xref: /linux/drivers/net/ethernet/ibm/emac/core.c (revision 836265d31631e28000fc8917ce697fc687a58724)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * drivers/net/ethernet/ibm/emac/core.c
4  *
5  * Driver for PowerPC 4xx on-chip ethernet controller.
6  *
7  * Copyright 2007 Benjamin Herrenschmidt, IBM Corp.
8  *                <benh@kernel.crashing.org>
9  *
10  * Based on the arch/ppc version of the driver:
11  *
12  * Copyright (c) 2004, 2005 Zultys Technologies.
13  * Eugene Surovegin <eugene.surovegin@zultys.com> or <ebs@ebshome.net>
14  *
15  * Based on original work by
16  * 	Matt Porter <mporter@kernel.crashing.org>
17  *	(c) 2003 Benjamin Herrenschmidt <benh@kernel.crashing.org>
18  *      Armin Kuster <akuster@mvista.com>
19  * 	Johnnie Peters <jpeters@mvista.com>
20  */
21 
22 #include <linux/module.h>
23 #include <linux/sched.h>
24 #include <linux/string.h>
25 #include <linux/errno.h>
26 #include <linux/delay.h>
27 #include <linux/types.h>
28 #include <linux/pci.h>
29 #include <linux/etherdevice.h>
30 #include <linux/skbuff.h>
31 #include <linux/crc32.h>
32 #include <linux/ethtool.h>
33 #include <linux/mii.h>
34 #include <linux/bitops.h>
35 #include <linux/of.h>
36 #include <linux/of_address.h>
37 #include <linux/of_irq.h>
38 #include <linux/of_net.h>
39 #include <linux/of_mdio.h>
40 #include <linux/of_platform.h>
41 #include <linux/platform_device.h>
42 #include <linux/slab.h>
43 
44 #include <asm/processor.h>
45 #include <asm/io.h>
46 #include <asm/dma.h>
47 #include <linux/uaccess.h>
48 #include <asm/dcr.h>
49 #include <asm/dcr-regs.h>
50 
51 #include "core.h"
52 
53 /*
54  * Lack of dma_unmap_???? calls is intentional.
55  *
56  * API-correct usage requires additional support state information to be
57  * maintained for every RX and TX buffer descriptor (BD). Unfortunately, due to
58  * EMAC design (e.g. TX buffer passed from network stack can be split into
59  * several BDs, dma_map_single/dma_map_page can be used to map particular BD),
60  * maintaining such information will add additional overhead.
61  * Current DMA API implementation for 4xx processors only ensures cache coherency
62  * and dma_unmap_???? routines are empty and are likely to stay this way.
63  * I decided to omit dma_unmap_??? calls because I don't want to add additional
64  * complexity just for the sake of following some abstract API, when it doesn't
65  * add any real benefit to the driver. I understand that this decision maybe
66  * controversial, but I really tried to make code API-correct and efficient
67  * at the same time and didn't come up with code I liked :(.                --ebs
68  */
69 
70 #define DRV_NAME        "emac"
71 #define DRV_VERSION     "3.54"
72 #define DRV_DESC        "PPC 4xx OCP EMAC driver"
73 
74 MODULE_DESCRIPTION(DRV_DESC);
75 MODULE_AUTHOR
76     ("Eugene Surovegin <eugene.surovegin@zultys.com> or <ebs@ebshome.net>");
77 MODULE_LICENSE("GPL");
78 
79 /* minimum number of free TX descriptors required to wake up TX process */
80 #define EMAC_TX_WAKEUP_THRESH		(NUM_TX_BUFF / 4)
81 
82 /* If packet size is less than this number, we allocate small skb and copy packet
83  * contents into it instead of just sending original big skb up
84  */
85 #define EMAC_RX_COPY_THRESH		CONFIG_IBM_EMAC_RX_COPY_THRESHOLD
86 
87 /* Since multiple EMACs share MDIO lines in various ways, we need
88  * to avoid re-using the same PHY ID in cases where the arch didn't
89  * setup precise phy_map entries
90  *
91  * XXX This is something that needs to be reworked as we can have multiple
92  * EMAC "sets" (multiple ASICs containing several EMACs) though we can
93  * probably require in that case to have explicit PHY IDs in the device-tree
94  */
95 static u32 busy_phy_map;
96 static DEFINE_MUTEX(emac_phy_map_lock);
97 
98 /* Having stable interface names is a doomed idea. However, it would be nice
99  * if we didn't have completely random interface names at boot too :-) It's
100  * just a matter of making everybody's life easier. Since we are doing
101  * threaded probing, it's a bit harder though. The base idea here is that
102  * we make up a list of all emacs in the device-tree before we register the
103  * driver. Every emac will then wait for the previous one in the list to
104  * initialize before itself. We should also keep that list ordered by
105  * cell_index.
106  * That list is only 4 entries long, meaning that additional EMACs don't
107  * get ordering guarantees unless EMAC_BOOT_LIST_SIZE is increased.
108  */
109 
110 #define EMAC_BOOT_LIST_SIZE	4
111 static struct device_node *emac_boot_list[EMAC_BOOT_LIST_SIZE];
112 
113 /* I don't want to litter system log with timeout errors
114  * when we have brain-damaged PHY.
115  */
116 static inline void emac_report_timeout_error(struct emac_instance *dev,
117 					     const char *error)
118 {
119 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX |
120 				  EMAC_FTR_460EX_PHY_CLK_FIX |
121 				  EMAC_FTR_440EP_PHY_CLK_FIX))
122 		DBG(dev, "%s" NL, error);
123 	else if (net_ratelimit())
124 		printk(KERN_ERR "%pOF: %s\n", dev->ofdev->dev.of_node, error);
125 }
126 
127 /* EMAC PHY clock workaround:
128  * 440EP/440GR has more sane SDR0_MFR register implementation than 440GX,
129  * which allows controlling each EMAC clock
130  */
131 static inline void emac_rx_clk_tx(struct emac_instance *dev)
132 {
133 #ifdef CONFIG_PPC_DCR_NATIVE
134 	if (emac_has_feature(dev, EMAC_FTR_440EP_PHY_CLK_FIX))
135 		dcri_clrset(SDR0, SDR0_MFR,
136 			    0, SDR0_MFR_ECS >> dev->cell_index);
137 #endif
138 }
139 
140 static inline void emac_rx_clk_default(struct emac_instance *dev)
141 {
142 #ifdef CONFIG_PPC_DCR_NATIVE
143 	if (emac_has_feature(dev, EMAC_FTR_440EP_PHY_CLK_FIX))
144 		dcri_clrset(SDR0, SDR0_MFR,
145 			    SDR0_MFR_ECS >> dev->cell_index, 0);
146 #endif
147 }
148 
149 /* PHY polling intervals */
150 #define PHY_POLL_LINK_ON	HZ
151 #define PHY_POLL_LINK_OFF	(HZ / 5)
152 
153 /* Graceful stop timeouts in us.
154  * We should allow up to 1 frame time (full-duplex, ignoring collisions)
155  */
156 #define STOP_TIMEOUT_10		1230
157 #define STOP_TIMEOUT_100	124
158 #define STOP_TIMEOUT_1000	13
159 #define STOP_TIMEOUT_1000_JUMBO	73
160 
161 static unsigned char default_mcast_addr[] = {
162 	0x01, 0x80, 0xC2, 0x00, 0x00, 0x01
163 };
164 
165 /* Please, keep in sync with struct ibm_emac_stats/ibm_emac_error_stats */
166 static const char emac_stats_keys[EMAC_ETHTOOL_STATS_COUNT][ETH_GSTRING_LEN] = {
167 	"rx_packets", "rx_bytes", "tx_packets", "tx_bytes", "rx_packets_csum",
168 	"tx_packets_csum", "tx_undo", "rx_dropped_stack", "rx_dropped_oom",
169 	"rx_dropped_error", "rx_dropped_resize", "rx_dropped_mtu",
170 	"rx_stopped", "rx_bd_errors", "rx_bd_overrun", "rx_bd_bad_packet",
171 	"rx_bd_runt_packet", "rx_bd_short_event", "rx_bd_alignment_error",
172 	"rx_bd_bad_fcs", "rx_bd_packet_too_long", "rx_bd_out_of_range",
173 	"rx_bd_in_range", "rx_parity", "rx_fifo_overrun", "rx_overrun",
174 	"rx_bad_packet", "rx_runt_packet", "rx_short_event",
175 	"rx_alignment_error", "rx_bad_fcs", "rx_packet_too_long",
176 	"rx_out_of_range", "rx_in_range", "tx_dropped", "tx_bd_errors",
177 	"tx_bd_bad_fcs", "tx_bd_carrier_loss", "tx_bd_excessive_deferral",
178 	"tx_bd_excessive_collisions", "tx_bd_late_collision",
179 	"tx_bd_multple_collisions", "tx_bd_single_collision",
180 	"tx_bd_underrun", "tx_bd_sqe", "tx_parity", "tx_underrun", "tx_sqe",
181 	"tx_errors"
182 };
183 
184 static irqreturn_t emac_irq(int irq, void *dev_instance);
185 static void emac_clean_tx_ring(struct emac_instance *dev);
186 static void __emac_set_multicast_list(struct emac_instance *dev);
187 
188 static inline int emac_phy_supports_gige(int phy_mode)
189 {
190 	return  phy_interface_mode_is_rgmii(phy_mode) ||
191 		phy_mode == PHY_INTERFACE_MODE_GMII ||
192 		phy_mode == PHY_INTERFACE_MODE_SGMII ||
193 		phy_mode == PHY_INTERFACE_MODE_TBI ||
194 		phy_mode == PHY_INTERFACE_MODE_RTBI;
195 }
196 
197 static inline int emac_phy_gpcs(int phy_mode)
198 {
199 	return  phy_mode == PHY_INTERFACE_MODE_SGMII ||
200 		phy_mode == PHY_INTERFACE_MODE_TBI ||
201 		phy_mode == PHY_INTERFACE_MODE_RTBI;
202 }
203 
204 static inline void emac_tx_enable(struct emac_instance *dev)
205 {
206 	struct emac_regs __iomem *p = dev->emacp;
207 	u32 r;
208 
209 	DBG(dev, "tx_enable" NL);
210 
211 	r = in_be32(&p->mr0);
212 	if (!(r & EMAC_MR0_TXE))
213 		out_be32(&p->mr0, r | EMAC_MR0_TXE);
214 }
215 
216 static void emac_tx_disable(struct emac_instance *dev)
217 {
218 	struct emac_regs __iomem *p = dev->emacp;
219 	u32 r;
220 
221 	DBG(dev, "tx_disable" NL);
222 
223 	r = in_be32(&p->mr0);
224 	if (r & EMAC_MR0_TXE) {
225 		int n = dev->stop_timeout;
226 		out_be32(&p->mr0, r & ~EMAC_MR0_TXE);
227 		while (!(in_be32(&p->mr0) & EMAC_MR0_TXI) && n) {
228 			udelay(1);
229 			--n;
230 		}
231 		if (unlikely(!n))
232 			emac_report_timeout_error(dev, "TX disable timeout");
233 	}
234 }
235 
236 static void emac_rx_enable(struct emac_instance *dev)
237 {
238 	struct emac_regs __iomem *p = dev->emacp;
239 	u32 r;
240 
241 	if (unlikely(test_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags)))
242 		goto out;
243 
244 	DBG(dev, "rx_enable" NL);
245 
246 	r = in_be32(&p->mr0);
247 	if (!(r & EMAC_MR0_RXE)) {
248 		if (unlikely(!(r & EMAC_MR0_RXI))) {
249 			/* Wait if previous async disable is still in progress */
250 			int n = dev->stop_timeout;
251 			while (!(r = in_be32(&p->mr0) & EMAC_MR0_RXI) && n) {
252 				udelay(1);
253 				--n;
254 			}
255 			if (unlikely(!n))
256 				emac_report_timeout_error(dev,
257 							  "RX disable timeout");
258 		}
259 		out_be32(&p->mr0, r | EMAC_MR0_RXE);
260 	}
261  out:
262 	;
263 }
264 
265 static void emac_rx_disable(struct emac_instance *dev)
266 {
267 	struct emac_regs __iomem *p = dev->emacp;
268 	u32 r;
269 
270 	DBG(dev, "rx_disable" NL);
271 
272 	r = in_be32(&p->mr0);
273 	if (r & EMAC_MR0_RXE) {
274 		int n = dev->stop_timeout;
275 		out_be32(&p->mr0, r & ~EMAC_MR0_RXE);
276 		while (!(in_be32(&p->mr0) & EMAC_MR0_RXI) && n) {
277 			udelay(1);
278 			--n;
279 		}
280 		if (unlikely(!n))
281 			emac_report_timeout_error(dev, "RX disable timeout");
282 	}
283 }
284 
285 static inline void emac_netif_stop(struct emac_instance *dev)
286 {
287 	netif_tx_lock_bh(dev->ndev);
288 	netif_addr_lock(dev->ndev);
289 	dev->no_mcast = 1;
290 	netif_addr_unlock(dev->ndev);
291 	netif_tx_unlock_bh(dev->ndev);
292 	netif_trans_update(dev->ndev);	/* prevent tx timeout */
293 	mal_poll_disable(dev->mal, &dev->commac);
294 	netif_tx_disable(dev->ndev);
295 }
296 
297 static inline void emac_netif_start(struct emac_instance *dev)
298 {
299 	netif_tx_lock_bh(dev->ndev);
300 	netif_addr_lock(dev->ndev);
301 	dev->no_mcast = 0;
302 	if (dev->mcast_pending && netif_running(dev->ndev))
303 		__emac_set_multicast_list(dev);
304 	netif_addr_unlock(dev->ndev);
305 	netif_tx_unlock_bh(dev->ndev);
306 
307 	netif_wake_queue(dev->ndev);
308 
309 	/* NOTE: unconditional netif_wake_queue is only appropriate
310 	 * so long as all callers are assured to have free tx slots
311 	 * (taken from tg3... though the case where that is wrong is
312 	 *  not terribly harmful)
313 	 */
314 	mal_poll_enable(dev->mal, &dev->commac);
315 }
316 
317 static inline void emac_rx_disable_async(struct emac_instance *dev)
318 {
319 	struct emac_regs __iomem *p = dev->emacp;
320 	u32 r;
321 
322 	DBG(dev, "rx_disable_async" NL);
323 
324 	r = in_be32(&p->mr0);
325 	if (r & EMAC_MR0_RXE)
326 		out_be32(&p->mr0, r & ~EMAC_MR0_RXE);
327 }
328 
329 static int emac_reset(struct emac_instance *dev)
330 {
331 	struct emac_regs __iomem *p = dev->emacp;
332 	int n = 20;
333 	bool __maybe_unused try_internal_clock = false;
334 
335 	DBG(dev, "reset" NL);
336 
337 	if (!dev->reset_failed) {
338 		/* 40x erratum suggests stopping RX channel before reset,
339 		 * we stop TX as well
340 		 */
341 		emac_rx_disable(dev);
342 		emac_tx_disable(dev);
343 	}
344 
345 #ifdef CONFIG_PPC_DCR_NATIVE
346 do_retry:
347 	/*
348 	 * PPC460EX/GT Embedded Processor Advanced User's Manual
349 	 * section 28.10.1 Mode Register 0 (EMACx_MR0) states:
350 	 * Note: The PHY must provide a TX Clk in order to perform a soft reset
351 	 * of the EMAC. If none is present, select the internal clock
352 	 * (SDR0_ETH_CFG[EMACx_PHY_CLK] = 1).
353 	 * After a soft reset, select the external clock.
354 	 *
355 	 * The AR8035-A PHY Meraki MR24 does not provide a TX Clk if the
356 	 * ethernet cable is not attached. This causes the reset to timeout
357 	 * and the PHY detection code in emac_init_phy() is unable to
358 	 * communicate and detect the AR8035-A PHY. As a result, the emac
359 	 * driver bails out early and the user has no ethernet.
360 	 * In order to stay compatible with existing configurations, the
361 	 * driver will temporarily switch to the internal clock, after
362 	 * the first reset fails.
