1#define VERSION "0.23"
2/* ns83820.c by Benjamin LaHaise with contributions.
3 *
4 * Questions/comments/discussion to linux-ns83820@kvack.org.
5 *
6 * $Revision: 1.34.2.23 $
7 *
8 * Copyright 2001 Benjamin LaHaise.
9 * Copyright 2001, 2002 Red Hat.
10 *
11 * Mmmm, chocolate vanilla mocha...
12 *
13 *
14 * This program is free software; you can redistribute it and/or modify
15 * it under the terms of the GNU General Public License as published by
16 * the Free Software Foundation; either version 2 of the License, or
17 * (at your option) any later version.
18 *
19 * This program is distributed in the hope that it will be useful,
20 * but WITHOUT ANY WARRANTY; without even the implied warranty of
21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22 * GNU General Public License for more details.
23 *
24 * You should have received a copy of the GNU General Public License
25 * along with this program; if not, see <http://www.gnu.org/licenses/>.
26 *
27 *
28 * ChangeLog
29 * =========
30 *	20010414	0.1 - created
31 *	20010622	0.2 - basic rx and tx.
32 *	20010711	0.3 - added duplex and link state detection support.
33 *	20010713	0.4 - zero copy, no hangs.
34 *			0.5 - 64 bit dma support (davem will hate me for this)
35 *			    - disable jumbo frames to avoid tx hangs
36 *			    - work around tx deadlocks on my 1.02 card via
37 *			      fiddling with TXCFG
38 *	20010810	0.6 - use pci dma api for ringbuffers, work on ia64
39 *	20010816	0.7 - misc cleanups
40 *	20010826	0.8 - fix critical zero copy bugs
41 *			0.9 - internal experiment
42 *	20010827	0.10 - fix ia64 unaligned access.
43 *	20010906	0.11 - accept all packets with checksum errors as
44 *			       otherwise fragments get lost
45 *			     - fix >> 32 bugs
46 *			0.12 - add statistics counters
47 *			     - add allmulti/promisc support
48 *	20011009	0.13 - hotplug support, other smaller pci api cleanups
49 *	20011204	0.13a - optical transceiver support added
50 *				by Michael Clark <michael@metaparadigm.com>
51 *	20011205	0.13b - call register_netdev earlier in initialization
52 *				suppress duplicate link status messages
53 *	20011117 	0.14 - ethtool GDRVINFO, GLINK support from jgarzik
54 *	20011204 	0.15	get ppc (big endian) working
55 *	20011218	0.16	various cleanups
56 *	20020310	0.17	speedups
57 *	20020610	0.18 -	actually use the pci dma api for highmem
58 *			     -	remove pci latency register fiddling
59 *			0.19 -	better bist support
60 *			     -	add ihr and reset_phy parameters
61 *			     -	gmii bus probing
62 *			     -	fix missed txok introduced during performance
63 *				tuning
64 *			0.20 -	fix stupid RFEN thinko.  i am such a smurf.
65 *	20040828	0.21 -	add hardware vlan accleration
66 *				by Neil Horman <nhorman@redhat.com>
67 *	20050406	0.22 -	improved DAC ifdefs from Andi Kleen
68 *			     -	removal of dead code from Adrian Bunk
69 *			     -	fix half duplex collision behaviour
70 * Driver Overview
71 * ===============
72 *
73 * This driver was originally written for the National Semiconductor
74 * 83820 chip, a 10/100/1000 Mbps 64 bit PCI ethernet NIC.  Hopefully
75 * this code will turn out to be a) clean, b) correct, and c) fast.
76 * With that in mind, I'm aiming to split the code up as much as
77 * reasonably possible.  At present there are X major sections that
78 * break down into a) packet receive, b) packet transmit, c) link
79 * management, d) initialization and configuration.  Where possible,
80 * these code paths are designed to run in parallel.
81 *
82 * This driver has been tested and found to work with the following
83 * cards (in no particular order):
84 *
85 *	Cameo		SOHO-GA2000T	SOHO-GA2500T
86 *	D-Link		DGE-500T
87 *	PureData	PDP8023Z-TG
88 *	SMC		SMC9452TX	SMC9462TX
89 *	Netgear		GA621
90 *
91 * Special thanks to SMC for providing hardware to test this driver on.
92 *
93 * Reports of success or failure would be greatly appreciated.
94 */
95//#define dprintk		printk
96#define dprintk(x...)		do { } while (0)
97
98#include <linux/module.h>
99#include <linux/moduleparam.h>
100#include <linux/types.h>
101#include <linux/pci.h>
102#include <linux/dma-mapping.h>
103#include <linux/netdevice.h>
104#include <linux/etherdevice.h>
105#include <linux/delay.h>
106#include <linux/workqueue.h>
107#include <linux/init.h>
108#include <linux/interrupt.h>
109#include <linux/ip.h>	/* for iph */
110#include <linux/in.h>	/* for IPPROTO_... */
111#include <linux/compiler.h>
112#include <linux/prefetch.h>
113#include <linux/ethtool.h>
114#include <linux/sched.h>
115#include <linux/timer.h>
116#include <linux/if_vlan.h>
117#include <linux/rtnetlink.h>
118#include <linux/jiffies.h>
119#include <linux/slab.h>
120
121#include <asm/io.h>
122#include <asm/uaccess.h>
123
124#define DRV_NAME "ns83820"
125
126/* Global parameters.  See module_param near the bottom. */
127static int ihr = 2;
128static int reset_phy = 0;
129static int lnksts = 0;		/* CFG_LNKSTS bit polarity */
130
131/* Dprintk is used for more interesting debug events */
132#undef Dprintk
133#define	Dprintk			dprintk
134
135/* tunables */
136#define RX_BUF_SIZE	1500	/* 8192 */
137#if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
138#define NS83820_VLAN_ACCEL_SUPPORT
139#endif
140
141/* Must not exceed ~65000. */
142#define NR_RX_DESC	64
143#define NR_TX_DESC	128
144
145/* not tunable */
146#define REAL_RX_BUF_SIZE (RX_BUF_SIZE + 14)	/* rx/tx mac addr + type */
147
148#define MIN_TX_DESC_FREE	8
149
150/* register defines */
151#define CFGCS		0x04
152
153#define CR_TXE		0x00000001
154#define CR_TXD		0x00000002
155/* Ramit : Here's a tip, don't do a RXD immediately followed by an RXE
156 * The Receive engine skips one descriptor and moves
157 * onto the next one!! */
158#define CR_RXE		0x00000004
159#define CR_RXD		0x00000008
160#define CR_TXR		0x00000010
161#define CR_RXR		0x00000020
162#define CR_SWI		0x00000080
163#define CR_RST		0x00000100
164
165#define PTSCR_EEBIST_FAIL       0x00000001
166#define PTSCR_EEBIST_EN         0x00000002
167#define PTSCR_EELOAD_EN         0x00000004
168#define PTSCR_RBIST_FAIL        0x000001b8
169#define PTSCR_RBIST_DONE        0x00000200
170#define PTSCR_RBIST_EN          0x00000400
171#define PTSCR_RBIST_RST         0x00002000
172
173#define MEAR_EEDI		0x00000001
174#define MEAR_EEDO		0x00000002
175#define MEAR_EECLK		0x00000004
176#define MEAR_EESEL		0x00000008
177#define MEAR_MDIO		0x00000010
178#define MEAR_MDDIR		0x00000020
179#define MEAR_MDC		0x00000040
180
181#define ISR_TXDESC3	0x40000000
182#define ISR_TXDESC2	0x20000000
183#define ISR_TXDESC1	0x10000000
184#define ISR_TXDESC0	0x08000000
185#define ISR_RXDESC3	0x04000000
186#define ISR_RXDESC2	0x02000000
187#define ISR_RXDESC1	0x01000000
188#define ISR_RXDESC0	0x00800000
189#define ISR_TXRCMP	0x00400000
190#define ISR_RXRCMP	0x00200000
191#define ISR_DPERR	0x00100000
192#define ISR_SSERR	0x00080000
193#define ISR_RMABT	0x00040000
194#define ISR_RTABT	0x00020000
195#define ISR_RXSOVR	0x00010000
196#define ISR_HIBINT	0x00008000
197#define ISR_PHY		0x00004000
198#define ISR_PME		0x00002000
199#define ISR_SWI		0x00001000
200#define ISR_MIB		0x00000800
201#define ISR_TXURN	0x00000400
202#define ISR_TXIDLE	0x00000200
203#define ISR_TXERR	0x00000100
204#define ISR_TXDESC	0x00000080
205#define ISR_TXOK	0x00000040
206#define ISR_RXORN	0x00000020
207#define ISR_RXIDLE	0x00000010
208#define ISR_RXEARLY	0x00000008
209#define ISR_RXERR	0x00000004
210#define ISR_RXDESC	0x00000002
211#define ISR_RXOK	0x00000001
212
213#define TXCFG_CSI	0x80000000
214#define TXCFG_HBI	0x40000000
215#define TXCFG_MLB	0x20000000
216#define TXCFG_ATP	0x10000000
217#define TXCFG_ECRETRY	0x00800000
218#define TXCFG_BRST_DIS	0x00080000
219#define TXCFG_MXDMA1024	0x00000000
220#define TXCFG_MXDMA512	0x00700000
221#define TXCFG_MXDMA256	0x00600000
222#define TXCFG_MXDMA128	0x00500000
223#define TXCFG_MXDMA64	0x00400000
224#define TXCFG_MXDMA32	0x00300000
225#define TXCFG_MXDMA16	0x00200000
226#define TXCFG_MXDMA8	0x00100000
227
228#define CFG_LNKSTS	0x80000000
229#define CFG_SPDSTS	0x60000000
230#define CFG_SPDSTS1	0x40000000
231#define CFG_SPDSTS0	0x20000000
232#define CFG_DUPSTS	0x10000000
233#define CFG_TBI_EN	0x01000000
234#define CFG_MODE_1000	0x00400000
235/* Ramit : Dont' ever use AUTO_1000, it never works and is buggy.
