1// SPDX-License-Identifier: GPL-2.0-or-later
   2#define VERSION "0.23"
   3/* ns83820.c by Benjamin LaHaise with contributions.
   4 *
   5 * Questions/comments/discussion to linux-ns83820@kvack.org.
   6 *
   7 * $Revision: 1.34.2.23 $
   8 *
   9 * Copyright 2001 Benjamin LaHaise.
  10 * Copyright 2001, 2002 Red Hat.
  11 *
  12 * Mmmm, chocolate vanilla mocha...
  13 *
  14 * ChangeLog
  15 * =========
  16 *	20010414	0.1 - created
  17 *	20010622	0.2 - basic rx and tx.
  18 *	20010711	0.3 - added duplex and link state detection support.
  19 *	20010713	0.4 - zero copy, no hangs.
  20 *			0.5 - 64 bit dma support (davem will hate me for this)
  21 *			    - disable jumbo frames to avoid tx hangs
  22 *			    - work around tx deadlocks on my 1.02 card via
  23 *			      fiddling with TXCFG
  24 *	20010810	0.6 - use pci dma api for ringbuffers, work on ia64
  25 *	20010816	0.7 - misc cleanups
  26 *	20010826	0.8 - fix critical zero copy bugs
  27 *			0.9 - internal experiment
  28 *	20010827	0.10 - fix ia64 unaligned access.
  29 *	20010906	0.11 - accept all packets with checksum errors as
  30 *			       otherwise fragments get lost
  31 *			     - fix >> 32 bugs
  32 *			0.12 - add statistics counters
  33 *			     - add allmulti/promisc support
  34 *	20011009	0.13 - hotplug support, other smaller pci api cleanups
  35 *	20011204	0.13a - optical transceiver support added
  36 *				by Michael Clark <michael@metaparadigm.com>
  37 *	20011205	0.13b - call register_netdev earlier in initialization
  38 *				suppress duplicate link status messages
  39 *	20011117 	0.14 - ethtool GDRVINFO, GLINK support from jgarzik
  40 *	20011204 	0.15	get ppc (big endian) working
  41 *	20011218	0.16	various cleanups
  42 *	20020310	0.17	speedups
  43 *	20020610	0.18 -	actually use the pci dma api for highmem
  44 *			     -	remove pci latency register fiddling
  45 *			0.19 -	better bist support
  46 *			     -	add ihr and reset_phy parameters
  47 *			     -	gmii bus probing
  48 *			     -	fix missed txok introduced during performance
  49 *				tuning
  50 *			0.20 -	fix stupid RFEN thinko.  i am such a smurf.
  51 *	20040828	0.21 -	add hardware vlan accleration
  52 *				by Neil Horman <nhorman@redhat.com>
  53 *	20050406	0.22 -	improved DAC ifdefs from Andi Kleen
  54 *			     -	removal of dead code from Adrian Bunk
  55 *			     -	fix half duplex collision behaviour
  56 * Driver Overview
  57 * ===============
  58 *
  59 * This driver was originally written for the National Semiconductor
  60 * 83820 chip, a 10/100/1000 Mbps 64 bit PCI ethernet NIC.  Hopefully
  61 * this code will turn out to be a) clean, b) correct, and c) fast.
  62 * With that in mind, I'm aiming to split the code up as much as
  63 * reasonably possible.  At present there are X major sections that
  64 * break down into a) packet receive, b) packet transmit, c) link
  65 * management, d) initialization and configuration.  Where possible,
  66 * these code paths are designed to run in parallel.
  67 *
  68 * This driver has been tested and found to work with the following
  69 * cards (in no particular order):
  70 *
  71 *	Cameo		SOHO-GA2000T	SOHO-GA2500T
  72 *	D-Link		DGE-500T
  73 *	PureData	PDP8023Z-TG
  74 *	SMC		SMC9452TX	SMC9462TX
  75 *	Netgear		GA621
  76 *
  77 * Special thanks to SMC for providing hardware to test this driver on.
  78 *
  79 * Reports of success or failure would be greatly appreciated.
  80 */
  81//#define dprintk		printk
  82#define dprintk(x...)		do { } while (0)
  83
  84#include <linux/module.h>
  85#include <linux/moduleparam.h>
  86#include <linux/types.h>
  87#include <linux/pci.h>
  88#include <linux/dma-mapping.h>
  89#include <linux/netdevice.h>
  90#include <linux/etherdevice.h>
  91#include <linux/delay.h>
  92#include <linux/workqueue.h>
  93#include <linux/init.h>
  94#include <linux/interrupt.h>
  95#include <linux/ip.h>	/* for iph */
  96#include <linux/in.h>	/* for IPPROTO_... */
  97#include <linux/compiler.h>
  98#include <linux/prefetch.h>
  99#include <linux/ethtool.h>
 100#include <linux/sched.h>
 101#include <linux/timer.h>
 102#include <linux/if_vlan.h>
 103#include <linux/rtnetlink.h>
 104#include <linux/jiffies.h>
 105#include <linux/slab.h>
 106
 107#include <asm/io.h>
 108#include <linux/uaccess.h>
 109
 110#define DRV_NAME "ns83820"
 111
 112/* Global parameters.  See module_param near the bottom. */
 113static int ihr = 2;
 114static int reset_phy = 0;
 115static int lnksts = 0;		/* CFG_LNKSTS bit polarity */
 116
 117/* Dprintk is used for more interesting debug events */
 118#undef Dprintk
 119#define	Dprintk			dprintk
 120
 121/* tunables */
 122#define RX_BUF_SIZE	1500	/* 8192 */
 123#if IS_ENABLED(CONFIG_VLAN_8021Q)
 124#define NS83820_VLAN_ACCEL_SUPPORT
 125#endif
 126
 127/* Must not exceed ~65000. */
 128#define NR_RX_DESC	64
 129#define NR_TX_DESC	128
 130
 131/* not tunable */
 132#define REAL_RX_BUF_SIZE (RX_BUF_SIZE + 14)	/* rx/tx mac addr + type */
 133
 134#define MIN_TX_DESC_FREE	8
 135
 136/* register defines */
 137#define CFGCS		0x04
 138
 139#define CR_TXE		0x00000001
 140#define CR_TXD		0x00000002
 141/* Ramit : Here's a tip, don't do a RXD immediately followed by an RXE
 142 * The Receive engine skips one descriptor and moves
 143 * onto the next one!! */
 144#define CR_RXE		0x00000004
 145#define CR_RXD		0x00000008
 146#define CR_TXR		0x00000010
 147#define CR_RXR		0x00000020
 148#define CR_SWI		0x00000080
 149#define CR_RST		0x00000100
 150
 151#define PTSCR_EEBIST_FAIL       0x00000001
 152#define PTSCR_EEBIST_EN         0x00000002
 153#define PTSCR_EELOAD_EN         0x00000004
 154#define PTSCR_RBIST_FAIL        0x000001b8
 155#define PTSCR_RBIST_DONE        0x00000200
 156#define PTSCR_RBIST_EN          0x00000400
 157#define PTSCR_RBIST_RST         0x00002000
 158
 159#define MEAR_EEDI		0x00000001
 160#define MEAR_EEDO		0x00000002
 161#define MEAR_EECLK		0x00000004
 162#define MEAR_EESEL		0x00000008
 163#define MEAR_MDIO		0x00000010
 164#define MEAR_MDDIR		0x00000020
 165#define MEAR_MDC		0x00000040
 166
 167#define ISR_TXDESC3	0x40000000
 168#define ISR_TXDESC2	0x20000000
 169#define ISR_TXDESC1	0x10000000
 170#define ISR_TXDESC0	0x08000000
 171#define ISR_RXDESC3	0x04000000
 172#define ISR_RXDESC2	0x02000000
 173#define ISR_RXDESC1	0x01000000
 174#define ISR_RXDESC0	0x00800000
 175#define ISR_TXRCMP	0x00400000
 176#define ISR_RXRCMP	0x00200000
 177#define ISR_DPERR	0x00100000
 178#define ISR_SSERR	0x00080000
 179#define ISR_RMABT	0x00040000
 180#define ISR_RTABT	0x00020000
 181#define ISR_RXSOVR	0x00010000
 182#define ISR_HIBINT	0x00008000
 183#define ISR_PHY		0x00004000
 184#define ISR_PME		0x00002000
 185#define ISR_SWI		0x00001000
 186#define ISR_MIB		0x00000800
 187#define ISR_TXURN	0x00000400
 188#define ISR_TXIDLE	0x00000200
 189#define ISR_TXERR	0x00000100
 190#define ISR_TXDESC	0x00000080
 191#define ISR_TXOK	0x00000040
 192#define ISR_RXORN	0x00000020
 193#define ISR_RXIDLE	0x00000010
 194#define ISR_RXEARLY	0x00000008
 195#define ISR_RXERR	0x00000004
 196#define ISR_RXDESC	0x00000002
 197#define ISR_RXOK	0x00000001
 198
 199#define TXCFG_CSI	0x80000000
 200#define TXCFG_HBI	0x40000000
 201#define TXCFG_MLB	0x20000000
 202#define TXCFG_ATP	0x10000000
 203#define TXCFG_ECRETRY	0x00800000
 204#define TXCFG_BRST_DIS	0x00080000
 205#define TXCFG_MXDMA1024	0x00000000
 206#define TXCFG_MXDMA512	0x00700000
 207#define TXCFG_MXDMA256	0x00600000
 208#define TXCFG_MXDMA128	0x00500000
 209#define TXCFG_MXDMA64	0x00400000
 210#define TXCFG_MXDMA32	0x00300000
 211#define TXCFG_MXDMA16	0x00200000
 212#define TXCFG_MXDMA8	0x00100000
 213
 214#define CFG_LNKSTS	0x80000000
 215#define CFG_SPDSTS	0x60000000
 216#define CFG_SPDSTS1	0x40000000
 217#define CFG_SPDSTS0	0x20000000
 218#define CFG_DUPSTS	0x10000000
 219#define CFG_TBI_EN	0x01000000
 220#define CFG_MODE_1000	0x00400000
 221/* Ramit : Dont' ever use AUTO_1000, it never works and is buggy.