363 	 */
364 	if (emac_has_feature(dev, EMAC_FTR_460EX_PHY_CLK_FIX)) {
365 		if (try_internal_clock || (dev->phy_address == 0xffffffff &&
366 					   dev->phy_map == 0xffffffff)) {
367 			/* No PHY: select internal loop clock before reset */
368 			dcri_clrset(SDR0, SDR0_ETH_CFG,
369 				    0, SDR0_ETH_CFG_ECS << dev->cell_index);
370 		} else {
371 			/* PHY present: select external clock before reset */
372 			dcri_clrset(SDR0, SDR0_ETH_CFG,
373 				    SDR0_ETH_CFG_ECS << dev->cell_index, 0);
374 		}
375 	}
376 #endif
377 
378 	out_be32(&p->mr0, EMAC_MR0_SRST);
379 	while ((in_be32(&p->mr0) & EMAC_MR0_SRST) && n)
380 		--n;
381 
382 #ifdef CONFIG_PPC_DCR_NATIVE
383 	if (emac_has_feature(dev, EMAC_FTR_460EX_PHY_CLK_FIX)) {
384 		if (!n && !try_internal_clock) {
385 			/* first attempt has timed out. */
386 			n = 20;
387 			try_internal_clock = true;
388 			goto do_retry;
389 		}
390 
391 		if (try_internal_clock || (dev->phy_address == 0xffffffff &&
392 					   dev->phy_map == 0xffffffff)) {
393 			/* No PHY: restore external clock source after reset */
394 			dcri_clrset(SDR0, SDR0_ETH_CFG,
395 				    SDR0_ETH_CFG_ECS << dev->cell_index, 0);
396 		}
397 	}
398 #endif
399 
400 	if (n) {
401 		dev->reset_failed = 0;
402 		return 0;
403 	} else {
404 		emac_report_timeout_error(dev, "reset timeout");
405 		dev->reset_failed = 1;
406 		return -ETIMEDOUT;
407 	}
408 }
409 
410 static void emac_hash_mc(struct emac_instance *dev)
411 {
412 	u32 __iomem *gaht_base = emac_gaht_base(dev);
413 	const int regs = EMAC_XAHT_REGS(dev);
414 	u32 gaht_temp[EMAC_XAHT_MAX_REGS];
415 	struct netdev_hw_addr *ha;
416 	int i;
417 
418 	DBG(dev, "hash_mc %d" NL, netdev_mc_count(dev->ndev));
419 
420 	memset(gaht_temp, 0, sizeof (gaht_temp));
421 
422 	netdev_for_each_mc_addr(ha, dev->ndev) {
423 		int slot, reg, mask;
424 		DBG2(dev, "mc %pM" NL, ha->addr);
425 
426 		slot = EMAC_XAHT_CRC_TO_SLOT(dev,
427 					     ether_crc(ETH_ALEN, ha->addr));
428 		reg = EMAC_XAHT_SLOT_TO_REG(dev, slot);
429 		mask = EMAC_XAHT_SLOT_TO_MASK(dev, slot);
430 
431 		gaht_temp[reg] |= mask;
432 	}
433 
434 	for (i = 0; i < regs; i++)
435 		out_be32(gaht_base + i, gaht_temp[i]);
436 }
437 
438 static inline u32 emac_iff2rmr(struct net_device *ndev)
439 {
440 	struct emac_instance *dev = netdev_priv(ndev);
441 	u32 r;
442 
443 	r = EMAC_RMR_SP | EMAC_RMR_SFCS | EMAC_RMR_IAE | EMAC_RMR_BAE;
444 
445 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
446 	    r |= EMAC4_RMR_BASE;
447 	else
448 	    r |= EMAC_RMR_BASE;
449 
450 	if (ndev->flags & IFF_PROMISC)
451 		r |= EMAC_RMR_PME;
452 	else if (ndev->flags & IFF_ALLMULTI ||
453 			 (netdev_mc_count(ndev) > EMAC_XAHT_SLOTS(dev)))
454 		r |= EMAC_RMR_PMME;
455 	else if (!netdev_mc_empty(ndev))
456 		r |= EMAC_RMR_MAE;
457 
458 	if (emac_has_feature(dev, EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE)) {
459 		r &= ~EMAC4_RMR_MJS_MASK;
460 		r |= EMAC4_RMR_MJS(ndev->mtu);
461 	}
462 
463 	return r;
464 }
465 
466 static u32 __emac_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
467 {
468 	u32 ret = EMAC_MR1_VLE | EMAC_MR1_IST | EMAC_MR1_TR0_MULT;
469 
470 	DBG2(dev, "__emac_calc_base_mr1" NL);
471 
472 	switch(tx_size) {
473 	case 2048:
474 		ret |= EMAC_MR1_TFS_2K;
475 		break;
476 	default:
477 		printk(KERN_WARNING "%s: Unknown Tx FIFO size %d\n",
478 		       dev->ndev->name, tx_size);
479 	}
480 
481 	switch(rx_size) {
482 	case 16384:
483 		ret |= EMAC_MR1_RFS_16K;
484 		break;
485 	case 4096:
486 		ret |= EMAC_MR1_RFS_4K;
487 		break;
488 	default:
489 		printk(KERN_WARNING "%s: Unknown Rx FIFO size %d\n",
490 		       dev->ndev->name, rx_size);
491 	}
492 
493 	return ret;
494 }
495 
496 static u32 __emac4_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
497 {
498 	u32 ret = EMAC_MR1_VLE | EMAC_MR1_IST | EMAC4_MR1_TR |
499 		EMAC4_MR1_OBCI(dev->opb_bus_freq / 1000000);
500 
501 	DBG2(dev, "__emac4_calc_base_mr1" NL);
502 
503 	switch(tx_size) {
504 	case 16384:
505 		ret |= EMAC4_MR1_TFS_16K;
506 		break;
507 	case 8192:
508 		ret |= EMAC4_MR1_TFS_8K;
509 		break;
510 	case 4096:
511 		ret |= EMAC4_MR1_TFS_4K;
512 		break;
513 	case 2048:
514 		ret |= EMAC4_MR1_TFS_2K;
515 		break;
516 	default:
517 		printk(KERN_WARNING "%s: Unknown Tx FIFO size %d\n",
518 		       dev->ndev->name, tx_size);
519 	}
520 
521 	switch(rx_size) {
522 	case 16384:
523 		ret |= EMAC4_MR1_RFS_16K;
524 		break;
525 	case 8192:
526 		ret |= EMAC4_MR1_RFS_8K;
527 		break;
528 	case 4096:
529 		ret |= EMAC4_MR1_RFS_4K;
530 		break;
531 	case 2048:
532 		ret |= EMAC4_MR1_RFS_2K;
533 		break;
534 	default:
535 		printk(KERN_WARNING "%s: Unknown Rx FIFO size %d\n",
536 		       dev->ndev->name, rx_size);
537 	}
538 
539 	return ret;
540 }
541 
542 static u32 emac_calc_base_mr1(struct emac_instance *dev, int tx_size, int rx_size)
543 {
544 	return emac_has_feature(dev, EMAC_FTR_EMAC4) ?
545 		__emac4_calc_base_mr1(dev, tx_size, rx_size) :
546 		__emac_calc_base_mr1(dev, tx_size, rx_size);
547 }
548 
549 static inline u32 emac_calc_trtr(struct emac_instance *dev, unsigned int size)
550 {
551 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
552 		return ((size >> 6) - 1) << EMAC_TRTR_SHIFT_EMAC4;
553 	else
554 		return ((size >> 6) - 1) << EMAC_TRTR_SHIFT;
555 }
556 
557 static inline u32 emac_calc_rwmr(struct emac_instance *dev,
558 				 unsigned int low, unsigned int high)
559 {
560 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
561 		return (low << 22) | ( (high & 0x3ff) << 6);
562 	else
563 		return (low << 23) | ( (high & 0x1ff) << 7);
564 }
565 
566 static int emac_configure(struct emac_instance *dev)
567 {
568 	struct emac_regs __iomem *p = dev->emacp;
569 	struct net_device *ndev = dev->ndev;
570 	int tx_size, rx_size, link = netif_carrier_ok(dev->ndev);
571 	u32 r, mr1 = 0;
572 
573 	DBG(dev, "configure" NL);
574 
575 	if (!link) {
576 		out_be32(&p->mr1, in_be32(&p->mr1)
577 			 | EMAC_MR1_FDE | EMAC_MR1_ILE);
578 		udelay(100);
579 	} else if (emac_reset(dev) < 0)
580 		return -ETIMEDOUT;
581 
582 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
583 		tah_reset(dev->tah_dev);
584 
585 	DBG(dev, " link = %d duplex = %d, pause = %d, asym_pause = %d\n",
586 	    link, dev->phy.duplex, dev->phy.pause, dev->phy.asym_pause);
587 
588 	/* Default fifo sizes */
589 	tx_size = dev->tx_fifo_size;
590 	rx_size = dev->rx_fifo_size;
591 
592 	/* No link, force loopback */
593 	if (!link)
594 		mr1 = EMAC_MR1_FDE | EMAC_MR1_ILE;
595 
596 	/* Check for full duplex */
597 	else if (dev->phy.duplex == DUPLEX_FULL)
598 		mr1 |= EMAC_MR1_FDE | EMAC_MR1_MWSW_001;
599 
600 	/* Adjust fifo sizes, mr1 and timeouts based on link speed */
601 	dev->stop_timeout = STOP_TIMEOUT_10;
602 	switch (dev->phy.speed) {
603 	case SPEED_1000:
604 		if (emac_phy_gpcs(dev->phy.mode)) {
605 			mr1 |= EMAC_MR1_MF_1000GPCS | EMAC_MR1_MF_IPPA(
606 				(dev->phy.gpcs_address != 0xffffffff) ?
607 				 dev->phy.gpcs_address : dev->phy.address);
608 
609 			/* Put some arbitrary OUI, Manuf & Rev IDs so we can
610 			 * identify this GPCS PHY later.
611 			 */
612 			out_be32(&p->u1.emac4.ipcr, 0xdeadbeef);
613 		} else
614 			mr1 |= EMAC_MR1_MF_1000;
615 
616 		/* Extended fifo sizes */
617 		tx_size = dev->tx_fifo_size_gige;
618 		rx_size = dev->rx_fifo_size_gige;
619 
620 		if (dev->ndev->mtu > ETH_DATA_LEN) {
621 			if (emac_has_feature(dev, EMAC_FTR_EMAC4))
622 				mr1 |= EMAC4_MR1_JPSM;
623 			else
624 				mr1 |= EMAC_MR1_JPSM;
625 			dev->stop_timeout = STOP_TIMEOUT_1000_JUMBO;
626 		} else
627 			dev->stop_timeout = STOP_TIMEOUT_1000;
628 		break;
629 	case SPEED_100:
630 		mr1 |= EMAC_MR1_MF_100;
631 		dev->stop_timeout = STOP_TIMEOUT_100;
632 		break;
633 	default: /* make gcc happy */
634 		break;
635 	}
636 
637 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
638 		rgmii_set_speed(dev->rgmii_dev, dev->rgmii_port,
639 				dev->phy.speed);
640 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
641 		zmii_set_speed(dev->zmii_dev, dev->zmii_port, dev->phy.speed);
642 
643 	/* on 40x erratum forces us to NOT use integrated flow control,
644 	 * let's hope it works on 44x ;)
645 	 */
646 	if (!emac_has_feature(dev, EMAC_FTR_NO_FLOW_CONTROL_40x) &&
647 	    dev->phy.duplex == DUPLEX_FULL) {
648 		if (dev->phy.pause)
649 			mr1 |= EMAC_MR1_EIFC | EMAC_MR1_APP;
650 		else if (dev->phy.asym_pause)
651 			mr1 |= EMAC_MR1_APP;
652 	}
653 
654 	/* Add base settings & fifo sizes & program MR1 */
655 	mr1 |= emac_calc_base_mr1(dev, tx_size, rx_size);
656 	out_be32(&p->mr1, mr1);
657 
658 	/* Set individual MAC address */
659 	out_be32(&p->iahr, (ndev->dev_addr[0] << 8) | ndev->dev_addr[1]);
660 	out_be32(&p->ialr, (ndev->dev_addr[2] << 24) |
661 		 (ndev->dev_addr[3] << 16) | (ndev->dev_addr[4] << 8) |
662 		 ndev->dev_addr[5]);
663 
664 	/* VLAN Tag Protocol ID */
665 	out_be32(&p->vtpid, 0x8100);
666 
667 	/* Receive mode register */
668 	r = emac_iff2rmr(ndev);
669 	if (r & EMAC_RMR_MAE)
670 		emac_hash_mc(dev);
671 	out_be32(&p->rmr, r);
672 
673 	/* FIFOs thresholds */
674 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
675 		r = EMAC4_TMR1((dev->mal_burst_size / dev->fifo_entry_size) + 1,
676 			       tx_size / 2 / dev->fifo_entry_size);
677 	else
678 		r = EMAC_TMR1((dev->mal_burst_size / dev->fifo_entry_size) + 1,
679 			      tx_size / 2 / dev->fifo_entry_size);
680 	out_be32(&p->tmr1, r);
681 	out_be32(&p->trtr, emac_calc_trtr(dev, tx_size / 2));
682 
683 	/* PAUSE frame is sent when RX FIFO reaches its high-water mark,
684 	   there should be still enough space in FIFO to allow the our link
685 	   partner time to process this frame and also time to send PAUSE
686 	   frame itself.
687 
688 	   Here is the worst case scenario for the RX FIFO "headroom"
689 	   (from "The Switch Book") (100Mbps, without preamble, inter-frame gap):
690 
691 	   1) One maximum-length frame on TX                    1522 bytes
692 	   2) One PAUSE frame time                                64 bytes
693 	   3) PAUSE frame decode time allowance                   64 bytes
694 	   4) One maximum-length frame on RX                    1522 bytes
695 	   5) Round-trip propagation delay of the link (100Mb)    15 bytes
696 	   ----------
697 	   3187 bytes
698 
699 	   I chose to set high-water mark to RX_FIFO_SIZE / 4 (1024 bytes)
700 	   low-water mark  to RX_FIFO_SIZE / 8 (512 bytes)
701 	 */
702 	r = emac_calc_rwmr(dev, rx_size / 8 / dev->fifo_entry_size,
703 			   rx_size / 4 / dev->fifo_entry_size);
704 	out_be32(&p->rwmr, r);
705 
706 	/* Set PAUSE timer to the maximum */
707 	out_be32(&p->ptr, 0xffff);
708 
709 	/* IRQ sources */
710 	r = EMAC_ISR_OVR | EMAC_ISR_BP | EMAC_ISR_SE |
711 		EMAC_ISR_ALE | EMAC_ISR_BFCS | EMAC_ISR_PTLE | EMAC_ISR_ORE |
712 		EMAC_ISR_IRE | EMAC_ISR_TE;
713 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
714 	    r |= EMAC4_ISR_TXPE | EMAC4_ISR_RXPE /* | EMAC4_ISR_TXUE |
715 						  EMAC4_ISR_RXOE | */;
716 	out_be32(&p->iser,  r);
717 
718 	/* We need to take GPCS PHY out of isolate mode after EMAC reset */
719 	if (emac_phy_gpcs(dev->phy.mode)) {
720 		if (dev->phy.gpcs_address != 0xffffffff)
721 			emac_mii_reset_gpcs(&dev->phy);
722 		else
723 			emac_mii_reset_phy(&dev->phy);
724 	}
725 
726 	return 0;
727 }
728 
729 static void emac_reinitialize(struct emac_instance *dev)
730 {
731 	DBG(dev, "reinitialize" NL);
732 
733 	emac_netif_stop(dev);
734 	if (!emac_configure(dev)) {
735 		emac_tx_enable(dev);
736 		emac_rx_enable(dev);
737 	}
738 	emac_netif_start(dev);
739 }
740 
741 static void emac_full_tx_reset(struct emac_instance *dev)
742 {
743 	DBG(dev, "full_tx_reset" NL);
744 
745 	emac_tx_disable(dev);
746 	mal_disable_tx_channel(dev->mal, dev->mal_tx_chan);
747 	emac_clean_tx_ring(dev);
748 	dev->tx_cnt = dev->tx_slot = dev->ack_slot = 0;
749 
750 	emac_configure(dev);
751 
752 	mal_enable_tx_channel(dev->mal, dev->mal_tx_chan);
753 	emac_tx_enable(dev);
754 	emac_rx_enable(dev);
755 }
756 
757 static void emac_reset_work(struct work_struct *work)
758 {
759 	struct emac_instance *dev = container_of(work, struct emac_instance, reset_work);
760 
761 	DBG(dev, "reset_work" NL);
762 
763 	mutex_lock(&dev->link_lock);
764 	if (dev->opened) {
765 		emac_netif_stop(dev);
766 		emac_full_tx_reset(dev);
767 		emac_netif_start(dev);
768 	}
769 	mutex_unlock(&dev->link_lock);
770 }
771 
772 static void emac_tx_timeout(struct net_device *ndev, unsigned int txqueue)
773 {
774 	struct emac_instance *dev = netdev_priv(ndev);
775 
776 	DBG(dev, "tx_timeout" NL);
777 
778 	schedule_work(&dev->reset_work);
779 }
780 
781 
782 static inline int emac_phy_done(struct emac_instance *dev, u32 stacr)
783 {
784 	int done = !!(stacr & EMAC_STACR_OC);
785 
786 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
787 		done = !done;
788 
789 	return done;
790 };
791 
792 static int __emac_mdio_read(struct emac_instance *dev, u8 id, u8 reg)
793 {
794 	struct emac_regs __iomem *p = dev->emacp;
795 	u32 r = 0;
796 	int n, err = -ETIMEDOUT;
797 
798 	mutex_lock(&dev->mdio_lock);
799 
800 	DBG2(dev, "mdio_read(%02x,%02x)" NL, id, reg);
801 
802 	/* Enable proper MDIO port */
803 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
804 		zmii_get_mdio(dev->zmii_dev, dev->zmii_port);
805 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
806 		rgmii_get_mdio(dev->rgmii_dev, dev->rgmii_port);
807 
808 	/* Wait for management interface to become idle */
809 	n = 20;
810 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
811 		udelay(1);
812 		if (!--n) {
813 			DBG2(dev, " -> timeout wait idle\n");
814 			goto bail;
815 		}
816 	}
817 
818 	/* Issue read command */
819 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
820 		r = EMAC4_STACR_BASE(dev->opb_bus_freq);
821 	else
822 		r = EMAC_STACR_BASE(dev->opb_bus_freq);
823 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
824 		r |= EMAC_STACR_OC;
825 	if (emac_has_feature(dev, EMAC_FTR_HAS_NEW_STACR))
826 		r |= EMACX_STACR_STAC_READ;
827 	else
828 		r |= EMAC_STACR_STAC_READ;
829 	r |= (reg & EMAC_STACR_PRA_MASK)
830 		| ((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT);
831 	out_be32(&p->stacr, r);
832 
833 	/* Wait for read to complete */
834 	n = 200;
835 	while (!emac_phy_done(dev, (r = in_be32(&p->stacr)))) {
836 		udelay(1);
837 		if (!--n) {
838 			DBG2(dev, " -> timeout wait complete\n");
839 			goto bail;
840 		}
841 	}
842 
843 	if (unlikely(r & EMAC_STACR_PHYE)) {
844 		DBG(dev, "mdio_read(%02x, %02x) failed" NL, id, reg);
845 		err = -EREMOTEIO;
846 		goto bail;
847 	}
848 
849 	r = ((r >> EMAC_STACR_PHYD_SHIFT) & EMAC_STACR_PHYD_MASK);
850 
851 	DBG2(dev, "mdio_read -> %04x" NL, r);
852 	err = 0;
853  bail:
854 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
855 		rgmii_put_mdio(dev->rgmii_dev, dev->rgmii_port);
856 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
857 		zmii_put_mdio(dev->zmii_dev, dev->zmii_port);
858 	mutex_unlock(&dev->mdio_lock);
859 
860 	return err == 0 ? r : err;
861 }
862 
863 static void __emac_mdio_write(struct emac_instance *dev, u8 id, u8 reg,
864 			      u16 val)
865 {
866 	struct emac_regs __iomem *p = dev->emacp;
867 	u32 r = 0;
868 	int n;
869 
870 	mutex_lock(&dev->mdio_lock);
871 
872 	DBG2(dev, "mdio_write(%02x,%02x,%04x)" NL, id, reg, val);
873 
874 	/* Enable proper MDIO port */
875 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
876 		zmii_get_mdio(dev->zmii_dev, dev->zmii_port);
877 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
878 		rgmii_get_mdio(dev->rgmii_dev, dev->rgmii_port);
879 
880 	/* Wait for management interface to be idle */
881 	n = 20;
882 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
883 		udelay(1);
884 		if (!--n) {
885 			DBG2(dev, " -> timeout wait idle\n");
886 			goto bail;
887 		}
888 	}
889 
890 	/* Issue write command */
891 	if (emac_has_feature(dev, EMAC_FTR_EMAC4))
892 		r = EMAC4_STACR_BASE(dev->opb_bus_freq);
893 	else
894 		r = EMAC_STACR_BASE(dev->opb_bus_freq);
895 	if (emac_has_feature(dev, EMAC_FTR_STACR_OC_INVERT))
896 		r |= EMAC_STACR_OC;
897 	if (emac_has_feature(dev, EMAC_FTR_HAS_NEW_STACR))
898 		r |= EMACX_STACR_STAC_WRITE;
899 	else
900 		r |= EMAC_STACR_STAC_WRITE;
901 	r |= (reg & EMAC_STACR_PRA_MASK) |
902 		((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT) |
903 		(val << EMAC_STACR_PHYD_SHIFT);
904 	out_be32(&p->stacr, r);
905 
906 	/* Wait for write to complete */
907 	n = 200;
908 	while (!emac_phy_done(dev, in_be32(&p->stacr))) {
909 		udelay(1);
910 		if (!--n) {
911 			DBG2(dev, " -> timeout wait complete\n");
912 			goto bail;
913 		}
914 	}
915  bail:
916 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
917 		rgmii_put_mdio(dev->rgmii_dev, dev->rgmii_port);
918 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
919 		zmii_put_mdio(dev->zmii_dev, dev->zmii_port);
920 	mutex_unlock(&dev->mdio_lock);
921 }
922 
923 static int emac_mdio_read(struct net_device *ndev, int id, int reg)
924 {
925 	struct emac_instance *dev = netdev_priv(ndev);
926 	int res;
927 
928 	res = __emac_mdio_read((dev->mdio_instance &&
929 				dev->phy.gpcs_address != id) ?
