xref: /openbmc/linux/drivers/net/ethernet/ibm/emac/core.c (revision 48c926cd)
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_mode == PHY_MODE_GMII ||
203 		phy_mode == PHY_MODE_RGMII ||
204 		phy_mode == PHY_MODE_SGMII ||
205 		phy_mode == PHY_MODE_TBI ||
206 		phy_mode == PHY_MODE_RTBI;
207 }
208 
209 static inline int emac_phy_gpcs(int phy_mode)
210 {
211 	return  phy_mode == PHY_MODE_SGMII ||
212 		phy_mode == PHY_MODE_TBI ||
213 		phy_mode == PHY_MODE_RTBI;
214 }
215 
216 static inline void emac_tx_enable(struct emac_instance *dev)
217 {
218 	struct emac_regs __iomem *p = dev->emacp;
219 	u32 r;
220 
221 	DBG(dev, "tx_enable" NL);
222 
223 	r = in_be32(&p->mr0);
224 	if (!(r & EMAC_MR0_TXE))
225 		out_be32(&p->mr0, r | EMAC_MR0_TXE);
226 }
227 
228 static void emac_tx_disable(struct emac_instance *dev)
229 {
230 	struct emac_regs __iomem *p = dev->emacp;
231 	u32 r;
232 
233 	DBG(dev, "tx_disable" NL);
234 
235 	r = in_be32(&p->mr0);
236 	if (r & EMAC_MR0_TXE) {
237 		int n = dev->stop_timeout;
238 		out_be32(&p->mr0, r & ~EMAC_MR0_TXE);
239 		while (!(in_be32(&p->mr0) & EMAC_MR0_TXI) && n) {
240 			udelay(1);
241 			--n;
242 		}
243 		if (unlikely(!n))
244 			emac_report_timeout_error(dev, "TX disable timeout");
245 	}
246 }
247 
248 static void emac_rx_enable(struct emac_instance *dev)
249 {
250 	struct emac_regs __iomem *p = dev->emacp;
251 	u32 r;
252 
253 	if (unlikely(test_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags)))
254 		goto out;
255 
256 	DBG(dev, "rx_enable" NL);
257 
258 	r = in_be32(&p->mr0);
259 	if (!(r & EMAC_MR0_RXE)) {
260 		if (unlikely(!(r & EMAC_MR0_RXI))) {
261 			/* Wait if previous async disable is still in progress */
262 			int n = dev->stop_timeout;
263 			while (!(r = in_be32(&p->mr0) & EMAC_MR0_RXI) && n) {
264 				udelay(1);
265 				--n;
266 			}
267 			if (unlikely(!n))
268 				emac_report_timeout_error(dev,
269 							  "RX disable timeout");
270 		}
271 		out_be32(&p->mr0, r | EMAC_MR0_RXE);
272 	}
273  out:
274 	;
275 }
276 
277 static void emac_rx_disable(struct emac_instance *dev)
278 {
279 	struct emac_regs __iomem *p = dev->emacp;
280 	u32 r;
281 
282 	DBG(dev, "rx_disable" NL);
283 
284 	r = in_be32(&p->mr0);
285 	if (r & EMAC_MR0_RXE) {
286 		int n = dev->stop_timeout;
287 		out_be32(&p->mr0, r & ~EMAC_MR0_RXE);
288 		while (!(in_be32(&p->mr0) & EMAC_MR0_RXI) && n) {
289 			udelay(1);
290 			--n;
291 		}
292 		if (unlikely(!n))
293 			emac_report_timeout_error(dev, "RX disable timeout");
294 	}
295 }
296 
297 static inline void emac_netif_stop(struct emac_instance *dev)
298 {
299 	netif_tx_lock_bh(dev->ndev);
300 	netif_addr_lock(dev->ndev);
301 	dev->no_mcast = 1;
302 	netif_addr_unlock(dev->ndev);
303 	netif_tx_unlock_bh(dev->ndev);
304 	netif_trans_update(dev->ndev);	/* prevent tx timeout */
305 	mal_poll_disable(dev->mal, &dev->commac);
306 	netif_tx_disable(dev->ndev);
307 }
308 
309 static inline void emac_netif_start(struct emac_instance *dev)
310 {
311 	netif_tx_lock_bh(dev->ndev);
312 	netif_addr_lock(dev->ndev);
313 	dev->no_mcast = 0;
314 	if (dev->mcast_pending && netif_running(dev->ndev))
315 		__emac_set_multicast_list(dev);
316 	netif_addr_unlock(dev->ndev);
317 	netif_tx_unlock_bh(dev->ndev);
318 
319 	netif_wake_queue(dev->ndev);
320 
321 	/* NOTE: unconditional netif_wake_queue is only appropriate
322 	 * so long as all callers are assured to have free tx slots
323 	 * (taken from tg3... though the case where that is wrong is
324 	 *  not terribly harmful)
325 	 */
326 	mal_poll_enable(dev->mal, &dev->commac);
327 }
328 
329 static inline void emac_rx_disable_async(struct emac_instance *dev)
330 {
331 	struct emac_regs __iomem *p = dev->emacp;
332 	u32 r;
333 
334 	DBG(dev, "rx_disable_async" NL);
335 
336 	r = in_be32(&p->mr0);
337 	if (r & EMAC_MR0_RXE)
338 		out_be32(&p->mr0, r & ~EMAC_MR0_RXE);
339 }
340 
341 static int emac_reset(struct emac_instance *dev)
342 {
343 	struct emac_regs __iomem *p = dev->emacp;
344 	int n = 20;
345 	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 4096:
498 		ret |= EMAC_MR1_RFS_4K;
499 		break;
500 	default:
501 		printk(KERN_WARNING "%s: Unknown Rx FIFO size %d\n",
502 		       dev->ndev->name, rx_size);
503 	}
504 
505 	return ret;
506 }
507 
508 static u32 __emac4_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
509 {
510 	u32 ret = EMAC_MR1_VLE | EMAC_MR1_IST | EMAC4_MR1_TR |
511 		EMAC4_MR1_OBCI(dev->opb_bus_freq / 1000000);
512 
513 	DBG2(dev, "__emac4_calc_base_mr1" NL);
514 
515 	switch(tx_size) {
516 	case 16384:
517 		ret |= EMAC4_MR1_TFS_16K;
518 		break;
519 	case 4096:
520 		ret |= EMAC4_MR1_TFS_4K;
521 		break;
522 	case 2048:
523 		ret |= EMAC4_MR1_TFS_2K;
524 		break;
525 	default:
526 		printk(KERN_WARNING "%s: Unknown Tx FIFO size %d\n",
527 		       dev->ndev->name, tx_size);
528 	}
529 
530 	switch(rx_size) {
531 	case 16384:
532 		ret |= EMAC4_MR1_RFS_16K;
533 		break;
534 	case 4096:
535 		ret |= EMAC4_MR1_RFS_4K;
536 		break;
537 	case 2048:
538 		ret |= EMAC4_MR1_RFS_2K;
539 		break;
540 	default:
541 		printk(KERN_WARNING "%s: Unknown Rx FIFO size %d\n",
542 		       dev->ndev->name, rx_size);
543 	}
544 
545 	return ret;
546 }
547 
548 static u32 emac_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
549 {
550 	return emac_has_feature(dev, EMAC_FTR_EMAC4) ?
551 		__emac4_calc_base_mr1(dev, tx_size, rx_size) :
552 		__emac_calc_base_mr1(dev, tx_size, rx_size);
553 }
554 
555 static inline u32 emac_calc_trtr(struct emac_instance *dev, unsigned int size)
556 {
557 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
558 		return ((size >> 6) - 1) << EMAC_TRTR_SHIFT_EMAC4;
559 	else
560 		return ((size >> 6) - 1) << EMAC_TRTR_SHIFT;
561 }
562 
563 static inline u32 emac_calc_rwmr(struct emac_instance *dev,
564 				 unsigned int low, unsigned int high)
565 {
566 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
567 		return (low << 22) | ( (high & 0x3ff) << 6);
568 	else
569 		return (low << 23) | ( (high & 0x1ff) << 7);
570 }
571 
572 static int emac_configure(struct emac_instance *dev)
573 {
574 	struct emac_regs __iomem *p = dev->emacp;
575 	struct net_device *ndev = dev->ndev;
576 	int tx_size, rx_size, link = netif_carrier_ok(dev->ndev);
577 	u32 r, mr1 = 0;
578 
579 	DBG(dev, "configure" NL);
580 
581 	if (!link) {
582 		out_be32(&p->mr1, in_be32(&p->mr1)
583 			 | EMAC_MR1_FDE | EMAC_MR1_ILE);
584 		udelay(100);
585 	} else if (emac_reset(dev) < 0)
586 		return -ETIMEDOUT;
587 
588 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
589 		tah_reset(dev->tah_dev);
590 
591 	DBG(dev, " link = %d duplex = %d, pause = %d, asym_pause = %d\n",
592 	    link, dev->phy.duplex, dev->phy.pause, dev->phy.asym_pause);
593 
594 	/* Default fifo sizes */
595 	tx_size = dev->tx_fifo_size;
596 	rx_size = dev->rx_fifo_size;
597 
598 	/* No link, force loopback */
599 	if (!link)
600 		mr1 = EMAC_MR1_FDE | EMAC_MR1_ILE;
601 
602 	/* Check for full duplex */
603 	else if (dev->phy.duplex == DUPLEX_FULL)
604 		mr1 |= EMAC_MR1_FDE | EMAC_MR1_MWSW_001;
605 
606 	/* Adjust fifo sizes, mr1 and timeouts based on link speed */
607 	dev->stop_timeout = STOP_TIMEOUT_10;
608 	switch (dev->phy.speed) {
609 	case SPEED_1000:
610 		if (emac_phy_gpcs(dev->phy.mode)) {
611 			mr1 |= EMAC_MR1_MF_1000GPCS | EMAC_MR1_MF_IPPA(
612 				(dev->phy.gpcs_address != 0xffffffff) ?
613 				 dev->phy.gpcs_address : dev->phy.address);
614 
615 			/* Put some arbitrary OUI, Manuf & Rev IDs so we can
616 			 * identify this GPCS PHY later.
617 			 */
618 			out_be32(&p->u1.emac4.ipcr, 0xdeadbeef);
619 		} else
620 			mr1 |= EMAC_MR1_MF_1000;
621 
622 		/* Extended fifo sizes */
623 		tx_size = dev->tx_fifo_size_gige;
624 		rx_size = dev->rx_fifo_size_gige;
625 
626 		if (dev->ndev->mtu > ETH_DATA_LEN) {
627 			if (emac_has_feature(dev, EMAC_FTR_EMAC4))
628 				mr1 |= EMAC4_MR1_JPSM;
629 			else
630 				mr1 |= EMAC_MR1_JPSM;
631 			dev->stop_timeout = STOP_TIMEOUT_1000_JUMBO;
632 		} else
633 			dev->stop_timeout = STOP_TIMEOUT_1000;
634 		break;
635 	case SPEED_100:
636 		mr1 |= EMAC_MR1_MF_100;
637 		dev->stop_timeout = STOP_TIMEOUT_100;
638 		break;
639 	default: /* make gcc happy */
640 		break;
641 	}
642 
643 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
644 		rgmii_set_speed(dev->rgmii_dev, dev->rgmii_port,
645 				dev->phy.speed);
646 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
647 		zmii_set_speed(dev->zmii_dev, dev->zmii_port, dev->phy.speed);
648 
649 	/* on 40x erratum forces us to NOT use integrated flow control,
650 	 * let's hope it works on 44x ;)
651 	 */
652 	if (!emac_has_feature(dev, EMAC_FTR_NO_FLOW_CONTROL_40x) &&
653 	    dev->phy.duplex == DUPLEX_FULL) {
654 		if (dev->phy.pause)
655 			mr1 |= EMAC_MR1_EIFC | EMAC_MR1_APP;
656 		else if (dev->phy.asym_pause)
657 			mr1 |= EMAC_MR1_APP;
658 	}
659 
660 	/* Add base settings & fifo sizes & program MR1 */
661 	mr1 |= emac_calc_base_mr1(dev, tx_size, rx_size);
662 	out_be32(&p->mr1, mr1);
663 
664 	/* Set individual MAC address */
665 	out_be32(&p->iahr, (ndev->dev_addr[0] << 8) | ndev->dev_addr[1]);
666 	out_be32(&p->ialr, (ndev->dev_addr[2] << 24) |
667 		 (ndev->dev_addr[3] << 16) | (ndev->dev_addr[4] << 8) |
668 		 ndev->dev_addr[5]);
669 
670 	/* VLAN Tag Protocol ID */
671 	out_be32(&p->vtpid, 0x8100);
672 
673 	/* Receive mode register */
674 	r = emac_iff2rmr(ndev);
675 	if (r & EMAC_RMR_MAE)
676 		emac_hash_mc(dev);
677 	out_be32(&p->rmr, r);
678 
679 	/* FIFOs thresholds */
680 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
681 		r = EMAC4_TMR1((dev->mal_burst_size / dev->fifo_entry_size) + 1,
682 			       tx_size / 2 / dev->fifo_entry_size);
683 	else
684 		r = EMAC_TMR1((dev->mal_burst_size / dev->fifo_entry_size) + 1,
685 			      tx_size / 2 / dev->fifo_entry_size);
686 	out_be32(&p->tmr1, r);
687 	out_be32(&p->trtr, emac_calc_trtr(dev, tx_size / 2));
688 
689 	/* PAUSE frame is sent when RX FIFO reaches its high-water mark,
690 	   there should be still enough space in FIFO to allow the our link
691 	   partner time to process this frame and also time to send PAUSE
692 	   frame itself.
