
2043 		if (!(f & SUPPORTED_Autoneg))
2044 			return -EINVAL;
2045 
2046 		mutex_lock(&dev->link_lock);
2047 		dev->phy.def->ops->setup_aneg(&dev->phy,
2048 					      (cmd->advertising & f) |
2049 					      (dev->phy.advertising &
2050 					       (ADVERTISED_Pause |
2051 						ADVERTISED_Asym_Pause)));
2052 		mutex_unlock(&dev->link_lock);
2053 	}
2054 	emac_force_link_update(dev);
2055 
2056 	return 0;
2057 }
2058 
2059 static void emac_ethtool_get_ringparam(struct net_device *ndev,
2060 				       struct ethtool_ringparam *rp)
2061 {
2062 	rp->rx_max_pending = rp->rx_pending = NUM_RX_BUFF;
2063 	rp->tx_max_pending = rp->tx_pending = NUM_TX_BUFF;
2064 }
2065 
2066 static void emac_ethtool_get_pauseparam(struct net_device *ndev,
2067 					struct ethtool_pauseparam *pp)
2068 {
2069 	struct emac_instance *dev = netdev_priv(ndev);
2070 
2071 	mutex_lock(&dev->link_lock);
2072 	if ((dev->phy.features & SUPPORTED_Autoneg) &&
2073 	    (dev->phy.advertising & (ADVERTISED_Pause | ADVERTISED_Asym_Pause)))
2074 		pp->autoneg = 1;
2075 
2076 	if (dev->phy.duplex == DUPLEX_FULL) {
2077 		if (dev->phy.pause)
2078 			pp->rx_pause = pp->tx_pause = 1;
2079 		else if (dev->phy.asym_pause)
2080 			pp->tx_pause = 1;
2081 	}
2082 	mutex_unlock(&dev->link_lock);
2083 }
2084 
2085 static int emac_get_regs_len(struct emac_instance *dev)
2086 {
2087 		return sizeof(struct emac_ethtool_regs_subhdr) +
2088 			sizeof(struct emac_regs);
2089 }
2090 
2091 static int emac_ethtool_get_regs_len(struct net_device *ndev)
2092 {
2093 	struct emac_instance *dev = netdev_priv(ndev);
2094 	int size;
2095 
2096 	size = sizeof(struct emac_ethtool_regs_hdr) +
2097 		emac_get_regs_len(dev) + mal_get_regs_len(dev->mal);
2098 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
2099 		size += zmii_get_regs_len(dev->zmii_dev);
2100 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
2101 		size += rgmii_get_regs_len(dev->rgmii_dev);
2102 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
2103 		size += tah_get_regs_len(dev->tah_dev);
2104 
2105 	return size;
2106 }
2107 
2108 static void *emac_dump_regs(struct emac_instance *dev, void *buf)
2109 {
2110 	struct emac_ethtool_regs_subhdr *hdr = buf;
2111 
2112 	hdr->index = dev->cell_index;
2113 	if (emac_has_feature(dev, EMAC_FTR_EMAC4SYNC)) {
2114 		hdr->version = EMAC4SYNC_ETHTOOL_REGS_VER;
2115 	} else if (emac_has_feature(dev, EMAC_FTR_EMAC4)) {
2116 		hdr->version = EMAC4_ETHTOOL_REGS_VER;
2117 	} else {
2118 		hdr->version = EMAC_ETHTOOL_REGS_VER;
2119 	}
2120 	memcpy_fromio(hdr + 1, dev->emacp, sizeof(struct emac_regs));
2121 	return (void *)(hdr + 1) + sizeof(struct emac_regs);
2122 }
2123 
2124 static void emac_ethtool_get_regs(struct net_device *ndev,
2125 				  struct ethtool_regs *regs, void *buf)
2126 {
2127 	struct emac_instance *dev = netdev_priv(ndev);
2128 	struct emac_ethtool_regs_hdr *hdr = buf;
2129 
2130 	hdr->components = 0;
2131 	buf = hdr + 1;
2132 
2133 	buf = mal_dump_regs(dev->mal, buf);
2134 	buf = emac_dump_regs(dev, buf);
2135 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII)) {
2136 		hdr->components |= EMAC_ETHTOOL_REGS_ZMII;
2137 		buf = zmii_dump_regs(dev->zmii_dev, buf);
2138 	}
2139 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII)) {
2140 		hdr->components |= EMAC_ETHTOOL_REGS_RGMII;
2141 		buf = rgmii_dump_regs(dev->rgmii_dev, buf);
2142 	}
2143 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH)) {
2144 		hdr->components |= EMAC_ETHTOOL_REGS_TAH;
2145 		buf = tah_dump_regs(dev->tah_dev, buf);
2146 	}
2147 }
2148 
2149 static int emac_ethtool_nway_reset(struct net_device *ndev)
2150 {
2151 	struct emac_instance *dev = netdev_priv(ndev);
2152 	int res = 0;
2153 
2154 	DBG(dev, "nway_reset" NL);
2155 
2156 	if (dev->phy.address < 0)
2157 		return -EOPNOTSUPP;
2158 
2159 	mutex_lock(&dev->link_lock);
2160 	if (!dev->phy.autoneg) {
2161 		res = -EINVAL;
2162 		goto out;
2163 	}
2164 
2165 	dev->phy.def->ops->setup_aneg(&dev->phy, dev->phy.advertising);
2166  out:
2167 	mutex_unlock(&dev->link_lock);
2168 	emac_force_link_update(dev);
2169 	return res;
2170 }
2171 
2172 static int emac_ethtool_get_sset_count(struct net_device *ndev, int stringset)
2173 {
2174 	if (stringset == ETH_SS_STATS)
2175 		return EMAC_ETHTOOL_STATS_COUNT;
2176 	else
2177 		return -EINVAL;
2178 }
2179 