 222 * Read the Phy response and then configure the MAC accordingly */
 223#define CFG_AUTO_1000	0x00200000
 224#define CFG_PINT_CTL	0x001c0000
 225#define CFG_PINT_DUPSTS	0x00100000
 226#define CFG_PINT_LNKSTS	0x00080000
 227#define CFG_PINT_SPDSTS	0x00040000
 228#define CFG_TMRTEST	0x00020000
 229#define CFG_MRM_DIS	0x00010000
 230#define CFG_MWI_DIS	0x00008000
 231#define CFG_T64ADDR	0x00004000
 232#define CFG_PCI64_DET	0x00002000
 233#define CFG_DATA64_EN	0x00001000
 234#define CFG_M64ADDR	0x00000800
 235#define CFG_PHY_RST	0x00000400
 236#define CFG_PHY_DIS	0x00000200
 237#define CFG_EXTSTS_EN	0x00000100
 238#define CFG_REQALG	0x00000080
 239#define CFG_SB		0x00000040
 240#define CFG_POW		0x00000020
 241#define CFG_EXD		0x00000010
 242#define CFG_PESEL	0x00000008
 243#define CFG_BROM_DIS	0x00000004
 244#define CFG_EXT_125	0x00000002
 245#define CFG_BEM		0x00000001
 246
 247#define EXTSTS_UDPPKT	0x00200000
 248#define EXTSTS_TCPPKT	0x00080000
 249#define EXTSTS_IPPKT	0x00020000
 250#define EXTSTS_VPKT	0x00010000
 251#define EXTSTS_VTG_MASK	0x0000ffff
 252
 253#define SPDSTS_POLARITY	(CFG_SPDSTS1 | CFG_SPDSTS0 | CFG_DUPSTS | (lnksts ? CFG_LNKSTS : 0))
 254
 255#define MIBC_MIBS	0x00000008
 256#define MIBC_ACLR	0x00000004
 257#define MIBC_FRZ	0x00000002
 258#define MIBC_WRN	0x00000001
 259
 260#define PCR_PSEN	(1 << 31)
 261#define PCR_PS_MCAST	(1 << 30)
 262#define PCR_PS_DA	(1 << 29)
 263#define PCR_STHI_8	(3 << 23)
 264#define PCR_STLO_4	(1 << 23)
 265#define PCR_FFHI_8K	(3 << 21)
 266#define PCR_FFLO_4K	(1 << 21)
 267#define PCR_PAUSE_CNT	0xFFFE
 268
 269#define RXCFG_AEP	0x80000000
 270#define RXCFG_ARP	0x40000000
 271#define RXCFG_STRIPCRC	0x20000000
 272#define RXCFG_RX_FD	0x10000000
 273#define RXCFG_ALP	0x08000000
 274#define RXCFG_AIRL	0x04000000
 275#define RXCFG_MXDMA512	0x00700000
 276#define RXCFG_DRTH	0x0000003e
 277#define RXCFG_DRTH0	0x00000002
 278
 279#define RFCR_RFEN	0x80000000
 280#define RFCR_AAB	0x40000000
 281#define RFCR_AAM	0x20000000
 282#define RFCR_AAU	0x10000000
 283#define RFCR_APM	0x08000000
 284#define RFCR_APAT	0x07800000
 285#define RFCR_APAT3	0x04000000
 286#define RFCR_APAT2	0x02000000
 287#define RFCR_APAT1	0x01000000
 288#define RFCR_APAT0	0x00800000
 289#define RFCR_AARP	0x00400000
 290#define RFCR_MHEN	0x00200000
 291#define RFCR_UHEN	0x00100000
 292#define RFCR_ULM	0x00080000
 293
 294#define VRCR_RUDPE	0x00000080
 295#define VRCR_RTCPE	0x00000040
 296#define VRCR_RIPE	0x00000020
 297#define VRCR_IPEN	0x00000010
 298#define VRCR_DUTF	0x00000008
 299#define VRCR_DVTF	0x00000004
 300#define VRCR_VTREN	0x00000002
 301#define VRCR_VTDEN	0x00000001
 302
 303#define VTCR_PPCHK	0x00000008
 304#define VTCR_GCHK	0x00000004
 305#define VTCR_VPPTI	0x00000002
 306#define VTCR_VGTI	0x00000001
 307
 308#define CR		0x00
 309#define CFG		0x04
 310#define MEAR		0x08
 311#define PTSCR		0x0c
 312#define	ISR		0x10
 313#define	IMR		0x14
 314#define	IER		0x18
 315#define	IHR		0x1c
 316#define TXDP		0x20
 317#define TXDP_HI		0x24
 318#define TXCFG		0x28
 319#define GPIOR		0x2c
 320#define RXDP		0x30
 321#define RXDP_HI		0x34
 322#define RXCFG		0x38
 323#define PQCR		0x3c
 324#define WCSR		0x40
 325#define PCR		0x44
 326#define RFCR		0x48
 327#define RFDR		0x4c
 328
 329#define SRR		0x58
 330
 331#define VRCR		0xbc
 332#define VTCR		0xc0
 333#define VDR		0xc4
 334#define CCSR		0xcc
 335
 336#define TBICR		0xe0
 337#define TBISR		0xe4
 338#define TANAR		0xe8
 339#define TANLPAR		0xec
 340#define TANER		0xf0
 341#define TESR		0xf4
 342
 343#define TBICR_MR_AN_ENABLE	0x00001000
 344#define TBICR_MR_RESTART_AN	0x00000200
 345
 346#define TBISR_MR_LINK_STATUS	0x00000020
 347#define TBISR_MR_AN_COMPLETE	0x00000004
 348
 349#define TANAR_PS2 		0x00000100
 350#define TANAR_PS1 		0x00000080
 351#define TANAR_HALF_DUP 		0x00000040
 352#define TANAR_FULL_DUP 		0x00000020
 353
 354#define GPIOR_GP5_OE		0x00000200
 355#define GPIOR_GP4_OE		0x00000100
 356#define GPIOR_GP3_OE		0x00000080
 357#define GPIOR_GP2_OE		0x00000040
 358#define GPIOR_GP1_OE		0x00000020
 359#define GPIOR_GP3_OUT		0x00000004
 360#define GPIOR_GP1_OUT		0x00000001
 361
 362#define LINK_AUTONEGOTIATE	0x01
 363#define LINK_DOWN		0x02
 364#define LINK_UP			0x04
 365
 366#define HW_ADDR_LEN	sizeof(dma_addr_t)
 367#define desc_addr_set(desc, addr)				\
 368	do {							\
 369		((desc)[0] = cpu_to_le32(addr));		\
 370		if (HW_ADDR_LEN == 8)		 		\
 371			(desc)[1] = cpu_to_le32(((u64)addr) >> 32);	\
 372	} while(0)
 373#define desc_addr_get(desc)					\
 374	(le32_to_cpu((desc)[0]) | \
 375	(HW_ADDR_LEN == 8 ? ((dma_addr_t)le32_to_cpu((desc)[1]))<<32 : 0))
 376
 377#define DESC_LINK		0
 378#define DESC_BUFPTR		(DESC_LINK + HW_ADDR_LEN/4)
 379#define DESC_CMDSTS		(DESC_BUFPTR + HW_ADDR_LEN/4)
 380#define DESC_EXTSTS		(DESC_CMDSTS + 4/4)
 381
 382#define CMDSTS_OWN	0x80000000
 383#define CMDSTS_MORE	0x40000000
 384#define CMDSTS_INTR	0x20000000
 385#define CMDSTS_ERR	0x10000000
 386#define CMDSTS_OK	0x08000000
 387#define CMDSTS_RUNT	0x00200000
 388#define CMDSTS_LEN_MASK	0x0000ffff
 389
 390#define CMDSTS_DEST_MASK	0x01800000
 391#define CMDSTS_DEST_SELF	0x00800000
 392#define CMDSTS_DEST_MULTI	0x01000000
 393
 394#define DESC_SIZE	8		/* Should be cache line sized */
 395
 396struct rx_info {
 397	spinlock_t	lock;
 398	int		up;
 399	unsigned long	idle;
 400
 401	struct sk_buff	*skbs[NR_RX_DESC];
 402
 403	__le32		*next_rx_desc;
 404	u16		next_rx, next_empty;
 405
 406	__le32		*descs;
 407	dma_addr_t	phy_descs;
 408};
 409
 410
 411struct ns83820 {
 412	u8			__iomem *base;
 413
 414	struct pci_dev		*pci_dev;
 415	struct net_device	*ndev;
 416
 417	struct rx_info		rx_info;
 418	struct tasklet_struct	rx_tasklet;
 419
 420	unsigned		ihr;
 421	struct work_struct	tq_refill;
 422
 423	/* protects everything below.  irqsave when using. */
 424	spinlock_t		misc_lock;
 425
 426	u32			CFG_cache;
 427
 428	u32			MEAR_cache;
 429	u32			IMR_cache;
 430
 431	unsigned		linkstate;
 432
 433	spinlock_t	tx_lock;
 434
 435	u16		tx_done_idx;
 436	u16		tx_idx;
 437	volatile u16	tx_free_idx;	/* idx of free desc chain */
 438	u16		tx_intr_idx;
 439
 440	atomic_t	nr_tx_skbs;
 441	struct sk_buff	*tx_skbs[NR_TX_DESC];
 442
 443	char		pad[16] __attribute__((aligned(16)));
 444	__le32		*tx_descs;
 445	dma_addr_t	tx_phy_descs;
 446
 447	struct timer_list	tx_watchdog;
 448};
 449
 450static inline struct ns83820 *PRIV(struct net_device *dev)
 451{
 452	return netdev_priv(dev);
 453}
 454
 455#define __kick_rx(dev)	writel(CR_RXE, dev->base + CR)
 456
 457static inline void kick_rx(struct net_device *ndev)
 458{
 459	struct ns83820 *dev = PRIV(ndev);
 460	dprintk("kick_rx: maybe kicking\n");
 461	if (test_and_clear_bit(0, &dev->rx_info.idle)) {
 462		dprintk("actually kicking\n");
 463		writel(dev->rx_info.phy_descs +
 464			(4 * DESC_SIZE * dev->rx_info.next_rx),
 465		       dev->base + RXDP);
 466		if (dev->rx_info.next_rx == dev->rx_info.next_empty)
 467			printk(KERN_DEBUG "%s: uh-oh: next_rx == next_empty???\n",
 468				ndev->name);
 469		__kick_rx(dev);
 470	}
 471}
 472
 473//free = (tx_done_idx + NR_TX_DESC-2 - free_idx) % NR_TX_DESC
 474#define start_tx_okay(dev)	\
 475	(((NR_TX_DESC-2 + dev->tx_done_idx - dev->tx_free_idx) % NR_TX_DESC) > MIN_TX_DESC_FREE)
 476
 477/* Packet Receiver
 478 *
 479 * The hardware supports linked lists of receive descriptors for
 480 * which ownership is transferred back and forth by means of an
 481 * ownership bit.  While the hardware does support the use of a
 482 * ring for receive descriptors, we only make use of a chain in
 483 * an attempt to reduce bus traffic under heavy load scenarios.
 484 * This will also make bugs a bit more obvious.  The current code
 485 * only makes use of a single rx chain; I hope to implement
 486 * priority based rx for version 1.0.  Goal: even under overload
 487 * conditions, still route realtime traffic with as low jitter as
 488 * possible.