236 * Read the Phy response and then configure the MAC accordingly */
237#define CFG_AUTO_1000	0x00200000
238#define CFG_PINT_CTL	0x001c0000
239#define CFG_PINT_DUPSTS	0x00100000
240#define CFG_PINT_LNKSTS	0x00080000
241#define CFG_PINT_SPDSTS	0x00040000
242#define CFG_TMRTEST	0x00020000
243#define CFG_MRM_DIS	0x00010000
244#define CFG_MWI_DIS	0x00008000
245#define CFG_T64ADDR	0x00004000
246#define CFG_PCI64_DET	0x00002000
247#define CFG_DATA64_EN	0x00001000
248#define CFG_M64ADDR	0x00000800
249#define CFG_PHY_RST	0x00000400
250#define CFG_PHY_DIS	0x00000200
251#define CFG_EXTSTS_EN	0x00000100
252#define CFG_REQALG	0x00000080
253#define CFG_SB		0x00000040
254#define CFG_POW		0x00000020
255#define CFG_EXD		0x00000010
256#define CFG_PESEL	0x00000008
257#define CFG_BROM_DIS	0x00000004
258#define CFG_EXT_125	0x00000002
259#define CFG_BEM		0x00000001
260
261#define EXTSTS_UDPPKT	0x00200000
262#define EXTSTS_TCPPKT	0x00080000
263#define EXTSTS_IPPKT	0x00020000
264#define EXTSTS_VPKT	0x00010000
265#define EXTSTS_VTG_MASK	0x0000ffff
266
267#define SPDSTS_POLARITY	(CFG_SPDSTS1 | CFG_SPDSTS0 | CFG_DUPSTS | (lnksts ? CFG_LNKSTS : 0))
268
269#define MIBC_MIBS	0x00000008
270#define MIBC_ACLR	0x00000004
271#define MIBC_FRZ	0x00000002
272#define MIBC_WRN	0x00000001
273
274#define PCR_PSEN	(1 << 31)
275#define PCR_PS_MCAST	(1 << 30)
276#define PCR_PS_DA	(1 << 29)
277#define PCR_STHI_8	(3 << 23)
278#define PCR_STLO_4	(1 << 23)
279#define PCR_FFHI_8K	(3 << 21)
280#define PCR_FFLO_4K	(1 << 21)
281#define PCR_PAUSE_CNT	0xFFFE
282
283#define RXCFG_AEP	0x80000000
284#define RXCFG_ARP	0x40000000
285#define RXCFG_STRIPCRC	0x20000000
286#define RXCFG_RX_FD	0x10000000
287#define RXCFG_ALP	0x08000000
288#define RXCFG_AIRL	0x04000000
289#define RXCFG_MXDMA512	0x00700000
290#define RXCFG_DRTH	0x0000003e
291#define RXCFG_DRTH0	0x00000002
292
293#define RFCR_RFEN	0x80000000
294#define RFCR_AAB	0x40000000
295#define RFCR_AAM	0x20000000
296#define RFCR_AAU	0x10000000
297#define RFCR_APM	0x08000000
298#define RFCR_APAT	0x07800000
299#define RFCR_APAT3	0x04000000
300#define RFCR_APAT2	0x02000000
301#define RFCR_APAT1	0x01000000
302#define RFCR_APAT0	0x00800000
303#define RFCR_AARP	0x00400000
304#define RFCR_MHEN	0x00200000
305#define RFCR_UHEN	0x00100000
306#define RFCR_ULM	0x00080000
307
308#define VRCR_RUDPE	0x00000080
309#define VRCR_RTCPE	0x00000040
310#define VRCR_RIPE	0x00000020
311#define VRCR_IPEN	0x00000010
312#define VRCR_DUTF	0x00000008
313#define VRCR_DVTF	0x00000004
314#define VRCR_VTREN	0x00000002
315#define VRCR_VTDEN	0x00000001
316
317#define VTCR_PPCHK	0x00000008
318#define VTCR_GCHK	0x00000004
319#define VTCR_VPPTI	0x00000002
320#define VTCR_VGTI	0x00000001
321
322#define CR		0x00
323#define CFG		0x04
324#define MEAR		0x08
325#define PTSCR		0x0c
326#define	ISR		0x10
327#define	IMR		0x14
328#define	IER		0x18
329#define	IHR		0x1c
330#define TXDP		0x20
331#define TXDP_HI		0x24
332#define TXCFG		0x28
333#define GPIOR		0x2c
334#define RXDP		0x30
335#define RXDP_HI		0x34
336#define RXCFG		0x38
337#define PQCR		0x3c
338#define WCSR		0x40
339#define PCR		0x44
340#define RFCR		0x48
341#define RFDR		0x4c
342
343#define SRR		0x58
344
345#define VRCR		0xbc
346#define VTCR		0xc0
347#define VDR		0xc4
348#define CCSR		0xcc
349
350#define TBICR		0xe0
351#define TBISR		0xe4
352#define TANAR		0xe8
353#define TANLPAR		0xec
354#define TANER		0xf0
355#define TESR		0xf4
356
357#define TBICR_MR_AN_ENABLE	0x00001000
358#define TBICR_MR_RESTART_AN	0x00000200
359
360#define TBISR_MR_LINK_STATUS	0x00000020
361#define TBISR_MR_AN_COMPLETE	0x00000004
362
363#define TANAR_PS2 		0x00000100
364#define TANAR_PS1 		0x00000080
365#define TANAR_HALF_DUP 		0x00000040
366#define TANAR_FULL_DUP 		0x00000020
367
368#define GPIOR_GP5_OE		0x00000200
369#define GPIOR_GP4_OE		0x00000100
370#define GPIOR_GP3_OE		0x00000080
371#define GPIOR_GP2_OE		0x00000040
372#define GPIOR_GP1_OE		0x00000020
373#define GPIOR_GP3_OUT		0x00000004
374#define GPIOR_GP1_OUT		0x00000001
375
376#define LINK_AUTONEGOTIATE	0x01
377#define LINK_DOWN		0x02
378#define LINK_UP			0x04
379
380#define HW_ADDR_LEN	sizeof(dma_addr_t)
381#define desc_addr_set(desc, addr)				\
382	do {							\
383		((desc)[0] = cpu_to_le32(addr));		\
384		if (HW_ADDR_LEN == 8)		 		\
385			(desc)[1] = cpu_to_le32(((u64)addr) >> 32);	\
386	} while(0)
387#define desc_addr_get(desc)					\
388	(le32_to_cpu((desc)[0]) | \
389	(HW_ADDR_LEN == 8 ? ((dma_addr_t)le32_to_cpu((desc)[1]))<<32 : 0))
390
391#define DESC_LINK		0
392#define DESC_BUFPTR		(DESC_LINK + HW_ADDR_LEN/4)
393#define DESC_CMDSTS		(DESC_BUFPTR + HW_ADDR_LEN/4)
394#define DESC_EXTSTS		(DESC_CMDSTS + 4/4)
395
396#define CMDSTS_OWN	0x80000000
397#define CMDSTS_MORE	0x40000000
398#define CMDSTS_INTR	0x20000000
399#define CMDSTS_ERR	0x10000000
400#define CMDSTS_OK	0x08000000
401#define CMDSTS_RUNT	0x00200000
402#define CMDSTS_LEN_MASK	0x0000ffff
403
404#define CMDSTS_DEST_MASK	0x01800000
405#define CMDSTS_DEST_SELF	0x00800000
406#define CMDSTS_DEST_MULTI	0x01000000
407
408#define DESC_SIZE	8		/* Should be cache line sized */
409
410struct rx_info {
411	spinlock_t	lock;
412	int		up;
413	unsigned long	idle;
414
415	struct sk_buff	*skbs[NR_RX_DESC];
416
417	__le32		*next_rx_desc;
418	u16		next_rx, next_empty;
419
420	__le32		*descs;
421	dma_addr_t	phy_descs;
422};
423
424
425struct ns83820 {
426	u8			__iomem *base;
427
428	struct pci_dev		*pci_dev;
429	struct net_device	*ndev;
430
431	struct rx_info		rx_info;
432	struct tasklet_struct	rx_tasklet;
433
434	unsigned		ihr;
435	struct work_struct	tq_refill;
436
437	/* protects everything below.  irqsave when using. */
438	spinlock_t		misc_lock;
439
440	u32			CFG_cache;
441
442	u32			MEAR_cache;
443	u32			IMR_cache;
444
445	unsigned		linkstate;
446
447	spinlock_t	tx_lock;
448
449	u16		tx_done_idx;
450	u16		tx_idx;
451	volatile u16	tx_free_idx;	/* idx of free desc chain */
452	u16		tx_intr_idx;
453
454	atomic_t	nr_tx_skbs;
455	struct sk_buff	*tx_skbs[NR_TX_DESC];
456
457	char		pad[16] __attribute__((aligned(16)));
458	__le32		*tx_descs;
459	dma_addr_t	tx_phy_descs;
460
461	struct timer_list	tx_watchdog;
462};
463
464static inline struct ns83820 *PRIV(struct net_device *dev)
465{
466	return netdev_priv(dev);
467}
468
469#define __kick_rx(dev)	writel(CR_RXE, dev->base + CR)
470
471static inline void kick_rx(struct net_device *ndev)
472{
473	struct ns83820 *dev = PRIV(ndev);
474	dprintk("kick_rx: maybe kicking\n");
475	if (test_and_clear_bit(0, &dev->rx_info.idle)) {
476		dprintk("actually kicking\n");
477		writel(dev->rx_info.phy_descs +
478			(4 * DESC_SIZE * dev->rx_info.next_rx),
479		       dev->base + RXDP);
480		if (dev->rx_info.next_rx == dev->rx_info.next_empty)
481			printk(KERN_DEBUG "%s: uh-oh: next_rx == next_empty???\n",
482				ndev->name);
483		__kick_rx(dev);
484	}
485}
486
487//free = (tx_done_idx + NR_TX_DESC-2 - free_idx) % NR_TX_DESC
488#define start_tx_okay(dev)	\
489	(((NR_TX_DESC-2 + dev->tx_done_idx - dev->tx_free_idx) % NR_TX_DESC) > MIN_TX_DESC_FREE)
490
491/* Packet Receiver
492 *
493 * The hardware supports linked lists of receive descriptors for
494 * which ownership is transferred back and forth by means of an
495 * ownership bit.  While the hardware does support the use of a
496 * ring for receive descriptors, we only make use of a chain in
497 * an attempt to reduce bus traffic under heavy load scenarios.
498 * This will also make bugs a bit more obvious.  The current code
499 * only makes use of a single rx chain; I hope to implement
500 * priority based rx for version 1.0.  Goal: even under overload
501 * conditions, still route realtime traffic with as low jitter as
502 * possible.
503 */
504static inline void build_rx_desc(struct ns83820 *dev, __le32 *desc, dma_addr_t link, dma_addr_t buf, u32 cmdsts, u32 extsts)
505{
506	desc_addr_set(desc + DESC_LINK, link);
507	desc_addr_set(desc + DESC_BUFPTR, buf);
508	desc[DESC_EXTSTS] = cpu_to_le32(extsts);
509	mb();
510	desc[DESC_CMDSTS] = cpu_to_le32(cmdsts);
511}
512
513#define nr_rx_empty(dev) ((NR_RX_DESC-2 + dev->rx_info.next_rx - dev->rx_info.next_empty) % NR_RX_DESC)
514static inline int ns83820_add_rx_skb(struct ns83820 *dev, struct sk_buff *skb)
515{
516	unsigned next_empty;
517	u32 cmdsts;
518	__le32 *sg;
519	dma_addr_t buf;
520
521	next_empty = dev->rx_info.next_empty;
522
523	/* don't overrun last rx marker */
524	if (unlikely(nr_rx_empty(dev) <= 2)) {
525		kfree_skb(skb);
526		return 1;
527	}
528
529#if 0
530	dprintk("next_empty[%d] nr_used[%d] next_rx[%d]\n",
531		dev->rx_info.next_empty,
532		dev->rx_info.nr_used,
533		dev->rx_info.next_rx
534		);
535#endif
536
537	sg = dev->rx_info.descs + (next_empty * DESC_SIZE);
538	BUG_ON(NULL != dev->rx_info.skbs[next_empty]);
539	dev->rx_info.skbs[next_empty] = skb;
540
541	dev->rx_info.next_empty = (next_empty + 1) % NR_RX_DESC;
542	cmdsts = REAL_RX_BUF_SIZE | CMDSTS_INTR;
543	buf = pci_map_single(dev->pci_dev, skb->data,
544			     REAL_RX_BUF_SIZE, PCI_DMA_FROMDEVICE);
545	build_rx_desc(dev, sg, 0, buf, cmdsts, 0);
546	/* update link of previous rx */
547	if (likely(next_empty != dev->rx_info.next_rx))
548		dev->rx_info.descs[((NR_RX_DESC + next_empty - 1) % NR_RX_DESC) * DESC_SIZE] = cpu_to_le32(dev->rx_info.phy_descs + (next_empty * DESC_SIZE * 4));
549
550	return 0;
551}
552
553static inline int rx_refill(struct net_device *ndev, gfp_t gfp)
554{
555	struct ns83820 *dev = PRIV(ndev);
556	unsigned i;
557	unsigned long flags = 0;
558
559	if (unlikely(nr_rx_empty(dev) <= 2))
560		return 0;
561
562	dprintk("rx_refill(%p)\n", ndev);
563	if (gfp == GFP_ATOMIC)
564		spin_lock_irqsave(&dev->rx_info.lock, flags);