2180 static void emac_ethtool_get_strings(struct net_device *ndev, u32 stringset,
2181 				     u8 * buf)
2182 {
2183 	if (stringset == ETH_SS_STATS)
2184 		memcpy(buf, &emac_stats_keys, sizeof(emac_stats_keys));
2185 }
2186 
2187 static void emac_ethtool_get_ethtool_stats(struct net_device *ndev,
2188 					   struct ethtool_stats *estats,
2189 					   u64 * tmp_stats)
2190 {
2191 	struct emac_instance *dev = netdev_priv(ndev);
2192 
2193 	memcpy(tmp_stats, &dev->stats, sizeof(dev->stats));
2194 	tmp_stats += sizeof(dev->stats) / sizeof(u64);
2195 	memcpy(tmp_stats, &dev->estats, sizeof(dev->estats));
2196 }
2197 
2198 static void emac_ethtool_get_drvinfo(struct net_device *ndev,
2199 				     struct ethtool_drvinfo *info)
2200 {
2201 	struct emac_instance *dev = netdev_priv(ndev);
2202 
2203 	strlcpy(info->driver, "ibm_emac", sizeof(info->driver));
2204 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2205 	snprintf(info->bus_info, sizeof(info->bus_info), "PPC 4xx EMAC-%d %s",
2206 		 dev->cell_index, dev->ofdev->dev.of_node->full_name);
2207 }
2208 
2209 static const struct ethtool_ops emac_ethtool_ops = {
2210 	.get_settings = emac_ethtool_get_settings,
2211 	.set_settings = emac_ethtool_set_settings,
2212 	.get_drvinfo = emac_ethtool_get_drvinfo,
2213 
2214 	.get_regs_len = emac_ethtool_get_regs_len,
2215 	.get_regs = emac_ethtool_get_regs,
2216 
2217 	.nway_reset = emac_ethtool_nway_reset,
2218 
2219 	.get_ringparam = emac_ethtool_get_ringparam,
2220 	.get_pauseparam = emac_ethtool_get_pauseparam,
2221 
2222 	.get_strings = emac_ethtool_get_strings,
2223 	.get_sset_count = emac_ethtool_get_sset_count,
2224 	.get_ethtool_stats = emac_ethtool_get_ethtool_stats,
2225 
2226 	.get_link = ethtool_op_get_link,
2227 };
2228 
2229 static int emac_ioctl(struct net_device *ndev, struct ifreq *rq, int cmd)
2230 {
2231 	struct emac_instance *dev = netdev_priv(ndev);
2232 	struct mii_ioctl_data *data = if_mii(rq);
2233 
2234 	DBG(dev, "ioctl %08x" NL, cmd);
2235 
2236 	if (dev->phy.address < 0)
2237 		return -EOPNOTSUPP;
2238 
2239 	switch (cmd) {
2240 	case SIOCGMIIPHY:
2241 		data->phy_id = dev->phy.address;
2242 		/* Fall through */
2243 	case SIOCGMIIREG:
2244 		data->val_out = emac_mdio_read(ndev, dev->phy.address,
2245 					       data->reg_num);
2246 		return 0;
2247 
2248 	case SIOCSMIIREG:
2249 		emac_mdio_write(ndev, dev->phy.address, data->reg_num,
2250 				data->val_in);
2251 		return 0;
2252 	default:
2253 		return -EOPNOTSUPP;
2254 	}
2255 }
2256 
2257 struct emac_depentry {
2258 	u32			phandle;
2259 	struct device_node	*node;
2260 	struct platform_device	*ofdev;
2261 	void			*drvdata;
2262 };
2263 
2264 #define	EMAC_DEP_MAL_IDX	0
2265 #define	EMAC_DEP_ZMII_IDX	1
2266 #define	EMAC_DEP_RGMII_IDX	2
2267 #define	EMAC_DEP_TAH_IDX	3
2268 #define	EMAC_DEP_MDIO_IDX	4
2269 #define	EMAC_DEP_PREV_IDX	5
2270 #define	EMAC_DEP_COUNT		6
2271 
2272 static int emac_check_deps(struct emac_instance *dev,
2273 			   struct emac_depentry *deps)
2274 {
2275 	int i, there = 0;
2276 	struct device_node *np;
2277 
2278 	for (i = 0; i < EMAC_DEP_COUNT; i++) {
2279 		/* no dependency on that item, allright */
2280 		if (deps[i].phandle == 0) {
2281 			there++;
2282 			continue;
2283 		}
2284 		/* special case for blist as the dependency might go away */
2285 		if (i == EMAC_DEP_PREV_IDX) {
2286 			np = *(dev->blist - 1);
2287 			if (np == NULL) {
2288 				deps[i].phandle = 0;
2289 				there++;
2290 				continue;
2291 			}
2292 			if (deps[i].node == NULL)
2293 				deps[i].node = of_node_get(np);
2294 		}
2295 		if (deps[i].node == NULL)
2296 			deps[i].node = of_find_node_by_phandle(deps[i].phandle);
2297 		if (deps[i].node == NULL)
2298 			continue;
2299 		if (deps[i].ofdev == NULL)
2300 			deps[i].ofdev = of_find_device_by_node(deps[i].node);
2301 		if (deps[i].ofdev == NULL)
2302 			continue;
2303 		if (deps[i].drvdata == NULL)
2304 			deps[i].drvdata = platform_get_drvdata(deps[i].ofdev);
2305 		if (deps[i].drvdata != NULL)
2306 			there++;
2307 	}
2308 	return there == EMAC_DEP_COUNT;
2309 }
2310 
2311 static void emac_put_deps(struct emac_instance *dev)
2312 {
2313 	of_dev_put(dev->mal_dev);
2314 	of_dev_put(dev->zmii_dev);
2315 	of_dev_put(dev->rgmii_dev);
2316 	of_dev_put(dev->mdio_dev);
2317 	of_dev_put(dev->tah_dev);
2318 }
2319 
