xref: /linux/drivers/net/ethernet/ibm/emac/core.c (revision e9f0878c4b2004ac19581274c1ae4c61ae3ca70e)
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/of_mdio.h>
46 #include <linux/slab.h>
47 
48 #include <asm/processor.h>
49 #include <asm/io.h>
50 #include <asm/dma.h>
51 #include <linux/uaccess.h>
52 #include <asm/dcr.h>
53 #include <asm/dcr-regs.h>
54 
55 #include "core.h"
56 
57 /*
58  * Lack of dma_unmap_???? calls is intentional.
59  *
60  * API-correct usage requires additional support state information to be
61  * maintained for every RX and TX buffer descriptor (BD). Unfortunately, due to
62  * EMAC design (e.g. TX buffer passed from network stack can be split into
63  * several BDs, dma_map_single/dma_map_page can be used to map particular BD),
64  * maintaining such information will add additional overhead.
65  * Current DMA API implementation for 4xx processors only ensures cache coherency
66  * and dma_unmap_???? routines are empty and are likely to stay this way.
67  * I decided to omit dma_unmap_??? calls because I don't want to add additional
68  * complexity just for the sake of following some abstract API, when it doesn't
69  * add any real benefit to the driver. I understand that this decision maybe
70  * controversial, but I really tried to make code API-correct and efficient
71  * at the same time and didn't come up with code I liked :(.                --ebs
72  */
73 
74 #define DRV_NAME        "emac"
75 #define DRV_VERSION     "3.54"
76 #define DRV_DESC        "PPC 4xx OCP EMAC driver"
77 
78 MODULE_DESCRIPTION(DRV_DESC);
79 MODULE_AUTHOR
80     ("Eugene Surovegin <eugene.surovegin@zultys.com> or <ebs@ebshome.net>");
81 MODULE_LICENSE("GPL");
82 
83 /* minimum number of free TX descriptors required to wake up TX process */
84 #define EMAC_TX_WAKEUP_THRESH		(NUM_TX_BUFF / 4)
85 
86 /* If packet size is less than this number, we allocate small skb and copy packet
87  * contents into it instead of just sending original big skb up
88  */
89 #define EMAC_RX_COPY_THRESH		CONFIG_IBM_EMAC_RX_COPY_THRESHOLD
90 
91 /* Since multiple EMACs share MDIO lines in various ways, we need
92  * to avoid re-using the same PHY ID in cases where the arch didn't
93  * setup precise phy_map entries
94  *
95  * XXX This is something that needs to be reworked as we can have multiple
96  * EMAC "sets" (multiple ASICs containing several EMACs) though we can
97  * probably require in that case to have explicit PHY IDs in the device-tree
98  */
99 static u32 busy_phy_map;
100 static DEFINE_MUTEX(emac_phy_map_lock);
101 
102 /* This is the wait queue used to wait on any event related to probe, that
103  * is discovery of MALs, other EMACs, ZMII/RGMIIs, etc...
104  */
105 static DECLARE_WAIT_QUEUE_HEAD(emac_probe_wait);
106 
107 /* Having stable interface names is a doomed idea. However, it would be nice
108  * if we didn't have completely random interface names at boot too :-) It's
109  * just a matter of making everybody's life easier. Since we are doing
110  * threaded probing, it's a bit harder though. The base idea here is that
111  * we make up a list of all emacs in the device-tree before we register the
112  * driver. Every emac will then wait for the previous one in the list to
113  * initialize before itself. We should also keep that list ordered by
114  * cell_index.
115  * That list is only 4 entries long, meaning that additional EMACs don't
116  * get ordering guarantees unless EMAC_BOOT_LIST_SIZE is increased.
117  */
118 
119 #define EMAC_BOOT_LIST_SIZE	4
120 static struct device_node *emac_boot_list[EMAC_BOOT_LIST_SIZE];
121 
122 /* How long should I wait for dependent devices ? */
123 #define EMAC_PROBE_DEP_TIMEOUT	(HZ * 5)
124 
125 /* I don't want to litter system log with timeout errors
126  * when we have brain-damaged PHY.
127  */
128 static inline void emac_report_timeout_error(struct emac_instance *dev,
129 					     const char *error)
130 {
131 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX |
132 				  EMAC_FTR_460EX_PHY_CLK_FIX |
133 				  EMAC_FTR_440EP_PHY_CLK_FIX))
134 		DBG(dev, "%s" NL, error);
135 	else if (net_ratelimit())
136 		printk(KERN_ERR "%pOF: %s\n", dev->ofdev->dev.of_node, 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_interface_mode_is_rgmii(phy_mode) ||
203 		phy_mode == PHY_INTERFACE_MODE_GMII ||
204 		phy_mode == PHY_INTERFACE_MODE_SGMII ||
205 		phy_mode == PHY_INTERFACE_MODE_TBI ||
206 		phy_mode == PHY_INTERFACE_MODE_RTBI;
207 }
208 
209 static inline int emac_phy_gpcs(int phy_mode)
210 {
211 	return  phy_mode == PHY_INTERFACE_MODE_SGMII ||
212 		phy_mode == PHY_INTERFACE_MODE_TBI ||
213 		phy_mode == PHY_INTERFACE_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 	bool __maybe_unused try_internal_clock = false;
346 
347 	DBG(dev, "reset" NL);
348 
349 	if (!dev->reset_failed) {
350 		/* 40x erratum suggests stopping RX channel before reset,
351 		 * we stop TX as well
352 		 */
353 		emac_rx_disable(dev);
354 		emac_tx_disable(dev);
355 	}
356 
357 #ifdef CONFIG_PPC_DCR_NATIVE
358 do_retry:
359 	/*
360 	 * PPC460EX/GT Embedded Processor Advanced User's Manual
361 	 * section 28.10.1 Mode Register 0 (EMACx_MR0) states:
362 	 * Note: The PHY must provide a TX Clk in order to perform a soft reset
363 	 * of the EMAC. If none is present, select the internal clock
364 	 * (SDR0_ETH_CFG[EMACx_PHY_CLK] = 1).
365 	 * After a soft reset, select the external clock.
366 	 *
367 	 * The AR8035-A PHY Meraki MR24 does not provide a TX Clk if the
368 	 * ethernet cable is not attached. This causes the reset to timeout
369 	 * and the PHY detection code in emac_init_phy() is unable to
370 	 * communicate and detect the AR8035-A PHY. As a result, the emac
371 	 * driver bails out early and the user has no ethernet.
372 	 * In order to stay compatible with existing configurations, the
373 	 * driver will temporarily switch to the internal clock, after
374 	 * the first reset fails.
375 	 */
376 	if (emac_has_feature(dev, EMAC_FTR_460EX_PHY_CLK_FIX)) {
377 		if (try_internal_clock || (dev->phy_address == 0xffffffff &&
378 					   dev->phy_map == 0xffffffff)) {
379 			/* No PHY: select internal loop clock before reset */
380 			dcri_clrset(SDR0, SDR0_ETH_CFG,
381 				    0, SDR0_ETH_CFG_ECS << dev->cell_index);
382 		} else {
383 			/* PHY present: select external clock before reset */
384 			dcri_clrset(SDR0, SDR0_ETH_CFG,
385 				    SDR0_ETH_CFG_ECS << dev->cell_index, 0);
386 		}
387 	}
388 #endif
389 
390 	out_be32(&p->mr0, EMAC_MR0_SRST);
391 	while ((in_be32(&p->mr0) & EMAC_MR0_SRST) && n)
392 		--n;
393 
394 #ifdef CONFIG_PPC_DCR_NATIVE
395 	if (emac_has_feature(dev, EMAC_FTR_460EX_PHY_CLK_FIX)) {
396 		if (!n && !try_internal_clock) {
397 			/* first attempt has timed out. */
398 			n = 20;
399 			try_internal_clock = true;
400 			goto do_retry;
401 		}
402 
403 		if (try_internal_clock || (dev->phy_address == 0xffffffff &&
404 					   dev->phy_map == 0xffffffff)) {
405 			/* No PHY: restore external clock source after reset */
406 			dcri_clrset(SDR0, SDR0_ETH_CFG,
407 				    SDR0_ETH_CFG_ECS << dev->cell_index, 0);
408 		}
409 	}
410 #endif
411 
412 	if (n) {
413 		dev->reset_failed = 0;
414 		return 0;
415 	} else {
416 		emac_report_timeout_error(dev, "reset timeout");
417 		dev->reset_failed = 1;
418 		return -ETIMEDOUT;
419 	}
420 }
421 
422 static void emac_hash_mc(struct emac_instance *dev)
423 {
424 	const int regs = EMAC_XAHT_REGS(dev);
425 	u32 *gaht_base = emac_gaht_base(dev);
426 	u32 gaht_temp[regs];
427 	struct netdev_hw_addr *ha;
428 	int i;
429 
430 	DBG(dev, "hash_mc %d" NL, netdev_mc_count(dev->ndev));
431 
432 	memset(gaht_temp, 0, sizeof (gaht_temp));
433 
434 	netdev_for_each_mc_addr(ha, dev->ndev) {
435 		int slot, reg, mask;
436 		DBG2(dev, "mc %pM" NL, ha->addr);
437 
438 		slot = EMAC_XAHT_CRC_TO_SLOT(dev,
439 					     ether_crc(ETH_ALEN, ha->addr));
440 		reg = EMAC_XAHT_SLOT_TO_REG(dev, slot);
441 		mask = EMAC_XAHT_SLOT_TO_MASK(dev, slot);
442 
443 		gaht_temp[reg] |= mask;
444 	}
445 
446 	for (i = 0; i < regs; i++)
447 		out_be32(gaht_base + i, gaht_temp[i]);
448 }
449 
450 static inline u32 emac_iff2rmr(struct net_device *ndev)
451 {
452 	struct emac_instance *dev = netdev_priv(ndev);
453 	u32 r;
454 
455 	r = EMAC_RMR_SP | EMAC_RMR_SFCS | EMAC_RMR_IAE | EMAC_RMR_BAE;
456 
457 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
458 	    r |= EMAC4_RMR_BASE;
459 	else
460 	    r |= EMAC_RMR_BASE;
461 
462 	if (ndev->flags & IFF_PROMISC)
463 		r |= EMAC_RMR_PME;
464 	else if (ndev->flags & IFF_ALLMULTI ||
465 			 (netdev_mc_count(ndev) > EMAC_XAHT_SLOTS(dev)))
466 		r |= EMAC_RMR_PMME;
467 	else if (!netdev_mc_empty(ndev))
468 		r |= EMAC_RMR_MAE;
469 
470 	if (emac_has_feature(dev, EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE)) {
471 		r &= ~EMAC4_RMR_MJS_MASK;
472 		r |= EMAC4_RMR_MJS(ndev->mtu);
473 	}
474 
475 	return r;
476 }
477 
478 static u32 __emac_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
479 {
480 	u32 ret = EMAC_MR1_VLE | EMAC_MR1_IST | EMAC_MR1_TR0_MULT;
481 
482 	DBG2(dev, "__emac_calc_base_mr1" NL);
483 
484 	switch(tx_size) {
485 	case 2048:
486 		ret |= EMAC_MR1_TFS_2K;
487 		break;
488 	default:
489 		printk(KERN_WARNING "%s: Unknown Tx FIFO size %d\n",
490 		       dev->ndev->name, tx_size);
491 	}
492 
493 	switch(rx_size) {
494 	case 16384:
495 		ret |= EMAC_MR1_RFS_16K;
496 		break;
497 	case 8192:
498 		ret |= EMAC4_MR1_RFS_8K;
499 		break;
500 	case 4096:
501 		ret |= EMAC_MR1_RFS_4K;
502 		break;
503 	default:
504 		printk(KERN_WARNING "%s: Unknown Rx FIFO size %d\n",
505 		       dev->ndev->name, rx_size);
506 	}
507 
508 	return ret;
509 }
510 
511 static u32 __emac4_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
512 {
513 	u32 ret = EMAC_MR1_VLE | EMAC_MR1_IST | EMAC4_MR1_TR |
514 		EMAC4_MR1_OBCI(dev->opb_bus_freq / 1000000);
515 
516 	DBG2(dev, "__emac4_calc_base_mr1" NL);
517 
518 	switch(tx_size) {
519 	case 16384:
520 		ret |= EMAC4_MR1_TFS_16K;
521 		break;
522 	case 8192:
523 		ret |= EMAC4_MR1_TFS_8K;
524 		break;
525 	case 4096:
526 		ret |= EMAC4_MR1_TFS_4K;
527 		break;
528 	case 2048:
529 		ret |= EMAC4_MR1_TFS_2K;
530 		break;
531 	default:
532 		printk(KERN_WARNING "%s: Unknown Tx FIFO size %d\n",
533 		       dev->ndev->name, tx_size);
534 	}
535 
536 	switch(rx_size) {
537 	case 16384:
538 		ret |= EMAC4_MR1_RFS_16K;
539 		break;
540 	case 4096:
541 		ret |= EMAC4_MR1_RFS_4K;
542 		break;
543 	case 2048:
544 		ret |= EMAC4_MR1_RFS_2K;
545 		break;
546 	default:
547 		printk(KERN_WARNING "%s: Unknown Rx FIFO size %d\n",
548 		       dev->ndev->name, rx_size);
549 	}
550 
551 	return ret;
552 }
553 
554 static u32 emac_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
555 {
556 	return emac_has_feature(dev, EMAC_FTR_EMAC4) ?
557 		__emac4_calc_base_mr1(dev, tx_size, rx_size) :
558 		__emac_calc_base_mr1(dev, tx_size, rx_size);
559 }
560 
561 static inline u32 emac_calc_trtr(struct emac_instance *dev, unsigned int size)
562 {
563 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
564 		return ((size >> 6) - 1) << EMAC_TRTR_SHIFT_EMAC4;
565 	else
566 		return ((size >> 6) - 1) << EMAC_TRTR_SHIFT;
567 }
568 
569 static inline u32 emac_calc_rwmr(struct emac_instance *dev,
570 				 unsigned int low, unsigned int high)
571 {
572 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
573 		return (low << 22) | ( (high & 0x3ff) << 6);
574 	else
575 		return (low << 23) | ( (high & 0x1ff) << 7);
576 }
577 
578 static int emac_configure(struct emac_instance *dev)
579 {
580 	struct emac_regs __iomem *p = dev->emacp;
581 	struct net_device *ndev = dev->ndev;
582 	int tx_size, rx_size, link = netif_carrier_ok(dev->ndev);
583 	u32 r, mr1 = 0;
584 
585 	DBG(dev, "configure" NL);
586 
587 	if (!link) {
588 		out_be32(&p->mr1, in_be32(&p->mr1)
589 			 | EMAC_MR1_FDE | EMAC_MR1_ILE);
590 		udelay(100);
591 	} else if (emac_reset(dev) < 0)
592 		return -ETIMEDOUT;
593 
594 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
595 		tah_reset(dev->tah_dev);
596 
597 	DBG(dev, " link = %d duplex = %d, pause = %d, asym_pause = %d\n",
598 	    link, dev->phy.duplex, dev->phy.pause, dev->phy.asym_pause);
599 
600 	/* Default fifo sizes */
601 	tx_size = dev->tx_fifo_size;
602 	rx_size = dev->rx_fifo_size;
603 
604 	/* No link, force loopback */
605 	if (!link)
606 		mr1 = EMAC_MR1_FDE | EMAC_MR1_ILE;
607 
608 	/* Check for full duplex */
609 	else if (dev->phy.duplex == DUPLEX_FULL)
610 		mr1 |= EMAC_MR1_FDE | EMAC_MR1_MWSW_001;
611 
612 	/* Adjust fifo sizes, mr1 and timeouts based on link speed */
613 	dev->stop_timeout = STOP_TIMEOUT_10;
614 	switch (dev->phy.speed) {
615 	case SPEED_1000:
616 		if (emac_phy_gpcs(dev->phy.mode)) {
617 			mr1 |= EMAC_MR1_MF_1000GPCS | EMAC_MR1_MF_IPPA(
618 				(dev->phy.gpcs_address != 0xffffffff) ?
619 				 dev->phy.gpcs_address : dev->phy.address);
620 
621 			/* Put some arbitrary OUI, Manuf & Rev IDs so we can
622 			 * identify this GPCS PHY later.