565	for (i=0; i<NR_RX_DESC; i++) {
566		struct sk_buff *skb;
567		long res;
568
569		/* extra 16 bytes for alignment */
570		skb = __netdev_alloc_skb(ndev, REAL_RX_BUF_SIZE+16, gfp);
571		if (unlikely(!skb))
572			break;
573
574		skb_reserve(skb, skb->data - PTR_ALIGN(skb->data, 16));
575		if (gfp != GFP_ATOMIC)
576			spin_lock_irqsave(&dev->rx_info.lock, flags);
577		res = ns83820_add_rx_skb(dev, skb);
578		if (gfp != GFP_ATOMIC)
579			spin_unlock_irqrestore(&dev->rx_info.lock, flags);
580		if (res) {
581			i = 1;
582			break;
583		}
584	}
585	if (gfp == GFP_ATOMIC)
586		spin_unlock_irqrestore(&dev->rx_info.lock, flags);
587
588	return i ? 0 : -ENOMEM;
589}
590
591static void rx_refill_atomic(struct net_device *ndev)
592{
593	rx_refill(ndev, GFP_ATOMIC);
594}
595
596/* REFILL */
597static inline void queue_refill(struct work_struct *work)
598{
599	struct ns83820 *dev = container_of(work, struct ns83820, tq_refill);
600	struct net_device *ndev = dev->ndev;
601
602	rx_refill(ndev, GFP_KERNEL);
603	if (dev->rx_info.up)
604		kick_rx(ndev);
605}
606
607static inline void clear_rx_desc(struct ns83820 *dev, unsigned i)
608{
609	build_rx_desc(dev, dev->rx_info.descs + (DESC_SIZE * i), 0, 0, CMDSTS_OWN, 0);
610}
611
612static void phy_intr(struct net_device *ndev)
613{
614	struct ns83820 *dev = PRIV(ndev);
615	static const char *speeds[] = { "10", "100", "1000", "1000(?)", "1000F" };
616	u32 cfg, new_cfg;
617	u32 tbisr, tanar, tanlpar;
618	int speed, fullduplex, newlinkstate;
619
620	cfg = readl(dev->base + CFG) ^ SPDSTS_POLARITY;
621
622	if (dev->CFG_cache & CFG_TBI_EN) {
623		/* we have an optical transceiver */
624		tbisr = readl(dev->base + TBISR);
625		tanar = readl(dev->base + TANAR);
626		tanlpar = readl(dev->base + TANLPAR);
627		dprintk("phy_intr: tbisr=%08x, tanar=%08x, tanlpar=%08x\n",
628			tbisr, tanar, tanlpar);
629
630		if ( (fullduplex = (tanlpar & TANAR_FULL_DUP) &&
631		      (tanar & TANAR_FULL_DUP)) ) {
632
633			/* both of us are full duplex */
634			writel(readl(dev->base + TXCFG)
635			       | TXCFG_CSI | TXCFG_HBI | TXCFG_ATP,
636			       dev->base + TXCFG);
637			writel(readl(dev->base + RXCFG) | RXCFG_RX_FD,
638			       dev->base + RXCFG);
639			/* Light up full duplex LED */
640			writel(readl(dev->base + GPIOR) | GPIOR_GP1_OUT,
641			       dev->base + GPIOR);
642
643		} else if (((tanlpar & TANAR_HALF_DUP) &&
644			    (tanar & TANAR_HALF_DUP)) ||
645			   ((tanlpar & TANAR_FULL_DUP) &&
646			    (tanar & TANAR_HALF_DUP)) ||
647			   ((tanlpar & TANAR_HALF_DUP) &&
648			    (tanar & TANAR_FULL_DUP))) {
649
650			/* one or both of us are half duplex */
651			writel((readl(dev->base + TXCFG)
652				& ~(TXCFG_CSI | TXCFG_HBI)) | TXCFG_ATP,
653			       dev->base + TXCFG);
654			writel(readl(dev->base + RXCFG) & ~RXCFG_RX_FD,
655			       dev->base + RXCFG);
656			/* Turn off full duplex LED */
657			writel(readl(dev->base + GPIOR) & ~GPIOR_GP1_OUT,
658			       dev->base + GPIOR);
659		}
660
661		speed = 4; /* 1000F */
662
663	} else {
664		/* we have a copper transceiver */
665		new_cfg = dev->CFG_cache & ~(CFG_SB | CFG_MODE_1000 | CFG_SPDSTS);
666
667		if (cfg & CFG_SPDSTS1)
668			new_cfg |= CFG_MODE_1000;
669		else
670			new_cfg &= ~CFG_MODE_1000;
671
672		speed = ((cfg / CFG_SPDSTS0) & 3);
673		fullduplex = (cfg & CFG_DUPSTS);
674
675		if (fullduplex) {
676			new_cfg |= CFG_SB;
677			writel(readl(dev->base + TXCFG)
678					| TXCFG_CSI | TXCFG_HBI,
679			       dev->base + TXCFG);
680			writel(readl(dev->base + RXCFG) | RXCFG_RX_FD,
681			       dev->base + RXCFG);
682		} else {
683			writel(readl(dev->base + TXCFG)
684					& ~(TXCFG_CSI | TXCFG_HBI),
685			       dev->base + TXCFG);
686			writel(readl(dev->base + RXCFG) & ~(RXCFG_RX_FD),
687			       dev->base + RXCFG);
688		}
689
690		if ((cfg & CFG_LNKSTS) &&
691		    ((new_cfg ^ dev->CFG_cache) != 0)) {
692			writel(new_cfg, dev->base + CFG);
693			dev->CFG_cache = new_cfg;
694		}
695
696		dev->CFG_cache &= ~CFG_SPDSTS;
697		dev->CFG_cache |= cfg & CFG_SPDSTS;
698	}
699
700	newlinkstate = (cfg & CFG_LNKSTS) ? LINK_UP : LINK_DOWN;
701
702	if (newlinkstate & LINK_UP &&
703	    dev->linkstate != newlinkstate) {
704		netif_start_queue(ndev);
705		netif_wake_queue(ndev);
706		printk(KERN_INFO "%s: link now %s mbps, %s duplex and up.\n",
707			ndev->name,
708			speeds[speed],
709			fullduplex ? "full" : "half");
710	} else if (newlinkstate & LINK_DOWN &&
711		   dev->linkstate != newlinkstate) {
712		netif_stop_queue(ndev);
713		printk(KERN_INFO "%s: link now down.\n", ndev->name);
714	}
715
716	dev->linkstate = newlinkstate;
717}
718
719static int ns83820_setup_rx(struct net_device *ndev)
720{
721	struct ns83820 *dev = PRIV(ndev);
722	unsigned i;
723	int ret;
724
725	dprintk("ns83820_setup_rx(%p)\n", ndev);
726
727	dev->rx_info.idle = 1;
728	dev->rx_info.next_rx = 0;
729	dev->rx_info.next_rx_desc = dev->rx_info.descs;
730	dev->rx_info.next_empty = 0;
731
732	for (i=0; i<NR_RX_DESC; i++)
733		clear_rx_desc(dev, i);
734
735	writel(0, dev->base + RXDP_HI);
736	writel(dev->rx_info.phy_descs, dev->base + RXDP);
737
738	ret = rx_refill(ndev, GFP_KERNEL);
739	if (!ret) {
740		dprintk("starting receiver\n");
741		/* prevent the interrupt handler from stomping on us */
742		spin_lock_irq(&dev->rx_info.lock);
743
744		writel(0x0001, dev->base + CCSR);
745		writel(0, dev->base + RFCR);
746		writel(0x7fc00000, dev->base + RFCR);
747		writel(0xffc00000, dev->base + RFCR);
748
749		dev->rx_info.up = 1;
750
751		phy_intr(ndev);
752
753		/* Okay, let it rip */
754		spin_lock(&dev->misc_lock);
755		dev->IMR_cache |= ISR_PHY;
756		dev->IMR_cache |= ISR_RXRCMP;
757		//dev->IMR_cache |= ISR_RXERR;
758		//dev->IMR_cache |= ISR_RXOK;
759		dev->IMR_cache |= ISR_RXORN;
760		dev->IMR_cache |= ISR_RXSOVR;
761		dev->IMR_cache |= ISR_RXDESC;
762		dev->IMR_cache |= ISR_RXIDLE;
763		dev->IMR_cache |= ISR_TXDESC;
764		dev->IMR_cache |= ISR_TXIDLE;
765
766		writel(dev->IMR_cache, dev->base + IMR);
767		writel(1, dev->base + IER);
768		spin_unlock(&dev->misc_lock);
769
770		kick_rx(ndev);
771
772		spin_unlock_irq(&dev->rx_info.lock);
773	}
774	return ret;
775}
776
777static void ns83820_cleanup_rx(struct ns83820 *dev)
778{
779	unsigned i;
780	unsigned long flags;
781
782	dprintk("ns83820_cleanup_rx(%p)\n", dev);
783
784	/* disable receive interrupts */
785	spin_lock_irqsave(&dev->misc_lock, flags);
786	dev->IMR_cache &= ~(ISR_RXOK | ISR_RXDESC | ISR_RXERR | ISR_RXEARLY | ISR_RXIDLE);
787	writel(dev->IMR_cache, dev->base + IMR);
788	spin_unlock_irqrestore(&dev->misc_lock, flags);
789
790	/* synchronize with the interrupt handler and kill it */
791	dev->rx_info.up = 0;
792	synchronize_irq(dev->pci_dev->irq);
793
794	/* touch the pci bus... */
795	readl(dev->base + IMR);
796
797	/* assumes the transmitter is already disabled and reset */
798	writel(0, dev->base + RXDP_HI);
799	writel(0, dev->base + RXDP);
800
801	for (i=0; i<NR_RX_DESC; i++) {
802		struct sk_buff *skb = dev->rx_info.skbs[i];
803		dev->rx_info.skbs[i] = NULL;
804		clear_rx_desc(dev, i);
805		kfree_skb(skb);
806	}
807}
808
809static void ns83820_rx_kick(struct net_device *ndev)
810{
811	struct ns83820 *dev = PRIV(ndev);
812	/*if (nr_rx_empty(dev) >= NR_RX_DESC/4)*/ {
813		if (dev->rx_info.up) {
814			rx_refill_atomic(ndev);
815			kick_rx(ndev);
816		}
817	}
818
819	if (dev->rx_info.up && nr_rx_empty(dev) > NR_RX_DESC*3/4)
820		schedule_work(&dev->tq_refill);
821	else
822		kick_rx(ndev);
823	if (dev->rx_info.idle)
824		printk(KERN_DEBUG "%s: BAD\n", ndev->name);
825}
826
827/* rx_irq
828 *
829 */
830static void rx_irq(struct net_device *ndev)
831{
832	struct ns83820 *dev = PRIV(ndev);
833	struct rx_info *info = &dev->rx_info;
834	unsigned next_rx;
835	int rx_rc, len;
836	u32 cmdsts;
837	__le32 *desc;
838	unsigned long flags;
839	int nr = 0;
840
841	dprintk("rx_irq(%p)\n", ndev);
842	dprintk("rxdp: %08x, descs: %08lx next_rx[%d]: %p next_empty[%d]: %p\n",
843		readl(dev->base + RXDP),
844		(long)(dev->rx_info.phy_descs),
845		(int)dev->rx_info.next_rx,
846		(dev->rx_info.descs + (DESC_SIZE * dev->rx_info.next_rx)),
847		(int)dev->rx_info.next_empty,
848		(dev->rx_info.descs + (DESC_SIZE * dev->rx_info.next_empty))
849		);
850
851	spin_lock_irqsave(&info->lock, flags);
852	if (!info->up)
853		goto out;
854
855	dprintk("walking descs\n");
856	next_rx = info->next_rx;
857	desc = info->next_rx_desc;
858	while ((CMDSTS_OWN & (cmdsts = le32_to_cpu(desc[DESC_CMDSTS]))) &&
859	       (cmdsts != CMDSTS_OWN)) {
860		struct sk_buff *skb;
861		u32 extsts = le32_to_cpu(desc[DESC_EXTSTS]);
862		dma_addr_t bufptr = desc_addr_get(desc + DESC_BUFPTR);
863
864		dprintk("cmdsts: %08x\n", cmdsts);
865		dprintk("link: %08x\n", cpu_to_le32(desc[DESC_LINK]));
866		dprintk("extsts: %08x\n", extsts);
867
868		skb = info->skbs[next_rx];
869		info->skbs[next_rx] = NULL;
870		info->next_rx = (next_rx + 1) % NR_RX_DESC;
871
872		mb();
873		clear_rx_desc(dev, next_rx);
874
875		pci_unmap_single(dev->pci_dev, bufptr,
876				 RX_BUF_SIZE, PCI_DMA_FROMDEVICE);
877		len = cmdsts & CMDSTS_LEN_MASK;
878#ifdef NS83820_VLAN_ACCEL_SUPPORT
879		/* NH: As was mentioned below, this chip is kinda
880		 * brain dead about vlan tag stripping.  Frames
881		 * that are 64 bytes with a vlan header appended
882		 * like arp frames, or pings, are flagged as Runts
883		 * when the tag is stripped and hardware.  This
884		 * also means that the OK bit in the descriptor
885		 * is cleared when the frame comes in so we have
886		 * to do a specific length check here to make sure
887		 * the frame would have been ok, had we not stripped
888		 * the tag.
