b1ad62d89eb982cb525b2358227e7bf98f595ca5
[powerpc.git] / drivers / net / ibm_emac / ibm_emac_core.c
1 /*
2  * drivers/net/ibm_emac/ibm_emac_core.c
3  *
4  * Driver for PowerPC 4xx on-chip ethernet controller.
5  *
6  * Copyright (c) 2004, 2005 Zultys Technologies.
7  * Eugene Surovegin <eugene.surovegin@zultys.com> or <ebs@ebshome.net>
8  *
9  * Based on original work by
10  *      Matt Porter <mporter@kernel.crashing.org>
11  *      (c) 2003 Benjamin Herrenschmidt <benh@kernel.crashing.org>
12  *      Armin Kuster <akuster@mvista.com>
13  *      Johnnie Peters <jpeters@mvista.com>
14  *
15  * This program is free software; you can redistribute  it and/or modify it
16  * under  the terms of  the GNU General  Public License as published by the
17  * Free Software Foundation;  either version 2 of the  License, or (at your
18  * option) any later version.
19  *
20  */
21
22 #include <linux/module.h>
23 #include <linux/kernel.h>
24 #include <linux/string.h>
25 #include <linux/errno.h>
26 #include <linux/interrupt.h>
27 #include <linux/delay.h>
28 #include <linux/init.h>
29 #include <linux/types.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/etherdevice.h>
33 #include <linux/skbuff.h>
34 #include <linux/crc32.h>
35 #include <linux/ethtool.h>
36 #include <linux/mii.h>
37 #include <linux/bitops.h>
38
39 #include <asm/processor.h>
40 #include <asm/io.h>
41 #include <asm/dma.h>
42 #include <asm/uaccess.h>
43 #include <asm/ocp.h>
44
45 #include "ibm_emac_core.h"
46 #include "ibm_emac_debug.h"
47
48 /*
49  * Lack of dma_unmap_???? calls is intentional.
50  *
51  * API-correct usage requires additional support state information to be 
52  * maintained for every RX and TX buffer descriptor (BD). Unfortunately, due to
53  * EMAC design (e.g. TX buffer passed from network stack can be split into
54  * several BDs, dma_map_single/dma_map_page can be used to map particular BD),
55  * maintaining such information will add additional overhead.
56  * Current DMA API implementation for 4xx processors only ensures cache coherency
57  * and dma_unmap_???? routines are empty and are likely to stay this way.
58  * I decided to omit dma_unmap_??? calls because I don't want to add additional
59  * complexity just for the sake of following some abstract API, when it doesn't
60  * add any real benefit to the driver. I understand that this decision maybe 
61  * controversial, but I really tried to make code API-correct and efficient 
62  * at the same time and didn't come up with code I liked :(.                --ebs
63  */
64
65 #define DRV_NAME        "emac"
66 #define DRV_VERSION     "3.54"
67 #define DRV_DESC        "PPC 4xx OCP EMAC driver"
68
69 MODULE_DESCRIPTION(DRV_DESC);
70 MODULE_AUTHOR
71     ("Eugene Surovegin <eugene.surovegin@zultys.com> or <ebs@ebshome.net>");
72 MODULE_LICENSE("GPL");
73
74 /* minimum number of free TX descriptors required to wake up TX process */
75 #define EMAC_TX_WAKEUP_THRESH           (NUM_TX_BUFF / 4)
76
77 /* If packet size is less than this number, we allocate small skb and copy packet 
78  * contents into it instead of just sending original big skb up
79  */
80 #define EMAC_RX_COPY_THRESH             CONFIG_IBM_EMAC_RX_COPY_THRESHOLD
81
82 /* Since multiple EMACs share MDIO lines in various ways, we need
83  * to avoid re-using the same PHY ID in cases where the arch didn't
84  * setup precise phy_map entries
85  */
86 static u32 busy_phy_map;
87
88 #if defined(CONFIG_IBM_EMAC_PHY_RX_CLK_FIX) && \
89     (defined(CONFIG_405EP) || defined(CONFIG_440EP) || defined(CONFIG_440GR))
90 /* 405EP has "EMAC to PHY Control Register" (CPC0_EPCTL) which can help us
91  * with PHY RX clock problem.
92  * 440EP/440GR has more sane SDR0_MFR register implementation than 440GX, which
93  * also allows controlling each EMAC clock
94  */
95 static inline void EMAC_RX_CLK_TX(int idx)
96 {
97         unsigned long flags;
98         local_irq_save(flags);
99
100 #if defined(CONFIG_405EP)
101         mtdcr(0xf3, mfdcr(0xf3) | (1 << idx));
102 #else /* CONFIG_440EP || CONFIG_440GR */
103         SDR_WRITE(DCRN_SDR_MFR, SDR_READ(DCRN_SDR_MFR) | (0x08000000 >> idx));
104 #endif
105
106         local_irq_restore(flags);
107 }
108
109 static inline void EMAC_RX_CLK_DEFAULT(int idx)
110 {
111         unsigned long flags;
112         local_irq_save(flags);
113
114 #if defined(CONFIG_405EP)
115         mtdcr(0xf3, mfdcr(0xf3) & ~(1 << idx));
116 #else /* CONFIG_440EP */
117         SDR_WRITE(DCRN_SDR_MFR, SDR_READ(DCRN_SDR_MFR) & ~(0x08000000 >> idx));
118 #endif
119
120         local_irq_restore(flags);
121 }
122 #else
123 #define EMAC_RX_CLK_TX(idx)             ((void)0)
124 #define EMAC_RX_CLK_DEFAULT(idx)        ((void)0)
125 #endif
126
127 #if defined(CONFIG_IBM_EMAC_PHY_RX_CLK_FIX) && defined(CONFIG_440GX)
128 /* We can switch Ethernet clock to the internal source through SDR0_MFR[ECS],
129  * unfortunately this is less flexible than 440EP case, because it's a global 
130  * setting for all EMACs, therefore we do this clock trick only during probe.
131  */
132 #define EMAC_CLK_INTERNAL               SDR_WRITE(DCRN_SDR_MFR, \
133                                             SDR_READ(DCRN_SDR_MFR) | 0x08000000)
134 #define EMAC_CLK_EXTERNAL               SDR_WRITE(DCRN_SDR_MFR, \
135                                             SDR_READ(DCRN_SDR_MFR) & ~0x08000000)
136 #else
137 #define EMAC_CLK_INTERNAL               ((void)0)
138 #define EMAC_CLK_EXTERNAL               ((void)0)
139 #endif
140
141 /* I don't want to litter system log with timeout errors 
142  * when we have brain-damaged PHY.
143  */
144 static inline void emac_report_timeout_error(struct ocp_enet_private *dev,
145                                              const char *error)
146 {
147 #if defined(CONFIG_IBM_EMAC_PHY_RX_CLK_FIX)
148         DBG("%d: %s" NL, dev->def->index, error);
149 #else
150         if (net_ratelimit())
151                 printk(KERN_ERR "emac%d: %s\n", dev->def->index, error);
152 #endif
153 }
154
155 /* PHY polling intervals */
156 #define PHY_POLL_LINK_ON        HZ
157 #define PHY_POLL_LINK_OFF       (HZ / 5)
158
159 /* Graceful stop timeouts in us. 
160  * We should allow up to 1 frame time (full-duplex, ignoring collisions) 
161  */
162 #define STOP_TIMEOUT_10         1230    
163 #define STOP_TIMEOUT_100        124
164 #define STOP_TIMEOUT_1000       13
165 #define STOP_TIMEOUT_1000_JUMBO 73
166
167 /* Please, keep in sync with struct ibm_emac_stats/ibm_emac_error_stats */
168 static const char emac_stats_keys[EMAC_ETHTOOL_STATS_COUNT][ETH_GSTRING_LEN] = {
169         "rx_packets", "rx_bytes", "tx_packets", "tx_bytes", "rx_packets_csum",
170         "tx_packets_csum", "tx_undo", "rx_dropped_stack", "rx_dropped_oom",
171         "rx_dropped_error", "rx_dropped_resize", "rx_dropped_mtu",
172         "rx_stopped", "rx_bd_errors", "rx_bd_overrun", "rx_bd_bad_packet",
173         "rx_bd_runt_packet", "rx_bd_short_event", "rx_bd_alignment_error",
174         "rx_bd_bad_fcs", "rx_bd_packet_too_long", "rx_bd_out_of_range",
175         "rx_bd_in_range", "rx_parity", "rx_fifo_overrun", "rx_overrun",
176         "rx_bad_packet", "rx_runt_packet", "rx_short_event",
177         "rx_alignment_error", "rx_bad_fcs", "rx_packet_too_long",
178         "rx_out_of_range", "rx_in_range", "tx_dropped", "tx_bd_errors",
179         "tx_bd_bad_fcs", "tx_bd_carrier_loss", "tx_bd_excessive_deferral",
180         "tx_bd_excessive_collisions", "tx_bd_late_collision",
181         "tx_bd_multple_collisions", "tx_bd_single_collision",
182         "tx_bd_underrun", "tx_bd_sqe", "tx_parity", "tx_underrun", "tx_sqe",
183         "tx_errors"
184 };
185
186 static irqreturn_t emac_irq(int irq, void *dev_instance);
187 static void emac_clean_tx_ring(struct ocp_enet_private *dev);
188
189 static inline int emac_phy_supports_gige(int phy_mode)
190 {
191         return  phy_mode == PHY_MODE_GMII ||
192                 phy_mode == PHY_MODE_RGMII ||
193                 phy_mode == PHY_MODE_TBI ||
194                 phy_mode == PHY_MODE_RTBI;
195 }
196
197 static inline int emac_phy_gpcs(int phy_mode)
198 {
199         return  phy_mode == PHY_MODE_TBI ||
200                 phy_mode == PHY_MODE_RTBI;
201 }
202
203 static inline void emac_tx_enable(struct ocp_enet_private *dev)
204 {
205         struct emac_regs __iomem *p = dev->emacp;
206         unsigned long flags;
207         u32 r;
208
209         local_irq_save(flags);
210
211         DBG("%d: tx_enable" NL, dev->def->index);
212
213         r = in_be32(&p->mr0);
214         if (!(r & EMAC_MR0_TXE))
215                 out_be32(&p->mr0, r | EMAC_MR0_TXE);
216         local_irq_restore(flags);
217 }
218
219 static void emac_tx_disable(struct ocp_enet_private *dev)
220 {
221         struct emac_regs __iomem *p = dev->emacp;
222         unsigned long flags;
223         u32 r;
224
225         local_irq_save(flags);
226
227         DBG("%d: tx_disable" NL, dev->def->index);
228
229         r = in_be32(&p->mr0);
230         if (r & EMAC_MR0_TXE) {
231                 int n = dev->stop_timeout;
232                 out_be32(&p->mr0, r & ~EMAC_MR0_TXE);
233                 while (!(in_be32(&p->mr0) & EMAC_MR0_TXI) && n) {