623 			 */
624 			out_be32(&p->u1.emac4.ipcr, 0xdeadbeef);
625 		} else
626 			mr1 |= EMAC_MR1_MF_1000;
627 
628 		/* Extended fifo sizes */
629 		tx_size = dev->tx_fifo_size_gige;
630 		rx_size = dev->rx_fifo_size_gige;
631 
632 		if (dev->ndev->mtu > ETH_DATA_LEN) {
633 			if (emac_has_feature(dev, EMAC_FTR_EMAC4))
634 				mr1 |= EMAC4_MR1_JPSM;
635 			else
636 				mr1 |= EMAC_MR1_JPSM;
637 			dev->stop_timeout = STOP_TIMEOUT_1000_JUMBO;
638 		} else
639 			dev->stop_timeout = STOP_TIMEOUT_1000;
640 		break;
641 	case SPEED_100:
642 		mr1 |= EMAC_MR1_MF_100;
643 		dev->stop_timeout = STOP_TIMEOUT_100;
644 		break;
645 	default: /* make gcc happy */
646 		break;
647 	}
648 
649 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
650 		rgmii_set_speed(dev->rgmii_dev, dev->rgmii_port,
651 				dev->phy.speed);
652 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
653 		zmii_set_speed(dev->zmii_dev, dev->zmii_port, dev->phy.speed);
654 
655 	/* on 40x erratum forces us to NOT use integrated flow control,
656 	 * let's hope it works on 44x ;)
657 	 */
658 	if (!emac_has_feature(dev, EMAC_FTR_NO_FLOW_CONTROL_40x) &&
659 	    dev->phy.duplex == DUPLEX_FULL) {
660 		if (dev->phy.pause)
661 			mr1 |= EMAC_MR1_EIFC | EMAC_MR1_APP;
662 		else if (dev->phy.asym_pause)
663 			mr1 |= EMAC_MR1_APP;
664 	}
665 
666 	/* Add base settings & fifo sizes & program MR1 */
667 	mr1 |= emac_calc_base_mr1(dev, tx_size, rx_size);
668 	out_be32(&p->mr1, mr1);
669 
670 	/* Set individual MAC address */
671 	out_be32(&p->iahr, (ndev->dev_addr[0] << 8) | ndev->dev_addr[1]);
672 	out_be32(&p->ialr, (ndev->dev_addr[2] << 24) |
673 		 (ndev->dev_addr[3] << 16) | (ndev->dev_addr[4] << 8) |
674 		 ndev->dev_addr[5]);
675 
676 	/* VLAN Tag Protocol ID */
677 	out_be32(&p->vtpid, 0x8100);
678 
679 	/* Receive mode register */
680 	r = emac_iff2rmr(ndev);
681 	if (r & EMAC_RMR_MAE)
682 		emac_hash_mc(dev);
683 	out_be32(&p->rmr, r);
684 
685 	/* FIFOs thresholds */
686 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
687 		r = EMAC4_TMR1((dev->mal_burst_size / dev->fifo_entry_size) + 1,
688 			       tx_size / 2 / dev->fifo_entry_size);
689 	else
690 		r = EMAC_TMR1((dev->mal_burst_size / dev->fifo_entry_size) + 1,
691 			      tx_size / 2 / dev->fifo_entry_size);
692 	out_be32(&p->tmr1, r);
693 	out_be32(&p->trtr, emac_calc_trtr(dev, tx_size / 2));
694 
695 	/* PAUSE frame is sent when RX FIFO reaches its high-water mark,
696 	   there should be still enough space in FIFO to allow the our link
697 	   partner time to process this frame and also time to send PAUSE
698 	   frame itself.
699 
700 	   Here is the worst case scenario for the RX FIFO "headroom"
701 	   (from "The Switch Book") (100Mbps, without preamble, inter-frame gap):
702 
703 	   1) One maximum-length frame on TX                    1522 bytes
704 	   2) One PAUSE frame time                                64 bytes
705 	   3) PAUSE frame decode time allowance                   64 bytes
706 	   4) One maximum-length frame on RX                    1522 bytes
707 	   5) Round-trip propagation delay of the link (100Mb)    15 bytes
708 	   ----------
709 	   3187 bytes
710 
711 	   I chose to set high-water mark to RX_FIFO_SIZE / 4 (1024 bytes)
712 	   low-water mark  to RX_FIFO_SIZE / 8 (512 bytes)
713 	 */
714 	r = emac_calc_rwmr(dev, rx_size / 8 / dev->fifo_entry_size,
715 			   rx_size / 4 / dev->fifo_entry_size);
716 	out_be32(&p->rwmr, r);
717 
718 	/* Set PAUSE timer to the maximum */
719 	out_be32(&p->ptr, 0xffff);
720 
721 	/* IRQ sources */
722 	r = EMAC_ISR_OVR | EMAC_ISR_BP | EMAC_ISR_SE |
723 		EMAC_ISR_ALE | EMAC_ISR_BFCS | EMAC_ISR_PTLE | EMAC_ISR_ORE |
724 		EMAC_ISR_IRE | EMAC_ISR_TE;
725 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
726 	    r |= EMAC4_ISR_TXPE | EMAC4_ISR_RXPE /* | EMAC4_ISR_TXUE |
727 						  EMAC4_ISR_RXOE | */;
728 	out_be32(&p->iser,  r);
729 
730 	/* We need to take GPCS PHY out of isolate mode after EMAC reset */
731 	if (emac_phy_gpcs(dev->phy.mode)) {
732 		if (dev->phy.gpcs_address != 0xffffffff)
733 			emac_mii_reset_gpcs(&dev->phy);
734 		else
735 			emac_mii_reset_phy(&dev->phy);
736 	}
737 
738 	return 0;
739 }
740 
741 static void emac_reinitialize(struct emac_instance *dev)
742 {
743 	DBG(dev, "reinitialize" NL);
744 
745 	emac_netif_stop(dev);
746 	if (!emac_configure(dev)) {
747 		emac_tx_enable(dev);
748 		emac_rx_enable(dev);
749 	}
750 	emac_netif_start(dev);
751 }
752 
753 static void emac_full_tx_reset(struct emac_instance *dev)
754 {
755 	DBG(dev, "full_tx_reset" NL);
756 
757 	emac_tx_disable(dev);
758 	mal_disable_tx_channel(dev->mal, dev->mal_tx_chan);
759 	emac_clean_tx_ring(dev);
760 	dev->tx_cnt = dev->tx_slot = dev->ack_slot = 0;
761 
762 	emac_configure(dev);
763 
764 	mal_enable_tx_channel(dev->mal, dev->mal_tx_chan);
765 	emac_tx_enable(dev);
766 	emac_rx_enable(dev);
767 }
768 
769 static void emac_reset_work(struct work_struct *work)
770 {
771 	struct emac_instance *dev = container_of(work, struct emac_instance, reset_work);
772 
773 	DBG(dev, "reset_work" NL);
774 
775 	mutex_lock(&dev->link_lock);
776 	if (dev->opened) {
777 		emac_netif_stop(dev);
778 		emac_full_tx_reset(dev);
779 		emac_netif_start(dev);
780 	}
781 	mutex_unlock(&dev->link_lock);
782 }
783 
784 static void emac_tx_timeout(struct net_device *ndev)
785 {
786 	struct emac_instance *dev = netdev_priv(ndev);
787 
788 	DBG(dev, "tx_timeout" NL);
789 
790 	schedule_work(&dev->reset_work);
791 }
792 
793 
794 static inline int emac_phy_done(struct emac_instance *dev, u32 stacr)
795 {
796 	int done = !!(stacr & EMAC_STACR_OC);
797 
798 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
799 		done = !done;
800 
801 	return done;
802 };
803 
804 static int __emac_mdio_read(struct emac_instance *dev, u8 id, u8 reg)
805 {
806 	struct emac_regs __iomem *p = dev->emacp;
807 	u32 r = 0;
808 	int n, err = -ETIMEDOUT;
809 
810 	mutex_lock(&dev->mdio_lock);
811 
812 	DBG2(dev, "mdio_read(%02x,%02x)" NL, id, reg);
813 
814 	/* Enable proper MDIO port */
815 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
816 		zmii_get_mdio(dev->zmii_dev, dev->zmii_port);
817 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
818 		rgmii_get_mdio(dev->rgmii_dev, dev->rgmii_port);
819 
820 	/* Wait for management interface to become idle */
821 	n = 20;
822 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
823 		udelay(1);
824 		if (!--n) {
825 			DBG2(dev, " -> timeout wait idle\n");
826 			goto bail;
827 		}
828 	}
829 
830 	/* Issue read command */
831 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
832 		r = EMAC4_STACR_BASE(dev->opb_bus_freq);
833 	else
834 		r = EMAC_STACR_BASE(dev->opb_bus_freq);
835 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
836 		r |= EMAC_STACR_OC;
837 	if (emac_has_feature(dev, EMAC_FTR_HAS_NEW_STACR))
838 		r |= EMACX_STACR_STAC_READ;
839 	else
840 		r |= EMAC_STACR_STAC_READ;
841 	r |= (reg & EMAC_STACR_PRA_MASK)
842 		| ((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT);
843 	out_be32(&p->stacr, r);
844 
845 	/* Wait for read to complete */
846 	n = 200;
847 	while (!emac_phy_done(dev, (r = in_be32(&p->stacr)))) {
848 		udelay(1);
849 		if (!--n) {
850 			DBG2(dev, " -> timeout wait complete\n");
851 			goto bail;
852 		}
853 	}
854 
855 	if (unlikely(r & EMAC_STACR_PHYE)) {
856 		DBG(dev, "mdio_read(%02x, %02x) failed" NL, id, reg);
857 		err = -EREMOTEIO;
858 		goto bail;
859 	}
860 
861 	r = ((r >> EMAC_STACR_PHYD_SHIFT) & EMAC_STACR_PHYD_MASK);
862 
863 	DBG2(dev, "mdio_read -> %04x" NL, r);
864 	err = 0;
865  bail:
866 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
867 		rgmii_put_mdio(dev->rgmii_dev, dev->rgmii_port);
868 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
869 		zmii_put_mdio(dev->zmii_dev, dev->zmii_port);
870 	mutex_unlock(&dev->mdio_lock);
871 
872 	return err == 0 ? r : err;
873 }
874 
875 static void __emac_mdio_write(struct emac_instance *dev, u8 id, u8 reg,
876 			      u16 val)
877 {
878 	struct emac_regs __iomem *p = dev->emacp;
879 	u32 r = 0;
880 	int n, err = -ETIMEDOUT;
881 
882 	mutex_lock(&dev->mdio_lock);
883 
884 	DBG2(dev, "mdio_write(%02x,%02x,%04x)" NL, id, reg, val);
885 
886 	/* Enable proper MDIO port */
887 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
888 		zmii_get_mdio(dev->zmii_dev, dev->zmii_port);
889 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
890 		rgmii_get_mdio(dev->rgmii_dev, dev->rgmii_port);
891 
892 	/* Wait for management interface to be idle */
893 	n = 20;
894 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
895 		udelay(1);
896 		if (!--n) {
897 			DBG2(dev, " -> timeout wait idle\n");
898 			goto bail;
899 		}
900 	}
901 
902 	/* Issue write command */
903 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
904 		r = EMAC4_STACR_BASE(dev->opb_bus_freq);
905 	else
906 		r = EMAC_STACR_BASE(dev->opb_bus_freq);
907 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
908 		r |= EMAC_STACR_OC;
909 	if (emac_has_feature(dev, EMAC_FTR_HAS_NEW_STACR))
910 		r |= EMACX_STACR_STAC_WRITE;
911 	else
912 		r |= EMAC_STACR_STAC_WRITE;
913 	r |= (reg & EMAC_STACR_PRA_MASK) |
914 		((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT) |
915 		(val << EMAC_STACR_PHYD_SHIFT);
916 	out_be32(&p->stacr, r);
917 
918 	/* Wait for write to complete */
919 	n = 200;
920 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
921 		udelay(1);
922 		if (!--n) {
923 			DBG2(dev, " -> timeout wait complete\n");
924 			goto bail;
925 		}
926 	}
927 	err = 0;
928  bail:
929 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
930 		rgmii_put_mdio(dev->rgmii_dev, dev->rgmii_port);
931 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
932 		zmii_put_mdio(dev->zmii_dev, dev->zmii_port);
933 	mutex_unlock(&dev->mdio_lock);
934 }
935 
936 static int emac_mdio_read(struct net_device *ndev, int id, int reg)
937 {
938 	struct emac_instance *dev = netdev_priv(ndev);
939 	int res;
940 
941 	res = __emac_mdio_read((dev->mdio_instance &&
942 				dev->phy.gpcs_address != id) ?
943 				dev->mdio_instance : dev,
944 			       (u8) id, (u8) reg);
945 	return res;
946 }
947 
948 static void emac_mdio_write(struct net_device *ndev, int id, int reg, int val)
949 {
950 	struct emac_instance *dev = netdev_priv(ndev);
951 
952 	__emac_mdio_write((dev->mdio_instance &&
953 			   dev->phy.gpcs_address != id) ?
954 			   dev->mdio_instance : dev,
955 			  (u8) id, (u8) reg, (u16) val);
956 }
957 
958 /* Tx lock BH */
959 static void __emac_set_multicast_list(struct emac_instance *dev)
960 {
961 	struct emac_regs __iomem *p = dev->emacp;
962 	u32 rmr = emac_iff2rmr(dev->ndev);
963 
964 	DBG(dev, "__multicast %08x" NL, rmr);
965 
966 	/* I decided to relax register access rules here to avoid
967 	 * full EMAC reset.
968 	 *
969 	 * There is a real problem with EMAC4 core if we use MWSW_001 bit
970 	 * in MR1 register and do a full EMAC reset.
971 	 * One TX BD status update is delayed and, after EMAC reset, it
972 	 * never happens, resulting in TX hung (it'll be recovered by TX
973 	 * timeout handler eventually, but this is just gross).
974 	 * So we either have to do full TX reset or try to cheat here :)
975 	 *
976 	 * The only required change is to RX mode register, so I *think* all
977 	 * we need is just to stop RX channel. This seems to work on all
978 	 * tested SoCs.                                                --ebs
979 	 *
980 	 * If we need the full reset, we might just trigger the workqueue
981 	 * and do it async... a bit nasty but should work --BenH
982 	 */
983 	dev->mcast_pending = 0;
984 	emac_rx_disable(dev);
985 	if (rmr & EMAC_RMR_MAE)
986 		emac_hash_mc(dev);
987 	out_be32(&p->rmr, rmr);
988 	emac_rx_enable(dev);
989 }
990 
991 /* Tx lock BH */
992 static void emac_set_multicast_list(struct net_device *ndev)
993 {
994 	struct emac_instance *dev = netdev_priv(ndev);
995 
996 	DBG(dev, "multicast" NL);
997 
998 	BUG_ON(!netif_running(dev->ndev));
999 
1000 	if (dev->no_mcast) {
1001 		dev->mcast_pending = 1;
1002 		return;
1003 	}
1004 
1005 	mutex_lock(&dev->link_lock);
1006 	__emac_set_multicast_list(dev);
1007 	mutex_unlock(&dev->link_lock);
1008 }
1009 
1010 static int emac_set_mac_address(struct net_device *ndev, void *sa)
1011 {
1012 	struct emac_instance *dev = netdev_priv(ndev);
1013 	struct sockaddr *addr = sa;
1014 	struct emac_regs __iomem *p = dev->emacp;
1015 
1016 	if (!is_valid_ether_addr(addr->sa_data))
1017 	       return -EADDRNOTAVAIL;
1018 
1019 	mutex_lock(&dev->link_lock);
1020 
1021 	memcpy(ndev->dev_addr, addr->sa_data, ndev->addr_len);
1022 
1023 	emac_rx_disable(dev);
1024 	emac_tx_disable(dev);
1025 	out_be32(&p->iahr, (ndev->dev_addr[0] << 8) | ndev->dev_addr[1]);
1026 	out_be32(&p->ialr, (ndev->dev_addr[2] << 24) |
1027 		(ndev->dev_addr[3] << 16) | (ndev->dev_addr[4] << 8) |
1028 		ndev->dev_addr[5]);
1029 	emac_tx_enable(dev);
1030 	emac_rx_enable(dev);
1031 
1032 	mutex_unlock(&dev->link_lock);
1033 
1034 	return 0;
1035 }
1036 
1037 static int emac_resize_rx_ring(struct emac_instance *dev, int new_mtu)
1038 {
1039 	int rx_sync_size = emac_rx_sync_size(new_mtu);
1040 	int rx_skb_size = emac_rx_skb_size(new_mtu);
1041 	int i, ret = 0;
1042 	int mr1_jumbo_bit_change = 0;
1043 
1044 	mutex_lock(&dev->link_lock);
1045 	emac_netif_stop(dev);
1046 	emac_rx_disable(dev);
1047 	mal_disable_rx_channel(dev->mal, dev->mal_rx_chan);
1048 
1049 	if (dev->rx_sg_skb) {
1050 		++dev->estats.rx_dropped_resize;
1051 		dev_kfree_skb(dev->rx_sg_skb);
1052 		dev->rx_sg_skb = NULL;
1053 	}
1054 
1055 	/* Make a first pass over RX ring and mark BDs ready, dropping
1056 	 * non-processed packets on the way. We need this as a separate pass
1057 	 * to simplify error recovery in the case of allocation failure later.