2320 static int emac_of_bus_notify(struct notifier_block *nb, unsigned long action,
2321 			      void *data)
2322 {
2323 	/* We are only intereted in device addition */
2324 	if (action == BUS_NOTIFY_BOUND_DRIVER)
2325 		wake_up_all(&emac_probe_wait);
2326 	return 0;
2327 }
2328 
2329 static struct notifier_block emac_of_bus_notifier = {
2330 	.notifier_call = emac_of_bus_notify
2331 };
2332 
2333 static int emac_wait_deps(struct emac_instance *dev)
2334 {
2335 	struct emac_depentry deps[EMAC_DEP_COUNT];
2336 	int i, err;
2337 
2338 	memset(&deps, 0, sizeof(deps));
2339 
2340 	deps[EMAC_DEP_MAL_IDX].phandle = dev->mal_ph;
2341 	deps[EMAC_DEP_ZMII_IDX].phandle = dev->zmii_ph;
2342 	deps[EMAC_DEP_RGMII_IDX].phandle = dev->rgmii_ph;
2343 	if (dev->tah_ph)
2344 		deps[EMAC_DEP_TAH_IDX].phandle = dev->tah_ph;
2345 	if (dev->mdio_ph)
2346 		deps[EMAC_DEP_MDIO_IDX].phandle = dev->mdio_ph;
2347 	if (dev->blist && dev->blist > emac_boot_list)
2348 		deps[EMAC_DEP_PREV_IDX].phandle = 0xffffffffu;
2349 	bus_register_notifier(&platform_bus_type, &emac_of_bus_notifier);
2350 	wait_event_timeout(emac_probe_wait,
2351 			   emac_check_deps(dev, deps),
2352 			   EMAC_PROBE_DEP_TIMEOUT);
2353 	bus_unregister_notifier(&platform_bus_type, &emac_of_bus_notifier);
2354 	err = emac_check_deps(dev, deps) ? 0 : -ENODEV;
2355 	for (i = 0; i < EMAC_DEP_COUNT; i++) {
2356 		of_node_put(deps[i].node);
2357 		if (err)
2358 			of_dev_put(deps[i].ofdev);
2359 	}
2360 	if (err == 0) {
2361 		dev->mal_dev = deps[EMAC_DEP_MAL_IDX].ofdev;
2362 		dev->zmii_dev = deps[EMAC_DEP_ZMII_IDX].ofdev;
2363 		dev->rgmii_dev = deps[EMAC_DEP_RGMII_IDX].ofdev;
2364 		dev->tah_dev = deps[EMAC_DEP_TAH_IDX].ofdev;
2365 		dev->mdio_dev = deps[EMAC_DEP_MDIO_IDX].ofdev;
2366 	}
2367 	of_dev_put(deps[EMAC_DEP_PREV_IDX].ofdev);
2368 	return err;
2369 }
2370 
2371 static int emac_read_uint_prop(struct device_node *np, const char *name,
2372 			       u32 *val, int fatal)
2373 {
2374 	int len;
2375 	const u32 *prop = of_get_property(np, name, &len);
2376 	if (prop == NULL || len < sizeof(u32)) {
2377 		if (fatal)
2378 			printk(KERN_ERR "%s: missing %s property\n",
2379 			       np->full_name, name);
2380 		return -ENODEV;
2381 	}
2382 	*val = *prop;
2383 	return 0;
2384 }
2385 
2386 static int emac_init_phy(struct emac_instance *dev)
2387 {
2388 	struct device_node *np = dev->ofdev->dev.of_node;
2389 	struct net_device *ndev = dev->ndev;
2390 	u32 phy_map, adv;
2391 	int i;
2392 
2393 	dev->phy.dev = ndev;
2394 	dev->phy.mode = dev->phy_mode;
2395 
2396 	/* PHY-less configuration.
2397 	 * XXX I probably should move these settings to the dev tree
2398 	 */
2399 	if (dev->phy_address == 0xffffffff && dev->phy_map == 0xffffffff) {
2400 		emac_reset(dev);
2401 
2402 		/* PHY-less configuration.
2403 		 * XXX I probably should move these settings to the dev tree
2404 		 */
2405 		dev->phy.address = -1;
2406 		dev->phy.features = SUPPORTED_MII;
2407 		if (emac_phy_supports_gige(dev->phy_mode))
2408 			dev->phy.features |= SUPPORTED_1000baseT_Full;
2409 		else
2410 			dev->phy.features |= SUPPORTED_100baseT_Full;
2411 		dev->phy.pause = 1;
2412 
2413 		return 0;
2414 	}
2415 
2416 	mutex_lock(&emac_phy_map_lock);
2417 	phy_map = dev->phy_map | busy_phy_map;
2418 
2419 	DBG(dev, "PHY maps %08x %08x" NL, dev->phy_map, busy_phy_map);
2420 
2421 	dev->phy.mdio_read = emac_mdio_read;
2422 	dev->phy.mdio_write = emac_mdio_write;
2423 
2424 	/* Enable internal clock source */
2425 #ifdef CONFIG_PPC_DCR_NATIVE
2426 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2427 		dcri_clrset(SDR0, SDR0_MFR, 0, SDR0_MFR_ECS);
2428 #endif
2429 	/* PHY clock workaround */
2430 	emac_rx_clk_tx(dev);
2431 
2432 	/* Enable internal clock source on 440GX*/
2433 #ifdef CONFIG_PPC_DCR_NATIVE
2434 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2435 		dcri_clrset(SDR0, SDR0_MFR, 0, SDR0_MFR_ECS);
2436 #endif
2437 	/* Configure EMAC with defaults so we can at least use MDIO
2438 	 * This is needed mostly for 440GX
2439 	 */
2440 	if (emac_phy_gpcs(dev->phy.mode)) {
2441 		/* XXX
2442 		 * Make GPCS PHY address equal to EMAC index.
2443 		 * We probably should take into account busy_phy_map
2444 		 * and/or phy_map here.
2445 		 *
2446 		 * Note that the busy_phy_map is currently global
2447 		 * while it should probably be per-ASIC...