930 				dev->mdio_instance : dev,
931 			       (u8) id, (u8) reg);
932 	return res;
933 }
934 
935 static void emac_mdio_write(struct net_device *ndev, int id, int reg, int val)
936 {
937 	struct emac_instance *dev = netdev_priv(ndev);
938 
939 	__emac_mdio_write((dev->mdio_instance &&
940 			   dev->phy.gpcs_address != id) ?
941 			   dev->mdio_instance : dev,
942 			  (u8) id, (u8) reg, (u16) val);
943 }
944 
945 /* Tx lock BH */
946 static void __emac_set_multicast_list(struct emac_instance *dev)
947 {
948 	struct emac_regs __iomem *p = dev->emacp;
949 	u32 rmr = emac_iff2rmr(dev->ndev);
950 
951 	DBG(dev, "__multicast %08x" NL, rmr);
952 
953 	/* I decided to relax register access rules here to avoid
954 	 * full EMAC reset.
955 	 *
956 	 * There is a real problem with EMAC4 core if we use MWSW_001 bit
957 	 * in MR1 register and do a full EMAC reset.
958 	 * One TX BD status update is delayed and, after EMAC reset, it
959 	 * never happens, resulting in TX hung (it'll be recovered by TX
960 	 * timeout handler eventually, but this is just gross).
961 	 * So we either have to do full TX reset or try to cheat here :)
962 	 *
963 	 * The only required change is to RX mode register, so I *think* all
964 	 * we need is just to stop RX channel. This seems to work on all
965 	 * tested SoCs.                                                --ebs
966 	 *
967 	 */
968 	dev->mcast_pending = 0;
969 	emac_rx_disable(dev);
970 	if (rmr & EMAC_RMR_MAE)
971 		emac_hash_mc(dev);
972 	out_be32(&p->rmr, rmr);
973 	emac_rx_enable(dev);
974 }
975 
976 /* Tx lock BH */
977 static void emac_set_multicast_list(struct net_device *ndev)
978 {
979 	struct emac_instance *dev = netdev_priv(ndev);
980 
981 	DBG(dev, "multicast" NL);
982 
983 	BUG_ON(!netif_running(dev->ndev));
984 
985 	if (dev->no_mcast) {
986 		dev->mcast_pending = 1;
987 		return;
988 	}
989 
990 	mutex_lock(&dev->link_lock);
991 	__emac_set_multicast_list(dev);
992 	mutex_unlock(&dev->link_lock);
993 }
994 
995 static int emac_set_mac_address(struct net_device *ndev, void *sa)
996 {
997 	struct emac_instance *dev = netdev_priv(ndev);
998 	struct sockaddr *addr = sa;
999 	struct emac_regs __iomem *p = dev->emacp;
1000 
1001 	if (!is_valid_ether_addr(addr->sa_data))
1002 	       return -EADDRNOTAVAIL;
1003 
1004 	mutex_lock(&dev->link_lock);
1005 
1006 	eth_hw_addr_set(ndev, addr->sa_data);
1007 
1008 	emac_rx_disable(dev);
1009 	emac_tx_disable(dev);
1010 	out_be32(&p->iahr, (ndev->dev_addr[0] << 8) | ndev->dev_addr[1]);
1011 	out_be32(&p->ialr, (ndev->dev_addr[2] << 24) |
1012 		(ndev->dev_addr[3] << 16) | (ndev->dev_addr[4] << 8) |
1013 		ndev->dev_addr[5]);
1014 	emac_tx_enable(dev);
1015 	emac_rx_enable(dev);
1016 
1017 	mutex_unlock(&dev->link_lock);
1018 
1019 	return 0;
1020 }
1021 
1022 static int emac_resize_rx_ring(struct emac_instance *dev, int new_mtu)
1023 {
1024 	int rx_sync_size = emac_rx_sync_size(new_mtu);
1025 	int rx_skb_size = emac_rx_skb_size(new_mtu);
1026 	int i, ret = 0;
1027 	int mr1_jumbo_bit_change = 0;
1028 
1029 	mutex_lock(&dev->link_lock);
1030 	emac_netif_stop(dev);
1031 	emac_rx_disable(dev);
1032 	mal_disable_rx_channel(dev->mal, dev->mal_rx_chan);
1033 
1034 	if (dev->rx_sg_skb) {
1035 		++dev->estats.rx_dropped_resize;
1036 		dev_kfree_skb(dev->rx_sg_skb);
1037 		dev->rx_sg_skb = NULL;
1038 	}
1039 
1040 	/* Make a first pass over RX ring and mark BDs ready, dropping
1041 	 * non-processed packets on the way. We need this as a separate pass
1042 	 * to simplify error recovery in the case of allocation failure later.
1043 	 */
1044 	for (i = 0; i < NUM_RX_BUFF; ++i) {
1045 		if (dev->rx_desc[i].ctrl & MAL_RX_CTRL_FIRST)
1046 			++dev->estats.rx_dropped_resize;
1047 
1048 		dev->rx_desc[i].data_len = 0;
1049 		dev->rx_desc[i].ctrl = MAL_RX_CTRL_EMPTY |
1050 		    (i == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1051 	}
1052 
1053 	/* Reallocate RX ring only if bigger skb buffers are required */
1054 	if (rx_skb_size <= dev->rx_skb_size)
1055 		goto skip;
1056 
1057 	/* Second pass, allocate new skbs */
1058 	for (i = 0; i < NUM_RX_BUFF; ++i) {
1059 		struct sk_buff *skb;
1060 
1061 		skb = netdev_alloc_skb_ip_align(dev->ndev, rx_skb_size);
1062 		if (!skb) {
1063 			ret = -ENOMEM;
1064 			goto oom;
1065 		}
1066 
1067 		BUG_ON(!dev->rx_skb[i]);
1068 		dev_kfree_skb(dev->rx_skb[i]);
1069 
1070 		dev->rx_desc[i].data_ptr =
1071 		    dma_map_single(&dev->ofdev->dev, skb->data - NET_IP_ALIGN,
1072 				   rx_sync_size, DMA_FROM_DEVICE)
1073 				   + NET_IP_ALIGN;
1074 		dev->rx_skb[i] = skb;
1075 	}
1076  skip:
1077 	/* Check if we need to change "Jumbo" bit in MR1 */
1078 	if (emac_has_feature(dev, EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE)) {
1079 		mr1_jumbo_bit_change = (new_mtu > ETH_DATA_LEN) ||
1080 				(dev->ndev->mtu > ETH_DATA_LEN);
1081 	} else {
1082 		mr1_jumbo_bit_change = (new_mtu > ETH_DATA_LEN) ^
1083 				(dev->ndev->mtu > ETH_DATA_LEN);
1084 	}
1085 
1086 	if (mr1_jumbo_bit_change) {
1087 		/* This is to prevent starting RX channel in emac_rx_enable() */
1088 		set_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1089 
1090 		WRITE_ONCE(dev->ndev->mtu, new_mtu);
1091 		emac_full_tx_reset(dev);
1092 	}
1093 
1094 	mal_set_rcbs(dev->mal, dev->mal_rx_chan, emac_rx_size(new_mtu));
1095  oom:
1096 	/* Restart RX */
1097 	clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1098 	dev->rx_slot = 0;
1099 	mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1100 	emac_rx_enable(dev);
1101 	emac_netif_start(dev);
1102 	mutex_unlock(&dev->link_lock);
1103 
1104 	return ret;
1105 }
1106 
1107 /* Process ctx, rtnl_lock semaphore */
1108 static int emac_change_mtu(struct net_device *ndev, int new_mtu)
1109 {
1110 	struct emac_instance *dev = netdev_priv(ndev);
1111 	int ret = 0;
1112 
1113 	DBG(dev, "change_mtu(%d)" NL, new_mtu);
1114 
1115 	if (netif_running(ndev)) {
1116 		/* Check if we really need to reinitialize RX ring */
1117 		if (emac_rx_skb_size(ndev->mtu) != emac_rx_skb_size(new_mtu))
1118 			ret = emac_resize_rx_ring(dev, new_mtu);
1119 	}
1120 
1121 	if (!ret) {
1122 		WRITE_ONCE(ndev->mtu, new_mtu);
1123 		dev->rx_skb_size = emac_rx_skb_size(new_mtu);
1124 		dev->rx_sync_size = emac_rx_sync_size(new_mtu);
1125 	}
1126 
1127 	return ret;
1128 }
1129 
1130 static void emac_clean_tx_ring(struct emac_instance *dev)
1131 {
1132 	int i;
1133 
1134 	for (i = 0; i < NUM_TX_BUFF; ++i) {
1135 		if (dev->tx_skb[i]) {
1136 			dev_kfree_skb(dev->tx_skb[i]);
1137 			dev->tx_skb[i] = NULL;
1138 			if (dev->tx_desc[i].ctrl & MAL_TX_CTRL_READY)
1139 				++dev->estats.tx_dropped;
1140 		}
1141 		dev->tx_desc[i].ctrl = 0;
1142 		dev->tx_desc[i].data_ptr = 0;
1143 	}
1144 }
1145 
1146 static void emac_clean_rx_ring(struct emac_instance *dev)
1147 {
1148 	int i;
1149 
1150 	for (i = 0; i < NUM_RX_BUFF; ++i)
1151 		if (dev->rx_skb[i]) {
1152 			dev->rx_desc[i].ctrl = 0;
1153 			dev_kfree_skb(dev->rx_skb[i]);
1154 			dev->rx_skb[i] = NULL;
1155 			dev->rx_desc[i].data_ptr = 0;
1156 		}
1157 
1158 	if (dev->rx_sg_skb) {
1159 		dev_kfree_skb(dev->rx_sg_skb);
1160 		dev->rx_sg_skb = NULL;
1161 	}
1162 }
1163 
1164 static int
1165 __emac_prepare_rx_skb(struct sk_buff *skb, struct emac_instance *dev, int slot)
1166 {
1167 	if (unlikely(!skb))
1168 		return -ENOMEM;
1169 
1170 	dev->rx_skb[slot] = skb;
1171 	dev->rx_desc[slot].data_len = 0;
1172 
1173 	dev->rx_desc[slot].data_ptr =
1174 	    dma_map_single(&dev->ofdev->dev, skb->data - NET_IP_ALIGN,
1175 			   dev->rx_sync_size, DMA_FROM_DEVICE) + NET_IP_ALIGN;
1176 	wmb();
1177 	dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1178 	    (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1179 
1180 	return 0;
1181 }
1182 
1183 static int
1184 emac_alloc_rx_skb(struct emac_instance *dev, int slot)
1185 {
1186 	struct sk_buff *skb;
1187 
1188 	skb = __netdev_alloc_skb_ip_align(dev->ndev, dev->rx_skb_size,
1189 					  GFP_KERNEL);
1190 
1191 	return __emac_prepare_rx_skb(skb, dev, slot);
1192 }
1193 
1194 static int
1195 emac_alloc_rx_skb_napi(struct emac_instance *dev, int slot)
1196 {
1197 	struct sk_buff *skb;
1198 
1199 	skb = napi_alloc_skb(&dev->mal->napi, dev->rx_skb_size);
1200 
1201 	return __emac_prepare_rx_skb(skb, dev, slot);
1202 }
1203 
1204 static void emac_print_link_status(struct emac_instance *dev)
1205 {
1206 	if (netif_carrier_ok(dev->ndev))
1207 		printk(KERN_INFO "%s: link is up, %d %s%s\n",
1208 		       dev->ndev->name, dev->phy.speed,
1209 		       dev->phy.duplex == DUPLEX_FULL ? "FDX" : "HDX",
1210 		       dev->phy.pause ? ", pause enabled" :
1211 		       dev->phy.asym_pause ? ", asymmetric pause enabled" : "");
1212 	else
1213 		printk(KERN_INFO "%s: link is down\n", dev->ndev->name);
1214 }
1215 
1216 /* Process ctx, rtnl_lock semaphore */
1217 static int emac_open(struct net_device *ndev)
1218 {
1219 	struct emac_instance *dev = netdev_priv(ndev);
1220 	int i;
1221 
1222 	DBG(dev, "open" NL);
1223 
1224 	/* Allocate RX ring */
1225 	for (i = 0; i < NUM_RX_BUFF; ++i)
1226 		if (emac_alloc_rx_skb(dev, i)) {
1227 			printk(KERN_ERR "%s: failed to allocate RX ring\n",
1228 			       ndev->name);
1229 			goto oom;
1230 		}
1231 
1232 	dev->tx_cnt = dev->tx_slot = dev->ack_slot = dev->rx_slot = 0;
1233 	clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1234 	dev->rx_sg_skb = NULL;
1235 
1236 	mutex_lock(&dev->link_lock);
1237 	dev->opened = 1;
1238 
1239 	/* Start PHY polling now.