1058 	 */
1059 	for (i = 0; i < NUM_RX_BUFF; ++i) {
1060 		if (dev->rx_desc[i].ctrl & MAL_RX_CTRL_FIRST)
1061 			++dev->estats.rx_dropped_resize;
1062 
1063 		dev->rx_desc[i].data_len = 0;
1064 		dev->rx_desc[i].ctrl = MAL_RX_CTRL_EMPTY |
1065 		    (i == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1066 	}
1067 
1068 	/* Reallocate RX ring only if bigger skb buffers are required */
1069 	if (rx_skb_size <= dev->rx_skb_size)
1070 		goto skip;
1071 
1072 	/* Second pass, allocate new skbs */
1073 	for (i = 0; i < NUM_RX_BUFF; ++i) {
1074 		struct sk_buff *skb = alloc_skb(rx_skb_size, GFP_ATOMIC);
1075 		if (!skb) {
1076 			ret = -ENOMEM;
1077 			goto oom;
1078 		}
1079 
1080 		BUG_ON(!dev->rx_skb[i]);
1081 		dev_kfree_skb(dev->rx_skb[i]);
1082 
1083 		skb_reserve(skb, EMAC_RX_SKB_HEADROOM + 2);
1084 		dev->rx_desc[i].data_ptr =
1085 		    dma_map_single(&dev->ofdev->dev, skb->data - 2, rx_sync_size,
1086 				   DMA_FROM_DEVICE) + 2;
1087 		dev->rx_skb[i] = skb;
1088 	}
1089  skip:
1090 	/* Check if we need to change "Jumbo" bit in MR1 */
1091 	if (emac_has_feature(dev, EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE)) {
1092 		mr1_jumbo_bit_change = (new_mtu > ETH_DATA_LEN) ||
1093 				(dev->ndev->mtu > ETH_DATA_LEN);
1094 	} else {
1095 		mr1_jumbo_bit_change = (new_mtu > ETH_DATA_LEN) ^
1096 				(dev->ndev->mtu > ETH_DATA_LEN);
1097 	}
1098 
1099 	if (mr1_jumbo_bit_change) {
1100 		/* This is to prevent starting RX channel in emac_rx_enable() */
1101 		set_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1102 
1103 		dev->ndev->mtu = new_mtu;
1104 		emac_full_tx_reset(dev);
1105 	}
1106 
1107 	mal_set_rcbs(dev->mal, dev->mal_rx_chan, emac_rx_size(new_mtu));
1108  oom:
1109 	/* Restart RX */
1110 	clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1111 	dev->rx_slot = 0;
1112 	mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1113 	emac_rx_enable(dev);
1114 	emac_netif_start(dev);
1115 	mutex_unlock(&dev->link_lock);
1116 
1117 	return ret;
1118 }
1119 
1120 /* Process ctx, rtnl_lock semaphore */
1121 static int emac_change_mtu(struct net_device *ndev, int new_mtu)
1122 {
1123 	struct emac_instance *dev = netdev_priv(ndev);
1124 	int ret = 0;
1125 
1126 	DBG(dev, "change_mtu(%d)" NL, new_mtu);
1127 
1128 	if (netif_running(ndev)) {
1129 		/* Check if we really need to reinitialize RX ring */
1130 		if (emac_rx_skb_size(ndev->mtu) != emac_rx_skb_size(new_mtu))
1131 			ret = emac_resize_rx_ring(dev, new_mtu);
1132 	}
1133 
1134 	if (!ret) {
1135 		ndev->mtu = new_mtu;
1136 		dev->rx_skb_size = emac_rx_skb_size(new_mtu);
1137 		dev->rx_sync_size = emac_rx_sync_size(new_mtu);
1138 	}
1139 
1140 	return ret;
1141 }
1142 
1143 static void emac_clean_tx_ring(struct emac_instance *dev)
1144 {
1145 	int i;
1146 
1147 	for (i = 0; i < NUM_TX_BUFF; ++i) {
1148 		if (dev->tx_skb[i]) {
1149 			dev_kfree_skb(dev->tx_skb[i]);
1150 			dev->tx_skb[i] = NULL;
1151 			if (dev->tx_desc[i].ctrl & MAL_TX_CTRL_READY)
1152 				++dev->estats.tx_dropped;
1153 		}
1154 		dev->tx_desc[i].ctrl = 0;
1155 		dev->tx_desc[i].data_ptr = 0;
1156 	}
1157 }
1158 
1159 static void emac_clean_rx_ring(struct emac_instance *dev)
1160 {
1161 	int i;
1162 
1163 	for (i = 0; i < NUM_RX_BUFF; ++i)
1164 		if (dev->rx_skb[i]) {
1165 			dev->rx_desc[i].ctrl = 0;
1166 			dev_kfree_skb(dev->rx_skb[i]);
1167 			dev->rx_skb[i] = NULL;
1168 			dev->rx_desc[i].data_ptr = 0;
1169 		}
1170 
1171 	if (dev->rx_sg_skb) {
1172 		dev_kfree_skb(dev->rx_sg_skb);
1173 		dev->rx_sg_skb = NULL;
1174 	}
1175 }
1176 
1177 static inline int emac_alloc_rx_skb(struct emac_instance *dev, int slot,
1178 				    gfp_t flags)
1179 {
1180 	struct sk_buff *skb = alloc_skb(dev->rx_skb_size, flags);
1181 	if (unlikely(!skb))
1182 		return -ENOMEM;
1183 
1184 	dev->rx_skb[slot] = skb;
1185 	dev->rx_desc[slot].data_len = 0;
1186 
1187 	skb_reserve(skb, EMAC_RX_SKB_HEADROOM + 2);
1188 	dev->rx_desc[slot].data_ptr =
1189 	    dma_map_single(&dev->ofdev->dev, skb->data - 2, dev->rx_sync_size,
1190 			   DMA_FROM_DEVICE) + 2;
1191 	wmb();
1192 	dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1193 	    (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1194 
1195 	return 0;
1196 }
1197 
1198 static void emac_print_link_status(struct emac_instance *dev)
1199 {
1200 	if (netif_carrier_ok(dev->ndev))
1201 		printk(KERN_INFO "%s: link is up, %d %s%s\n",
1202 		       dev->ndev->name, dev->phy.speed,
1203 		       dev->phy.duplex == DUPLEX_FULL ? "FDX" : "HDX",
1204 		       dev->phy.pause ? ", pause enabled" :
1205 		       dev->phy.asym_pause ? ", asymmetric pause enabled" : "");
1206 	else
1207 		printk(KERN_INFO "%s: link is down\n", dev->ndev->name);
1208 }
1209 
1210 /* Process ctx, rtnl_lock semaphore */
1211 static int emac_open(struct net_device *ndev)
1212 {
1213 	struct emac_instance *dev = netdev_priv(ndev);
1214 	int err, i;
1215 
1216 	DBG(dev, "open" NL);
1217 
1218 	/* Setup error IRQ handler */
1219 	err = request_irq(dev->emac_irq, emac_irq, 0, "EMAC", dev);
1220 	if (err) {
1221 		printk(KERN_ERR "%s: failed to request IRQ %d\n",
1222 		       ndev->name, dev->emac_irq);
1223 		return err;
1224 	}
1225 
1226 	/* Allocate RX ring */
1227 	for (i = 0; i < NUM_RX_BUFF; ++i)
1228 		if (emac_alloc_rx_skb(dev, i, GFP_KERNEL)) {
1229 			printk(KERN_ERR "%s: failed to allocate RX ring\n",
1230 			       ndev->name);
1231 			goto oom;
1232 		}
1233 
1234 	dev->tx_cnt = dev->tx_slot = dev->ack_slot = dev->rx_slot = 0;
1235 	clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1236 	dev->rx_sg_skb = NULL;
1237 
1238 	mutex_lock(&dev->link_lock);
1239 	dev->opened = 1;
1240 
1241 	/* Start PHY polling now.
1242 	 */
1243 	if (dev->phy.address >= 0) {
1244 		int link_poll_interval;
1245 		if (dev->phy.def->ops->poll_link(&dev->phy)) {
1246 			dev->phy.def->ops->read_link(&dev->phy);
1247 			emac_rx_clk_default(dev);
1248 			netif_carrier_on(dev->ndev);
1249 			link_poll_interval = PHY_POLL_LINK_ON;
1250 		} else {
1251 			emac_rx_clk_tx(dev);
1252 			netif_carrier_off(dev->ndev);
1253 			link_poll_interval = PHY_POLL_LINK_OFF;
1254 		}
1255 		dev->link_polling = 1;
1256 		wmb();
1257 		schedule_delayed_work(&dev->link_work, link_poll_interval);
1258 		emac_print_link_status(dev);
1259 	} else
1260 		netif_carrier_on(dev->ndev);
1261 
1262 	/* Required for Pause packet support in EMAC */
1263 	dev_mc_add_global(ndev, default_mcast_addr);
1264 
1265 	emac_configure(dev);
1266 	mal_poll_add(dev->mal, &dev->commac);
1267 	mal_enable_tx_channel(dev->mal, dev->mal_tx_chan);
1268 	mal_set_rcbs(dev->mal, dev->mal_rx_chan, emac_rx_size(ndev->mtu));
1269 	mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1270 	emac_tx_enable(dev);
1271 	emac_rx_enable(dev);
1272 	emac_netif_start(dev);
1273 
1274 	mutex_unlock(&dev->link_lock);
1275 
1276 	return 0;
1277  oom:
1278 	emac_clean_rx_ring(dev);
1279 	free_irq(dev->emac_irq, dev);
1280 
1281 	return -ENOMEM;
1282 }
1283 
1284 /* BHs disabled */
1285 #if 0
1286 static int emac_link_differs(struct emac_instance *dev)
1287 {
1288 	u32 r = in_be32(&dev->emacp->mr1);
1289 
1290 	int duplex = r & EMAC_MR1_FDE ? DUPLEX_FULL : DUPLEX_HALF;
1291 	int speed, pause, asym_pause;
1292 
1293 	if (r & EMAC_MR1_MF_1000)
1294 		speed = SPEED_1000;
1295 	else if (r & EMAC_MR1_MF_100)
1296 		speed = SPEED_100;
1297 	else
1298 		speed = SPEED_10;
1299 
1300 	switch (r & (EMAC_MR1_EIFC | EMAC_MR1_APP)) {
1301 	case (EMAC_MR1_EIFC | EMAC_MR1_APP):
1302 		pause = 1;
1303 		asym_pause = 0;
1304 		break;
1305 	case EMAC_MR1_APP:
1306 		pause = 0;
1307 		asym_pause = 1;
1308 		break;
1309 	default:
1310 		pause = asym_pause = 0;
1311 	}
1312 	return speed != dev->phy.speed || duplex != dev->phy.duplex ||
1313 	    pause != dev->phy.pause || asym_pause != dev->phy.asym_pause;
1314 }
1315 #endif
1316 
1317 static void emac_link_timer(struct work_struct *work)
1318 {
1319 	struct emac_instance *dev =
1320 		container_of(to_delayed_work(work),
1321 			     struct emac_instance, link_work);
1322 	int link_poll_interval;
1323 
1324 	mutex_lock(&dev->link_lock);
1325 	DBG2(dev, "link timer" NL);
1326 
1327 	if (!dev->opened)
1328 		goto bail;
1329 
1330 	if (dev->phy.def->ops->poll_link(&dev->phy)) {
1331 		if (!netif_carrier_ok(dev->ndev)) {
1332 			emac_rx_clk_default(dev);
1333 			/* Get new link parameters */
1334 			dev->phy.def->ops->read_link(&dev->phy);
1335 
1336 			netif_carrier_on(dev->ndev);
1337 			emac_netif_stop(dev);
1338 			emac_full_tx_reset(dev);
1339 			emac_netif_start(dev);
1340 			emac_print_link_status(dev);
1341 		}
1342 		link_poll_interval = PHY_POLL_LINK_ON;
1343 	} else {
1344 		if (netif_carrier_ok(dev->ndev)) {
1345 			emac_rx_clk_tx(dev);
1346 			netif_carrier_off(dev->ndev);
1347 			netif_tx_disable(dev->ndev);
1348 			emac_reinitialize(dev);
1349 			emac_print_link_status(dev);
1350 		}
1351 		link_poll_interval = PHY_POLL_LINK_OFF;
1352 	}
1353 	schedule_delayed_work(&dev->link_work, link_poll_interval);
1354  bail:
1355 	mutex_unlock(&dev->link_lock);
1356 }
1357 
1358 static void emac_force_link_update(struct emac_instance *dev)
1359 {
1360 	netif_carrier_off(dev->ndev);
1361 	smp_rmb();
1362 	if (dev->link_polling) {
1363 		cancel_delayed_work_sync(&dev->link_work);
1364 		if (dev->link_polling)
1365 			schedule_delayed_work(&dev->link_work,  PHY_POLL_LINK_OFF);
1366 	}
1367 }
1368 
1369 /* Process ctx, rtnl_lock semaphore */
1370 static int emac_close(struct net_device *ndev)
1371 {
1372 	struct emac_instance *dev = netdev_priv(ndev);
1373 
1374 	DBG(dev, "close" NL);
1375 
1376 	if (dev->phy.address >= 0) {
1377 		dev->link_polling = 0;
1378 		cancel_delayed_work_sync(&dev->link_work);
1379 	}
1380 	mutex_lock(&dev->link_lock);
1381 	emac_netif_stop(dev);
1382 	dev->opened = 0;
1383 	mutex_unlock(&dev->link_lock);
1384 
1385 	emac_rx_disable(dev);
1386 	emac_tx_disable(dev);
1387 	mal_disable_rx_channel(dev->mal, dev->mal_rx_chan);
1388 	mal_disable_tx_channel(dev->mal, dev->mal_tx_chan);
1389 	mal_poll_del(dev->mal, &dev->commac);
1390 
1391 	emac_clean_tx_ring(dev);
1392 	emac_clean_rx_ring(dev);
1393 
1394 	free_irq(dev->emac_irq, dev);
1395 
1396 	netif_carrier_off(ndev);
1397 
1398 	return 0;
1399 }
1400 
1401 static inline u16 emac_tx_csum(struct emac_instance *dev,
1402 			       struct sk_buff *skb)
1403 {
1404 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH) &&
1405 		(skb->ip_summed == CHECKSUM_PARTIAL)) {
1406 		++dev->stats.tx_packets_csum;
1407 		return EMAC_TX_CTRL_TAH_CSUM;
1408 	}
1409 	return 0;
1410 }
1411 
1412 static inline int emac_xmit_finish(struct emac_instance *dev, int len)
1413 {
1414 	struct emac_regs __iomem *p = dev->emacp;
1415 	struct net_device *ndev = dev->ndev;
1416 
1417 	/* Send the packet out. If the if makes a significant perf
1418 	 * difference, then we can store the TMR0 value in "dev"
1419 	 * instead
1420 	 */
1421 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
1422 		out_be32(&p->tmr0, EMAC4_TMR0_XMIT);
1423 	else
1424 		out_be32(&p->tmr0, EMAC_TMR0_XMIT);
1425 
1426 	if (unlikely(++dev->tx_cnt == NUM_TX_BUFF)) {
1427 		netif_stop_queue(ndev);
1428 		DBG2(dev, "stopped TX queue" NL);
1429 	}
1430 
1431 	netif_trans_update(ndev);
1432 	++dev->stats.tx_packets;
1433 	dev->stats.tx_bytes += len;
1434 
1435 	return NETDEV_TX_OK;
1436 }
1437 
1438 /* Tx lock BH */
1439 static int emac_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1440 {
1441 	struct emac_instance *dev = netdev_priv(ndev);
1442 	unsigned int len = skb->len;
1443 	int slot;
1444 
1445 	u16 ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1446 	    MAL_TX_CTRL_LAST | emac_tx_csum(dev, skb);
1447 
1448 	slot = dev->tx_slot++;
1449 	if (dev->tx_slot == NUM_TX_BUFF) {
1450 		dev->tx_slot = 0;
1451 		ctrl |= MAL_TX_CTRL_WRAP;
1452 	}
1453 
1454 	DBG2(dev, "xmit(%u) %d" NL, len, slot);
1455 
1456 	dev->tx_skb[slot] = skb;
1457 	dev->tx_desc[slot].data_ptr = dma_map_single(&dev->ofdev->dev,
1458 						     skb->data, len,
1459 						     DMA_TO_DEVICE);
1460 	dev->tx_desc[slot].data_len = (u16) len;
1461 	wmb();
1462 	dev->tx_desc[slot].ctrl = ctrl;
1463 
1464 	return emac_xmit_finish(dev, len);
1465 }
1466 
1467 static inline int emac_xmit_split(struct emac_instance *dev, int slot,
1468 				  u32 pd, int len, int last, u16 base_ctrl)
1469 {
1470 	while (1) {
1471 		u16 ctrl = base_ctrl;
1472 		int chunk = min(len, MAL_MAX_TX_SIZE);
1473 		len -= chunk;
1474 
1475 		slot = (slot + 1) % NUM_TX_BUFF;
1476 
1477 		if (last && !len)
1478 			ctrl |= MAL_TX_CTRL_LAST;
1479 		if (slot == NUM_TX_BUFF - 1)
1480 			ctrl |= MAL_TX_CTRL_WRAP;
1481 
1482 		dev->tx_skb[slot] = NULL;
1483 		dev->tx_desc[slot].data_ptr = pd;
1484 		dev->tx_desc[slot].data_len = (u16) chunk;
1485 		dev->tx_desc[slot].ctrl = ctrl;
1486 		++dev->tx_cnt;
1487 
1488 		if (!len)
1489 			break;
1490 
1491 		pd += chunk;
1492 	}
1493 	return slot;
1494 }
1495 
1496 /* Tx lock BH disabled (SG version for TAH equipped EMACs) */
1497 static int emac_start_xmit_sg(struct sk_buff *skb, struct net_device *ndev)
1498 {
1499 	struct emac_instance *dev = netdev_priv(ndev);
1500 	int nr_frags = skb_shinfo(skb)->nr_frags;
1501 	int len = skb->len, chunk;
1502 	int slot, i;
1503 	u16 ctrl;
1504 	u32 pd;
1505 
1506 	/* This is common "fast" path */
1507 	if (likely(!nr_frags && len <= MAL_MAX_TX_SIZE))
1508 		return emac_start_xmit(skb, ndev);
1509 
1510 	len -= skb->data_len;
1511 
1512 	/* Note, this is only an *estimation*, we can still run out of empty
1513 	 * slots because of the additional fragmentation into
1514 	 * MAL_MAX_TX_SIZE-sized chunks
1515 	 */
1516 	if (unlikely(dev->tx_cnt + nr_frags + mal_tx_chunks(len) > NUM_TX_BUFF))
1517 		goto stop_queue;
1518 
1519 	ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1520 	    emac_tx_csum(dev, skb);
1521 	slot = dev->tx_slot;
1522 
1523 	/* skb data */
1524 	dev->tx_skb[slot] = NULL;
1525 	chunk = min(len, MAL_MAX_TX_SIZE);
1526 	dev->tx_desc[slot].data_ptr = pd =
1527 	    dma_map_single(&dev->ofdev->dev, skb->data, len, DMA_TO_DEVICE);
1528 	dev->tx_desc[slot].data_len = (u16) chunk;
1529 	len -= chunk;
1530 	if (unlikely(len))
1531 		slot = emac_xmit_split(dev, slot, pd + chunk, len, !nr_frags,
1532 				       ctrl);
1533 	/* skb fragments */
1534 	for (i = 0; i < nr_frags; ++i) {
1535 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1536 		len = skb_frag_size(frag);
1537 
1538 		if (unlikely(dev->tx_cnt + mal_tx_chunks(len) >= NUM_TX_BUFF))
1539 			goto undo_frame;
1540 
1541 		pd = skb_frag_dma_map(&dev->ofdev->dev, frag, 0, len,
1542 				      DMA_TO_DEVICE);
1543 
1544 		slot = emac_xmit_split(dev, slot, pd, len, i == nr_frags - 1,
1545 				       ctrl);
1546 	}
1547 
1548 	DBG2(dev, "xmit_sg(%u) %d - %d" NL, skb->len, dev->tx_slot, slot);
1549 
1550 	/* Attach skb to the last slot so we don't release it too early */
1551 	dev->tx_skb[slot] = skb;
1552 
1553 	/* Send the packet out */
1554 	if (dev->tx_slot == NUM_TX_BUFF - 1)
1555 		ctrl |= MAL_TX_CTRL_WRAP;
1556 	wmb();
1557 	dev->tx_desc[dev->tx_slot].ctrl = ctrl;
1558 	dev->tx_slot = (slot + 1) % NUM_TX_BUFF;
1559 
1560 	return emac_xmit_finish(dev, skb->len);
1561 
1562  undo_frame:
1563 	/* Well, too bad. Our previous estimation was overly optimistic.