693 
694 	   Here is the worst case scenario for the RX FIFO "headroom"
695 	   (from "The Switch Book") (100Mbps, without preamble, inter-frame gap):
696 
697 	   1) One maximum-length frame on TX                    1522 bytes
698 	   2) One PAUSE frame time                                64 bytes
699 	   3) PAUSE frame decode time allowance                   64 bytes
700 	   4) One maximum-length frame on RX                    1522 bytes
701 	   5) Round-trip propagation delay of the link (100Mb)    15 bytes
702 	   ----------
703 	   3187 bytes
704 
705 	   I chose to set high-water mark to RX_FIFO_SIZE / 4 (1024 bytes)
706 	   low-water mark  to RX_FIFO_SIZE / 8 (512 bytes)
707 	 */
708 	r = emac_calc_rwmr(dev, rx_size / 8 / dev->fifo_entry_size,
709 			   rx_size / 4 / dev->fifo_entry_size);
710 	out_be32(&p->rwmr, r);
711 
712 	/* Set PAUSE timer to the maximum */
713 	out_be32(&p->ptr, 0xffff);
714 
715 	/* IRQ sources */
716 	r = EMAC_ISR_OVR | EMAC_ISR_BP | EMAC_ISR_SE |
717 		EMAC_ISR_ALE | EMAC_ISR_BFCS | EMAC_ISR_PTLE | EMAC_ISR_ORE |
718 		EMAC_ISR_IRE | EMAC_ISR_TE;
719 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
720 	    r |= EMAC4_ISR_TXPE | EMAC4_ISR_RXPE /* | EMAC4_ISR_TXUE |
721 						  EMAC4_ISR_RXOE | */;
722 	out_be32(&p->iser,  r);
723 
724 	/* We need to take GPCS PHY out of isolate mode after EMAC reset */
725 	if (emac_phy_gpcs(dev->phy.mode)) {
726 		if (dev->phy.gpcs_address != 0xffffffff)
727 			emac_mii_reset_gpcs(&dev->phy);
728 		else
729 			emac_mii_reset_phy(&dev->phy);
730 	}
731 
732 	return 0;
733 }
734 
735 static void emac_reinitialize(struct emac_instance *dev)
736 {
737 	DBG(dev, "reinitialize" NL);
738 
739 	emac_netif_stop(dev);
740 	if (!emac_configure(dev)) {
741 		emac_tx_enable(dev);
742 		emac_rx_enable(dev);
743 	}
744 	emac_netif_start(dev);
745 }
746 
747 static void emac_full_tx_reset(struct emac_instance *dev)
748 {
749 	DBG(dev, "full_tx_reset" NL);
750 
751 	emac_tx_disable(dev);
752 	mal_disable_tx_channel(dev->mal, dev->mal_tx_chan);
753 	emac_clean_tx_ring(dev);
754 	dev->tx_cnt = dev->tx_slot = dev->ack_slot = 0;
755 
756 	emac_configure(dev);
757 
758 	mal_enable_tx_channel(dev->mal, dev->mal_tx_chan);
759 	emac_tx_enable(dev);
760 	emac_rx_enable(dev);
761 }
762 
763 static void emac_reset_work(struct work_struct *work)
764 {
765 	struct emac_instance *dev = container_of(work, struct emac_instance, reset_work);
766 
767 	DBG(dev, "reset_work" NL);
768 
769 	mutex_lock(&dev->link_lock);
770 	if (dev->opened) {
771 		emac_netif_stop(dev);
772 		emac_full_tx_reset(dev);
773 		emac_netif_start(dev);
774 	}
775 	mutex_unlock(&dev->link_lock);
776 }
777 
778 static void emac_tx_timeout(struct net_device *ndev)
779 {
780 	struct emac_instance *dev = netdev_priv(ndev);
781 
782 	DBG(dev, "tx_timeout" NL);
783 
784 	schedule_work(&dev->reset_work);
785 }
786 
787 
788 static inline int emac_phy_done(struct emac_instance *dev, u32 stacr)
789 {
790 	int done = !!(stacr & EMAC_STACR_OC);
791 
792 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
793 		done = !done;
794 
795 	return done;
796 };
797 
798 static int __emac_mdio_read(struct emac_instance *dev, u8 id, u8 reg)
799 {
800 	struct emac_regs __iomem *p = dev->emacp;
801 	u32 r = 0;
802 	int n, err = -ETIMEDOUT;
803 
804 	mutex_lock(&dev->mdio_lock);
805 
806 	DBG2(dev, "mdio_read(%02x,%02x)" NL, id, reg);
807 
808 	/* Enable proper MDIO port */
809 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
810 		zmii_get_mdio(dev->zmii_dev, dev->zmii_port);
811 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
812 		rgmii_get_mdio(dev->rgmii_dev, dev->rgmii_port);
813 
814 	/* Wait for management interface to become idle */
815 	n = 20;
816 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
817 		udelay(1);
818 		if (!--n) {
819 			DBG2(dev, " -> timeout wait idle\n");
820 			goto bail;
821 		}
822 	}
823 
824 	/* Issue read command */
825 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
826 		r = EMAC4_STACR_BASE(dev->opb_bus_freq);
827 	else
828 		r = EMAC_STACR_BASE(dev->opb_bus_freq);
829 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
830 		r |= EMAC_STACR_OC;
831 	if (emac_has_feature(dev, EMAC_FTR_HAS_NEW_STACR))
832 		r |= EMACX_STACR_STAC_READ;
833 	else
834 		r |= EMAC_STACR_STAC_READ;
835 	r |= (reg & EMAC_STACR_PRA_MASK)
836 		| ((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT);
837 	out_be32(&p->stacr, r);
838 
839 	/* Wait for read to complete */
840 	n = 200;
841 	while (!emac_phy_done(dev, (r = in_be32(&p->stacr)))) {
842 		udelay(1);
843 		if (!--n) {
844 			DBG2(dev, " -> timeout wait complete\n");
845 			goto bail;
846 		}
847 	}
848 
849 	if (unlikely(r & EMAC_STACR_PHYE)) {
850 		DBG(dev, "mdio_read(%02x, %02x) failed" NL, id, reg);
851 		err = -EREMOTEIO;
852 		goto bail;
853 	}
854 
855 	r = ((r >> EMAC_STACR_PHYD_SHIFT) & EMAC_STACR_PHYD_MASK);
856 
857 	DBG2(dev, "mdio_read -> %04x" NL, r);
858 	err = 0;
859  bail:
860 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
861 		rgmii_put_mdio(dev->rgmii_dev, dev->rgmii_port);
862 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
863 		zmii_put_mdio(dev->zmii_dev, dev->zmii_port);
864 	mutex_unlock(&dev->mdio_lock);
865 
866 	return err == 0 ? r : err;
867 }
868 
869 static void __emac_mdio_write(struct emac_instance *dev, u8 id, u8 reg,
870 			      u16 val)
871 {
872 	struct emac_regs __iomem *p = dev->emacp;
873 	u32 r = 0;
874 	int n, err = -ETIMEDOUT;
875 
876 	mutex_lock(&dev->mdio_lock);
877 
878 	DBG2(dev, "mdio_write(%02x,%02x,%04x)" NL, id, reg, val);
879 
880 	/* Enable proper MDIO port */
881 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
882 		zmii_get_mdio(dev->zmii_dev, dev->zmii_port);
883 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
884 		rgmii_get_mdio(dev->rgmii_dev, dev->rgmii_port);
885 
886 	/* Wait for management interface to be idle */
887 	n = 20;
888 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
889 		udelay(1);
890 		if (!--n) {
891 			DBG2(dev, " -> timeout wait idle\n");
892 			goto bail;
893 		}
894 	}
895 
896 	/* Issue write command */
897 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
898 		r = EMAC4_STACR_BASE(dev->opb_bus_freq);
899 	else
900 		r = EMAC_STACR_BASE(dev->opb_bus_freq);
901 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
902 		r |= EMAC_STACR_OC;
903 	if (emac_has_feature(dev, EMAC_FTR_HAS_NEW_STACR))
904 		r |= EMACX_STACR_STAC_WRITE;
905 	else
906 		r |= EMAC_STACR_STAC_WRITE;
907 	r |= (reg & EMAC_STACR_PRA_MASK) |
908 		((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT) |
909 		(val << EMAC_STACR_PHYD_SHIFT);
910 	out_be32(&p->stacr, r);
911 
912 	/* Wait for write to complete */
913 	n = 200;
914 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
915 		udelay(1);
916 		if (!--n) {
917 			DBG2(dev, " -> timeout wait complete\n");
918 			goto bail;
919 		}
920 	}
921 	err = 0;
922  bail:
923 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
924 		rgmii_put_mdio(dev->rgmii_dev, dev->rgmii_port);
925 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
926 		zmii_put_mdio(dev->zmii_dev, dev->zmii_port);
927 	mutex_unlock(&dev->mdio_lock);
928 }
929 
930 static int emac_mdio_read(struct net_device *ndev, int id, int reg)
931 {
932 	struct emac_instance *dev = netdev_priv(ndev);
933 	int res;
934 
935 	res = __emac_mdio_read((dev->mdio_instance &&
936 				dev->phy.gpcs_address != id) ?
937 				dev->mdio_instance : dev,
938 			       (u8) id, (u8) reg);
939 	return res;
940 }
941 
942 static void emac_mdio_write(struct net_device *ndev, int id, int reg, int val)
943 {
944 	struct emac_instance *dev = netdev_priv(ndev);
945 
946 	__emac_mdio_write((dev->mdio_instance &&
947 			   dev->phy.gpcs_address != id) ?
948 			   dev->mdio_instance : dev,
949 			  (u8) id, (u8) reg, (u16) val);
950 }
951 
952 /* Tx lock BH */
953 static void __emac_set_multicast_list(struct emac_instance *dev)
954 {
955 	struct emac_regs __iomem *p = dev->emacp;
956 	u32 rmr = emac_iff2rmr(dev->ndev);
957 
958 	DBG(dev, "__multicast %08x" NL, rmr);
959 
960 	/* I decided to relax register access rules here to avoid
961 	 * full EMAC reset.
962 	 *
963 	 * There is a real problem with EMAC4 core if we use MWSW_001 bit
964 	 * in MR1 register and do a full EMAC reset.
965 	 * One TX BD status update is delayed and, after EMAC reset, it
966 	 * never happens, resulting in TX hung (it'll be recovered by TX
967 	 * timeout handler eventually, but this is just gross).
968 	 * So we either have to do full TX reset or try to cheat here :)
969 	 *
970 	 * The only required change is to RX mode register, so I *think* all
971 	 * we need is just to stop RX channel. This seems to work on all
972 	 * tested SoCs.                                                --ebs
973 	 *
974 	 * If we need the full reset, we might just trigger the workqueue
975 	 * and do it async... a bit nasty but should work --BenH
976 	 */
977 	dev->mcast_pending = 0;
978 	emac_rx_disable(dev);
979 	if (rmr & EMAC_RMR_MAE)
980 		emac_hash_mc(dev);
981 	out_be32(&p->rmr, rmr);
982 	emac_rx_enable(dev);
983 }
984 
985 /* Tx lock BH */
986 static void emac_set_multicast_list(struct net_device *ndev)
987 {
988 	struct emac_instance *dev = netdev_priv(ndev);
989 
990 	DBG(dev, "multicast" NL);
991 
992 	BUG_ON(!netif_running(dev->ndev));
993 
994 	if (dev->no_mcast) {
995 		dev->mcast_pending = 1;
996 		return;
997 	}
998 
999 	mutex_lock(&dev->link_lock);
1000 	__emac_set_multicast_list(dev);
1001 	mutex_unlock(&dev->link_lock);
1002 }
1003 
1004 static int emac_set_mac_address(struct net_device *ndev, void *sa)
1005 {
1006 	struct emac_instance *dev = netdev_priv(ndev);
1007 	struct sockaddr *addr = sa;
1008 	struct emac_regs __iomem *p = dev->emacp;
1009 
1010 	if (!is_valid_ether_addr(addr->sa_data))
1011 	       return -EADDRNOTAVAIL;
1012 
1013 	mutex_lock(&dev->link_lock);
1014 
1015 	memcpy(ndev->dev_addr, addr->sa_data, ndev->addr_len);
1016 
1017 	emac_rx_disable(dev);
1018 	emac_tx_disable(dev);
1019 	out_be32(&p->iahr, (ndev->dev_addr[0] << 8) | ndev->dev_addr[1]);
1020 	out_be32(&p->ialr, (ndev->dev_addr[2] << 24) |
1021 		(ndev->dev_addr[3] << 16) | (ndev->dev_addr[4] << 8) |
1022 		ndev->dev_addr[5]);
1023 	emac_tx_enable(dev);
1024 	emac_rx_enable(dev);
1025 
1026 	mutex_unlock(&dev->link_lock);
1027 
1028 	return 0;
1029 }
1030 
1031 static int emac_resize_rx_ring(struct emac_instance *dev, int new_mtu)
1032 {
1033 	int rx_sync_size = emac_rx_sync_size(new_mtu);
1034 	int rx_skb_size = emac_rx_skb_size(new_mtu);
1035 	int i, ret = 0;
1036 	int mr1_jumbo_bit_change = 0;
1037 
1038 	mutex_lock(&dev->link_lock);
1039 	emac_netif_stop(dev);
1040 	emac_rx_disable(dev);
1041 	mal_disable_rx_channel(dev->mal, dev->mal_rx_chan);
1042 
1043 	if (dev->rx_sg_skb) {
1044 		++dev->estats.rx_dropped_resize;
1045 		dev_kfree_skb(dev->rx_sg_skb);
1046 		dev->rx_sg_skb = NULL;
1047 	}
1048 
1049 	/* Make a first pass over RX ring and mark BDs ready, dropping
1050 	 * non-processed packets on the way. We need this as a separate pass
1051 	 * to simplify error recovery in the case of allocation failure later.
1052 	 */
1053 	for (i = 0; i < NUM_RX_BUFF; ++i) {
1054 		if (dev->rx_desc[i].ctrl & MAL_RX_CTRL_FIRST)
1055 			++dev->estats.rx_dropped_resize;
1056 
1057 		dev->rx_desc[i].data_len = 0;
1058 		dev->rx_desc[i].ctrl = MAL_RX_CTRL_EMPTY |
1059 		    (i == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1060 	}
1061 
1062 	/* Reallocate RX ring only if bigger skb buffers are required */
1063 	if (rx_skb_size <= dev->rx_skb_size)
1064 		goto skip;
1065 
1066 	/* Second pass, allocate new skbs */
1067 	for (i = 0; i < NUM_RX_BUFF; ++i) {
1068 		struct sk_buff *skb = alloc_skb(rx_skb_size, GFP_ATOMIC);
1069 		if (!skb) {
1070 			ret = -ENOMEM;
1071 			goto oom;
1072 		}
1073 
1074 		BUG_ON(!dev->rx_skb[i]);
1075 		dev_kfree_skb(dev->rx_skb[i]);
1076 
1077 		skb_reserve(skb, EMAC_RX_SKB_HEADROOM + 2);
1078 		dev->rx_desc[i].data_ptr =
1079 		    dma_map_single(&dev->ofdev->dev, skb->data - 2, rx_sync_size,
1080 				   DMA_FROM_DEVICE) + 2;
1081 		dev->rx_skb[i] = skb;
1082 	}
1083  skip:
1084 	/* Check if we need to change "Jumbo" bit in MR1 */
1085 	if (emac_has_feature(dev, EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE)) {
1086 		mr1_jumbo_bit_change = (new_mtu > ETH_DATA_LEN) ||
1087 				(dev->ndev->mtu > ETH_DATA_LEN);
1088 	} else {
1089 		mr1_jumbo_bit_change = (new_mtu > ETH_DATA_LEN) ^
1090 				(dev->ndev->mtu > ETH_DATA_LEN);
1091 	}
1092 
1093 	if (mr1_jumbo_bit_change) {
1094 		/* This is to prevent starting RX channel in emac_rx_enable() */
1095 		set_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1096 
1097 		dev->ndev->mtu = new_mtu;
1098 		emac_full_tx_reset(dev);
1099 	}
1100 
1101 	mal_set_rcbs(dev->mal, dev->mal_rx_chan, emac_rx_size(new_mtu));
1102  oom:
1103 	/* Restart RX */
1104 	clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1105 	dev->rx_slot = 0;
1106 	mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1107 	emac_rx_enable(dev);
1108 	emac_netif_start(dev);
1109 	mutex_unlock(&dev->link_lock);
1110 
1111 	return ret;
1112 }
1113 
1114 /* Process ctx, rtnl_lock semaphore */
1115 static int emac_change_mtu(struct net_device *ndev, int new_mtu)
1116 {
1117 	struct emac_instance *dev = netdev_priv(ndev);
1118 	int ret = 0;
1119 
1120 	DBG(dev, "change_mtu(%d)" NL, new_mtu);
1121 
1122 	if (netif_running(ndev)) {
1123 		/* Check if we really need to reinitialize RX ring */
1124 		if (emac_rx_skb_size(ndev->mtu) != emac_rx_skb_size(new_mtu))
1125 			ret = emac_resize_rx_ring(dev, new_mtu);
1126 	}
1127 
1128 	if (!ret) {
1129 		ndev->mtu = new_mtu;
1130 		dev->rx_skb_size = emac_rx_skb_size(new_mtu);
1131 		dev->rx_sync_size = emac_rx_sync_size(new_mtu);
1132 	}
1133 
1134 	return ret;
1135 }
1136 
1137 static void emac_clean_tx_ring(struct emac_instance *dev)
1138 {
1139 	int i;
1140 
1141 	for (i = 0; i < NUM_TX_BUFF; ++i) {
1142 		if (dev->tx_skb[i]) {
1143 			dev_kfree_skb(dev->tx_skb[i]);
1144 			dev->tx_skb[i] = NULL;
1145 			if (dev->tx_desc[i].ctrl & MAL_TX_CTRL_READY)
1146 				++dev->estats.tx_dropped;
1147 		}
1148 		dev->tx_desc[i].ctrl = 0;
1149 		dev->tx_desc[i].data_ptr = 0;
1150 	}
1151 }
1152 
1153 static void emac_clean_rx_ring(struct emac_instance *dev)
1154 {
1155 	int i;
1156 
1157 	for (i = 0; i < NUM_RX_BUFF; ++i)
1158 		if (dev->rx_skb[i]) {
1159 			dev->rx_desc[i].ctrl = 0;
1160 			dev_kfree_skb(dev->rx_skb[i]);
1161 			dev->rx_skb[i] = NULL;
1162 			dev->rx_desc[i].data_ptr = 0;
1163 		}
1164 
1165 	if (dev->rx_sg_skb) {
1166 		dev_kfree_skb(dev->rx_sg_skb);
1167 		dev->rx_sg_skb = NULL;
1168 	}
1169 }
1170 
1171 static inline int emac_alloc_rx_skb(struct emac_instance *dev, int slot,
1172 				    gfp_t flags)
1173 {
1174 	struct sk_buff *skb = alloc_skb(dev->rx_skb_size, flags);
1175 	if (unlikely(!skb))
1176 		return -ENOMEM;
1177 
1178 	dev->rx_skb[slot] = skb;
1179 	dev->rx_desc[slot].data_len = 0;
1180 
1181 	skb_reserve(skb, EMAC_RX_SKB_HEADROOM + 2);
1182 	dev->rx_desc[slot].data_ptr =
1183 	    dma_map_single(&dev->ofdev->dev, skb->data - 2, dev->rx_sync_size,
1184 			   DMA_FROM_DEVICE) + 2;
1185 	wmb();
1186 	dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1187 	    (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1188 
1189 	return 0;
1190 }
1191 
1192 static void emac_print_link_status(struct emac_instance *dev)
1193 {
1194 	if (netif_carrier_ok(dev->ndev))
1195 		printk(KERN_INFO "%s: link is up, %d %s%s\n",
1196 		       dev->ndev->name, dev->phy.speed,
1197 		       dev->phy.duplex == DUPLEX_FULL ? "FDX" : "HDX",
1198 		       dev->phy.pause ? ", pause enabled" :
1199 		       dev->phy.asym_pause ? ", asymmetric pause enabled" : "");
1200 	else
1201 		printk(KERN_INFO "%s: link is down\n", dev->ndev->name);
1202 }
1203 
1204 /* Process ctx, rtnl_lock semaphore */
1205 static int emac_open(struct net_device *ndev)
1206 {
1207 	struct emac_instance *dev = netdev_priv(ndev);
1208 	int err, i;
1209 
1210 	DBG(dev, "open" NL);
1211 
1212 	/* Setup error IRQ handler */
1213 	err = request_irq(dev->emac_irq, emac_irq, 0, "EMAC", dev);
1214 	if (err) {
1215 		printk(KERN_ERR "%s: failed to request IRQ %d\n",
1216 		       ndev->name, dev->emac_irq);
1217 		return err;
1218 	}
1219 
1220 	/* Allocate RX ring */
1221 	for (i = 0; i < NUM_RX_BUFF; ++i)
1222 		if (emac_alloc_rx_skb(dev, i, GFP_KERNEL)) {
1223 			printk(KERN_ERR "%s: failed to allocate RX ring\n",
1224 			       ndev->name);
1225 			goto oom;
1226 		}
1227 
1228 	dev->tx_cnt = dev->tx_slot = dev->ack_slot = dev->rx_slot = 0;
1229 	clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1230 	dev->rx_sg_skb = NULL;
1231 
1232 	mutex_lock(&dev->link_lock);
1233 	dev->opened = 1;
1234 
1235 	/* Start PHY polling now.