1240 	 */
1241 	if (dev->phy.address >= 0) {
1242 		int link_poll_interval;
1243 		if (dev->phy.def->ops->poll_link(&dev->phy)) {
1244 			dev->phy.def->ops->read_link(&dev->phy);
1245 			emac_rx_clk_default(dev);
1246 			netif_carrier_on(dev->ndev);
1247 			link_poll_interval = PHY_POLL_LINK_ON;
1248 		} else {
1249 			emac_rx_clk_tx(dev);
1250 			netif_carrier_off(dev->ndev);
1251 			link_poll_interval = PHY_POLL_LINK_OFF;
1252 		}
1253 		dev->link_polling = 1;
1254 		wmb();
1255 		schedule_delayed_work(&dev->link_work, link_poll_interval);
1256 		emac_print_link_status(dev);
1257 	} else
1258 		netif_carrier_on(dev->ndev);
1259 
1260 	/* Required for Pause packet support in EMAC */
1261 	dev_mc_add_global(ndev, default_mcast_addr);
1262 
1263 	emac_configure(dev);
1264 	mal_poll_add(dev->mal, &dev->commac);
1265 	mal_enable_tx_channel(dev->mal, dev->mal_tx_chan);
1266 	mal_set_rcbs(dev->mal, dev->mal_rx_chan, emac_rx_size(ndev->mtu));
1267 	mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1268 	emac_tx_enable(dev);
1269 	emac_rx_enable(dev);
1270 	emac_netif_start(dev);
1271 
1272 	mutex_unlock(&dev->link_lock);
1273 
1274 	return 0;
1275  oom:
1276 	emac_clean_rx_ring(dev);
1277 	return -ENOMEM;
1278 }
1279 
1280 /* BHs disabled */
1281 #if 0
1282 static int emac_link_differs(struct emac_instance *dev)
1283 {
1284 	u32 r = in_be32(&dev->emacp->mr1);
1285 
1286 	int duplex = r & EMAC_MR1_FDE ? DUPLEX_FULL : DUPLEX_HALF;
1287 	int speed, pause, asym_pause;
1288 
1289 	if (r & EMAC_MR1_MF_1000)
1290 		speed = SPEED_1000;
1291 	else if (r & EMAC_MR1_MF_100)
1292 		speed = SPEED_100;
1293 	else
1294 		speed = SPEED_10;
1295 
1296 	switch (r & (EMAC_MR1_EIFC | EMAC_MR1_APP)) {
1297 	case (EMAC_MR1_EIFC | EMAC_MR1_APP):
1298 		pause = 1;
1299 		asym_pause = 0;
1300 		break;
1301 	case EMAC_MR1_APP:
1302 		pause = 0;
1303 		asym_pause = 1;
1304 		break;
1305 	default:
1306 		pause = asym_pause = 0;
1307 	}
1308 	return speed != dev->phy.speed || duplex != dev->phy.duplex ||
1309 	    pause != dev->phy.pause || asym_pause != dev->phy.asym_pause;
1310 }
1311 #endif
1312 
1313 static void emac_link_timer(struct work_struct *work)
1314 {
1315 	struct emac_instance *dev =
1316 		container_of(to_delayed_work(work),
1317 			     struct emac_instance, link_work);
1318 	int link_poll_interval;
1319 
1320 	mutex_lock(&dev->link_lock);
1321 	DBG2(dev, "link timer" NL);
1322 
1323 	if (!dev->opened)
1324 		goto bail;
1325 
1326 	if (dev->phy.def->ops->poll_link(&dev->phy)) {
1327 		if (!netif_carrier_ok(dev->ndev)) {
1328 			emac_rx_clk_default(dev);
1329 			/* Get new link parameters */
1330 			dev->phy.def->ops->read_link(&dev->phy);
1331 
1332 			netif_carrier_on(dev->ndev);
1333 			emac_netif_stop(dev);
1334 			emac_full_tx_reset(dev);
1335 			emac_netif_start(dev);
1336 			emac_print_link_status(dev);
1337 		}
1338 		link_poll_interval = PHY_POLL_LINK_ON;
1339 	} else {
1340 		if (netif_carrier_ok(dev->ndev)) {
1341 			emac_rx_clk_tx(dev);
1342 			netif_carrier_off(dev->ndev);
1343 			netif_tx_disable(dev->ndev);
1344 			emac_reinitialize(dev);
1345 			emac_print_link_status(dev);
1346 		}
1347 		link_poll_interval = PHY_POLL_LINK_OFF;
1348 	}
1349 	schedule_delayed_work(&dev->link_work, link_poll_interval);
1350  bail:
1351 	mutex_unlock(&dev->link_lock);
1352 }
1353 
1354 static void emac_force_link_update(struct emac_instance *dev)
1355 {
1356 	netif_carrier_off(dev->ndev);
1357 	smp_rmb();
1358 	if (dev->link_polling) {
1359 		cancel_delayed_work_sync(&dev->link_work);
1360 		if (dev->link_polling)
1361 			schedule_delayed_work(&dev->link_work,  PHY_POLL_LINK_OFF);
1362 	}
1363 }
1364 
1365 /* Process ctx, rtnl_lock semaphore */
1366 static int emac_close(struct net_device *ndev)
1367 {
1368 	struct emac_instance *dev = netdev_priv(ndev);
1369 
1370 	DBG(dev, "close" NL);
1371 
1372 	if (dev->phy.address >= 0) {
1373 		dev->link_polling = 0;
1374 		cancel_delayed_work_sync(&dev->link_work);
1375 	}
1376 	mutex_lock(&dev->link_lock);
1377 	emac_netif_stop(dev);
1378 	dev->opened = 0;
1379 	mutex_unlock(&dev->link_lock);
1380 
1381 	emac_rx_disable(dev);
1382 	emac_tx_disable(dev);
1383 	mal_disable_rx_channel(dev->mal, dev->mal_rx_chan);
1384 	mal_disable_tx_channel(dev->mal, dev->mal_tx_chan);
1385 	mal_poll_del(dev->mal, &dev->commac);
1386 
1387 	emac_clean_tx_ring(dev);
1388 	emac_clean_rx_ring(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 netdev_tx_t 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 netdev_tx_t 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 netdev_tx_t
1492 emac_start_xmit_sg(struct sk_buff *skb, struct net_device *ndev)
1493 {
1494 	struct emac_instance *dev = netdev_priv(ndev);
1495 	int nr_frags = skb_shinfo(skb)->nr_frags;
1496 	int len = skb->len, chunk;
1497 	int slot, i;
1498 	u16 ctrl;
1499 	u32 pd;
1500 
1501 	/* This is common "fast" path */
1502 	if (likely(!nr_frags && len <= MAL_MAX_TX_SIZE))
1503 		return emac_start_xmit(skb, ndev);
1504 
1505 	len -= skb->data_len;
1506 
1507 	/* Note, this is only an *estimation*, we can still run out of empty
1508 	 * slots because of the additional fragmentation into
1509 	 * MAL_MAX_TX_SIZE-sized chunks
1510 	 */
1511 	if (unlikely(dev->tx_cnt + nr_frags + mal_tx_chunks(len) > NUM_TX_BUFF))
1512 		goto stop_queue;
1513 
1514 	ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1515 	    emac_tx_csum(dev, skb);
1516 	slot = dev->tx_slot;
1517 
1518 	/* skb data */
1519 	dev->tx_skb[slot] = NULL;
1520 	chunk = min(len, MAL_MAX_TX_SIZE);
1521 	dev->tx_desc[slot].data_ptr = pd =
1522 	    dma_map_single(&dev->ofdev->dev, skb->data, len, DMA_TO_DEVICE);
1523 	dev->tx_desc[slot].data_len = (u16) chunk;
1524 	len -= chunk;
1525 	if (unlikely(len))
1526 		slot = emac_xmit_split(dev, slot, pd + chunk, len, !nr_frags,
1527 				       ctrl);
1528 	/* skb fragments */
1529 	for (i = 0; i < nr_frags; ++i) {
1530 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1531 		len = skb_frag_size(frag);
1532 
1533 		if (unlikely(dev->tx_cnt + mal_tx_chunks(len) >= NUM_TX_BUFF))
1534 			goto undo_frame;
1535 
1536 		pd = skb_frag_dma_map(&dev->ofdev->dev, frag, 0, len,
1537 				      DMA_TO_DEVICE);
1538 
1539 		slot = emac_xmit_split(dev, slot, pd, len, i == nr_frags - 1,
1540 				       ctrl);
1541 	}
1542 
1543 	DBG2(dev, "xmit_sg(%u) %d - %d" NL, skb->len, dev->tx_slot, slot);
1544 
1545 	/* Attach skb to the last slot so we don't release it too early */
1546 	dev->tx_skb[slot] = skb;
1547 
1548 	/* Send the packet out */
1549 	if (dev->tx_slot == NUM_TX_BUFF - 1)
1550 		ctrl |= MAL_TX_CTRL_WRAP;
1551 	wmb();
1552 	dev->tx_desc[dev->tx_slot].ctrl = ctrl;
1553 	dev->tx_slot = (slot + 1) % NUM_TX_BUFF;
1554 
1555 	return emac_xmit_finish(dev, skb->len);
1556 
1557  undo_frame:
1558 	/* Well, too bad. Our previous estimation was overly optimistic.
1559 	 * Undo everything.
1560 	 */
1561 	while (slot != dev->tx_slot) {
1562 		dev->tx_desc[slot].ctrl = 0;
1563 		--dev->tx_cnt;
1564 		if (--slot < 0)
1565 			slot = NUM_TX_BUFF - 1;
1566 	}
1567 	++dev->estats.tx_undo;
1568 
1569  stop_queue:
1570 	netif_stop_queue(ndev);
1571 	DBG2(dev, "stopped TX queue" NL);
1572 	return NETDEV_TX_BUSY;
1573 }
1574 
1575 /* Tx lock BHs */
1576 static void emac_parse_tx_error(struct emac_instance *dev, u16 ctrl)
1577 {
1578 	struct emac_error_stats *st = &dev->estats;
1579 
1580 	DBG(dev, "BD TX error %04x" NL, ctrl);
1581 
1582 	++st->tx_bd_errors;
1583 	if (ctrl & EMAC_TX_ST_BFCS)
1584 		++st->tx_bd_bad_fcs;
1585 	if (ctrl & EMAC_TX_ST_LCS)
1586 		++st->tx_bd_carrier_loss;
1587 	if (ctrl & EMAC_TX_ST_ED)
1588 		++st->tx_bd_excessive_deferral;
1589 	if (ctrl & EMAC_TX_ST_EC)
1590 		++st->tx_bd_excessive_collisions;
1591 	if (ctrl & EMAC_TX_ST_LC)
1592 		++st->tx_bd_late_collision;
1593 	if (ctrl & EMAC_TX_ST_MC)
1594 		++st->tx_bd_multple_collisions;
1595 	if (ctrl & EMAC_TX_ST_SC)
1596 		++st->tx_bd_single_collision;
1597 	if (ctrl & EMAC_TX_ST_UR)
1598 		++st->tx_bd_underrun;
1599 	if (ctrl & EMAC_TX_ST_SQE)
1600 		++st->tx_bd_sqe;
1601 }
1602 
1603 static void emac_poll_tx(void *param)
1604 {
1605 	struct emac_instance *dev = param;
1606 	u32 bad_mask;
1607 
1608 	DBG2(dev, "poll_tx, %d %d" NL, dev->tx_cnt, dev->ack_slot);
1609 
1610 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
1611 		bad_mask = EMAC_IS_BAD_TX_TAH;
1612 	else
1613 		bad_mask = EMAC_IS_BAD_TX;
1614 
1615 	netif_tx_lock_bh(dev->ndev);
1616 	if (dev->tx_cnt) {
1617 		u16 ctrl;
1618 		int slot = dev->ack_slot, n = 0;
1619 	again:
1620 		ctrl = dev->tx_desc[slot].ctrl;
1621 		if (!(ctrl & MAL_TX_CTRL_READY)) {
1622 			struct sk_buff *skb = dev->tx_skb[slot];
1623 			++n;
1624 
1625 			if (skb) {
1626 				dev_kfree_skb(skb);
1627 				dev->tx_skb[slot] = NULL;
1628 			}
1629 			slot = (slot + 1) % NUM_TX_BUFF;
1630 
1631 			if (unlikely(ctrl & bad_mask))
1632 				emac_parse_tx_error(dev, ctrl);
1633 
1634 			if (--dev->tx_cnt)
1635 				goto again;
1636 		}
1637 		if (n) {
1638 			dev->ack_slot = slot;
1639 			if (netif_queue_stopped(dev->ndev) &&
1640 			    dev->tx_cnt < EMAC_TX_WAKEUP_THRESH)
1641 				netif_wake_queue(dev->ndev);
1642 
1643 			DBG2(dev, "tx %d pkts" NL, n);
1644 		}
1645 	}
1646 	netif_tx_unlock_bh(dev->ndev);
1647 }
1648 
1649 static inline void emac_recycle_rx_skb(struct emac_instance *dev, int slot,
1650 				       int len)
1651 {
1652 	struct sk_buff *skb = dev->rx_skb[slot];
1653 
1654 	DBG2(dev, "recycle %d %d" NL, slot, len);
1655 
1656 	if (len)
1657 		dma_map_single(&dev->ofdev->dev, skb->data - NET_IP_ALIGN,
1658 			       SKB_DATA_ALIGN(len + NET_IP_ALIGN),
1659 			       DMA_FROM_DEVICE);
1660 
1661 	dev->rx_desc[slot].data_len = 0;
1662 	wmb();
1663 	dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1664 	    (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1665 }
1666 
1667 static void emac_parse_rx_error(struct emac_instance *dev, u16 ctrl)
1668 {
1669 	struct emac_error_stats *st = &dev->estats;
1670 
1671 	DBG(dev, "BD RX error %04x" NL, ctrl);
1672 
1673 	++st->rx_bd_errors;
1674 	if (ctrl & EMAC_RX_ST_OE)
1675 		++st->rx_bd_overrun;
1676 	if (ctrl & EMAC_RX_ST_BP)
1677 		++st->rx_bd_bad_packet;
1678 	if (ctrl & EMAC_RX_ST_RP)
1679 		++st->rx_bd_runt_packet;
1680 	if (ctrl & EMAC_RX_ST_SE)
1681 		++st->rx_bd_short_event;
1682 	if (ctrl & EMAC_RX_ST_AE)
1683 		++st->rx_bd_alignment_error;
1684 	if (ctrl & EMAC_RX_ST_BFCS)
1685 		++st->rx_bd_bad_fcs;
1686 	if (ctrl & EMAC_RX_ST_PTL)
1687 		++st->rx_bd_packet_too_long;
1688 	if (ctrl & EMAC_RX_ST_ORE)
1689 		++st->rx_bd_out_of_range;
1690 	if (ctrl & EMAC_RX_ST_IRE)
1691 		++st->rx_bd_in_range;
1692 }
1693 
1694 static inline void emac_rx_csum(struct emac_instance *dev,
1695 				struct sk_buff *skb, u16 ctrl)
1696 {
1697 #ifdef CONFIG_IBM_EMAC_TAH
1698 	if (!ctrl && dev->tah_dev) {
1699 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1700 		++dev->stats.rx_packets_csum;
1701 	}
1702 #endif
1703 }
1704 
1705 static inline int emac_rx_sg_append(struct emac_instance *dev, int slot)
1706 {
1707 	if (likely(dev->rx_sg_skb != NULL)) {
1708 		int len = dev->rx_desc[slot].data_len;
1709 		int tot_len = dev->rx_sg_skb->len + len;
1710 
1711 		if (unlikely(tot_len + NET_IP_ALIGN > dev->rx_skb_size)) {
1712 			++dev->estats.rx_dropped_mtu;
1713 			dev_kfree_skb(dev->rx_sg_skb);
1714 			dev->rx_sg_skb = NULL;
1715 		} else {
1716 			memcpy(skb_tail_pointer(dev->rx_sg_skb),
1717 					 dev->rx_skb[slot]->data, len);