1564 	 * Undo everything.
1565 	 */
1566 	while (slot != dev->tx_slot) {
1567 		dev->tx_desc[slot].ctrl = 0;
1568 		--dev->tx_cnt;
1569 		if (--slot < 0)
1570 			slot = NUM_TX_BUFF - 1;
1571 	}
1572 	++dev->estats.tx_undo;
1573 
1574  stop_queue:
1575 	netif_stop_queue(ndev);
1576 	DBG2(dev, "stopped TX queue" NL);
1577 	return NETDEV_TX_BUSY;
1578 }
1579 
1580 /* Tx lock BHs */
1581 static void emac_parse_tx_error(struct emac_instance *dev, u16 ctrl)
1582 {
1583 	struct emac_error_stats *st = &dev->estats;
1584 
1585 	DBG(dev, "BD TX error %04x" NL, ctrl);
1586 
1587 	++st->tx_bd_errors;
1588 	if (ctrl & EMAC_TX_ST_BFCS)
1589 		++st->tx_bd_bad_fcs;
1590 	if (ctrl & EMAC_TX_ST_LCS)
1591 		++st->tx_bd_carrier_loss;
1592 	if (ctrl & EMAC_TX_ST_ED)
1593 		++st->tx_bd_excessive_deferral;
1594 	if (ctrl & EMAC_TX_ST_EC)
1595 		++st->tx_bd_excessive_collisions;
1596 	if (ctrl & EMAC_TX_ST_LC)
1597 		++st->tx_bd_late_collision;
1598 	if (ctrl & EMAC_TX_ST_MC)
1599 		++st->tx_bd_multple_collisions;
1600 	if (ctrl & EMAC_TX_ST_SC)
1601 		++st->tx_bd_single_collision;
1602 	if (ctrl & EMAC_TX_ST_UR)
1603 		++st->tx_bd_underrun;
1604 	if (ctrl & EMAC_TX_ST_SQE)
1605 		++st->tx_bd_sqe;
1606 }
1607 
1608 static void emac_poll_tx(void *param)
1609 {
1610 	struct emac_instance *dev = param;
1611 	u32 bad_mask;
1612 
1613 	DBG2(dev, "poll_tx, %d %d" NL, dev->tx_cnt, dev->ack_slot);
1614 
1615 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
1616 		bad_mask = EMAC_IS_BAD_TX_TAH;
1617 	else
1618 		bad_mask = EMAC_IS_BAD_TX;
1619 
1620 	netif_tx_lock_bh(dev->ndev);
1621 	if (dev->tx_cnt) {
1622 		u16 ctrl;
1623 		int slot = dev->ack_slot, n = 0;
1624 	again:
1625 		ctrl = dev->tx_desc[slot].ctrl;
1626 		if (!(ctrl & MAL_TX_CTRL_READY)) {
1627 			struct sk_buff *skb = dev->tx_skb[slot];
1628 			++n;
1629 
1630 			if (skb) {
1631 				dev_kfree_skb(skb);
1632 				dev->tx_skb[slot] = NULL;
1633 			}
1634 			slot = (slot + 1) % NUM_TX_BUFF;
1635 
1636 			if (unlikely(ctrl & bad_mask))
1637 				emac_parse_tx_error(dev, ctrl);
1638 
1639 			if (--dev->tx_cnt)
1640 				goto again;
1641 		}
1642 		if (n) {
1643 			dev->ack_slot = slot;
1644 			if (netif_queue_stopped(dev->ndev) &&
1645 			    dev->tx_cnt < EMAC_TX_WAKEUP_THRESH)
1646 				netif_wake_queue(dev->ndev);
1647 
1648 			DBG2(dev, "tx %d pkts" NL, n);
1649 		}
1650 	}
1651 	netif_tx_unlock_bh(dev->ndev);
1652 }
1653 
1654 static inline void emac_recycle_rx_skb(struct emac_instance *dev, int slot,
1655 				       int len)
1656 {
1657 	struct sk_buff *skb = dev->rx_skb[slot];
1658 
1659 	DBG2(dev, "recycle %d %d" NL, slot, len);
1660 
1661 	if (len)
1662 		dma_map_single(&dev->ofdev->dev, skb->data - 2,
1663 			       EMAC_DMA_ALIGN(len + 2), DMA_FROM_DEVICE);
1664 
1665 	dev->rx_desc[slot].data_len = 0;
1666 	wmb();
1667 	dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1668 	    (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1669 }
1670 
1671 static void emac_parse_rx_error(struct emac_instance *dev, u16 ctrl)
1672 {
1673 	struct emac_error_stats *st = &dev->estats;
1674 
1675 	DBG(dev, "BD RX error %04x" NL, ctrl);
1676 
1677 	++st->rx_bd_errors;
1678 	if (ctrl & EMAC_RX_ST_OE)
1679 		++st->rx_bd_overrun;
1680 	if (ctrl & EMAC_RX_ST_BP)
1681 		++st->rx_bd_bad_packet;
1682 	if (ctrl & EMAC_RX_ST_RP)
1683 		++st->rx_bd_runt_packet;
1684 	if (ctrl & EMAC_RX_ST_SE)
1685 		++st->rx_bd_short_event;
1686 	if (ctrl & EMAC_RX_ST_AE)
1687 		++st->rx_bd_alignment_error;
1688 	if (ctrl & EMAC_RX_ST_BFCS)
1689 		++st->rx_bd_bad_fcs;
1690 	if (ctrl & EMAC_RX_ST_PTL)
1691 		++st->rx_bd_packet_too_long;
1692 	if (ctrl & EMAC_RX_ST_ORE)
1693 		++st->rx_bd_out_of_range;
1694 	if (ctrl & EMAC_RX_ST_IRE)
1695 		++st->rx_bd_in_range;
1696 }
1697 
1698 static inline void emac_rx_csum(struct emac_instance *dev,
1699 				struct sk_buff *skb, u16 ctrl)
1700 {
1701 #ifdef CONFIG_IBM_EMAC_TAH
1702 	if (!ctrl && dev->tah_dev) {
1703 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1704 		++dev->stats.rx_packets_csum;
1705 	}
1706 #endif
1707 }
1708 
1709 static inline int emac_rx_sg_append(struct emac_instance *dev, int slot)
1710 {
1711 	if (likely(dev->rx_sg_skb != NULL)) {
1712 		int len = dev->rx_desc[slot].data_len;
1713 		int tot_len = dev->rx_sg_skb->len + len;
1714 
1715 		if (unlikely(tot_len + 2 > dev->rx_skb_size)) {
1716 			++dev->estats.rx_dropped_mtu;
1717 			dev_kfree_skb(dev->rx_sg_skb);
1718 			dev->rx_sg_skb = NULL;
1719 		} else {
1720 			memcpy(skb_tail_pointer(dev->rx_sg_skb),
1721 					 dev->rx_skb[slot]->data, len);
1722 			skb_put(dev->rx_sg_skb, len);
1723 			emac_recycle_rx_skb(dev, slot, len);
1724 			return 0;
1725 		}
1726 	}
1727 	emac_recycle_rx_skb(dev, slot, 0);
1728 	return -1;
1729 }
1730 
1731 /* NAPI poll context */
1732 static int emac_poll_rx(void *param, int budget)
1733 {
1734 	struct emac_instance *dev = param;
1735 	int slot = dev->rx_slot, received = 0;
1736 
1737 	DBG2(dev, "poll_rx(%d)" NL, budget);
1738 
1739  again:
1740 	while (budget > 0) {
1741 		int len;
1742 		struct sk_buff *skb;
1743 		u16 ctrl = dev->rx_desc[slot].ctrl;
1744 
1745 		if (ctrl & MAL_RX_CTRL_EMPTY)
1746 			break;
1747 
1748 		skb = dev->rx_skb[slot];
1749 		mb();
1750 		len = dev->rx_desc[slot].data_len;
1751 
1752 		if (unlikely(!MAL_IS_SINGLE_RX(ctrl)))
1753 			goto sg;
1754 
1755 		ctrl &= EMAC_BAD_RX_MASK;
1756 		if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1757 			emac_parse_rx_error(dev, ctrl);
1758 			++dev->estats.rx_dropped_error;
1759 			emac_recycle_rx_skb(dev, slot, 0);
1760 			len = 0;
1761 			goto next;
1762 		}
1763 
1764 		if (len < ETH_HLEN) {
1765 			++dev->estats.rx_dropped_stack;
1766 			emac_recycle_rx_skb(dev, slot, len);
1767 			goto next;
1768 		}
1769 
1770 		if (len && len < EMAC_RX_COPY_THRESH) {
1771 			struct sk_buff *copy_skb =
1772 			    alloc_skb(len + EMAC_RX_SKB_HEADROOM + 2, GFP_ATOMIC);
1773 			if (unlikely(!copy_skb))
1774 				goto oom;
1775 
1776 			skb_reserve(copy_skb, EMAC_RX_SKB_HEADROOM + 2);
1777 			memcpy(copy_skb->data - 2, skb->data - 2, len + 2);
1778 			emac_recycle_rx_skb(dev, slot, len);
1779 			skb = copy_skb;
1780 		} else if (unlikely(emac_alloc_rx_skb(dev, slot, GFP_ATOMIC)))
1781 			goto oom;
1782 
1783 		skb_put(skb, len);
1784 	push_packet:
1785 		skb->protocol = eth_type_trans(skb, dev->ndev);
1786 		emac_rx_csum(dev, skb, ctrl);
1787 
1788 		if (unlikely(netif_receive_skb(skb) == NET_RX_DROP))
1789 			++dev->estats.rx_dropped_stack;
1790 	next:
1791 		++dev->stats.rx_packets;
1792 	skip:
1793 		dev->stats.rx_bytes += len;
1794 		slot = (slot + 1) % NUM_RX_BUFF;
1795 		--budget;
1796 		++received;
1797 		continue;
1798 	sg:
1799 		if (ctrl & MAL_RX_CTRL_FIRST) {
1800 			BUG_ON(dev->rx_sg_skb);
1801 			if (unlikely(emac_alloc_rx_skb(dev, slot, GFP_ATOMIC))) {
1802 				DBG(dev, "rx OOM %d" NL, slot);
1803 				++dev->estats.rx_dropped_oom;
1804 				emac_recycle_rx_skb(dev, slot, 0);
1805 			} else {
1806 				dev->rx_sg_skb = skb;
1807 				skb_put(skb, len);
1808 			}
1809 		} else if (!emac_rx_sg_append(dev, slot) &&
1810 			   (ctrl & MAL_RX_CTRL_LAST)) {
1811 
1812 			skb = dev->rx_sg_skb;
1813 			dev->rx_sg_skb = NULL;
1814 
1815 			ctrl &= EMAC_BAD_RX_MASK;
1816 			if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1817 				emac_parse_rx_error(dev, ctrl);
1818 				++dev->estats.rx_dropped_error;
1819 				dev_kfree_skb(skb);
1820 				len = 0;
1821 			} else
1822 				goto push_packet;
1823 		}
1824 		goto skip;
1825 	oom:
1826 		DBG(dev, "rx OOM %d" NL, slot);
1827 		/* Drop the packet and recycle skb */
1828 		++dev->estats.rx_dropped_oom;
1829 		emac_recycle_rx_skb(dev, slot, 0);
1830 		goto next;
1831 	}
1832 
1833 	if (received) {
1834 		DBG2(dev, "rx %d BDs" NL, received);
1835 		dev->rx_slot = slot;
1836 	}
1837 
1838 	if (unlikely(budget && test_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags))) {
1839 		mb();
1840 		if (!(dev->rx_desc[slot].ctrl & MAL_RX_CTRL_EMPTY)) {
1841 			DBG2(dev, "rx restart" NL);
1842 			received = 0;
1843 			goto again;
1844 		}
1845 
1846 		if (dev->rx_sg_skb) {
1847 			DBG2(dev, "dropping partial rx packet" NL);
1848 			++dev->estats.rx_dropped_error;
1849 			dev_kfree_skb(dev->rx_sg_skb);
1850 			dev->rx_sg_skb = NULL;
1851 		}
1852 
1853 		clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1854 		mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1855 		emac_rx_enable(dev);
1856 		dev->rx_slot = 0;
1857 	}
1858 	return received;
1859 }
1860 
1861 /* NAPI poll context */
1862 static int emac_peek_rx(void *param)
1863 {
1864 	struct emac_instance *dev = param;
1865 
1866 	return !(dev->rx_desc[dev->rx_slot].ctrl & MAL_RX_CTRL_EMPTY);
1867 }
1868 
1869 /* NAPI poll context */
1870 static int emac_peek_rx_sg(void *param)
1871 {
1872 	struct emac_instance *dev = param;
1873 
1874 	int slot = dev->rx_slot;
1875 	while (1) {
1876 		u16 ctrl = dev->rx_desc[slot].ctrl;
1877 		if (ctrl & MAL_RX_CTRL_EMPTY)
1878 			return 0;
1879 		else if (ctrl & MAL_RX_CTRL_LAST)
1880 			return 1;
1881 
1882 		slot = (slot + 1) % NUM_RX_BUFF;
1883 
1884 		/* I'm just being paranoid here :) */
1885 		if (unlikely(slot == dev->rx_slot))
1886 			return 0;
1887 	}
1888 }
1889 
1890 /* Hard IRQ */
1891 static void emac_rxde(void *param)
1892 {
1893 	struct emac_instance *dev = param;
1894 
1895 	++dev->estats.rx_stopped;
1896 	emac_rx_disable_async(dev);
1897 }
1898 
1899 /* Hard IRQ */
1900 static irqreturn_t emac_irq(int irq, void *dev_instance)
1901 {
1902 	struct emac_instance *dev = dev_instance;
1903 	struct emac_regs __iomem *p = dev->emacp;
1904 	struct emac_error_stats *st = &dev->estats;
1905 	u32 isr;
1906 
1907 	spin_lock(&dev->lock);
1908 
1909 	isr = in_be32(&p->isr);
1910 	out_be32(&p->isr, isr);
1911 
1912 	DBG(dev, "isr = %08x" NL, isr);
1913 
1914 	if (isr & EMAC4_ISR_TXPE)
1915 		++st->tx_parity;
1916 	if (isr & EMAC4_ISR_RXPE)
1917 		++st->rx_parity;
1918 	if (isr & EMAC4_ISR_TXUE)
1919 		++st->tx_underrun;
1920 	if (isr & EMAC4_ISR_RXOE)
1921 		++st->rx_fifo_overrun;
1922 	if (isr & EMAC_ISR_OVR)
1923 		++st->rx_overrun;
1924 	if (isr & EMAC_ISR_BP)
1925 		++st->rx_bad_packet;
1926 	if (isr & EMAC_ISR_RP)
1927 		++st->rx_runt_packet;
1928 	if (isr & EMAC_ISR_SE)
1929 		++st->rx_short_event;
1930 	if (isr & EMAC_ISR_ALE)
1931 		++st->rx_alignment_error;
1932 	if (isr & EMAC_ISR_BFCS)
1933 		++st->rx_bad_fcs;
1934 	if (isr & EMAC_ISR_PTLE)
1935 		++st->rx_packet_too_long;
1936 	if (isr & EMAC_ISR_ORE)
1937 		++st->rx_out_of_range;
1938 	if (isr & EMAC_ISR_IRE)
1939 		++st->rx_in_range;
1940 	if (isr & EMAC_ISR_SQE)
1941 		++st->tx_sqe;
1942 	if (isr & EMAC_ISR_TE)
1943 		++st->tx_errors;
1944 
1945 	spin_unlock(&dev->lock);
1946 
1947 	return IRQ_HANDLED;
1948 }
1949 
1950 static struct net_device_stats *emac_stats(struct net_device *ndev)
1951 {
1952 	struct emac_instance *dev = netdev_priv(ndev);
1953 	struct emac_stats *st = &dev->stats;
1954 	struct emac_error_stats *est = &dev->estats;
1955 	struct net_device_stats *nst = &ndev->stats;
1956 	unsigned long flags;
1957 
1958 	DBG2(dev, "stats" NL);
1959 
1960 	/* Compute "legacy" statistics */
1961 	spin_lock_irqsave(&dev->lock, flags);
1962 	nst->rx_packets = (unsigned long)st->rx_packets;
1963 	nst->rx_bytes = (unsigned long)st->rx_bytes;
1964 	nst->tx_packets = (unsigned long)st->tx_packets;
1965 	nst->tx_bytes = (unsigned long)st->tx_bytes;
1966 	nst->rx_dropped = (unsigned long)(est->rx_dropped_oom +
1967 					  est->rx_dropped_error +
1968 					  est->rx_dropped_resize +
1969 					  est->rx_dropped_mtu);
1970 	nst->tx_dropped = (unsigned long)est->tx_dropped;
1971 
1972 	nst->rx_errors = (unsigned long)est->rx_bd_errors;
1973 	nst->rx_fifo_errors = (unsigned long)(est->rx_bd_overrun +
1974 					      est->rx_fifo_overrun +
1975 					      est->rx_overrun);