889		 */
890		if (likely((CMDSTS_OK & cmdsts) ||
891			((cmdsts & CMDSTS_RUNT) && len >= 56))) {
892#else
893		if (likely(CMDSTS_OK & cmdsts)) {
894#endif
895			skb_put(skb, len);
896			if (unlikely(!skb))
897				goto netdev_mangle_me_harder_failed;
898			if (cmdsts & CMDSTS_DEST_MULTI)
899				ndev->stats.multicast++;
900			ndev->stats.rx_packets++;
901			ndev->stats.rx_bytes += len;
902			if ((extsts & 0x002a0000) && !(extsts & 0x00540000)) {
903				skb->ip_summed = CHECKSUM_UNNECESSARY;
904			} else {
905				skb_checksum_none_assert(skb);
906			}
907			skb->protocol = eth_type_trans(skb, ndev);
908#ifdef NS83820_VLAN_ACCEL_SUPPORT
909			if(extsts & EXTSTS_VPKT) {
910				unsigned short tag;
911
912				tag = ntohs(extsts & EXTSTS_VTG_MASK);
913				__vlan_hwaccel_put_tag(skb, htons(ETH_P_IPV6), tag);
914			}
915#endif
916			rx_rc = netif_rx(skb);
917			if (NET_RX_DROP == rx_rc) {
918netdev_mangle_me_harder_failed:
919				ndev->stats.rx_dropped++;
920			}
921		} else {
922			kfree_skb(skb);
923		}
924
925		nr++;
926		next_rx = info->next_rx;
927		desc = info->descs + (DESC_SIZE * next_rx);
928	}
929	info->next_rx = next_rx;
930	info->next_rx_desc = info->descs + (DESC_SIZE * next_rx);
931
932out:
933	if (0 && !nr) {
934		Dprintk("dazed: cmdsts_f: %08x\n", cmdsts);
935	}
936
937	spin_unlock_irqrestore(&info->lock, flags);
938}
939
940static void rx_action(unsigned long _dev)
941{
942	struct net_device *ndev = (void *)_dev;
943	struct ns83820 *dev = PRIV(ndev);
944	rx_irq(ndev);
945	writel(ihr, dev->base + IHR);
946
947	spin_lock_irq(&dev->misc_lock);
948	dev->IMR_cache |= ISR_RXDESC;
949	writel(dev->IMR_cache, dev->base + IMR);
950	spin_unlock_irq(&dev->misc_lock);
951
952	rx_irq(ndev);
953	ns83820_rx_kick(ndev);
954}
955
956/* Packet Transmit code
957 */
958static inline void kick_tx(struct ns83820 *dev)
959{
960	dprintk("kick_tx(%p): tx_idx=%d free_idx=%d\n",
961		dev, dev->tx_idx, dev->tx_free_idx);
962	writel(CR_TXE, dev->base + CR);
963}
964
965/* No spinlock needed on the transmit irq path as the interrupt handler is
966 * serialized.
967 */
968static void do_tx_done(struct net_device *ndev)
969{
970	struct ns83820 *dev = PRIV(ndev);
971	u32 cmdsts, tx_done_idx;
972	__le32 *desc;
973
974	dprintk("do_tx_done(%p)\n", ndev);
975	tx_done_idx = dev->tx_done_idx;
976	desc = dev->tx_descs + (tx_done_idx * DESC_SIZE);
977
978	dprintk("tx_done_idx=%d free_idx=%d cmdsts=%08x\n",
979		tx_done_idx, dev->tx_free_idx, le32_to_cpu(desc[DESC_CMDSTS]));
980	while ((tx_done_idx != dev->tx_free_idx) &&
981	       !(CMDSTS_OWN & (cmdsts = le32_to_cpu(desc[DESC_CMDSTS]))) ) {
982		struct sk_buff *skb;
983		unsigned len;
984		dma_addr_t addr;
985
986		if (cmdsts & CMDSTS_ERR)
987			ndev->stats.tx_errors++;
988		if (cmdsts & CMDSTS_OK)
989			ndev->stats.tx_packets++;
990		if (cmdsts & CMDSTS_OK)
991			ndev->stats.tx_bytes += cmdsts & 0xffff;
992
993		dprintk("tx_done_idx=%d free_idx=%d cmdsts=%08x\n",
994			tx_done_idx, dev->tx_free_idx, cmdsts);
995		skb = dev->tx_skbs[tx_done_idx];
996		dev->tx_skbs[tx_done_idx] = NULL;
997		dprintk("done(%p)\n", skb);
998
999		len = cmdsts & CMDSTS_LEN_MASK;
1000		addr = desc_addr_get(desc + DESC_BUFPTR);
1001		if (skb) {
1002			pci_unmap_single(dev->pci_dev,
1003					addr,
1004					len,
1005					PCI_DMA_TODEVICE);
1006			dev_kfree_skb_irq(skb);
1007			atomic_dec(&dev->nr_tx_skbs);
1008		} else
1009			pci_unmap_page(dev->pci_dev,
1010					addr,
1011					len,
1012					PCI_DMA_TODEVICE);
1013
1014		tx_done_idx = (tx_done_idx + 1) % NR_TX_DESC;
1015		dev->tx_done_idx = tx_done_idx;
1016		desc[DESC_CMDSTS] = cpu_to_le32(0);
1017		mb();
1018		desc = dev->tx_descs + (tx_done_idx * DESC_SIZE);
1019	}
1020
1021	/* Allow network stack to resume queueing packets after we've
1022	 * finished transmitting at least 1/4 of the packets in the queue.
1023	 */
1024	if (netif_queue_stopped(ndev) && start_tx_okay(dev)) {
1025		dprintk("start_queue(%p)\n", ndev);
1026		netif_start_queue(ndev);
1027		netif_wake_queue(ndev);
1028	}
1029}
1030
1031static void ns83820_cleanup_tx(struct ns83820 *dev)
1032{
1033	unsigned i;
1034
1035	for (i=0; i<NR_TX_DESC; i++) {
1036		struct sk_buff *skb = dev->tx_skbs[i];
1037		dev->tx_skbs[i] = NULL;
1038		if (skb) {
1039			__le32 *desc = dev->tx_descs + (i * DESC_SIZE);
1040			pci_unmap_single(dev->pci_dev,
1041					desc_addr_get(desc + DESC_BUFPTR),
1042					le32_to_cpu(desc[DESC_CMDSTS]) & CMDSTS_LEN_MASK,
1043					PCI_DMA_TODEVICE);
1044			dev_kfree_skb_irq(skb);
1045			atomic_dec(&dev->nr_tx_skbs);
1046		}
1047	}
1048
1049	memset(dev->tx_descs, 0, NR_TX_DESC * DESC_SIZE * 4);
1050}
1051
1052/* transmit routine.  This code relies on the network layer serializing
1053 * its calls in, but will run happily in parallel with the interrupt
1054 * handler.  This code currently has provisions for fragmenting tx buffers
1055 * while trying to track down a bug in either the zero copy code or
1056 * the tx fifo (hence the MAX_FRAG_LEN).
1057 */
1058static netdev_tx_t ns83820_hard_start_xmit(struct sk_buff *skb,
1059					   struct net_device *ndev)
1060{
1061	struct ns83820 *dev = PRIV(ndev);
1062	u32 free_idx, cmdsts, extsts;
1063	int nr_free, nr_frags;
1064	unsigned tx_done_idx, last_idx;
1065	dma_addr_t buf;
1066	unsigned len;
1067	skb_frag_t *frag;
1068	int stopped = 0;
1069	int do_intr = 0;
1070	volatile __le32 *first_desc;
1071
1072	dprintk("ns83820_hard_start_xmit\n");
1073
1074	nr_frags =  skb_shinfo(skb)->nr_frags;
1075again:
1076	if (unlikely(dev->CFG_cache & CFG_LNKSTS)) {
1077		netif_stop_queue(ndev);
1078		if (unlikely(dev->CFG_cache & CFG_LNKSTS))
1079			return NETDEV_TX_BUSY;
1080		netif_start_queue(ndev);
1081	}
1082
1083	last_idx = free_idx = dev->tx_free_idx;
1084	tx_done_idx = dev->tx_done_idx;
1085	nr_free = (tx_done_idx + NR_TX_DESC-2 - free_idx) % NR_TX_DESC;
1086	nr_free -= 1;
1087	if (nr_free <= nr_frags) {
1088		dprintk("stop_queue - not enough(%p)\n", ndev);
1089		netif_stop_queue(ndev);
1090
1091		/* Check again: we may have raced with a tx done irq */
1092		if (dev->tx_done_idx != tx_done_idx) {
1093			dprintk("restart queue(%p)\n", ndev);
1094			netif_start_queue(ndev);
1095			goto again;
1096		}
1097		return NETDEV_TX_BUSY;
1098	}
1099
1100	if (free_idx == dev->tx_intr_idx) {
1101		do_intr = 1;
1102		dev->tx_intr_idx = (dev->tx_intr_idx + NR_TX_DESC/4) % NR_TX_DESC;
1103	}
1104
1105	nr_free -= nr_frags;
1106	if (nr_free < MIN_TX_DESC_FREE) {
1107		dprintk("stop_queue - last entry(%p)\n", ndev);
1108		netif_stop_queue(ndev);
1109		stopped = 1;
1110	}
1111
1112	frag = skb_shinfo(skb)->frags;
1113	if (!nr_frags)
1114		frag = NULL;
1115	extsts = 0;
1116	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1117		extsts |= EXTSTS_IPPKT;
1118		if (IPPROTO_TCP == ip_hdr(skb)->protocol)
1119			extsts |= EXTSTS_TCPPKT;
1120		else if (IPPROTO_UDP == ip_hdr(skb)->protocol)
1121			extsts |= EXTSTS_UDPPKT;
1122	}
1123
1124#ifdef NS83820_VLAN_ACCEL_SUPPORT
1125	if (skb_vlan_tag_present(skb)) {
1126		/* fetch the vlan tag info out of the
1127		 * ancillary data if the vlan code
1128		 * is using hw vlan acceleration
1129		 */
1130		short tag = skb_vlan_tag_get(skb);
1131		extsts |= (EXTSTS_VPKT | htons(tag));
1132	}
1133#endif
1134
1135	len = skb->len;
1136	if (nr_frags)
1137		len -= skb->data_len;
1138	buf = pci_map_single(dev->pci_dev, skb->data, len, PCI_DMA_TODEVICE);
1139
1140	first_desc = dev->tx_descs + (free_idx * DESC_SIZE);
1141
1142	for (;;) {
1143		volatile __le32 *desc = dev->tx_descs + (free_idx * DESC_SIZE);
1144
1145		dprintk("frag[%3u]: %4u @ 0x%08Lx\n", free_idx, len,
1146			(unsigned long long)buf);
1147		last_idx = free_idx;
1148		free_idx = (free_idx + 1) % NR_TX_DESC;
1149		desc[DESC_LINK] = cpu_to_le32(dev->tx_phy_descs + (free_idx * DESC_SIZE * 4));
1150		desc_addr_set(desc + DESC_BUFPTR, buf);
1151		desc[DESC_EXTSTS] = cpu_to_le32(extsts);
1152
1153		cmdsts = ((nr_frags) ? CMDSTS_MORE : do_intr ? CMDSTS_INTR : 0);
1154		cmdsts |= (desc == first_desc) ? 0 : CMDSTS_OWN;
1155		cmdsts |= len;
1156		desc[DESC_CMDSTS] = cpu_to_le32(cmdsts);
1157
1158		if (!nr_frags)
1159			break;
1160
1161		buf = skb_frag_dma_map(&dev->pci_dev->dev, frag, 0,
1162				       skb_frag_size(frag), DMA_TO_DEVICE);
1163		dprintk("frag: buf=%08Lx  page=%08lx offset=%08lx\n",
1164			(long long)buf, (long) page_to_pfn(frag->page),
1165			frag->page_offset);
1166		len = skb_frag_size(frag);
1167		frag++;
1168		nr_frags--;
1169	}
1170	dprintk("done pkt\n");
1171
1172	spin_lock_irq(&dev->tx_lock);
1173	dev->tx_skbs[last_idx] = skb;
1174	first_desc[DESC_CMDSTS] |= cpu_to_le32(CMDSTS_OWN);
1175	dev->tx_free_idx = free_idx;
1176	atomic_inc(&dev->nr_tx_skbs);
1177	spin_unlock_irq(&dev->tx_lock);
1178
1179	kick_tx(dev);
1180
1181	/* Check again: we may have raced with a tx done irq */
1182	if (stopped && (dev->tx_done_idx != tx_done_idx) && start_tx_okay(dev))
1183		netif_start_queue(ndev);
1184
1185	return NETDEV_TX_OK;
1186}
1187
1188static void ns83820_update_stats(struct ns83820 *dev)
1189{
1190	struct net_device *ndev = dev->ndev;
1191	u8 __iomem *base = dev->base;
1192
1193	/* the DP83820 will freeze counters, so we need to read all of them */
1194	ndev->stats.rx_errors		+= readl(base + 0x60) & 0xffff;
1195	ndev->stats.rx_crc_errors	+= readl(base + 0x64) & 0xffff;
1196	ndev->stats.rx_missed_errors	+= readl(base + 0x68) & 0xffff;
1197	ndev->stats.rx_frame_errors	+= readl(base + 0x6c) & 0xffff;
1198	/*ndev->stats.rx_symbol_errors +=*/ readl(base + 0x70);
1199	ndev->stats.rx_length_errors	+= readl(base + 0x74) & 0xffff;
1200	ndev->stats.rx_length_errors	+= readl(base + 0x78) & 0xffff;
1201	/*ndev->stats.rx_badopcode_errors += */ readl(base + 0x7c);
1202	/*ndev->stats.rx_pause_count += */  readl(base + 0x80);
1203	/*ndev->stats.tx_pause_count += */  readl(base + 0x84);
1204	ndev->stats.tx_carrier_errors	+= readl(base + 0x88) & 0xff;
1205}
1206
1207static struct net_device_stats *ns83820_get_stats(struct net_device *ndev)
1208{
1209	struct ns83820 *dev = PRIV(ndev);
1210
1211	/* somewhat overkill */
1212	spin_lock_irq(&dev->misc_lock);
1213	ns83820_update_stats(dev);
1214	spin_unlock_irq(&dev->misc_lock);
1215
1216	return &ndev->stats;
1217}
1218
1219/* Let ethtool retrieve info */
1220static int ns83820_get_settings(struct net_device *ndev,
1221				struct ethtool_cmd *cmd)
1222{
1223	struct ns83820 *dev = PRIV(ndev);
1224	u32 cfg, tanar, tbicr;
1225	int fullduplex   = 0;
1226
1227	/*
1228	 * Here's the list of available ethtool commands from other drivers:
1229	 *	cmd->advertising =
1230	 *	ethtool_cmd_speed_set(cmd, ...)