234                         udelay(1);
235                         --n;
236                 }       
237                 if (unlikely(!n))
238                         emac_report_timeout_error(dev, "TX disable timeout");
239         }
240         local_irq_restore(flags);
241 }
242
243 static void emac_rx_enable(struct ocp_enet_private *dev)
244 {
245         struct emac_regs __iomem *p = dev->emacp;
246         unsigned long flags;
247         u32 r;
248
249         local_irq_save(flags);
250         if (unlikely(dev->commac.rx_stopped))
251                 goto out;
252
253         DBG("%d: rx_enable" NL, dev->def->index);
254
255         r = in_be32(&p->mr0);
256         if (!(r & EMAC_MR0_RXE)) {
257                 if (unlikely(!(r & EMAC_MR0_RXI))) {
258                         /* Wait if previous async disable is still in progress */
259                         int n = dev->stop_timeout;
260                         while (!(r = in_be32(&p->mr0) & EMAC_MR0_RXI) && n) {
261                                 udelay(1);
262                                 --n;
263                         }       
264                         if (unlikely(!n))
265                                 emac_report_timeout_error(dev,
266                                                           "RX disable timeout");
267                 }
268                 out_be32(&p->mr0, r | EMAC_MR0_RXE);
269         }
270       out:
271         local_irq_restore(flags);
272 }
273
274 static void emac_rx_disable(struct ocp_enet_private *dev)
275 {
276         struct emac_regs __iomem *p = dev->emacp;
277         unsigned long flags;
278         u32 r;
279
280         local_irq_save(flags);
281
282         DBG("%d: rx_disable" NL, dev->def->index);
283
284         r = in_be32(&p->mr0);
285         if (r & EMAC_MR0_RXE) {
286                 int n = dev->stop_timeout;
287                 out_be32(&p->mr0, r & ~EMAC_MR0_RXE);
288                 while (!(in_be32(&p->mr0) & EMAC_MR0_RXI) && n) {
289                         udelay(1);
290                         --n;
291                 }       
292                 if (unlikely(!n))
293                         emac_report_timeout_error(dev, "RX disable timeout");
294         }
295         local_irq_restore(flags);
296 }
297
298 static inline void emac_rx_disable_async(struct ocp_enet_private *dev)
299 {
300         struct emac_regs __iomem *p = dev->emacp;
301         unsigned long flags;
302         u32 r;
303
304         local_irq_save(flags);
305
306         DBG("%d: rx_disable_async" NL, dev->def->index);
307
308         r = in_be32(&p->mr0);
309         if (r & EMAC_MR0_RXE)
310                 out_be32(&p->mr0, r & ~EMAC_MR0_RXE);
311         local_irq_restore(flags);
312 }
313
314 static int emac_reset(struct ocp_enet_private *dev)
315 {
316         struct emac_regs __iomem *p = dev->emacp;
317         unsigned long flags;
318         int n = 20;
319
320         DBG("%d: reset" NL, dev->def->index);
321
322         local_irq_save(flags);
323
324         if (!dev->reset_failed) {
325                 /* 40x erratum suggests stopping RX channel before reset,
326                  * we stop TX as well
327                  */
328                 emac_rx_disable(dev);
329                 emac_tx_disable(dev);
330         }
331
332         out_be32(&p->mr0, EMAC_MR0_SRST);
333         while ((in_be32(&p->mr0) & EMAC_MR0_SRST) && n)
334                 --n;
335         local_irq_restore(flags);
336
337         if (n) {
338                 dev->reset_failed = 0;
339                 return 0;
340         } else {
341                 emac_report_timeout_error(dev, "reset timeout");
342                 dev->reset_failed = 1;
343                 return -ETIMEDOUT;
344         }
345 }
346
347 static void emac_hash_mc(struct ocp_enet_private *dev)
348 {
349         struct emac_regs __iomem *p = dev->emacp;
350         u16 gaht[4] = { 0 };
351         struct dev_mc_list *dmi;
352
353         DBG("%d: hash_mc %d" NL, dev->def->index, dev->ndev->mc_count);
354
355         for (dmi = dev->ndev->mc_list; dmi; dmi = dmi->next) {
356                 int bit;
357                 DBG2("%d: mc %02x:%02x:%02x:%02x:%02x:%02x" NL,
358                      dev->def->index,
359                      dmi->dmi_addr[0], dmi->dmi_addr[1], dmi->dmi_addr[2],
360                      dmi->dmi_addr[3], dmi->dmi_addr[4], dmi->dmi_addr[5]);
361
362                 bit = 63 - (ether_crc(ETH_ALEN, dmi->dmi_addr) >> 26);
363                 gaht[bit >> 4] |= 0x8000 >> (bit & 0x0f);
364         }
365         out_be32(&p->gaht1, gaht[0]);
366         out_be32(&p->gaht2, gaht[1]);
367         out_be32(&p->gaht3, gaht[2]);
368         out_be32(&p->gaht4, gaht[3]);
369 }
370
371 static inline u32 emac_iff2rmr(struct net_device *ndev)
372 {
373         u32 r = EMAC_RMR_SP | EMAC_RMR_SFCS | EMAC_RMR_IAE | EMAC_RMR_BAE |
374             EMAC_RMR_BASE;
375
376         if (ndev->flags & IFF_PROMISC)
377                 r |= EMAC_RMR_PME;
378         else if (ndev->flags & IFF_ALLMULTI || ndev->mc_count > 32)
379                 r |= EMAC_RMR_PMME;
380         else if (ndev->mc_count > 0)
381                 r |= EMAC_RMR_MAE;
382
383         return r;
384 }
385
386 static inline int emac_opb_mhz(void)
387 {
388         return (ocp_sys_info.opb_bus_freq + 500000) / 1000000;
389 }
390
391 /* BHs disabled */
392 static int emac_configure(struct ocp_enet_private *dev)
393 {
394         struct emac_regs __iomem *p = dev->emacp;
395         struct net_device *ndev = dev->ndev;
396         int gige;
397         u32 r;
398
399         DBG("%d: configure" NL, dev->def->index);
400
401         if (emac_reset(dev) < 0)
402                 return -ETIMEDOUT;
403
404         tah_reset(dev->tah_dev);
405
406         /* Mode register */
407         r = EMAC_MR1_BASE(emac_opb_mhz()) | EMAC_MR1_VLE | EMAC_MR1_IST;
408         if (dev->phy.duplex == DUPLEX_FULL)
409                 r |= EMAC_MR1_FDE | EMAC_MR1_MWSW_001;
410         dev->stop_timeout = STOP_TIMEOUT_10;
411         switch (dev->phy.speed) {
412         case SPEED_1000:
413                 if (emac_phy_gpcs(dev->phy.mode)) {
414                         r |= EMAC_MR1_MF_1000GPCS |
415                             EMAC_MR1_MF_IPPA(dev->phy.address);
416
417                         /* Put some arbitrary OUI, Manuf & Rev IDs so we can
418                          * identify this GPCS PHY later.
419                          */
420                         out_be32(&p->ipcr, 0xdeadbeef);
421                 } else
422                         r |= EMAC_MR1_MF_1000;
423                 r |= EMAC_MR1_RFS_16K;
424                 gige = 1;
425
426                 if (dev->ndev->mtu > ETH_DATA_LEN) {
427                         r |= EMAC_MR1_JPSM;
428                         dev->stop_timeout = STOP_TIMEOUT_1000_JUMBO;
429                 } else
430                         dev->stop_timeout = STOP_TIMEOUT_1000;
431                 break;
432         case SPEED_100:
433                 r |= EMAC_MR1_MF_100;
434                 dev->stop_timeout = STOP_TIMEOUT_100;
435                 /* Fall through */
436         default:
437                 r |= EMAC_MR1_RFS_4K;
438                 gige = 0;
439                 break;
440         }
441
442         if (dev->rgmii_dev)
443                 rgmii_set_speed(dev->rgmii_dev, dev->rgmii_input,
444                                 dev->phy.speed);
445         else
446                 zmii_set_speed(dev->zmii_dev, dev->zmii_input, dev->phy.speed);
447
448 #if !defined(CONFIG_40x)
449         /* on 40x erratum forces us to NOT use integrated flow control, 
450          * let's hope it works on 44x ;)
451          */
452         if (dev->phy.duplex == DUPLEX_FULL) {
453                 if (dev->phy.pause)
454                         r |= EMAC_MR1_EIFC | EMAC_MR1_APP;
455                 else if (dev->phy.asym_pause)
456                         r |= EMAC_MR1_APP;
457         }
458 #endif
459         out_be32(&p->mr1, r);
460
461         /* Set individual MAC address */
462         out_be32(&p->iahr, (ndev->dev_addr[0] << 8) | ndev->dev_addr[1]);
463         out_be32(&p->ialr, (ndev->dev_addr[2] << 24) |
464                  (ndev->dev_addr[3] << 16) | (ndev->dev_addr[4] << 8) |
465                  ndev->dev_addr[5]);
466
467         /* VLAN Tag Protocol ID */
468         out_be32(&p->vtpid, 0x8100);
469
470         /* Receive mode register */
471         r = emac_iff2rmr(ndev);
472         if (r & EMAC_RMR_MAE)
473                 emac_hash_mc(dev);
474         out_be32(&p->rmr, r);
475
476         /* FIFOs thresholds */
477         r = EMAC_TMR1((EMAC_MAL_BURST_SIZE / EMAC_FIFO_ENTRY_SIZE) + 1,
478                       EMAC_TX_FIFO_SIZE / 2 / EMAC_FIFO_ENTRY_SIZE);
479         out_be32(&p->tmr1, r);
480         out_be32(&p->trtr, EMAC_TRTR(EMAC_TX_FIFO_SIZE / 2));
481
482         /* PAUSE frame is sent when RX FIFO reaches its high-water mark,
483            there should be still enough space in FIFO to allow the our link
484            partner time to process this frame and also time to send PAUSE 
485            frame itself.