1976 	nst->rx_frame_errors = (unsigned long)(est->rx_bd_alignment_error +
1977 					       est->rx_alignment_error);
1978 	nst->rx_crc_errors = (unsigned long)(est->rx_bd_bad_fcs +
1979 					     est->rx_bad_fcs);
1980 	nst->rx_length_errors = (unsigned long)(est->rx_bd_runt_packet +
1981 						est->rx_bd_short_event +
1982 						est->rx_bd_packet_too_long +
1983 						est->rx_bd_out_of_range +
1984 						est->rx_bd_in_range +
1985 						est->rx_runt_packet +
1986 						est->rx_short_event +
1987 						est->rx_packet_too_long +
1988 						est->rx_out_of_range +
1989 						est->rx_in_range);
1990 
1991 	nst->tx_errors = (unsigned long)(est->tx_bd_errors + est->tx_errors);
1992 	nst->tx_fifo_errors = (unsigned long)(est->tx_bd_underrun +
1993 					      est->tx_underrun);
1994 	nst->tx_carrier_errors = (unsigned long)est->tx_bd_carrier_loss;
1995 	nst->collisions = (unsigned long)(est->tx_bd_excessive_deferral +
1996 					  est->tx_bd_excessive_collisions +
1997 					  est->tx_bd_late_collision +
1998 					  est->tx_bd_multple_collisions);
1999 	spin_unlock_irqrestore(&dev->lock, flags);
2000 	return nst;
2001 }
2002 
2003 static struct mal_commac_ops emac_commac_ops = {
2004 	.poll_tx = &emac_poll_tx,
2005 	.poll_rx = &emac_poll_rx,
2006 	.peek_rx = &emac_peek_rx,
2007 	.rxde = &emac_rxde,
2008 };
2009 
2010 static struct mal_commac_ops emac_commac_sg_ops = {
2011 	.poll_tx = &emac_poll_tx,
2012 	.poll_rx = &emac_poll_rx,
2013 	.peek_rx = &emac_peek_rx_sg,
2014 	.rxde = &emac_rxde,
2015 };
2016 
2017 /* Ethtool support */
2018 static int emac_ethtool_get_link_ksettings(struct net_device *ndev,
2019 					   struct ethtool_link_ksettings *cmd)
2020 {
2021 	struct emac_instance *dev = netdev_priv(ndev);
2022 	u32 supported, advertising;
2023 
2024 	supported = dev->phy.features;
2025 	cmd->base.port = PORT_MII;
2026 	cmd->base.phy_address = dev->phy.address;
2027 
2028 	mutex_lock(&dev->link_lock);
2029 	advertising = dev->phy.advertising;
2030 	cmd->base.autoneg = dev->phy.autoneg;
2031 	cmd->base.speed = dev->phy.speed;
2032 	cmd->base.duplex = dev->phy.duplex;
2033 	mutex_unlock(&dev->link_lock);
2034 
2035 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
2036 						supported);
2037 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
2038 						advertising);
2039 
2040 	return 0;
2041 }
2042 
2043 static int
2044 emac_ethtool_set_link_ksettings(struct net_device *ndev,
2045 				const struct ethtool_link_ksettings *cmd)
2046 {
2047 	struct emac_instance *dev = netdev_priv(ndev);
2048 	u32 f = dev->phy.features;
2049 	u32 advertising;
2050 
2051 	ethtool_convert_link_mode_to_legacy_u32(&advertising,
2052 						cmd->link_modes.advertising);
2053 
2054 	DBG(dev, "set_settings(%d, %d, %d, 0x%08x)" NL,
2055 	    cmd->base.autoneg, cmd->base.speed, cmd->base.duplex, advertising);
2056 
2057 	/* Basic sanity checks */
2058 	if (dev->phy.address < 0)
2059 		return -EOPNOTSUPP;
2060 	if (cmd->base.autoneg != AUTONEG_ENABLE &&
2061 	    cmd->base.autoneg != AUTONEG_DISABLE)
2062 		return -EINVAL;
2063 	if (cmd->base.autoneg == AUTONEG_ENABLE && advertising == 0)
2064 		return -EINVAL;
2065 	if (cmd->base.duplex != DUPLEX_HALF && cmd->base.duplex != DUPLEX_FULL)
2066 		return -EINVAL;
2067 
2068 	if (cmd->base.autoneg == AUTONEG_DISABLE) {
2069 		switch (cmd->base.speed) {
2070 		case SPEED_10:
2071 			if (cmd->base.duplex == DUPLEX_HALF &&
2072 			    !(f & SUPPORTED_10baseT_Half))
2073 				return -EINVAL;
2074 			if (cmd->base.duplex == DUPLEX_FULL &&
2075 			    !(f & SUPPORTED_10baseT_Full))
2076 				return -EINVAL;
2077 			break;
2078 		case SPEED_100:
2079 			if (cmd->base.duplex == DUPLEX_HALF &&
2080 			    !(f & SUPPORTED_100baseT_Half))
2081 				return -EINVAL;
2082 			if (cmd->base.duplex == DUPLEX_FULL &&
2083 			    !(f & SUPPORTED_100baseT_Full))
2084 				return -EINVAL;
2085 			break;
2086 		case SPEED_1000:
2087 			if (cmd->base.duplex == DUPLEX_HALF &&
2088 			    !(f & SUPPORTED_1000baseT_Half))
2089 				return -EINVAL;
2090 			if (cmd->base.duplex == DUPLEX_FULL &&
2091 			    !(f & SUPPORTED_1000baseT_Full))
2092 				return -EINVAL;
2093 			break;
2094 		default:
2095 			return -EINVAL;
2096 		}
2097 
2098 		mutex_lock(&dev->link_lock);
2099 		dev->phy.def->ops->setup_forced(&dev->phy, cmd->base.speed,
2100 						cmd->base.duplex);
2101 		mutex_unlock(&dev->link_lock);
2102 
2103 	} else {
2104 		if (!(f & SUPPORTED_Autoneg))
2105 			return -EINVAL;
2106 
2107 		mutex_lock(&dev->link_lock);
2108 		dev->phy.def->ops->setup_aneg(&dev->phy,
2109 					      (advertising & f) |
2110 					      (dev->phy.advertising &
2111 					       (ADVERTISED_Pause |
2112 						ADVERTISED_Asym_Pause)));
2113 		mutex_unlock(&dev->link_lock);
2114 	}
2115 	emac_force_link_update(dev);
2116 
2117 	return 0;
2118 }
2119 
2120 static void emac_ethtool_get_ringparam(struct net_device *ndev,
2121 				       struct ethtool_ringparam *rp)
2122 {
2123 	rp->rx_max_pending = rp->rx_pending = NUM_RX_BUFF;
2124 	rp->tx_max_pending = rp->tx_pending = NUM_TX_BUFF;
2125 }
2126 
2127 static void emac_ethtool_get_pauseparam(struct net_device *ndev,
2128 					struct ethtool_pauseparam *pp)
2129 {
2130 	struct emac_instance *dev = netdev_priv(ndev);
2131 
2132 	mutex_lock(&dev->link_lock);
2133 	if ((dev->phy.features & SUPPORTED_Autoneg) &&
2134 	    (dev->phy.advertising & (ADVERTISED_Pause | ADVERTISED_Asym_Pause)))
2135 		pp->autoneg = 1;
2136 
2137 	if (dev->phy.duplex == DUPLEX_FULL) {
2138 		if (dev->phy.pause)
2139 			pp->rx_pause = pp->tx_pause = 1;
2140 		else if (dev->phy.asym_pause)
2141 			pp->tx_pause = 1;
2142 	}
2143 	mutex_unlock(&dev->link_lock);
2144 }
2145 
2146 static int emac_get_regs_len(struct emac_instance *dev)
2147 {
2148 		return sizeof(struct emac_ethtool_regs_subhdr) +
2149 			sizeof(struct emac_regs);
2150 }
2151 
2152 static int emac_ethtool_get_regs_len(struct net_device *ndev)
2153 {
2154 	struct emac_instance *dev = netdev_priv(ndev);
2155 	int size;
2156 
2157 	size = sizeof(struct emac_ethtool_regs_hdr) +
2158 		emac_get_regs_len(dev) + mal_get_regs_len(dev->mal);
2159 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
2160 		size += zmii_get_regs_len(dev->zmii_dev);
2161 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
2162 		size += rgmii_get_regs_len(dev->rgmii_dev);
2163 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
2164 		size += tah_get_regs_len(dev->tah_dev);
2165 
2166 	return size;
2167 }
2168 
2169 static void *emac_dump_regs(struct emac_instance *dev, void *buf)
2170 {
2171 	struct emac_ethtool_regs_subhdr *hdr = buf;
2172 
2173 	hdr->index = dev->cell_index;
2174 	if (emac_has_feature(dev, EMAC_FTR_EMAC4SYNC)) {
2175 		hdr->version = EMAC4SYNC_ETHTOOL_REGS_VER;
2176 	} else if (emac_has_feature(dev, EMAC_FTR_EMAC4)) {
2177 		hdr->version = EMAC4_ETHTOOL_REGS_VER;
2178 	} else {
2179 		hdr->version = EMAC_ETHTOOL_REGS_VER;
2180 	}
2181 	memcpy_fromio(hdr + 1, dev->emacp, sizeof(struct emac_regs));
2182 	return (void *)(hdr + 1) + sizeof(struct emac_regs);
2183 }
2184 
2185 static void emac_ethtool_get_regs(struct net_device *ndev,
2186 				  struct ethtool_regs *regs, void *buf)
2187 {
2188 	struct emac_instance *dev = netdev_priv(ndev);
2189 	struct emac_ethtool_regs_hdr *hdr = buf;
2190 
2191 	hdr->components = 0;
2192 	buf = hdr + 1;
2193 
2194 	buf = mal_dump_regs(dev->mal, buf);
2195 	buf = emac_dump_regs(dev, buf);
2196 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII)) {
2197 		hdr->components |= EMAC_ETHTOOL_REGS_ZMII;
2198 		buf = zmii_dump_regs(dev->zmii_dev, buf);
2199 	}
2200 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII)) {
2201 		hdr->components |= EMAC_ETHTOOL_REGS_RGMII;
2202 		buf = rgmii_dump_regs(dev->rgmii_dev, buf);
2203 	}
2204 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH)) {
2205 		hdr->components |= EMAC_ETHTOOL_REGS_TAH;
2206 		buf = tah_dump_regs(dev->tah_dev, buf);
2207 	}
2208 }
2209 
2210 static int emac_ethtool_nway_reset(struct net_device *ndev)
2211 {
2212 	struct emac_instance *dev = netdev_priv(ndev);
2213 	int res = 0;
2214 
2215 	DBG(dev, "nway_reset" NL);
2216 
2217 	if (dev->phy.address < 0)
2218 		return -EOPNOTSUPP;
2219 
2220 	mutex_lock(&dev->link_lock);
2221 	if (!dev->phy.autoneg) {
2222 		res = -EINVAL;
2223 		goto out;
2224 	}
2225 
2226 	dev->phy.def->ops->setup_aneg(&dev->phy, dev->phy.advertising);
2227  out:
2228 	mutex_unlock(&dev->link_lock);
2229 	emac_force_link_update(dev);
2230 	return res;
2231 }
2232 
2233 static int emac_ethtool_get_sset_count(struct net_device *ndev, int stringset)
2234 {
2235 	if (stringset == ETH_SS_STATS)
2236 		return EMAC_ETHTOOL_STATS_COUNT;
2237 	else
2238 		return -EINVAL;
2239 }
2240 
2241 static void emac_ethtool_get_strings(struct net_device *ndev, u32 stringset,
2242 				     u8 * buf)
2243 {
2244 	if (stringset == ETH_SS_STATS)
2245 		memcpy(buf, &emac_stats_keys, sizeof(emac_stats_keys));
2246 }
2247 
2248 static void emac_ethtool_get_ethtool_stats(struct net_device *ndev,
2249 					   struct ethtool_stats *estats,
2250 					   u64 * tmp_stats)
2251 {
2252 	struct emac_instance *dev = netdev_priv(ndev);
2253 
2254 	memcpy(tmp_stats, &dev->stats, sizeof(dev->stats));
2255 	tmp_stats += sizeof(dev->stats) / sizeof(u64);
2256 	memcpy(tmp_stats, &dev->estats, sizeof(dev->estats));
2257 }
2258 
2259 static void emac_ethtool_get_drvinfo(struct net_device *ndev,
2260 				     struct ethtool_drvinfo *info)
2261 {
2262 	struct emac_instance *dev = netdev_priv(ndev);
2263 
2264 	strlcpy(info->driver, "ibm_emac", sizeof(info->driver));
2265 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2266 	snprintf(info->bus_info, sizeof(info->bus_info), "PPC 4xx EMAC-%d %pOF",
2267 		 dev->cell_index, dev->ofdev->dev.of_node);
2268 }
2269 
2270 static const struct ethtool_ops emac_ethtool_ops = {
2271 	.get_drvinfo = emac_ethtool_get_drvinfo,
2272 
2273 	.get_regs_len = emac_ethtool_get_regs_len,
2274 	.get_regs = emac_ethtool_get_regs,
2275 
2276 	.nway_reset = emac_ethtool_nway_reset,
2277 
2278 	.get_ringparam = emac_ethtool_get_ringparam,
2279 	.get_pauseparam = emac_ethtool_get_pauseparam,
2280 
2281 	.get_strings = emac_ethtool_get_strings,
2282 	.get_sset_count = emac_ethtool_get_sset_count,
2283 	.get_ethtool_stats = emac_ethtool_get_ethtool_stats,
2284 
2285 	.get_link = ethtool_op_get_link,
2286 	.get_link_ksettings = emac_ethtool_get_link_ksettings,
2287 	.set_link_ksettings = emac_ethtool_set_link_ksettings,
2288 };
2289 
2290 static int emac_ioctl(struct net_device *ndev, struct ifreq *rq, int cmd)
2291 {
2292 	struct emac_instance *dev = netdev_priv(ndev);
2293 	struct mii_ioctl_data *data = if_mii(rq);
2294 
2295 	DBG(dev, "ioctl %08x" NL, cmd);
2296 
2297 	if (dev->phy.address < 0)
2298 		return -EOPNOTSUPP;
2299 
2300 	switch (cmd) {
2301 	case SIOCGMIIPHY:
2302 		data->phy_id = dev->phy.address;
2303 		/* Fall through */
2304 	case SIOCGMIIREG:
2305 		data->val_out = emac_mdio_read(ndev, dev->phy.address,
2306 					       data->reg_num);
2307 		return 0;
2308 
2309 	case SIOCSMIIREG:
2310 		emac_mdio_write(ndev, dev->phy.address, data->reg_num,
2311 				data->val_in);
2312 		return 0;
2313 	default:
2314 		return -EOPNOTSUPP;
2315 	}
2316 }
2317 
2318 struct emac_depentry {
2319 	u32			phandle;
2320 	struct device_node	*node;
2321 	struct platform_device	*ofdev;
2322 	void			*drvdata;
2323 };
2324 
2325 #define	EMAC_DEP_MAL_IDX	0
2326 #define	EMAC_DEP_ZMII_IDX	1
2327 #define	EMAC_DEP_RGMII_IDX	2
2328 #define	EMAC_DEP_TAH_IDX	3
2329 #define	EMAC_DEP_MDIO_IDX	4
2330 #define	EMAC_DEP_PREV_IDX	5
2331 #define	EMAC_DEP_COUNT		6
2332 
2333 static int emac_check_deps(struct emac_instance *dev,
2334 			   struct emac_depentry *deps)
2335 {
2336 	int i, there = 0;
2337 	struct device_node *np;
2338 
2339 	for (i = 0; i < EMAC_DEP_COUNT; i++) {