1231	 *	cmd->duplex =
1232	 *	cmd->port = 0;
1233	 *	cmd->phy_address =
1234	 *	cmd->transceiver = 0;
1235	 *	cmd->autoneg =
1236	 *	cmd->maxtxpkt = 0;
1237	 *	cmd->maxrxpkt = 0;
1238	 */
1239
1240	/* read current configuration */
1241	cfg   = readl(dev->base + CFG) ^ SPDSTS_POLARITY;
1242	tanar = readl(dev->base + TANAR);
1243	tbicr = readl(dev->base + TBICR);
1244
1245	fullduplex = (cfg & CFG_DUPSTS) ? 1 : 0;
1246
1247	cmd->supported = SUPPORTED_Autoneg;
1248
1249	if (dev->CFG_cache & CFG_TBI_EN) {
1250		/* we have optical interface */
1251		cmd->supported |= SUPPORTED_1000baseT_Half |
1252					SUPPORTED_1000baseT_Full |
1253					SUPPORTED_FIBRE;
1254		cmd->port       = PORT_FIBRE;
1255	} else {
1256		/* we have copper */
1257		cmd->supported |= SUPPORTED_10baseT_Half |
1258			SUPPORTED_10baseT_Full | SUPPORTED_100baseT_Half |
1259			SUPPORTED_100baseT_Full | SUPPORTED_1000baseT_Half |
1260			SUPPORTED_1000baseT_Full |
1261			SUPPORTED_MII;
1262		cmd->port = PORT_MII;
1263	}
1264
1265	cmd->duplex = fullduplex ? DUPLEX_FULL : DUPLEX_HALF;
1266	switch (cfg / CFG_SPDSTS0 & 3) {
1267	case 2:
1268		ethtool_cmd_speed_set(cmd, SPEED_1000);
1269		break;
1270	case 1:
1271		ethtool_cmd_speed_set(cmd, SPEED_100);
1272		break;
1273	default:
1274		ethtool_cmd_speed_set(cmd, SPEED_10);
1275		break;
1276	}
1277	cmd->autoneg = (tbicr & TBICR_MR_AN_ENABLE)
1278		? AUTONEG_ENABLE : AUTONEG_DISABLE;
1279	return 0;
1280}
1281
1282/* Let ethool change settings*/
1283static int ns83820_set_settings(struct net_device *ndev,
1284				struct ethtool_cmd *cmd)
1285{
1286	struct ns83820 *dev = PRIV(ndev);
1287	u32 cfg, tanar;
1288	int have_optical = 0;
1289	int fullduplex   = 0;
1290
1291	/* read current configuration */
1292	cfg = readl(dev->base + CFG) ^ SPDSTS_POLARITY;
1293	tanar = readl(dev->base + TANAR);
1294
1295	if (dev->CFG_cache & CFG_TBI_EN) {
1296		/* we have optical */
1297		have_optical = 1;
1298		fullduplex   = (tanar & TANAR_FULL_DUP);
1299
1300	} else {
1301		/* we have copper */
1302		fullduplex = cfg & CFG_DUPSTS;
1303	}
1304
1305	spin_lock_irq(&dev->misc_lock);
1306	spin_lock(&dev->tx_lock);
1307
1308	/* Set duplex */
1309	if (cmd->duplex != fullduplex) {
1310		if (have_optical) {
1311			/*set full duplex*/
1312			if (cmd->duplex == DUPLEX_FULL) {
1313				/* force full duplex */
1314				writel(readl(dev->base + TXCFG)
1315					| TXCFG_CSI | TXCFG_HBI | TXCFG_ATP,
1316					dev->base + TXCFG);
1317				writel(readl(dev->base + RXCFG) | RXCFG_RX_FD,
1318					dev->base + RXCFG);
1319				/* Light up full duplex LED */
1320				writel(readl(dev->base + GPIOR) | GPIOR_GP1_OUT,
1321					dev->base + GPIOR);
1322			} else {
1323				/*TODO: set half duplex */
1324			}
1325
1326		} else {
1327			/*we have copper*/
1328			/* TODO: Set duplex for copper cards */
1329		}
1330		printk(KERN_INFO "%s: Duplex set via ethtool\n",
1331		ndev->name);
1332	}
1333
1334	/* Set autonegotiation */
1335	if (1) {
1336		if (cmd->autoneg == AUTONEG_ENABLE) {
1337			/* restart auto negotiation */
1338			writel(TBICR_MR_AN_ENABLE | TBICR_MR_RESTART_AN,
1339				dev->base + TBICR);
1340			writel(TBICR_MR_AN_ENABLE, dev->base + TBICR);
1341				dev->linkstate = LINK_AUTONEGOTIATE;
1342
1343			printk(KERN_INFO "%s: autoneg enabled via ethtool\n",
1344				ndev->name);
1345		} else {
1346			/* disable auto negotiation */
1347			writel(0x00000000, dev->base + TBICR);
1348		}
1349
1350		printk(KERN_INFO "%s: autoneg %s via ethtool\n", ndev->name,
1351				cmd->autoneg ? "ENABLED" : "DISABLED");
1352	}
1353
1354	phy_intr(ndev);
1355	spin_unlock(&dev->tx_lock);
1356	spin_unlock_irq(&dev->misc_lock);
1357
1358	return 0;
1359}
1360/* end ethtool get/set support -df */
1361
1362static void ns83820_get_drvinfo(struct net_device *ndev, struct ethtool_drvinfo *info)
1363{
1364	struct ns83820 *dev = PRIV(ndev);
1365	strlcpy(info->driver, "ns83820", sizeof(info->driver));
1366	strlcpy(info->version, VERSION, sizeof(info->version));
1367	strlcpy(info->bus_info, pci_name(dev->pci_dev), sizeof(info->bus_info));
1368}
1369
1370static u32 ns83820_get_link(struct net_device *ndev)
1371{
1372	struct ns83820 *dev = PRIV(ndev);
1373	u32 cfg = readl(dev->base + CFG) ^ SPDSTS_POLARITY;
1374	return cfg & CFG_LNKSTS ? 1 : 0;
1375}
1376
1377static const struct ethtool_ops ops = {
1378	.get_settings    = ns83820_get_settings,
1379	.set_settings    = ns83820_set_settings,
1380	.get_drvinfo     = ns83820_get_drvinfo,
1381	.get_link        = ns83820_get_link
1382};
1383
1384static inline void ns83820_disable_interrupts(struct ns83820 *dev)
1385{
1386	writel(0, dev->base + IMR);
1387	writel(0, dev->base + IER);
1388	readl(dev->base + IER);
1389}
1390
1391/* this function is called in irq context from the ISR */
1392static void ns83820_mib_isr(struct ns83820 *dev)
1393{
1394	unsigned long flags;
1395	spin_lock_irqsave(&dev->misc_lock, flags);
1396	ns83820_update_stats(dev);
1397	spin_unlock_irqrestore(&dev->misc_lock, flags);
1398}
1399
1400static void ns83820_do_isr(struct net_device *ndev, u32 isr);
1401static irqreturn_t ns83820_irq(int foo, void *data)
1402{
1403	struct net_device *ndev = data;
1404	struct ns83820 *dev = PRIV(ndev);
1405	u32 isr;
1406	dprintk("ns83820_irq(%p)\n", ndev);
1407
1408	dev->ihr = 0;
1409
1410	isr = readl(dev->base + ISR);
1411	dprintk("irq: %08x\n", isr);
1412	ns83820_do_isr(ndev, isr);
1413	return IRQ_HANDLED;
1414}
1415
1416static void ns83820_do_isr(struct net_device *ndev, u32 isr)
1417{
1418	struct ns83820 *dev = PRIV(ndev);
1419	unsigned long flags;
1420
1421#ifdef DEBUG
1422	if (isr & ~(ISR_PHY | ISR_RXDESC | ISR_RXEARLY | ISR_RXOK | ISR_RXERR | ISR_TXIDLE | ISR_TXOK | ISR_TXDESC))
1423		Dprintk("odd isr? 0x%08x\n", isr);
1424#endif
1425
1426	if (ISR_RXIDLE & isr) {
1427		dev->rx_info.idle = 1;
1428		Dprintk("oh dear, we are idle\n");
1429		ns83820_rx_kick(ndev);
1430	}
1431
1432	if ((ISR_RXDESC | ISR_RXOK) & isr) {
1433		prefetch(dev->rx_info.next_rx_desc);
1434
1435		spin_lock_irqsave(&dev->misc_lock, flags);
1436		dev->IMR_cache &= ~(ISR_RXDESC | ISR_RXOK);
1437		writel(dev->IMR_cache, dev->base + IMR);
1438		spin_unlock_irqrestore(&dev->misc_lock, flags);
1439
1440		tasklet_schedule(&dev->rx_tasklet);
1441		//rx_irq(ndev);
1442		//writel(4, dev->base + IHR);
1443	}
1444
1445	if ((ISR_RXIDLE | ISR_RXORN | ISR_RXDESC | ISR_RXOK | ISR_RXERR) & isr)
1446		ns83820_rx_kick(ndev);
1447
1448	if (unlikely(ISR_RXSOVR & isr)) {
1449		//printk("overrun: rxsovr\n");
1450		ndev->stats.rx_fifo_errors++;
1451	}
1452
1453	if (unlikely(ISR_RXORN & isr)) {
1454		//printk("overrun: rxorn\n");
1455		ndev->stats.rx_fifo_errors++;
1456	}
1457
1458	if ((ISR_RXRCMP & isr) && dev->rx_info.up)
1459		writel(CR_RXE, dev->base + CR);
1460
1461	if (ISR_TXIDLE & isr) {
1462		u32 txdp;
1463		txdp = readl(dev->base + TXDP);
1464		dprintk("txdp: %08x\n", txdp);
1465		txdp -= dev->tx_phy_descs;
1466		dev->tx_idx = txdp / (DESC_SIZE * 4);
1467		if (dev->tx_idx >= NR_TX_DESC) {
1468			printk(KERN_ALERT "%s: BUG -- txdp out of range\n", ndev->name);
1469			dev->tx_idx = 0;
1470		}
1471		/* The may have been a race between a pci originated read
1472		 * and the descriptor update from the cpu.  Just in case,
1473		 * kick the transmitter if the hardware thinks it is on a
1474		 * different descriptor than we are.