486
487            Here is the worst case scenario for the RX FIFO "headroom"
488            (from "The Switch Book") (100Mbps, without preamble, inter-frame gap):
489
490            1) One maximum-length frame on TX                    1522 bytes
491            2) One PAUSE frame time                                64 bytes
492            3) PAUSE frame decode time allowance                   64 bytes
493            4) One maximum-length frame on RX                    1522 bytes
494            5) Round-trip propagation delay of the link (100Mb)    15 bytes
495            ----------       
496            3187 bytes
497
498            I chose to set high-water mark to RX_FIFO_SIZE / 4 (1024 bytes)
499            low-water mark  to RX_FIFO_SIZE / 8 (512 bytes)
500          */
501         r = EMAC_RWMR(EMAC_RX_FIFO_SIZE(gige) / 8 / EMAC_FIFO_ENTRY_SIZE,
502                       EMAC_RX_FIFO_SIZE(gige) / 4 / EMAC_FIFO_ENTRY_SIZE);
503         out_be32(&p->rwmr, r);
504
505         /* Set PAUSE timer to the maximum */
506         out_be32(&p->ptr, 0xffff);
507
508         /* IRQ sources */
509         out_be32(&p->iser, EMAC_ISR_TXPE | EMAC_ISR_RXPE | /* EMAC_ISR_TXUE |
510                  EMAC_ISR_RXOE | */ EMAC_ISR_OVR | EMAC_ISR_BP | EMAC_ISR_SE |
511                  EMAC_ISR_ALE | EMAC_ISR_BFCS | EMAC_ISR_PTLE | EMAC_ISR_ORE |
512                  EMAC_ISR_IRE | EMAC_ISR_TE);
513                  
514         /* We need to take GPCS PHY out of isolate mode after EMAC reset */
515         if (emac_phy_gpcs(dev->phy.mode)) 
516                 mii_reset_phy(&dev->phy);
517                  
518         return 0;
519 }
520
521 /* BHs disabled */
522 static void emac_reinitialize(struct ocp_enet_private *dev)
523 {
524         DBG("%d: reinitialize" NL, dev->def->index);
525
526         if (!emac_configure(dev)) {
527                 emac_tx_enable(dev);
528                 emac_rx_enable(dev);
529         }
530 }
531
532 /* BHs disabled */
533 static void emac_full_tx_reset(struct net_device *ndev)
534 {
535         struct ocp_enet_private *dev = ndev->priv;
536         struct ocp_func_emac_data *emacdata = dev->def->additions;
537
538         DBG("%d: full_tx_reset" NL, dev->def->index);
539
540         emac_tx_disable(dev);
541         mal_disable_tx_channel(dev->mal, emacdata->mal_tx_chan);
542         emac_clean_tx_ring(dev);
543         dev->tx_cnt = dev->tx_slot = dev->ack_slot = 0;
544
545         emac_configure(dev);
546
547         mal_enable_tx_channel(dev->mal, emacdata->mal_tx_chan);
548         emac_tx_enable(dev);
549         emac_rx_enable(dev);
550
551         netif_wake_queue(ndev);
552 }
553
554 static int __emac_mdio_read(struct ocp_enet_private *dev, u8 id, u8 reg)
555 {
556         struct emac_regs __iomem *p = dev->emacp;
557         u32 r;
558         int n;
559
560         DBG2("%d: mdio_read(%02x,%02x)" NL, dev->def->index, id, reg);
561
562         /* Enable proper MDIO port */
563         zmii_enable_mdio(dev->zmii_dev, dev->zmii_input);
564
565         /* Wait for management interface to become idle */
566         n = 10;
567         while (!emac_phy_done(in_be32(&p->stacr))) {
568                 udelay(1);
569                 if (!--n)
570                         goto to;
571         }
572
573         /* Issue read command */
574         out_be32(&p->stacr,
575                  EMAC_STACR_BASE(emac_opb_mhz()) | EMAC_STACR_STAC_READ |
576                  (reg & EMAC_STACR_PRA_MASK)
577                  | ((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT)
578                  | EMAC_STACR_START);
579
580         /* Wait for read to complete */
581         n = 100;
582         while (!emac_phy_done(r = in_be32(&p->stacr))) {
583                 udelay(1);
584                 if (!--n)
585                         goto to;
586         }
587
588         if (unlikely(r & EMAC_STACR_PHYE)) {
589                 DBG("%d: mdio_read(%02x, %02x) failed" NL, dev->def->index,
590                     id, reg);
591                 return -EREMOTEIO;
592         }
593
594         r = ((r >> EMAC_STACR_PHYD_SHIFT) & EMAC_STACR_PHYD_MASK);
595         DBG2("%d: mdio_read -> %04x" NL, dev->def->index, r);
596         return r;
597       to:
598         DBG("%d: MII management interface timeout (read)" NL, dev->def->index);
599         return -ETIMEDOUT;
600 }
601
602 static void __emac_mdio_write(struct ocp_enet_private *dev, u8 id, u8 reg,
603                               u16 val)
604 {
605         struct emac_regs __iomem *p = dev->emacp;
606         int n;
607
608         DBG2("%d: mdio_write(%02x,%02x,%04x)" NL, dev->def->index, id, reg,
609              val);
610
611         /* Enable proper MDIO port */
612         zmii_enable_mdio(dev->zmii_dev, dev->zmii_input);
613
614         /* Wait for management interface to be idle */
615         n = 10;
616         while (!emac_phy_done(in_be32(&p->stacr))) {
617                 udelay(1);
618                 if (!--n)
619                         goto to;
620         }
621
622         /* Issue write command */
623         out_be32(&p->stacr,
624                  EMAC_STACR_BASE(emac_opb_mhz()) | EMAC_STACR_STAC_WRITE |
625                  (reg & EMAC_STACR_PRA_MASK) |
626                  ((id & EMAC_STACR_PCDA_MASK) << EMAC_STACR_PCDA_SHIFT) |
627                  (val << EMAC_STACR_PHYD_SHIFT) | EMAC_STACR_START);
628
629         /* Wait for write to complete */
630         n = 100;
631         while (!emac_phy_done(in_be32(&p->stacr))) {
632                 udelay(1);
633                 if (!--n)
634                         goto to;
635         }
636         return;
637       to:
638         DBG("%d: MII management interface timeout (write)" NL, dev->def->index);
639 }
640
641 static int emac_mdio_read(struct net_device *ndev, int id, int reg)
642 {
643         struct ocp_enet_private *dev = ndev->priv;
644         int res;
645
646         local_bh_disable();
647         res = __emac_mdio_read(dev->mdio_dev ? dev->mdio_dev : dev, (u8) id,
648                                (u8) reg);
649         local_bh_enable();
650         return res;
651 }
652
653 static void emac_mdio_write(struct net_device *ndev, int id, int reg, int val)
654 {
655         struct ocp_enet_private *dev = ndev->priv;
656
657         local_bh_disable();
658         __emac_mdio_write(dev->mdio_dev ? dev->mdio_dev : dev, (u8) id,
659                           (u8) reg, (u16) val);
660         local_bh_enable();
661 }
662
663 /* BHs disabled */
664 static void emac_set_multicast_list(struct net_device *ndev)
665 {
666         struct ocp_enet_private *dev = ndev->priv;
667         struct emac_regs __iomem *p = dev->emacp;
668         u32 rmr = emac_iff2rmr(ndev);
669
670         DBG("%d: multicast %08x" NL, dev->def->index, rmr);
671         BUG_ON(!netif_running(dev->ndev));
672
673         /* I decided to relax register access rules here to avoid
674          * full EMAC reset.
675          *
676          * There is a real problem with EMAC4 core if we use MWSW_001 bit 
677          * in MR1 register and do a full EMAC reset.
678          * One TX BD status update is delayed and, after EMAC reset, it 
679          * never happens, resulting in TX hung (it'll be recovered by TX 
680          * timeout handler eventually, but this is just gross).
681          * So we either have to do full TX reset or try to cheat here :)
682          *
683          * The only required change is to RX mode register, so I *think* all
684          * we need is just to stop RX channel. This seems to work on all
685          * tested SoCs.                                                --ebs
686          */
687         emac_rx_disable(dev);
688         if (rmr & EMAC_RMR_MAE)
689                 emac_hash_mc(dev);
690         out_be32(&p->rmr, rmr);
691         emac_rx_enable(dev);
692 }
693
694 /* BHs disabled */
695 static int emac_resize_rx_ring(struct ocp_enet_private *dev, int new_mtu)
696 {
697         struct ocp_func_emac_data *emacdata = dev->def->additions;
698         int rx_sync_size = emac_rx_sync_size(new_mtu);
699         int rx_skb_size = emac_rx_skb_size(new_mtu);
700         int i, ret = 0;
701
702         emac_rx_disable(dev);
703         mal_disable_rx_channel(dev->mal, emacdata->mal_rx_chan);
704
705         if (dev->rx_sg_skb) {
706                 ++dev->estats.rx_dropped_resize;
707                 dev_kfree_skb(dev->rx_sg_skb);
708                 dev->rx_sg_skb = NULL;
709         }
710
711         /* Make a first pass over RX ring and mark BDs ready, dropping 
712          * non-processed packets on the way. We need this as a separate pass
713          * to simplify error recovery in the case of allocation failure later.
714          */
715         for (i = 0; i < NUM_RX_BUFF; ++i) {
716                 if (dev->rx_desc[i].ctrl & MAL_RX_CTRL_FIRST)
717                         ++dev->estats.rx_dropped_resize;
718
719                 dev->rx_desc[i].data_len = 0;
720                 dev->rx_desc[i].ctrl = MAL_RX_CTRL_EMPTY |
721                     (i == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
722         }
723
724         /* Reallocate RX ring only if bigger skb buffers are required */
725         if (rx_skb_size <= dev->rx_skb_size)
726                 goto skip;
727
728         /* Second pass, allocate new skbs */
729         for (i = 0; i < NUM_RX_BUFF; ++i) {
730                 struct sk_buff *skb = alloc_skb(rx_skb_size, GFP_ATOMIC);
731                 if (!skb) {
732                         ret = -ENOMEM;
733                         goto oom;
734                 }
735
736                 BUG_ON(!dev->rx_skb[i]);
737                 dev_kfree_skb(dev->rx_skb[i]);
738
739                 skb_reserve(skb, EMAC_RX_SKB_HEADROOM + 2);
740                 dev->rx_desc[i].data_ptr =
741                     dma_map_single(dev->ldev, skb->data - 2, rx_sync_size,
742                                    DMA_FROM_DEVICE) + 2;
743                 dev->rx_skb[i] = skb;
744         }
745       skip:
746         /* Check if we need to change "Jumbo" bit in MR1 */
747         if ((new_mtu > ETH_DATA_LEN) ^ (dev->ndev->mtu > ETH_DATA_LEN)) {
748                 /* This is to prevent starting RX channel in emac_rx_enable() */
749                 dev->commac.rx_stopped = 1;
750
751                 dev->ndev->mtu = new_mtu;
752                 emac_full_tx_reset(dev->ndev);
753         }
754
755         mal_set_rcbs(dev->mal, emacdata->mal_rx_chan, emac_rx_size(new_mtu));
756       oom:
757         /* Restart RX */
758         dev->commac.rx_stopped = dev->rx_slot = 0;
759         mal_enable_rx_channel(dev->mal, emacdata->mal_rx_chan);
760         emac_rx_enable(dev);
761
762         return ret;
763 }
764
765 /* Process ctx, rtnl_lock semaphore */
766 static int emac_change_mtu(struct net_device *ndev, int new_mtu)
767 {
768         struct ocp_enet_private *dev = ndev->priv;
769         int ret = 0;
770
771         if (new_mtu < EMAC_MIN_MTU || new_mtu > EMAC_MAX_MTU)
772                 return -EINVAL;
773
774         DBG("%d: change_mtu(%d)" NL, dev->def->index, new_mtu);
775
776         local_bh_disable();
777         if (netif_running(ndev)) {
778                 /* Check if we really need to reinitalize RX ring */
779                 if (emac_rx_skb_size(ndev->mtu) != emac_rx_skb_size(new_mtu))
780                         ret = emac_resize_rx_ring(dev, new_mtu);
781         }
782
783         if (!ret) {
784                 ndev->mtu = new_mtu;
785                 dev->rx_skb_size = emac_rx_skb_size(new_mtu);
786                 dev->rx_sync_size = emac_rx_sync_size(new_mtu);
787         }       
788         local_bh_enable();
789
790         return ret;
791 }
792
793 static void emac_clean_tx_ring(struct ocp_enet_private *dev)
794 {
795         int i;
796         for (i = 0; i < NUM_TX_BUFF; ++i) {
797                 if (dev->tx_skb[i]) {
798                         dev_kfree_skb(dev->tx_skb[i]);
799                         dev->tx_skb[i] = NULL;
800                         if (dev->tx_desc[i].ctrl & MAL_TX_CTRL_READY)
801                                 ++dev->estats.tx_dropped;
802                 }
803                 dev->tx_desc[i].ctrl = 0;
804                 dev->tx_desc[i].data_ptr = 0;
805         }
806 }
807
808 static void emac_clean_rx_ring(struct ocp_enet_private *dev)
809 {
810         int i;
811         for (i = 0; i < NUM_RX_BUFF; ++i)
812                 if (dev->rx_skb[i]) {
813                         dev->rx_desc[i].ctrl = 0;
814                         dev_kfree_skb(dev->rx_skb[i]);
815                         dev->rx_skb[i] = NULL;
816                         dev->rx_desc[i].data_ptr = 0;
817                 }
818
819         if (dev->rx_sg_skb) {
820                 dev_kfree_skb(dev->rx_sg_skb);
821                 dev->rx_sg_skb = NULL;
822         }
823 }
824
825 static inline int emac_alloc_rx_skb(struct ocp_enet_private *dev, int slot,
826                                     gfp_t flags)
827 {
828         struct sk_buff *skb = alloc_skb(dev->rx_skb_size, flags);
829         if (unlikely(!skb))
830                 return -ENOMEM;
831
832         dev->rx_skb[slot] = skb;
833         dev->rx_desc[slot].data_len = 0;
834
835         skb_reserve(skb, EMAC_RX_SKB_HEADROOM + 2);
836         dev->rx_desc[slot].data_ptr = 
837             dma_map_single(dev->ldev, skb->data - 2, dev->rx_sync_size, 
838                            DMA_FROM_DEVICE) + 2;