2340 		/* no dependency on that item, allright */
2341 		if (deps[i].phandle == 0) {
2342 			there++;
2343 			continue;
2344 		}
2345 		/* special case for blist as the dependency might go away */
2346 		if (i == EMAC_DEP_PREV_IDX) {
2347 			np = *(dev->blist - 1);
2348 			if (np == NULL) {
2349 				deps[i].phandle = 0;
2350 				there++;
2351 				continue;
2352 			}
2353 			if (deps[i].node == NULL)
2354 				deps[i].node = of_node_get(np);
2355 		}
2356 		if (deps[i].node == NULL)
2357 			deps[i].node = of_find_node_by_phandle(deps[i].phandle);
2358 		if (deps[i].node == NULL)
2359 			continue;
2360 		if (deps[i].ofdev == NULL)
2361 			deps[i].ofdev = of_find_device_by_node(deps[i].node);
2362 		if (deps[i].ofdev == NULL)
2363 			continue;
2364 		if (deps[i].drvdata == NULL)
2365 			deps[i].drvdata = platform_get_drvdata(deps[i].ofdev);
2366 		if (deps[i].drvdata != NULL)
2367 			there++;
2368 	}
2369 	return there == EMAC_DEP_COUNT;
2370 }
2371 
2372 static void emac_put_deps(struct emac_instance *dev)
2373 {
2374 	of_dev_put(dev->mal_dev);
2375 	of_dev_put(dev->zmii_dev);
2376 	of_dev_put(dev->rgmii_dev);
2377 	of_dev_put(dev->mdio_dev);
2378 	of_dev_put(dev->tah_dev);
2379 }
2380 
2381 static int emac_of_bus_notify(struct notifier_block *nb, unsigned long action,
2382 			      void *data)
2383 {
2384 	/* We are only intereted in device addition */
2385 	if (action == BUS_NOTIFY_BOUND_DRIVER)
2386 		wake_up_all(&emac_probe_wait);
2387 	return 0;
2388 }
2389 
2390 static struct notifier_block emac_of_bus_notifier = {
2391 	.notifier_call = emac_of_bus_notify
2392 };
2393 
2394 static int emac_wait_deps(struct emac_instance *dev)
2395 {
2396 	struct emac_depentry deps[EMAC_DEP_COUNT];
2397 	int i, err;
2398 
2399 	memset(&deps, 0, sizeof(deps));
2400 
2401 	deps[EMAC_DEP_MAL_IDX].phandle = dev->mal_ph;
2402 	deps[EMAC_DEP_ZMII_IDX].phandle = dev->zmii_ph;
2403 	deps[EMAC_DEP_RGMII_IDX].phandle = dev->rgmii_ph;
2404 	if (dev->tah_ph)
2405 		deps[EMAC_DEP_TAH_IDX].phandle = dev->tah_ph;
2406 	if (dev->mdio_ph)
2407 		deps[EMAC_DEP_MDIO_IDX].phandle = dev->mdio_ph;
2408 	if (dev->blist && dev->blist > emac_boot_list)
2409 		deps[EMAC_DEP_PREV_IDX].phandle = 0xffffffffu;
2410 	bus_register_notifier(&platform_bus_type, &emac_of_bus_notifier);
2411 	wait_event_timeout(emac_probe_wait,
2412 			   emac_check_deps(dev, deps),
2413 			   EMAC_PROBE_DEP_TIMEOUT);
2414 	bus_unregister_notifier(&platform_bus_type, &emac_of_bus_notifier);
2415 	err = emac_check_deps(dev, deps) ? 0 : -ENODEV;
2416 	for (i = 0; i < EMAC_DEP_COUNT; i++) {
2417 		of_node_put(deps[i].node);
2418 		if (err)
2419 			of_dev_put(deps[i].ofdev);
2420 	}
2421 	if (err == 0) {
2422 		dev->mal_dev = deps[EMAC_DEP_MAL_IDX].ofdev;
2423 		dev->zmii_dev = deps[EMAC_DEP_ZMII_IDX].ofdev;
2424 		dev->rgmii_dev = deps[EMAC_DEP_RGMII_IDX].ofdev;
2425 		dev->tah_dev = deps[EMAC_DEP_TAH_IDX].ofdev;
2426 		dev->mdio_dev = deps[EMAC_DEP_MDIO_IDX].ofdev;
2427 	}
2428 	of_dev_put(deps[EMAC_DEP_PREV_IDX].ofdev);
2429 	return err;
2430 }
2431 
2432 static int emac_read_uint_prop(struct device_node *np, const char *name,
2433 			       u32 *val, int fatal)
2434 {
2435 	int len;
2436 	const u32 *prop = of_get_property(np, name, &len);
2437 	if (prop == NULL || len < sizeof(u32)) {
2438 		if (fatal)
2439 			printk(KERN_ERR "%pOF: missing %s property\n",
2440 			       np, name);
2441 		return -ENODEV;
2442 	}
2443 	*val = *prop;
2444 	return 0;
2445 }
2446 
2447 static void emac_adjust_link(struct net_device *ndev)
2448 {
2449 	struct emac_instance *dev = netdev_priv(ndev);
2450 	struct phy_device *phy = dev->phy_dev;
2451 
2452 	dev->phy.autoneg = phy->autoneg;
2453 	dev->phy.speed = phy->speed;
2454 	dev->phy.duplex = phy->duplex;
2455 	dev->phy.pause = phy->pause;
2456 	dev->phy.asym_pause = phy->asym_pause;
2457 	dev->phy.advertising = phy->advertising;
2458 }
2459 
2460 static int emac_mii_bus_read(struct mii_bus *bus, int addr, int regnum)
2461 {
2462 	int ret = emac_mdio_read(bus->priv, addr, regnum);
2463 	/* This is a workaround for powered down ports/phys.
2464 	 * In the wild, this was seen on the Cisco Meraki MX60(W).
2465 	 * This hardware disables ports as part of the handoff
2466 	 * procedure. Accessing the ports will lead to errors
2467 	 * (-ETIMEDOUT, -EREMOTEIO) that do more harm than good.
2468 	 */
2469 	return ret < 0 ? 0xffff : ret;
2470 }
2471 
2472 static int emac_mii_bus_write(struct mii_bus *bus, int addr,
2473 			      int regnum, u16 val)
2474 {
2475 	emac_mdio_write(bus->priv, addr, regnum, val);
2476 	return 0;
2477 }
2478 
2479 static int emac_mii_bus_reset(struct mii_bus *bus)
2480 {
2481 	struct emac_instance *dev = netdev_priv(bus->priv);
2482 
2483 	return emac_reset(dev);
2484 }
2485 
2486 static int emac_mdio_phy_start_aneg(struct mii_phy *phy,
2487 				    struct phy_device *phy_dev)
2488 {
2489 	phy_dev->autoneg = phy->autoneg;
2490 	phy_dev->speed = phy->speed;
2491 	phy_dev->duplex = phy->duplex;
2492 	phy_dev->advertising = phy->advertising;
2493 	return phy_start_aneg(phy_dev);
2494 }
2495 
2496 static int emac_mdio_setup_aneg(struct mii_phy *phy, u32 advertise)
2497 {
2498 	struct net_device *ndev = phy->dev;
2499 	struct emac_instance *dev = netdev_priv(ndev);
2500 
2501 	phy->autoneg = AUTONEG_ENABLE;
2502 	phy->advertising = advertise;
2503 	return emac_mdio_phy_start_aneg(phy, dev->phy_dev);
2504 }
2505 
2506 static int emac_mdio_setup_forced(struct mii_phy *phy, int speed, int fd)
2507 {
2508 	struct net_device *ndev = phy->dev;
2509 	struct emac_instance *dev = netdev_priv(ndev);
2510 
2511 	phy->autoneg = AUTONEG_DISABLE;
2512 	phy->speed = speed;
2513 	phy->duplex = fd;
2514 	return emac_mdio_phy_start_aneg(phy, dev->phy_dev);
2515 }
2516 
2517 static int emac_mdio_poll_link(struct mii_phy *phy)
2518 {
2519 	struct net_device *ndev = phy->dev;
2520 	struct emac_instance *dev = netdev_priv(ndev);
2521 	int res;
2522 
2523 	res = phy_read_status(dev->phy_dev);
2524 	if (res) {
2525 		dev_err(&dev->ofdev->dev, "link update failed (%d).", res);
2526 		return ethtool_op_get_link(ndev);
2527 	}
2528 
2529 	return dev->phy_dev->link;
2530 }
2531 
2532 static int emac_mdio_read_link(struct mii_phy *phy)
2533 {
2534 	struct net_device *ndev = phy->dev;
2535 	struct emac_instance *dev = netdev_priv(ndev);
2536 	struct phy_device *phy_dev = dev->phy_dev;
2537 	int res;
2538 
2539 	res = phy_read_status(phy_dev);
2540 	if (res)
2541 		return res;
2542 
2543 	phy->speed = phy_dev->speed;
2544 	phy->duplex = phy_dev->duplex;
2545 	phy->pause = phy_dev->pause;
2546 	phy->asym_pause = phy_dev->asym_pause;
2547 	return 0;
2548 }
2549 
2550 static int emac_mdio_init_phy(struct mii_phy *phy)
2551 {
2552 	struct net_device *ndev = phy->dev;
2553 	struct emac_instance *dev = netdev_priv(ndev);
2554 
2555 	phy_start(dev->phy_dev);
2556 	return phy_init_hw(dev->phy_dev);
2557 }
2558 
2559 static const struct mii_phy_ops emac_dt_mdio_phy_ops = {
2560 	.init		= emac_mdio_init_phy,
2561 	.setup_aneg	= emac_mdio_setup_aneg,
2562 	.setup_forced	= emac_mdio_setup_forced,
2563 	.poll_link	= emac_mdio_poll_link,
2564 	.read_link	= emac_mdio_read_link,
2565 };
2566 
2567 static int emac_dt_mdio_probe(struct emac_instance *dev)
2568 {
2569 	struct device_node *mii_np;
2570 	int res;
2571 
2572 	mii_np = of_get_child_by_name(dev->ofdev->dev.of_node, "mdio");
2573 	if (!mii_np) {
2574 		dev_err(&dev->ofdev->dev, "no mdio definition found.");
2575 		return -ENODEV;
2576 	}
2577 
2578 	if (!of_device_is_available(mii_np)) {
2579 		res = -ENODEV;
2580 		goto put_node;
2581 	}
2582 
2583 	dev->mii_bus = devm_mdiobus_alloc(&dev->ofdev->dev);
2584 	if (!dev->mii_bus) {
2585 		res = -ENOMEM;
2586 		goto put_node;
2587 	}
2588 
2589 	dev->mii_bus->priv = dev->ndev;
2590 	dev->mii_bus->parent = dev->ndev->dev.parent;
2591 	dev->mii_bus->name = "emac_mdio";
2592 	dev->mii_bus->read = &emac_mii_bus_read;
2593 	dev->mii_bus->write = &emac_mii_bus_write;
2594 	dev->mii_bus->reset = &emac_mii_bus_reset;
2595 	snprintf(dev->mii_bus->id, MII_BUS_ID_SIZE, "%s", dev->ofdev->name);
2596 	res = of_mdiobus_register(dev->mii_bus, mii_np);
2597 	if (res) {
2598 		dev_err(&dev->ofdev->dev, "cannot register MDIO bus %s (%d)",
2599 			dev->mii_bus->name, res);
2600 	}
2601 
2602  put_node:
2603 	of_node_put(mii_np);
2604 	return res;
2605 }
2606 
2607 static int emac_dt_phy_connect(struct emac_instance *dev,
2608 			       struct device_node *phy_handle)
2609 {
2610 	dev->phy.def = devm_kzalloc(&dev->ofdev->dev, sizeof(*dev->phy.def),
2611 				    GFP_KERNEL);
2612 	if (!dev->phy.def)
2613 		return -ENOMEM;
2614 
2615 	dev->phy_dev = of_phy_connect(dev->ndev, phy_handle, &emac_adjust_link,
2616 				      0, dev->phy_mode);
2617 	if (!dev->phy_dev) {
2618 		dev_err(&dev->ofdev->dev, "failed to connect to PHY.\n");
2619 		return -ENODEV;
2620 	}
2621 
2622 	dev->phy.def->phy_id = dev->phy_dev->drv->phy_id;
2623 	dev->phy.def->phy_id_mask = dev->phy_dev->drv->phy_id_mask;
2624 	dev->phy.def->name = dev->phy_dev->drv->name;
2625 	dev->phy.def->ops = &emac_dt_mdio_phy_ops;
2626 	dev->phy.features = dev->phy_dev->supported;
2627 	dev->phy.address = dev->phy_dev->mdio.addr;
2628 	dev->phy.mode = dev->phy_dev->interface;
2629 	return 0;
2630 }
2631 
2632 static int emac_dt_phy_probe(struct emac_instance *dev)
2633 {
2634 	struct device_node *np = dev->ofdev->dev.of_node;
2635 	struct device_node *phy_handle;
2636 	int res = 1;
2637 
2638 	phy_handle = of_parse_phandle(np, "phy-handle", 0);
2639 
2640 	if (phy_handle) {
2641 		res = emac_dt_mdio_probe(dev);
2642 		if (!res) {
2643 			res = emac_dt_phy_connect(dev, phy_handle);
2644 			if (res)
2645 				mdiobus_unregister(dev->mii_bus);
2646 		}
2647 	}
2648 
2649 	of_node_put(phy_handle);
2650 	return res;
2651 }
2652 
2653 static int emac_init_phy(struct emac_instance *dev)
2654 {
2655 	struct device_node *np = dev->ofdev->dev.of_node;
2656 	struct net_device *ndev = dev->ndev;
2657 	u32 phy_map, adv;
2658 	int i;
2659 
2660 	dev->phy.dev = ndev;
2661 	dev->phy.mode = dev->phy_mode;
2662 
2663 	/* PHY-less configuration. */
2664 	if ((dev->phy_address == 0xffffffff && dev->phy_map == 0xffffffff) ||
2665 	    of_phy_is_fixed_link(np)) {
2666 		emac_reset(dev);
2667 
2668 		/* PHY-less configuration. */
2669 		dev->phy.address = -1;
2670 		dev->phy.features = SUPPORTED_MII;
2671 		if (emac_phy_supports_gige(dev->phy_mode))
2672 			dev->phy.features |= SUPPORTED_1000baseT_Full;
2673 		else
2674 			dev->phy.features |= SUPPORTED_100baseT_Full;
2675 		dev->phy.pause = 1;
2676 
2677 		if (of_phy_is_fixed_link(np)) {
2678 			int res = emac_dt_mdio_probe(dev);
2679 
2680 			if (!res) {
2681 				res = of_phy_register_fixed_link(np);
2682 				if (res)
2683 					mdiobus_unregister(dev->mii_bus);
2684 			}
2685 			return res;
2686 		}
2687 		return 0;
2688 	}
2689 
2690 	mutex_lock(&emac_phy_map_lock);
2691 	phy_map = dev->phy_map | busy_phy_map;
2692 
2693 	DBG(dev, "PHY maps %08x %08x" NL, dev->phy_map, busy_phy_map);
2694 
2695 	dev->phy.mdio_read = emac_mdio_read;
2696 	dev->phy.mdio_write = emac_mdio_write;
2697 
2698 	/* Enable internal clock source */
2699 #ifdef CONFIG_PPC_DCR_NATIVE
2700 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2701 		dcri_clrset(SDR0, SDR0_MFR, 0, SDR0_MFR_ECS);
2702 #endif
2703 	/* PHY clock workaround */
2704 	emac_rx_clk_tx(dev);
2705 
2706 	/* Enable internal clock source on 440GX*/
2707 #ifdef CONFIG_PPC_DCR_NATIVE
2708 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2709 		dcri_clrset(SDR0, SDR0_MFR, 0, SDR0_MFR_ECS);
2710 #endif
2711 	/* Configure EMAC with defaults so we can at least use MDIO
2712 	 * This is needed mostly for 440GX
2713 	 */
2714 	if (emac_phy_gpcs(dev->phy.mode)) {
2715 		/* XXX
2716 		 * Make GPCS PHY address equal to EMAC index.