 489 */
 490static inline void build_rx_desc(struct ns83820 *dev, __le32 *desc, dma_addr_t link, dma_addr_t buf, u32 cmdsts, u32 extsts)
 491{
 492	desc_addr_set(desc + DESC_LINK, link);
 493	desc_addr_set(desc + DESC_BUFPTR, buf);
 494	desc[DESC_EXTSTS] = cpu_to_le32(extsts);
 495	mb();
 496	desc[DESC_CMDSTS] = cpu_to_le32(cmdsts);
 497}
 498
 499#define nr_rx_empty(dev) ((NR_RX_DESC-2 + dev->rx_info.next_rx - dev->rx_info.next_empty) % NR_RX_DESC)
 500static inline int ns83820_add_rx_skb(struct ns83820 *dev, struct sk_buff *skb)
 501{
 502	unsigned next_empty;
 503	u32 cmdsts;
 504	__le32 *sg;
 505	dma_addr_t buf;
 506
 507	next_empty = dev->rx_info.next_empty;
 508
 509	/* don't overrun last rx marker */
 510	if (unlikely(nr_rx_empty(dev) <= 2)) {
 511		kfree_skb(skb);
 512		return 1;
 513	}
 514
 515#if 0
 516	dprintk("next_empty[%d] nr_used[%d] next_rx[%d]\n",
 517		dev->rx_info.next_empty,
 518		dev->rx_info.nr_used,
 519		dev->rx_info.next_rx
 520		);
 521#endif
 522
 523	sg = dev->rx_info.descs + (next_empty * DESC_SIZE);
 524	BUG_ON(NULL != dev->rx_info.skbs[next_empty]);
 525	dev->rx_info.skbs[next_empty] = skb;
 526
 527	dev->rx_info.next_empty = (next_empty + 1) % NR_RX_DESC;
 528	cmdsts = REAL_RX_BUF_SIZE | CMDSTS_INTR;
 529	buf = dma_map_single(&dev->pci_dev->dev, skb->data, REAL_RX_BUF_SIZE,
 530			     DMA_FROM_DEVICE);
 531	build_rx_desc(dev, sg, 0, buf, cmdsts, 0);
 532	/* update link of previous rx */
 533	if (likely(next_empty != dev->rx_info.next_rx))
 534		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));
 535
 536	return 0;
 537}
 538
 539static inline int rx_refill(struct net_device *ndev, gfp_t gfp)
 540{
 541	struct ns83820 *dev = PRIV(ndev);
 542	unsigned i;
 543	unsigned long flags = 0;
 544
 545	if (unlikely(nr_rx_empty(dev) <= 2))
 546		return 0;
 547
 548	dprintk("rx_refill(%p)\n", ndev);
 549	if (gfp == GFP_ATOMIC)
 550		spin_lock_irqsave(&dev->rx_info.lock, flags);
 551	for (i=0; i<NR_RX_DESC; i++) {
 552		struct sk_buff *skb;
 553		long res;
 554
 555		/* extra 16 bytes for alignment */
 556		skb = __netdev_alloc_skb(ndev, REAL_RX_BUF_SIZE+16, gfp);
 557		if (unlikely(!skb))
 558			break;
 559
 560		skb_reserve(skb, skb->data - PTR_ALIGN(skb->data, 16));
 561		if (gfp != GFP_ATOMIC)
 562			spin_lock_irqsave(&dev->rx_info.lock, flags);
 563		res = ns83820_add_rx_skb(dev, skb);
 564		if (gfp != GFP_ATOMIC)
 565			spin_unlock_irqrestore(&dev->rx_info.lock, flags);
 566		if (res) {
 567			i = 1;
 568			break;
 569		}
 570	}
 571	if (gfp == GFP_ATOMIC)
 572		spin_unlock_irqrestore(&dev->rx_info.lock, flags);
 573
 574	return i ? 0 : -ENOMEM;
 575}
 576
 577static void rx_refill_atomic(struct net_device *ndev)
 578{
 579	rx_refill(ndev, GFP_ATOMIC);
 580}
 581
 582/* REFILL */
 583static inline void queue_refill(struct work_struct *work)
 584{
 585	struct ns83820 *dev = container_of(work, struct ns83820, tq_refill);
 586	struct net_device *ndev = dev->ndev;
 587
 588	rx_refill(ndev, GFP_KERNEL);
 589	if (dev->rx_info.up)
 590		kick_rx(ndev);
 591}
 592
 593static inline void clear_rx_desc(struct ns83820 *dev, unsigned i)
 594{
 595	build_rx_desc(dev, dev->rx_info.descs + (DESC_SIZE * i), 0, 0, CMDSTS_OWN, 0);
 596}
 597
 598static void phy_intr(struct net_device *ndev)
 599{
 600	struct ns83820 *dev = PRIV(ndev);
 601	static const char *speeds[] = { "10", "100", "1000", "1000(?)", "1000F" };
 602	u32 cfg, new_cfg;
 603	u32 tanar, tanlpar;
 604	int speed, fullduplex, newlinkstate;
 605
 606	cfg = readl(dev->base + CFG) ^ SPDSTS_POLARITY;
 607
 608	if (dev->CFG_cache & CFG_TBI_EN) {
 609		u32 __maybe_unused tbisr;
 610
 611		/* we have an optical transceiver */
 612		tbisr = readl(dev->base + TBISR);
 613		tanar = readl(dev->base + TANAR);
 614		tanlpar = readl(dev->base + TANLPAR);
 615		dprintk("phy_intr: tbisr=%08x, tanar=%08x, tanlpar=%08x\n",
 616			tbisr, tanar, tanlpar);
 617
 618		if ( (fullduplex = (tanlpar & TANAR_FULL_DUP) &&
 619		      (tanar & TANAR_FULL_DUP)) ) {
 620
 621			/* both of us are full duplex */
 622			writel(readl(dev->base + TXCFG)
 623			       | TXCFG_CSI | TXCFG_HBI | TXCFG_ATP,
 624			       dev->base + TXCFG);
 625			writel(readl(dev->base + RXCFG) | RXCFG_RX_FD,
 626			       dev->base + RXCFG);
 627			/* Light up full duplex LED */
 628			writel(readl(dev->base + GPIOR) | GPIOR_GP1_OUT,
 629			       dev->base + GPIOR);
 630
 631		} else if (((tanlpar & TANAR_HALF_DUP) &&
 632			    (tanar & TANAR_HALF_DUP)) ||
 633			   ((tanlpar & TANAR_FULL_DUP) &&
 634			    (tanar & TANAR_HALF_DUP)) ||
 635			   ((tanlpar & TANAR_HALF_DUP) &&
 636			    (tanar & TANAR_FULL_DUP))) {
 637
 638			/* one or both of us are half duplex */
 639			writel((readl(dev->base + TXCFG)
 640				& ~(TXCFG_CSI | TXCFG_HBI)) | TXCFG_ATP,
 641			       dev->base + TXCFG);
 642			writel(readl(dev->base + RXCFG) & ~RXCFG_RX_FD,
 643			       dev->base + RXCFG);
 644			/* Turn off full duplex LED */
 645			writel(readl(dev->base + GPIOR) & ~GPIOR_GP1_OUT,
 646			       dev->base + GPIOR);
 647		}
 648
 649		speed = 4; /* 1000F */
 650
 651	} else {
 652		/* we have a copper transceiver */
 653		new_cfg = dev->CFG_cache & ~(CFG_SB | CFG_MODE_1000 | CFG_SPDSTS);
 654
 655		if (cfg & CFG_SPDSTS1)
 656			new_cfg |= CFG_MODE_1000;
 657		else
 658			new_cfg &= ~CFG_MODE_1000;
 659
 660		speed = ((cfg / CFG_SPDSTS0) & 3);
 661		fullduplex = (cfg & CFG_DUPSTS);
 662
 663		if (fullduplex) {
 664			new_cfg |= CFG_SB;
 665			writel(readl(dev->base + TXCFG)
 666					| TXCFG_CSI | TXCFG_HBI,
 667			       dev->base + TXCFG);
 668			writel(readl(dev->base + RXCFG) | RXCFG_RX_FD,
 669			       dev->base + RXCFG);
 670		} else {
 671			writel(readl(dev->base + TXCFG)
 672					& ~(TXCFG_CSI | TXCFG_HBI),
 673			       dev->base + TXCFG);
 674			writel(readl(dev->base + RXCFG) & ~(RXCFG_RX_FD),
 675			       dev->base + RXCFG);
 676		}
 677
 678		if ((cfg & CFG_LNKSTS) &&
 679		    ((new_cfg ^ dev->CFG_cache) != 0)) {
 680			writel(new_cfg, dev->base + CFG);
 681			dev->CFG_cache = new_cfg;
 682		}
 683
 684		dev->CFG_cache &= ~CFG_SPDSTS;
 685		dev->CFG_cache |= cfg & CFG_SPDSTS;
 686	}
 687
 688	newlinkstate = (cfg & CFG_LNKSTS) ? LINK_UP : LINK_DOWN;
 689
 690	if (newlinkstate & LINK_UP &&
 691	    dev->linkstate != newlinkstate) {
 692		netif_start_queue(ndev);
 693		netif_wake_queue(ndev);
 694		printk(KERN_INFO "%s: link now %s mbps, %s duplex and up.\n",
 695			ndev->name,
 696			speeds[speed],
 697			fullduplex ? "full" : "half");
 698	} else if (newlinkstate & LINK_DOWN &&
 699		   dev->linkstate != newlinkstate) {
 700		netif_stop_queue(ndev);
 701		printk(KERN_INFO "%s: link now down.\n", ndev->name);
 702	}
 703
 704	dev->linkstate = newlinkstate;
 705}
 706
 707static int ns83820_setup_rx(struct net_device *ndev)
 708{
 709	struct ns83820 *dev = PRIV(ndev);
 710	unsigned i;
 711	int ret;
 712
 713	dprintk("ns83820_setup_rx(%p)\n", ndev);
 714
 715	dev->rx_info.idle = 1;
 716	dev->rx_info.next_rx = 0;
 717	dev->rx_info.next_rx_desc = dev->rx_info.descs;
 718	dev->rx_info.next_empty = 0;
 719
 720	for (i=0; i<NR_RX_DESC; i++)
 721		clear_rx_desc(dev, i);
 722
 723	writel(0, dev->base + RXDP_HI);
 724	writel(dev->rx_info.phy_descs, dev->base + RXDP);
 725
 726	ret = rx_refill(ndev, GFP_KERNEL);
 727	if (!ret) {
 728		dprintk("starting receiver\n");
 729		/* prevent the interrupt handler from stomping on us */
 730		spin_lock_irq(&dev->rx_info.lock);
 731
 732		writel(0x0001, dev->base + CCSR);
 733		writel(0, dev->base + RFCR);
 734		writel(0x7fc00000, dev->base + RFCR);
 735		writel(0xffc00000, dev->base + RFCR);
 736
 737		dev->rx_info.up = 1;
 738
 739		phy_intr(ndev);
 740
 741		/* Okay, let it rip */
 742		spin_lock(&dev->misc_lock);
 743		dev->IMR_cache |= ISR_PHY;
 744		dev->IMR_cache |= ISR_RXRCMP;
 745		//dev->IMR_cache |= ISR_RXERR;
 746		//dev->IMR_cache |= ISR_RXOK;
 747		dev->IMR_cache |= ISR_RXORN;
 748		dev->IMR_cache |= ISR_RXSOVR;
 749		dev->IMR_cache |= ISR_RXDESC;
 750		dev->IMR_cache |= ISR_RXIDLE;
 751		dev->IMR_cache |= ISR_TXDESC;
 752		dev->IMR_cache |= ISR_TXIDLE;
 753
 754		writel(dev->IMR_cache, dev->base + IMR);
 755		writel(1, dev->base + IER);
 756		spin_unlock(&dev->misc_lock);
 757
 758		kick_rx(ndev);
 759
 760		spin_unlock_irq(&dev->rx_info.lock);
 761	}
 762	return ret;
 763}
 764
 765static void ns83820_cleanup_rx(struct ns83820 *dev)
 766{
 767	unsigned i;
 768	unsigned long flags;
 769
 770	dprintk("ns83820_cleanup_rx(%p)\n", dev);
 771
 772	/* disable receive interrupts */
 773	spin_lock_irqsave(&dev->misc_lock, flags);
 774	dev->IMR_cache &= ~(ISR_RXOK | ISR_RXDESC | ISR_RXERR | ISR_RXEARLY | ISR_RXIDLE);
 775	writel(dev->IMR_cache, dev->base + IMR);
 776	spin_unlock_irqrestore(&dev->misc_lock, flags);
 777
 778	/* synchronize with the interrupt handler and kill it */
 779	dev->rx_info.up = 0;
 780	synchronize_irq(dev->pci_dev->irq);
 781
 782	/* touch the pci bus... */
 783	readl(dev->base + IMR);
 784
 785	/* assumes the transmitter is already disabled and reset */
 786	writel(0, dev->base + RXDP_HI);
 787	writel(0, dev->base + RXDP);
 788
 789	for (i=0; i<NR_RX_DESC; i++) {
 790		struct sk_buff *skb = dev->rx_info.skbs[i];
 791		dev->rx_info.skbs[i] = NULL;
 792		clear_rx_desc(dev, i);
 793		kfree_skb(skb);
 794	}
 795}
 796
 797static void ns83820_rx_kick(struct net_device *ndev)
 798{
 799	struct ns83820 *dev = PRIV(ndev);
 800	/*if (nr_rx_empty(dev) >= NR_RX_DESC/4)*/ {
 801		if (dev->rx_info.up) {
 802			rx_refill_atomic(ndev);
 803			kick_rx(ndev);
 804		}
 805	}
 806
 807	if (dev->rx_info.up && nr_rx_empty(dev) > NR_RX_DESC*3/4)
 808		schedule_work(&dev->tq_refill);
 809	else
 810		kick_rx(ndev);
 811	if (dev->rx_info.idle)
 812		printk(KERN_DEBUG "%s: BAD\n", ndev->name);
 813}
 814
 815/* rx_irq
 816 *
 817 */
 818static void rx_irq(struct net_device *ndev)
 819{
 820	struct ns83820 *dev = PRIV(ndev);
 821	struct rx_info *info = &dev->rx_info;
 822	unsigned next_rx;
 823	int rx_rc, len;
 824	u32 cmdsts;
 825	__le32 *desc;
 826	unsigned long flags;
 827	int nr = 0;
 828
 829	dprintk("rx_irq(%p)\n", ndev);
 830	dprintk("rxdp: %08x, descs: %08lx next_rx[%d]: %p next_empty[%d]: %p\n",
 831		readl(dev->base + RXDP),
 832		(long)(dev->rx_info.phy_descs),
 833		(int)dev->rx_info.next_rx,
 834		(dev->rx_info.descs + (DESC_SIZE * dev->rx_info.next_rx)),
 835		(int)dev->rx_info.next_empty,
 836		(dev->rx_info.descs + (DESC_SIZE * dev->rx_info.next_empty))
 837		);
 838
 839	spin_lock_irqsave(&info->lock, flags);
 840	if (!info->up)
 841		goto out;
 842
 843	dprintk("walking descs\n");
 844	next_rx = info->next_rx;
 845	desc = info->next_rx_desc;
 846	while ((CMDSTS_OWN & (cmdsts = le32_to_cpu(desc[DESC_CMDSTS]))) &&
 847	       (cmdsts != CMDSTS_OWN)) {
 848		struct sk_buff *skb;
 849		u32 extsts = le32_to_cpu(desc[DESC_EXTSTS]);
 850		dma_addr_t bufptr = desc_addr_get(desc + DESC_BUFPTR);
 851
 852		dprintk("cmdsts: %08x\n", cmdsts);
 853		dprintk("link: %08x\n", cpu_to_le32(desc[DESC_LINK]));
 854		dprintk("extsts: %08x\n", extsts);
 855
 856		skb = info->skbs[next_rx];
 857		info->skbs[next_rx] = NULL;
 858		info->next_rx = (next_rx + 1) % NR_RX_DESC;
 859
 860		mb();
 861		clear_rx_desc(dev, next_rx);
 862
 863		dma_unmap_single(&dev->pci_dev->dev, bufptr, RX_BUF_SIZE,
 864				 DMA_FROM_DEVICE);
 865		len = cmdsts & CMDSTS_LEN_MASK;
 866#ifdef NS83820_VLAN_ACCEL_SUPPORT
 867		/* NH: As was mentioned below, this chip is kinda
 868		 * brain dead about vlan tag stripping.  Frames
 869		 * that are 64 bytes with a vlan header appended
 870		 * like arp frames, or pings, are flagged as Runts
 871		 * when the tag is stripped and hardware.  This
 872		 * also means that the OK bit in the descriptor
 873		 * is cleared when the frame comes in so we have
 874		 * to do a specific length check here to make sure
 875		 * the frame would have been ok, had we not stripped
 876		 * the tag.