1236 	 */
1237 	if (dev->phy.address >= 0) {
1238 		int link_poll_interval;
1239 		if (dev->phy.def->ops->poll_link(&dev->phy)) {
1240 			dev->phy.def->ops->read_link(&dev->phy);
1241 			emac_rx_clk_default(dev);
1242 			netif_carrier_on(dev->ndev);
1243 			link_poll_interval = PHY_POLL_LINK_ON;
1244 		} else {
1245 			emac_rx_clk_tx(dev);
1246 			netif_carrier_off(dev->ndev);
1247 			link_poll_interval = PHY_POLL_LINK_OFF;
1248 		}
1249 		dev->link_polling = 1;
1250 		wmb();
1251 		schedule_delayed_work(&dev->link_work, link_poll_interval);
1252 		emac_print_link_status(dev);
1253 	} else
1254 		netif_carrier_on(dev->ndev);
1255 
1256 	/* Required for Pause packet support in EMAC */
1257 	dev_mc_add_global(ndev, default_mcast_addr);
1258 
1259 	emac_configure(dev);
1260 	mal_poll_add(dev->mal, &dev->commac);
1261 	mal_enable_tx_channel(dev->mal, dev->mal_tx_chan);
1262 	mal_set_rcbs(dev->mal, dev->mal_rx_chan, emac_rx_size(ndev->mtu));
1263 	mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1264 	emac_tx_enable(dev);
1265 	emac_rx_enable(dev);
1266 	emac_netif_start(dev);
1267 
1268 	mutex_unlock(&dev->link_lock);
1269 
1270 	return 0;
1271  oom:
1272 	emac_clean_rx_ring(dev);
1273 	free_irq(dev->emac_irq, dev);
1274 
1275 	return -ENOMEM;
1276 }
1277 
1278 /* BHs disabled */
1279 #if 0
1280 static int emac_link_differs(struct emac_instance *dev)
1281 {
1282 	u32 r = in_be32(&dev->emacp->mr1);
1283 
1284 	int duplex = r & EMAC_MR1_FDE ? DUPLEX_FULL : DUPLEX_HALF;
1285 	int speed, pause, asym_pause;
1286 
1287 	if (r & EMAC_MR1_MF_1000)
1288 		speed = SPEED_1000;
1289 	else if (r & EMAC_MR1_MF_100)
1290 		speed = SPEED_100;
1291 	else
1292 		speed = SPEED_10;
1293 
1294 	switch (r & (EMAC_MR1_EIFC | EMAC_MR1_APP)) {
1295 	case (EMAC_MR1_EIFC | EMAC_MR1_APP):
1296 		pause = 1;
1297 		asym_pause = 0;
1298 		break;
1299 	case EMAC_MR1_APP:
1300 		pause = 0;
1301 		asym_pause = 1;
1302 		break;
1303 	default:
1304 		pause = asym_pause = 0;
1305 	}
1306 	return speed != dev->phy.speed || duplex != dev->phy.duplex ||
1307 	    pause != dev->phy.pause || asym_pause != dev->phy.asym_pause;
1308 }
1309 #endif
1310 
1311 static void emac_link_timer(struct work_struct *work)
1312 {
1313 	struct emac_instance *dev =
1314 		container_of(to_delayed_work(work),
1315 			     struct emac_instance, link_work);
1316 	int link_poll_interval;
1317 
1318 	mutex_lock(&dev->link_lock);
1319 	DBG2(dev, "link timer" NL);
1320 
1321 	if (!dev->opened)
1322 		goto bail;
1323 
1324 	if (dev->phy.def->ops->poll_link(&dev->phy)) {
1325 		if (!netif_carrier_ok(dev->ndev)) {
1326 			emac_rx_clk_default(dev);
1327 			/* Get new link parameters */
1328 			dev->phy.def->ops->read_link(&dev->phy);
1329 
1330 			netif_carrier_on(dev->ndev);
1331 			emac_netif_stop(dev);
1332 			emac_full_tx_reset(dev);
1333 			emac_netif_start(dev);
1334 			emac_print_link_status(dev);
1335 		}
1336 		link_poll_interval = PHY_POLL_LINK_ON;
1337 	} else {
1338 		if (netif_carrier_ok(dev->ndev)) {
1339 			emac_rx_clk_tx(dev);
1340 			netif_carrier_off(dev->ndev);
1341 			netif_tx_disable(dev->ndev);
1342 			emac_reinitialize(dev);
1343 			emac_print_link_status(dev);
1344 		}
1345 		link_poll_interval = PHY_POLL_LINK_OFF;
1346 	}
1347 	schedule_delayed_work(&dev->link_work, link_poll_interval);
1348  bail:
1349 	mutex_unlock(&dev->link_lock);
1350 }
1351 
1352 static void emac_force_link_update(struct emac_instance *dev)
1353 {
1354 	netif_carrier_off(dev->ndev);
1355 	smp_rmb();
1356 	if (dev->link_polling) {
1357 		cancel_delayed_work_sync(&dev->link_work);
1358 		if (dev->link_polling)
1359 			schedule_delayed_work(&dev->link_work,  PHY_POLL_LINK_OFF);
1360 	}
1361 }
1362 
1363 /* Process ctx, rtnl_lock semaphore */
1364 static int emac_close(struct net_device *ndev)
1365 {
1366 	struct emac_instance *dev = netdev_priv(ndev);
1367 
1368 	DBG(dev, "close" NL);
1369 
1370 	if (dev->phy.address >= 0) {
1371 		dev->link_polling = 0;
1372 		cancel_delayed_work_sync(&dev->link_work);
1373 	}
1374 	mutex_lock(&dev->link_lock);
1375 	emac_netif_stop(dev);
1376 	dev->opened = 0;
1377 	mutex_unlock(&dev->link_lock);
1378 
1379 	emac_rx_disable(dev);
1380 	emac_tx_disable(dev);
1381 	mal_disable_rx_channel(dev->mal, dev->mal_rx_chan);
1382 	mal_disable_tx_channel(dev->mal, dev->mal_tx_chan);
1383 	mal_poll_del(dev->mal, &dev->commac);
1384 
1385 	emac_clean_tx_ring(dev);
1386 	emac_clean_rx_ring(dev);
1387 
1388 	free_irq(dev->emac_irq, dev);
1389 
1390 	netif_carrier_off(ndev);
1391 
1392 	return 0;
1393 }
1394 
1395 static inline u16 emac_tx_csum(struct emac_instance *dev,
1396 			       struct sk_buff *skb)
1397 {
1398 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH) &&
1399 		(skb->ip_summed == CHECKSUM_PARTIAL)) {
1400 		++dev->stats.tx_packets_csum;
1401 		return EMAC_TX_CTRL_TAH_CSUM;
1402 	}
1403 	return 0;
1404 }
1405 
1406 static inline int emac_xmit_finish(struct emac_instance *dev, int len)
1407 {
1408 	struct emac_regs __iomem *p = dev->emacp;
1409 	struct net_device *ndev = dev->ndev;
1410 
1411 	/* Send the packet out. If the if makes a significant perf
1412 	 * difference, then we can store the TMR0 value in "dev"
1413 	 * instead
1414 	 */
1415 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
1416 		out_be32(&p->tmr0, EMAC4_TMR0_XMIT);
1417 	else
1418 		out_be32(&p->tmr0, EMAC_TMR0_XMIT);
1419 
1420 	if (unlikely(++dev->tx_cnt == NUM_TX_BUFF)) {
1421 		netif_stop_queue(ndev);
1422 		DBG2(dev, "stopped TX queue" NL);
1423 	}
1424 
1425 	netif_trans_update(ndev);
1426 	++dev->stats.tx_packets;
1427 	dev->stats.tx_bytes += len;
1428 
1429 	return NETDEV_TX_OK;
1430 }
1431 
1432 /* Tx lock BH */
1433 static int emac_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1434 {
1435 	struct emac_instance *dev = netdev_priv(ndev);
1436 	unsigned int len = skb->len;
1437 	int slot;
1438 
1439 	u16 ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1440 	    MAL_TX_CTRL_LAST | emac_tx_csum(dev, skb);
1441 
1442 	slot = dev->tx_slot++;
1443 	if (dev->tx_slot == NUM_TX_BUFF) {
1444 		dev->tx_slot = 0;
1445 		ctrl |= MAL_TX_CTRL_WRAP;
1446 	}
1447 
1448 	DBG2(dev, "xmit(%u) %d" NL, len, slot);
1449 
1450 	dev->tx_skb[slot] = skb;
1451 	dev->tx_desc[slot].data_ptr = dma_map_single(&dev->ofdev->dev,
1452 						     skb->data, len,
1453 						     DMA_TO_DEVICE);
1454 	dev->tx_desc[slot].data_len = (u16) len;
1455 	wmb();
1456 	dev->tx_desc[slot].ctrl = ctrl;
1457 
1458 	return emac_xmit_finish(dev, len);
1459 }
1460 
1461 static inline int emac_xmit_split(struct emac_instance *dev, int slot,
1462 				  u32 pd, int len, int last, u16 base_ctrl)
1463 {
1464 	while (1) {
1465 		u16 ctrl = base_ctrl;
1466 		int chunk = min(len, MAL_MAX_TX_SIZE);
1467 		len -= chunk;
1468 
1469 		slot = (slot + 1) % NUM_TX_BUFF;
1470 
1471 		if (last && !len)
1472 			ctrl |= MAL_TX_CTRL_LAST;
1473 		if (slot == NUM_TX_BUFF - 1)
1474 			ctrl |= MAL_TX_CTRL_WRAP;
1475 
1476 		dev->tx_skb[slot] = NULL;
1477 		dev->tx_desc[slot].data_ptr = pd;
1478 		dev->tx_desc[slot].data_len = (u16) chunk;
1479 		dev->tx_desc[slot].ctrl = ctrl;
1480 		++dev->tx_cnt;
1481 
1482 		if (!len)
1483 			break;
1484 
1485 		pd += chunk;
1486 	}
1487 	return slot;
1488 }
1489 
1490 /* Tx lock BH disabled (SG version for TAH equipped EMACs) */
1491 static int emac_start_xmit_sg(struct sk_buff *skb, struct net_device *ndev)
1492 {
1493 	struct emac_instance *dev = netdev_priv(ndev);
1494 	int nr_frags = skb_shinfo(skb)->nr_frags;
1495 	int len = skb->len, chunk;
1496 	int slot, i;
1497 	u16 ctrl;
1498 	u32 pd;
1499 
1500 	/* This is common "fast" path */
1501 	if (likely(!nr_frags && len <= MAL_MAX_TX_SIZE))
1502 		return emac_start_xmit(skb, ndev);
1503 
1504 	len -= skb->data_len;
1505 
1506 	/* Note, this is only an *estimation*, we can still run out of empty
1507 	 * slots because of the additional fragmentation into
1508 	 * MAL_MAX_TX_SIZE-sized chunks
1509 	 */
1510 	if (unlikely(dev->tx_cnt + nr_frags + mal_tx_chunks(len) > NUM_TX_BUFF))
1511 		goto stop_queue;
1512 
1513 	ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1514 	    emac_tx_csum(dev, skb);
1515 	slot = dev->tx_slot;
1516 
1517 	/* skb data */
1518 	dev->tx_skb[slot] = NULL;
1519 	chunk = min(len, MAL_MAX_TX_SIZE);
1520 	dev->tx_desc[slot].data_ptr = pd =
1521 	    dma_map_single(&dev->ofdev->dev, skb->data, len, DMA_TO_DEVICE);
1522 	dev->tx_desc[slot].data_len = (u16) chunk;
1523 	len -= chunk;
1524 	if (unlikely(len))
1525 		slot = emac_xmit_split(dev, slot, pd + chunk, len, !nr_frags,
1526 				       ctrl);
1527 	/* skb fragments */
1528 	for (i = 0; i < nr_frags; ++i) {
1529 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1530 		len = skb_frag_size(frag);
1531 
1532 		if (unlikely(dev->tx_cnt + mal_tx_chunks(len) >= NUM_TX_BUFF))
1533 			goto undo_frame;
1534 
1535 		pd = skb_frag_dma_map(&dev->ofdev->dev, frag, 0, len,
1536 				      DMA_TO_DEVICE);
1537 
1538 		slot = emac_xmit_split(dev, slot, pd, len, i == nr_frags - 1,
1539 				       ctrl);
1540 	}
1541 
1542 	DBG2(dev, "xmit_sg(%u) %d - %d" NL, skb->len, dev->tx_slot, slot);
1543 
1544 	/* Attach skb to the last slot so we don't release it too early */
1545 	dev->tx_skb[slot] = skb;
1546 
1547 	/* Send the packet out */
1548 	if (dev->tx_slot == NUM_TX_BUFF - 1)
1549 		ctrl |= MAL_TX_CTRL_WRAP;
1550 	wmb();
1551 	dev->tx_desc[dev->tx_slot].ctrl = ctrl;
1552 	dev->tx_slot = (slot + 1) % NUM_TX_BUFF;
1553 
1554 	return emac_xmit_finish(dev, skb->len);
1555 
1556  undo_frame:
1557 	/* Well, too bad. Our previous estimation was overly optimistic.