2448 		 */
2449 		dev->phy.gpcs_address = dev->gpcs_address;
2450 		if (dev->phy.gpcs_address == 0xffffffff)
2451 			dev->phy.address = dev->cell_index;
2452 	}
2453 
2454 	emac_configure(dev);
2455 
2456 	if (dev->phy_address != 0xffffffff)
2457 		phy_map = ~(1 << dev->phy_address);
2458 
2459 	for (i = 0; i < 0x20; phy_map >>= 1, ++i)
2460 		if (!(phy_map & 1)) {
2461 			int r;
2462 			busy_phy_map |= 1 << i;
2463 
2464 			/* Quick check if there is a PHY at the address */
2465 			r = emac_mdio_read(dev->ndev, i, MII_BMCR);
2466 			if (r == 0xffff || r < 0)
2467 				continue;
2468 			if (!emac_mii_phy_probe(&dev->phy, i))
2469 				break;
2470 		}
2471 
2472 	/* Enable external clock source */
2473 #ifdef CONFIG_PPC_DCR_NATIVE
2474 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2475 		dcri_clrset(SDR0, SDR0_MFR, SDR0_MFR_ECS, 0);
2476 #endif
2477 	mutex_unlock(&emac_phy_map_lock);
2478 	if (i == 0x20) {
2479 		printk(KERN_WARNING "%s: can't find PHY!\n", np->full_name);
2480 		return -ENXIO;
2481 	}
2482 
2483 	/* Init PHY */
2484 	if (dev->phy.def->ops->init)
2485 		dev->phy.def->ops->init(&dev->phy);
2486 
2487 	/* Disable any PHY features not supported by the platform */
2488 	dev->phy.def->features &= ~dev->phy_feat_exc;
2489 	dev->phy.features &= ~dev->phy_feat_exc;
2490 
2491 	/* Setup initial link parameters */
2492 	if (dev->phy.features & SUPPORTED_Autoneg) {
2493 		adv = dev->phy.features;
2494 		if (!emac_has_feature(dev, EMAC_FTR_NO_FLOW_CONTROL_40x))
2495 			adv |= ADVERTISED_Pause | ADVERTISED_Asym_Pause;
2496 		/* Restart autonegotiation */
2497 		dev->phy.def->ops->setup_aneg(&dev->phy, adv);
2498 	} else {
2499 		u32 f = dev->phy.def->features;
2500 		int speed = SPEED_10, fd = DUPLEX_HALF;
2501 
2502 		/* Select highest supported speed/duplex */
2503 		if (f & SUPPORTED_1000baseT_Full) {
2504 			speed = SPEED_1000;
2505 			fd = DUPLEX_FULL;
2506 		} else if (f & SUPPORTED_1000baseT_Half)
2507 			speed = SPEED_1000;
2508 		else if (f & SUPPORTED_100baseT_Full) {
2509 			speed = SPEED_100;
2510 			fd = DUPLEX_FULL;
2511 		} else if (f & SUPPORTED_100baseT_Half)
2512 			speed = SPEED_100;
2513 		else if (f & SUPPORTED_10baseT_Full)
2514 			fd = DUPLEX_FULL;
2515 
2516 		/* Force link parameters */
2517 		dev->phy.def->ops->setup_forced(&dev->phy, speed, fd);
2518 	}
2519 	return 0;
2520 }
2521 
2522 static int emac_init_config(struct emac_instance *dev)
2523 {
2524 	struct device_node *np = dev->ofdev->dev.of_node;
2525 	const void *p;
2526 
2527 	/* Read config from device-tree */
2528 	if (emac_read_uint_prop(np, "mal-device", &dev->mal_ph, 1))
2529 		return -ENXIO;
2530 	if (emac_read_uint_prop(np, "mal-tx-channel", &dev->mal_tx_chan, 1))
2531 		return -ENXIO;
2532 	if (emac_read_uint_prop(np, "mal-rx-channel", &dev->mal_rx_chan, 1))
2533 		return -ENXIO;
2534 	if (emac_read_uint_prop(np, "cell-index", &dev->cell_index, 1))
2535 		return -ENXIO;
2536 	if (emac_read_uint_prop(np, "max-frame-size", &dev->max_mtu, 0))
2537 		dev->max_mtu = 1500;
2538 	if (emac_read_uint_prop(np, "rx-fifo-size", &dev->rx_fifo_size, 0))
2539 		dev->rx_fifo_size = 2048;
2540 	if (emac_read_uint_prop(np, "tx-fifo-size", &dev->tx_fifo_size, 0))
2541 		dev->tx_fifo_size = 2048;
2542 	if (emac_read_uint_prop(np, "rx-fifo-size-gige", &dev->rx_fifo_size_gige, 0))
2543 		dev->rx_fifo_size_gige = dev->rx_fifo_size;
2544 	if (emac_read_uint_prop(np, "tx-fifo-size-gige", &dev->tx_fifo_size_gige, 0))
2545 		dev->tx_fifo_size_gige = dev->tx_fifo_size;
2546 	if (emac_read_uint_prop(np, "phy-address", &dev->phy_address, 0))
2547 		dev->phy_address = 0xffffffff;
2548 	if (emac_read_uint_prop(np, "phy-map", &dev->phy_map, 0))
2549 		dev->phy_map = 0xffffffff;
2550 	if (emac_read_uint_prop(np, "gpcs-address", &dev->gpcs_address, 0))
2551 		dev->gpcs_address = 0xffffffff;
2552 	if (emac_read_uint_prop(np->parent, "clock-frequency", &dev->opb_bus_freq, 1))
2553 		return -ENXIO;
2554 	if (emac_read_uint_prop(np, "tah-device", &dev->tah_ph, 0))
2555 		dev->tah_ph = 0;
2556 	if (emac_read_uint_prop(np, "tah-channel", &dev->tah_port, 0))
2557 		dev->tah_port = 0;
2558 	if (emac_read_uint_prop(np, "mdio-device", &dev->mdio_ph, 0))
2559 		dev->mdio_ph = 0;
2560 	if (emac_read_uint_prop(np, "zmii-device", &dev->zmii_ph, 0))
2561 		dev->zmii_ph = 0;
2562 	if (emac_read_uint_prop(np, "zmii-channel", &dev->zmii_port, 0))