 877		 */
 878		if (likely((CMDSTS_OK & cmdsts) ||
 879			((cmdsts & CMDSTS_RUNT) && len >= 56))) {
 880#else
 881		if (likely(CMDSTS_OK & cmdsts)) {
 882#endif
 883			skb_put(skb, len);
 884			if (unlikely(!skb))
 885				goto netdev_mangle_me_harder_failed;
 886			if (cmdsts & CMDSTS_DEST_MULTI)
 887				ndev->stats.multicast++;
 888			ndev->stats.rx_packets++;
 889			ndev->stats.rx_bytes += len;
 890			if ((extsts & 0x002a0000) && !(extsts & 0x00540000)) {
 891				skb->ip_summed = CHECKSUM_UNNECESSARY;
 892			} else {
 893				skb_checksum_none_assert(skb);
 894			}
 895			skb->protocol = eth_type_trans(skb, ndev);
 896#ifdef NS83820_VLAN_ACCEL_SUPPORT
 897			if(extsts & EXTSTS_VPKT) {
 898				unsigned short tag;
 899
 900				tag = ntohs(extsts & EXTSTS_VTG_MASK);
 901				__vlan_hwaccel_put_tag(skb, htons(ETH_P_IPV6), tag);
 902			}
 903#endif
 904			rx_rc = netif_rx(skb);
 905			if (NET_RX_DROP == rx_rc) {
 906netdev_mangle_me_harder_failed:
 907				ndev->stats.rx_dropped++;
 908			}
 909		} else {
 910			dev_kfree_skb_irq(skb);
 911		}
 912
 913		nr++;
 914		next_rx = info->next_rx;
 915		desc = info->descs + (DESC_SIZE * next_rx);
 916	}
 917	info->next_rx = next_rx;
 918	info->next_rx_desc = info->descs + (DESC_SIZE * next_rx);
 919
 920out:
 921	if (0 && !nr) {
 922		Dprintk("dazed: cmdsts_f: %08x\n", cmdsts);
 923	}
 924
 925	spin_unlock_irqrestore(&info->lock, flags);
 926}
 927
 928static void rx_action(struct tasklet_struct *t)
 929{
 930	struct ns83820 *dev = from_tasklet(dev, t, rx_tasklet);
 931	struct net_device *ndev = dev->ndev;
 932	rx_irq(ndev);
 933	writel(ihr, dev->base + IHR);
 934
 935	spin_lock_irq(&dev->misc_lock);
 936	dev->IMR_cache |= ISR_RXDESC;
 937	writel(dev->IMR_cache, dev->base + IMR);
 938	spin_unlock_irq(&dev->misc_lock);
 939
 940	rx_irq(ndev);
 941	ns83820_rx_kick(ndev);
 942}
 943
 944/* Packet Transmit code
 945 */
 946static inline void kick_tx(struct ns83820 *dev)
 947{
 948	dprintk("kick_tx(%p): tx_idx=%d free_idx=%d\n",
 949		dev, dev->tx_idx, dev->tx_free_idx);
 950	writel(CR_TXE, dev->base + CR);
 951}
 952
 953/* No spinlock needed on the transmit irq path as the interrupt handler is
 954 * serialized.
 955 */
 956static void do_tx_done(struct net_device *ndev)
 957{
 958	struct ns83820 *dev = PRIV(ndev);
 959	u32 cmdsts, tx_done_idx;
 960	__le32 *desc;
 961
 962	dprintk("do_tx_done(%p)\n", ndev);
 963	tx_done_idx = dev->tx_done_idx;
 964	desc = dev->tx_descs + (tx_done_idx * DESC_SIZE);
 965
 966	dprintk("tx_done_idx=%d free_idx=%d cmdsts=%08x\n",
 967		tx_done_idx, dev->tx_free_idx, le32_to_cpu(desc[DESC_CMDSTS]));
 968	while ((tx_done_idx != dev->tx_free_idx) &&
 969	       !(CMDSTS_OWN & (cmdsts = le32_to_cpu(desc[DESC_CMDSTS]))) ) {
 970		struct sk_buff *skb;
 971		unsigned len;
 972		dma_addr_t addr;
 973
 974		if (cmdsts & CMDSTS_ERR)
 975			ndev->stats.tx_errors++;
 976		if (cmdsts & CMDSTS_OK)
 977			ndev->stats.tx_packets++;
 978		if (cmdsts & CMDSTS_OK)
 979			ndev->stats.tx_bytes += cmdsts & 0xffff;
 980
 981		dprintk("tx_done_idx=%d free_idx=%d cmdsts=%08x\n",
 982			tx_done_idx, dev->tx_free_idx, cmdsts);
 983		skb = dev->tx_skbs[tx_done_idx];
 984		dev->tx_skbs[tx_done_idx] = NULL;
 985		dprintk("done(%p)\n", skb);
 986
 987		len = cmdsts & CMDSTS_LEN_MASK;
 988		addr = desc_addr_get(desc + DESC_BUFPTR);
 989		if (skb) {
 990			dma_unmap_single(&dev->pci_dev->dev, addr, len,
 991					 DMA_TO_DEVICE);
 992			dev_consume_skb_irq(skb);
 993			atomic_dec(&dev->nr_tx_skbs);
 994		} else
 995			dma_unmap_page(&dev->pci_dev->dev, addr, len,
 996				       DMA_TO_DEVICE);
 997
 998		tx_done_idx = (tx_done_idx + 1) % NR_TX_DESC;
 999		dev->tx_done_idx = tx_done_idx;
1000		desc[DESC_CMDSTS] = cpu_to_le32(0);
1001		mb();
1002		desc = dev->tx_descs + (tx_done_idx * DESC_SIZE);
1003	}
1004
1005	/* Allow network stack to resume queueing packets after we've
1006	 * finished transmitting at least 1/4 of the packets in the queue.
1007	 */
1008	if (netif_queue_stopped(ndev) && start_tx_okay(dev)) {
1009		dprintk("start_queue(%p)\n", ndev);
1010		netif_start_queue(ndev);
1011		netif_wake_queue(ndev);
1012	}
1013}
1014
1015static void ns83820_cleanup_tx(struct ns83820 *dev)
1016{
1017	unsigned i;
1018
1019	for (i=0; i<NR_TX_DESC; i++) {
1020		struct sk_buff *skb = dev->tx_skbs[i];
1021		dev->tx_skbs[i] = NULL;
1022		if (skb) {
1023			__le32 *desc = dev->tx_descs + (i * DESC_SIZE);
1024			dma_unmap_single(&dev->pci_dev->dev,
1025					 desc_addr_get(desc + DESC_BUFPTR),
1026					 le32_to_cpu(desc[DESC_CMDSTS]) & CMDSTS_LEN_MASK,
1027					 DMA_TO_DEVICE);
1028			dev_kfree_skb_irq(skb);
1029			atomic_dec(&dev->nr_tx_skbs);
1030		}
1031	}
1032
1033	memset(dev->tx_descs, 0, NR_TX_DESC * DESC_SIZE * 4);
1034}
1035
1036/* transmit routine.  This code relies on the network layer serializing
1037 * its calls in, but will run happily in parallel with the interrupt
1038 * handler.  This code currently has provisions for fragmenting tx buffers
1039 * while trying to track down a bug in either the zero copy code or
1040 * the tx fifo (hence the MAX_FRAG_LEN).