1558 	 * Undo everything.
1559 	 */
1560 	while (slot != dev->tx_slot) {
1561 		dev->tx_desc[slot].ctrl = 0;
1562 		--dev->tx_cnt;
1563 		if (--slot < 0)
1564 			slot = NUM_TX_BUFF - 1;
1565 	}
1566 	++dev->estats.tx_undo;
1567 
1568  stop_queue:
1569 	netif_stop_queue(ndev);
1570 	DBG2(dev, "stopped TX queue" NL);
1571 	return NETDEV_TX_BUSY;
1572 }
1573 
1574 /* Tx lock BHs */
1575 static void emac_parse_tx_error(struct emac_instance *dev, u16 ctrl)
1576 {
1577 	struct emac_error_stats *st = &dev->estats;
1578 
1579 	DBG(dev, "BD TX error %04x" NL, ctrl);
1580 
1581 	++st->tx_bd_errors;
1582 	if (ctrl & EMAC_TX_ST_BFCS)
1583 		++st->tx_bd_bad_fcs;
1584 	if (ctrl & EMAC_TX_ST_LCS)
1585 		++st->tx_bd_carrier_loss;
1586 	if (ctrl & EMAC_TX_ST_ED)
1587 		++st->tx_bd_excessive_deferral;
1588 	if (ctrl & EMAC_TX_ST_EC)
1589 		++st->tx_bd_excessive_collisions;
1590 	if (ctrl & EMAC_TX_ST_LC)
1591 		++st->tx_bd_late_collision;
1592 	if (ctrl & EMAC_TX_ST_MC)
1593 		++st->tx_bd_multple_collisions;
1594 	if (ctrl & EMAC_TX_ST_SC)
1595 		++st->tx_bd_single_collision;
1596 	if (ctrl & EMAC_TX_ST_UR)
1597 		++st->tx_bd_underrun;
1598 	if (ctrl & EMAC_TX_ST_SQE)
1599 		++st->tx_bd_sqe;
1600 }
1601 
1602 static void emac_poll_tx(void *param)
1603 {
1604 	struct emac_instance *dev = param;
1605 	u32 bad_mask;
1606 
1607 	DBG2(dev, "poll_tx, %d %d" NL, dev->tx_cnt, dev->ack_slot);
1608 
1609 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
1610 		bad_mask = EMAC_IS_BAD_TX_TAH;
1611 	else
1612 		bad_mask = EMAC_IS_BAD_TX;
1613 
1614 	netif_tx_lock_bh(dev->ndev);
1615 	if (dev->tx_cnt) {
1616 		u16 ctrl;
1617 		int slot = dev->ack_slot, n = 0;
1618 	again:
1619 		ctrl = dev->tx_desc[slot].ctrl;
1620 		if (!(ctrl & MAL_TX_CTRL_READY)) {
1621 			struct sk_buff *skb = dev->tx_skb[slot];
1622 			++n;
1623 
1624 			if (skb) {
1625 				dev_kfree_skb(skb);
1626 				dev->tx_skb[slot] = NULL;
1627 			}
1628 			slot = (slot + 1) % NUM_TX_BUFF;
1629 
1630 			if (unlikely(ctrl & bad_mask))
1631 				emac_parse_tx_error(dev, ctrl);
1632 
1633 			if (--dev->tx_cnt)
1634 				goto again;
1635 		}
1636 		if (n) {
1637 			dev->ack_slot = slot;
1638 			if (netif_queue_stopped(dev->ndev) &&
1639 			    dev->tx_cnt < EMAC_TX_WAKEUP_THRESH)
1640 				netif_wake_queue(dev->ndev);
1641 
1642 			DBG2(dev, "tx %d pkts" NL, n);
1643 		}
1644 	}
1645 	netif_tx_unlock_bh(dev->ndev);
1646 }
1647 
1648 static inline void emac_recycle_rx_skb(struct emac_instance *dev, int slot,
1649 				       int len)
1650 {
1651 	struct sk_buff *skb = dev->rx_skb[slot];
1652 
1653 	DBG2(dev, "recycle %d %d" NL, slot, len);
1654 
1655 	if (len)
1656 		dma_map_single(&dev->ofdev->dev, skb->data - 2,
1657 			       EMAC_DMA_ALIGN(len + 2), DMA_FROM_DEVICE);
1658 
1659 	dev->rx_desc[slot].data_len = 0;
1660 	wmb();
1661 	dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1662 	    (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1663 }
1664 
1665 static void emac_parse_rx_error(struct emac_instance *dev, u16 ctrl)
1666 {
1667 	struct emac_error_stats *st = &dev->estats;
1668 
1669 	DBG(dev, "BD RX error %04x" NL, ctrl);
1670 
1671 	++st->rx_bd_errors;
1672 	if (ctrl & EMAC_RX_ST_OE)
1673 		++st->rx_bd_overrun;
1674 	if (ctrl & EMAC_RX_ST_BP)
1675 		++st->rx_bd_bad_packet;
1676 	if (ctrl & EMAC_RX_ST_RP)
1677 		++st->rx_bd_runt_packet;
1678 	if (ctrl & EMAC_RX_ST_SE)
1679 		++st->rx_bd_short_event;
1680 	if (ctrl & EMAC_RX_ST_AE)
1681 		++st->rx_bd_alignment_error;
1682 	if (ctrl & EMAC_RX_ST_BFCS)
1683 		++st->rx_bd_bad_fcs;
1684 	if (ctrl & EMAC_RX_ST_PTL)
1685 		++st->rx_bd_packet_too_long;
1686 	if (ctrl & EMAC_RX_ST_ORE)
1687 		++st->rx_bd_out_of_range;
1688 	if (ctrl & EMAC_RX_ST_IRE)
1689 		++st->rx_bd_in_range;
1690 }
1691 
1692 static inline void emac_rx_csum(struct emac_instance *dev,
1693 				struct sk_buff *skb, u16 ctrl)
1694 {
1695 #ifdef CONFIG_IBM_EMAC_TAH
1696 	if (!ctrl && dev->tah_dev) {
1697 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1698 		++dev->stats.rx_packets_csum;
1699 	}
1700 #endif
1701 }
1702 
1703 static inline int emac_rx_sg_append(struct emac_instance *dev, int slot)
1704 {
1705 	if (likely(dev->rx_sg_skb != NULL)) {
1706 		int len = dev->rx_desc[slot].data_len;
1707 		int tot_len = dev->rx_sg_skb->len + len;
1708 
1709 		if (unlikely(tot_len + 2 > dev->rx_skb_size)) {
1710 			++dev->estats.rx_dropped_mtu;
1711 			dev_kfree_skb(dev->rx_sg_skb);
1712 			dev->rx_sg_skb = NULL;
1713 		} else {
1714 			memcpy(skb_tail_pointer(dev->rx_sg_skb),
1715 					 dev->rx_skb[slot]->data, len);
1716 			skb_put(dev->rx_sg_skb, len);
1717 			emac_recycle_rx_skb(dev, slot, len);
1718 			return 0;
1719 		}
1720 	}
1721 	emac_recycle_rx_skb(dev, slot, 0);
1722 	return -1;
1723 }
1724 
1725 /* NAPI poll context */
1726 static int emac_poll_rx(void *param, int budget)
1727 {
1728 	struct emac_instance *dev = param;
1729 	int slot = dev->rx_slot, received = 0;
1730 
1731 	DBG2(dev, "poll_rx(%d)" NL, budget);
1732 
1733  again:
1734 	while (budget > 0) {
1735 		int len;
1736 		struct sk_buff *skb;
1737 		u16 ctrl = dev->rx_desc[slot].ctrl;
1738 
1739 		if (ctrl & MAL_RX_CTRL_EMPTY)
1740 			break;
1741 
1742 		skb = dev->rx_skb[slot];
1743 		mb();
1744 		len = dev->rx_desc[slot].data_len;
1745 
1746 		if (unlikely(!MAL_IS_SINGLE_RX(ctrl)))
1747 			goto sg;
1748 
1749 		ctrl &= EMAC_BAD_RX_MASK;
1750 		if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1751 			emac_parse_rx_error(dev, ctrl);
1752 			++dev->estats.rx_dropped_error;
1753 			emac_recycle_rx_skb(dev, slot, 0);
1754 			len = 0;
1755 			goto next;
1756 		}
1757 
1758 		if (len < ETH_HLEN) {
1759 			++dev->estats.rx_dropped_stack;
1760 			emac_recycle_rx_skb(dev, slot, len);
1761 			goto next;
1762 		}
1763 
1764 		if (len && len < EMAC_RX_COPY_THRESH) {
1765 			struct sk_buff *copy_skb =
1766 			    alloc_skb(len + EMAC_RX_SKB_HEADROOM + 2, GFP_ATOMIC);
1767 			if (unlikely(!copy_skb))
1768 				goto oom;
1769 
1770 			skb_reserve(copy_skb, EMAC_RX_SKB_HEADROOM + 2);
1771 			memcpy(copy_skb->data - 2, skb->data - 2, len + 2);
1772 			emac_recycle_rx_skb(dev, slot, len);
1773 			skb = copy_skb;
1774 		} else if (unlikely(emac_alloc_rx_skb(dev, slot, GFP_ATOMIC)))
1775 			goto oom;
1776 
1777 		skb_put(skb, len);
1778 	push_packet:
1779 		skb->protocol = eth_type_trans(skb, dev->ndev);
1780 		emac_rx_csum(dev, skb, ctrl);
1781 
1782 		if (unlikely(netif_receive_skb(skb) == NET_RX_DROP))
1783 			++dev->estats.rx_dropped_stack;
1784 	next:
1785 		++dev->stats.rx_packets;
1786 	skip:
1787 		dev->stats.rx_bytes += len;
1788 		slot = (slot + 1) % NUM_RX_BUFF;
1789 		--budget;
1790 		++received;
1791 		continue;
1792 	sg:
1793 		if (ctrl & MAL_RX_CTRL_FIRST) {
1794 			BUG_ON(dev->rx_sg_skb);
1795 			if (unlikely(emac_alloc_rx_skb(dev, slot, GFP_ATOMIC))) {
1796 				DBG(dev, "rx OOM %d" NL, slot);
1797 				++dev->estats.rx_dropped_oom;
1798 				emac_recycle_rx_skb(dev, slot, 0);
1799 			} else {
1800 				dev->rx_sg_skb = skb;
1801 				skb_put(skb, len);
1802 			}
1803 		} else if (!emac_rx_sg_append(dev, slot) &&
1804 			   (ctrl & MAL_RX_CTRL_LAST)) {
1805 
1806 			skb = dev->rx_sg_skb;
1807 			dev->rx_sg_skb = NULL;
1808 
1809 			ctrl &= EMAC_BAD_RX_MASK;
1810 			if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1811 				emac_parse_rx_error(dev, ctrl);
1812 				++dev->estats.rx_dropped_error;
1813 				dev_kfree_skb(skb);
1814 				len = 0;
1815 			} else
1816 				goto push_packet;
1817 		}
1818 		goto skip;
1819 	oom:
1820 		DBG(dev, "rx OOM %d" NL, slot);
1821 		/* Drop the packet and recycle skb */
1822 		++dev->estats.rx_dropped_oom;
1823 		emac_recycle_rx_skb(dev, slot, 0);
1824 		goto next;
1825 	}
1826 
1827 	if (received) {
1828 		DBG2(dev, "rx %d BDs" NL, received);
1829 		dev->rx_slot = slot;
1830 	}
1831 
1832 	if (unlikely(budget && test_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags))) {
1833 		mb();
1834 		if (!(dev->rx_desc[slot].ctrl & MAL_RX_CTRL_EMPTY)) {
1835 			DBG2(dev, "rx restart" NL);
1836 			received = 0;
1837 			goto again;
1838 		}
1839 
1840 		if (dev->rx_sg_skb) {
1841 			DBG2(dev, "dropping partial rx packet" NL);
1842 			++dev->estats.rx_dropped_error;
1843 			dev_kfree_skb(dev->rx_sg_skb);
1844 			dev->rx_sg_skb = NULL;
1845 		}
1846 
1847 		clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1848 		mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1849 		emac_rx_enable(dev);
1850 		dev->rx_slot = 0;
1851 	}
1852 	return received;
1853 }
1854 
1855 /* NAPI poll context */
1856 static int emac_peek_rx(void *param)
1857 {
1858 	struct emac_instance *dev = param;
1859 
1860 	return !(dev->rx_desc[dev->rx_slot].ctrl & MAL_RX_CTRL_EMPTY);
1861 }
1862 
1863 /* NAPI poll context */
1864 static int emac_peek_rx_sg(void *param)
1865 {
1866 	struct emac_instance *dev = param;
1867 
1868 	int slot = dev->rx_slot;
1869 	while (1) {
1870 		u16 ctrl = dev->rx_desc[slot].ctrl;
1871 		if (ctrl & MAL_RX_CTRL_EMPTY)
1872 			return 0;
1873 		else if (ctrl & MAL_RX_CTRL_LAST)
1874 			return 1;
1875 
1876 		slot = (slot + 1) % NUM_RX_BUFF;
1877 
1878 		/* I'm just being paranoid here :) */
1879 		if (unlikely(slot == dev->rx_slot))
1880 			return 0;
1881 	}
1882 }
1883 
1884 /* Hard IRQ */
1885 static void emac_rxde(void *param)
1886 {
1887 	struct emac_instance *dev = param;
1888 
1889 	++dev->estats.rx_stopped;
1890 	emac_rx_disable_async(dev);
1891 }
1892 
1893 /* Hard IRQ */
1894 static irqreturn_t emac_irq(int irq, void *dev_instance)
1895 {
1896 	struct emac_instance *dev = dev_instance;
1897 	struct emac_regs __iomem *p = dev->emacp;
1898 	struct emac_error_stats *st = &dev->estats;
1899 	u32 isr;
1900 
1901 	spin_lock(&dev->lock);
1902 
1903 	isr = in_be32(&p->isr);
1904 	out_be32(&p->isr, isr);
1905 
1906 	DBG(dev, "isr = %08x" NL, isr);
1907 
1908 	if (isr & EMAC4_ISR_TXPE)
1909 		++st->tx_parity;
1910 	if (isr & EMAC4_ISR_RXPE)
1911 		++st->rx_parity;
1912 	if (isr & EMAC4_ISR_TXUE)
1913 		++st->tx_underrun;
1914 	if (isr & EMAC4_ISR_RXOE)
1915 		++st->rx_fifo_overrun;
1916 	if (isr & EMAC_ISR_OVR)
1917 		++st->rx_overrun;
1918 	if (isr & EMAC_ISR_BP)
1919 		++st->rx_bad_packet;
1920 	if (isr & EMAC_ISR_RP)
1921 		++st->rx_runt_packet;
1922 	if (isr & EMAC_ISR_SE)
1923 		++st->rx_short_event;
1924 	if (isr & EMAC_ISR_ALE)
1925 		++st->rx_alignment_error;
1926 	if (isr & EMAC_ISR_BFCS)
1927 		++st->rx_bad_fcs;
1928 	if (isr & EMAC_ISR_PTLE)
1929 		++st->rx_packet_too_long;
1930 	if (isr & EMAC_ISR_ORE)
1931 		++st->rx_out_of_range;
1932 	if (isr & EMAC_ISR_IRE)
1933 		++st->rx_in_range;
1934 	if (isr & EMAC_ISR_SQE)
1935 		++st->tx_sqe;
1936 	if (isr & EMAC_ISR_TE)
1937 		++st->tx_errors;
1938 
1939 	spin_unlock(&dev->lock);
1940 
1941 	return IRQ_HANDLED;
1942 }
1943 
1944 static struct net_device_stats *emac_stats(struct net_device *ndev)
1945 {
1946 	struct emac_instance *dev = netdev_priv(ndev);
1947 	struct emac_stats *st = &dev->stats;
1948 	struct emac_error_stats *est = &dev->estats;
1949 	struct net_device_stats *nst = &ndev->stats;
1950 	unsigned long flags;
1951 
1952 	DBG2(dev, "stats" NL);
1953 
1954 	/* Compute "legacy" statistics */
1955 	spin_lock_irqsave(&dev->lock, flags);
1956 	nst->rx_packets = (unsigned long)st->rx_packets;
1957 	nst->rx_bytes = (unsigned long)st->rx_bytes;
1958 	nst->tx_packets = (unsigned long)st->tx_packets;
1959 	nst->tx_bytes = (unsigned long)st->tx_bytes;
1960 	nst->rx_dropped = (unsigned long)(est->rx_dropped_oom +
1961 					  est->rx_dropped_error +
1962 					  est->rx_dropped_resize +
1963 					  est->rx_dropped_mtu);
1964 	nst->tx_dropped = (unsigned long)est->tx_dropped;
1965 
1966 	nst->rx_errors = (unsigned long)est->rx_bd_errors;
1967 	nst->rx_fifo_errors = (unsigned long)(est->rx_bd_overrun +
1968 					      est->rx_fifo_overrun +
1969 					      est->rx_overrun);