839         barrier();
840         dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
841             (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
842
843         return 0;
844 }
845
846 static void emac_print_link_status(struct ocp_enet_private *dev)
847 {
848         if (netif_carrier_ok(dev->ndev))
849                 printk(KERN_INFO "%s: link is up, %d %s%s\n",
850                        dev->ndev->name, dev->phy.speed,
851                        dev->phy.duplex == DUPLEX_FULL ? "FDX" : "HDX",
852                        dev->phy.pause ? ", pause enabled" :
853                        dev->phy.asym_pause ? ", assymetric pause enabled" : "");
854         else
855                 printk(KERN_INFO "%s: link is down\n", dev->ndev->name);
856 }
857
858 /* Process ctx, rtnl_lock semaphore */
859 static int emac_open(struct net_device *ndev)
860 {
861         struct ocp_enet_private *dev = ndev->priv;
862         struct ocp_func_emac_data *emacdata = dev->def->additions;
863         int err, i;
864
865         DBG("%d: open" NL, dev->def->index);
866
867         /* Setup error IRQ handler */
868         err = request_irq(dev->def->irq, emac_irq, 0, "EMAC", dev);
869         if (err) {
870                 printk(KERN_ERR "%s: failed to request IRQ %d\n",
871                        ndev->name, dev->def->irq);
872                 return err;
873         }
874
875         /* Allocate RX ring */
876         for (i = 0; i < NUM_RX_BUFF; ++i)
877                 if (emac_alloc_rx_skb(dev, i, GFP_KERNEL)) {
878                         printk(KERN_ERR "%s: failed to allocate RX ring\n",
879                                ndev->name);
880                         goto oom;
881                 }
882
883         local_bh_disable();
884         dev->tx_cnt = dev->tx_slot = dev->ack_slot = dev->rx_slot =
885             dev->commac.rx_stopped = 0;
886         dev->rx_sg_skb = NULL;
887
888         if (dev->phy.address >= 0) {
889                 int link_poll_interval;
890                 if (dev->phy.def->ops->poll_link(&dev->phy)) {
891                         dev->phy.def->ops->read_link(&dev->phy);
892                         EMAC_RX_CLK_DEFAULT(dev->def->index);
893                         netif_carrier_on(dev->ndev);
894                         link_poll_interval = PHY_POLL_LINK_ON;
895                 } else {
896                         EMAC_RX_CLK_TX(dev->def->index);
897                         netif_carrier_off(dev->ndev);
898                         link_poll_interval = PHY_POLL_LINK_OFF;
899                 }
900                 mod_timer(&dev->link_timer, jiffies + link_poll_interval);
901                 emac_print_link_status(dev);
902         } else
903                 netif_carrier_on(dev->ndev);
904
905         emac_configure(dev);
906         mal_poll_add(dev->mal, &dev->commac);
907         mal_enable_tx_channel(dev->mal, emacdata->mal_tx_chan);
908         mal_set_rcbs(dev->mal, emacdata->mal_rx_chan, emac_rx_size(ndev->mtu));
909         mal_enable_rx_channel(dev->mal, emacdata->mal_rx_chan);
910         emac_tx_enable(dev);
911         emac_rx_enable(dev);
912         netif_start_queue(ndev);
913         local_bh_enable();
914
915         return 0;
916       oom:
917         emac_clean_rx_ring(dev);
918         free_irq(dev->def->irq, dev);
919         return -ENOMEM;
920 }
921
922 /* BHs disabled */
923 static int emac_link_differs(struct ocp_enet_private *dev)
924 {
925         u32 r = in_be32(&dev->emacp->mr1);
926
927         int duplex = r & EMAC_MR1_FDE ? DUPLEX_FULL : DUPLEX_HALF;
928         int speed, pause, asym_pause;
929
930         if (r & (EMAC_MR1_MF_1000 | EMAC_MR1_MF_1000GPCS))
931                 speed = SPEED_1000;
932         else if (r & EMAC_MR1_MF_100)
933                 speed = SPEED_100;
934         else
935                 speed = SPEED_10;
936
937         switch (r & (EMAC_MR1_EIFC | EMAC_MR1_APP)) {
938         case (EMAC_MR1_EIFC | EMAC_MR1_APP):
939                 pause = 1;
940                 asym_pause = 0;
941                 break;
942         case EMAC_MR1_APP:
943                 pause = 0;
944                 asym_pause = 1;
945                 break;
946         default:
947                 pause = asym_pause = 0;
948         }
949         return speed != dev->phy.speed || duplex != dev->phy.duplex ||
950             pause != dev->phy.pause || asym_pause != dev->phy.asym_pause;
951 }
952
953 /* BHs disabled */
954 static void emac_link_timer(unsigned long data)
955 {
956         struct ocp_enet_private *dev = (struct ocp_enet_private *)data;
957         int link_poll_interval;
958
959         DBG2("%d: link timer" NL, dev->def->index);
960
961         if (dev->phy.def->ops->poll_link(&dev->phy)) {
962                 if (!netif_carrier_ok(dev->ndev)) {
963                         EMAC_RX_CLK_DEFAULT(dev->def->index);
964
965                         /* Get new link parameters */
966                         dev->phy.def->ops->read_link(&dev->phy);
967
968                         if (dev->tah_dev || emac_link_differs(dev))
969                                 emac_full_tx_reset(dev->ndev);
970
971                         netif_carrier_on(dev->ndev);
972                         emac_print_link_status(dev);
973                 }
974                 link_poll_interval = PHY_POLL_LINK_ON;
975         } else {
976                 if (netif_carrier_ok(dev->ndev)) {
977                         EMAC_RX_CLK_TX(dev->def->index);
978 #if defined(CONFIG_IBM_EMAC_PHY_RX_CLK_FIX)
979                         emac_reinitialize(dev);
980 #endif
981                         netif_carrier_off(dev->ndev);
982                         emac_print_link_status(dev);
983                 }
984
985                 /* Retry reset if the previous attempt failed.
986                  * This is needed mostly for CONFIG_IBM_EMAC_PHY_RX_CLK_FIX
987                  * case, but I left it here because it shouldn't trigger for
988                  * sane PHYs anyway.
989                  */
990                 if (unlikely(dev->reset_failed))
991                         emac_reinitialize(dev);
992
993                 link_poll_interval = PHY_POLL_LINK_OFF;
994         }
995         mod_timer(&dev->link_timer, jiffies + link_poll_interval);
996 }
997
998 /* BHs disabled */
999 static void emac_force_link_update(struct ocp_enet_private *dev)
1000 {
1001         netif_carrier_off(dev->ndev);
1002         if (timer_pending(&dev->link_timer))
1003                 mod_timer(&dev->link_timer, jiffies + PHY_POLL_LINK_OFF);
1004 }
1005
1006 /* Process ctx, rtnl_lock semaphore */
1007 static int emac_close(struct net_device *ndev)
1008 {
1009         struct ocp_enet_private *dev = ndev->priv;
1010         struct ocp_func_emac_data *emacdata = dev->def->additions;
1011
1012         DBG("%d: close" NL, dev->def->index);
1013
1014         local_bh_disable();
1015
1016         if (dev->phy.address >= 0)
1017                 del_timer_sync(&dev->link_timer);
1018
1019         netif_stop_queue(ndev);
1020         emac_rx_disable(dev);
1021         emac_tx_disable(dev);
1022         mal_disable_rx_channel(dev->mal, emacdata->mal_rx_chan);
1023         mal_disable_tx_channel(dev->mal, emacdata->mal_tx_chan);
1024         mal_poll_del(dev->mal, &dev->commac);
1025         local_bh_enable();
1026
1027         emac_clean_tx_ring(dev);
1028         emac_clean_rx_ring(dev);
1029         free_irq(dev->def->irq, dev);
1030
1031         return 0;
1032 }
1033
1034 static inline u16 emac_tx_csum(struct ocp_enet_private *dev,
1035                                struct sk_buff *skb)
1036 {
1037 #if defined(CONFIG_IBM_EMAC_TAH)
1038         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1039                 ++dev->stats.tx_packets_csum;
1040                 return EMAC_TX_CTRL_TAH_CSUM;
1041         }
1042 #endif
1043         return 0;
1044 }
1045
1046 static inline int emac_xmit_finish(struct ocp_enet_private *dev, int len)
1047 {
1048         struct emac_regs __iomem *p = dev->emacp;
1049         struct net_device *ndev = dev->ndev;
1050
1051         /* Send the packet out */
1052         out_be32(&p->tmr0, EMAC_TMR0_XMIT);
1053
1054         if (unlikely(++dev->tx_cnt == NUM_TX_BUFF)) {
1055                 netif_stop_queue(ndev);
1056                 DBG2("%d: stopped TX queue" NL, dev->def->index);
1057         }
1058
1059         ndev->trans_start = jiffies;
1060         ++dev->stats.tx_packets;
1061         dev->stats.tx_bytes += len;
1062
1063         return 0;
1064 }
1065
1066 /* BHs disabled */
1067 static int emac_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1068 {
1069         struct ocp_enet_private *dev = ndev->priv;
1070         unsigned int len = skb->len;
1071         int slot;
1072
1073         u16 ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1074             MAL_TX_CTRL_LAST | emac_tx_csum(dev, skb);
1075
1076         slot = dev->tx_slot++;
1077         if (dev->tx_slot == NUM_TX_BUFF) {
1078                 dev->tx_slot = 0;
1079                 ctrl |= MAL_TX_CTRL_WRAP;
1080         }
1081
1082         DBG2("%d: xmit(%u) %d" NL, dev->def->index, len, slot);
1083
1084         dev->tx_skb[slot] = skb;
1085         dev->tx_desc[slot].data_ptr = dma_map_single(dev->ldev, skb->data, len,
1086                                                      DMA_TO_DEVICE);
1087         dev->tx_desc[slot].data_len = (u16) len;
1088         barrier();
1089         dev->tx_desc[slot].ctrl = ctrl;
1090
1091         return emac_xmit_finish(dev, len);
1092 }
1093
1094 #if defined(CONFIG_IBM_EMAC_TAH)
1095 static inline int emac_xmit_split(struct ocp_enet_private *dev, int slot,
1096                                   u32 pd, int len, int last, u16 base_ctrl)
1097 {
1098         while (1) {
1099                 u16 ctrl = base_ctrl;
1100                 int chunk = min(len, MAL_MAX_TX_SIZE);
1101                 len -= chunk;
1102
1103                 slot = (slot + 1) % NUM_TX_BUFF;
1104
1105                 if (last && !len)
1106                         ctrl |= MAL_TX_CTRL_LAST;
1107                 if (slot == NUM_TX_BUFF - 1)
1108                         ctrl |= MAL_TX_CTRL_WRAP;
1109
1110                 dev->tx_skb[slot] = NULL;
1111                 dev->tx_desc[slot].data_ptr = pd;
1112                 dev->tx_desc[slot].data_len = (u16) chunk;
1113                 dev->tx_desc[slot].ctrl = ctrl;
1114                 ++dev->tx_cnt;
1115
1116                 if (!len)
1117                         break;
1118
1119                 pd += chunk;
1120         }
1121         return slot;
1122 }
1123
1124 /* BHs disabled (SG version for TAH equipped EMACs) */
1125 static int emac_start_xmit_sg(struct sk_buff *skb, struct net_device *ndev)
1126 {
1127         struct ocp_enet_private *dev = ndev->priv;
1128         int nr_frags = skb_shinfo(skb)->nr_frags;
1129         int len = skb->len, chunk;
1130         int slot, i;
1131         u16 ctrl;
1132         u32 pd;
1133
1134         /* This is common "fast" path */
1135         if (likely(!nr_frags && len <= MAL_MAX_TX_SIZE))
1136                 return emac_start_xmit(skb, ndev);
1137
1138         len -= skb->data_len;
1139
1140         /* Note, this is only an *estimation*, we can still run out of empty
1141          * slots because of the additional fragmentation into
1142          * MAL_MAX_TX_SIZE-sized chunks
1143          */
1144         if (unlikely(dev->tx_cnt + nr_frags + mal_tx_chunks(len) > NUM_TX_BUFF))
1145                 goto stop_queue;
1146
1147         ctrl = EMAC_TX_CTRL_GFCS | EMAC_TX_CTRL_GP | MAL_TX_CTRL_READY |
1148             emac_tx_csum(dev, skb);
1149         slot = dev->tx_slot;
1150
1151         /* skb data */
1152         dev->tx_skb[slot] = NULL;
1153         chunk = min(len, MAL_MAX_TX_SIZE);
1154         dev->tx_desc[slot].data_ptr = pd =
1155             dma_map_single(dev->ldev, skb->data, len, DMA_TO_DEVICE);
1156         dev->tx_desc[slot].data_len = (u16) chunk;
1157         len -= chunk;
1158         if (unlikely(len))
1159                 slot = emac_xmit_split(dev, slot, pd + chunk, len, !nr_frags,
1160                                        ctrl);
1161         /* skb fragments */
1162         for (i = 0; i < nr_frags; ++i) {
1163                 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1164                 len = frag->size;
1165
1166                 if (unlikely(dev->tx_cnt + mal_tx_chunks(len) >= NUM_TX_BUFF))
1167                         goto undo_frame;
1168
1169                 pd = dma_map_page(dev->ldev, frag->page, frag->page_offset, len,
1170                                   DMA_TO_DEVICE);
1171
1172                 slot = emac_xmit_split(dev, slot, pd, len, i == nr_frags - 1,
1173                                        ctrl);
1174         }
1175
1176         DBG2("%d: xmit_sg(%u) %d - %d" NL, dev->def->index, skb->len,
1177              dev->tx_slot, slot);
1178
1179         /* Attach skb to the last slot so we don't release it too early */
1180         dev->tx_skb[slot] = skb;
1181
1182         /* Send the packet out */
1183         if (dev->tx_slot == NUM_TX_BUFF - 1)
1184                 ctrl |= MAL_TX_CTRL_WRAP;
1185         barrier();
1186         dev->tx_desc[dev->tx_slot].ctrl = ctrl;
1187         dev->tx_slot = (slot + 1) % NUM_TX_BUFF;
1188
1189         return emac_xmit_finish(dev, skb->len);
1190
1191       undo_frame:
1192         /* Well, too bad. Our previous estimation was overly optimistic. 