1041 */
1042static netdev_tx_t ns83820_hard_start_xmit(struct sk_buff *skb,
1043					   struct net_device *ndev)
1044{
1045	struct ns83820 *dev = PRIV(ndev);
1046	u32 free_idx, cmdsts, extsts;
1047	int nr_free, nr_frags;
1048	unsigned tx_done_idx, last_idx;
1049	dma_addr_t buf;
1050	unsigned len;
1051	skb_frag_t *frag;
1052	int stopped = 0;
1053	int do_intr = 0;
1054	volatile __le32 *first_desc;
1055
1056	dprintk("ns83820_hard_start_xmit\n");
1057
1058	nr_frags =  skb_shinfo(skb)->nr_frags;
1059again:
1060	if (unlikely(dev->CFG_cache & CFG_LNKSTS)) {
1061		netif_stop_queue(ndev);
1062		if (unlikely(dev->CFG_cache & CFG_LNKSTS))
1063			return NETDEV_TX_BUSY;
1064		netif_start_queue(ndev);
1065	}
1066
1067	last_idx = free_idx = dev->tx_free_idx;
1068	tx_done_idx = dev->tx_done_idx;
1069	nr_free = (tx_done_idx + NR_TX_DESC-2 - free_idx) % NR_TX_DESC;
1070	nr_free -= 1;
1071	if (nr_free <= nr_frags) {
1072		dprintk("stop_queue - not enough(%p)\n", ndev);
1073		netif_stop_queue(ndev);
1074
1075		/* Check again: we may have raced with a tx done irq */
1076		if (dev->tx_done_idx != tx_done_idx) {
1077			dprintk("restart queue(%p)\n", ndev);
1078			netif_start_queue(ndev);
1079			goto again;
1080		}
1081		return NETDEV_TX_BUSY;
1082	}
1083
1084	if (free_idx == dev->tx_intr_idx) {
1085		do_intr = 1;
1086		dev->tx_intr_idx = (dev->tx_intr_idx + NR_TX_DESC/4) % NR_TX_DESC;
1087	}
1088
1089	nr_free -= nr_frags;
1090	if (nr_free < MIN_TX_DESC_FREE) {
1091		dprintk("stop_queue - last entry(%p)\n", ndev);
1092		netif_stop_queue(ndev);
1093		stopped = 1;
1094	}
1095
1096	frag = skb_shinfo(skb)->frags;
1097	if (!nr_frags)
1098		frag = NULL;
1099	extsts = 0;
1100	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1101		extsts |= EXTSTS_IPPKT;
1102		if (IPPROTO_TCP == ip_hdr(skb)->protocol)
1103			extsts |= EXTSTS_TCPPKT;
1104		else if (IPPROTO_UDP == ip_hdr(skb)->protocol)
1105			extsts |= EXTSTS_UDPPKT;
1106	}
1107
1108#ifdef NS83820_VLAN_ACCEL_SUPPORT
1109	if (skb_vlan_tag_present(skb)) {
1110		/* fetch the vlan tag info out of the
1111		 * ancillary data if the vlan code
1112		 * is using hw vlan acceleration
1113		 */
1114		short tag = skb_vlan_tag_get(skb);
1115		extsts |= (EXTSTS_VPKT | htons(tag));
1116	}
1117#endif
1118
1119	len = skb->len;
1120	if (nr_frags)
1121		len -= skb->data_len;
1122	buf = dma_map_single(&dev->pci_dev->dev, skb->data, len,
1123			     DMA_TO_DEVICE);
1124
1125	first_desc = dev->tx_descs + (free_idx * DESC_SIZE);
1126
1127	for (;;) {
1128		volatile __le32 *desc = dev->tx_descs + (free_idx * DESC_SIZE);
1129
1130		dprintk("frag[%3u]: %4u @ 0x%08Lx\n", free_idx, len,
1131			(unsigned long long)buf);
1132		last_idx = free_idx;
1133		free_idx = (free_idx + 1) % NR_TX_DESC;
1134		desc[DESC_LINK] = cpu_to_le32(dev->tx_phy_descs + (free_idx * DESC_SIZE * 4));
1135		desc_addr_set(desc + DESC_BUFPTR, buf);
1136		desc[DESC_EXTSTS] = cpu_to_le32(extsts);
1137
1138		cmdsts = ((nr_frags) ? CMDSTS_MORE : do_intr ? CMDSTS_INTR : 0);
1139		cmdsts |= (desc == first_desc) ? 0 : CMDSTS_OWN;
1140		cmdsts |= len;
1141		desc[DESC_CMDSTS] = cpu_to_le32(cmdsts);
1142
1143		if (!nr_frags)
1144			break;
1145
1146		buf = skb_frag_dma_map(&dev->pci_dev->dev, frag, 0,
1147				       skb_frag_size(frag), DMA_TO_DEVICE);
1148		dprintk("frag: buf=%08Lx  page=%08lx offset=%08lx\n",
1149			(long long)buf, (long) page_to_pfn(frag->page),
1150			frag->page_offset);
1151		len = skb_frag_size(frag);
1152		frag++;
1153		nr_frags--;
1154	}
1155	dprintk("done pkt\n");
1156
1157	spin_lock_irq(&dev->tx_lock);
1158	dev->tx_skbs[last_idx] = skb;
1159	first_desc[DESC_CMDSTS] |= cpu_to_le32(CMDSTS_OWN);
1160	dev->tx_free_idx = free_idx;
1161	atomic_inc(&dev->nr_tx_skbs);
1162	spin_unlock_irq(&dev->tx_lock);
1163
1164	kick_tx(dev);
1165
1166	/* Check again: we may have raced with a tx done irq */
1167	if (stopped && (dev->tx_done_idx != tx_done_idx) && start_tx_okay(dev))
1168		netif_start_queue(ndev);
1169
1170	return NETDEV_TX_OK;
1171}
1172
1173static void ns83820_update_stats(struct ns83820 *dev)
1174{
1175	struct net_device *ndev = dev->ndev;
1176	u8 __iomem *base = dev->base;
1177
1178	/* the DP83820 will freeze counters, so we need to read all of them */
1179	ndev->stats.rx_errors		+= readl(base + 0x60) & 0xffff;
1180	ndev->stats.rx_crc_errors	+= readl(base + 0x64) & 0xffff;
1181	ndev->stats.rx_missed_errors	+= readl(base + 0x68) & 0xffff;
1182	ndev->stats.rx_frame_errors	+= readl(base + 0x6c) & 0xffff;
1183	/*ndev->stats.rx_symbol_errors +=*/ readl(base + 0x70);
1184	ndev->stats.rx_length_errors	+= readl(base + 0x74) & 0xffff;
1185	ndev->stats.rx_length_errors	+= readl(base + 0x78) & 0xffff;
1186	/*ndev->stats.rx_badopcode_errors += */ readl(base + 0x7c);
1187	/*ndev->stats.rx_pause_count += */  readl(base + 0x80);
1188	/*ndev->stats.tx_pause_count += */  readl(base + 0x84);
1189	ndev->stats.tx_carrier_errors	+= readl(base + 0x88) & 0xff;
1190}
1191
1192static struct net_device_stats *ns83820_get_stats(struct net_device *ndev)
1193{
1194	struct ns83820 *dev = PRIV(ndev);
1195
1196	/* somewhat overkill */
1197	spin_lock_irq(&dev->misc_lock);
1198	ns83820_update_stats(dev);
1199	spin_unlock_irq(&dev->misc_lock);
1200
1201	return &ndev->stats;
1202}
1203
1204/* Let ethtool retrieve info */
1205static int ns83820_get_link_ksettings(struct net_device *ndev,
1206				      struct ethtool_link_ksettings *cmd)
1207{
1208	struct ns83820 *dev = PRIV(ndev);
1209	u32 cfg, tbicr;
1210	int fullduplex   = 0;
1211	u32 supported;
1212
1213	/*
1214	 * Here's the list of available ethtool commands from other drivers:
1215	 *	cmd->advertising =
1216	 *	ethtool_cmd_speed_set(cmd, ...)
1217	 *	cmd->duplex =
1218	 *	cmd->port = 0;
1219	 *	cmd->phy_address =
1220	 *	cmd->transceiver = 0;
1221	 *	cmd->autoneg =
1222	 *	cmd->maxtxpkt = 0;
1223	 *	cmd->maxrxpkt = 0;
1224	 */
1225
1226	/* read current configuration */
1227	cfg   = readl(dev->base + CFG) ^ SPDSTS_POLARITY;
1228	readl(dev->base + TANAR);
1229	tbicr = readl(dev->base + TBICR);
1230
1231	fullduplex = (cfg & CFG_DUPSTS) ? 1 : 0;
1232
1233	supported = SUPPORTED_Autoneg;
1234
1235	if (dev->CFG_cache & CFG_TBI_EN) {
1236		/* we have optical interface */
1237		supported |= SUPPORTED_1000baseT_Half |
1238					SUPPORTED_1000baseT_Full |
1239					SUPPORTED_FIBRE;
1240		cmd->base.port       = PORT_FIBRE;
1241	} else {
1242		/* we have copper */
1243		supported |= SUPPORTED_10baseT_Half |
1244			SUPPORTED_10baseT_Full | SUPPORTED_100baseT_Half |
1245			SUPPORTED_100baseT_Full | SUPPORTED_1000baseT_Half |
1246			SUPPORTED_1000baseT_Full |
1247			SUPPORTED_MII;
1248		cmd->base.port = PORT_MII;
1249	}
1250
1251	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
1252						supported);
1253
1254	cmd->base.duplex = fullduplex ? DUPLEX_FULL : DUPLEX_HALF;
1255	switch (cfg / CFG_SPDSTS0 & 3) {
1256	case 2:
1257		cmd->base.speed = SPEED_1000;
1258		break;
1259	case 1:
1260		cmd->base.speed = SPEED_100;
1261		break;
1262	default:
1263		cmd->base.speed = SPEED_10;
1264		break;
1265	}
1266	cmd->base.autoneg = (tbicr & TBICR_MR_AN_ENABLE)
1267		? AUTONEG_ENABLE : AUTONEG_DISABLE;
1268	return 0;
1269}
1270
1271/* Let ethool change settings*/
1272static int ns83820_set_link_ksettings(struct net_device *ndev,
1273				      const struct ethtool_link_ksettings *cmd)
1274{
1275	struct ns83820 *dev = PRIV(ndev);
1276	u32 cfg, tanar;
1277	int have_optical = 0;
1278	int fullduplex   = 0;
1279
1280	/* read current configuration */
1281	cfg = readl(dev->base + CFG) ^ SPDSTS_POLARITY;
1282	tanar = readl(dev->base + TANAR);
1283
1284	if (dev->CFG_cache & CFG_TBI_EN) {
1285		/* we have optical */
1286		have_optical = 1;
1287		fullduplex   = (tanar & TANAR_FULL_DUP);
1288
1289	} else {
1290		/* we have copper */
1291		fullduplex = cfg & CFG_DUPSTS;
1292	}
1293
1294	spin_lock_irq(&dev->misc_lock);
1295	spin_lock(&dev->tx_lock);
1296
1297	/* Set duplex */
1298	if (cmd->base.duplex != fullduplex) {
1299		if (have_optical) {
1300			/*set full duplex*/
1301			if (cmd->base.duplex == DUPLEX_FULL) {
1302				/* force full duplex */
1303				writel(readl(dev->base + TXCFG)
1304					| TXCFG_CSI | TXCFG_HBI | TXCFG_ATP,
1305					dev->base + TXCFG);
1306				writel(readl(dev->base + RXCFG) | RXCFG_RX_FD,
1307					dev->base + RXCFG);
1308				/* Light up full duplex LED */
1309				writel(readl(dev->base + GPIOR) | GPIOR_GP1_OUT,
1310					dev->base + GPIOR);
1311			} else {
1312				/*TODO: set half duplex */
1313			}
1314
1315		} else {
1316			/*we have copper*/
1317			/* TODO: Set duplex for copper cards */
1318		}
1319		printk(KERN_INFO "%s: Duplex set via ethtool\n",
1320		ndev->name);
1321	}
1322
1323	/* Set autonegotiation */
1324	if (1) {
1325		if (cmd->base.autoneg == AUTONEG_ENABLE) {
1326			/* restart auto negotiation */
1327			writel(TBICR_MR_AN_ENABLE | TBICR_MR_RESTART_AN,
1328				dev->base + TBICR);
1329			writel(TBICR_MR_AN_ENABLE, dev->base + TBICR);
1330				dev->linkstate = LINK_AUTONEGOTIATE;
1331
1332			printk(KERN_INFO "%s: autoneg enabled via ethtool\n",
1333				ndev->name);
1334		} else {
1335			/* disable auto negotiation */
1336			writel(0x00000000, dev->base + TBICR);
1337		}
1338
1339		printk(KERN_INFO "%s: autoneg %s via ethtool\n", ndev->name,
1340				cmd->base.autoneg ? "ENABLED" : "DISABLED");
1341	}
1342
1343	phy_intr(ndev);
1344	spin_unlock(&dev->tx_lock);
1345	spin_unlock_irq(&dev->misc_lock);
1346
1347	return 0;
1348}
1349/* end ethtool get/set support -df */
1350
1351static void ns83820_get_drvinfo(struct net_device *ndev, struct ethtool_drvinfo *info)
1352{
1353	struct ns83820 *dev = PRIV(ndev);
1354	strscpy(info->driver, "ns83820", sizeof(info->driver));
1355	strscpy(info->version, VERSION, sizeof(info->version));
1356	strscpy(info->bus_info, pci_name(dev->pci_dev), sizeof(info->bus_info));
1357}
1358
1359static u32 ns83820_get_link(struct net_device *ndev)
1360{
1361	struct ns83820 *dev = PRIV(ndev);
1362	u32 cfg = readl(dev->base + CFG) ^ SPDSTS_POLARITY;
1363	return cfg & CFG_LNKSTS ? 1 : 0;
1364}
1365
1366static const struct ethtool_ops ops = {
1367	.get_drvinfo     = ns83820_get_drvinfo,
1368	.get_link        = ns83820_get_link,
1369	.get_link_ksettings = ns83820_get_link_ksettings,
1370	.set_link_ksettings = ns83820_set_link_ksettings,
1371};
1372
1373static inline void ns83820_disable_interrupts(struct ns83820 *dev)
1374{
1375	writel(0, dev->base + IMR);
1376	writel(0, dev->base + IER);
1377	readl(dev->base + IER);
1378}
1379
1380/* this function is called in irq context from the ISR */
1381static void ns83820_mib_isr(struct ns83820 *dev)
1382{
1383	unsigned long flags;
1384	spin_lock_irqsave(&dev->misc_lock, flags);
1385	ns83820_update_stats(dev);
1386	spin_unlock_irqrestore(&dev->misc_lock, flags);
1387}
1388
1389static void ns83820_do_isr(struct net_device *ndev, u32 isr);
1390static irqreturn_t ns83820_irq(int foo, void *data)
1391{
1392	struct net_device *ndev = data;
1393	struct ns83820 *dev = PRIV(ndev);
1394	u32 isr;
1395	dprintk("ns83820_irq(%p)\n", ndev);
1396
1397	dev->ihr = 0;
1398
1399	isr = readl(dev->base + ISR);
1400	dprintk("irq: %08x\n", isr);
1401	ns83820_do_isr(ndev, isr);
1402	return IRQ_HANDLED;
1403}
1404
1405static void ns83820_do_isr(struct net_device *ndev, u32 isr)
1406{
1407	struct ns83820 *dev = PRIV(ndev);
1408	unsigned long flags;
1409
1410#ifdef DEBUG
1411	if (isr & ~(ISR_PHY | ISR_RXDESC | ISR_RXEARLY | ISR_RXOK | ISR_RXERR | ISR_TXIDLE | ISR_TXOK | ISR_TXDESC))
1412		Dprintk("odd isr? 0x%08x\n", isr);
1413#endif
1414
1415	if (ISR_RXIDLE & isr) {
1416		dev->rx_info.idle = 1;
1417		Dprintk("oh dear, we are idle\n");
1418		ns83820_rx_kick(ndev);
1419	}
1420
1421	if ((ISR_RXDESC | ISR_RXOK) & isr) {
1422		prefetch(dev->rx_info.next_rx_desc);
1423
1424		spin_lock_irqsave(&dev->misc_lock, flags);
1425		dev->IMR_cache &= ~(ISR_RXDESC | ISR_RXOK);
1426		writel(dev->IMR_cache, dev->base + IMR);
1427		spin_unlock_irqrestore(&dev->misc_lock, flags);
1428
1429		tasklet_schedule(&dev->rx_tasklet);
1430		//rx_irq(ndev);
1431		//writel(4, dev->base + IHR);
1432	}
1433
1434	if ((ISR_RXIDLE | ISR_RXORN | ISR_RXDESC | ISR_RXOK | ISR_RXERR) & isr)
1435		ns83820_rx_kick(ndev);
1436
1437	if (unlikely(ISR_RXSOVR & isr)) {
1438		//printk("overrun: rxsovr\n");
1439		ndev->stats.rx_fifo_errors++;
1440	}
1441
1442	if (unlikely(ISR_RXORN & isr)) {
1443		//printk("overrun: rxorn\n");
1444		ndev->stats.rx_fifo_errors++;
1445	}
1446
1447	if ((ISR_RXRCMP & isr) && dev->rx_info.up)
1448		writel(CR_RXE, dev->base + CR);
1449
1450	if (ISR_TXIDLE & isr) {
1451		u32 txdp;
1452		txdp = readl(dev->base + TXDP);
1453		dprintk("txdp: %08x\n", txdp);
1454		txdp -= dev->tx_phy_descs;
1455		dev->tx_idx = txdp / (DESC_SIZE * 4);
1456		if (dev->tx_idx >= NR_TX_DESC) {
1457			printk(KERN_ALERT "%s: BUG -- txdp out of range\n", ndev->name);
1458			dev->tx_idx = 0;
1459		}
1460		/* The may have been a race between a pci originated read
1461		 * and the descriptor update from the cpu.  Just in case,
1462		 * kick the transmitter if the hardware thinks it is on a
1463		 * different descriptor than we are.