1475		 */
1476		if (dev->tx_idx != dev->tx_free_idx)
1477			kick_tx(dev);
1478	}
1479
1480	/* Defer tx ring processing until more than a minimum amount of
1481	 * work has accumulated
1482	 */
1483	if ((ISR_TXDESC | ISR_TXIDLE | ISR_TXOK | ISR_TXERR) & isr) {
1484		spin_lock_irqsave(&dev->tx_lock, flags);
1485		do_tx_done(ndev);
1486		spin_unlock_irqrestore(&dev->tx_lock, flags);
1487
1488		/* Disable TxOk if there are no outstanding tx packets.
1489		 */
1490		if ((dev->tx_done_idx == dev->tx_free_idx) &&
1491		    (dev->IMR_cache & ISR_TXOK)) {
1492			spin_lock_irqsave(&dev->misc_lock, flags);
1493			dev->IMR_cache &= ~ISR_TXOK;
1494			writel(dev->IMR_cache, dev->base + IMR);
1495			spin_unlock_irqrestore(&dev->misc_lock, flags);
1496		}
1497	}
1498
1499	/* The TxIdle interrupt can come in before the transmit has
1500	 * completed.  Normally we reap packets off of the combination
1501	 * of TxDesc and TxIdle and leave TxOk disabled (since it
1502	 * occurs on every packet), but when no further irqs of this
1503	 * nature are expected, we must enable TxOk.
1504	 */
1505	if ((ISR_TXIDLE & isr) && (dev->tx_done_idx != dev->tx_free_idx)) {
1506		spin_lock_irqsave(&dev->misc_lock, flags);
1507		dev->IMR_cache |= ISR_TXOK;
1508		writel(dev->IMR_cache, dev->base + IMR);
1509		spin_unlock_irqrestore(&dev->misc_lock, flags);
1510	}
1511
1512	/* MIB interrupt: one of the statistics counters is about to overflow */
1513	if (unlikely(ISR_MIB & isr))
1514		ns83820_mib_isr(dev);
1515
1516	/* PHY: Link up/down/negotiation state change */
1517	if (unlikely(ISR_PHY & isr))
1518		phy_intr(ndev);
1519
1520#if 0	/* Still working on the interrupt mitigation strategy */
1521	if (dev->ihr)
1522		writel(dev->ihr, dev->base + IHR);
1523#endif
1524}
1525
1526static void ns83820_do_reset(struct ns83820 *dev, u32 which)
1527{
1528	Dprintk("resetting chip...\n");
1529	writel(which, dev->base + CR);
1530	do {
1531		schedule();
1532	} while (readl(dev->base + CR) & which);
1533	Dprintk("okay!\n");
1534}
1535
1536static int ns83820_stop(struct net_device *ndev)
1537{
1538	struct ns83820 *dev = PRIV(ndev);
1539
1540	/* FIXME: protect against interrupt handler? */
1541	del_timer_sync(&dev->tx_watchdog);
1542
1543	ns83820_disable_interrupts(dev);
1544
1545	dev->rx_info.up = 0;
1546	synchronize_irq(dev->pci_dev->irq);
1547
1548	ns83820_do_reset(dev, CR_RST);
1549
1550	synchronize_irq(dev->pci_dev->irq);
1551
1552	spin_lock_irq(&dev->misc_lock);
1553	dev->IMR_cache &= ~(ISR_TXURN | ISR_TXIDLE | ISR_TXERR | ISR_TXDESC | ISR_TXOK);
1554	spin_unlock_irq(&dev->misc_lock);
1555
1556	ns83820_cleanup_rx(dev);
1557	ns83820_cleanup_tx(dev);
1558
1559	return 0;
1560}
1561
1562static void ns83820_tx_timeout(struct net_device *ndev)
1563{
1564	struct ns83820 *dev = PRIV(ndev);
1565        u32 tx_done_idx;
1566	__le32 *desc;
1567	unsigned long flags;
1568
1569	spin_lock_irqsave(&dev->tx_lock, flags);
1570
1571	tx_done_idx = dev->tx_done_idx;
1572	desc = dev->tx_descs + (tx_done_idx * DESC_SIZE);
1573
1574	printk(KERN_INFO "%s: tx_timeout: tx_done_idx=%d free_idx=%d cmdsts=%08x\n",
1575		ndev->name,
1576		tx_done_idx, dev->tx_free_idx, le32_to_cpu(desc[DESC_CMDSTS]));
1577
1578#if defined(DEBUG)
1579	{
1580		u32 isr;
1581		isr = readl(dev->base + ISR);
1582		printk("irq: %08x imr: %08x\n", isr, dev->IMR_cache);
1583		ns83820_do_isr(ndev, isr);
1584	}
1585#endif
1586
1587	do_tx_done(ndev);
1588
1589	tx_done_idx = dev->tx_done_idx;
1590	desc = dev->tx_descs + (tx_done_idx * DESC_SIZE);
1591
1592	printk(KERN_INFO "%s: after: tx_done_idx=%d free_idx=%d cmdsts=%08x\n",
1593		ndev->name,
1594		tx_done_idx, dev->tx_free_idx, le32_to_cpu(desc[DESC_CMDSTS]));
1595
1596	spin_unlock_irqrestore(&dev->tx_lock, flags);
1597}
1598
1599static void ns83820_tx_watch(unsigned long data)
1600{
1601	struct net_device *ndev = (void *)data;
1602	struct ns83820 *dev = PRIV(ndev);
1603
1604#if defined(DEBUG)
1605	printk("ns83820_tx_watch: %u %u %d\n",
1606		dev->tx_done_idx, dev->tx_free_idx, atomic_read(&dev->nr_tx_skbs)
1607		);
1608#endif
1609
1610	if (time_after(jiffies, dev_trans_start(ndev) + 1*HZ) &&
1611	    dev->tx_done_idx != dev->tx_free_idx) {
1612		printk(KERN_DEBUG "%s: ns83820_tx_watch: %u %u %d\n",
1613			ndev->name,
1614			dev->tx_done_idx, dev->tx_free_idx,
1615			atomic_read(&dev->nr_tx_skbs));
1616		ns83820_tx_timeout(ndev);
1617	}
1618
1619	mod_timer(&dev->tx_watchdog, jiffies + 2*HZ);
1620}
1621
1622static int ns83820_open(struct net_device *ndev)
1623{
1624	struct ns83820 *dev = PRIV(ndev);
1625	unsigned i;
1626	u32 desc;
1627	int ret;
1628
1629	dprintk("ns83820_open\n");
1630
1631	writel(0, dev->base + PQCR);
1632
1633	ret = ns83820_setup_rx(ndev);
1634	if (ret)
1635		goto failed;
1636
1637	memset(dev->tx_descs, 0, 4 * NR_TX_DESC * DESC_SIZE);
1638	for (i=0; i<NR_TX_DESC; i++) {
1639		dev->tx_descs[(i * DESC_SIZE) + DESC_LINK]
1640				= cpu_to_le32(
1641				  dev->tx_phy_descs
1642				  + ((i+1) % NR_TX_DESC) * DESC_SIZE * 4);
1643	}
1644
1645	dev->tx_idx = 0;
1646	dev->tx_done_idx = 0;
1647	desc = dev->tx_phy_descs;
1648	writel(0, dev->base + TXDP_HI);
1649	writel(desc, dev->base + TXDP);
1650
1651	init_timer(&dev->tx_watchdog);
1652	dev->tx_watchdog.data = (unsigned long)ndev;
1653	dev->tx_watchdog.function = ns83820_tx_watch;
1654	mod_timer(&dev->tx_watchdog, jiffies + 2*HZ);
1655
1656	netif_start_queue(ndev);	/* FIXME: wait for phy to come up */
1657
1658	return 0;
1659
1660failed:
1661	ns83820_stop(ndev);
1662	return ret;
1663}
1664
1665static void ns83820_getmac(struct ns83820 *dev, u8 *mac)
1666{
1667	unsigned i;
1668	for (i=0; i<3; i++) {
1669		u32 data;
1670
1671		/* Read from the perfect match memory: this is loaded by
1672		 * the chip from the EEPROM via the EELOAD self test.