1193          * Undo everything.
1194          */
1195         while (slot != dev->tx_slot) {
1196                 dev->tx_desc[slot].ctrl = 0;
1197                 --dev->tx_cnt;
1198                 if (--slot < 0)
1199                         slot = NUM_TX_BUFF - 1;
1200         }
1201         ++dev->estats.tx_undo;
1202
1203       stop_queue:
1204         netif_stop_queue(ndev);
1205         DBG2("%d: stopped TX queue" NL, dev->def->index);
1206         return 1;
1207 }
1208 #else
1209 # define emac_start_xmit_sg     emac_start_xmit
1210 #endif  /* !defined(CONFIG_IBM_EMAC_TAH) */
1211
1212 /* BHs disabled */
1213 static void emac_parse_tx_error(struct ocp_enet_private *dev, u16 ctrl)
1214 {
1215         struct ibm_emac_error_stats *st = &dev->estats;
1216         DBG("%d: BD TX error %04x" NL, dev->def->index, ctrl);
1217
1218         ++st->tx_bd_errors;
1219         if (ctrl & EMAC_TX_ST_BFCS)
1220                 ++st->tx_bd_bad_fcs;
1221         if (ctrl & EMAC_TX_ST_LCS)
1222                 ++st->tx_bd_carrier_loss;
1223         if (ctrl & EMAC_TX_ST_ED)
1224                 ++st->tx_bd_excessive_deferral;
1225         if (ctrl & EMAC_TX_ST_EC)
1226                 ++st->tx_bd_excessive_collisions;
1227         if (ctrl & EMAC_TX_ST_LC)
1228                 ++st->tx_bd_late_collision;
1229         if (ctrl & EMAC_TX_ST_MC)
1230                 ++st->tx_bd_multple_collisions;
1231         if (ctrl & EMAC_TX_ST_SC)
1232                 ++st->tx_bd_single_collision;
1233         if (ctrl & EMAC_TX_ST_UR)
1234                 ++st->tx_bd_underrun;
1235         if (ctrl & EMAC_TX_ST_SQE)
1236                 ++st->tx_bd_sqe;
1237 }
1238
1239 static void emac_poll_tx(void *param)
1240 {
1241         struct ocp_enet_private *dev = param;
1242         DBG2("%d: poll_tx, %d %d" NL, dev->def->index, dev->tx_cnt,
1243              dev->ack_slot);
1244
1245         if (dev->tx_cnt) {
1246                 u16 ctrl;
1247                 int slot = dev->ack_slot, n = 0;
1248               again:
1249                 ctrl = dev->tx_desc[slot].ctrl;
1250                 if (!(ctrl & MAL_TX_CTRL_READY)) {
1251                         struct sk_buff *skb = dev->tx_skb[slot];
1252                         ++n;
1253
1254                         if (skb) {
1255                                 dev_kfree_skb(skb);
1256                                 dev->tx_skb[slot] = NULL;
1257                         }
1258                         slot = (slot + 1) % NUM_TX_BUFF;
1259
1260                         if (unlikely(EMAC_IS_BAD_TX(ctrl)))
1261                                 emac_parse_tx_error(dev, ctrl);
1262
1263                         if (--dev->tx_cnt)
1264                                 goto again;
1265                 }
1266                 if (n) {
1267                         dev->ack_slot = slot;
1268                         if (netif_queue_stopped(dev->ndev) &&
1269                             dev->tx_cnt < EMAC_TX_WAKEUP_THRESH)
1270                                 netif_wake_queue(dev->ndev);
1271
1272                         DBG2("%d: tx %d pkts" NL, dev->def->index, n);
1273                 }
1274         }
1275 }
1276
1277 static inline void emac_recycle_rx_skb(struct ocp_enet_private *dev, int slot,
1278                                        int len)
1279 {
1280         struct sk_buff *skb = dev->rx_skb[slot];
1281         DBG2("%d: recycle %d %d" NL, dev->def->index, slot, len);
1282
1283         if (len) 
1284                 dma_map_single(dev->ldev, skb->data - 2, 
1285                                EMAC_DMA_ALIGN(len + 2), DMA_FROM_DEVICE);
1286
1287         dev->rx_desc[slot].data_len = 0;
1288         barrier();
1289         dev->rx_desc[slot].ctrl = MAL_RX_CTRL_EMPTY |
1290             (slot == (NUM_RX_BUFF - 1) ? MAL_RX_CTRL_WRAP : 0);
1291 }
1292
1293 static void emac_parse_rx_error(struct ocp_enet_private *dev, u16 ctrl)
1294 {
1295         struct ibm_emac_error_stats *st = &dev->estats;
1296         DBG("%d: BD RX error %04x" NL, dev->def->index, ctrl);
1297
1298         ++st->rx_bd_errors;
1299         if (ctrl & EMAC_RX_ST_OE)
1300                 ++st->rx_bd_overrun;
1301         if (ctrl & EMAC_RX_ST_BP)
1302                 ++st->rx_bd_bad_packet;
1303         if (ctrl & EMAC_RX_ST_RP)
1304                 ++st->rx_bd_runt_packet;
1305         if (ctrl & EMAC_RX_ST_SE)
1306                 ++st->rx_bd_short_event;
1307         if (ctrl & EMAC_RX_ST_AE)
1308                 ++st->rx_bd_alignment_error;
1309         if (ctrl & EMAC_RX_ST_BFCS)
1310                 ++st->rx_bd_bad_fcs;
1311         if (ctrl & EMAC_RX_ST_PTL)
1312                 ++st->rx_bd_packet_too_long;
1313         if (ctrl & EMAC_RX_ST_ORE)
1314                 ++st->rx_bd_out_of_range;
1315         if (ctrl & EMAC_RX_ST_IRE)
1316                 ++st->rx_bd_in_range;
1317 }
1318
1319 static inline void emac_rx_csum(struct ocp_enet_private *dev,
1320                                 struct sk_buff *skb, u16 ctrl)
1321 {
1322 #if defined(CONFIG_IBM_EMAC_TAH)
1323         if (!ctrl && dev->tah_dev) {
1324                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1325                 ++dev->stats.rx_packets_csum;
1326         }
1327 #endif
1328 }
1329
1330 static inline int emac_rx_sg_append(struct ocp_enet_private *dev, int slot)
1331 {
1332         if (likely(dev->rx_sg_skb != NULL)) {
1333                 int len = dev->rx_desc[slot].data_len;
1334                 int tot_len = dev->rx_sg_skb->len + len;
1335
1336                 if (unlikely(tot_len + 2 > dev->rx_skb_size)) {
1337                         ++dev->estats.rx_dropped_mtu;
1338                         dev_kfree_skb(dev->rx_sg_skb);
1339                         dev->rx_sg_skb = NULL;
1340                 } else {
1341                         cacheable_memcpy(dev->rx_sg_skb->tail,
1342                                          dev->rx_skb[slot]->data, len);
1343                         skb_put(dev->rx_sg_skb, len);
1344                         emac_recycle_rx_skb(dev, slot, len);
1345                         return 0;
1346                 }
1347         }
1348         emac_recycle_rx_skb(dev, slot, 0);
1349         return -1;
1350 }
1351
1352 /* BHs disabled */
1353 static int emac_poll_rx(void *param, int budget)
1354 {
1355         struct ocp_enet_private *dev = param;
1356         int slot = dev->rx_slot, received = 0;
1357
1358         DBG2("%d: poll_rx(%d)" NL, dev->def->index, budget);
1359
1360       again:
1361         while (budget > 0) {
1362                 int len;
1363                 struct sk_buff *skb;
1364                 u16 ctrl = dev->rx_desc[slot].ctrl;
1365
1366                 if (ctrl & MAL_RX_CTRL_EMPTY)
1367                         break;
1368
1369                 skb = dev->rx_skb[slot];
1370                 barrier();
1371                 len = dev->rx_desc[slot].data_len;
1372
1373                 if (unlikely(!MAL_IS_SINGLE_RX(ctrl)))
1374                         goto sg;
1375
1376                 ctrl &= EMAC_BAD_RX_MASK;
1377                 if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1378                         emac_parse_rx_error(dev, ctrl);
1379                         ++dev->estats.rx_dropped_error;
1380                         emac_recycle_rx_skb(dev, slot, 0);
1381                         len = 0;
1382                         goto next;
1383                 }
1384
1385                 if (len && len < EMAC_RX_COPY_THRESH) {
1386                         struct sk_buff *copy_skb =
1387                             alloc_skb(len + EMAC_RX_SKB_HEADROOM + 2, GFP_ATOMIC);
1388                         if (unlikely(!copy_skb))
1389                                 goto oom;
1390
1391                         skb_reserve(copy_skb, EMAC_RX_SKB_HEADROOM + 2);
1392                         cacheable_memcpy(copy_skb->data - 2, skb->data - 2,
1393                                          len + 2);
1394                         emac_recycle_rx_skb(dev, slot, len);
1395                         skb = copy_skb;
1396                 } else if (unlikely(emac_alloc_rx_skb(dev, slot, GFP_ATOMIC)))
1397                         goto oom;
1398
1399                 skb_put(skb, len);
1400               push_packet:
1401                 skb->protocol = eth_type_trans(skb, dev->ndev);
1402                 emac_rx_csum(dev, skb, ctrl);
1403
1404                 if (unlikely(netif_receive_skb(skb) == NET_RX_DROP))
1405                         ++dev->estats.rx_dropped_stack;
1406               next:
1407                 ++dev->stats.rx_packets;
1408               skip:
1409                 dev->stats.rx_bytes += len;
1410                 slot = (slot + 1) % NUM_RX_BUFF;
1411                 --budget;
1412                 ++received;
1413                 continue;
1414               sg:
1415                 if (ctrl & MAL_RX_CTRL_FIRST) {
1416                         BUG_ON(dev->rx_sg_skb);
1417                         if (unlikely(emac_alloc_rx_skb(dev, slot, GFP_ATOMIC))) {
1418                                 DBG("%d: rx OOM %d" NL, dev->def->index, slot);
1419                                 ++dev->estats.rx_dropped_oom;
1420                                 emac_recycle_rx_skb(dev, slot, 0);
1421                         } else {
1422                                 dev->rx_sg_skb = skb;
1423                                 skb_put(skb, len);
1424                         }
1425                 } else if (!emac_rx_sg_append(dev, slot) &&
1426                            (ctrl & MAL_RX_CTRL_LAST)) {
1427
1428                         skb = dev->rx_sg_skb;
1429                         dev->rx_sg_skb = NULL;
1430
1431                         ctrl &= EMAC_BAD_RX_MASK;
1432                         if (unlikely(ctrl && ctrl != EMAC_RX_TAH_BAD_CSUM)) {
1433                                 emac_parse_rx_error(dev, ctrl);
1434                                 ++dev->estats.rx_dropped_error;
1435                                 dev_kfree_skb(skb);
1436                                 len = 0;
1437                         } else
1438                                 goto push_packet;
1439                 }
1440                 goto skip;
1441               oom:
1442                 DBG("%d: rx OOM %d" NL, dev->def->index, slot);
1443                 /* Drop the packet and recycle skb */
1444                 ++dev->estats.rx_dropped_oom;
1445                 emac_recycle_rx_skb(dev, slot, 0);
1446                 goto next;
1447         }
1448
1449         if (received) {
1450                 DBG2("%d: rx %d BDs" NL, dev->def->index, received);
1451                 dev->rx_slot = slot;
1452         }
1453
1454         if (unlikely(budget && dev->commac.rx_stopped)) {
1455                 struct ocp_func_emac_data *emacdata = dev->def->additions;
1456
1457                 barrier();
1458                 if (!(dev->rx_desc[slot].ctrl & MAL_RX_CTRL_EMPTY)) {