2563 		dev->zmii_port = 0xffffffff;
2564 	if (emac_read_uint_prop(np, "rgmii-device", &dev->rgmii_ph, 0))
2565 		dev->rgmii_ph = 0;
2566 	if (emac_read_uint_prop(np, "rgmii-channel", &dev->rgmii_port, 0))
2567 		dev->rgmii_port = 0xffffffff;
2568 	if (emac_read_uint_prop(np, "fifo-entry-size", &dev->fifo_entry_size, 0))
2569 		dev->fifo_entry_size = 16;
2570 	if (emac_read_uint_prop(np, "mal-burst-size", &dev->mal_burst_size, 0))
2571 		dev->mal_burst_size = 256;
2572 
2573 	/* PHY mode needs some decoding */
2574 	dev->phy_mode = of_get_phy_mode(np);
2575 	if (dev->phy_mode < 0)
2576 		dev->phy_mode = PHY_MODE_NA;
2577 
2578 	/* Check EMAC version */
2579 	if (of_device_is_compatible(np, "ibm,emac4sync")) {
2580 		dev->features |= (EMAC_FTR_EMAC4 | EMAC_FTR_EMAC4SYNC);
2581 		if (of_device_is_compatible(np, "ibm,emac-460ex") ||
2582 		    of_device_is_compatible(np, "ibm,emac-460gt"))
2583 			dev->features |= EMAC_FTR_460EX_PHY_CLK_FIX;
2584 		if (of_device_is_compatible(np, "ibm,emac-405ex") ||
2585 		    of_device_is_compatible(np, "ibm,emac-405exr"))
2586 			dev->features |= EMAC_FTR_440EP_PHY_CLK_FIX;
2587 		if (of_device_is_compatible(np, "ibm,emac-apm821xx")) {
2588 			dev->features |= (EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE |
2589 					  EMAC_FTR_APM821XX_NO_HALF_DUPLEX |
2590 					  EMAC_FTR_460EX_PHY_CLK_FIX);
2591 		}
2592 	} else if (of_device_is_compatible(np, "ibm,emac4")) {
2593 		dev->features |= EMAC_FTR_EMAC4;
2594 		if (of_device_is_compatible(np, "ibm,emac-440gx"))
2595 			dev->features |= EMAC_FTR_440GX_PHY_CLK_FIX;
2596 	} else {
2597 		if (of_device_is_compatible(np, "ibm,emac-440ep") ||
2598 		    of_device_is_compatible(np, "ibm,emac-440gr"))
2599 			dev->features |= EMAC_FTR_440EP_PHY_CLK_FIX;
2600 		if (of_device_is_compatible(np, "ibm,emac-405ez")) {
2601 #ifdef CONFIG_IBM_EMAC_NO_FLOW_CTRL
2602 			dev->features |= EMAC_FTR_NO_FLOW_CONTROL_40x;
2603 #else
2604 			printk(KERN_ERR "%s: Flow control not disabled!\n",
2605 					np->full_name);
2606 			return -ENXIO;
2607 #endif
2608 		}
2609 
2610 	}
2611 
2612 	/* Fixup some feature bits based on the device tree */
2613 	if (of_get_property(np, "has-inverted-stacr-oc", NULL))
2614 		dev->features |= EMAC_FTR_STACR_OC_INVERT;
2615 	if (of_get_property(np, "has-new-stacr-staopc", NULL))
2616 		dev->features |= EMAC_FTR_HAS_NEW_STACR;
2617 
2618 	/* CAB lacks the appropriate properties */
2619 	if (of_device_is_compatible(np, "ibm,emac-axon"))
2620 		dev->features |= EMAC_FTR_HAS_NEW_STACR |
2621 			EMAC_FTR_STACR_OC_INVERT;
2622 
2623 	/* Enable TAH/ZMII/RGMII features as found */
2624 	if (dev->tah_ph != 0) {
2625 #ifdef CONFIG_IBM_EMAC_TAH
2626 		dev->features |= EMAC_FTR_HAS_TAH;
2627 #else
2628 		printk(KERN_ERR "%s: TAH support not enabled !\n",
2629 		       np->full_name);
2630 		return -ENXIO;
2631 #endif
2632 	}
2633 
2634 	if (dev->zmii_ph != 0) {
2635 #ifdef CONFIG_IBM_EMAC_ZMII
2636 		dev->features |= EMAC_FTR_HAS_ZMII;
2637 #else
2638 		printk(KERN_ERR "%s: ZMII support not enabled !\n",
2639 		       np->full_name);
2640 		return -ENXIO;
2641 #endif
2642 	}
2643 
2644 	if (dev->rgmii_ph != 0) {
2645 #ifdef CONFIG_IBM_EMAC_RGMII
2646 		dev->features |= EMAC_FTR_HAS_RGMII;
2647 #else
2648 		printk(KERN_ERR "%s: RGMII support not enabled !\n",
2649 		       np->full_name);
2650 		return -ENXIO;
2651 #endif
2652 	}
2653 
2654 	/* Read MAC-address */
2655 	p = of_get_property(np, "local-mac-address", NULL);
2656 	if (p == NULL) {
2657 		printk(KERN_ERR "%s: Can't find local-mac-address property\n",
2658 		       np->full_name);
2659 		return -ENXIO;
2660 	}
2661 	memcpy(dev->ndev->dev_addr, p, ETH_ALEN);
2662 
2663 	/* IAHT and GAHT filter parameterization */
2664 	if (emac_has_feature(dev, EMAC_FTR_EMAC4SYNC)) {
2665 		dev->xaht_slots_shift = EMAC4SYNC_XAHT_SLOTS_SHIFT;
2666 		dev->xaht_width_shift = EMAC4SYNC_XAHT_WIDTH_SHIFT;
2667 	} else {
2668 		dev->xaht_slots_shift = EMAC4_XAHT_SLOTS_SHIFT;
2669 		dev->xaht_width_shift = EMAC4_XAHT_WIDTH_SHIFT;
2670 	}
2671 
2672 	DBG(dev, "features     : 0x%08x / 0x%08x\n", dev->features, EMAC_FTRS_POSSIBLE);
2673 	DBG(dev, "tx_fifo_size : %d (%d gige)\n", dev->tx_fifo_size, dev->tx_fifo_size_gige);
2674 	DBG(dev, "rx_fifo_size : %d (%d gige)\n", dev->rx_fifo_size, dev->rx_fifo_size_gige);