1673		 */
1674		writel(i*2, dev->base + RFCR);
1675		data = readl(dev->base + RFDR);
1676
1677		*mac++ = data;
1678		*mac++ = data >> 8;
1679	}
1680}
1681
1682static int ns83820_change_mtu(struct net_device *ndev, int new_mtu)
1683{
1684	if (new_mtu > RX_BUF_SIZE)
1685		return -EINVAL;
1686	ndev->mtu = new_mtu;
1687	return 0;
1688}
1689
1690static void ns83820_set_multicast(struct net_device *ndev)
1691{
1692	struct ns83820 *dev = PRIV(ndev);
1693	u8 __iomem *rfcr = dev->base + RFCR;
1694	u32 and_mask = 0xffffffff;
1695	u32 or_mask = 0;
1696	u32 val;
1697
1698	if (ndev->flags & IFF_PROMISC)
1699		or_mask |= RFCR_AAU | RFCR_AAM;
1700	else
1701		and_mask &= ~(RFCR_AAU | RFCR_AAM);
1702
1703	if (ndev->flags & IFF_ALLMULTI || netdev_mc_count(ndev))
1704		or_mask |= RFCR_AAM;
1705	else
1706		and_mask &= ~RFCR_AAM;
1707
1708	spin_lock_irq(&dev->misc_lock);
1709	val = (readl(rfcr) & and_mask) | or_mask;
1710	/* Ramit : RFCR Write Fix doc says RFEN must be 0 modify other bits */
1711	writel(val & ~RFCR_RFEN, rfcr);
1712	writel(val, rfcr);
1713	spin_unlock_irq(&dev->misc_lock);
1714}
1715
1716static void ns83820_run_bist(struct net_device *ndev, const char *name, u32 enable, u32 done, u32 fail)
1717{
1718	struct ns83820 *dev = PRIV(ndev);
1719	int timed_out = 0;
1720	unsigned long start;
1721	u32 status;
1722	int loops = 0;
1723
1724	dprintk("%s: start %s\n", ndev->name, name);
1725
1726	start = jiffies;
1727
1728	writel(enable, dev->base + PTSCR);
1729	for (;;) {
1730		loops++;
1731		status = readl(dev->base + PTSCR);
1732		if (!(status & enable))
1733			break;
1734		if (status & done)
1735			break;
1736		if (status & fail)
1737			break;
1738		if (time_after_eq(jiffies, start + HZ)) {
1739			timed_out = 1;
1740			break;
1741		}
1742		schedule_timeout_uninterruptible(1);
1743	}
1744
1745	if (status & fail)
1746		printk(KERN_INFO "%s: %s failed! (0x%08x & 0x%08x)\n",
1747			ndev->name, name, status, fail);
1748	else if (timed_out)
1749		printk(KERN_INFO "%s: run_bist %s timed out! (%08x)\n",
1750			ndev->name, name, status);
1751
1752	dprintk("%s: done %s in %d loops\n", ndev->name, name, loops);
1753}
1754
1755#ifdef PHY_CODE_IS_FINISHED
1756static void ns83820_mii_write_bit(struct ns83820 *dev, int bit)
1757{
1758	/* drive MDC low */
1759	dev->MEAR_cache &= ~MEAR_MDC;
1760	writel(dev->MEAR_cache, dev->base + MEAR);
1761	readl(dev->base + MEAR);
1762
1763	/* enable output, set bit */
1764	dev->MEAR_cache |= MEAR_MDDIR;
1765	if (bit)
1766		dev->MEAR_cache |= MEAR_MDIO;
1767	else
1768		dev->MEAR_cache &= ~MEAR_MDIO;
1769
1770	/* set the output bit */
1771	writel(dev->MEAR_cache, dev->base + MEAR);
1772	readl(dev->base + MEAR);
1773
1774	/* Wait.  Max clock rate is 2.5MHz, this way we come in under 1MHz */
1775	udelay(1);
1776
1777	/* drive MDC high causing the data bit to be latched */
1778	dev->MEAR_cache |= MEAR_MDC;
1779	writel(dev->MEAR_cache, dev->base + MEAR);
1780	readl(dev->base + MEAR);
1781
1782	/* Wait again... */
1783	udelay(1);
1784}
1785
1786static int ns83820_mii_read_bit(struct ns83820 *dev)
1787{
1788	int bit;
1789
1790	/* drive MDC low, disable output */
1791	dev->MEAR_cache &= ~MEAR_MDC;
1792	dev->MEAR_cache &= ~MEAR_MDDIR;
1793	writel(dev->MEAR_cache, dev->base + MEAR);
1794	readl(dev->base + MEAR);
1795
1796	/* Wait.  Max clock rate is 2.5MHz, this way we come in under 1MHz */
1797	udelay(1);
1798
1799	/* drive MDC high causing the data bit to be latched */
1800	bit = (readl(dev->base + MEAR) & MEAR_MDIO) ? 1 : 0;
1801	dev->MEAR_cache |= MEAR_MDC;
1802	writel(dev->MEAR_cache, dev->base + MEAR);
1803
1804	/* Wait again... */
1805	udelay(1);
1806
1807	return bit;
1808}
1809
1810static unsigned ns83820_mii_read_reg(struct ns83820 *dev, unsigned phy, unsigned reg)
1811{
1812	unsigned data = 0;
1813	int i;
1814
1815	/* read some garbage so that we eventually sync up */
1816	for (i=0; i<64; i++)
1817		ns83820_mii_read_bit(dev);
1818
1819	ns83820_mii_write_bit(dev, 0);	/* start */
1820	ns83820_mii_write_bit(dev, 1);
1821	ns83820_mii_write_bit(dev, 1);	/* opcode read */
1822	ns83820_mii_write_bit(dev, 0);
1823
1824	/* write out the phy address: 5 bits, msb first */
1825	for (i=0; i<5; i++)
1826		ns83820_mii_write_bit(dev, phy & (0x10 >> i));
1827
1828	/* write out the register address, 5 bits, msb first */
1829	for (i=0; i<5; i++)
1830		ns83820_mii_write_bit(dev, reg & (0x10 >> i));
1831
1832	ns83820_mii_read_bit(dev);	/* turn around cycles */
1833	ns83820_mii_read_bit(dev);
1834
1835	/* read in the register data, 16 bits msb first */
1836	for (i=0; i<16; i++) {
1837		data <<= 1;
1838		data |= ns83820_mii_read_bit(dev);
1839	}
1840
1841	return data;
1842}
1843
1844static unsigned ns83820_mii_write_reg(struct ns83820 *dev, unsigned phy, unsigned reg, unsigned data)
1845{
1846	int i;
1847
1848	/* read some garbage so that we eventually sync up */
1849	for (i=0; i<64; i++)
1850		ns83820_mii_read_bit(dev);
1851
1852	ns83820_mii_write_bit(dev, 0);	/* start */
1853	ns83820_mii_write_bit(dev, 1);
1854	ns83820_mii_write_bit(dev, 0);	/* opcode read */
1855	ns83820_mii_write_bit(dev, 1);
1856
1857	/* write out the phy address: 5 bits, msb first */
1858	for (i=0; i<5; i++)
1859		ns83820_mii_write_bit(dev, phy & (0x10 >> i));
1860
1861	/* write out the register address, 5 bits, msb first */
1862	for (i=0; i<5; i++)
1863		ns83820_mii_write_bit(dev, reg & (0x10 >> i));
1864
1865	ns83820_mii_read_bit(dev);	/* turn around cycles */
1866	ns83820_mii_read_bit(dev);
1867
1868	/* read in the register data, 16 bits msb first */
1869	for (i=0; i<16; i++)
1870		ns83820_mii_write_bit(dev, (data >> (15 - i)) & 1);
1871
1872	return data;
1873}
1874
1875static void ns83820_probe_phy(struct net_device *ndev)
1876{
1877	struct ns83820 *dev = PRIV(ndev);
1878	static int first;
1879	int i;
1880#define MII_PHYIDR1	0x02
1881#define MII_PHYIDR2	0x03
1882
1883#if 0
1884	if (!first) {
1885		unsigned tmp;
1886		ns83820_mii_read_reg(dev, 1, 0x09);
1887		ns83820_mii_write_reg(dev, 1, 0x10, 0x0d3e);
1888
1889		tmp = ns83820_mii_read_reg(dev, 1, 0x00);
1890		ns83820_mii_write_reg(dev, 1, 0x00, tmp | 0x8000);
1891		udelay(1300);
1892		ns83820_mii_read_reg(dev, 1, 0x09);
1893	}
1894#endif
1895	first = 1;
1896
1897	for (i=1; i<2; i++) {
1898		int j;
1899		unsigned a, b;
1900		a = ns83820_mii_read_reg(dev, i, MII_PHYIDR1);
1901		b = ns83820_mii_read_reg(dev, i, MII_PHYIDR2);
1902
1903		//printk("%s: phy %d: 0x%04x 0x%04x\n",
1904		//	ndev->name, i, a, b);
1905
1906		for (j=0; j<0x16; j+=4) {
1907			dprintk("%s: [0x%02x] %04x %04x %04x %04x\n",
1908				ndev->name, j,
1909				ns83820_mii_read_reg(dev, i, 0 + j),
1910				ns83820_mii_read_reg(dev, i, 1 + j),
1911				ns83820_mii_read_reg(dev, i, 2 + j),
1912				ns83820_mii_read_reg(dev, i, 3 + j)
1913				);
1914		}
1915	}
1916	{
1917		unsigned a, b;
1918		/* read firmware version: memory addr is 0x8402 and 0x8403 */
1919		ns83820_mii_write_reg(dev, 1, 0x16, 0x000d);
1920		ns83820_mii_write_reg(dev, 1, 0x1e, 0x810e);
1921		a = ns83820_mii_read_reg(dev, 1, 0x1d);
1922
1923		ns83820_mii_write_reg(dev, 1, 0x16, 0x000d);
1924		ns83820_mii_write_reg(dev, 1, 0x1e, 0x810e);
1925		b = ns83820_mii_read_reg(dev, 1, 0x1d);
1926		dprintk("version: 0x%04x 0x%04x\n", a, b);
1927	}
1928}
1929#endif
1930
1931static const struct net_device_ops netdev_ops = {
1932	.ndo_open		= ns83820_open,
1933	.ndo_stop		= ns83820_stop,
1934	.ndo_start_xmit		= ns83820_hard_start_xmit,
1935	.ndo_get_stats		= ns83820_get_stats,
1936	.ndo_change_mtu		= ns83820_change_mtu,
1937	.ndo_set_rx_mode	= ns83820_set_multicast,
1938	.ndo_validate_addr	= eth_validate_addr,
1939	.ndo_set_mac_address	= eth_mac_addr,
1940	.ndo_tx_timeout		= ns83820_tx_timeout,
1941};
1942
1943static int ns83820_init_one(struct pci_dev *pci_dev,
1944			    const struct pci_device_id *id)
1945{
1946	struct net_device *ndev;
1947	struct ns83820 *dev;
1948	long addr;
1949	int err;
1950	int using_dac = 0;
1951
1952	/* See if we can set the dma mask early on; failure is fatal. */
1953	if (sizeof(dma_addr_t) == 8 &&
1954		!pci_set_dma_mask(pci_dev, DMA_BIT_MASK(64))) {
1955		using_dac = 1;
1956	} else if (!pci_set_dma_mask(pci_dev, DMA_BIT_MASK(32))) {
1957		using_dac = 0;
1958	} else {
1959		dev_warn(&pci_dev->dev, "pci_set_dma_mask failed!\n");
1960		return -ENODEV;
1961	}
1962
1963	ndev = alloc_etherdev(sizeof(struct ns83820));
1964	err = -ENOMEM;
1965	if (!ndev)
1966		goto out;
1967
1968	dev = PRIV(ndev);
1969	dev->ndev = ndev;
1970
1971	spin_lock_init(&dev->rx_info.lock);
1972	spin_lock_init(&dev->tx_lock);
1973	spin_lock_init(&dev->misc_lock);
1974	dev->pci_dev = pci_dev;
1975
1976	SET_NETDEV_DEV(ndev, &pci_dev->dev);
1977
1978	INIT_WORK(&dev->tq_refill, queue_refill);
1979	tasklet_init(&dev->rx_tasklet, rx_action, (unsigned long)ndev);
1980
1981	err = pci_enable_device(pci_dev);
1982	if (err) {
1983		dev_info(&pci_dev->dev, "pci_enable_dev failed: %d\n", err);
1984		goto out_free;
1985	}
1986
1987	pci_set_master(pci_dev);
1988	addr = pci_resource_start(pci_dev, 1);
1989	dev->base = ioremap_nocache(addr, PAGE_SIZE);
1990	dev->tx_descs = pci_alloc_consistent(pci_dev,
1991			4 * DESC_SIZE * NR_TX_DESC, &dev->tx_phy_descs);
1992	dev->rx_info.descs = pci_alloc_consistent(pci_dev,
1993			4 * DESC_SIZE * NR_RX_DESC, &dev->rx_info.phy_descs);
1994	err = -ENOMEM;
1995	if (!dev->base || !dev->tx_descs || !dev->rx_info.descs)
1996		goto out_disable;
1997
1998	dprintk("%p: %08lx  %p: %08lx\n",
1999		dev->tx_descs, (long)dev->tx_phy_descs,
2000		dev->rx_info.descs, (long)dev->rx_info.phy_descs);
2001
2002	ns83820_disable_interrupts(dev);
2003
2004	dev->IMR_cache = 0;
2005
2006	err = request_irq(pci_dev->irq, ns83820_irq, IRQF_SHARED,
2007			  DRV_NAME, ndev);
2008	if (err) {
2009		dev_info(&pci_dev->dev, "unable to register irq %d, err %d\n",
2010			pci_dev->irq, err);
2011		goto out_disable;
2012	}
2013
2014	/*
2015	 * FIXME: we are holding rtnl_lock() over obscenely long area only
2016	 * because some of the setup code uses dev->name.  It's Wrong(tm) -
2017	 * we should be using driver-specific names for all that stuff.
2018	 * For now that will do, but we really need to come back and kill
2019	 * most of the dev_alloc_name() users later.
2020	 */
2021	rtnl_lock();
2022	err = dev_alloc_name(ndev, ndev->name);
2023	if (err < 0) {
2024		dev_info(&pci_dev->dev, "unable to get netdev name: %d\n", err);
2025		goto out_free_irq;
2026	}
2027
2028	printk("%s: ns83820.c: 0x22c: %08x, subsystem: %04x:%04x\n",
2029		ndev->name, le32_to_cpu(readl(dev->base + 0x22c)),
2030		pci_dev->subsystem_vendor, pci_dev->subsystem_device);
2031
2032	ndev->netdev_ops = &netdev_ops;
2033	ndev->ethtool_ops = &ops;
2034	ndev->watchdog_timeo = 5 * HZ;
2035	pci_set_drvdata(pci_dev, ndev);
2036
2037	ns83820_do_reset(dev, CR_RST);
2038
2039	/* Must reset the ram bist before running it */
2040	writel(PTSCR_RBIST_RST, dev->base + PTSCR);
2041	ns83820_run_bist(ndev, "sram bist",   PTSCR_RBIST_EN,
2042			 PTSCR_RBIST_DONE, PTSCR_RBIST_FAIL);
2043	ns83820_run_bist(ndev, "eeprom bist", PTSCR_EEBIST_EN, 0,
2044			 PTSCR_EEBIST_FAIL);
2045	ns83820_run_bist(ndev, "eeprom load", PTSCR_EELOAD_EN, 0, 0);
2046
2047	/* I love config registers */
2048	dev->CFG_cache = readl(dev->base + CFG);
2049
2050	if ((dev->CFG_cache & CFG_PCI64_DET)) {
2051		printk(KERN_INFO "%s: detected 64 bit PCI data bus.\n",
2052			ndev->name);
2053		/*dev->CFG_cache |= CFG_DATA64_EN;*/
2054		if (!(dev->CFG_cache & CFG_DATA64_EN))
2055			printk(KERN_INFO "%s: EEPROM did not enable 64 bit bus.  Disabled.\n",
2056				ndev->name);
2057	} else
2058		dev->CFG_cache &= ~(CFG_DATA64_EN);
2059
2060	dev->CFG_cache &= (CFG_TBI_EN  | CFG_MRM_DIS   | CFG_MWI_DIS |
2061			   CFG_T64ADDR | CFG_DATA64_EN | CFG_EXT_125 |
2062			   CFG_M64ADDR);
2063	dev->CFG_cache |= CFG_PINT_DUPSTS | CFG_PINT_LNKSTS | CFG_PINT_SPDSTS |
2064			  CFG_EXTSTS_EN   | CFG_EXD         | CFG_PESEL;
2065	dev->CFG_cache |= CFG_REQALG;
2066	dev->CFG_cache |= CFG_POW;
2067	dev->CFG_cache |= CFG_TMRTEST;
2068
2069	/* When compiled with 64 bit addressing, we must always enable
2070	 * the 64 bit descriptor format.