1459                         DBG2("%d: rx restart" NL, dev->def->index);
1460                         received = 0;
1461                         goto again;
1462                 }
1463
1464                 if (dev->rx_sg_skb) {
1465                         DBG2("%d: dropping partial rx packet" NL,
1466                              dev->def->index);
1467                         ++dev->estats.rx_dropped_error;
1468                         dev_kfree_skb(dev->rx_sg_skb);
1469                         dev->rx_sg_skb = NULL;
1470                 }
1471
1472                 dev->commac.rx_stopped = 0;
1473                 mal_enable_rx_channel(dev->mal, emacdata->mal_rx_chan);
1474                 emac_rx_enable(dev);
1475                 dev->rx_slot = 0;
1476         }
1477         return received;
1478 }
1479
1480 /* BHs disabled */
1481 static int emac_peek_rx(void *param)
1482 {
1483         struct ocp_enet_private *dev = param;
1484         return !(dev->rx_desc[dev->rx_slot].ctrl & MAL_RX_CTRL_EMPTY);
1485 }
1486
1487 /* BHs disabled */
1488 static int emac_peek_rx_sg(void *param)
1489 {
1490         struct ocp_enet_private *dev = param;
1491         int slot = dev->rx_slot;
1492         while (1) {
1493                 u16 ctrl = dev->rx_desc[slot].ctrl;
1494                 if (ctrl & MAL_RX_CTRL_EMPTY)
1495                         return 0;
1496                 else if (ctrl & MAL_RX_CTRL_LAST)
1497                         return 1;
1498
1499                 slot = (slot + 1) % NUM_RX_BUFF;
1500
1501                 /* I'm just being paranoid here :) */
1502                 if (unlikely(slot == dev->rx_slot))
1503                         return 0;
1504         }
1505 }
1506
1507 /* Hard IRQ */
1508 static void emac_rxde(void *param)
1509 {
1510         struct ocp_enet_private *dev = param;
1511         ++dev->estats.rx_stopped;
1512         emac_rx_disable_async(dev);
1513 }
1514
1515 /* Hard IRQ */
1516 static irqreturn_t emac_irq(int irq, void *dev_instance)
1517 {
1518         struct ocp_enet_private *dev = dev_instance;
1519         struct emac_regs __iomem *p = dev->emacp;
1520         struct ibm_emac_error_stats *st = &dev->estats;
1521
1522         u32 isr = in_be32(&p->isr);
1523         out_be32(&p->isr, isr);
1524
1525         DBG("%d: isr = %08x" NL, dev->def->index, isr);
1526
1527         if (isr & EMAC_ISR_TXPE)
1528                 ++st->tx_parity;
1529         if (isr & EMAC_ISR_RXPE)
1530                 ++st->rx_parity;
1531         if (isr & EMAC_ISR_TXUE)
1532                 ++st->tx_underrun;
1533         if (isr & EMAC_ISR_RXOE)
1534                 ++st->rx_fifo_overrun;
1535         if (isr & EMAC_ISR_OVR)
1536                 ++st->rx_overrun;
1537         if (isr & EMAC_ISR_BP)
1538                 ++st->rx_bad_packet;
1539         if (isr & EMAC_ISR_RP)
1540                 ++st->rx_runt_packet;
1541         if (isr & EMAC_ISR_SE)
1542                 ++st->rx_short_event;
1543         if (isr & EMAC_ISR_ALE)
1544                 ++st->rx_alignment_error;
1545         if (isr & EMAC_ISR_BFCS)
1546                 ++st->rx_bad_fcs;
1547         if (isr & EMAC_ISR_PTLE)
1548                 ++st->rx_packet_too_long;
1549         if (isr & EMAC_ISR_ORE)
1550                 ++st->rx_out_of_range;
1551         if (isr & EMAC_ISR_IRE)
1552                 ++st->rx_in_range;
1553         if (isr & EMAC_ISR_SQE)
1554                 ++st->tx_sqe;
1555         if (isr & EMAC_ISR_TE)
1556                 ++st->tx_errors;
1557
1558         return IRQ_HANDLED;
1559 }
1560
1561 static struct net_device_stats *emac_stats(struct net_device *ndev)
1562 {
1563         struct ocp_enet_private *dev = ndev->priv;
1564         struct ibm_emac_stats *st = &dev->stats;
1565         struct ibm_emac_error_stats *est = &dev->estats;
1566         struct net_device_stats *nst = &dev->nstats;
1567
1568         DBG2("%d: stats" NL, dev->def->index);
1569
1570         /* Compute "legacy" statistics */
1571         local_irq_disable();
1572         nst->rx_packets = (unsigned long)st->rx_packets;
1573         nst->rx_bytes = (unsigned long)st->rx_bytes;
1574         nst->tx_packets = (unsigned long)st->tx_packets;
1575         nst->tx_bytes = (unsigned long)st->tx_bytes;
1576         nst->rx_dropped = (unsigned long)(est->rx_dropped_oom +
1577                                           est->rx_dropped_error +
1578                                           est->rx_dropped_resize +
1579                                           est->rx_dropped_mtu);
1580         nst->tx_dropped = (unsigned long)est->tx_dropped;
1581
1582         nst->rx_errors = (unsigned long)est->rx_bd_errors;
1583         nst->rx_fifo_errors = (unsigned long)(est->rx_bd_overrun +
1584                                               est->rx_fifo_overrun +
1585                                               est->rx_overrun);
1586         nst->rx_frame_errors = (unsigned long)(est->rx_bd_alignment_error +
1587                                                est->rx_alignment_error);
1588         nst->rx_crc_errors = (unsigned long)(est->rx_bd_bad_fcs +
1589                                              est->rx_bad_fcs);
1590         nst->rx_length_errors = (unsigned long)(est->rx_bd_runt_packet +
1591                                                 est->rx_bd_short_event +
1592                                                 est->rx_bd_packet_too_long +
1593                                                 est->rx_bd_out_of_range +
1594                                                 est->rx_bd_in_range +
1595                                                 est->rx_runt_packet +
1596                                                 est->rx_short_event +
1597                                                 est->rx_packet_too_long +
1598                                                 est->rx_out_of_range +
1599                                                 est->rx_in_range);
1600
1601         nst->tx_errors = (unsigned long)(est->tx_bd_errors + est->tx_errors);
1602         nst->tx_fifo_errors = (unsigned long)(est->tx_bd_underrun +
1603                                               est->tx_underrun);
1604         nst->tx_carrier_errors = (unsigned long)est->tx_bd_carrier_loss;
1605         nst->collisions = (unsigned long)(est->tx_bd_excessive_deferral +
1606                                           est->tx_bd_excessive_collisions +
1607                                           est->tx_bd_late_collision +
1608                                           est->tx_bd_multple_collisions);
1609         local_irq_enable();
1610         return nst;
1611 }
1612
1613 static void emac_remove(struct ocp_device *ocpdev)
1614 {
1615         struct ocp_enet_private *dev = ocp_get_drvdata(ocpdev);
1616
1617         DBG("%d: remove" NL, dev->def->index);
1618
1619         ocp_set_drvdata(ocpdev, NULL);
1620         unregister_netdev(dev->ndev);
1621
1622         tah_fini(dev->tah_dev);
1623         rgmii_fini(dev->rgmii_dev, dev->rgmii_input);
1624         zmii_fini(dev->zmii_dev, dev->zmii_input);
1625
1626         emac_dbg_register(dev->def->index, NULL);
1627
1628         mal_unregister_commac(dev->mal, &dev->commac);
1629         iounmap(dev->emacp);
1630         kfree(dev->ndev);
1631 }
1632
1633 static struct mal_commac_ops emac_commac_ops = {
1634         .poll_tx = &emac_poll_tx,
1635         .poll_rx = &emac_poll_rx,
1636         .peek_rx = &emac_peek_rx,
1637         .rxde = &emac_rxde,
1638 };
1639
1640 static struct mal_commac_ops emac_commac_sg_ops = {
1641         .poll_tx = &emac_poll_tx,
1642         .poll_rx = &emac_poll_rx,
1643         .peek_rx = &emac_peek_rx_sg,
1644         .rxde = &emac_rxde,
1645 };
1646
1647 /* Ethtool support */
1648 static int emac_ethtool_get_settings(struct net_device *ndev,
1649                                      struct ethtool_cmd *cmd)
1650 {
1651         struct ocp_enet_private *dev = ndev->priv;
1652
1653         cmd->supported = dev->phy.features;
1654         cmd->port = PORT_MII;
1655         cmd->phy_address = dev->phy.address;
1656         cmd->transceiver =
1657             dev->phy.address >= 0 ? XCVR_EXTERNAL : XCVR_INTERNAL;
1658
1659         local_bh_disable();
1660         cmd->advertising = dev->phy.advertising;
1661         cmd->autoneg = dev->phy.autoneg;
1662         cmd->speed = dev->phy.speed;
1663         cmd->duplex = dev->phy.duplex;
1664         local_bh_enable();
1665
1666         return 0;
1667 }
1668
1669 static int emac_ethtool_set_settings(struct net_device *ndev,
1670                                      struct ethtool_cmd *cmd)
1671 {
1672         struct ocp_enet_private *dev = ndev->priv;
1673         u32 f = dev->phy.features;
1674
1675         DBG("%d: set_settings(%d, %d, %d, 0x%08x)" NL, dev->def->index,
1676             cmd->autoneg, cmd->speed, cmd->duplex, cmd->advertising);
1677
1678         /* Basic sanity checks */
1679         if (dev->phy.address < 0)
1680                 return -EOPNOTSUPP;
1681         if (cmd->autoneg != AUTONEG_ENABLE && cmd->autoneg != AUTONEG_DISABLE)
1682                 return -EINVAL;
1683         if (cmd->autoneg == AUTONEG_ENABLE && cmd->advertising == 0)
1684                 return -EINVAL;
1685         if (cmd->duplex != DUPLEX_HALF && cmd->duplex != DUPLEX_FULL)
1686                 return -EINVAL;
1687
1688         if (cmd->autoneg == AUTONEG_DISABLE) {
1689                 switch (cmd->speed) {
1690                 case SPEED_10:
1691                         if (cmd->duplex == DUPLEX_HALF
1692                             && !(f & SUPPORTED_10baseT_Half))
1693                                 return -EINVAL;
1694                         if (cmd->duplex == DUPLEX_FULL
1695                             && !(f & SUPPORTED_10baseT_Full))
1696                                 return -EINVAL;
1697                         break;
1698                 case SPEED_100:
1699                         if (cmd->duplex == DUPLEX_HALF
1700                             && !(f & SUPPORTED_100baseT_Half))
1701                                 return -EINVAL;
1702                         if (cmd->duplex == DUPLEX_FULL
1703                             && !(f & SUPPORTED_100baseT_Full))
1704                                 return -EINVAL;
1705                         break;
1706                 case SPEED_1000:
1707                         if (cmd->duplex == DUPLEX_HALF
1708                             && !(f & SUPPORTED_1000baseT_Half))
1709                                 return -EINVAL;
1710                         if (cmd->duplex == DUPLEX_FULL
1711                             && !(f & SUPPORTED_1000baseT_Full))
1712                                 return -EINVAL;
1713                         break;
1714                 default:
1715                         return -EINVAL;
1716                 }
1717
1718                 local_bh_disable();