1970 	nst->rx_frame_errors = (unsigned long)(est->rx_bd_alignment_error +
1971 					       est->rx_alignment_error);
1972 	nst->rx_crc_errors = (unsigned long)(est->rx_bd_bad_fcs +
1973 					     est->rx_bad_fcs);
1974 	nst->rx_length_errors = (unsigned long)(est->rx_bd_runt_packet +
1975 						est->rx_bd_short_event +
1976 						est->rx_bd_packet_too_long +
1977 						est->rx_bd_out_of_range +
1978 						est->rx_bd_in_range +
1979 						est->rx_runt_packet +
1980 						est->rx_short_event +
1981 						est->rx_packet_too_long +
1982 						est->rx_out_of_range +
1983 						est->rx_in_range);
1984 
1985 	nst->tx_errors = (unsigned long)(est->tx_bd_errors + est->tx_errors);
1986 	nst->tx_fifo_errors = (unsigned long)(est->tx_bd_underrun +
1987 					      est->tx_underrun);
1988 	nst->tx_carrier_errors = (unsigned long)est->tx_bd_carrier_loss;
1989 	nst->collisions = (unsigned long)(est->tx_bd_excessive_deferral +
1990 					  est->tx_bd_excessive_collisions +
1991 					  est->tx_bd_late_collision +
1992 					  est->tx_bd_multple_collisions);
1993 	spin_unlock_irqrestore(&dev->lock, flags);
1994 	return nst;
1995 }
1996 
1997 static struct mal_commac_ops emac_commac_ops = {
1998 	.poll_tx = &emac_poll_tx,
1999 	.poll_rx = &emac_poll_rx,
2000 	.peek_rx = &emac_peek_rx,
2001 	.rxde = &emac_rxde,
2002 };
2003 
2004 static struct mal_commac_ops emac_commac_sg_ops = {
2005 	.poll_tx = &emac_poll_tx,
2006 	.poll_rx = &emac_poll_rx,
2007 	.peek_rx = &emac_peek_rx_sg,
2008 	.rxde = &emac_rxde,
2009 };
2010 
2011 /* Ethtool support */
2012 static int emac_ethtool_get_link_ksettings(struct net_device *ndev,
2013 					   struct ethtool_link_ksettings *cmd)
2014 {
2015 	struct emac_instance *dev = netdev_priv(ndev);
2016 	u32 supported, advertising;
2017 
2018 	supported = dev->phy.features;
2019 	cmd->base.port = PORT_MII;
2020 	cmd->base.phy_address = dev->phy.address;
2021 
2022 	mutex_lock(&dev->link_lock);
2023 	advertising = dev->phy.advertising;
2024 	cmd->base.autoneg = dev->phy.autoneg;
2025 	cmd->base.speed = dev->phy.speed;
2026 	cmd->base.duplex = dev->phy.duplex;
2027 	mutex_unlock(&dev->link_lock);
2028 
2029 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
2030 						supported);
2031 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
2032 						advertising);
2033 
2034 	return 0;
2035 }
2036 
2037 static int
2038 emac_ethtool_set_link_ksettings(struct net_device *ndev,
2039 				const struct ethtool_link_ksettings *cmd)
2040 {
2041 	struct emac_instance *dev = netdev_priv(ndev);
2042 	u32 f = dev->phy.features;
2043 	u32 advertising;
2044 
2045 	ethtool_convert_link_mode_to_legacy_u32(&advertising,
2046 						cmd->link_modes.advertising);
2047 
2048 	DBG(dev, "set_settings(%d, %d, %d, 0x%08x)" NL,
2049 	    cmd->base.autoneg, cmd->base.speed, cmd->base.duplex, advertising);
2050 
2051 	/* Basic sanity checks */
2052 	if (dev->phy.address < 0)
2053 		return -EOPNOTSUPP;
2054 	if (cmd->base.autoneg != AUTONEG_ENABLE &&
2055 	    cmd->base.autoneg != AUTONEG_DISABLE)
2056 		return -EINVAL;
2057 	if (cmd->base.autoneg == AUTONEG_ENABLE && advertising == 0)
2058 		return -EINVAL;
2059 	if (cmd->base.duplex != DUPLEX_HALF && cmd->base.duplex != DUPLEX_FULL)
2060 		return -EINVAL;
2061 
2062 	if (cmd->base.autoneg == AUTONEG_DISABLE) {
2063 		switch (cmd->base.speed) {
2064 		case SPEED_10:
2065 			if (cmd->base.duplex == DUPLEX_HALF &&
2066 			    !(f & SUPPORTED_10baseT_Half))
2067 				return -EINVAL;
2068 			if (cmd->base.duplex == DUPLEX_FULL &&
2069 			    !(f & SUPPORTED_10baseT_Full))
2070 				return -EINVAL;
2071 			break;
2072 		case SPEED_100:
2073 			if (cmd->base.duplex == DUPLEX_HALF &&
2074 			    !(f & SUPPORTED_100baseT_Half))
2075 				return -EINVAL;
2076 			if (cmd->base.duplex == DUPLEX_FULL &&
2077 			    !(f & SUPPORTED_100baseT_Full))
2078 				return -EINVAL;
2079 			break;
2080 		case SPEED_1000:
2081 			if (cmd->base.duplex == DUPLEX_HALF &&
2082 			    !(f & SUPPORTED_1000baseT_Half))
2083 				return -EINVAL;
2084 			if (cmd->base.duplex == DUPLEX_FULL &&
2085 			    !(f & SUPPORTED_1000baseT_Full))
2086 				return -EINVAL;
2087 			break;
2088 		default:
2089 			return -EINVAL;
2090 		}
2091 
2092 		mutex_lock(&dev->link_lock);
2093 		dev->phy.def->ops->setup_forced(&dev->phy, cmd->base.speed,
2094 						cmd->base.duplex);
2095 		mutex_unlock(&dev->link_lock);
2096 
2097 	} else {
2098 		if (!(f & SUPPORTED_Autoneg))
2099 			return -EINVAL;
2100 
2101 		mutex_lock(&dev->link_lock);
2102 		dev->phy.def->ops->setup_aneg(&dev->phy,
2103 					      (advertising & f) |
2104 					      (dev->phy.advertising &
2105 					       (ADVERTISED_Pause |
2106 						ADVERTISED_Asym_Pause)));
2107 		mutex_unlock(&dev->link_lock);
2108 	}
2109 	emac_force_link_update(dev);
2110 
2111 	return 0;
2112 }
2113 
2114 static void emac_ethtool_get_ringparam(struct net_device *ndev,
2115 				       struct ethtool_ringparam *rp)
2116 {
2117 	rp->rx_max_pending = rp->rx_pending = NUM_RX_BUFF;
2118 	rp->tx_max_pending = rp->tx_pending = NUM_TX_BUFF;
2119 }
2120 
2121 static void emac_ethtool_get_pauseparam(struct net_device *ndev,
2122 					struct ethtool_pauseparam *pp)
2123 {
2124 	struct emac_instance *dev = netdev_priv(ndev);
2125 
2126 	mutex_lock(&dev->link_lock);
2127 	if ((dev->phy.features & SUPPORTED_Autoneg) &&
2128 	    (dev->phy.advertising & (ADVERTISED_Pause | ADVERTISED_Asym_Pause)))
2129 		pp->autoneg = 1;
2130 
2131 	if (dev->phy.duplex == DUPLEX_FULL) {
2132 		if (dev->phy.pause)
2133 			pp->rx_pause = pp->tx_pause = 1;
2134 		else if (dev->phy.asym_pause)
2135 			pp->tx_pause = 1;
2136 	}
2137 	mutex_unlock(&dev->link_lock);
2138 }
2139 
2140 static int emac_get_regs_len(struct emac_instance *dev)
2141 {
2142 		return sizeof(struct emac_ethtool_regs_subhdr) +
2143 			sizeof(struct emac_regs);
2144 }
2145 
2146 static int emac_ethtool_get_regs_len(struct net_device *ndev)
2147 {
2148 	struct emac_instance *dev = netdev_priv(ndev);
2149 	int size;
2150 
2151 	size = sizeof(struct emac_ethtool_regs_hdr) +
2152 		emac_get_regs_len(dev) + mal_get_regs_len(dev->mal);
2153 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
2154 		size += zmii_get_regs_len(dev->zmii_dev);
2155 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
2156 		size += rgmii_get_regs_len(dev->rgmii_dev);
2157 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
2158 		size += tah_get_regs_len(dev->tah_dev);
2159 
2160 	return size;
2161 }
2162 
2163 static void *emac_dump_regs(struct emac_instance *dev, void *buf)
2164 {
2165 	struct emac_ethtool_regs_subhdr *hdr = buf;
2166 
2167 	hdr->index = dev->cell_index;
2168 	if (emac_has_feature(dev, EMAC_FTR_EMAC4SYNC)) {
2169 		hdr->version = EMAC4SYNC_ETHTOOL_REGS_VER;
2170 	} else if (emac_has_feature(dev, EMAC_FTR_EMAC4)) {
2171 		hdr->version = EMAC4_ETHTOOL_REGS_VER;
2172 	} else {
2173 		hdr->version = EMAC_ETHTOOL_REGS_VER;
2174 	}
2175 	memcpy_fromio(hdr + 1, dev->emacp, sizeof(struct emac_regs));
2176 	return (void *)(hdr + 1) + sizeof(struct emac_regs);
2177 }
2178 
2179 static void emac_ethtool_get_regs(struct net_device *ndev,
2180 				  struct ethtool_regs *regs, void *buf)
2181 {
2182 	struct emac_instance *dev = netdev_priv(ndev);
2183 	struct emac_ethtool_regs_hdr *hdr = buf;
2184 
2185 	hdr->components = 0;
2186 	buf = hdr + 1;
2187 
2188 	buf = mal_dump_regs(dev->mal, buf);
2189 	buf = emac_dump_regs(dev, buf);
2190 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII)) {
2191 		hdr->components |= EMAC_ETHTOOL_REGS_ZMII;
2192 		buf = zmii_dump_regs(dev->zmii_dev, buf);
2193 	}
2194 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII)) {
2195 		hdr->components |= EMAC_ETHTOOL_REGS_RGMII;
2196 		buf = rgmii_dump_regs(dev->rgmii_dev, buf);
2197 	}
2198 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH)) {
2199 		hdr->components |= EMAC_ETHTOOL_REGS_TAH;
2200 		buf = tah_dump_regs(dev->tah_dev, buf);
2201 	}
2202 }
2203 
2204 static int emac_ethtool_nway_reset(struct net_device *ndev)
2205 {
2206 	struct emac_instance *dev = netdev_priv(ndev);
2207 	int res = 0;
2208 
2209 	DBG(dev, "nway_reset" NL);
2210 
2211 	if (dev->phy.address < 0)
2212 		return -EOPNOTSUPP;
2213 
2214 	mutex_lock(&dev->link_lock);
2215 	if (!dev->phy.autoneg) {
2216 		res = -EINVAL;
2217 		goto out;
2218 	}
2219 
2220 	dev->phy.def->ops->setup_aneg(&dev->phy, dev->phy.advertising);
2221  out:
2222 	mutex_unlock(&dev->link_lock);
2223 	emac_force_link_update(dev);
2224 	return res;
2225 }
2226 
2227 static int emac_ethtool_get_sset_count(struct net_device *ndev, int stringset)
2228 {
2229 	if (stringset == ETH_SS_STATS)
2230 		return EMAC_ETHTOOL_STATS_COUNT;
2231 	else
2232 		return -EINVAL;
2233 }
2234 
2235 static void emac_ethtool_get_strings(struct net_device *ndev, u32 stringset,
2236 				     u8 * buf)
2237 {
2238 	if (stringset == ETH_SS_STATS)
2239 		memcpy(buf, &emac_stats_keys, sizeof(emac_stats_keys));
2240 }
2241 
2242 static void emac_ethtool_get_ethtool_stats(struct net_device *ndev,
2243 					   struct ethtool_stats *estats,
2244 					   u64 * tmp_stats)
2245 {
2246 	struct emac_instance *dev = netdev_priv(ndev);
2247 
2248 	memcpy(tmp_stats, &dev->stats, sizeof(dev->stats));
2249 	tmp_stats += sizeof(dev->stats) / sizeof(u64);
2250 	memcpy(tmp_stats, &dev->estats, sizeof(dev->estats));
2251 }
2252 
2253 static void emac_ethtool_get_drvinfo(struct net_device *ndev,
2254 				     struct ethtool_drvinfo *info)
2255 {
2256 	struct emac_instance *dev = netdev_priv(ndev);
2257 
2258 	strlcpy(info->driver, "ibm_emac", sizeof(info->driver));
2259 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2260 	snprintf(info->bus_info, sizeof(info->bus_info), "PPC 4xx EMAC-%d %pOF",
2261 		 dev->cell_index, dev->ofdev->dev.of_node);
2262 }
2263 
2264 static const struct ethtool_ops emac_ethtool_ops = {
2265 	.get_drvinfo = emac_ethtool_get_drvinfo,
2266 
2267 	.get_regs_len = emac_ethtool_get_regs_len,
2268 	.get_regs = emac_ethtool_get_regs,
2269 
2270 	.nway_reset = emac_ethtool_nway_reset,
2271 
2272 	.get_ringparam = emac_ethtool_get_ringparam,
2273 	.get_pauseparam = emac_ethtool_get_pauseparam,
2274 
2275 	.get_strings = emac_ethtool_get_strings,
2276 	.get_sset_count = emac_ethtool_get_sset_count,
2277 	.get_ethtool_stats = emac_ethtool_get_ethtool_stats,
2278 
2279 	.get_link = ethtool_op_get_link,
2280 	.get_link_ksettings = emac_ethtool_get_link_ksettings,
2281 	.set_link_ksettings = emac_ethtool_set_link_ksettings,
2282 };
2283 
2284 static int emac_ioctl(struct net_device *ndev, struct ifreq *rq, int cmd)
2285 {
2286 	struct emac_instance *dev = netdev_priv(ndev);
2287 	struct mii_ioctl_data *data = if_mii(rq);
2288 
2289 	DBG(dev, "ioctl %08x" NL, cmd);
2290 
2291 	if (dev->phy.address < 0)
2292 		return -EOPNOTSUPP;
2293 
2294 	switch (cmd) {
2295 	case SIOCGMIIPHY:
2296 		data->phy_id = dev->phy.address;
2297 		/* Fall through */
2298 	case SIOCGMIIREG:
2299 		data->val_out = emac_mdio_read(ndev, dev->phy.address,
2300 					       data->reg_num);
2301 		return 0;
2302 
2303 	case SIOCSMIIREG:
2304 		emac_mdio_write(ndev, dev->phy.address, data->reg_num,
2305 				data->val_in);
2306 		return 0;
2307 	default:
2308 		return -EOPNOTSUPP;
2309 	}
2310 }
2311 
2312 struct emac_depentry {
2313 	u32			phandle;
2314 	struct device_node	*node;
2315 	struct platform_device	*ofdev;
2316 	void			*drvdata;
2317 };
2318 
2319 #define	EMAC_DEP_MAL_IDX	0
2320 #define	EMAC_DEP_ZMII_IDX	1
2321 #define	EMAC_DEP_RGMII_IDX	2
2322 #define	EMAC_DEP_TAH_IDX	3
2323 #define	EMAC_DEP_MDIO_IDX	4
2324 #define	EMAC_DEP_PREV_IDX	5
2325 #define	EMAC_DEP_COUNT		6
2326 
2327 static int emac_check_deps(struct emac_instance *dev,
2328 			   struct emac_depentry *deps)
2329 {
2330 	int i, there = 0;
2331 	struct device_node *np;
2332 
2333 	for (i = 0; i < EMAC_DEP_COUNT; i++) {