2071	 */
2072	if (sizeof(dma_addr_t) == 8)
2073		dev->CFG_cache |= CFG_M64ADDR;
2074	if (using_dac)
2075		dev->CFG_cache |= CFG_T64ADDR;
2076
2077	/* Big endian mode does not seem to do what the docs suggest */
2078	dev->CFG_cache &= ~CFG_BEM;
2079
2080	/* setup optical transceiver if we have one */
2081	if (dev->CFG_cache & CFG_TBI_EN) {
2082		printk(KERN_INFO "%s: enabling optical transceiver\n",
2083			ndev->name);
2084		writel(readl(dev->base + GPIOR) | 0x3e8, dev->base + GPIOR);
2085
2086		/* setup auto negotiation feature advertisement */
2087		writel(readl(dev->base + TANAR)
2088		       | TANAR_HALF_DUP | TANAR_FULL_DUP,
2089		       dev->base + TANAR);
2090
2091		/* start auto negotiation */
2092		writel(TBICR_MR_AN_ENABLE | TBICR_MR_RESTART_AN,
2093		       dev->base + TBICR);
2094		writel(TBICR_MR_AN_ENABLE, dev->base + TBICR);
2095		dev->linkstate = LINK_AUTONEGOTIATE;
2096
2097		dev->CFG_cache |= CFG_MODE_1000;
2098	}
2099
2100	writel(dev->CFG_cache, dev->base + CFG);
2101	dprintk("CFG: %08x\n", dev->CFG_cache);
2102
2103	if (reset_phy) {
2104		printk(KERN_INFO "%s: resetting phy\n", ndev->name);
2105		writel(dev->CFG_cache | CFG_PHY_RST, dev->base + CFG);
2106		msleep(10);
2107		writel(dev->CFG_cache, dev->base + CFG);
2108	}
2109
2110#if 0	/* Huh?  This sets the PCI latency register.  Should be done via
2111	 * the PCI layer.  FIXME.
2112	 */
2113	if (readl(dev->base + SRR))
2114		writel(readl(dev->base+0x20c) | 0xfe00, dev->base + 0x20c);
2115#endif
2116
2117	/* Note!  The DMA burst size interacts with packet
2118	 * transmission, such that the largest packet that
2119	 * can be transmitted is 8192 - FLTH - burst size.
2120	 * If only the transmit fifo was larger...
2121	 */
2122	/* Ramit : 1024 DMA is not a good idea, it ends up banging
2123	 * some DELL and COMPAQ SMP systems */
2124	writel(TXCFG_CSI | TXCFG_HBI | TXCFG_ATP | TXCFG_MXDMA512
2125		| ((1600 / 32) * 0x100),
2126		dev->base + TXCFG);
2127
2128	/* Flush the interrupt holdoff timer */
2129	writel(0x000, dev->base + IHR);
2130	writel(0x100, dev->base + IHR);
2131	writel(0x000, dev->base + IHR);
2132
2133	/* Set Rx to full duplex, don't accept runt, errored, long or length
2134	 * range errored packets.  Use 512 byte DMA.
2135	 */
2136	/* Ramit : 1024 DMA is not a good idea, it ends up banging
2137	 * some DELL and COMPAQ SMP systems
2138	 * Turn on ALP, only we are accpeting Jumbo Packets */
2139	writel(RXCFG_AEP | RXCFG_ARP | RXCFG_AIRL | RXCFG_RX_FD
2140		| RXCFG_STRIPCRC
2141		//| RXCFG_ALP
2142		| (RXCFG_MXDMA512) | 0, dev->base + RXCFG);
2143
2144	/* Disable priority queueing */
2145	writel(0, dev->base + PQCR);
2146
2147	/* Enable IP checksum validation and detetion of VLAN headers.
2148	 * Note: do not set the reject options as at least the 0x102
2149	 * revision of the chip does not properly accept IP fragments
2150	 * at least for UDP.
2151	 */
2152	/* Ramit : Be sure to turn on RXCFG_ARP if VLAN's are enabled, since
2153	 * the MAC it calculates the packetsize AFTER stripping the VLAN
2154	 * header, and if a VLAN Tagged packet of 64 bytes is received (like
2155	 * a ping with a VLAN header) then the card, strips the 4 byte VLAN
2156	 * tag and then checks the packet size, so if RXCFG_ARP is not enabled,
2157	 * it discrards it!.  These guys......
2158	 * also turn on tag stripping if hardware acceleration is enabled
2159	 */
2160#ifdef NS83820_VLAN_ACCEL_SUPPORT
2161#define VRCR_INIT_VALUE (VRCR_IPEN|VRCR_VTDEN|VRCR_VTREN)
2162#else
2163#define VRCR_INIT_VALUE (VRCR_IPEN|VRCR_VTDEN)
2164#endif
2165	writel(VRCR_INIT_VALUE, dev->base + VRCR);
2166
2167	/* Enable per-packet TCP/UDP/IP checksumming
2168	 * and per packet vlan tag insertion if
2169	 * vlan hardware acceleration is enabled
2170	 */
2171#ifdef NS83820_VLAN_ACCEL_SUPPORT
2172#define VTCR_INIT_VALUE (VTCR_PPCHK|VTCR_VPPTI)
2173#else
2174#define VTCR_INIT_VALUE VTCR_PPCHK
2175#endif
2176	writel(VTCR_INIT_VALUE, dev->base + VTCR);
2177
2178	/* Ramit : Enable async and sync pause frames */
2179	/* writel(0, dev->base + PCR); */
2180	writel((PCR_PS_MCAST | PCR_PS_DA | PCR_PSEN | PCR_FFLO_4K |
2181		PCR_FFHI_8K | PCR_STLO_4 | PCR_STHI_8 | PCR_PAUSE_CNT),
2182		dev->base + PCR);
2183
2184	/* Disable Wake On Lan */
2185	writel(0, dev->base + WCSR);
2186
2187	ns83820_getmac(dev, ndev->dev_addr);
2188
2189	/* Yes, we support dumb IP checksum on transmit */
2190	ndev->features |= NETIF_F_SG;
2191	ndev->features |= NETIF_F_IP_CSUM;
2192
2193#ifdef NS83820_VLAN_ACCEL_SUPPORT
2194	/* We also support hardware vlan acceleration */
2195	ndev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2196#endif
2197
2198	if (using_dac) {
2199		printk(KERN_INFO "%s: using 64 bit addressing.\n",
2200			ndev->name);
2201		ndev->features |= NETIF_F_HIGHDMA;
2202	}
2203
2204	printk(KERN_INFO "%s: ns83820 v" VERSION ": DP83820 v%u.%u: %pM io=0x%08lx irq=%d f=%s\n",
2205		ndev->name,
2206		(unsigned)readl(dev->base + SRR) >> 8,
2207		(unsigned)readl(dev->base + SRR) & 0xff,
2208		ndev->dev_addr, addr, pci_dev->irq,
2209		(ndev->features & NETIF_F_HIGHDMA) ? "h,sg" : "sg"
2210		);
2211
2212#ifdef PHY_CODE_IS_FINISHED
2213	ns83820_probe_phy(ndev);
2214#endif
2215
2216	err = register_netdevice(ndev);
2217	if (err) {
2218		printk(KERN_INFO "ns83820: unable to register netdev: %d\n", err);
2219		goto out_cleanup;
2220	}
2221	rtnl_unlock();
2222
2223	return 0;
2224
2225out_cleanup:
2226	ns83820_disable_interrupts(dev); /* paranoia */
2227out_free_irq:
2228	rtnl_unlock();
2229	free_irq(pci_dev->irq, ndev);
2230out_disable:
2231	if (dev->base)
2232		iounmap(dev->base);
2233	pci_free_consistent(pci_dev, 4 * DESC_SIZE * NR_TX_DESC, dev->tx_descs, dev->tx_phy_descs);
2234	pci_free_consistent(pci_dev, 4 * DESC_SIZE * NR_RX_DESC, dev->rx_info.descs, dev->rx_info.phy_descs);
2235	pci_disable_device(pci_dev);
2236out_free:
2237	free_netdev(ndev);
2238out:
2239	return err;
2240}
2241
2242static void ns83820_remove_one(struct pci_dev *pci_dev)
2243{
2244	struct net_device *ndev = pci_get_drvdata(pci_dev);
2245	struct ns83820 *dev = PRIV(ndev); /* ok even if NULL */
2246
2247	if (!ndev)			/* paranoia */
2248		return;
2249
2250	ns83820_disable_interrupts(dev); /* paranoia */
2251
2252	unregister_netdev(ndev);
2253	free_irq(dev->pci_dev->irq, ndev);
2254	iounmap(dev->base);
2255	pci_free_consistent(dev->pci_dev, 4 * DESC_SIZE * NR_TX_DESC,
2256			dev->tx_descs, dev->tx_phy_descs);
2257	pci_free_consistent(dev->pci_dev, 4 * DESC_SIZE * NR_RX_DESC,
2258			dev->rx_info.descs, dev->rx_info.phy_descs);
2259	pci_disable_device(dev->pci_dev);
2260	free_netdev(ndev);
2261}
2262
2263static const struct pci_device_id ns83820_pci_tbl[] = {
2264	{ 0x100b, 0x0022, PCI_ANY_ID, PCI_ANY_ID, 0, .driver_data = 0, },
2265	{ 0, },
2266};
2267
2268static struct pci_driver driver = {
2269	.name		= "ns83820",
2270	.id_table	= ns83820_pci_tbl,
2271	.probe		= ns83820_init_one,
2272	.remove		= ns83820_remove_one,
2273#if 0	/* FIXME: implement */
2274	.suspend	= ,
2275	.resume		= ,
2276#endif
2277};
2278
2279
2280static int __init ns83820_init(void)
2281{
2282	printk(KERN_INFO "ns83820.c: National Semiconductor DP83820 10/100/1000 driver.\n");
2283	return pci_register_driver(&driver);
2284}
2285
2286static void __exit ns83820_exit(void)
2287{
2288	pci_unregister_driver(&driver);
2289}
2290
2291MODULE_AUTHOR("Benjamin LaHaise <bcrl@kvack.org>");
2292MODULE_DESCRIPTION("National Semiconductor DP83820 10/100/1000 driver");
2293MODULE_LICENSE("GPL");
2294
2295MODULE_DEVICE_TABLE(pci, ns83820_pci_tbl);
2296
2297module_param(lnksts, int, 0);
2298MODULE_PARM_DESC(lnksts, "Polarity of LNKSTS bit");
2299
2300module_param(ihr, int, 0);
2301MODULE_PARM_DESC(ihr, "Time in 100 us increments to delay interrupts (range 0-127)");
2302
2303module_param(reset_phy, int, 0);
2304MODULE_PARM_DESC(reset_phy, "Set to 1 to reset the PHY on startup");
2305
2306module_init(ns83820_init);
2307module_exit(ns83820_exit);
2308