1464		 */
1465		if (dev->tx_idx != dev->tx_free_idx)
1466			kick_tx(dev);
1467	}
1468
1469	/* Defer tx ring processing until more than a minimum amount of
1470	 * work has accumulated
1471	 */
1472	if ((ISR_TXDESC | ISR_TXIDLE | ISR_TXOK | ISR_TXERR) & isr) {
1473		spin_lock_irqsave(&dev->tx_lock, flags);
1474		do_tx_done(ndev);
1475		spin_unlock_irqrestore(&dev->tx_lock, flags);
1476
1477		/* Disable TxOk if there are no outstanding tx packets.
1478		 */
1479		if ((dev->tx_done_idx == dev->tx_free_idx) &&
1480		    (dev->IMR_cache & ISR_TXOK)) {
1481			spin_lock_irqsave(&dev->misc_lock, flags);
1482			dev->IMR_cache &= ~ISR_TXOK;
1483			writel(dev->IMR_cache, dev->base + IMR);
1484			spin_unlock_irqrestore(&dev->misc_lock, flags);
1485		}
1486	}
1487
1488	/* The TxIdle interrupt can come in before the transmit has
1489	 * completed.  Normally we reap packets off of the combination
1490	 * of TxDesc and TxIdle and leave TxOk disabled (since it
1491	 * occurs on every packet), but when no further irqs of this
1492	 * nature are expected, we must enable TxOk.
1493	 */
1494	if ((ISR_TXIDLE & isr) && (dev->tx_done_idx != dev->tx_free_idx)) {
1495		spin_lock_irqsave(&dev->misc_lock, flags);
1496		dev->IMR_cache |= ISR_TXOK;
1497		writel(dev->IMR_cache, dev->base + IMR);
1498		spin_unlock_irqrestore(&dev->misc_lock, flags);
1499	}
1500
1501	/* MIB interrupt: one of the statistics counters is about to overflow */
1502	if (unlikely(ISR_MIB & isr))
1503		ns83820_mib_isr(dev);
1504
1505	/* PHY: Link up/down/negotiation state change */
1506	if (unlikely(ISR_PHY & isr))
1507		phy_intr(ndev);
1508
1509#if 0	/* Still working on the interrupt mitigation strategy */
1510	if (dev->ihr)
1511		writel(dev->ihr, dev->base + IHR);
1512#endif
1513}
1514
1515static void ns83820_do_reset(struct ns83820 *dev, u32 which)
1516{
1517	Dprintk("resetting chip...\n");
1518	writel(which, dev->base + CR);
1519	do {
1520		schedule();
1521	} while (readl(dev->base + CR) & which);
1522	Dprintk("okay!\n");
1523}
1524
1525static int ns83820_stop(struct net_device *ndev)
1526{
1527	struct ns83820 *dev = PRIV(ndev);
1528
1529	/* FIXME: protect against interrupt handler? */
1530	del_timer_sync(&dev->tx_watchdog);
1531
1532	ns83820_disable_interrupts(dev);
1533
1534	dev->rx_info.up = 0;
1535	synchronize_irq(dev->pci_dev->irq);
1536
1537	ns83820_do_reset(dev, CR_RST);
1538
1539	synchronize_irq(dev->pci_dev->irq);
1540
1541	spin_lock_irq(&dev->misc_lock);
1542	dev->IMR_cache &= ~(ISR_TXURN | ISR_TXIDLE | ISR_TXERR | ISR_TXDESC | ISR_TXOK);
1543	spin_unlock_irq(&dev->misc_lock);
1544
1545	ns83820_cleanup_rx(dev);
1546	ns83820_cleanup_tx(dev);
1547
1548	return 0;
1549}
1550
1551static void ns83820_tx_timeout(struct net_device *ndev, unsigned int txqueue)
1552{
1553	struct ns83820 *dev = PRIV(ndev);
1554        u32 tx_done_idx;
1555	__le32 *desc;
1556	unsigned long flags;
1557
1558	spin_lock_irqsave(&dev->tx_lock, flags);
1559
1560	tx_done_idx = dev->tx_done_idx;
1561	desc = dev->tx_descs + (tx_done_idx * DESC_SIZE);
1562
1563	printk(KERN_INFO "%s: tx_timeout: tx_done_idx=%d free_idx=%d cmdsts=%08x\n",
1564		ndev->name,
1565		tx_done_idx, dev->tx_free_idx, le32_to_cpu(desc[DESC_CMDSTS]));
1566
1567#if defined(DEBUG)
1568	{
1569		u32 isr;
1570		isr = readl(dev->base + ISR);
1571		printk("irq: %08x imr: %08x\n", isr, dev->IMR_cache);
1572		ns83820_do_isr(ndev, isr);
1573	}
1574#endif
1575
1576	do_tx_done(ndev);
1577
1578	tx_done_idx = dev->tx_done_idx;
1579	desc = dev->tx_descs + (tx_done_idx * DESC_SIZE);
1580
1581	printk(KERN_INFO "%s: after: tx_done_idx=%d free_idx=%d cmdsts=%08x\n",
1582		ndev->name,
1583		tx_done_idx, dev->tx_free_idx, le32_to_cpu(desc[DESC_CMDSTS]));
1584
1585	spin_unlock_irqrestore(&dev->tx_lock, flags);
1586}
1587
1588static void ns83820_tx_watch(struct timer_list *t)
1589{
1590	struct ns83820 *dev = from_timer(dev, t, tx_watchdog);
1591	struct net_device *ndev = dev->ndev;
1592
1593#if defined(DEBUG)
1594	printk("ns83820_tx_watch: %u %u %d\n",
1595		dev->tx_done_idx, dev->tx_free_idx, atomic_read(&dev->nr_tx_skbs)
1596		);
1597#endif
1598
1599	if (time_after(jiffies, dev_trans_start(ndev) + 1*HZ) &&
1600	    dev->tx_done_idx != dev->tx_free_idx) {
1601		printk(KERN_DEBUG "%s: ns83820_tx_watch: %u %u %d\n",
1602			ndev->name,
1603			dev->tx_done_idx, dev->tx_free_idx,
1604			atomic_read(&dev->nr_tx_skbs));
1605		ns83820_tx_timeout(ndev, UINT_MAX);
1606	}
1607
1608	mod_timer(&dev->tx_watchdog, jiffies + 2*HZ);
1609}
1610
1611static int ns83820_open(struct net_device *ndev)
1612{
1613	struct ns83820 *dev = PRIV(ndev);
1614	unsigned i;
1615	u32 desc;
1616	int ret;
1617
1618	dprintk("ns83820_open\n");
1619
1620	writel(0, dev->base + PQCR);
1621
1622	ret = ns83820_setup_rx(ndev);
1623	if (ret)
1624		goto failed;
1625
1626	memset(dev->tx_descs, 0, 4 * NR_TX_DESC * DESC_SIZE);
1627	for (i=0; i<NR_TX_DESC; i++) {
1628		dev->tx_descs[(i * DESC_SIZE) + DESC_LINK]
1629				= cpu_to_le32(
1630				  dev->tx_phy_descs
1631				  + ((i+1) % NR_TX_DESC) * DESC_SIZE * 4);
1632	}
1633
1634	dev->tx_idx = 0;
1635	dev->tx_done_idx = 0;
1636	desc = dev->tx_phy_descs;
1637	writel(0, dev->base + TXDP_HI);
1638	writel(desc, dev->base + TXDP);
1639
1640	timer_setup(&dev->tx_watchdog, ns83820_tx_watch, 0);
1641	mod_timer(&dev->tx_watchdog, jiffies + 2*HZ);
1642
1643	netif_start_queue(ndev);	/* FIXME: wait for phy to come up */
1644
1645	return 0;
1646
1647failed:
1648	ns83820_stop(ndev);
1649	return ret;
1650}
1651
1652static void ns83820_getmac(struct ns83820 *dev, struct net_device *ndev)
1653{
1654	u8 mac[ETH_ALEN];
1655	unsigned i;
1656
1657	for (i=0; i<3; i++) {
1658		u32 data;
1659
1660		/* Read from the perfect match memory: this is loaded by
1661		 * the chip from the EEPROM via the EELOAD self test.