2334 		/* no dependency on that item, allright */
2335 		if (deps[i].phandle == 0) {
2336 			there++;
2337 			continue;
2338 		}
2339 		/* special case for blist as the dependency might go away */
2340 		if (i == EMAC_DEP_PREV_IDX) {
2341 			np = *(dev->blist - 1);
2342 			if (np == NULL) {
2343 				deps[i].phandle = 0;
2344 				there++;
2345 				continue;
2346 			}
2347 			if (deps[i].node == NULL)
2348 				deps[i].node = of_node_get(np);
2349 		}
2350 		if (deps[i].node == NULL)
2351 			deps[i].node = of_find_node_by_phandle(deps[i].phandle);
2352 		if (deps[i].node == NULL)
2353 			continue;
2354 		if (deps[i].ofdev == NULL)
2355 			deps[i].ofdev = of_find_device_by_node(deps[i].node);
2356 		if (deps[i].ofdev == NULL)
2357 			continue;
2358 		if (deps[i].drvdata == NULL)
2359 			deps[i].drvdata = platform_get_drvdata(deps[i].ofdev);
2360 		if (deps[i].drvdata != NULL)
2361 			there++;
2362 	}
2363 	return there == EMAC_DEP_COUNT;
2364 }
2365 
2366 static void emac_put_deps(struct emac_instance *dev)
2367 {
2368 	of_dev_put(dev->mal_dev);
2369 	of_dev_put(dev->zmii_dev);
2370 	of_dev_put(dev->rgmii_dev);
2371 	of_dev_put(dev->mdio_dev);
2372 	of_dev_put(dev->tah_dev);
2373 }
2374 
2375 static int emac_of_bus_notify(struct notifier_block *nb, unsigned long action,
2376 			      void *data)
2377 {
2378 	/* We are only intereted in device addition */
2379 	if (action == BUS_NOTIFY_BOUND_DRIVER)
2380 		wake_up_all(&emac_probe_wait);
2381 	return 0;
2382 }
2383 
2384 static struct notifier_block emac_of_bus_notifier = {
2385 	.notifier_call = emac_of_bus_notify
2386 };
2387 
2388 static int emac_wait_deps(struct emac_instance *dev)
2389 {
2390 	struct emac_depentry deps[EMAC_DEP_COUNT];
2391 	int i, err;
2392 
2393 	memset(&deps, 0, sizeof(deps));
2394 
2395 	deps[EMAC_DEP_MAL_IDX].phandle = dev->mal_ph;
2396 	deps[EMAC_DEP_ZMII_IDX].phandle = dev->zmii_ph;
2397 	deps[EMAC_DEP_RGMII_IDX].phandle = dev->rgmii_ph;
2398 	if (dev->tah_ph)
2399 		deps[EMAC_DEP_TAH_IDX].phandle = dev->tah_ph;
2400 	if (dev->mdio_ph)
2401 		deps[EMAC_DEP_MDIO_IDX].phandle = dev->mdio_ph;
2402 	if (dev->blist && dev->blist > emac_boot_list)
2403 		deps[EMAC_DEP_PREV_IDX].phandle = 0xffffffffu;
2404 	bus_register_notifier(&platform_bus_type, &emac_of_bus_notifier);
2405 	wait_event_timeout(emac_probe_wait,
2406 			   emac_check_deps(dev, deps),
2407 			   EMAC_PROBE_DEP_TIMEOUT);
2408 	bus_unregister_notifier(&platform_bus_type, &emac_of_bus_notifier);
2409 	err = emac_check_deps(dev, deps) ? 0 : -ENODEV;
2410 	for (i = 0; i < EMAC_DEP_COUNT; i++) {
2411 		of_node_put(deps[i].node);
2412 		if (err)
2413 			of_dev_put(deps[i].ofdev);
2414 	}
2415 	if (err == 0) {
2416 		dev->mal_dev = deps[EMAC_DEP_MAL_IDX].ofdev;
2417 		dev->zmii_dev = deps[EMAC_DEP_ZMII_IDX].ofdev;
2418 		dev->rgmii_dev = deps[EMAC_DEP_RGMII_IDX].ofdev;
2419 		dev->tah_dev = deps[EMAC_DEP_TAH_IDX].ofdev;
2420 		dev->mdio_dev = deps[EMAC_DEP_MDIO_IDX].ofdev;
2421 	}
2422 	of_dev_put(deps[EMAC_DEP_PREV_IDX].ofdev);
2423 	return err;
2424 }
2425 
2426 static int emac_read_uint_prop(struct device_node *np, const char *name,
2427 			       u32 *val, int fatal)
2428 {
2429 	int len;
2430 	const u32 *prop = of_get_property(np, name, &len);
2431 	if (prop == NULL || len < sizeof(u32)) {
2432 		if (fatal)
2433 			printk(KERN_ERR "%pOF: missing %s property\n",
2434 			       np, name);
2435 		return -ENODEV;
2436 	}
2437 	*val = *prop;
2438 	return 0;
2439 }
2440 
2441 static void emac_adjust_link(struct net_device *ndev)
2442 {
2443 	struct emac_instance *dev = netdev_priv(ndev);
2444 	struct phy_device *phy = dev->phy_dev;
2445 
2446 	dev->phy.autoneg = phy->autoneg;
2447 	dev->phy.speed = phy->speed;
2448 	dev->phy.duplex = phy->duplex;
2449 	dev->phy.pause = phy->pause;
2450 	dev->phy.asym_pause = phy->asym_pause;
2451 	dev->phy.advertising = phy->advertising;
2452 }
2453 
2454 static int emac_mii_bus_read(struct mii_bus *bus, int addr, int regnum)
2455 {
2456 	int ret = emac_mdio_read(bus->priv, addr, regnum);
2457 	/* This is a workaround for powered down ports/phys.
2458 	 * In the wild, this was seen on the Cisco Meraki MX60(W).
2459 	 * This hardware disables ports as part of the handoff
2460 	 * procedure. Accessing the ports will lead to errors
2461 	 * (-ETIMEDOUT, -EREMOTEIO) that do more harm than good.
2462 	 */
2463 	return ret < 0 ? 0xffff : ret;
2464 }
2465 
2466 static int emac_mii_bus_write(struct mii_bus *bus, int addr,
2467 			      int regnum, u16 val)
2468 {
2469 	emac_mdio_write(bus->priv, addr, regnum, val);
2470 	return 0;
2471 }
2472 
2473 static int emac_mii_bus_reset(struct mii_bus *bus)
2474 {
2475 	struct emac_instance *dev = netdev_priv(bus->priv);
2476 
2477 	return emac_reset(dev);
2478 }
2479 
2480 static int emac_mdio_phy_start_aneg(struct mii_phy *phy,
2481 				    struct phy_device *phy_dev)
2482 {
2483 	phy_dev->autoneg = phy->autoneg;
2484 	phy_dev->speed = phy->speed;
2485 	phy_dev->duplex = phy->duplex;
2486 	phy_dev->advertising = phy->advertising;
2487 	return phy_start_aneg(phy_dev);
2488 }
2489 
2490 static int emac_mdio_setup_aneg(struct mii_phy *phy, u32 advertise)
2491 {
2492 	struct net_device *ndev = phy->dev;
2493 	struct emac_instance *dev = netdev_priv(ndev);
2494 
2495 	phy->autoneg = AUTONEG_ENABLE;
2496 	phy->advertising = advertise;
2497 	return emac_mdio_phy_start_aneg(phy, dev->phy_dev);
2498 }
2499 
2500 static int emac_mdio_setup_forced(struct mii_phy *phy, int speed, int fd)
2501 {
2502 	struct net_device *ndev = phy->dev;
2503 	struct emac_instance *dev = netdev_priv(ndev);
2504 
2505 	phy->autoneg = AUTONEG_DISABLE;
2506 	phy->speed = speed;
2507 	phy->duplex = fd;
2508 	return emac_mdio_phy_start_aneg(phy, dev->phy_dev);
2509 }
2510 
2511 static int emac_mdio_poll_link(struct mii_phy *phy)
2512 {
2513 	struct net_device *ndev = phy->dev;
2514 	struct emac_instance *dev = netdev_priv(ndev);
2515 	int res;
2516 
2517 	res = phy_read_status(dev->phy_dev);
2518 	if (res) {
2519 		dev_err(&dev->ofdev->dev, "link update failed (%d).", res);
2520 		return ethtool_op_get_link(ndev);
2521 	}
2522 
2523 	return dev->phy_dev->link;
2524 }
2525 
2526 static int emac_mdio_read_link(struct mii_phy *phy)
2527 {
2528 	struct net_device *ndev = phy->dev;
2529 	struct emac_instance *dev = netdev_priv(ndev);
2530 	struct phy_device *phy_dev = dev->phy_dev;
2531 	int res;
2532 
2533 	res = phy_read_status(phy_dev);
2534 	if (res)
2535 		return res;
2536 
2537 	phy->speed = phy_dev->speed;
2538 	phy->duplex = phy_dev->duplex;
2539 	phy->pause = phy_dev->pause;
2540 	phy->asym_pause = phy_dev->asym_pause;
2541 	return 0;
2542 }
2543 
2544 static int emac_mdio_init_phy(struct mii_phy *phy)
2545 {
2546 	struct net_device *ndev = phy->dev;
2547 	struct emac_instance *dev = netdev_priv(ndev);
2548 
2549 	phy_start(dev->phy_dev);
2550 	return phy_init_hw(dev->phy_dev);
2551 }
2552 
2553 static const struct mii_phy_ops emac_dt_mdio_phy_ops = {
2554 	.init		= emac_mdio_init_phy,
2555 	.setup_aneg	= emac_mdio_setup_aneg,
2556 	.setup_forced	= emac_mdio_setup_forced,
2557 	.poll_link	= emac_mdio_poll_link,
2558 	.read_link	= emac_mdio_read_link,
2559 };
2560 
2561 static int emac_dt_mdio_probe(struct emac_instance *dev)
2562 {
2563 	struct device_node *mii_np;
2564 	int res;
2565 
2566 	mii_np = of_get_child_by_name(dev->ofdev->dev.of_node, "mdio");
2567 	if (!mii_np) {
2568 		dev_err(&dev->ofdev->dev, "no mdio definition found.");
2569 		return -ENODEV;
2570 	}
2571 
2572 	if (!of_device_is_available(mii_np)) {
2573 		res = -ENODEV;
2574 		goto put_node;
2575 	}
2576 
2577 	dev->mii_bus = devm_mdiobus_alloc(&dev->ofdev->dev);
2578 	if (!dev->mii_bus) {
2579 		res = -ENOMEM;
2580 		goto put_node;
2581 	}
2582 
2583 	dev->mii_bus->priv = dev->ndev;
2584 	dev->mii_bus->parent = dev->ndev->dev.parent;
2585 	dev->mii_bus->name = "emac_mdio";
2586 	dev->mii_bus->read = &emac_mii_bus_read;
2587 	dev->mii_bus->write = &emac_mii_bus_write;
2588 	dev->mii_bus->reset = &emac_mii_bus_reset;
2589 	snprintf(dev->mii_bus->id, MII_BUS_ID_SIZE, "%s", dev->ofdev->name);
2590 	res = of_mdiobus_register(dev->mii_bus, mii_np);
2591 	if (res) {
2592 		dev_err(&dev->ofdev->dev, "cannot register MDIO bus %s (%d)",
2593 			dev->mii_bus->name, res);
2594 	}
2595 
2596  put_node:
2597 	of_node_put(mii_np);
2598 	return res;
2599 }
2600 
2601 static int emac_dt_phy_connect(struct emac_instance *dev,
2602 			       struct device_node *phy_handle)
2603 {
2604 	dev->phy.def = devm_kzalloc(&dev->ofdev->dev, sizeof(*dev->phy.def),
2605 				    GFP_KERNEL);
2606 	if (!dev->phy.def)
2607 		return -ENOMEM;
2608 
2609 	dev->phy_dev = of_phy_connect(dev->ndev, phy_handle, &emac_adjust_link,
2610 				      0, dev->phy_mode);
2611 	if (!dev->phy_dev) {
2612 		dev_err(&dev->ofdev->dev, "failed to connect to PHY.\n");
2613 		return -ENODEV;
2614 	}
2615 
2616 	dev->phy.def->phy_id = dev->phy_dev->drv->phy_id;
2617 	dev->phy.def->phy_id_mask = dev->phy_dev->drv->phy_id_mask;
2618 	dev->phy.def->name = dev->phy_dev->drv->name;
2619 	dev->phy.def->ops = &emac_dt_mdio_phy_ops;
2620 	dev->phy.features = dev->phy_dev->supported;
2621 	dev->phy.address = dev->phy_dev->mdio.addr;
2622 	dev->phy.mode = dev->phy_dev->interface;
2623 	return 0;
2624 }
2625 
2626 static int emac_dt_phy_probe(struct emac_instance *dev)
2627 {
2628 	struct device_node *np = dev->ofdev->dev.of_node;
2629 	struct device_node *phy_handle;
2630 	int res = 1;
2631 
2632 	phy_handle = of_parse_phandle(np, "phy-handle", 0);
2633 
2634 	if (phy_handle) {
2635 		res = emac_dt_mdio_probe(dev);
2636 		if (!res) {
2637 			res = emac_dt_phy_connect(dev, phy_handle);
2638 			if (res)
2639 				mdiobus_unregister(dev->mii_bus);
2640 		}
2641 	}
2642 
2643 	of_node_put(phy_handle);
2644 	return res;
2645 }
2646 
2647 static int emac_init_phy(struct emac_instance *dev)
2648 {
2649 	struct device_node *np = dev->ofdev->dev.of_node;
2650 	struct net_device *ndev = dev->ndev;
2651 	u32 phy_map, adv;
2652 	int i;
2653 
2654 	dev->phy.dev = ndev;
2655 	dev->phy.mode = dev->phy_mode;
2656 
2657 	/* PHY-less configuration. */
2658 	if ((dev->phy_address == 0xffffffff && dev->phy_map == 0xffffffff) ||
2659 	    of_phy_is_fixed_link(np)) {
2660 		emac_reset(dev);
2661 
2662 		/* PHY-less configuration. */
2663 		dev->phy.address = -1;
2664 		dev->phy.features = SUPPORTED_MII;
2665 		if (emac_phy_supports_gige(dev->phy_mode))
2666 			dev->phy.features |= SUPPORTED_1000baseT_Full;
2667 		else
2668 			dev->phy.features |= SUPPORTED_100baseT_Full;
2669 		dev->phy.pause = 1;
2670 
2671 		if (of_phy_is_fixed_link(np)) {
2672 			int res = emac_dt_mdio_probe(dev);
2673 
2674 			if (!res) {
2675 				res = of_phy_register_fixed_link(np);
2676 				if (res)
2677 					mdiobus_unregister(dev->mii_bus);
2678 			}
2679 			return res;
2680 		}
2681 		return 0;
2682 	}
2683 
2684 	mutex_lock(&emac_phy_map_lock);
2685 	phy_map = dev->phy_map | busy_phy_map;
2686 
2687 	DBG(dev, "PHY maps %08x %08x" NL, dev->phy_map, busy_phy_map);
2688 
2689 	dev->phy.mdio_read = emac_mdio_read;
2690 	dev->phy.mdio_write = emac_mdio_write;
2691 
2692 	/* Enable internal clock source */
2693 #ifdef CONFIG_PPC_DCR_NATIVE
2694 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2695 		dcri_clrset(SDR0, SDR0_MFR, 0, SDR0_MFR_ECS);
2696 #endif
2697 	/* PHY clock workaround */
2698 	emac_rx_clk_tx(dev);
2699 
2700 	/* Enable internal clock source on 440GX*/
2701 #ifdef CONFIG_PPC_DCR_NATIVE
2702 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2703 		dcri_clrset(SDR0, SDR0_MFR, 0, SDR0_MFR_ECS);
2704 #endif
2705 	/* Configure EMAC with defaults so we can at least use MDIO
2706 	 * This is needed mostly for 440GX
2707 	 */
2708 	if (emac_phy_gpcs(dev->phy.mode)) {
2709 		/* XXX
2710 		 * Make GPCS PHY address equal to EMAC index.