1718 			skb_put(dev->rx_sg_skb, len);
1719 			emac_recycle_rx_skb(dev, slot, len);
1720 			return 0;
1721 		}
1722 	}
1723 	emac_recycle_rx_skb(dev, slot, 0);
1724 	return -1;
1725 }
1726 
1727 /* NAPI poll context */
1728 static int emac_poll_rx(void *param, int budget)
1729 {
1730 	struct emac_instance *dev = param;
1731 	int slot = dev->rx_slot, received = 0;
1732 
1733 	DBG2(dev, "poll_rx(%d)" NL, budget);
1734 
1735  again:
1736 	while (budget > 0) {
1737 		int len;
1738 		struct sk_buff *skb;
1739 		u16 ctrl = dev->rx_desc[slot].ctrl;
1740 
1741 		if (ctrl & MAL_RX_CTRL_EMPTY)
1742 			break;
1743 
1744 		skb = dev->rx_skb[slot];
1745 		mb();
1746 		len = dev->rx_desc[slot].data_len;
1747 
1748 		if (unlikely(!MAL_IS_SINGLE_RX(ctrl)))
1749 			goto sg;
1750 
1751 		ctrl &= EMAC_BAD_RX_MASK;
1752 		if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1753 			emac_parse_rx_error(dev, ctrl);
1754 			++dev->estats.rx_dropped_error;
1755 			emac_recycle_rx_skb(dev, slot, 0);
1756 			len = 0;
1757 			goto next;
1758 		}
1759 
1760 		if (len < ETH_HLEN) {
1761 			++dev->estats.rx_dropped_stack;
1762 			emac_recycle_rx_skb(dev, slot, len);
1763 			goto next;
1764 		}
1765 
1766 		if (len && len < EMAC_RX_COPY_THRESH) {
1767 			struct sk_buff *copy_skb;
1768 
1769 			copy_skb = napi_alloc_skb(&dev->mal->napi, len);
1770 			if (unlikely(!copy_skb))
1771 				goto oom;
1772 
1773 			memcpy(copy_skb->data - NET_IP_ALIGN,
1774 			       skb->data - NET_IP_ALIGN,
1775 			       len + NET_IP_ALIGN);
1776 			emac_recycle_rx_skb(dev, slot, len);
1777 			skb = copy_skb;
1778 		} else if (unlikely(emac_alloc_rx_skb_napi(dev, slot)))
1779 			goto oom;
1780 
1781 		skb_put(skb, len);
1782 	push_packet:
1783 		skb->protocol = eth_type_trans(skb, dev->ndev);
1784 		emac_rx_csum(dev, skb, ctrl);
1785 
1786 		if (unlikely(netif_receive_skb(skb) == NET_RX_DROP))
1787 			++dev->estats.rx_dropped_stack;
1788 	next:
1789 		++dev->stats.rx_packets;
1790 	skip:
1791 		dev->stats.rx_bytes += len;
1792 		slot = (slot + 1) % NUM_RX_BUFF;
1793 		--budget;
1794 		++received;
1795 		continue;
1796 	sg:
1797 		if (ctrl & MAL_RX_CTRL_FIRST) {
1798 			BUG_ON(dev->rx_sg_skb);
1799 			if (unlikely(emac_alloc_rx_skb_napi(dev, slot))) {
1800 				DBG(dev, "rx OOM %d" NL, slot);
1801 				++dev->estats.rx_dropped_oom;
1802 				emac_recycle_rx_skb(dev, slot, 0);
1803 			} else {
1804 				dev->rx_sg_skb = skb;
1805 				skb_put(skb, len);
1806 			}
1807 		} else if (!emac_rx_sg_append(dev, slot) &&
1808 			   (ctrl & MAL_RX_CTRL_LAST)) {
1809 
1810 			skb = dev->rx_sg_skb;
1811 			dev->rx_sg_skb = NULL;
1812 
1813 			ctrl &= EMAC_BAD_RX_MASK;
1814 			if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1815 				emac_parse_rx_error(dev, ctrl);
1816 				++dev->estats.rx_dropped_error;
1817 				dev_kfree_skb(skb);
1818 				len = 0;
1819 			} else
1820 				goto push_packet;
1821 		}
1822 		goto skip;
1823 	oom:
1824 		DBG(dev, "rx OOM %d" NL, slot);
1825 		/* Drop the packet and recycle skb */
1826 		++dev->estats.rx_dropped_oom;
1827 		emac_recycle_rx_skb(dev, slot, 0);
1828 		goto next;
1829 	}
1830 
1831 	if (received) {
1832 		DBG2(dev, "rx %d BDs" NL, received);
1833 		dev->rx_slot = slot;
1834 	}
1835 
1836 	if (unlikely(budget && test_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags))) {
1837 		mb();
1838 		if (!(dev->rx_desc[slot].ctrl & MAL_RX_CTRL_EMPTY)) {
1839 			DBG2(dev, "rx restart" NL);
1840 			received = 0;
1841 			goto again;
1842 		}
1843 
1844 		if (dev->rx_sg_skb) {
1845 			DBG2(dev, "dropping partial rx packet" NL);
1846 			++dev->estats.rx_dropped_error;
1847 			dev_kfree_skb(dev->rx_sg_skb);
1848 			dev->rx_sg_skb = NULL;
1849 		}
1850 
1851 		clear_bit(MAL_COMMAC_RX_STOPPED, &dev->commac.flags);
1852 		mal_enable_rx_channel(dev->mal, dev->mal_rx_chan);
1853 		emac_rx_enable(dev);
1854 		dev->rx_slot = 0;
1855 	}
1856 	return received;
1857 }
1858 
1859 /* NAPI poll context */
1860 static int emac_peek_rx(void *param)
1861 {
1862 	struct emac_instance *dev = param;
1863 
1864 	return !(dev->rx_desc[dev->rx_slot].ctrl & MAL_RX_CTRL_EMPTY);
1865 }
1866 
1867 /* NAPI poll context */
1868 static int emac_peek_rx_sg(void *param)
1869 {
1870 	struct emac_instance *dev = param;
1871 
1872 	int slot = dev->rx_slot;
1873 	while (1) {
1874 		u16 ctrl = dev->rx_desc[slot].ctrl;
1875 		if (ctrl & MAL_RX_CTRL_EMPTY)
1876 			return 0;
1877 		else if (ctrl & MAL_RX_CTRL_LAST)
1878 			return 1;
1879 
1880 		slot = (slot + 1) % NUM_RX_BUFF;
1881 
1882 		/* I'm just being paranoid here :) */
1883 		if (unlikely(slot == dev->rx_slot))
1884 			return 0;
1885 	}
1886 }
1887 
1888 /* Hard IRQ */
1889 static void emac_rxde(void *param)
1890 {
1891 	struct emac_instance *dev = param;
1892 
1893 	++dev->estats.rx_stopped;
1894 	emac_rx_disable_async(dev);
1895 }
1896 
1897 /* Hard IRQ */
1898 static irqreturn_t emac_irq(int irq, void *dev_instance)
1899 {
1900 	struct emac_instance *dev = dev_instance;
1901 	struct emac_regs __iomem *p = dev->emacp;
1902 	struct emac_error_stats *st = &dev->estats;
1903 	u32 isr;
1904 
1905 	spin_lock(&dev->lock);
1906 
1907 	isr = in_be32(&p->isr);
1908 	out_be32(&p->isr, isr);
1909 
1910 	DBG(dev, "isr = %08x" NL, isr);
1911 
1912 	if (isr & EMAC4_ISR_TXPE)
1913 		++st->tx_parity;
1914 	if (isr & EMAC4_ISR_RXPE)
1915 		++st->rx_parity;
1916 	if (isr & EMAC4_ISR_TXUE)
1917 		++st->tx_underrun;
1918 	if (isr & EMAC4_ISR_RXOE)
1919 		++st->rx_fifo_overrun;
1920 	if (isr & EMAC_ISR_OVR)
1921 		++st->rx_overrun;
1922 	if (isr & EMAC_ISR_BP)
1923 		++st->rx_bad_packet;
1924 	if (isr & EMAC_ISR_RP)
1925 		++st->rx_runt_packet;
1926 	if (isr & EMAC_ISR_SE)
1927 		++st->rx_short_event;
1928 	if (isr & EMAC_ISR_ALE)
1929 		++st->rx_alignment_error;
1930 	if (isr & EMAC_ISR_BFCS)
1931 		++st->rx_bad_fcs;
1932 	if (isr & EMAC_ISR_PTLE)
1933 		++st->rx_packet_too_long;
1934 	if (isr & EMAC_ISR_ORE)
1935 		++st->rx_out_of_range;
1936 	if (isr & EMAC_ISR_IRE)
1937 		++st->rx_in_range;
1938 	if (isr & EMAC_ISR_SQE)
1939 		++st->tx_sqe;
1940 	if (isr & EMAC_ISR_TE)
1941 		++st->tx_errors;
1942 
1943 	spin_unlock(&dev->lock);
1944 
1945 	return IRQ_HANDLED;
1946 }
1947 
1948 static struct net_device_stats *emac_stats(struct net_device *ndev)
1949 {
1950 	struct emac_instance *dev = netdev_priv(ndev);
1951 	struct emac_stats *st = &dev->stats;
1952 	struct emac_error_stats *est = &dev->estats;
1953 	struct net_device_stats *nst = &ndev->stats;
1954 	unsigned long flags;
1955 
1956 	DBG2(dev, "stats" NL);
1957 
1958 	/* Compute "legacy" statistics */
1959 	spin_lock_irqsave(&dev->lock, flags);
1960 	nst->rx_packets = (unsigned long)st->rx_packets;
1961 	nst->rx_bytes = (unsigned long)st->rx_bytes;
1962 	nst->tx_packets = (unsigned long)st->tx_packets;
1963 	nst->tx_bytes = (unsigned long)st->tx_bytes;
1964 	nst->rx_dropped = (unsigned long)(est->rx_dropped_oom +
1965 					  est->rx_dropped_error +
1966 					  est->rx_dropped_resize +
1967 					  est->rx_dropped_mtu);
1968 	nst->tx_dropped = (unsigned long)est->tx_dropped;
1969 
1970 	nst->rx_errors = (unsigned long)est->rx_bd_errors;
1971 	nst->rx_fifo_errors = (unsigned long)(est->rx_bd_overrun +
1972 					      est->rx_fifo_overrun +
1973 					      est->rx_overrun);
1974 	nst->rx_frame_errors = (unsigned long)(est->rx_bd_alignment_error +
1975 					       est->rx_alignment_error);
1976 	nst->rx_crc_errors = (unsigned long)(est->rx_bd_bad_fcs +
1977 					     est->rx_bad_fcs);
1978 	nst->rx_length_errors = (unsigned long)(est->rx_bd_runt_packet +
1979 						est->rx_bd_short_event +
1980 						est->rx_bd_packet_too_long +
1981 						est->rx_bd_out_of_range +
1982 						est->rx_bd_in_range +
1983 						est->rx_runt_packet +
1984 						est->rx_short_event +
1985 						est->rx_packet_too_long +
1986 						est->rx_out_of_range +
1987 						est->rx_in_range);
1988 
1989 	nst->tx_errors = (unsigned long)(est->tx_bd_errors + est->tx_errors);
1990 	nst->tx_fifo_errors = (unsigned long)(est->tx_bd_underrun +
1991 					      est->tx_underrun);
1992 	nst->tx_carrier_errors = (unsigned long)est->tx_bd_carrier_loss;
1993 	nst->collisions = (unsigned long)(est->tx_bd_excessive_deferral +
1994 					  est->tx_bd_excessive_collisions +
1995 					  est->tx_bd_late_collision +
1996 					  est->tx_bd_multple_collisions);
1997 	spin_unlock_irqrestore(&dev->lock, flags);
1998 	return nst;
1999 }
2000 
2001 static struct mal_commac_ops emac_commac_ops = {
2002 	.poll_tx = &emac_poll_tx,
2003 	.poll_rx = &emac_poll_rx,
2004 	.peek_rx = &emac_peek_rx,
2005 	.rxde = &emac_rxde,
2006 };
2007 
2008 static struct mal_commac_ops emac_commac_sg_ops = {
2009 	.poll_tx = &emac_poll_tx,
2010 	.poll_rx = &emac_poll_rx,
2011 	.peek_rx = &emac_peek_rx_sg,
2012 	.rxde = &emac_rxde,
2013 };
2014 
2015 /* Ethtool support */
2016 static int emac_ethtool_get_link_ksettings(struct net_device *ndev,
2017 					   struct ethtool_link_ksettings *cmd)
2018 {
2019 	struct emac_instance *dev = netdev_priv(ndev);
2020 	u32 supported, advertising;
2021 
2022 	supported = dev->phy.features;
2023 	cmd->base.port = PORT_MII;
2024 	cmd->base.phy_address = dev->phy.address;
2025 
2026 	mutex_lock(&dev->link_lock);
2027 	advertising = dev->phy.advertising;
2028 	cmd->base.autoneg = dev->phy.autoneg;
2029 	cmd->base.speed = dev->phy.speed;
2030 	cmd->base.duplex = dev->phy.duplex;
2031 	mutex_unlock(&dev->link_lock);
2032 
2033 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
2034 						supported);
2035 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
2036 						advertising);
2037 
2038 	return 0;
2039 }
2040 
2041 static int
2042 emac_ethtool_set_link_ksettings(struct net_device *ndev,
2043 				const struct ethtool_link_ksettings *cmd)
2044 {
2045 	struct emac_instance *dev = netdev_priv(ndev);
2046 	u32 f = dev->phy.features;
2047 	u32 advertising;
2048 
2049 	ethtool_convert_link_mode_to_legacy_u32(&advertising,
2050 						cmd->link_modes.advertising);
2051 
2052 	DBG(dev, "set_settings(%d, %d, %d, 0x%08x)" NL,
2053 	    cmd->base.autoneg, cmd->base.speed, cmd->base.duplex, advertising);
2054 
2055 	/* Basic sanity checks */
2056 	if (dev->phy.address < 0)
2057 		return -EOPNOTSUPP;
2058 	if (cmd->base.autoneg != AUTONEG_ENABLE &&
2059 	    cmd->base.autoneg != AUTONEG_DISABLE)
2060 		return -EINVAL;
2061 	if (cmd->base.autoneg == AUTONEG_ENABLE && advertising == 0)
2062 		return -EINVAL;
2063 	if (cmd->base.duplex != DUPLEX_HALF && cmd->base.duplex != DUPLEX_FULL)
2064 		return -EINVAL;
2065 
2066 	if (cmd->base.autoneg == AUTONEG_DISABLE) {
2067 		switch (cmd->base.speed) {
2068 		case SPEED_10:
2069 			if (cmd->base.duplex == DUPLEX_HALF &&
2070 			    !(f & SUPPORTED_10baseT_Half))
2071 				return -EINVAL;
2072 			if (cmd->base.duplex == DUPLEX_FULL &&
2073 			    !(f & SUPPORTED_10baseT_Full))
2074 				return -EINVAL;
2075 			break;
2076 		case SPEED_100:
2077 			if (cmd->base.duplex == DUPLEX_HALF &&
2078 			    !(f & SUPPORTED_100baseT_Half))
2079 				return -EINVAL;
2080 			if (cmd->base.duplex == DUPLEX_FULL &&
2081 			    !(f & SUPPORTED_100baseT_Full))
2082 				return -EINVAL;
2083 			break;
2084 		case SPEED_1000:
2085 			if (cmd->base.duplex == DUPLEX_HALF &&
2086 			    !(f & SUPPORTED_1000baseT_Half))
2087 				return -EINVAL;
2088 			if (cmd->base.duplex == DUPLEX_FULL &&
2089 			    !(f & SUPPORTED_1000baseT_Full))
2090 				return -EINVAL;
2091 			break;
2092 		default:
2093 			return -EINVAL;
2094 		}
2095 