1662		 */
1663		writel(i*2, dev->base + RFCR);
1664		data = readl(dev->base + RFDR);
1665
1666		mac[i * 2] = data;
1667		mac[i * 2 + 1] = data >> 8;
1668	}
1669	eth_hw_addr_set(ndev, mac);
1670}
1671
1672static void ns83820_set_multicast(struct net_device *ndev)
1673{
1674	struct ns83820 *dev = PRIV(ndev);
1675	u8 __iomem *rfcr = dev->base + RFCR;
1676	u32 and_mask = 0xffffffff;
1677	u32 or_mask = 0;
1678	u32 val;
1679
1680	if (ndev->flags & IFF_PROMISC)
1681		or_mask |= RFCR_AAU | RFCR_AAM;
1682	else
1683		and_mask &= ~(RFCR_AAU | RFCR_AAM);
1684
1685	if (ndev->flags & IFF_ALLMULTI || netdev_mc_count(ndev))
1686		or_mask |= RFCR_AAM;
1687	else
1688		and_mask &= ~RFCR_AAM;
1689
1690	spin_lock_irq(&dev->misc_lock);
1691	val = (readl(rfcr) & and_mask) | or_mask;
1692	/* Ramit : RFCR Write Fix doc says RFEN must be 0 modify other bits */
1693	writel(val & ~RFCR_RFEN, rfcr);
1694	writel(val, rfcr);
1695	spin_unlock_irq(&dev->misc_lock);
1696}
1697
1698static void ns83820_run_bist(struct net_device *ndev, const char *name, u32 enable, u32 done, u32 fail)
1699{
1700	struct ns83820 *dev = PRIV(ndev);
1701	int timed_out = 0;
1702	unsigned long start;
1703	u32 status;
1704	int loops = 0;
1705
1706	dprintk("%s: start %s\n", ndev->name, name);
1707
1708	start = jiffies;
1709
1710	writel(enable, dev->base + PTSCR);
1711	for (;;) {
1712		loops++;
1713		status = readl(dev->base + PTSCR);
1714		if (!(status & enable))
1715			break;
1716		if (status & done)
1717			break;
1718		if (status & fail)
1719			break;
1720		if (time_after_eq(jiffies, start + HZ)) {
1721			timed_out = 1;
1722			break;
1723		}
1724		schedule_timeout_uninterruptible(1);
1725	}
1726
1727	if (status & fail)
1728		printk(KERN_INFO "%s: %s failed! (0x%08x & 0x%08x)\n",
1729			ndev->name, name, status, fail);
1730	else if (timed_out)
1731		printk(KERN_INFO "%s: run_bist %s timed out! (%08x)\n",
1732			ndev->name, name, status);
1733
1734	dprintk("%s: done %s in %d loops\n", ndev->name, name, loops);
1735}
1736
1737#ifdef PHY_CODE_IS_FINISHED
1738static void ns83820_mii_write_bit(struct ns83820 *dev, int bit)
1739{
1740	/* drive MDC low */
1741	dev->MEAR_cache &= ~MEAR_MDC;
1742	writel(dev->MEAR_cache, dev->base + MEAR);
1743	readl(dev->base + MEAR);
1744
1745	/* enable output, set bit */
1746	dev->MEAR_cache |= MEAR_MDDIR;
1747	if (bit)
1748		dev->MEAR_cache |= MEAR_MDIO;
1749	else
1750		dev->MEAR_cache &= ~MEAR_MDIO;
1751
1752	/* set the output bit */
1753	writel(dev->MEAR_cache, dev->base + MEAR);
1754	readl(dev->base + MEAR);
1755
1756	/* Wait.  Max clock rate is 2.5MHz, this way we come in under 1MHz */
1757	udelay(1);
1758
1759	/* drive MDC high causing the data bit to be latched */
1760	dev->MEAR_cache |= MEAR_MDC;
1761	writel(dev->MEAR_cache, dev->base + MEAR);
1762	readl(dev->base + MEAR);
1763
1764	/* Wait again... */
1765	udelay(1);
1766}
1767
1768static int ns83820_mii_read_bit(struct ns83820 *dev)
1769{
1770	int bit;
1771
1772	/* drive MDC low, disable output */
1773	dev->MEAR_cache &= ~MEAR_MDC;
1774	dev->MEAR_cache &= ~MEAR_MDDIR;
1775	writel(dev->MEAR_cache, dev->base + MEAR);
1776	readl(dev->base + MEAR);
1777
1778	/* Wait.  Max clock rate is 2.5MHz, this way we come in under 1MHz */
1779	udelay(1);
1780
1781	/* drive MDC high causing the data bit to be latched */
1782	bit = (readl(dev->base + MEAR) & MEAR_MDIO) ? 1 : 0;
1783	dev->MEAR_cache |= MEAR_MDC;
1784	writel(dev->MEAR_cache, dev->base + MEAR);
1785
1786	/* Wait again... */
1787	udelay(1);
1788
1789	return bit;
1790}
1791
1792static unsigned ns83820_mii_read_reg(struct ns83820 *dev, unsigned phy, unsigned reg)
1793{
1794	unsigned data = 0;
1795	int i;
1796
1797	/* read some garbage so that we eventually sync up */
1798	for (i=0; i<64; i++)
1799		ns83820_mii_read_bit(dev);
1800
1801	ns83820_mii_write_bit(dev, 0);	/* start */
1802	ns83820_mii_write_bit(dev, 1);
1803	ns83820_mii_write_bit(dev, 1);	/* opcode read */
1804	ns83820_mii_write_bit(dev, 0);
1805
1806	/* write out the phy address: 5 bits, msb first */
1807	for (i=0; i<5; i++)
1808		ns83820_mii_write_bit(dev, phy & (0x10 >> i));
1809
1810	/* write out the register address, 5 bits, msb first */
1811	for (i=0; i<5; i++)
1812		ns83820_mii_write_bit(dev, reg & (0x10 >> i));
1813
1814	ns83820_mii_read_bit(dev);	/* turn around cycles */
1815	ns83820_mii_read_bit(dev);
1816
1817	/* read in the register data, 16 bits msb first */
1818	for (i=0; i<16; i++) {
1819		data <<= 1;
1820		data |= ns83820_mii_read_bit(dev);
1821	}
1822
1823	return data;
1824}
1825
1826static unsigned ns83820_mii_write_reg(struct ns83820 *dev, unsigned phy, unsigned reg, unsigned data)
1827{
1828	int i;
1829
1830	/* read some garbage so that we eventually sync up */
1831	for (i=0; i<64; i++)
1832		ns83820_mii_read_bit(dev);
1833
1834	ns83820_mii_write_bit(dev, 0);	/* start */
1835	ns83820_mii_write_bit(dev, 1);
1836	ns83820_mii_write_bit(dev, 0);	/* opcode read */
1837	ns83820_mii_write_bit(dev, 1);
1838
1839	/* write out the phy address: 5 bits, msb first */
1840	for (i=0; i<5; i++)
1841		ns83820_mii_write_bit(dev, phy & (0x10 >> i));
1842
1843	/* write out the register address, 5 bits, msb first */
1844	for (i=0; i<5; i++)
1845		ns83820_mii_write_bit(dev, reg & (0x10 >> i));
1846
1847	ns83820_mii_read_bit(dev);	/* turn around cycles */
1848	ns83820_mii_read_bit(dev);
1849
1850	/* read in the register data, 16 bits msb first */
1851	for (i=0; i<16; i++)
1852		ns83820_mii_write_bit(dev, (data >> (15 - i)) & 1);
1853
1854	return data;
1855}
1856
1857static void ns83820_probe_phy(struct net_device *ndev)
1858{
1859	struct ns83820 *dev = PRIV(ndev);
1860	int j;
1861	unsigned a, b;
1862
1863	for (j = 0; j < 0x16; j += 4) {
1864		dprintk("%s: [0x%02x] %04x %04x %04x %04x\n",
1865			ndev->name, j,
1866			ns83820_mii_read_reg(dev, 1, 0 + j),
1867			ns83820_mii_read_reg(dev, 1, 1 + j),
1868			ns83820_mii_read_reg(dev, 1, 2 + j),
1869			ns83820_mii_read_reg(dev, 1, 3 + j)
1870			);
1871	}
1872
1873	/* read firmware version: memory addr is 0x8402 and 0x8403 */
1874	ns83820_mii_write_reg(dev, 1, 0x16, 0x000d);
1875	ns83820_mii_write_reg(dev, 1, 0x1e, 0x810e);
1876	a = ns83820_mii_read_reg(dev, 1, 0x1d);
1877
1878	ns83820_mii_write_reg(dev, 1, 0x16, 0x000d);
1879	ns83820_mii_write_reg(dev, 1, 0x1e, 0x810e);
1880	b = ns83820_mii_read_reg(dev, 1, 0x1d);
1881	dprintk("version: 0x%04x 0x%04x\n", a, b);
1882}
1883#endif
1884
1885static const struct net_device_ops netdev_ops = {
1886	.ndo_open		= ns83820_open,
1887	.ndo_stop		= ns83820_stop,
1888	.ndo_start_xmit		= ns83820_hard_start_xmit,
1889	.ndo_get_stats		= ns83820_get_stats,
1890	.ndo_set_rx_mode	= ns83820_set_multicast,
1891	.ndo_validate_addr	= eth_validate_addr,
1892	.ndo_set_mac_address	= eth_mac_addr,
1893	.ndo_tx_timeout		= ns83820_tx_timeout,
1894};
1895
1896static int ns83820_init_one(struct pci_dev *pci_dev,
1897			    const struct pci_device_id *id)
1898{
1899	struct net_device *ndev;
1900	struct ns83820 *dev;
1901	long addr;
1902	int err;
1903	int using_dac = 0;
1904
1905	/* See if we can set the dma mask early on; failure is fatal. */
1906	if (sizeof(dma_addr_t) == 8 &&
1907		!dma_set_mask(&pci_dev->dev, DMA_BIT_MASK(64))) {
1908		using_dac = 1;
1909	} else if (!dma_set_mask(&pci_dev->dev, DMA_BIT_MASK(32))) {
1910		using_dac = 0;
1911	} else {
1912		dev_warn(&pci_dev->dev, "dma_set_mask failed!\n");
1913		return -ENODEV;
1914	}
1915
1916	ndev = alloc_etherdev(sizeof(struct ns83820));
1917	err = -ENOMEM;
1918	if (!ndev)
1919		goto out;
1920
1921	dev = PRIV(ndev);
1922	dev->ndev = ndev;
1923
1924	spin_lock_init(&dev->rx_info.lock);
1925	spin_lock_init(&dev->tx_lock);
1926	spin_lock_init(&dev->misc_lock);
1927	dev->pci_dev = pci_dev;
1928
1929	SET_NETDEV_DEV(ndev, &pci_dev->dev);
1930
1931	INIT_WORK(&dev->tq_refill, queue_refill);
1932	tasklet_setup(&dev->rx_tasklet, rx_action);
1933
1934	err = pci_enable_device(pci_dev);
1935	if (err) {
1936		dev_info(&pci_dev->dev, "pci_enable_dev failed: %d\n", err);
1937		goto out_free;
1938	}
1939
1940	pci_set_master(pci_dev);
1941	addr = pci_resource_start(pci_dev, 1);
1942	dev->base = ioremap(addr, PAGE_SIZE);
1943	dev->tx_descs = dma_alloc_coherent(&pci_dev->dev,
1944					   4 * DESC_SIZE * NR_TX_DESC,
1945					   &dev->tx_phy_descs, GFP_KERNEL);
1946	dev->rx_info.descs = dma_alloc_coherent(&pci_dev->dev,
1947						4 * DESC_SIZE * NR_RX_DESC,
1948						&dev->rx_info.phy_descs, GFP_KERNEL);
1949	err = -ENOMEM;
1950	if (!dev->base || !dev->tx_descs || !dev->rx_info.descs)
1951		goto out_disable;
1952
1953	dprintk("%p: %08lx  %p: %08lx\n",
1954		dev->tx_descs, (long)dev->tx_phy_descs,
1955		dev->rx_info.descs, (long)dev->rx_info.phy_descs);
1956
1957	ns83820_disable_interrupts(dev);
1958
1959	dev->IMR_cache = 0;
1960
1961	err = request_irq(pci_dev->irq, ns83820_irq, IRQF_SHARED,
1962			  DRV_NAME, ndev);
1963	if (err) {
1964		dev_info(&pci_dev->dev, "unable to register irq %d, err %d\n",
1965			pci_dev->irq, err);
1966		goto out_disable;
1967	}
1968
1969	/*
1970	 * FIXME: we are holding rtnl_lock() over obscenely long area only
1971	 * because some of the setup code uses dev->name.  It's Wrong(tm) -
1972	 * we should be using driver-specific names for all that stuff.
1973	 * For now that will do, but we really need to come back and kill
1974	 * most of the dev_alloc_name() users later.