2675 	DBG(dev, "max_mtu      : %d\n", dev->max_mtu);
2676 	DBG(dev, "OPB freq     : %d\n", dev->opb_bus_freq);
2677 
2678 	return 0;
2679 }
2680 
2681 static const struct net_device_ops emac_netdev_ops = {
2682 	.ndo_open		= emac_open,
2683 	.ndo_stop		= emac_close,
2684 	.ndo_get_stats		= emac_stats,
2685 	.ndo_set_rx_mode	= emac_set_multicast_list,
2686 	.ndo_do_ioctl		= emac_ioctl,
2687 	.ndo_tx_timeout		= emac_tx_timeout,
2688 	.ndo_validate_addr	= eth_validate_addr,
2689 	.ndo_set_mac_address	= eth_mac_addr,
2690 	.ndo_start_xmit		= emac_start_xmit,
2691 	.ndo_change_mtu		= eth_change_mtu,
2692 };
2693 
2694 static const struct net_device_ops emac_gige_netdev_ops = {
2695 	.ndo_open		= emac_open,
2696 	.ndo_stop		= emac_close,
2697 	.ndo_get_stats		= emac_stats,
2698 	.ndo_set_rx_mode	= emac_set_multicast_list,
2699 	.ndo_do_ioctl		= emac_ioctl,
2700 	.ndo_tx_timeout		= emac_tx_timeout,
2701 	.ndo_validate_addr	= eth_validate_addr,
2702 	.ndo_set_mac_address	= eth_mac_addr,
2703 	.ndo_start_xmit		= emac_start_xmit_sg,
2704 	.ndo_change_mtu		= emac_change_mtu,
2705 };
2706 
2707 static int emac_probe(struct platform_device *ofdev)
2708 {
2709 	struct net_device *ndev;
2710 	struct emac_instance *dev;
2711 	struct device_node *np = ofdev->dev.of_node;
2712 	struct device_node **blist = NULL;
2713 	int err, i;
2714 
2715 	/* Skip unused/unwired EMACS.  We leave the check for an unused
2716 	 * property here for now, but new flat device trees should set a
2717 	 * status property to "disabled" instead.
2718 	 */
2719 	if (of_get_property(np, "unused", NULL) || !of_device_is_available(np))
2720 		return -ENODEV;
2721 
2722 	/* Find ourselves in the bootlist if we are there */
2723 	for (i = 0; i < EMAC_BOOT_LIST_SIZE; i++)
2724 		if (emac_boot_list[i] == np)
2725 			blist = &emac_boot_list[i];
2726 
2727 	/* Allocate our net_device structure */
2728 	err = -ENOMEM;
2729 	ndev = alloc_etherdev(sizeof(struct emac_instance));
2730 	if (!ndev)
2731 		goto err_gone;
2732 
2733 	dev = netdev_priv(ndev);
2734 	dev->ndev = ndev;
2735 	dev->ofdev = ofdev;
2736 	dev->blist = blist;
2737 	SET_NETDEV_DEV(ndev, &ofdev->dev);
2738 
2739 	/* Initialize some embedded data structures */
2740 	mutex_init(&dev->mdio_lock);
2741 	mutex_init(&dev->link_lock);
2742 	spin_lock_init(&dev->lock);
2743 	INIT_WORK(&dev->reset_work, emac_reset_work);
2744 
2745 	/* Init various config data based on device-tree */
2746 	err = emac_init_config(dev);
2747 	if (err != 0)
2748 		goto err_free;
2749 
2750 	/* Get interrupts. EMAC irq is mandatory, WOL irq is optional */
2751 	dev->emac_irq = irq_of_parse_and_map(np, 0);
2752 	dev->wol_irq = irq_of_parse_and_map(np, 1);
2753 	if (dev->emac_irq == NO_IRQ) {
2754 		printk(KERN_ERR "%s: Can't map main interrupt\n", np->full_name);
2755 		goto err_free;
2756 	}
2757 	ndev->irq = dev->emac_irq;
2758 
2759 	/* Map EMAC regs */
2760 	if (of_address_to_resource(np, 0, &dev->rsrc_regs)) {
2761 		printk(KERN_ERR "%s: Can't get registers address\n",
2762 		       np->full_name);
2763 		goto err_irq_unmap;
2764 	}
2765 	// TODO : request_mem_region
2766 	dev->emacp = ioremap(dev->rsrc_regs.start,
2767 			     resource_size(&dev->rsrc_regs));
2768 	if (dev->emacp == NULL) {
2769 		printk(KERN_ERR "%s: Can't map device registers!\n",
2770 		       np->full_name);
2771 		err = -ENOMEM;
2772 		goto err_irq_unmap;
2773 	}
2774 
2775 	/* Wait for dependent devices */
2776 	err = emac_wait_deps(dev);
2777 	if (err) {
2778 		printk(KERN_ERR
2779 		       "%s: Timeout waiting for dependent devices\n",
2780 		       np->full_name);
2781 		/*  display more info about what's missing ? */
2782 		goto err_reg_unmap;
2783 	}
2784 	dev->mal = platform_get_drvdata(dev->mal_dev);
2785 	if (dev->mdio_dev != NULL)
2786 		dev->mdio_instance = platform_get_drvdata(dev->mdio_dev);
2787 
2788 	/* Register with MAL */
2789 	dev->commac.ops = &emac_commac_ops;
2790 	dev->commac.dev = dev;
2791 	dev->commac.tx_chan_mask = MAL_CHAN_MASK(dev->mal_tx_chan);
2792 	dev->commac.rx_chan_mask = MAL_CHAN_MASK(dev->mal_rx_chan);
2793 	err = mal_register_commac(dev->mal, &dev->commac);
2794 	if (err) {
2795 		printk(KERN_ERR "%s: failed to register with mal %s!\n",
2796 		       np->full_name, dev->mal_dev->dev.of_node->full_name);
2797 		goto err_rel_deps;
2798 	}
2799 	dev->rx_skb_size = emac_rx_skb_size(ndev->mtu);