2096 		mutex_lock(&dev->link_lock);
2097 		dev->phy.def->ops->setup_forced(&dev->phy, cmd->base.speed,
2098 						cmd->base.duplex);
2099 		mutex_unlock(&dev->link_lock);
2100 
2101 	} else {
2102 		if (!(f & SUPPORTED_Autoneg))
2103 			return -EINVAL;
2104 
2105 		mutex_lock(&dev->link_lock);
2106 		dev->phy.def->ops->setup_aneg(&dev->phy,
2107 					      (advertising & f) |
2108 					      (dev->phy.advertising &
2109 					       (ADVERTISED_Pause |
2110 						ADVERTISED_Asym_Pause)));
2111 		mutex_unlock(&dev->link_lock);
2112 	}
2113 	emac_force_link_update(dev);
2114 
2115 	return 0;
2116 }
2117 
2118 static void
2119 emac_ethtool_get_ringparam(struct net_device *ndev,
2120 			   struct ethtool_ringparam *rp,
2121 			   struct kernel_ethtool_ringparam *kernel_rp,
2122 			   struct netlink_ext_ack *extack)
2123 {
2124 	rp->rx_max_pending = rp->rx_pending = NUM_RX_BUFF;
2125 	rp->tx_max_pending = rp->tx_pending = NUM_TX_BUFF;
2126 }
2127 
2128 static void emac_ethtool_get_pauseparam(struct net_device *ndev,
2129 					struct ethtool_pauseparam *pp)
2130 {
2131 	struct emac_instance *dev = netdev_priv(ndev);
2132 
2133 	mutex_lock(&dev->link_lock);
2134 	if ((dev->phy.features & SUPPORTED_Autoneg) &&
2135 	    (dev->phy.advertising & (ADVERTISED_Pause | ADVERTISED_Asym_Pause)))
2136 		pp->autoneg = 1;
2137 
2138 	if (dev->phy.duplex == DUPLEX_FULL) {
2139 		if (dev->phy.pause)
2140 			pp->rx_pause = pp->tx_pause = 1;
2141 		else if (dev->phy.asym_pause)
2142 			pp->tx_pause = 1;
2143 	}
2144 	mutex_unlock(&dev->link_lock);
2145 }
2146 
2147 static int emac_get_regs_len(struct emac_instance *dev)
2148 {
2149 		return sizeof(struct emac_ethtool_regs_subhdr) +
2150 			sizeof(struct emac_regs);
2151 }
2152 
2153 static int emac_ethtool_get_regs_len(struct net_device *ndev)
2154 {
2155 	struct emac_instance *dev = netdev_priv(ndev);
2156 	int size;
2157 
2158 	size = sizeof(struct emac_ethtool_regs_hdr) +
2159 		emac_get_regs_len(dev) + mal_get_regs_len(dev->mal);
2160 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
2161 		size += zmii_get_regs_len(dev->zmii_dev);
2162 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
2163 		size += rgmii_get_regs_len(dev->rgmii_dev);
2164 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
2165 		size += tah_get_regs_len(dev->tah_dev);
2166 
2167 	return size;
2168 }
2169 
2170 static void *emac_dump_regs(struct emac_instance *dev, void *buf)
2171 {
2172 	struct emac_ethtool_regs_subhdr *hdr = buf;
2173 
2174 	hdr->index = dev->cell_index;
2175 	if (emac_has_feature(dev, EMAC_FTR_EMAC4SYNC)) {
2176 		hdr->version = EMAC4SYNC_ETHTOOL_REGS_VER;
2177 	} else if (emac_has_feature(dev, EMAC_FTR_EMAC4)) {
2178 		hdr->version = EMAC4_ETHTOOL_REGS_VER;
2179 	} else {
2180 		hdr->version = EMAC_ETHTOOL_REGS_VER;
2181 	}
2182 	memcpy_fromio(hdr + 1, dev->emacp, sizeof(struct emac_regs));
2183 	return (void *)(hdr + 1) + sizeof(struct emac_regs);
2184 }
2185 
2186 static void emac_ethtool_get_regs(struct net_device *ndev,
2187 				  struct ethtool_regs *regs, void *buf)
2188 {
2189 	struct emac_instance *dev = netdev_priv(ndev);
2190 	struct emac_ethtool_regs_hdr *hdr = buf;
2191 
2192 	hdr->components = 0;
2193 	buf = hdr + 1;
2194 
2195 	buf = mal_dump_regs(dev->mal, buf);
2196 	buf = emac_dump_regs(dev, buf);
2197 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII)) {
2198 		hdr->components |= EMAC_ETHTOOL_REGS_ZMII;
2199 		buf = zmii_dump_regs(dev->zmii_dev, buf);
2200 	}
2201 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII)) {
2202 		hdr->components |= EMAC_ETHTOOL_REGS_RGMII;
2203 		buf = rgmii_dump_regs(dev->rgmii_dev, buf);
2204 	}
2205 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH)) {
2206 		hdr->components |= EMAC_ETHTOOL_REGS_TAH;
2207 		buf = tah_dump_regs(dev->tah_dev, buf);
2208 	}
2209 }
2210 
2211 static int emac_ethtool_nway_reset(struct net_device *ndev)
2212 {
2213 	struct emac_instance *dev = netdev_priv(ndev);
2214 	int res = 0;
2215 
2216 	DBG(dev, "nway_reset" NL);
2217 
2218 	if (dev->phy.address < 0)
2219 		return -EOPNOTSUPP;
2220 
2221 	mutex_lock(&dev->link_lock);
2222 	if (!dev->phy.autoneg) {
2223 		res = -EINVAL;
2224 		goto out;
2225 	}
2226 
2227 	dev->phy.def->ops->setup_aneg(&dev->phy, dev->phy.advertising);
2228  out:
2229 	mutex_unlock(&dev->link_lock);
2230 	emac_force_link_update(dev);
2231 	return res;
2232 }
2233 
2234 static int emac_ethtool_get_sset_count(struct net_device *ndev, int stringset)
2235 {
2236 	if (stringset == ETH_SS_STATS)
2237 		return EMAC_ETHTOOL_STATS_COUNT;
2238 	else
2239 		return -EINVAL;
2240 }
2241 
2242 static void emac_ethtool_get_strings(struct net_device *ndev, u32 stringset,
2243 				     u8 * buf)
2244 {
2245 	if (stringset == ETH_SS_STATS)
2246 		memcpy(buf, &emac_stats_keys, sizeof(emac_stats_keys));
2247 }
2248 
2249 static void emac_ethtool_get_ethtool_stats(struct net_device *ndev,
2250 					   struct ethtool_stats *estats,
2251 					   u64 * tmp_stats)
2252 {
2253 	struct emac_instance *dev = netdev_priv(ndev);
2254 
2255 	memcpy(tmp_stats, &dev->stats, sizeof(dev->stats));
2256 	tmp_stats += sizeof(dev->stats) / sizeof(u64);
2257 	memcpy(tmp_stats, &dev->estats, sizeof(dev->estats));
2258 }
2259 
2260 static void emac_ethtool_get_drvinfo(struct net_device *ndev,
2261 				     struct ethtool_drvinfo *info)
2262 {
2263 	struct emac_instance *dev = netdev_priv(ndev);
2264 
2265 	strscpy(info->driver, "ibm_emac", sizeof(info->driver));
2266 	strscpy(info->version, DRV_VERSION, sizeof(info->version));
2267 	snprintf(info->bus_info, sizeof(info->bus_info), "PPC 4xx EMAC-%d %pOF",
2268 		 dev->cell_index, dev->ofdev->dev.of_node);
2269 }
2270 
2271 static const struct ethtool_ops emac_ethtool_ops = {
2272 	.get_drvinfo = emac_ethtool_get_drvinfo,
2273 
2274 	.get_regs_len = emac_ethtool_get_regs_len,
2275 	.get_regs = emac_ethtool_get_regs,
2276 
2277 	.nway_reset = emac_ethtool_nway_reset,
2278 
2279 	.get_ringparam = emac_ethtool_get_ringparam,
2280 	.get_pauseparam = emac_ethtool_get_pauseparam,
2281 
2282 	.get_strings = emac_ethtool_get_strings,
2283 	.get_sset_count = emac_ethtool_get_sset_count,
2284 	.get_ethtool_stats = emac_ethtool_get_ethtool_stats,
2285 
2286 	.get_link = ethtool_op_get_link,
2287 	.get_link_ksettings = emac_ethtool_get_link_ksettings,
2288 	.set_link_ksettings = emac_ethtool_set_link_ksettings,
2289 };
2290 
2291 static int emac_ioctl(struct net_device *ndev, struct ifreq *rq, int cmd)
2292 {
2293 	struct emac_instance *dev = netdev_priv(ndev);
2294 	struct mii_ioctl_data *data = if_mii(rq);
2295 
2296 	DBG(dev, "ioctl %08x" NL, cmd);
2297 
2298 	if (dev->phy.address < 0)
2299 		return -EOPNOTSUPP;
2300 
2301 	switch (cmd) {
2302 	case SIOCGMIIPHY:
2303 		data->phy_id = dev->phy.address;
2304 		fallthrough;
2305 	case SIOCGMIIREG:
2306 		data->val_out = emac_mdio_read(ndev, dev->phy.address,
2307 					       data->reg_num);
2308 		return 0;
2309 
2310 	case SIOCSMIIREG:
2311 		emac_mdio_write(ndev, dev->phy.address, data->reg_num,
2312 				data->val_in);
2313 		return 0;
2314 	default:
2315 		return -EOPNOTSUPP;
2316 	}
2317 }
2318 
2319 struct emac_depentry {
2320 	u32			phandle;
2321 	struct device_node	*node;
2322 	struct platform_device	*ofdev;
2323 	void			*drvdata;
2324 };
2325 
2326 #define	EMAC_DEP_MAL_IDX	0
2327 #define	EMAC_DEP_ZMII_IDX	1
2328 #define	EMAC_DEP_RGMII_IDX	2
2329 #define	EMAC_DEP_TAH_IDX	3
2330 #define	EMAC_DEP_MDIO_IDX	4
2331 #define	EMAC_DEP_PREV_IDX	5
2332 #define	EMAC_DEP_COUNT		6
2333 
2334 static int emac_check_deps(struct emac_instance *dev,
2335 			   struct emac_depentry *deps)
2336 {
2337 	int i, there = 0;
2338 	struct device_node *np;
2339 
2340 	for (i = 0; i < EMAC_DEP_COUNT; i++) {
2341 		/* no dependency on that item, allright */
2342 		if (deps[i].phandle == 0) {
2343 			there++;
2344 			continue;
2345 		}
2346 		/* special case for blist as the dependency might go away */
2347 		if (i == EMAC_DEP_PREV_IDX) {
2348 			np = *(dev->blist - 1);
2349 			if (np == NULL) {
2350 				deps[i].phandle = 0;
2351 				there++;
2352 				continue;
2353 			}
2354 			if (deps[i].node == NULL)
2355 				deps[i].node = of_node_get(np);
2356 		}
2357 		if (deps[i].node == NULL)
2358 			deps[i].node = of_find_node_by_phandle(deps[i].phandle);
2359 		if (deps[i].node == NULL)
2360 			continue;
2361 		if (deps[i].ofdev == NULL)
2362 			deps[i].ofdev = of_find_device_by_node(deps[i].node);
2363 		if (deps[i].ofdev == NULL)
2364 			continue;
2365 		if (deps[i].drvdata == NULL)
2366 			deps[i].drvdata = platform_get_drvdata(deps[i].ofdev);
2367 		if (deps[i].drvdata != NULL)
2368 			there++;
2369 	}
2370 	if (there != EMAC_DEP_COUNT)
2371 		return -EPROBE_DEFER;
2372 	return 0;
2373 }
2374 
2375 static void emac_put_deps(struct emac_instance *dev)
2376 {
2377 	platform_device_put(dev->mal_dev);
2378 	platform_device_put(dev->zmii_dev);
2379 	platform_device_put(dev->rgmii_dev);
2380 	platform_device_put(dev->mdio_dev);
2381 	platform_device_put(dev->tah_dev);
2382 }
2383 
2384 static int emac_wait_deps(struct emac_instance *dev)
2385 {
2386 	struct emac_depentry deps[EMAC_DEP_COUNT];
2387 	int i, err;
2388 
2389 	memset(&deps, 0, sizeof(deps));
2390 
2391 	deps[EMAC_DEP_MAL_IDX].phandle = dev->mal_ph;
2392 	deps[EMAC_DEP_ZMII_IDX].phandle = dev->zmii_ph;
2393 	deps[EMAC_DEP_RGMII_IDX].phandle = dev->rgmii_ph;
2394 	if (dev->tah_ph)
2395 		deps[EMAC_DEP_TAH_IDX].phandle = dev->tah_ph;
2396 	if (dev->mdio_ph)
2397 		deps[EMAC_DEP_MDIO_IDX].phandle = dev->mdio_ph;
2398 	if (dev->blist && dev->blist > emac_boot_list)
2399 		deps[EMAC_DEP_PREV_IDX].phandle = 0xffffffffu;
2400 	err = emac_check_deps(dev, deps);
2401 	for (i = 0; i < EMAC_DEP_COUNT; i++) {
2402 		of_node_put(deps[i].node);
2403 		if (err)
2404 			platform_device_put(deps[i].ofdev);
2405 	}
2406 	if (!err) {
2407 		dev->mal_dev = deps[EMAC_DEP_MAL_IDX].ofdev;
2408 		dev->zmii_dev = deps[EMAC_DEP_ZMII_IDX].ofdev;
2409 		dev->rgmii_dev = deps[EMAC_DEP_RGMII_IDX].ofdev;
2410 		dev->tah_dev = deps[EMAC_DEP_TAH_IDX].ofdev;
2411 		dev->mdio_dev = deps[EMAC_DEP_MDIO_IDX].ofdev;
2412 	}
2413 	platform_device_put(deps[EMAC_DEP_PREV_IDX].ofdev);
2414 	return err;
2415 }
2416 
2417 static int emac_read_uint_prop(struct device_node *np, const char *name,
2418 			       u32 *val, int fatal)
2419 {
2420 	int err;
2421 
2422 	err = of_property_read_u32(np, name, val);
2423 	if (err) {
2424 		if (fatal)
2425 			pr_err("%pOF: missing %s property", np, name);
2426 		return err;
2427 	}
2428 	return 0;
2429 }
2430 
2431 static void emac_adjust_link(struct net_device *ndev)
2432 {
2433 	struct emac_instance *dev = netdev_priv(ndev);
2434 	struct phy_device *phy = ndev->phydev;
2435 
2436 	dev->phy.autoneg = phy->autoneg;
2437 	dev->phy.speed = phy->speed;
2438 	dev->phy.duplex = phy->duplex;
2439 	dev->phy.pause = phy->pause;
2440 	dev->phy.asym_pause = phy->asym_pause;
2441 	ethtool_convert_link_mode_to_legacy_u32(&dev->phy.advertising,
2442 						phy->advertising);
2443 }
2444 
2445 static int emac_mii_bus_read(struct mii_bus *bus, int addr, int regnum)
2446 {
2447 	int ret = emac_mdio_read(bus->priv, addr, regnum);
2448 	/* This is a workaround for powered down ports/phys.
2449 	 * In the wild, this was seen on the Cisco Meraki MX60(W).
2450 	 * This hardware disables ports as part of the handoff
2451 	 * procedure. Accessing the ports will lead to errors
2452 	 * (-ETIMEDOUT, -EREMOTEIO) that do more harm than good.