2717 		 * We probably should take into account busy_phy_map
2718 		 * and/or phy_map here.
2719 		 *
2720 		 * Note that the busy_phy_map is currently global
2721 		 * while it should probably be per-ASIC...
2722 		 */
2723 		dev->phy.gpcs_address = dev->gpcs_address;
2724 		if (dev->phy.gpcs_address == 0xffffffff)
2725 			dev->phy.address = dev->cell_index;
2726 	}
2727 
2728 	emac_configure(dev);
2729 
2730 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII)) {
2731 		int res = emac_dt_phy_probe(dev);
2732 
2733 		switch (res) {
2734 		case 1:
2735 			/* No phy-handle property configured.
2736 			 * Continue with the existing phy probe
2737 			 * and setup code.
2738 			 */
2739 			break;
2740 
2741 		case 0:
2742 			mutex_unlock(&emac_phy_map_lock);
2743 			goto init_phy;
2744 
2745 		default:
2746 			mutex_unlock(&emac_phy_map_lock);
2747 			dev_err(&dev->ofdev->dev, "failed to attach dt phy (%d).\n",
2748 				res);
2749 			return res;
2750 		}
2751 	}
2752 
2753 	if (dev->phy_address != 0xffffffff)
2754 		phy_map = ~(1 << dev->phy_address);
2755 
2756 	for (i = 0; i < 0x20; phy_map >>= 1, ++i)
2757 		if (!(phy_map & 1)) {
2758 			int r;
2759 			busy_phy_map |= 1 << i;
2760 
2761 			/* Quick check if there is a PHY at the address */
2762 			r = emac_mdio_read(dev->ndev, i, MII_BMCR);
2763 			if (r == 0xffff || r < 0)
2764 				continue;
2765 			if (!emac_mii_phy_probe(&dev->phy, i))
2766 				break;
2767 		}
2768 
2769 	/* Enable external clock source */
2770 #ifdef CONFIG_PPC_DCR_NATIVE
2771 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2772 		dcri_clrset(SDR0, SDR0_MFR, SDR0_MFR_ECS, 0);
2773 #endif
2774 	mutex_unlock(&emac_phy_map_lock);
2775 	if (i == 0x20) {
2776 		printk(KERN_WARNING "%pOF: can't find PHY!\n", np);
2777 		return -ENXIO;
2778 	}
2779 
2780  init_phy:
2781 	/* Init PHY */
2782 	if (dev->phy.def->ops->init)
2783 		dev->phy.def->ops->init(&dev->phy);
2784 
2785 	/* Disable any PHY features not supported by the platform */
2786 	dev->phy.def->features &= ~dev->phy_feat_exc;
2787 	dev->phy.features &= ~dev->phy_feat_exc;
2788 
2789 	/* Setup initial link parameters */
2790 	if (dev->phy.features & SUPPORTED_Autoneg) {
2791 		adv = dev->phy.features;
2792 		if (!emac_has_feature(dev, EMAC_FTR_NO_FLOW_CONTROL_40x))
2793 			adv |= ADVERTISED_Pause | ADVERTISED_Asym_Pause;
2794 		/* Restart autonegotiation */
2795 		dev->phy.def->ops->setup_aneg(&dev->phy, adv);
2796 	} else {
2797 		u32 f = dev->phy.def->features;
2798 		int speed = SPEED_10, fd = DUPLEX_HALF;
2799 
2800 		/* Select highest supported speed/duplex */
2801 		if (f & SUPPORTED_1000baseT_Full) {
2802 			speed = SPEED_1000;
2803 			fd = DUPLEX_FULL;
2804 		} else if (f & SUPPORTED_1000baseT_Half)
2805 			speed = SPEED_1000;
2806 		else if (f & SUPPORTED_100baseT_Full) {
2807 			speed = SPEED_100;
2808 			fd = DUPLEX_FULL;
2809 		} else if (f & SUPPORTED_100baseT_Half)
2810 			speed = SPEED_100;
2811 		else if (f & SUPPORTED_10baseT_Full)
2812 			fd = DUPLEX_FULL;
2813 
2814 		/* Force link parameters */
2815 		dev->phy.def->ops->setup_forced(&dev->phy, speed, fd);
2816 	}
2817 	return 0;
2818 }
2819 
2820 static int emac_init_config(struct emac_instance *dev)
2821 {
2822 	struct device_node *np = dev->ofdev->dev.of_node;
2823 	const void *p;
2824 
2825 	/* Read config from device-tree */
2826 	if (emac_read_uint_prop(np, "mal-device", &dev->mal_ph, 1))
2827 		return -ENXIO;
2828 	if (emac_read_uint_prop(np, "mal-tx-channel", &dev->mal_tx_chan, 1))
2829 		return -ENXIO;
2830 	if (emac_read_uint_prop(np, "mal-rx-channel", &dev->mal_rx_chan, 1))
2831 		return -ENXIO;
2832 	if (emac_read_uint_prop(np, "cell-index", &dev->cell_index, 1))
2833 		return -ENXIO;
2834 	if (emac_read_uint_prop(np, "max-frame-size", &dev->max_mtu, 0))
2835 		dev->max_mtu = ETH_DATA_LEN;
2836 	if (emac_read_uint_prop(np, "rx-fifo-size", &dev->rx_fifo_size, 0))
2837 		dev->rx_fifo_size = 2048;
2838 	if (emac_read_uint_prop(np, "tx-fifo-size", &dev->tx_fifo_size, 0))
2839 		dev->tx_fifo_size = 2048;
2840 	if (emac_read_uint_prop(np, "rx-fifo-size-gige", &dev->rx_fifo_size_gige, 0))
2841 		dev->rx_fifo_size_gige = dev->rx_fifo_size;
2842 	if (emac_read_uint_prop(np, "tx-fifo-size-gige", &dev->tx_fifo_size_gige, 0))
2843 		dev->tx_fifo_size_gige = dev->tx_fifo_size;
2844 	if (emac_read_uint_prop(np, "phy-address", &dev->phy_address, 0))
2845 		dev->phy_address = 0xffffffff;
2846 	if (emac_read_uint_prop(np, "phy-map", &dev->phy_map, 0))
2847 		dev->phy_map = 0xffffffff;
2848 	if (emac_read_uint_prop(np, "gpcs-address", &dev->gpcs_address, 0))
2849 		dev->gpcs_address = 0xffffffff;
2850 	if (emac_read_uint_prop(np->parent, "clock-frequency", &dev->opb_bus_freq, 1))
2851 		return -ENXIO;
2852 	if (emac_read_uint_prop(np, "tah-device", &dev->tah_ph, 0))
2853 		dev->tah_ph = 0;
2854 	if (emac_read_uint_prop(np, "tah-channel", &dev->tah_port, 0))
2855 		dev->tah_port = 0;
2856 	if (emac_read_uint_prop(np, "mdio-device", &dev->mdio_ph, 0))
2857 		dev->mdio_ph = 0;
2858 	if (emac_read_uint_prop(np, "zmii-device", &dev->zmii_ph, 0))
2859 		dev->zmii_ph = 0;
2860 	if (emac_read_uint_prop(np, "zmii-channel", &dev->zmii_port, 0))
2861 		dev->zmii_port = 0xffffffff;
2862 	if (emac_read_uint_prop(np, "rgmii-device", &dev->rgmii_ph, 0))
2863 		dev->rgmii_ph = 0;
2864 	if (emac_read_uint_prop(np, "rgmii-channel", &dev->rgmii_port, 0))
2865 		dev->rgmii_port = 0xffffffff;
2866 	if (emac_read_uint_prop(np, "fifo-entry-size", &dev->fifo_entry_size, 0))
2867 		dev->fifo_entry_size = 16;
2868 	if (emac_read_uint_prop(np, "mal-burst-size", &dev->mal_burst_size, 0))
2869 		dev->mal_burst_size = 256;
2870 
2871 	/* PHY mode needs some decoding */
2872 	dev->phy_mode = of_get_phy_mode(np);
2873 	if (dev->phy_mode < 0)
2874 		dev->phy_mode = PHY_INTERFACE_MODE_NA;
2875 
2876 	/* Check EMAC version */
2877 	if (of_device_is_compatible(np, "ibm,emac4sync")) {
2878 		dev->features |= (EMAC_FTR_EMAC4 | EMAC_FTR_EMAC4SYNC);
2879 		if (of_device_is_compatible(np, "ibm,emac-460ex") ||
2880 		    of_device_is_compatible(np, "ibm,emac-460gt"))
2881 			dev->features |= EMAC_FTR_460EX_PHY_CLK_FIX;
2882 		if (of_device_is_compatible(np, "ibm,emac-405ex") ||
2883 		    of_device_is_compatible(np, "ibm,emac-405exr"))
2884 			dev->features |= EMAC_FTR_440EP_PHY_CLK_FIX;
2885 		if (of_device_is_compatible(np, "ibm,emac-apm821xx")) {
2886 			dev->features |= (EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE |
2887 					  EMAC_FTR_APM821XX_NO_HALF_DUPLEX |
2888 					  EMAC_FTR_460EX_PHY_CLK_FIX);
2889 		}
2890 	} else if (of_device_is_compatible(np, "ibm,emac4")) {
2891 		dev->features |= EMAC_FTR_EMAC4;
2892 		if (of_device_is_compatible(np, "ibm,emac-440gx"))
2893 			dev->features |= EMAC_FTR_440GX_PHY_CLK_FIX;
2894 	} else {
2895 		if (of_device_is_compatible(np, "ibm,emac-440ep") ||
2896 		    of_device_is_compatible(np, "ibm,emac-440gr"))
2897 			dev->features |= EMAC_FTR_440EP_PHY_CLK_FIX;
2898 		if (of_device_is_compatible(np, "ibm,emac-405ez")) {
2899 #ifdef CONFIG_IBM_EMAC_NO_FLOW_CTRL
2900 			dev->features |= EMAC_FTR_NO_FLOW_CONTROL_40x;
2901 #else
2902 			printk(KERN_ERR "%pOF: Flow control not disabled!\n",
2903 					np);
2904 			return -ENXIO;
2905 #endif
2906 		}
2907 
2908 	}
2909 
2910 	/* Fixup some feature bits based on the device tree */
2911 	if (of_get_property(np, "has-inverted-stacr-oc", NULL))
2912 		dev->features |= EMAC_FTR_STACR_OC_INVERT;
2913 	if (of_get_property(np, "has-new-stacr-staopc", NULL))
2914 		dev->features |= EMAC_FTR_HAS_NEW_STACR;
2915 
2916 	/* CAB lacks the appropriate properties */
2917 	if (of_device_is_compatible(np, "ibm,emac-axon"))
2918 		dev->features |= EMAC_FTR_HAS_NEW_STACR |
2919 			EMAC_FTR_STACR_OC_INVERT;
2920 
2921 	/* Enable TAH/ZMII/RGMII features as found */
2922 	if (dev->tah_ph != 0) {
2923 #ifdef CONFIG_IBM_EMAC_TAH
2924 		dev->features |= EMAC_FTR_HAS_TAH;
2925 #else
2926 		printk(KERN_ERR "%pOF: TAH support not enabled !\n", np);
2927 		return -ENXIO;
2928 #endif
2929 	}
2930 
2931 	if (dev->zmii_ph != 0) {
2932 #ifdef CONFIG_IBM_EMAC_ZMII
2933 		dev->features |= EMAC_FTR_HAS_ZMII;
2934 #else
2935 		printk(KERN_ERR "%pOF: ZMII support not enabled !\n", np);
2936 		return -ENXIO;
2937 #endif
2938 	}
2939 
2940 	if (dev->rgmii_ph != 0) {
2941 #ifdef CONFIG_IBM_EMAC_RGMII
2942 		dev->features |= EMAC_FTR_HAS_RGMII;
2943 #else
2944 		printk(KERN_ERR "%pOF: RGMII support not enabled !\n", np);
2945 		return -ENXIO;
2946 #endif
2947 	}
2948 
2949 	/* Read MAC-address */
2950 	p = of_get_property(np, "local-mac-address", NULL);
2951 	if (p == NULL) {
2952 		printk(KERN_ERR "%pOF: Can't find local-mac-address property\n",
2953 		       np);
2954 		return -ENXIO;
2955 	}
2956 	memcpy(dev->ndev->dev_addr, p, ETH_ALEN);
2957 
2958 	/* IAHT and GAHT filter parameterization */
2959 	if (emac_has_feature(dev, EMAC_FTR_EMAC4SYNC)) {
2960 		dev->xaht_slots_shift = EMAC4SYNC_XAHT_SLOTS_SHIFT;
2961 		dev->xaht_width_shift = EMAC4SYNC_XAHT_WIDTH_SHIFT;
2962 	} else {
2963 		dev->xaht_slots_shift = EMAC4_XAHT_SLOTS_SHIFT;
2964 		dev->xaht_width_shift = EMAC4_XAHT_WIDTH_SHIFT;
2965 	}
2966 
2967 	DBG(dev, "features     : 0x%08x / 0x%08x\n", dev->features, EMAC_FTRS_POSSIBLE);
2968 	DBG(dev, "tx_fifo_size : %d (%d gige)\n", dev->tx_fifo_size, dev->tx_fifo_size_gige);
2969 	DBG(dev, "rx_fifo_size : %d (%d gige)\n", dev->rx_fifo_size, dev->rx_fifo_size_gige);
2970 	DBG(dev, "max_mtu      : %d\n", dev->max_mtu);
2971 	DBG(dev, "OPB freq     : %d\n", dev->opb_bus_freq);
2972 
2973 	return 0;
2974 }
2975 
2976 static const struct net_device_ops emac_netdev_ops = {
2977 	.ndo_open		= emac_open,
2978 	.ndo_stop		= emac_close,
2979 	.ndo_get_stats		= emac_stats,
2980 	.ndo_set_rx_mode	= emac_set_multicast_list,
2981 	.ndo_do_ioctl		= emac_ioctl,
2982 	.ndo_tx_timeout		= emac_tx_timeout,
2983 	.ndo_validate_addr	= eth_validate_addr,
2984 	.ndo_set_mac_address	= emac_set_mac_address,
2985 	.ndo_start_xmit		= emac_start_xmit,
2986 };
2987 
2988 static const struct net_device_ops emac_gige_netdev_ops = {
2989 	.ndo_open		= emac_open,
2990 	.ndo_stop		= emac_close,
2991 	.ndo_get_stats		= emac_stats,
2992 	.ndo_set_rx_mode	= emac_set_multicast_list,
2993 	.ndo_do_ioctl		= emac_ioctl,
2994 	.ndo_tx_timeout		= emac_tx_timeout,
2995 	.ndo_validate_addr	= eth_validate_addr,
2996 	.ndo_set_mac_address	= emac_set_mac_address,
2997 	.ndo_start_xmit		= emac_start_xmit_sg,
2998 	.ndo_change_mtu		= emac_change_mtu,
2999 };
3000 
3001 static int emac_probe(struct platform_device *ofdev)
3002 {
3003 	struct net_device *ndev;
3004 	struct emac_instance *dev;
3005 	struct device_node *np = ofdev->dev.of_node;
3006 	struct device_node **blist = NULL;
3007 	int err, i;
3008 
3009 	/* Skip unused/unwired EMACS.  We leave the check for an unused
3010 	 * property here for now, but new flat device trees should set a
3011 	 * status property to "disabled" instead.