2800 	dev->rx_sync_size = emac_rx_sync_size(ndev->mtu);
2801 
2802 	/* Get pointers to BD rings */
2803 	dev->tx_desc =
2804 	    dev->mal->bd_virt + mal_tx_bd_offset(dev->mal, dev->mal_tx_chan);
2805 	dev->rx_desc =
2806 	    dev->mal->bd_virt + mal_rx_bd_offset(dev->mal, dev->mal_rx_chan);
2807 
2808 	DBG(dev, "tx_desc %p" NL, dev->tx_desc);
2809 	DBG(dev, "rx_desc %p" NL, dev->rx_desc);
2810 
2811 	/* Clean rings */
2812 	memset(dev->tx_desc, 0, NUM_TX_BUFF * sizeof(struct mal_descriptor));
2813 	memset(dev->rx_desc, 0, NUM_RX_BUFF * sizeof(struct mal_descriptor));
2814 	memset(dev->tx_skb, 0, NUM_TX_BUFF * sizeof(struct sk_buff *));
2815 	memset(dev->rx_skb, 0, NUM_RX_BUFF * sizeof(struct sk_buff *));
2816 
2817 	/* Attach to ZMII, if needed */
2818 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII) &&
2819 	    (err = zmii_attach(dev->zmii_dev, dev->zmii_port, &dev->phy_mode)) != 0)
2820 		goto err_unreg_commac;
2821 
2822 	/* Attach to RGMII, if needed */
2823 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII) &&
2824 	    (err = rgmii_attach(dev->rgmii_dev, dev->rgmii_port, dev->phy_mode)) != 0)
2825 		goto err_detach_zmii;
2826 
2827 	/* Attach to TAH, if needed */
2828 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH) &&
2829 	    (err = tah_attach(dev->tah_dev, dev->tah_port)) != 0)
2830 		goto err_detach_rgmii;
2831 
2832 	/* Set some link defaults before we can find out real parameters */
2833 	dev->phy.speed = SPEED_100;
2834 	dev->phy.duplex = DUPLEX_FULL;
2835 	dev->phy.autoneg = AUTONEG_DISABLE;
2836 	dev->phy.pause = dev->phy.asym_pause = 0;
2837 	dev->stop_timeout = STOP_TIMEOUT_100;
2838 	INIT_DELAYED_WORK(&dev->link_work, emac_link_timer);
2839 
2840 	/* Some SoCs like APM821xx does not support Half Duplex mode. */
2841 	if (emac_has_feature(dev, EMAC_FTR_APM821XX_NO_HALF_DUPLEX)) {
2842 		dev->phy_feat_exc = (SUPPORTED_1000baseT_Half |
2843 				     SUPPORTED_100baseT_Half |
2844 				     SUPPORTED_10baseT_Half);
2845 	}
2846 
2847 	/* Find PHY if any */
2848 	err = emac_init_phy(dev);
2849 	if (err != 0)
2850 		goto err_detach_tah;
2851 
2852 	if (dev->tah_dev) {
2853 		ndev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG;
2854 		ndev->features |= ndev->hw_features | NETIF_F_RXCSUM;
2855 	}
2856 	ndev->watchdog_timeo = 5 * HZ;
2857 	if (emac_phy_supports_gige(dev->phy_mode)) {
2858 		ndev->netdev_ops = &emac_gige_netdev_ops;
2859 		dev->commac.ops = &emac_commac_sg_ops;
2860 	} else
2861 		ndev->netdev_ops = &emac_netdev_ops;
2862 	ndev->ethtool_ops = &emac_ethtool_ops;
2863 
2864 	netif_carrier_off(ndev);
2865 
2866 	err = register_netdev(ndev);
2867 	if (err) {
2868 		printk(KERN_ERR "%s: failed to register net device (%d)!\n",
2869 		       np->full_name, err);
2870 		goto err_detach_tah;
2871 	}
2872 
2873 	/* Set our drvdata last as we don't want them visible until we are
2874 	 * fully initialized
2875 	 */
2876 	wmb();
2877 	platform_set_drvdata(ofdev, dev);
2878 
2879 	/* There's a new kid in town ! Let's tell everybody */
2880 	wake_up_all(&emac_probe_wait);
2881 
2882 
2883 	printk(KERN_INFO "%s: EMAC-%d %s, MAC %pM\n",
2884 	       ndev->name, dev->cell_index, np->full_name, ndev->dev_addr);
2885 
2886 	if (dev->phy_mode == PHY_MODE_SGMII)
2887 		printk(KERN_NOTICE "%s: in SGMII mode\n", ndev->name);
2888 
2889 	if (dev->phy.address >= 0)
2890 		printk("%s: found %s PHY (0x%02x)\n", ndev->name,
2891 		       dev->phy.def->name, dev->phy.address);
2892 
2893 	emac_dbg_register(dev);
2894 
2895 	/* Life is good */
2896 	return 0;
2897 
2898 	/* I have a bad feeling about this ... */
2899 
2900  err_detach_tah:
2901 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
2902 		tah_detach(dev->tah_dev, dev->tah_port);
2903  err_detach_rgmii:
2904 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
2905 		rgmii_detach(dev->rgmii_dev, dev->rgmii_port);
2906  err_detach_zmii:
2907 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
2908 		zmii_detach(dev->zmii_dev, dev->zmii_port);
2909  err_unreg_commac:
2910 	mal_unregister_commac(dev->mal, &dev->commac);
2911  err_rel_deps:
2912 	emac_put_deps(dev);
2913  err_reg_unmap:
2914 	iounmap(dev->emacp);
2915  err_irq_unmap:
2916 	if (dev->wol_irq != NO_IRQ)
2917 		irq_dispose_mapping(dev->wol_irq);
2918 	if (dev->emac_irq != NO_IRQ)
2919 		irq_dispose_mapping(dev->emac_irq);