2453 	 */
2454 	return ret < 0 ? 0xffff : ret;
2455 }
2456 
2457 static int emac_mii_bus_write(struct mii_bus *bus, int addr,
2458 			      int regnum, u16 val)
2459 {
2460 	emac_mdio_write(bus->priv, addr, regnum, val);
2461 	return 0;
2462 }
2463 
2464 static int emac_mii_bus_reset(struct mii_bus *bus)
2465 {
2466 	struct emac_instance *dev = netdev_priv(bus->priv);
2467 
2468 	return emac_reset(dev);
2469 }
2470 
2471 static int emac_mdio_phy_start_aneg(struct mii_phy *phy,
2472 				    struct phy_device *phy_dev)
2473 {
2474 	phy_dev->autoneg = phy->autoneg;
2475 	phy_dev->speed = phy->speed;
2476 	phy_dev->duplex = phy->duplex;
2477 	ethtool_convert_legacy_u32_to_link_mode(phy_dev->advertising,
2478 						phy->advertising);
2479 	return phy_start_aneg(phy_dev);
2480 }
2481 
2482 static int emac_mdio_setup_aneg(struct mii_phy *phy, u32 advertise)
2483 {
2484 	struct net_device *ndev = phy->dev;
2485 
2486 	phy->autoneg = AUTONEG_ENABLE;
2487 	phy->advertising = advertise;
2488 	return emac_mdio_phy_start_aneg(phy, ndev->phydev);
2489 }
2490 
2491 static int emac_mdio_setup_forced(struct mii_phy *phy, int speed, int fd)
2492 {
2493 	struct net_device *ndev = phy->dev;
2494 
2495 	phy->autoneg = AUTONEG_DISABLE;
2496 	phy->speed = speed;
2497 	phy->duplex = fd;
2498 	return emac_mdio_phy_start_aneg(phy, ndev->phydev);
2499 }
2500 
2501 static int emac_mdio_poll_link(struct mii_phy *phy)
2502 {
2503 	struct net_device *ndev = phy->dev;
2504 	struct emac_instance *dev = netdev_priv(ndev);
2505 	int res;
2506 
2507 	res = phy_read_status(ndev->phydev);
2508 	if (res) {
2509 		dev_err(&dev->ofdev->dev, "link update failed (%d).", res);
2510 		return ethtool_op_get_link(ndev);
2511 	}
2512 
2513 	return ndev->phydev->link;
2514 }
2515 
2516 static int emac_mdio_read_link(struct mii_phy *phy)
2517 {
2518 	struct net_device *ndev = phy->dev;
2519 	struct phy_device *phy_dev = ndev->phydev;
2520 	int res;
2521 
2522 	res = phy_read_status(phy_dev);
2523 	if (res)
2524 		return res;
2525 
2526 	phy->speed = phy_dev->speed;
2527 	phy->duplex = phy_dev->duplex;
2528 	phy->pause = phy_dev->pause;
2529 	phy->asym_pause = phy_dev->asym_pause;
2530 	return 0;
2531 }
2532 
2533 static int emac_mdio_init_phy(struct mii_phy *phy)
2534 {
2535 	struct net_device *ndev = phy->dev;
2536 
2537 	phy_start(ndev->phydev);
2538 	return phy_init_hw(ndev->phydev);
2539 }
2540 
2541 static const struct mii_phy_ops emac_dt_mdio_phy_ops = {
2542 	.init		= emac_mdio_init_phy,
2543 	.setup_aneg	= emac_mdio_setup_aneg,
2544 	.setup_forced	= emac_mdio_setup_forced,
2545 	.poll_link	= emac_mdio_poll_link,
2546 	.read_link	= emac_mdio_read_link,
2547 };
2548 
2549 static int emac_dt_mdio_probe(struct emac_instance *dev)
2550 {
2551 	struct device_node *mii_np;
2552 	struct mii_bus *bus;
2553 	int res;
2554 
2555 	mii_np = of_get_child_by_name(dev->ofdev->dev.of_node, "mdio");
2556 	if (!mii_np) {
2557 		dev_err(&dev->ofdev->dev, "no mdio definition found.");
2558 		return -ENODEV;
2559 	}
2560 
2561 	if (!of_device_is_available(mii_np)) {
2562 		res = -ENODEV;
2563 		goto put_node;
2564 	}
2565 
2566 	bus = devm_mdiobus_alloc(&dev->ofdev->dev);
2567 	if (!bus) {
2568 		res = -ENOMEM;
2569 		goto put_node;
2570 	}
2571 
2572 	bus->priv = dev->ndev;
2573 	bus->parent = dev->ndev->dev.parent;
2574 	bus->name = "emac_mdio";
2575 	bus->read = &emac_mii_bus_read;
2576 	bus->write = &emac_mii_bus_write;
2577 	bus->reset = &emac_mii_bus_reset;
2578 	snprintf(bus->id, MII_BUS_ID_SIZE, "%s", dev->ofdev->name);
2579 	res = devm_of_mdiobus_register(&dev->ofdev->dev, bus, mii_np);
2580 	if (res) {
2581 		dev_err(&dev->ofdev->dev, "cannot register MDIO bus %s (%d)",
2582 			bus->name, res);
2583 	}
2584 
2585  put_node:
2586 	of_node_put(mii_np);
2587 	return res;
2588 }
2589 
2590 static int emac_dt_phy_connect(struct emac_instance *dev,
2591 			       struct device_node *phy_handle)
2592 {
2593 	struct phy_device *phy_dev;
2594 
2595 	dev->phy.def = devm_kzalloc(&dev->ofdev->dev, sizeof(*dev->phy.def),
2596 				    GFP_KERNEL);
2597 	if (!dev->phy.def)
2598 		return -ENOMEM;
2599 
2600 	phy_dev = of_phy_connect(dev->ndev, phy_handle, &emac_adjust_link, 0,
2601 				 dev->phy_mode);
2602 	if (!phy_dev) {
2603 		dev_err(&dev->ofdev->dev, "failed to connect to PHY.\n");
2604 		return -ENODEV;
2605 	}
2606 
2607 	dev->phy.def->phy_id = phy_dev->drv->phy_id;
2608 	dev->phy.def->phy_id_mask = phy_dev->drv->phy_id_mask;
2609 	dev->phy.def->name = phy_dev->drv->name;
2610 	dev->phy.def->ops = &emac_dt_mdio_phy_ops;
2611 	ethtool_convert_link_mode_to_legacy_u32(&dev->phy.features,
2612 						phy_dev->supported);
2613 	dev->phy.address = phy_dev->mdio.addr;
2614 	dev->phy.mode = phy_dev->interface;
2615 	return 0;
2616 }
2617 
2618 static int emac_dt_phy_probe(struct emac_instance *dev)
2619 {
2620 	struct device_node *np = dev->ofdev->dev.of_node;
2621 	struct device_node *phy_handle;
2622 	int res = 1;
2623 
2624 	phy_handle = of_parse_phandle(np, "phy-handle", 0);
2625 
2626 	if (phy_handle) {
2627 		res = emac_dt_mdio_probe(dev);
2628 		if (!res) {
2629 			res = emac_dt_phy_connect(dev, phy_handle);
2630 		}
2631 	}
2632 
2633 	of_node_put(phy_handle);
2634 	return res;
2635 }
2636 
2637 static int emac_init_phy(struct emac_instance *dev)
2638 {
2639 	struct device_node *np = dev->ofdev->dev.of_node;
2640 	struct net_device *ndev = dev->ndev;
2641 	u32 phy_map, adv;
2642 	int i;
2643 
2644 	dev->phy.dev = ndev;
2645 	dev->phy.mode = dev->phy_mode;
2646 
2647 	/* PHY-less configuration. */
2648 	if ((dev->phy_address == 0xffffffff && dev->phy_map == 0xffffffff) ||
2649 	    of_phy_is_fixed_link(np)) {
2650 		emac_reset(dev);
2651 
2652 		/* PHY-less configuration. */
2653 		dev->phy.address = -1;
2654 		dev->phy.features = SUPPORTED_MII;
2655 		if (emac_phy_supports_gige(dev->phy_mode))
2656 			dev->phy.features |= SUPPORTED_1000baseT_Full;
2657 		else
2658 			dev->phy.features |= SUPPORTED_100baseT_Full;
2659 		dev->phy.pause = 1;
2660 
2661 		if (of_phy_is_fixed_link(np)) {
2662 			int res = emac_dt_mdio_probe(dev);
2663 
2664 			if (res)
2665 				return res;
2666 
2667 			res = of_phy_register_fixed_link(np);
2668 			ndev->phydev = of_phy_find_device(np);
2669 			if (res || !ndev->phydev)
2670 				return res ? res : -EINVAL;
2671 			emac_adjust_link(dev->ndev);
2672 			put_device(&ndev->phydev->mdio.dev);
2673 		}
2674 		return 0;
2675 	}
2676 
2677 	mutex_lock(&emac_phy_map_lock);
2678 	phy_map = dev->phy_map | busy_phy_map;
2679 
2680 	DBG(dev, "PHY maps %08x %08x" NL, dev->phy_map, busy_phy_map);
2681 
2682 	dev->phy.mdio_read = emac_mdio_read;
2683 	dev->phy.mdio_write = emac_mdio_write;
2684 
2685 	/* Enable internal clock source */
2686 #ifdef CONFIG_PPC_DCR_NATIVE
2687 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2688 		dcri_clrset(SDR0, SDR0_MFR, 0, SDR0_MFR_ECS);
2689 #endif
2690 	/* PHY clock workaround */
2691 	emac_rx_clk_tx(dev);
2692 
2693 	/* Enable internal clock source on 440GX*/
2694 #ifdef CONFIG_PPC_DCR_NATIVE
2695 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2696 		dcri_clrset(SDR0, SDR0_MFR, 0, SDR0_MFR_ECS);
2697 #endif
2698 	/* Configure EMAC with defaults so we can at least use MDIO
2699 	 * This is needed mostly for 440GX
2700 	 */
2701 	if (emac_phy_gpcs(dev->phy.mode)) {
2702 		/* XXX
2703 		 * Make GPCS PHY address equal to EMAC index.
2704 		 * We probably should take into account busy_phy_map
2705 		 * and/or phy_map here.
2706 		 *
2707 		 * Note that the busy_phy_map is currently global
2708 		 * while it should probably be per-ASIC...
2709 		 */
2710 		dev->phy.gpcs_address = dev->gpcs_address;
2711 		if (dev->phy.gpcs_address == 0xffffffff)
2712 			dev->phy.address = dev->cell_index;
2713 	}
2714 
2715 	emac_configure(dev);
2716 
2717 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII)) {
2718 		int res = emac_dt_phy_probe(dev);
2719 
2720 		switch (res) {
2721 		case 1:
2722 			/* No phy-handle property configured.
2723 			 * Continue with the existing phy probe
2724 			 * and setup code.
2725 			 */
2726 			break;
2727 
2728 		case 0:
2729 			mutex_unlock(&emac_phy_map_lock);
2730 			goto init_phy;
2731 
2732 		default:
2733 			mutex_unlock(&emac_phy_map_lock);
2734 			dev_err(&dev->ofdev->dev, "failed to attach dt phy (%d).\n",
2735 				res);
2736 			return res;
2737 		}
2738 	}
2739 
2740 	if (dev->phy_address != 0xffffffff)
2741 		phy_map = ~(1 << dev->phy_address);
2742 
2743 	for (i = 0; i < 0x20; phy_map >>= 1, ++i)
2744 		if (!(phy_map & 1)) {
2745 			int r;
2746 			busy_phy_map |= 1 << i;
2747 
2748 			/* Quick check if there is a PHY at the address */
2749 			r = emac_mdio_read(dev->ndev, i, MII_BMCR);
2750 			if (r == 0xffff || r < 0)
2751 				continue;
2752 			if (!emac_mii_phy_probe(&dev->phy, i))
2753 				break;
2754 		}
2755 
2756 	/* Enable external clock source */
2757 #ifdef CONFIG_PPC_DCR_NATIVE
2758 	if (emac_has_feature(dev, EMAC_FTR_440GX_PHY_CLK_FIX))
2759 		dcri_clrset(SDR0, SDR0_MFR, SDR0_MFR_ECS, 0);
2760 #endif
2761 	mutex_unlock(&emac_phy_map_lock);
2762 	if (i == 0x20) {
2763 		printk(KERN_WARNING "%pOF: can't find PHY!\n", np);
2764 		return -ENXIO;
2765 	}
2766 
2767  init_phy:
2768 	/* Init PHY */
2769 	if (dev->phy.def->ops->init)
2770 		dev->phy.def->ops->init(&dev->phy);
2771 
2772 	/* Disable any PHY features not supported by the platform */
2773 	dev->phy.def->features &= ~dev->phy_feat_exc;
2774 	dev->phy.features &= ~dev->phy_feat_exc;
2775 
2776 	/* Setup initial link parameters */
2777 	if (dev->phy.features & SUPPORTED_Autoneg) {
2778 		adv = dev->phy.features;
2779 		if (!emac_has_feature(dev, EMAC_FTR_NO_FLOW_CONTROL_40x))
2780 			adv |= ADVERTISED_Pause | ADVERTISED_Asym_Pause;
2781 		/* Restart autonegotiation */
2782 		dev->phy.def->ops->setup_aneg(&dev->phy, adv);
2783 	} else {
2784 		u32 f = dev->phy.def->features;
2785 		int speed = SPEED_10, fd = DUPLEX_HALF;
2786 
2787 		/* Select highest supported speed/duplex */
2788 		if (f & SUPPORTED_1000baseT_Full) {
2789 			speed = SPEED_1000;
2790 			fd = DUPLEX_FULL;
2791 		} else if (f & SUPPORTED_1000baseT_Half)
2792 			speed = SPEED_1000;
2793 		else if (f & SUPPORTED_100baseT_Full) {
2794 			speed = SPEED_100;
2795 			fd = DUPLEX_FULL;
2796 		} else if (f & SUPPORTED_100baseT_Half)
2797 			speed = SPEED_100;
2798 		else if (f & SUPPORTED_10baseT_Full)
2799 			fd = DUPLEX_FULL;
2800 
2801 		/* Force link parameters */
2802 		dev->phy.def->ops->setup_forced(&dev->phy, speed, fd);
2803 	}
2804 	return 0;
2805 }
2806 
2807 static int emac_init_config(struct emac_instance *dev)
2808 {
2809 	struct device_node *np = dev->ofdev->dev.of_node;
2810 	int err;
2811 
2812 	/* Read config from device-tree */
2813 	if (emac_read_uint_prop(np, "mal-device", &dev->mal_ph, 1))
2814 		return -ENXIO;
2815 	if (emac_read_uint_prop(np, "mal-tx-channel", &dev->mal_tx_chan, 1))
2816 		return -ENXIO;
2817 	if (emac_read_uint_prop(np, "mal-rx-channel", &dev->mal_rx_chan, 1))
2818 		return -ENXIO;
2819 	if (emac_read_uint_prop(np, "cell-index", &dev->cell_index, 1))
2820 		return -ENXIO;
2821 	if (emac_read_uint_prop(np, "max-frame-size", &dev->max_mtu, 0))
2822 		dev->max_mtu = ETH_DATA_LEN;
2823 	if (emac_read_uint_prop(np, "rx-fifo-size", &dev->rx_fifo_size, 0))
2824 		dev->rx_fifo_size = 2048;
2825 	if (emac_read_uint_prop(np, "tx-fifo-size", &dev->tx_fifo_size, 0))
2826 		dev->tx_fifo_size = 2048;
2827 	if (emac_read_uint_prop(np, "rx-fifo-size-gige", &dev->rx_fifo_size_gige, 0))
2828 		dev->rx_fifo_size_gige = dev->rx_fifo_size;
2829 	if (emac_read_uint_prop(np, "tx-fifo-size-gige", &dev->tx_fifo_size_gige, 0))
2830 		dev->tx_fifo_size_gige = dev->tx_fifo_size;
2831 	if (emac_read_uint_prop(np, "phy-address", &dev->phy_address, 0))
2832 		dev->phy_address = 0xffffffff;
2833 	if (emac_read_uint_prop(np, "phy-map", &dev->phy_map, 0))
2834 		dev->phy_map = 0xffffffff;
2835 	if (emac_read_uint_prop(np, "gpcs-address", &dev->gpcs_address, 0))
2836 		dev->gpcs_address = 0xffffffff;
2837 	if (emac_read_uint_prop(np->parent, "clock-frequency", &dev->opb_bus_freq, 1))
2838 		return -ENXIO;
2839 	if (emac_read_uint_prop(np, "tah-device", &dev->tah_ph, 0))
2840 		dev->tah_ph = 0;
2841 	if (emac_read_uint_prop(np, "tah-channel", &dev->tah_port, 0))
2842 		dev->tah_port = 0;
2843 	if (emac_read_uint_prop(np, "mdio-device", &dev->mdio_ph, 0))
2844 		dev->mdio_ph = 0;
2845 	if (emac_read_uint_prop(np, "zmii-device", &dev->zmii_ph, 0))
2846 		dev->zmii_ph = 0;
2847 	if (emac_read_uint_prop(np, "zmii-channel", &dev->zmii_port, 0))
2848 		dev->zmii_port = 0xffffffff;
2849 	if (emac_read_uint_prop(np, "rgmii-device", &dev->rgmii_ph, 0))
2850 		dev->rgmii_ph = 0;
2851 	if (emac_read_uint_prop(np, "rgmii-channel", &dev->rgmii_port, 0))
2852 		dev->rgmii_port = 0xffffffff;
2853 	if (emac_read_uint_prop(np, "fifo-entry-size", &dev->fifo_entry_size, 0))
2854 		dev->fifo_entry_size = 16;
2855 	if (emac_read_uint_prop(np, "mal-burst-size", &dev->mal_burst_size, 0))
2856 		dev->mal_burst_size = 256;
2857 
2858 	/* PHY mode needs some decoding */
2859 	err = of_get_phy_mode(np, &dev->phy_mode);
2860 	if (err)
2861 		dev->phy_mode = PHY_INTERFACE_MODE_NA;
2862 
2863 	/* Check EMAC version */
2864 	if (of_device_is_compatible(np, "ibm,emac4sync")) {
2865 		dev->features |= (EMAC_FTR_EMAC4 | EMAC_FTR_EMAC4SYNC);
2866 		if (of_device_is_compatible(np, "ibm,emac-460ex") ||
2867 		    of_device_is_compatible(np, "ibm,emac-460gt"))
2868 			dev->features |= EMAC_FTR_460EX_PHY_CLK_FIX;
2869 		if (of_device_is_compatible(np, "ibm,emac-405ex") ||
2870 		    of_device_is_compatible(np, "ibm,emac-405exr"))
2871 			dev->features |= EMAC_FTR_440EP_PHY_CLK_FIX;
2872 		if (of_device_is_compatible(np, "ibm,emac-apm821xx")) {
2873 			dev->features |= (EMAC_APM821XX_REQ_JUMBO_FRAME_SIZE |
2874 					  EMAC_FTR_APM821XX_NO_HALF_DUPLEX |
2875 					  EMAC_FTR_460EX_PHY_CLK_FIX);
2876 		}
2877 	} else if (of_device_is_compatible(np, "ibm,emac4")) {
2878 		dev->features |= EMAC_FTR_EMAC4;
2879 		if (of_device_is_compatible(np, "ibm,emac-440gx"))
2880 			dev->features |= EMAC_FTR_440GX_PHY_CLK_FIX;
2881 	} else {
2882 		if (of_device_is_compatible(np, "ibm,emac-440ep") ||
2883 		    of_device_is_compatible(np, "ibm,emac-440gr"))
2884 			dev->features |= EMAC_FTR_440EP_PHY_CLK_FIX;
2885 		if (of_device_is_compatible(np, "ibm,emac-405ez")) {
2886 #ifdef CONFIG_IBM_EMAC_NO_FLOW_CTRL
2887 			dev->features |= EMAC_FTR_NO_FLOW_CONTROL_40x;
2888 #else
2889 			printk(KERN_ERR "%pOF: Flow control not disabled!\n",
2890 					np);
2891 			return -ENXIO;
2892 #endif
2893 		}
2894 
2895 	}
2896 
2897 	/* Fixup some feature bits based on the device tree */
2898 	if (of_property_read_bool(np, "has-inverted-stacr-oc"))
2899 		dev->features |= EMAC_FTR_STACR_OC_INVERT;
2900 	if (of_property_read_bool(np, "has-new-stacr-staopc"))
2901 		dev->features |= EMAC_FTR_HAS_NEW_STACR;
2902 
2903 	/* CAB lacks the appropriate properties */
2904 	if (of_device_is_compatible(np, "ibm,emac-axon"))
2905 		dev->features |= EMAC_FTR_HAS_NEW_STACR |
2906 			EMAC_FTR_STACR_OC_INVERT;
2907 
2908 	/* Enable TAH/ZMII/RGMII features as found */
2909 	if (dev->tah_ph != 0) {
2910 #ifdef CONFIG_IBM_EMAC_TAH
2911 		dev->features |= EMAC_FTR_HAS_TAH;
2912 #else
2913 		printk(KERN_ERR "%pOF: TAH support not enabled !\n", np);
2914 		return -ENXIO;
2915 #endif
2916 	}
2917 
2918 	if (dev->zmii_ph != 0) {
2919 #ifdef CONFIG_IBM_EMAC_ZMII
2920 		dev->features |= EMAC_FTR_HAS_ZMII;
2921 #else
2922 		printk(KERN_ERR "%pOF: ZMII support not enabled !\n", np);
2923 		return -ENXIO;
2924 #endif
2925 	}
2926 
2927 	if (dev->rgmii_ph != 0) {
2928 #ifdef CONFIG_IBM_EMAC_RGMII
2929 		dev->features |= EMAC_FTR_HAS_RGMII;
2930 #else
2931 		printk(KERN_ERR "%pOF: RGMII support not enabled !\n", np);
2932 		return -ENXIO;
2933 #endif
2934 	}
2935 
2936 	/* Read MAC-address */
2937 	err = of_get_ethdev_address(np, dev->ndev);
2938 	if (err)
2939 		return dev_err_probe(&dev->ofdev->dev, err,
2940 				     "Can't get valid [local-]mac-address from OF !\n");
2941 
2942 	/* IAHT and GAHT filter parameterization */
2943 	if (emac_has_feature(dev, EMAC_FTR_EMAC4SYNC)) {
2944 		dev->xaht_slots_shift = EMAC4SYNC_XAHT_SLOTS_SHIFT;
2945 		dev->xaht_width_shift = EMAC4SYNC_XAHT_WIDTH_SHIFT;
2946 	} else {
2947 		dev->xaht_slots_shift = EMAC4_XAHT_SLOTS_SHIFT;
2948 		dev->xaht_width_shift = EMAC4_XAHT_WIDTH_SHIFT;
2949 	}
2950 
2951 	/* This should never happen */
2952 	if (WARN_ON(EMAC_XAHT_REGS(dev) > EMAC_XAHT_MAX_REGS))
2953 		return -ENXIO;
2954 
2955 	DBG(dev, "features     : 0x%08x / 0x%08x\n", dev->features, EMAC_FTRS_POSSIBLE);
2956 	DBG(dev, "tx_fifo_size : %d (%d gige)\n", dev->tx_fifo_size, dev->tx_fifo_size_gige);
2957 	DBG(dev, "rx_fifo_size : %d (%d gige)\n", dev->rx_fifo_size, dev->rx_fifo_size_gige);
2958 	DBG(dev, "max_mtu      : %d\n", dev->max_mtu);
2959 	DBG(dev, "OPB freq     : %d\n", dev->opb_bus_freq);
2960 
2961 	return 0;
2962 }
2963 
2964 static const struct net_device_ops emac_netdev_ops = {
2965 	.ndo_open		= emac_open,
2966 	.ndo_stop		= emac_close,
2967 	.ndo_get_stats		= emac_stats,
2968 	.ndo_set_rx_mode	= emac_set_multicast_list,
2969 	.ndo_eth_ioctl		= emac_ioctl,
2970 	.ndo_tx_timeout		= emac_tx_timeout,
2971 	.ndo_validate_addr	= eth_validate_addr,
2972 	.ndo_set_mac_address	= emac_set_mac_address,
2973 	.ndo_start_xmit		= emac_start_xmit,
2974 };
2975 
2976 static const struct net_device_ops emac_gige_netdev_ops = {
2977 	.ndo_open		= emac_open,
2978 	.ndo_stop		= emac_close,
2979 	.ndo_get_stats		= emac_stats,
2980 	.ndo_set_rx_mode	= emac_set_multicast_list,
2981 	.ndo_eth_ioctl		= emac_ioctl,
2982 	.ndo_tx_timeout		= emac_tx_timeout,
2983 	.ndo_validate_addr	= eth_validate_addr,
2984 	.ndo_set_mac_address	= emac_set_mac_address,
2985 	.ndo_start_xmit		= emac_start_xmit_sg,
2986 	.ndo_change_mtu		= emac_change_mtu,
2987 };
2988 
2989 static int emac_probe(struct platform_device *ofdev)
2990 {
2991 	struct net_device *ndev;
2992 	struct emac_instance *dev;
2993 	struct device_node *np = ofdev->dev.of_node;
2994 	struct device_node **blist = NULL;
2995 	int err, i;
2996 
2997 	/* Skip unused/unwired EMACS.  We leave the check for an unused
2998 	 * property here for now, but new flat device trees should set a
2999 	 * status property to "disabled" instead.