3012 	 */
3013 	if (of_get_property(np, "unused", NULL) || !of_device_is_available(np))
3014 		return -ENODEV;
3015 
3016 	/* Find ourselves in the bootlist if we are there */
3017 	for (i = 0; i < EMAC_BOOT_LIST_SIZE; i++)
3018 		if (emac_boot_list[i] == np)
3019 			blist = &emac_boot_list[i];
3020 
3021 	/* Allocate our net_device structure */
3022 	err = -ENOMEM;
3023 	ndev = alloc_etherdev(sizeof(struct emac_instance));
3024 	if (!ndev)
3025 		goto err_gone;
3026 
3027 	dev = netdev_priv(ndev);
3028 	dev->ndev = ndev;
3029 	dev->ofdev = ofdev;
3030 	dev->blist = blist;
3031 	SET_NETDEV_DEV(ndev, &ofdev->dev);
3032 
3033 	/* Initialize some embedded data structures */
3034 	mutex_init(&dev->mdio_lock);
3035 	mutex_init(&dev->link_lock);
3036 	spin_lock_init(&dev->lock);
3037 	INIT_WORK(&dev->reset_work, emac_reset_work);
3038 
3039 	/* Init various config data based on device-tree */
3040 	err = emac_init_config(dev);
3041 	if (err)
3042 		goto err_free;
3043 
3044 	/* Get interrupts. EMAC irq is mandatory, WOL irq is optional */
3045 	dev->emac_irq = irq_of_parse_and_map(np, 0);
3046 	dev->wol_irq = irq_of_parse_and_map(np, 1);
3047 	if (!dev->emac_irq) {
3048 		printk(KERN_ERR "%pOF: Can't map main interrupt\n", np);
3049 		err = -ENODEV;
3050 		goto err_free;
3051 	}
3052 	ndev->irq = dev->emac_irq;
3053 
3054 	/* Map EMAC regs */
3055 	// TODO : platform_get_resource() and devm_ioremap_resource()
3056 	dev->emacp = of_iomap(np, 0);
3057 	if (dev->emacp == NULL) {
3058 		printk(KERN_ERR "%pOF: Can't map device registers!\n", np);
3059 		err = -ENOMEM;
3060 		goto err_irq_unmap;
3061 	}
3062 
3063 	/* Wait for dependent devices */
3064 	err = emac_wait_deps(dev);
3065 	if (err) {
3066 		printk(KERN_ERR
3067 		       "%pOF: Timeout waiting for dependent devices\n", np);
3068 		/*  display more info about what's missing ? */
3069 		goto err_reg_unmap;
3070 	}
3071 	dev->mal = platform_get_drvdata(dev->mal_dev);
3072 	if (dev->mdio_dev != NULL)
3073 		dev->mdio_instance = platform_get_drvdata(dev->mdio_dev);
3074 
3075 	/* Register with MAL */
3076 	dev->commac.ops = &emac_commac_ops;
3077 	dev->commac.dev = dev;
3078 	dev->commac.tx_chan_mask = MAL_CHAN_MASK(dev->mal_tx_chan);
3079 	dev->commac.rx_chan_mask = MAL_CHAN_MASK(dev->mal_rx_chan);
3080 	err = mal_register_commac(dev->mal, &dev->commac);
3081 	if (err) {
3082 		printk(KERN_ERR "%pOF: failed to register with mal %pOF!\n",
3083 		       np, dev->mal_dev->dev.of_node);
3084 		goto err_rel_deps;
3085 	}
3086 	dev->rx_skb_size = emac_rx_skb_size(ndev->mtu);
3087 	dev->rx_sync_size = emac_rx_sync_size(ndev->mtu);
3088 
3089 	/* Get pointers to BD rings */
3090 	dev->tx_desc =
3091 	    dev->mal->bd_virt + mal_tx_bd_offset(dev->mal, dev->mal_tx_chan);
3092 	dev->rx_desc =
3093 	    dev->mal->bd_virt + mal_rx_bd_offset(dev->mal, dev->mal_rx_chan);
3094 
3095 	DBG(dev, "tx_desc %p" NL, dev->tx_desc);
3096 	DBG(dev, "rx_desc %p" NL, dev->rx_desc);
3097 
3098 	/* Clean rings */
3099 	memset(dev->tx_desc, 0, NUM_TX_BUFF * sizeof(struct mal_descriptor));
3100 	memset(dev->rx_desc, 0, NUM_RX_BUFF * sizeof(struct mal_descriptor));
3101 	memset(dev->tx_skb, 0, NUM_TX_BUFF * sizeof(struct sk_buff *));
3102 	memset(dev->rx_skb, 0, NUM_RX_BUFF * sizeof(struct sk_buff *));
3103 
3104 	/* Attach to ZMII, if needed */
3105 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII) &&
3106 	    (err = zmii_attach(dev->zmii_dev, dev->zmii_port, &dev->phy_mode)) != 0)
3107 		goto err_unreg_commac;
3108 
3109 	/* Attach to RGMII, if needed */
3110 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII) &&
3111 	    (err = rgmii_attach(dev->rgmii_dev, dev->rgmii_port, dev->phy_mode)) != 0)
3112 		goto err_detach_zmii;
3113 
3114 	/* Attach to TAH, if needed */
3115 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH) &&
3116 	    (err = tah_attach(dev->tah_dev, dev->tah_port)) != 0)
3117 		goto err_detach_rgmii;
3118 
3119 	/* Set some link defaults before we can find out real parameters */
3120 	dev->phy.speed = SPEED_100;
3121 	dev->phy.duplex = DUPLEX_FULL;
3122 	dev->phy.autoneg = AUTONEG_DISABLE;
3123 	dev->phy.pause = dev->phy.asym_pause = 0;
3124 	dev->stop_timeout = STOP_TIMEOUT_100;
3125 	INIT_DELAYED_WORK(&dev->link_work, emac_link_timer);
3126 
3127 	/* Some SoCs like APM821xx does not support Half Duplex mode. */
3128 	if (emac_has_feature(dev, EMAC_FTR_APM821XX_NO_HALF_DUPLEX)) {
3129 		dev->phy_feat_exc = (SUPPORTED_1000baseT_Half |
3130 				     SUPPORTED_100baseT_Half |
3131 				     SUPPORTED_10baseT_Half);
3132 	}
3133 
3134 	/* Find PHY if any */
3135 	err = emac_init_phy(dev);
3136 	if (err != 0)
3137 		goto err_detach_tah;
3138 
3139 	if (dev->tah_dev) {
3140 		ndev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG;
3141 		ndev->features |= ndev->hw_features | NETIF_F_RXCSUM;
3142 	}
3143 	ndev->watchdog_timeo = 5 * HZ;
3144 	if (emac_phy_supports_gige(dev->phy_mode)) {
3145 		ndev->netdev_ops = &emac_gige_netdev_ops;
3146 		dev->commac.ops = &emac_commac_sg_ops;
3147 	} else
3148 		ndev->netdev_ops = &emac_netdev_ops;
3149 	ndev->ethtool_ops = &emac_ethtool_ops;
3150 
3151 	/* MTU range: 46 - 1500 or whatever is in OF */
3152 	ndev->min_mtu = EMAC_MIN_MTU;
3153 	ndev->max_mtu = dev->max_mtu;
3154 
3155 	netif_carrier_off(ndev);
3156 
3157 	err = register_netdev(ndev);
3158 	if (err) {
3159 		printk(KERN_ERR "%pOF: failed to register net device (%d)!\n",
3160 		       np, err);
3161 		goto err_detach_tah;
3162 	}
3163 
3164 	/* Set our drvdata last as we don't want them visible until we are
3165 	 * fully initialized
3166 	 */
3167 	wmb();
3168 	platform_set_drvdata(ofdev, dev);
3169 
3170 	/* There's a new kid in town ! Let's tell everybody */
3171 	wake_up_all(&emac_probe_wait);
3172 
3173 
3174 	printk(KERN_INFO "%s: EMAC-%d %pOF, MAC %pM\n",
3175 	       ndev->name, dev->cell_index, np, ndev->dev_addr);
3176 
3177 	if (dev->phy_mode == PHY_INTERFACE_MODE_SGMII)
3178 		printk(KERN_NOTICE "%s: in SGMII mode\n", ndev->name);
3179 
3180 	if (dev->phy.address >= 0)
3181 		printk("%s: found %s PHY (0x%02x)\n", ndev->name,
3182 		       dev->phy.def->name, dev->phy.address);
3183 
3184 	/* Life is good */
3185 	return 0;
3186 
3187 	/* I have a bad feeling about this ... */
3188 
3189  err_detach_tah:
3190 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
3191 		tah_detach(dev->tah_dev, dev->tah_port);
3192  err_detach_rgmii:
3193 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
3194 		rgmii_detach(dev->rgmii_dev, dev->rgmii_port);
3195  err_detach_zmii:
3196 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
3197 		zmii_detach(dev->zmii_dev, dev->zmii_port);
3198  err_unreg_commac:
3199 	mal_unregister_commac(dev->mal, &dev->commac);
3200  err_rel_deps:
3201 	emac_put_deps(dev);
3202  err_reg_unmap:
3203 	iounmap(dev->emacp);
3204  err_irq_unmap:
3205 	if (dev->wol_irq)
3206 		irq_dispose_mapping(dev->wol_irq);
3207 	if (dev->emac_irq)
3208 		irq_dispose_mapping(dev->emac_irq);
3209  err_free:
3210 	free_netdev(ndev);
3211  err_gone:
3212 	/* if we were on the bootlist, remove us as we won't show up and
3213 	 * wake up all waiters to notify them in case they were waiting
3214 	 * on us
3215 	 */
3216 	if (blist) {
3217 		*blist = NULL;
3218 		wake_up_all(&emac_probe_wait);
3219 	}
3220 	return err;
3221 }
3222 
3223 static int emac_remove(struct platform_device *ofdev)
3224 {
3225 	struct emac_instance *dev = platform_get_drvdata(ofdev);
3226 
3227 	DBG(dev, "remove" NL);
3228 
3229 	unregister_netdev(dev->ndev);
3230 
3231 	cancel_work_sync(&dev->reset_work);
3232 
3233 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
3234 		tah_detach(dev->tah_dev, dev->tah_port);
3235 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
3236 		rgmii_detach(dev->rgmii_dev, dev->rgmii_port);
3237 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
3238 		zmii_detach(dev->zmii_dev, dev->zmii_port);
3239 
3240 	if (dev->phy_dev)
3241 		phy_disconnect(dev->phy_dev);
3242 
3243 	if (dev->mii_bus)
3244 		mdiobus_unregister(dev->mii_bus);
3245 
3246 	busy_phy_map &= ~(1 << dev->phy.address);
3247 	DBG(dev, "busy_phy_map now %#x" NL, busy_phy_map);
3248 
3249 	mal_unregister_commac(dev->mal, &dev->commac);
3250 	emac_put_deps(dev);
3251 
3252 	iounmap(dev->emacp);
3253 
3254 	if (dev->wol_irq)
3255 		irq_dispose_mapping(dev->wol_irq);
3256 	if (dev->emac_irq)
3257 		irq_dispose_mapping(dev->emac_irq);
3258 
3259 	free_netdev(dev->ndev);
3260 
3261 	return 0;
3262 }
3263 
3264 /* XXX Features in here should be replaced by properties... */
3265 static const struct of_device_id emac_match[] =
3266 {
3267 	{
3268 		.type		= "network",
3269 		.compatible	= "ibm,emac",
3270 	},
3271 	{
3272 		.type		= "network",
3273 		.compatible	= "ibm,emac4",
3274 	},
3275 	{
3276 		.type		= "network",
3277 		.compatible	= "ibm,emac4sync",
3278 	},
3279 	{},
3280 };
3281 MODULE_DEVICE_TABLE(of, emac_match);
3282 
3283 static struct platform_driver emac_driver = {
3284 	.driver = {
3285 		.name = "emac",
3286 		.of_match_table = emac_match,
3287 	},
3288 	.probe = emac_probe,
3289 	.remove = emac_remove,
3290 };
3291 
3292 static void __init emac_make_bootlist(void)
3293 {
3294 	struct device_node *np = NULL;
3295 	int j, max, i = 0;
3296 	int cell_indices[EMAC_BOOT_LIST_SIZE];
3297 
3298 	/* Collect EMACs */
3299 	while((np = of_find_all_nodes(np)) != NULL) {
3300 		const u32 *idx;
3301 
3302 		if (of_match_node(emac_match, np) == NULL)
3303 			continue;
3304 		if (of_get_property(np, "unused", NULL))
3305 			continue;
3306 		idx = of_get_property(np, "cell-index", NULL);
3307 		if (idx == NULL)
3308 			continue;
3309 		cell_indices[i] = *idx;
3310 		emac_boot_list[i++] = of_node_get(np);
3311 		if (i >= EMAC_BOOT_LIST_SIZE) {
3312 			of_node_put(np);
3313 			break;
3314 		}
3315 	}
3316 	max = i;
3317 
3318 	/* Bubble sort them (doh, what a creative algorithm :-) */
3319 	for (i = 0; max > 1 && (i < (max - 1)); i++)
3320 		for (j = i; j < max; j++) {
3321 			if (cell_indices[i] > cell_indices[j]) {
3322 				swap(emac_boot_list[i], emac_boot_list[j]);
3323 				swap(cell_indices[i], cell_indices[j]);
3324 			}
3325 		}
3326 }
3327 
3328 static int __init emac_init(void)
3329 {
3330 	int rc;
3331 
3332 	printk(KERN_INFO DRV_DESC ", version " DRV_VERSION "\n");
3333 
3334 	/* Build EMAC boot list */
3335 	emac_make_bootlist();
3336 
3337 	/* Init submodules */
3338 	rc = mal_init();
3339 	if (rc)
3340 		goto err;
3341 	rc = zmii_init();
3342 	if (rc)
3343 		goto err_mal;
3344 	rc = rgmii_init();
3345 	if (rc)
3346 		goto err_zmii;
3347 	rc = tah_init();
3348 	if (rc)
3349 		goto err_rgmii;
3350 	rc = platform_driver_register(&emac_driver);
3351 	if (rc)
3352 		goto err_tah;
3353 
3354 	return 0;
3355 
3356  err_tah:
3357 	tah_exit();
3358  err_rgmii:
3359 	rgmii_exit();
3360  err_zmii:
3361 	zmii_exit();
3362  err_mal:
3363 	mal_exit();
3364  err:
3365 	return rc;
3366 }
3367 
3368 static void __exit emac_exit(void)
3369 {
3370 	int i;
3371 
3372 	platform_driver_unregister(&emac_driver);
3373 
3374 	tah_exit();
3375 	rgmii_exit();
3376 	zmii_exit();
3377 	mal_exit();
3378 
3379 	/* Destroy EMAC boot list */
3380 	for (i = 0; i < EMAC_BOOT_LIST_SIZE; i++)
3381 		of_node_put(emac_boot_list[i]);
3382 }
3383 
3384 module_init(emac_init);
3385 module_exit(emac_exit);
3386