1719                 dev->phy.def->ops->setup_forced(&dev->phy, cmd->speed,
1720                                                 cmd->duplex);
1721
1722         } else {
1723                 if (!(f & SUPPORTED_Autoneg))
1724                         return -EINVAL;
1725
1726                 local_bh_disable();
1727                 dev->phy.def->ops->setup_aneg(&dev->phy,
1728                                               (cmd->advertising & f) |
1729                                               (dev->phy.advertising &
1730                                                (ADVERTISED_Pause |
1731                                                 ADVERTISED_Asym_Pause)));
1732         }
1733         emac_force_link_update(dev);
1734         local_bh_enable();
1735
1736         return 0;
1737 }
1738
1739 static void emac_ethtool_get_ringparam(struct net_device *ndev,
1740                                        struct ethtool_ringparam *rp)
1741 {
1742         rp->rx_max_pending = rp->rx_pending = NUM_RX_BUFF;
1743         rp->tx_max_pending = rp->tx_pending = NUM_TX_BUFF;
1744 }
1745
1746 static void emac_ethtool_get_pauseparam(struct net_device *ndev,
1747                                         struct ethtool_pauseparam *pp)
1748 {
1749         struct ocp_enet_private *dev = ndev->priv;
1750
1751         local_bh_disable();
1752         if ((dev->phy.features & SUPPORTED_Autoneg) &&
1753             (dev->phy.advertising & (ADVERTISED_Pause | ADVERTISED_Asym_Pause)))
1754                 pp->autoneg = 1;
1755
1756         if (dev->phy.duplex == DUPLEX_FULL) {
1757                 if (dev->phy.pause)
1758                         pp->rx_pause = pp->tx_pause = 1;
1759                 else if (dev->phy.asym_pause)
1760                         pp->tx_pause = 1;
1761         }
1762         local_bh_enable();
1763 }
1764
1765 static u32 emac_ethtool_get_rx_csum(struct net_device *ndev)
1766 {
1767         struct ocp_enet_private *dev = ndev->priv;
1768         return dev->tah_dev != 0;
1769 }
1770
1771 static int emac_get_regs_len(struct ocp_enet_private *dev)
1772 {
1773         return sizeof(struct emac_ethtool_regs_subhdr) + EMAC_ETHTOOL_REGS_SIZE;
1774 }
1775
1776 static int emac_ethtool_get_regs_len(struct net_device *ndev)
1777 {
1778         struct ocp_enet_private *dev = ndev->priv;
1779         return sizeof(struct emac_ethtool_regs_hdr) +
1780             emac_get_regs_len(dev) + mal_get_regs_len(dev->mal) +
1781             zmii_get_regs_len(dev->zmii_dev) +
1782             rgmii_get_regs_len(dev->rgmii_dev) +
1783             tah_get_regs_len(dev->tah_dev);
1784 }
1785
1786 static void *emac_dump_regs(struct ocp_enet_private *dev, void *buf)
1787 {
1788         struct emac_ethtool_regs_subhdr *hdr = buf;
1789
1790         hdr->version = EMAC_ETHTOOL_REGS_VER;
1791         hdr->index = dev->def->index;
1792         memcpy_fromio(hdr + 1, dev->emacp, EMAC_ETHTOOL_REGS_SIZE);
1793         return ((void *)(hdr + 1) + EMAC_ETHTOOL_REGS_SIZE);
1794 }
1795
1796 static void emac_ethtool_get_regs(struct net_device *ndev,
1797                                   struct ethtool_regs *regs, void *buf)
1798 {
1799         struct ocp_enet_private *dev = ndev->priv;
1800         struct emac_ethtool_regs_hdr *hdr = buf;
1801
1802         hdr->components = 0;
1803         buf = hdr + 1;
1804
1805         local_irq_disable();
1806         buf = mal_dump_regs(dev->mal, buf);
1807         buf = emac_dump_regs(dev, buf);
1808         if (dev->zmii_dev) {
1809                 hdr->components |= EMAC_ETHTOOL_REGS_ZMII;
1810                 buf = zmii_dump_regs(dev->zmii_dev, buf);
1811         }
1812         if (dev->rgmii_dev) {
1813                 hdr->components |= EMAC_ETHTOOL_REGS_RGMII;
1814                 buf = rgmii_dump_regs(dev->rgmii_dev, buf);
1815         }
1816         if (dev->tah_dev) {
1817                 hdr->components |= EMAC_ETHTOOL_REGS_TAH;
1818                 buf = tah_dump_regs(dev->tah_dev, buf);
1819         }
1820         local_irq_enable();
1821 }
1822
1823 static int emac_ethtool_nway_reset(struct net_device *ndev)
1824 {
1825         struct ocp_enet_private *dev = ndev->priv;
1826         int res = 0;
1827
1828         DBG("%d: nway_reset" NL, dev->def->index);
1829
1830         if (dev->phy.address < 0)
1831                 return -EOPNOTSUPP;
1832
1833         local_bh_disable();
1834         if (!dev->phy.autoneg) {
1835                 res = -EINVAL;
1836                 goto out;
1837         }
1838
1839         dev->phy.def->ops->setup_aneg(&dev->phy, dev->phy.advertising);
1840         emac_force_link_update(dev);
1841
1842       out:
1843         local_bh_enable();
1844         return res;
1845 }
1846
1847 static int emac_ethtool_get_stats_count(struct net_device *ndev)
1848 {
1849         return EMAC_ETHTOOL_STATS_COUNT;
1850 }
1851
1852 static void emac_ethtool_get_strings(struct net_device *ndev, u32 stringset,
1853                                      u8 * buf)
1854 {
1855         if (stringset == ETH_SS_STATS)
1856                 memcpy(buf, &emac_stats_keys, sizeof(emac_stats_keys));
1857 }
1858
1859 static void emac_ethtool_get_ethtool_stats(struct net_device *ndev,
1860                                            struct ethtool_stats *estats,
1861                                            u64 * tmp_stats)
1862 {
1863         struct ocp_enet_private *dev = ndev->priv;
1864         local_irq_disable();
1865         memcpy(tmp_stats, &dev->stats, sizeof(dev->stats));
1866         tmp_stats += sizeof(dev->stats) / sizeof(u64);
1867         memcpy(tmp_stats, &dev->estats, sizeof(dev->estats));
1868         local_irq_enable();
1869 }
1870
1871 static void emac_ethtool_get_drvinfo(struct net_device *ndev,
1872                                      struct ethtool_drvinfo *info)
1873 {
1874         struct ocp_enet_private *dev = ndev->priv;
1875
1876         strcpy(info->driver, "ibm_emac");
1877         strcpy(info->version, DRV_VERSION);
1878         info->fw_version[0] = '\0';
1879         sprintf(info->bus_info, "PPC 4xx EMAC %d", dev->def->index);
1880         info->n_stats = emac_ethtool_get_stats_count(ndev);
1881         info->regdump_len = emac_ethtool_get_regs_len(ndev);
1882 }
1883
1884 static const struct ethtool_ops emac_ethtool_ops = {
1885         .get_settings = emac_ethtool_get_settings,
1886         .set_settings = emac_ethtool_set_settings,
1887         .get_drvinfo = emac_ethtool_get_drvinfo,
1888
1889         .get_regs_len = emac_ethtool_get_regs_len,
1890         .get_regs = emac_ethtool_get_regs,
1891
1892         .nway_reset = emac_ethtool_nway_reset,
1893
1894         .get_ringparam = emac_ethtool_get_ringparam,
1895         .get_pauseparam = emac_ethtool_get_pauseparam,
1896
1897         .get_rx_csum = emac_ethtool_get_rx_csum,
1898
1899         .get_strings = emac_ethtool_get_strings,
1900         .get_stats_count = emac_ethtool_get_stats_count,
1901         .get_ethtool_stats = emac_ethtool_get_ethtool_stats,
1902
1903         .get_link = ethtool_op_get_link,
1904         .get_tx_csum = ethtool_op_get_tx_csum,
1905         .get_sg = ethtool_op_get_sg,
1906 };
1907
1908 static int emac_ioctl(struct net_device *ndev, struct ifreq *rq, int cmd)
1909 {
1910         struct ocp_enet_private *dev = ndev->priv;
1911         uint16_t *data = (uint16_t *) & rq->ifr_ifru;
1912
1913         DBG("%d: ioctl %08x" NL, dev->def->index, cmd);
1914
1915         if (dev->phy.address < 0)
1916                 return -EOPNOTSUPP;
1917
1918         switch (cmd) {
1919         case SIOCGMIIPHY:
1920         case SIOCDEVPRIVATE:
1921                 data[0] = dev->phy.address;
1922                 /* Fall through */
1923         case SIOCGMIIREG:
1924         case SIOCDEVPRIVATE + 1:
1925                 data[3] = emac_mdio_read(ndev, dev->phy.address, data[1]);
1926                 return 0;
1927
1928         case SIOCSMIIREG:
1929         case SIOCDEVPRIVATE + 2:
1930                 if (!capable(CAP_NET_ADMIN))
1931                         return -EPERM;
1932                 emac_mdio_write(ndev, dev->phy.address, data[1], data[2]);
1933                 return 0;
1934         default:
1935                 return -EOPNOTSUPP;
1936         }
1937 }
1938
1939 static int __init emac_probe(struct ocp_device *ocpdev)
1940 {
1941         struct ocp_func_emac_data *emacdata = ocpdev->def->additions;
1942         struct net_device *ndev;
1943         struct ocp_device *maldev;
1944         struct ocp_enet_private *dev;
1945         int err, i;
1946
1947         DBG("%d: probe" NL, ocpdev->def->index);
1948
1949         if (!emacdata) {
1950                 printk(KERN_ERR "emac%d: Missing additional data!\n",
1951                        ocpdev->def->index);
1952                 return -ENODEV;
1953         }
1954
1955         /* Allocate our net_device structure */
1956         ndev = alloc_etherdev(sizeof(struct ocp_enet_private));
1957         if (!ndev) {
1958                 printk(KERN_ERR "emac%d: could not allocate ethernet device!\n",
1959                        ocpdev->def->index);
1960                 return -ENOMEM;
1961         }
1962         dev = ndev->priv;
1963         dev->ndev = ndev;
1964         dev->ldev = &ocpdev->dev;
1965         dev->def = ocpdev->def;
1966         SET_MODULE_OWNER(ndev);
1967
1968         /* Find MAL device we are connected to */
1969         maldev =
1970             ocp_find_device(OCP_VENDOR_IBM, OCP_FUNC_MAL, emacdata->mal_idx);
1971         if (!maldev) {
1972                 printk(KERN_ERR "emac%d: unknown mal%d device!\n",
1973                        dev->def->index, emacdata->mal_idx);
1974                 err = -ENODEV;
1975                 goto out;
1976         }
1977         dev->mal = ocp_get_drvdata(maldev);
1978         if (!dev->mal) {
1979                 printk(KERN_ERR "emac%d: mal%d hasn't been initialized yet!\n",
1980                        dev->def->index, emacdata->mal_idx);
1981                 err = -ENODEV;
1982                 goto out;
1983         }
1984
1985         /* Register with MAL */
1986         dev->commac.ops = &emac_commac_ops;
1987         dev->commac.dev = dev;
1988         dev->commac.tx_chan_mask = MAL_CHAN_MASK(emacdata->mal_tx_chan);
1989         dev->commac.rx_chan_mask = MAL_CHAN_MASK(emacdata->mal_rx_chan);
1990         err = mal_register_commac(dev->mal, &dev->commac);
1991         if (err) {
1992                 printk(KERN_ERR "emac%d: failed to register with mal%d!\n",
1993                        dev->def->index, emacdata->mal_idx);
1994                 goto out;
1995         }
1996         dev->rx_skb_size = emac_rx_skb_size(ndev->mtu);
1997         dev->rx_sync_size = emac_rx_sync_size(ndev->mtu);
1998
1999         /* Get pointers to BD rings */
2000         dev->tx_desc =
2001             dev->mal->bd_virt + mal_tx_bd_offset(dev->mal,
2002                                                  emacdata->mal_tx_chan);