3000 	 */
3001 	if (of_property_read_bool(np, "unused") || !of_device_is_available(np))
3002 		return -ENODEV;
3003 
3004 	/* Find ourselves in the bootlist if we are there */
3005 	for (i = 0; i < EMAC_BOOT_LIST_SIZE; i++)
3006 		if (emac_boot_list[i] == np)
3007 			blist = &emac_boot_list[i];
3008 
3009 	/* Allocate our net_device structure */
3010 	err = -ENOMEM;
3011 	ndev = devm_alloc_etherdev(&ofdev->dev, sizeof(struct emac_instance));
3012 	if (!ndev)
3013 		goto err_gone;
3014 
3015 	dev = netdev_priv(ndev);
3016 	dev->ndev = ndev;
3017 	dev->ofdev = ofdev;
3018 	dev->blist = blist;
3019 	SET_NETDEV_DEV(ndev, &ofdev->dev);
3020 
3021 	/* Initialize some embedded data structures */
3022 	mutex_init(&dev->mdio_lock);
3023 	mutex_init(&dev->link_lock);
3024 	spin_lock_init(&dev->lock);
3025 	INIT_WORK(&dev->reset_work, emac_reset_work);
3026 
3027 	/* Init various config data based on device-tree */
3028 	err = emac_init_config(dev);
3029 	if (err)
3030 		goto err_gone;
3031 
3032 	/* Get interrupts. EMAC irq is mandatory */
3033 	dev->emac_irq = irq_of_parse_and_map(np, 0);
3034 	if (!dev->emac_irq) {
3035 		printk(KERN_ERR "%pOF: Can't map main interrupt\n", np);
3036 		err = -ENODEV;
3037 		goto err_gone;
3038 	}
3039 
3040 	/* Setup error IRQ handler */
3041 	err = devm_request_irq(&ofdev->dev, dev->emac_irq, emac_irq, 0, "EMAC",
3042 			       dev);
3043 	if (err) {
3044 		dev_err_probe(&ofdev->dev, err, "failed to request IRQ %d",
3045 			      dev->emac_irq);
3046 		goto err_gone;
3047 	}
3048 
3049 	ndev->irq = dev->emac_irq;
3050 
3051 	/* Map EMAC regs */
3052 	// TODO : platform_get_resource() and devm_ioremap_resource()
3053 	dev->emacp = devm_of_iomap(&ofdev->dev, np, 0, NULL);
3054 	if (!dev->emacp) {
3055 		dev_err(&ofdev->dev, "can't map device registers");
3056 		err = -ENOMEM;
3057 		goto err_gone;
3058 	}
3059 
3060 	/* Wait for dependent devices */
3061 	err = emac_wait_deps(dev);
3062 	if (err)
3063 		goto err_gone;
3064 	dev->mal = platform_get_drvdata(dev->mal_dev);
3065 	if (dev->mdio_dev != NULL)
3066 		dev->mdio_instance = platform_get_drvdata(dev->mdio_dev);
3067 
3068 	/* Register with MAL */
3069 	dev->commac.ops = &emac_commac_ops;
3070 	dev->commac.dev = dev;
3071 	dev->commac.tx_chan_mask = MAL_CHAN_MASK(dev->mal_tx_chan);
3072 	dev->commac.rx_chan_mask = MAL_CHAN_MASK(dev->mal_rx_chan);
3073 	err = mal_register_commac(dev->mal, &dev->commac);
3074 	if (err) {
3075 		printk(KERN_ERR "%pOF: failed to register with mal %pOF!\n",
3076 		       np, dev->mal_dev->dev.of_node);
3077 		goto err_rel_deps;
3078 	}
3079 	dev->rx_skb_size = emac_rx_skb_size(ndev->mtu);
3080 	dev->rx_sync_size = emac_rx_sync_size(ndev->mtu);
3081 
3082 	/* Get pointers to BD rings */
3083 	dev->tx_desc =
3084 	    dev->mal->bd_virt + mal_tx_bd_offset(dev->mal, dev->mal_tx_chan);
3085 	dev->rx_desc =
3086 	    dev->mal->bd_virt + mal_rx_bd_offset(dev->mal, dev->mal_rx_chan);
3087 
3088 	DBG(dev, "tx_desc %p" NL, dev->tx_desc);
3089 	DBG(dev, "rx_desc %p" NL, dev->rx_desc);
3090 
3091 	/* Clean rings */
3092 	memset(dev->tx_desc, 0, NUM_TX_BUFF * sizeof(struct mal_descriptor));
3093 	memset(dev->rx_desc, 0, NUM_RX_BUFF * sizeof(struct mal_descriptor));
3094 	memset(dev->tx_skb, 0, NUM_TX_BUFF * sizeof(struct sk_buff *));
3095 	memset(dev->rx_skb, 0, NUM_RX_BUFF * sizeof(struct sk_buff *));
3096 
3097 	/* Attach to ZMII, if needed */
3098 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII) &&
3099 	    (err = zmii_attach(dev->zmii_dev, dev->zmii_port, &dev->phy_mode)) != 0)
3100 		goto err_unreg_commac;
3101 
3102 	/* Attach to RGMII, if needed */
3103 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII) &&
3104 	    (err = rgmii_attach(dev->rgmii_dev, dev->rgmii_port, dev->phy_mode)) != 0)
3105 		goto err_detach_zmii;
3106 
3107 	/* Attach to TAH, if needed */
3108 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH) &&
3109 	    (err = tah_attach(dev->tah_dev, dev->tah_port)) != 0)
3110 		goto err_detach_rgmii;
3111 
3112 	/* Set some link defaults before we can find out real parameters */
3113 	dev->phy.speed = SPEED_100;
3114 	dev->phy.duplex = DUPLEX_FULL;
3115 	dev->phy.autoneg = AUTONEG_DISABLE;
3116 	dev->phy.pause = dev->phy.asym_pause = 0;
3117 	dev->stop_timeout = STOP_TIMEOUT_100;
3118 	INIT_DELAYED_WORK(&dev->link_work, emac_link_timer);
3119 
3120 	/* Some SoCs like APM821xx does not support Half Duplex mode. */
3121 	if (emac_has_feature(dev, EMAC_FTR_APM821XX_NO_HALF_DUPLEX)) {
3122 		dev->phy_feat_exc = (SUPPORTED_1000baseT_Half |
3123 				     SUPPORTED_100baseT_Half |
3124 				     SUPPORTED_10baseT_Half);
3125 	}
3126 
3127 	/* Find PHY if any */
3128 	err = emac_init_phy(dev);
3129 	if (err != 0)
3130 		goto err_detach_tah;
3131 
3132 	if (dev->tah_dev) {
3133 		ndev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG;
3134 		ndev->features |= ndev->hw_features | NETIF_F_RXCSUM;
3135 	}
3136 	ndev->watchdog_timeo = 5 * HZ;
3137 	if (emac_phy_supports_gige(dev->phy_mode)) {
3138 		ndev->netdev_ops = &emac_gige_netdev_ops;
3139 		dev->commac.ops = &emac_commac_sg_ops;
3140 	} else
3141 		ndev->netdev_ops = &emac_netdev_ops;
3142 	ndev->ethtool_ops = &emac_ethtool_ops;
3143 
3144 	/* MTU range: 46 - 1500 or whatever is in OF */
3145 	ndev->min_mtu = EMAC_MIN_MTU;
3146 	ndev->max_mtu = dev->max_mtu;
3147 
3148 	netif_carrier_off(ndev);
3149 
3150 	err = devm_register_netdev(&ofdev->dev, ndev);
3151 	if (err) {
3152 		printk(KERN_ERR "%pOF: failed to register net device (%d)!\n",
3153 		       np, err);
3154 		goto err_detach_tah;
3155 	}
3156 
3157 	/* Set our drvdata last as we don't want them visible until we are
3158 	 * fully initialized
3159 	 */
3160 	wmb();
3161 	platform_set_drvdata(ofdev, dev);
3162 
3163 	printk(KERN_INFO "%s: EMAC-%d %pOF, MAC %pM\n",
3164 	       ndev->name, dev->cell_index, np, ndev->dev_addr);
3165 
3166 	if (dev->phy_mode == PHY_INTERFACE_MODE_SGMII)
3167 		printk(KERN_NOTICE "%s: in SGMII mode\n", ndev->name);
3168 
3169 	if (dev->phy.address >= 0)
3170 		printk("%s: found %s PHY (0x%02x)\n", ndev->name,
3171 		       dev->phy.def->name, dev->phy.address);
3172 
3173 	/* Life is good */
3174 	return 0;
3175 
3176 	/* I have a bad feeling about this ... */
3177 
3178  err_detach_tah:
3179 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
3180 		tah_detach(dev->tah_dev, dev->tah_port);
3181  err_detach_rgmii:
3182 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
3183 		rgmii_detach(dev->rgmii_dev, dev->rgmii_port);
3184  err_detach_zmii:
3185 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
3186 		zmii_detach(dev->zmii_dev, dev->zmii_port);
3187  err_unreg_commac:
3188 	mal_unregister_commac(dev->mal, &dev->commac);
3189  err_rel_deps:
3190 	emac_put_deps(dev);
3191  err_gone:
3192 	if (blist)
3193 		*blist = NULL;
3194 	return err;
3195 }
3196 
3197 static void emac_remove(struct platform_device *ofdev)
3198 {
3199 	struct emac_instance *dev = platform_get_drvdata(ofdev);
3200 
3201 	DBG(dev, "remove" NL);
3202 
3203 	cancel_work_sync(&dev->reset_work);
3204 
3205 	if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
3206 		tah_detach(dev->tah_dev, dev->tah_port);
3207 	if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
3208 		rgmii_detach(dev->rgmii_dev, dev->rgmii_port);
3209 	if (emac_has_feature(dev, EMAC_FTR_HAS_ZMII))
3210 		zmii_detach(dev->zmii_dev, dev->zmii_port);
3211 
3212 	busy_phy_map &= ~(1 << dev->phy.address);
3213 	DBG(dev, "busy_phy_map now %#x" NL, busy_phy_map);
3214 
3215 	mal_unregister_commac(dev->mal, &dev->commac);
3216 	emac_put_deps(dev);
3217 }
3218 
3219 /* XXX Features in here should be replaced by properties... */
3220 static const struct of_device_id emac_match[] =
3221 {
3222 	{
3223 		.type		= "network",
3224 		.compatible	= "ibm,emac",
3225 	},
3226 	{
3227 		.type		= "network",
3228 		.compatible	= "ibm,emac4",
3229 	},
3230 	{
3231 		.type		= "network",
3232 		.compatible	= "ibm,emac4sync",
3233 	},
3234 	{},
3235 };
3236 MODULE_DEVICE_TABLE(of, emac_match);
3237 
3238 static struct platform_driver emac_driver = {
3239 	.driver = {
3240 		.name = "emac",
3241 		.of_match_table = emac_match,
3242 	},
3243 	.probe = emac_probe,
3244 	.remove_new = emac_remove,
3245 };
3246 
3247 static void __init emac_make_bootlist(void)
3248 {
3249 	struct device_node *np = NULL;
3250 	int j, max, i = 0;
3251 	int cell_indices[EMAC_BOOT_LIST_SIZE];
3252 
3253 	/* Collect EMACs */
3254 	while((np = of_find_all_nodes(np)) != NULL) {
3255 		u32 idx;
3256 
3257 		if (of_match_node(emac_match, np) == NULL)
3258 			continue;
3259 		if (of_property_read_bool(np, "unused"))
3260 			continue;
3261 		if (of_property_read_u32(np, "cell-index", &idx))
3262 			continue;
3263 		cell_indices[i] = idx;
3264 		emac_boot_list[i++] = of_node_get(np);
3265 		if (i >= EMAC_BOOT_LIST_SIZE) {
3266 			of_node_put(np);
3267 			break;
3268 		}
3269 	}
3270 	max = i;
3271 
3272 	/* Bubble sort them (doh, what a creative algorithm :-) */
3273 	for (i = 0; max > 1 && (i < (max - 1)); i++)
3274 		for (j = i; j < max; j++) {
3275 			if (cell_indices[i] > cell_indices[j]) {
3276 				swap(emac_boot_list[i], emac_boot_list[j]);
3277 				swap(cell_indices[i], cell_indices[j]);
3278 			}
3279 		}
3280 }
3281 
3282 static int __init emac_init(void)
3283 {
3284 	int rc;
3285 
3286 	printk(KERN_INFO DRV_DESC ", version " DRV_VERSION "\n");
3287 
3288 	/* Build EMAC boot list */
3289 	emac_make_bootlist();
3290 
3291 	/* Init submodules */
3292 	rc = mal_init();
3293 	if (rc)
3294 		goto err;
3295 	rc = zmii_init();
3296 	if (rc)
3297 		goto err_mal;
3298 	rc = rgmii_init();
3299 	if (rc)
3300 		goto err_zmii;
3301 	rc = tah_init();
3302 	if (rc)
3303 		goto err_rgmii;
3304 	rc = platform_driver_register(&emac_driver);
3305 	if (rc)
3306 		goto err_tah;
3307 
3308 	return 0;
3309 
3310  err_tah:
3311 	tah_exit();
3312  err_rgmii:
3313 	rgmii_exit();
3314  err_zmii:
3315 	zmii_exit();
3316  err_mal:
3317 	mal_exit();
3318  err:
3319 	return rc;
3320 }
3321 
3322 static void __exit emac_exit(void)
3323 {
3324 	int i;
3325 
3326 	platform_driver_unregister(&emac_driver);
3327 
3328 	tah_exit();
3329 	rgmii_exit();
3330 	zmii_exit();
3331 	mal_exit();
3332 
3333 	/* Destroy EMAC boot list */
3334 	for (i = 0; i < EMAC_BOOT_LIST_SIZE; i++)
3335 		of_node_put(emac_boot_list[i]);
3336 }
3337 
3338 module_init(emac_init);
3339 module_exit(emac_exit);
3340