2711 		 * We probably should take into account busy_phy_map
2712 		 * and/or phy_map here.
2713 		 *
2714 		 * Note that the busy_phy_map is currently global
2715 		 * while it should probably be per-ASIC...
2716 		 */
2717 		dev->phy.gpcs_address = dev->gpcs_address;
2718 		if (dev->phy.gpcs_address == 0xffffffff)
2719 			dev->phy.address = dev->cell_index;
2720 	}
2721 
2722 	emac_configure(dev);
2723 
2724 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII)) {
2725 		int res = emac_dt_phy_probe(dev);
2726 
2727 		switch (res) {
2728 		case 1:
2729 			/* No phy-handle property configured.
2730 			 * Continue with the existing phy probe
2731 			 * and setup code.
2732 			 */
2733 			break;
2734 
2735 		case 0:
2736 			mutex_unlock(&emac_phy_map_lock);
2737 			goto init_phy;
2738 
2739 		default:
2740 			mutex_unlock(&emac_phy_map_lock);
2741 			dev_err(&dev->ofdev->dev, "failed to attach dt phy (%d).\n",
2742 				res);
2743 			return res;
2744 		}
2745 	}
2746 
2747 	if (dev->phy_address != 0xffffffff)
2748 		phy_map = ~(1 << dev->phy_address);
2749 
2750 	for (i = 0; i < 0x20; phy_map >>= 1, ++i)
2751 		if (!(phy_map & 1)) {
2752 			int r;
2753 			busy_phy_map |= 1 << i;
2754 
2755 			/* Quick check if there is a PHY at the address */
2756 			r = emac_mdio_read(dev->ndev, i, MII_BMCR);
2757 			if (r == 0xffff || r < 0)
2758 				continue;
2759 			if (!emac_mii_phy_probe(&dev->phy, i))
2760 				break;
2761 		}
2762 
2763 	/* Enable external clock source */
2764 #ifdef CONFIG_PPC_DCR_NATIVE
2765 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2766 		dcri_clrset(SDR0, SDR0_MFR, SDR0_MFR_ECS, 0);
2767 #endif
2768 	mutex_unlock(&emac_phy_map_lock);
2769 	if (i == 0x20) {
2770 		printk(KERN_WARNING "%pOF: can't find PHY!\n", np);
2771 		return -ENXIO;
2772 	}
2773 
2774  init_phy:
2775 	/* Init PHY */
2776 	if (dev->phy.def->ops->init)
2777 		dev->phy.def->ops->init(&dev->phy);
2778 
2779 	/* Disable any PHY features not supported by the platform */
2780 	dev->phy.def->features &= ~dev->phy_feat_exc;
2781 	dev->phy.features &= ~dev->phy_feat_exc;
2782 
2783 	/* Setup initial link parameters */
2784 	if (dev->phy.features & SUPPORTED_Autoneg) {
2785 		adv = dev->phy.features;
2786 		if (!emac_has_feature(dev, EMAC_FTR_NO_FLOW_CONTROL_40x))
2787 			adv |= ADVERTISED_Pause | ADVERTISED_Asym_Pause;
2788 		/* Restart autonegotiation */
2789 		dev->phy.def->ops->setup_aneg(&dev->phy, adv);
2790 	} else {
2791 		u32 f = dev->phy.def->features;
2792 		int speed = SPEED_10, fd = DUPLEX_HALF;
2793 
2794 		/* Select highest supported speed/duplex */
2795 		if (f & SUPPORTED_1000baseT_Full) {
2796 			speed = SPEED_1000;
2797 			fd = DUPLEX_FULL;
2798 		} else if (f & SUPPORTED_1000baseT_Half)
2799 			speed = SPEED_1000;
2800 		else if (f & SUPPORTED_100baseT_Full) {
2801 			speed = SPEED_100;
2802 			fd = DUPLEX_FULL;
2803 		} else if (f & SUPPORTED_100baseT_Half)
2804 			speed = SPEED_100;
2805 		else if (f & SUPPORTED_10baseT_Full)
2806 			fd = DUPLEX_FULL;
2807 
2808 		/* Force link parameters */
2809 		dev->phy.def->ops->setup_forced(&dev->phy, speed, fd);
2810 	}
2811 	return 0;
2812 }
2813 
2814 static int emac_init_config(struct emac_instance *dev)
2815 {
2816 	struct device_node *np = dev->ofdev->dev.of_node;
2817 	const void *p;
2818 
2819 	/* Read config from device-tree */
2820 	if (emac_read_uint_prop(np, "mal-device", &dev->mal_ph, 1))
2821 		return -ENXIO;
2822 	if (emac_read_uint_prop(np, "mal-tx-channel", &dev->mal_tx_chan, 1))
2823 		return -ENXIO;
2824 	if (emac_read_uint_prop(np, "mal-rx-channel", &dev->mal_rx_chan, 1))
2825 		return -ENXIO;
2826 	if (emac_read_uint_prop(np, "cell-index", &dev->cell_index, 1))
2827 		return -ENXIO;
2828 	if (emac_read_uint_prop(np, "max-frame-size", &dev->max_mtu, 0))
2829 		dev->max_mtu = ETH_DATA_LEN;
2830 	if (emac_read_uint_prop(np, "rx-fifo-size", &dev->rx_fifo_size, 0))
2831 		dev->rx_fifo_size = 2048;
2832 	if (emac_read_uint_prop(np, "tx-fifo-size", &dev->tx_fifo_size, 0))
2833 		dev->tx_fifo_size = 2048;
2834 	if (emac_read_uint_prop(np, "rx-fifo-size-gige", &dev->rx_fifo_size_gige, 0))
2835 		dev->rx_fifo_size_gige = dev->rx_fifo_size;
2836 	if (emac_read_uint_prop(np, "tx-fifo-size-gige", &dev->tx_fifo_size_gige, 0))
2837 		dev->tx_fifo_size_gige = dev->tx_fifo_size;
2838 	if (emac_read_uint_prop(np, "phy-address", &dev->phy_address, 0))
2839 		dev->phy_address = 0xffffffff;
2840 	if (emac_read_uint_prop(np, "phy-map", &dev->phy_map, 0))
2841 		dev->phy_map = 0xffffffff;
2842 	if (emac_read_uint_prop(np, "gpcs-address", &dev->gpcs_address, 0))
2843 		dev->gpcs_address = 0xffffffff;
2844 	if (emac_read_uint_prop(np->parent, "clock-frequency", &dev->opb_bus_freq, 1))
2845 		return -ENXIO;
2846 	if (emac_read_uint_prop(np, "tah-device", &dev->tah_ph, 0))
2847 		dev->tah_ph = 0;
2848 	if (emac_read_uint_prop(np, "tah-channel", &dev->tah_port, 0))
2849 		dev->tah_port = 0;
2850 	if (emac_read_uint_prop(np, "mdio-device", &dev->mdio_ph, 0))
2851 		dev->mdio_ph = 0;
2852 	if (emac_read_uint_prop(np, "zmii-device", &dev->zmii_ph, 0))
2853 		dev->zmii_ph = 0;
2854 	if (emac_read_uint_prop(np, "zmii-channel", &dev->zmii_port, 0))
2855 		dev->zmii_port = 0xffffffff;
2856 	if (emac_read_uint_prop(np, "rgmii-device", &dev->rgmii_ph, 0))
2857 		dev->rgmii_ph = 0;
2858 	if (emac_read_uint_prop(np, "rgmii-channel", &dev->rgmii_port, 0))
2859 		dev->rgmii_port = 0xffffffff;
2860 	if (emac_read_uint_prop(np, "fifo-entry-size", &dev->fifo_entry_size, 0))
2861 		dev->fifo_entry_size = 16;
2862 	if (emac_read_uint_prop(np, "mal-burst-size", &dev->mal_burst_size, 0))
2863 		dev->mal_burst_size = 256;
2864 
2865 	/* PHY mode needs some decoding */
2866 	dev->phy_mode = of_get_phy_mode(np);
2867 	if (dev->phy_mode < 0)
2868 		dev->phy_mode = PHY_MODE_NA;
2869 
2870 	/* Check EMAC version */
2871 	if (of_device_is_compatible(np, "ibm,emac4sync")) {
2872 		dev->features |= (EMAC_FTR_EMAC4 | EMAC_FTR_EMAC4SYNC);
2873 		if (of_device_is_compatible(np, "ibm,emac-460ex") ||
2874 		    of_device_is_compatible(np, "ibm,emac-460gt"))
2875 			dev->features |= EMAC_FTR_460EX_PHY_CLK_FIX;
2876 		if (of_device_is_compatible(np, "ibm,emac-405ex") ||
2877 		    of_device_is_compatible(np, "ibm,emac-405exr"))
2878 			dev->features |= EMAC_FTR_440EP_PHY_CLK_FIX;
2879 		if (of_device_is_compatible(np, "ibm,emac-apm821xx")) {
2880 			dev->features |= (EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE |
2881 					  EMAC_FTR_APM821XX_NO_HALF_DUPLEX |
2882 					  EMAC_FTR_460EX_PHY_CLK_FIX);
2883 		}
2884 	} else if (of_device_is_compatible(np, "ibm,emac4")) {
2885 		dev->features |= EMAC_FTR_EMAC4;
2886 		if (of_device_is_compatible(np, "ibm,emac-440gx"))
2887 			dev->features |= EMAC_FTR_440GX_PHY_CLK_FIX;
2888 	} else {
2889 		if (of_device_is_compatible(np, "ibm,emac-440ep") ||
2890 		    of_device_is_compatible(np, "ibm,emac-440gr"))
2891 			dev->features |= EMAC_FTR_440EP_PHY_CLK_FIX;
2892 		if (of_device_is_compatible(np, "ibm,emac-405ez")) {
2893 #ifdef CONFIG_IBM_EMAC_NO_FLOW_CTRL
2894 			dev->features |= EMAC_FTR_NO_FLOW_CONTROL_40x;
2895 #else
2896 			printk(KERN_ERR "%pOF: Flow control not disabled!\n",
2897 					np);
2898 			return -ENXIO;
2899 #endif
2900 		}
2901 
2902 	}
2903 
2904 	/* Fixup some feature bits based on the device tree */
2905 	if (of_get_property(np, "has-inverted-stacr-oc", NULL))
2906 		dev->features |= EMAC_FTR_STACR_OC_INVERT;
2907 	if (of_get_property(np, "has-new-stacr-staopc", NULL))
2908 		dev->features |= EMAC_FTR_HAS_NEW_STACR;
2909 
2910 	/* CAB lacks the appropriate properties */
2911 	if (of_device_is_compatible(np, "ibm,emac-axon"))
2912 		dev->features |= EMAC_FTR_HAS_NEW_STACR |
2913 			EMAC_FTR_STACR_OC_INVERT;
2914 
2915 	/* Enable TAH/ZMII/RGMII features as found */
2916 	if (dev->tah_ph != 0) {
2917 #ifdef CONFIG_IBM_EMAC_TAH
2918 		dev->features |= EMAC_FTR_HAS_TAH;
2919 #else
2920 		printk(KERN_ERR "%pOF: TAH support not enabled !\n", np);
2921 		return -ENXIO;
2922 #endif
2923 	}
2924 
2925 	if (dev->zmii_ph != 0) {
2926 #ifdef CONFIG_IBM_EMAC_ZMII
2927 		dev->features |= EMAC_FTR_HAS_ZMII;
2928 #else
2929 		printk(KERN_ERR "%pOF: ZMII support not enabled !\n", np);
2930 		return -ENXIO;
2931 #endif
2932 	}
2933 
2934 	if (dev->rgmii_ph != 0) {
2935 #ifdef CONFIG_IBM_EMAC_RGMII
2936 		dev->features |= EMAC_FTR_HAS_RGMII;
2937 #else
2938 		printk(KERN_ERR "%pOF: RGMII support not enabled !\n", np);
2939 		return -ENXIO;
2940 #endif
2941 	}
2942 
2943 	/* Read MAC-address */
2944 	p = of_get_property(np, "local-mac-address", NULL);
2945 	if (p == NULL) {
2946 		printk(KERN_ERR "%pOF: Can't find local-mac-address property\n",
2947 		       np);
2948 		return -ENXIO;
2949 	}
2950 	memcpy(dev->ndev->dev_addr, p, ETH_ALEN);
2951 
2952 	/* IAHT and GAHT filter parameterization */
2953 	if (emac_has_feature(dev, EMAC_FTR_EMAC4SYNC)) {
2954 		dev->xaht_slots_shift = EMAC4SYNC_XAHT_SLOTS_SHIFT;
2955 		dev->xaht_width_shift = EMAC4SYNC_XAHT_WIDTH_SHIFT;
2956 	} else {
2957 		dev->xaht_slots_shift = EMAC4_XAHT_SLOTS_SHIFT;
2958 		dev->xaht_width_shift = EMAC4_XAHT_WIDTH_SHIFT;
2959 	}
2960 
2961 	DBG(dev, "features     : 0x%08x / 0x%08x\n", dev->features, EMAC_FTRS_POSSIBLE);
2962 	DBG(dev, "tx_fifo_size : %d (%d gige)\n", dev->tx_fifo_size, dev->tx_fifo_size_gige);
2963 	DBG(dev, "rx_fifo_size : %d (%d gige)\n", dev->rx_fifo_size, dev->rx_fifo_size_gige);
2964 	DBG(dev, "max_mtu      : %d\n", dev->max_mtu);
2965 	DBG(dev, "OPB freq     : %d\n", dev->opb_bus_freq);
2966 
2967 	return 0;
2968 }
2969 
2970 static const struct net_device_ops emac_netdev_ops = {
2971 	.ndo_open		= emac_open,
2972 	.ndo_stop		= emac_close,
2973 	.ndo_get_stats		= emac_stats,
2974 	.ndo_set_rx_mode	= emac_set_multicast_list,
2975 	.ndo_do_ioctl		= emac_ioctl,
2976 	.ndo_tx_timeout		= emac_tx_timeout,
2977 	.ndo_validate_addr	= eth_validate_addr,
2978 	.ndo_set_mac_address	= emac_set_mac_address,
2979 	.ndo_start_xmit		= emac_start_xmit,
2980 };
2981 
2982 static const struct net_device_ops emac_gige_netdev_ops = {
2983 	.ndo_open		= emac_open,
2984 	.ndo_stop		= emac_close,
2985 	.ndo_get_stats		= emac_stats,
2986 	.ndo_set_rx_mode	= emac_set_multicast_list,
2987 	.ndo_do_ioctl		= emac_ioctl,
2988 	.ndo_tx_timeout		= emac_tx_timeout,
2989 	.ndo_validate_addr	= eth_validate_addr,
2990 	.ndo_set_mac_address	= emac_set_mac_address,
2991 	.ndo_start_xmit		= emac_start_xmit_sg,
2992 	.ndo_change_mtu		= emac_change_mtu,
2993 };
2994 
2995 static int emac_probe(struct platform_device *ofdev)
2996 {
2997 	struct net_device *ndev;
2998 	struct emac_instance *dev;
2999 	struct device_node *np = ofdev->dev.of_node;
3000 	struct device_node **blist = NULL;
3001 	int err, i;
3002 
3003 	/* Skip unused/unwired EMACS.  We leave the check for an unused
3004 	 * property here for now, but new flat device trees should set a
3005 	 * status property to "disabled" instead.