2003         dev->rx_desc =
2004             dev->mal->bd_virt + mal_rx_bd_offset(dev->mal,
2005                                                  emacdata->mal_rx_chan);
2006
2007         DBG("%d: tx_desc %p" NL, ocpdev->def->index, dev->tx_desc);
2008         DBG("%d: rx_desc %p" NL, ocpdev->def->index, dev->rx_desc);
2009
2010         /* Clean rings */
2011         memset(dev->tx_desc, 0, NUM_TX_BUFF * sizeof(struct mal_descriptor));
2012         memset(dev->rx_desc, 0, NUM_RX_BUFF * sizeof(struct mal_descriptor));
2013
2014         /* If we depend on another EMAC for MDIO, check whether it was probed already */
2015         if (emacdata->mdio_idx >= 0 && emacdata->mdio_idx != ocpdev->def->index) {
2016                 struct ocp_device *mdiodev =
2017                     ocp_find_device(OCP_VENDOR_IBM, OCP_FUNC_EMAC,
2018                                     emacdata->mdio_idx);
2019                 if (!mdiodev) {
2020                         printk(KERN_ERR "emac%d: unknown emac%d device!\n",
2021                                dev->def->index, emacdata->mdio_idx);
2022                         err = -ENODEV;
2023                         goto out2;
2024                 }
2025                 dev->mdio_dev = ocp_get_drvdata(mdiodev);
2026                 if (!dev->mdio_dev) {
2027                         printk(KERN_ERR
2028                                "emac%d: emac%d hasn't been initialized yet!\n",
2029                                dev->def->index, emacdata->mdio_idx);
2030                         err = -ENODEV;
2031                         goto out2;
2032                 }
2033         }
2034
2035         /* Attach to ZMII, if needed */
2036         if ((err = zmii_attach(dev)) != 0)
2037                 goto out2;
2038
2039         /* Attach to RGMII, if needed */
2040         if ((err = rgmii_attach(dev)) != 0)
2041                 goto out3;
2042
2043         /* Attach to TAH, if needed */
2044         if ((err = tah_attach(dev)) != 0)
2045                 goto out4;
2046
2047         /* Map EMAC regs */
2048         dev->emacp = ioremap(dev->def->paddr, sizeof(struct emac_regs));
2049         if (!dev->emacp) {
2050                 printk(KERN_ERR "emac%d: could not ioremap device registers!\n",
2051                        dev->def->index);
2052                 err = -ENOMEM;
2053                 goto out5;
2054         }
2055
2056         /* Fill in MAC address */
2057         for (i = 0; i < 6; ++i)
2058                 ndev->dev_addr[i] = emacdata->mac_addr[i];
2059
2060         /* Set some link defaults before we can find out real parameters */
2061         dev->phy.speed = SPEED_100;
2062         dev->phy.duplex = DUPLEX_FULL;
2063         dev->phy.autoneg = AUTONEG_DISABLE;
2064         dev->phy.pause = dev->phy.asym_pause = 0;
2065         dev->stop_timeout = STOP_TIMEOUT_100;
2066         init_timer(&dev->link_timer);
2067         dev->link_timer.function = emac_link_timer;
2068         dev->link_timer.data = (unsigned long)dev;
2069
2070         /* Find PHY if any */
2071         dev->phy.dev = ndev;
2072         dev->phy.mode = emacdata->phy_mode;
2073         if (emacdata->phy_map != 0xffffffff) {
2074                 u32 phy_map = emacdata->phy_map | busy_phy_map;
2075                 u32 adv;
2076
2077                 DBG("%d: PHY maps %08x %08x" NL, dev->def->index,
2078                     emacdata->phy_map, busy_phy_map);
2079
2080                 EMAC_RX_CLK_TX(dev->def->index);
2081
2082                 dev->phy.mdio_read = emac_mdio_read;
2083                 dev->phy.mdio_write = emac_mdio_write;
2084
2085                 /* Configure EMAC with defaults so we can at least use MDIO
2086                  * This is needed mostly for 440GX
2087                  */
2088                 if (emac_phy_gpcs(dev->phy.mode)) {
2089                         /* XXX
2090                          * Make GPCS PHY address equal to EMAC index.
2091                          * We probably should take into account busy_phy_map
2092                          * and/or phy_map here.
2093                          */
2094                         dev->phy.address = dev->def->index;
2095                 }
2096                 
2097                 emac_configure(dev);
2098
2099                 for (i = 0; i < 0x20; phy_map >>= 1, ++i)
2100                         if (!(phy_map & 1)) {
2101                                 int r;
2102                                 busy_phy_map |= 1 << i;
2103
2104                                 /* Quick check if there is a PHY at the address */
2105                                 r = emac_mdio_read(dev->ndev, i, MII_BMCR);
2106                                 if (r == 0xffff || r < 0)
2107                                         continue;
2108                                 if (!mii_phy_probe(&dev->phy, i))
2109                                         break;
2110                         }
2111                 if (i == 0x20) {
2112                         printk(KERN_WARNING "emac%d: can't find PHY!\n",
2113                                dev->def->index);
2114                         goto out6;
2115                 }
2116
2117                 /* Init PHY */
2118                 if (dev->phy.def->ops->init)
2119                         dev->phy.def->ops->init(&dev->phy);
2120                 
2121                 /* Disable any PHY features not supported by the platform */
2122                 dev->phy.def->features &= ~emacdata->phy_feat_exc;
2123
2124                 /* Setup initial link parameters */
2125                 if (dev->phy.features & SUPPORTED_Autoneg) {
2126                         adv = dev->phy.features;
2127 #if !defined(CONFIG_40x)
2128                         adv |= ADVERTISED_Pause | ADVERTISED_Asym_Pause;
2129 #endif
2130                         /* Restart autonegotiation */
2131                         dev->phy.def->ops->setup_aneg(&dev->phy, adv);
2132                 } else {
2133                         u32 f = dev->phy.def->features;
2134                         int speed = SPEED_10, fd = DUPLEX_HALF;
2135
2136                         /* Select highest supported speed/duplex */
2137                         if (f & SUPPORTED_1000baseT_Full) {
2138                                 speed = SPEED_1000;
2139                                 fd = DUPLEX_FULL;
2140                         } else if (f & SUPPORTED_1000baseT_Half)
2141                                 speed = SPEED_1000;
2142                         else if (f & SUPPORTED_100baseT_Full) {
2143                                 speed = SPEED_100;
2144                                 fd = DUPLEX_FULL;
2145                         } else if (f & SUPPORTED_100baseT_Half)
2146                                 speed = SPEED_100;
2147                         else if (f & SUPPORTED_10baseT_Full)
2148                                 fd = DUPLEX_FULL;
2149
2150                         /* Force link parameters */
2151                         dev->phy.def->ops->setup_forced(&dev->phy, speed, fd);
2152                 }
2153         } else {
2154                 emac_reset(dev);
2155
2156                 /* PHY-less configuration.
2157                  * XXX I probably should move these settings to emacdata
2158                  */
2159                 dev->phy.address = -1;
2160                 dev->phy.features = SUPPORTED_100baseT_Full | SUPPORTED_MII;
2161                 dev->phy.pause = 1;
2162         }
2163
2164         /* Fill in the driver function table */
2165         ndev->open = &emac_open;
2166         if (dev->tah_dev) {
2167                 ndev->hard_start_xmit = &emac_start_xmit_sg;
2168                 ndev->features |= NETIF_F_IP_CSUM | NETIF_F_SG;
2169         } else
2170                 ndev->hard_start_xmit = &emac_start_xmit;
2171         ndev->tx_timeout = &emac_full_tx_reset;
2172         ndev->watchdog_timeo = 5 * HZ;
2173         ndev->stop = &emac_close;
2174         ndev->get_stats = &emac_stats;
2175         ndev->set_multicast_list = &emac_set_multicast_list;
2176         ndev->do_ioctl = &emac_ioctl;
2177         if (emac_phy_supports_gige(emacdata->phy_mode)) {
2178                 ndev->change_mtu = &emac_change_mtu;
2179                 dev->commac.ops = &emac_commac_sg_ops;
2180         }
2181         SET_ETHTOOL_OPS(ndev, &emac_ethtool_ops);
2182
2183         netif_carrier_off(ndev);
2184         netif_stop_queue(ndev);
2185
2186         err = register_netdev(ndev);
2187         if (err) {
2188                 printk(KERN_ERR "emac%d: failed to register net device (%d)!\n",
2189                        dev->def->index, err);
2190                 goto out6;
2191         }
2192
2193         ocp_set_drvdata(ocpdev, dev);
2194
2195         printk("%s: emac%d, MAC %02x:%02x:%02x:%02x:%02x:%02x\n",
2196                ndev->name, dev->def->index,
2197                ndev->dev_addr[0], ndev->dev_addr[1], ndev->dev_addr[2],
2198                ndev->dev_addr[3], ndev->dev_addr[4], ndev->dev_addr[5]);
2199
2200         if (dev->phy.address >= 0)
2201                 printk("%s: found %s PHY (0x%02x)\n", ndev->name,
2202                        dev->phy.def->name, dev->phy.address);
2203
2204         emac_dbg_register(dev->def->index, dev);
2205
2206         return 0;
2207       out6:
2208         iounmap(dev->emacp);
2209       out5:
2210         tah_fini(dev->tah_dev);
2211       out4:
2212         rgmii_fini(dev->rgmii_dev, dev->rgmii_input);
2213       out3:
2214         zmii_fini(dev->zmii_dev, dev->zmii_input);
2215       out2:
2216         mal_unregister_commac(dev->mal, &dev->commac);
2217       out:
2218         kfree(ndev);
2219         return err;
2220 }
2221
2222 static struct ocp_device_id emac_ids[] = {
2223         { .vendor = OCP_VENDOR_IBM, .function = OCP_FUNC_EMAC },
2224         { .vendor = OCP_VENDOR_INVALID}
2225 };
2226
2227 static struct ocp_driver emac_driver = {
2228         .name = "emac",
2229         .id_table = emac_ids,
2230         .probe = emac_probe,
2231         .remove = emac_remove,
2232 };
2233
2234 static int __init emac_init(void)
2235 {
2236         printk(KERN_INFO DRV_DESC ", version " DRV_VERSION "\n");
2237
2238         DBG(": init" NL);
2239
2240         if (mal_init())
2241                 return -ENODEV;
2242
2243         EMAC_CLK_INTERNAL;
2244         if (ocp_register_driver(&emac_driver)) {
2245                 EMAC_CLK_EXTERNAL;
2246                 ocp_unregister_driver(&emac_driver);
2247                 mal_exit();
2248                 return -ENODEV;
2249         }
2250         EMAC_CLK_EXTERNAL;
2251
2252         emac_init_debug();
2253         return 0;
2254 }
2255
2256 static void __exit emac_exit(void)
2257 {
2258         DBG(": exit" NL);
2259         ocp_unregister_driver(&emac_driver);
2260         mal_exit();
2261         emac_fini_debug();
2262 }
2263
2264 module_init(emac_init);
2265 module_exit(emac_exit);