2920  err_free:
2921 	free_netdev(ndev);
2922  err_gone:
2923 	/* if we were on the bootlist, remove us as we won't show up and
2924 	 * wake up all waiters to notify them in case they were waiting
2925 	 * on us
2926 	 */
2927 	if (blist) {
2928 		*blist = NULL;
2929 		wake_up_all(&emac_probe_wait);
2930 	}
2931 	return err;
2932 }
2933 
2934 static int emac_remove(struct platform_device *ofdev)
2935 {
2936 	struct emac_instance *dev = platform_get_drvdata(ofdev);
2937 
2938 	DBG(dev, "remove" NL);
2939 
2940 	unregister_netdev(dev->ndev);
2941 
2942 	cancel_work_sync(&dev->reset_work);
2943 
2944 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
2945 		tah_detach(dev->tah_dev, dev->tah_port);
2946 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
2947 		rgmii_detach(dev->rgmii_dev, dev->rgmii_port);
2948 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
2949 		zmii_detach(dev->zmii_dev, dev->zmii_port);
2950 
2951 	busy_phy_map &= ~(1 << dev->phy.address);
2952 	DBG(dev, "busy_phy_map now %#x" NL, busy_phy_map);
2953 
2954 	mal_unregister_commac(dev->mal, &dev->commac);
2955 	emac_put_deps(dev);
2956 
2957 	emac_dbg_unregister(dev);
2958 	iounmap(dev->emacp);
2959 
2960 	if (dev->wol_irq != NO_IRQ)
2961 		irq_dispose_mapping(dev->wol_irq);
2962 	if (dev->emac_irq != NO_IRQ)
2963 		irq_dispose_mapping(dev->emac_irq);
2964 
2965 	free_netdev(dev->ndev);
2966 
2967 	return 0;
2968 }
2969 
2970 /* XXX Features in here should be replaced by properties... */
2971 static const struct of_device_id emac_match[] =
2972 {
2973 	{
2974 		.type		= "network",
2975 		.compatible	= "ibm,emac",
2976 	},
2977 	{
2978 		.type		= "network",
2979 		.compatible	= "ibm,emac4",
2980 	},
2981 	{
2982 		.type		= "network",
2983 		.compatible	= "ibm,emac4sync",
2984 	},
2985 	{},
2986 };
2987 MODULE_DEVICE_TABLE(of, emac_match);
2988 
2989 static struct platform_driver emac_driver = {
2990 	.driver = {
2991 		.name = "emac",
2992 		.of_match_table = emac_match,
2993 	},
2994 	.probe = emac_probe,
2995 	.remove = emac_remove,
2996 };
2997 
2998 static void __init emac_make_bootlist(void)
2999 {
3000 	struct device_node *np = NULL;
3001 	int j, max, i = 0;
3002 	int cell_indices[EMAC_BOOT_LIST_SIZE];
3003 
3004 	/* Collect EMACs */
3005 	while((np = of_find_all_nodes(np)) != NULL) {
3006 		const u32 *idx;
3007 
3008 		if (of_match_node(emac_match, np) == NULL)
3009 			continue;
3010 		if (of_get_property(np, "unused", NULL))
3011 			continue;
3012 		idx = of_get_property(np, "cell-index", NULL);
3013 		if (idx == NULL)
3014 			continue;
3015 		cell_indices[i] = *idx;
3016 		emac_boot_list[i++] = of_node_get(np);
3017 		if (i >= EMAC_BOOT_LIST_SIZE) {
3018 			of_node_put(np);
3019 			break;
3020 		}
3021 	}
3022 	max = i;
3023 
3024 	/* Bubble sort them (doh, what a creative algorithm :-) */
3025 	for (i = 0; max > 1 && (i < (max - 1)); i++)
3026 		for (j = i; j < max; j++) {
3027 			if (cell_indices[i] > cell_indices[j]) {
3028 				swap(emac_boot_list[i], emac_boot_list[j]);
3029 				swap(cell_indices[i], cell_indices[j]);
3030 			}
3031 		}
3032 }
3033 
3034 static int __init emac_init(void)
3035 {
3036 	int rc;
3037 
3038 	printk(KERN_INFO DRV_DESC ", version " DRV_VERSION "\n");
3039 
3040 	/* Init debug stuff */
3041 	emac_init_debug();
3042 
3043 	/* Build EMAC boot list */
3044 	emac_make_bootlist();
3045 
3046 	/* Init submodules */
3047 	rc = mal_init();
3048 	if (rc)
3049 		goto err;
3050 	rc = zmii_init();
3051 	if (rc)
3052 		goto err_mal;
3053 	rc = rgmii_init();
3054 	if (rc)
3055 		goto err_zmii;
3056 	rc = tah_init();
3057 	if (rc)
3058 		goto err_rgmii;
3059 	rc = platform_driver_register(&emac_driver);
3060 	if (rc)
3061 		goto err_tah;
3062 
3063 	return 0;
3064 
3065  err_tah:
3066 	tah_exit();
3067  err_rgmii:
3068 	rgmii_exit();
3069  err_zmii:
3070 	zmii_exit();
3071  err_mal:
3072 	mal_exit();
3073  err:
3074 	return rc;
3075 }
3076 
3077 static void __exit emac_exit(void)
3078 {
3079 	int i;
3080 
3081 	platform_driver_unregister(&emac_driver);
3082 
3083 	tah_exit();
3084 	rgmii_exit();
3085 	zmii_exit();
3086 	mal_exit();
3087 	emac_fini_debug();
3088 
3089 	/* Destroy EMAC boot list */
3090 	for (i = 0; i < EMAC_BOOT_LIST_SIZE; i++)
3091 		of_node_put(emac_boot_list[i]);
3092 }
3093 
3094 module_init(emac_init);
3095 module_exit(emac_exit);
3096