3006 	 */
3007 	if (of_get_property(np, "unused", NULL) || !of_device_is_available(np))
3008 		return -ENODEV;
3009 
3010 	/* Find ourselves in the bootlist if we are there */
3011 	for (i = 0; i < EMAC_BOOT_LIST_SIZE; i++)
3012 		if (emac_boot_list[i] == np)
3013 			blist = &emac_boot_list[i];
3014 
3015 	/* Allocate our net_device structure */
3016 	err = -ENOMEM;
3017 	ndev = alloc_etherdev(sizeof(struct emac_instance));
3018 	if (!ndev)
3019 		goto err_gone;
3020 
3021 	dev = netdev_priv(ndev);
3022 	dev->ndev = ndev;
3023 	dev->ofdev = ofdev;
3024 	dev->blist = blist;
3025 	SET_NETDEV_DEV(ndev, &ofdev->dev);
3026 
3027 	/* Initialize some embedded data structures */
3028 	mutex_init(&dev->mdio_lock);
3029 	mutex_init(&dev->link_lock);
3030 	spin_lock_init(&dev->lock);
3031 	INIT_WORK(&dev->reset_work, emac_reset_work);
3032 
3033 	/* Init various config data based on device-tree */
3034 	err = emac_init_config(dev);
3035 	if (err)
3036 		goto err_free;
3037 
3038 	/* Get interrupts. EMAC irq is mandatory, WOL irq is optional */
3039 	dev->emac_irq = irq_of_parse_and_map(np, 0);
3040 	dev->wol_irq = irq_of_parse_and_map(np, 1);
3041 	if (!dev->emac_irq) {
3042 		printk(KERN_ERR "%pOF: Can't map main interrupt\n", np);
3043 		err = -ENODEV;
3044 		goto err_free;
3045 	}
3046 	ndev->irq = dev->emac_irq;
3047 
3048 	/* Map EMAC regs */
3049 	// TODO : platform_get_resource() and devm_ioremap_resource()
3050 	dev->emacp = of_iomap(np, 0);
3051 	if (dev->emacp == NULL) {
3052 		printk(KERN_ERR "%pOF: Can't map device registers!\n", np);
3053 		err = -ENOMEM;
3054 		goto err_irq_unmap;
3055 	}
3056 
3057 	/* Wait for dependent devices */
3058 	err = emac_wait_deps(dev);
3059 	if (err) {
3060 		printk(KERN_ERR
3061 		       "%pOF: Timeout waiting for dependent devices\n", np);
3062 		/*  display more info about what's missing ? */
3063 		goto err_reg_unmap;
3064 	}
3065 	dev->mal = platform_get_drvdata(dev->mal_dev);
3066 	if (dev->mdio_dev != NULL)
3067 		dev->mdio_instance = platform_get_drvdata(dev->mdio_dev);
3068 
3069 	/* Register with MAL */
3070 	dev->commac.ops = &emac_commac_ops;
3071 	dev->commac.dev = dev;
3072 	dev->commac.tx_chan_mask = MAL_CHAN_MASK(dev->mal_tx_chan);
3073 	dev->commac.rx_chan_mask = MAL_CHAN_MASK(dev->mal_rx_chan);
3074 	err = mal_register_commac(dev->mal, &dev->commac);
3075 	if (err) {
3076 		printk(KERN_ERR "%pOF: failed to register with mal %pOF!\n",
3077 		       np, dev->mal_dev->dev.of_node);
3078 		goto err_rel_deps;
3079 	}
3080 	dev->rx_skb_size = emac_rx_skb_size(ndev->mtu);
3081 	dev->rx_sync_size = emac_rx_sync_size(ndev->mtu);
3082 
3083 	/* Get pointers to BD rings */
3084 	dev->tx_desc =
3085 	    dev->mal->bd_virt + mal_tx_bd_offset(dev->mal, dev->mal_tx_chan);
3086 	dev->rx_desc =
3087 	    dev->mal->bd_virt + mal_rx_bd_offset(dev->mal, dev->mal_rx_chan);
3088 
3089 	DBG(dev, "tx_desc %p" NL, dev->tx_desc);
3090 	DBG(dev, "rx_desc %p" NL, dev->rx_desc);
3091 
3092 	/* Clean rings */
3093 	memset(dev->tx_desc, 0, NUM_TX_BUFF * sizeof(struct mal_descriptor));
3094 	memset(dev->rx_desc, 0, NUM_RX_BUFF * sizeof(struct mal_descriptor));
3095 	memset(dev->tx_skb, 0, NUM_TX_BUFF * sizeof(struct sk_buff *));
3096 	memset(dev->rx_skb, 0, NUM_RX_BUFF * sizeof(struct sk_buff *));
3097 
3098 	/* Attach to ZMII, if needed */
3099 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII) &&
3100 	    (err = zmii_attach(dev->zmii_dev, dev->zmii_port, &dev->phy_mode)) != 0)
3101 		goto err_unreg_commac;
3102 
3103 	/* Attach to RGMII, if needed */
3104 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII) &&
3105 	    (err = rgmii_attach(dev->rgmii_dev, dev->rgmii_port, dev->phy_mode)) != 0)
3106 		goto err_detach_zmii;
3107 
3108 	/* Attach to TAH, if needed */
3109 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH) &&
3110 	    (err = tah_attach(dev->tah_dev, dev->tah_port)) != 0)
3111 		goto err_detach_rgmii;
3112 
3113 	/* Set some link defaults before we can find out real parameters */
3114 	dev->phy.speed = SPEED_100;
3115 	dev->phy.duplex = DUPLEX_FULL;
3116 	dev->phy.autoneg = AUTONEG_DISABLE;
3117 	dev->phy.pause = dev->phy.asym_pause = 0;
3118 	dev->stop_timeout = STOP_TIMEOUT_100;
3119 	INIT_DELAYED_WORK(&dev->link_work, emac_link_timer);
3120 
3121 	/* Some SoCs like APM821xx does not support Half Duplex mode. */
3122 	if (emac_has_feature(dev, EMAC_FTR_APM821XX_NO_HALF_DUPLEX)) {
3123 		dev->phy_feat_exc = (SUPPORTED_1000baseT_Half |
3124 				     SUPPORTED_100baseT_Half |
3125 				     SUPPORTED_10baseT_Half);
3126 	}
3127 
3128 	/* Find PHY if any */
3129 	err = emac_init_phy(dev);
3130 	if (err != 0)
3131 		goto err_detach_tah;
3132 
3133 	if (dev->tah_dev) {
3134 		ndev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG;
3135 		ndev->features |= ndev->hw_features | NETIF_F_RXCSUM;
3136 	}
3137 	ndev->watchdog_timeo = 5 * HZ;
3138 	if (emac_phy_supports_gige(dev->phy_mode)) {
3139 		ndev->netdev_ops = &emac_gige_netdev_ops;
3140 		dev->commac.ops = &emac_commac_sg_ops;
3141 	} else
3142 		ndev->netdev_ops = &emac_netdev_ops;
3143 	ndev->ethtool_ops = &emac_ethtool_ops;
3144 
3145 	/* MTU range: 46 - 1500 or whatever is in OF */
3146 	ndev->min_mtu = EMAC_MIN_MTU;
3147 	ndev->max_mtu = dev->max_mtu;
3148 
3149 	netif_carrier_off(ndev);
3150 
3151 	err = register_netdev(ndev);
3152 	if (err) {
3153 		printk(KERN_ERR "%pOF: failed to register net device (%d)!\n",
3154 		       np, err);
3155 		goto err_detach_tah;
3156 	}
3157 
3158 	/* Set our drvdata last as we don't want them visible until we are
3159 	 * fully initialized
3160 	 */
3161 	wmb();
3162 	platform_set_drvdata(ofdev, dev);
3163 
3164 	/* There's a new kid in town ! Let's tell everybody */
3165 	wake_up_all(&emac_probe_wait);
3166 
3167 
3168 	printk(KERN_INFO "%s: EMAC-%d %pOF, MAC %pM\n",
3169 	       ndev->name, dev->cell_index, np, ndev->dev_addr);
3170 
3171 	if (dev->phy_mode == PHY_MODE_SGMII)
3172 		printk(KERN_NOTICE "%s: in SGMII mode\n", ndev->name);
3173 
3174 	if (dev->phy.address >= 0)
3175 		printk("%s: found %s PHY (0x%02x)\n", ndev->name,
3176 		       dev->phy.def->name, dev->phy.address);
3177 
3178 	/* Life is good */
3179 	return 0;
3180 
3181 	/* I have a bad feeling about this ... */
3182 
3183  err_detach_tah:
3184 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
3185 		tah_detach(dev->tah_dev, dev->tah_port);
3186  err_detach_rgmii:
3187 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
3188 		rgmii_detach(dev->rgmii_dev, dev->rgmii_port);
3189  err_detach_zmii:
3190 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
3191 		zmii_detach(dev->zmii_dev, dev->zmii_port);
3192  err_unreg_commac:
3193 	mal_unregister_commac(dev->mal, &dev->commac);
3194  err_rel_deps:
3195 	emac_put_deps(dev);
3196  err_reg_unmap:
3197 	iounmap(dev->emacp);
3198  err_irq_unmap:
3199 	if (dev->wol_irq)
3200 		irq_dispose_mapping(dev->wol_irq);
3201 	if (dev->emac_irq)
3202 		irq_dispose_mapping(dev->emac_irq);
3203  err_free:
3204 	free_netdev(ndev);
3205  err_gone:
3206 	/* if we were on the bootlist, remove us as we won't show up and
3207 	 * wake up all waiters to notify them in case they were waiting
3208 	 * on us
3209 	 */
3210 	if (blist) {
3211 		*blist = NULL;
3212 		wake_up_all(&emac_probe_wait);
3213 	}
3214 	return err;
3215 }
3216 
3217 static int emac_remove(struct platform_device *ofdev)
3218 {
3219 	struct emac_instance *dev = platform_get_drvdata(ofdev);
3220 
3221 	DBG(dev, "remove" NL);
3222 
3223 	unregister_netdev(dev->ndev);
3224 
3225 	cancel_work_sync(&dev->reset_work);
3226 
3227 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
3228 		tah_detach(dev->tah_dev, dev->tah_port);
3229 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
3230 		rgmii_detach(dev->rgmii_dev, dev->rgmii_port);
3231 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
3232 		zmii_detach(dev->zmii_dev, dev->zmii_port);
3233 
3234 	if (dev->phy_dev)
3235 		phy_disconnect(dev->phy_dev);
3236 
3237 	if (dev->mii_bus)
3238 		mdiobus_unregister(dev->mii_bus);
3239 
3240 	busy_phy_map &= ~(1 << dev->phy.address);
3241 	DBG(dev, "busy_phy_map now %#x" NL, busy_phy_map);
3242 
3243 	mal_unregister_commac(dev->mal, &dev->commac);
3244 	emac_put_deps(dev);
3245 
3246 	iounmap(dev->emacp);
3247 
3248 	if (dev->wol_irq)
3249 		irq_dispose_mapping(dev->wol_irq);
3250 	if (dev->emac_irq)
3251 		irq_dispose_mapping(dev->emac_irq);
3252 
3253 	free_netdev(dev->ndev);
3254 
3255 	return 0;
3256 }
3257 
3258 /* XXX Features in here should be replaced by properties... */
3259 static const struct of_device_id emac_match[] =
3260 {
3261 	{
3262 		.type		= "network",
3263 		.compatible	= "ibm,emac",
3264 	},
3265 	{
3266 		.type		= "network",
3267 		.compatible	= "ibm,emac4",
3268 	},
3269 	{
3270 		.type		= "network",
3271 		.compatible	= "ibm,emac4sync",
3272 	},
3273 	{},
3274 };
3275 MODULE_DEVICE_TABLE(of, emac_match);
3276 
3277 static struct platform_driver emac_driver = {
3278 	.driver = {
3279 		.name = "emac",
3280 		.of_match_table = emac_match,
3281 	},
3282 	.probe = emac_probe,
3283 	.remove = emac_remove,
3284 };
3285 
3286 static void __init emac_make_bootlist(void)
3287 {
3288 	struct device_node *np = NULL;
3289 	int j, max, i = 0;
3290 	int cell_indices[EMAC_BOOT_LIST_SIZE];
3291 
3292 	/* Collect EMACs */
3293 	while((np = of_find_all_nodes(np)) != NULL) {
3294 		const u32 *idx;
3295 
3296 		if (of_match_node(emac_match, np) == NULL)
3297 			continue;
3298 		if (of_get_property(np, "unused", NULL))
3299 			continue;
3300 		idx = of_get_property(np, "cell-index", NULL);
3301 		if (idx == NULL)
3302 			continue;
3303 		cell_indices[i] = *idx;
3304 		emac_boot_list[i++] = of_node_get(np);
3305 		if (i >= EMAC_BOOT_LIST_SIZE) {
3306 			of_node_put(np);
3307 			break;
3308 		}
3309 	}
3310 	max = i;
3311 
3312 	/* Bubble sort them (doh, what a creative algorithm :-) */
3313 	for (i = 0; max > 1 && (i < (max - 1)); i++)
3314 		for (j = i; j < max; j++) {
3315 			if (cell_indices[i] > cell_indices[j]) {
3316 				swap(emac_boot_list[i], emac_boot_list[j]);
3317 				swap(cell_indices[i], cell_indices[j]);
3318 			}
3319 		}
3320 }
3321 
3322 static int __init emac_init(void)
3323 {
3324 	int rc;
3325 
3326 	printk(KERN_INFO DRV_DESC ", version " DRV_VERSION "\n");
3327 
3328 	/* Build EMAC boot list */
3329 	emac_make_bootlist();
3330 
3331 	/* Init submodules */
3332 	rc = mal_init();
3333 	if (rc)
3334 		goto err;
3335 	rc = zmii_init();
3336 	if (rc)
3337 		goto err_mal;
3338 	rc = rgmii_init();
3339 	if (rc)
3340 		goto err_zmii;
3341 	rc = tah_init();
3342 	if (rc)
3343 		goto err_rgmii;
3344 	rc = platform_driver_register(&emac_driver);
3345 	if (rc)
3346 		goto err_tah;
3347 
3348 	return 0;
3349 
3350  err_tah:
3351 	tah_exit();
3352  err_rgmii:
3353 	rgmii_exit();
3354  err_zmii:
3355 	zmii_exit();
3356  err_mal:
3357 	mal_exit();
3358  err:
3359 	return rc;
3360 }
3361 
3362 static void __exit emac_exit(void)
3363 {
3364 	int i;
3365 
3366 	platform_driver_unregister(&emac_driver);
3367 
3368 	tah_exit();
3369 	rgmii_exit();
3370 	zmii_exit();
3371 	mal_exit();
3372 
3373 	/* Destroy EMAC boot list */
3374 	for (i = 0; i < EMAC_BOOT_LIST_SIZE; i++)
3375 		of_node_put(emac_boot_list[i]);
3376 }
3377 
3378 module_init(emac_init);
3379 module_exit(emac_exit);
3380