1975	 */
1976	rtnl_lock();
1977	err = dev_alloc_name(ndev, ndev->name);
1978	if (err < 0) {
1979		dev_info(&pci_dev->dev, "unable to get netdev name: %d\n", err);
1980		goto out_free_irq;
1981	}
1982
1983	printk("%s: ns83820.c: 0x22c: %08x, subsystem: %04x:%04x\n",
1984		ndev->name, le32_to_cpu(readl(dev->base + 0x22c)),
1985		pci_dev->subsystem_vendor, pci_dev->subsystem_device);
1986
1987	ndev->netdev_ops = &netdev_ops;
1988	ndev->ethtool_ops = &ops;
1989	ndev->watchdog_timeo = 5 * HZ;
1990	pci_set_drvdata(pci_dev, ndev);
1991
1992	ns83820_do_reset(dev, CR_RST);
1993
1994	/* Must reset the ram bist before running it */
1995	writel(PTSCR_RBIST_RST, dev->base + PTSCR);
1996	ns83820_run_bist(ndev, "sram bist",   PTSCR_RBIST_EN,
1997			 PTSCR_RBIST_DONE, PTSCR_RBIST_FAIL);
1998	ns83820_run_bist(ndev, "eeprom bist", PTSCR_EEBIST_EN, 0,
1999			 PTSCR_EEBIST_FAIL);
2000	ns83820_run_bist(ndev, "eeprom load", PTSCR_EELOAD_EN, 0, 0);
2001
2002	/* I love config registers */
2003	dev->CFG_cache = readl(dev->base + CFG);
2004
2005	if ((dev->CFG_cache & CFG_PCI64_DET)) {
2006		printk(KERN_INFO "%s: detected 64 bit PCI data bus.\n",
2007			ndev->name);
2008		/*dev->CFG_cache |= CFG_DATA64_EN;*/
2009		if (!(dev->CFG_cache & CFG_DATA64_EN))
2010			printk(KERN_INFO "%s: EEPROM did not enable 64 bit bus.  Disabled.\n",
2011				ndev->name);
2012	} else
2013		dev->CFG_cache &= ~(CFG_DATA64_EN);
2014
2015	dev->CFG_cache &= (CFG_TBI_EN  | CFG_MRM_DIS   | CFG_MWI_DIS |
2016			   CFG_T64ADDR | CFG_DATA64_EN | CFG_EXT_125 |
2017			   CFG_M64ADDR);
2018	dev->CFG_cache |= CFG_PINT_DUPSTS | CFG_PINT_LNKSTS | CFG_PINT_SPDSTS |
2019			  CFG_EXTSTS_EN   | CFG_EXD         | CFG_PESEL;
2020	dev->CFG_cache |= CFG_REQALG;
2021	dev->CFG_cache |= CFG_POW;
2022	dev->CFG_cache |= CFG_TMRTEST;
2023
2024	/* When compiled with 64 bit addressing, we must always enable
2025	 * the 64 bit descriptor format.
2026	 */
2027	if (sizeof(dma_addr_t) == 8)
2028		dev->CFG_cache |= CFG_M64ADDR;
2029	if (using_dac)
2030		dev->CFG_cache |= CFG_T64ADDR;
2031
2032	/* Big endian mode does not seem to do what the docs suggest */
2033	dev->CFG_cache &= ~CFG_BEM;
2034
2035	/* setup optical transceiver if we have one */
2036	if (dev->CFG_cache & CFG_TBI_EN) {
2037		printk(KERN_INFO "%s: enabling optical transceiver\n",
2038			ndev->name);
2039		writel(readl(dev->base + GPIOR) | 0x3e8, dev->base + GPIOR);
2040
2041		/* setup auto negotiation feature advertisement */
2042		writel(readl(dev->base + TANAR)
2043		       | TANAR_HALF_DUP | TANAR_FULL_DUP,
2044		       dev->base + TANAR);
2045
2046		/* start auto negotiation */
2047		writel(TBICR_MR_AN_ENABLE | TBICR_MR_RESTART_AN,
2048		       dev->base + TBICR);
2049		writel(TBICR_MR_AN_ENABLE, dev->base + TBICR);
2050		dev->linkstate = LINK_AUTONEGOTIATE;
2051
2052		dev->CFG_cache |= CFG_MODE_1000;
2053	}
2054
2055	writel(dev->CFG_cache, dev->base + CFG);
2056	dprintk("CFG: %08x\n", dev->CFG_cache);
2057
2058	if (reset_phy) {
2059		printk(KERN_INFO "%s: resetting phy\n", ndev->name);
2060		writel(dev->CFG_cache | CFG_PHY_RST, dev->base + CFG);
2061		msleep(10);
2062		writel(dev->CFG_cache, dev->base + CFG);
2063	}
2064
2065#if 0	/* Huh?  This sets the PCI latency register.  Should be done via
2066	 * the PCI layer.  FIXME.
2067	 */
2068	if (readl(dev->base + SRR))
2069		writel(readl(dev->base+0x20c) | 0xfe00, dev->base + 0x20c);
2070#endif
2071
2072	/* Note!  The DMA burst size interacts with packet
2073	 * transmission, such that the largest packet that
2074	 * can be transmitted is 8192 - FLTH - burst size.
2075	 * If only the transmit fifo was larger...
2076	 */
2077	/* Ramit : 1024 DMA is not a good idea, it ends up banging
2078	 * some DELL and COMPAQ SMP systems */
2079	writel(TXCFG_CSI | TXCFG_HBI | TXCFG_ATP | TXCFG_MXDMA512
2080		| ((1600 / 32) * 0x100),
2081		dev->base + TXCFG);
2082
2083	/* Flush the interrupt holdoff timer */
2084	writel(0x000, dev->base + IHR);
2085	writel(0x100, dev->base + IHR);
2086	writel(0x000, dev->base + IHR);
2087
2088	/* Set Rx to full duplex, don't accept runt, errored, long or length
2089	 * range errored packets.  Use 512 byte DMA.
2090	 */
2091	/* Ramit : 1024 DMA is not a good idea, it ends up banging
2092	 * some DELL and COMPAQ SMP systems
2093	 * Turn on ALP, only we are accpeting Jumbo Packets */
2094	writel(RXCFG_AEP | RXCFG_ARP | RXCFG_AIRL | RXCFG_RX_FD
2095		| RXCFG_STRIPCRC
2096		//| RXCFG_ALP
2097		| (RXCFG_MXDMA512) | 0, dev->base + RXCFG);
2098
2099	/* Disable priority queueing */
2100	writel(0, dev->base + PQCR);
2101
2102	/* Enable IP checksum validation and detetion of VLAN headers.
2103	 * Note: do not set the reject options as at least the 0x102
2104	 * revision of the chip does not properly accept IP fragments
2105	 * at least for UDP.
2106	 */
2107	/* Ramit : Be sure to turn on RXCFG_ARP if VLAN's are enabled, since
2108	 * the MAC it calculates the packetsize AFTER stripping the VLAN
2109	 * header, and if a VLAN Tagged packet of 64 bytes is received (like
2110	 * a ping with a VLAN header) then the card, strips the 4 byte VLAN
2111	 * tag and then checks the packet size, so if RXCFG_ARP is not enabled,
2112	 * it discrards it!.  These guys......
2113	 * also turn on tag stripping if hardware acceleration is enabled
2114	 */
2115#ifdef NS83820_VLAN_ACCEL_SUPPORT
2116#define VRCR_INIT_VALUE (VRCR_IPEN|VRCR_VTDEN|VRCR_VTREN)
2117#else
2118#define VRCR_INIT_VALUE (VRCR_IPEN|VRCR_VTDEN)
2119#endif
2120	writel(VRCR_INIT_VALUE, dev->base + VRCR);
2121
2122	/* Enable per-packet TCP/UDP/IP checksumming
2123	 * and per packet vlan tag insertion if
2124	 * vlan hardware acceleration is enabled
2125	 */
2126#ifdef NS83820_VLAN_ACCEL_SUPPORT
2127#define VTCR_INIT_VALUE (VTCR_PPCHK|VTCR_VPPTI)
2128#else
2129#define VTCR_INIT_VALUE VTCR_PPCHK
2130#endif
2131	writel(VTCR_INIT_VALUE, dev->base + VTCR);
2132
2133	/* Ramit : Enable async and sync pause frames */
2134	/* writel(0, dev->base + PCR); */
2135	writel((PCR_PS_MCAST | PCR_PS_DA | PCR_PSEN | PCR_FFLO_4K |
2136		PCR_FFHI_8K | PCR_STLO_4 | PCR_STHI_8 | PCR_PAUSE_CNT),
2137		dev->base + PCR);
2138
2139	/* Disable Wake On Lan */
2140	writel(0, dev->base + WCSR);
2141
2142	ns83820_getmac(dev, ndev);
2143
2144	/* Yes, we support dumb IP checksum on transmit */
2145	ndev->features |= NETIF_F_SG;
2146	ndev->features |= NETIF_F_IP_CSUM;
2147
2148	ndev->min_mtu = 0;
2149
2150#ifdef NS83820_VLAN_ACCEL_SUPPORT
2151	/* We also support hardware vlan acceleration */
2152	ndev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2153#endif
2154
2155	if (using_dac) {
2156		printk(KERN_INFO "%s: using 64 bit addressing.\n",
2157			ndev->name);
2158		ndev->features |= NETIF_F_HIGHDMA;
2159	}
2160
2161	printk(KERN_INFO "%s: ns83820 v" VERSION ": DP83820 v%u.%u: %pM io=0x%08lx irq=%d f=%s\n",
2162		ndev->name,
2163		(unsigned)readl(dev->base + SRR) >> 8,
2164		(unsigned)readl(dev->base + SRR) & 0xff,
2165		ndev->dev_addr, addr, pci_dev->irq,
2166		(ndev->features & NETIF_F_HIGHDMA) ? "h,sg" : "sg"
2167		);
2168
2169#ifdef PHY_CODE_IS_FINISHED
2170	ns83820_probe_phy(ndev);
2171#endif
2172
2173	err = register_netdevice(ndev);
2174	if (err) {
2175		printk(KERN_INFO "ns83820: unable to register netdev: %d\n", err);
2176		goto out_cleanup;
2177	}
2178	rtnl_unlock();
2179
2180	return 0;
2181
2182out_cleanup:
2183	ns83820_disable_interrupts(dev); /* paranoia */
2184out_free_irq:
2185	rtnl_unlock();
2186	free_irq(pci_dev->irq, ndev);
2187out_disable:
2188	if (dev->base)
2189		iounmap(dev->base);
2190	dma_free_coherent(&pci_dev->dev, 4 * DESC_SIZE * NR_TX_DESC,
2191			  dev->tx_descs, dev->tx_phy_descs);
2192	dma_free_coherent(&pci_dev->dev, 4 * DESC_SIZE * NR_RX_DESC,
2193			  dev->rx_info.descs, dev->rx_info.phy_descs);
2194	pci_disable_device(pci_dev);
2195out_free:
2196	free_netdev(ndev);
2197out:
2198	return err;
2199}
2200
2201static void ns83820_remove_one(struct pci_dev *pci_dev)
2202{
2203	struct net_device *ndev = pci_get_drvdata(pci_dev);
2204	struct ns83820 *dev = PRIV(ndev); /* ok even if NULL */
2205
2206	if (!ndev)			/* paranoia */
2207		return;
2208
2209	ns83820_disable_interrupts(dev); /* paranoia */
2210
2211	unregister_netdev(ndev);
2212	free_irq(dev->pci_dev->irq, ndev);
2213	iounmap(dev->base);
2214	dma_free_coherent(&dev->pci_dev->dev, 4 * DESC_SIZE * NR_TX_DESC,
2215			  dev->tx_descs, dev->tx_phy_descs);
2216	dma_free_coherent(&dev->pci_dev->dev, 4 * DESC_SIZE * NR_RX_DESC,
2217			  dev->rx_info.descs, dev->rx_info.phy_descs);
2218	pci_disable_device(dev->pci_dev);
2219	free_netdev(ndev);
2220}
2221
2222static const struct pci_device_id ns83820_pci_tbl[] = {
2223	{ 0x100b, 0x0022, PCI_ANY_ID, PCI_ANY_ID, 0, .driver_data = 0, },
2224	{ 0, },
2225};
2226
2227static struct pci_driver driver = {
2228	.name		= "ns83820",
2229	.id_table	= ns83820_pci_tbl,
2230	.probe		= ns83820_init_one,
2231	.remove		= ns83820_remove_one,
2232#if 0	/* FIXME: implement */
2233	.suspend	= ,
2234	.resume		= ,
2235#endif
2236};
2237
2238
2239static int __init ns83820_init(void)
2240{
2241	printk(KERN_INFO "ns83820.c: National Semiconductor DP83820 10/100/1000 driver.\n");
2242	return pci_register_driver(&driver);
2243}
2244
2245static void __exit ns83820_exit(void)
2246{
2247	pci_unregister_driver(&driver);
2248}
2249
2250MODULE_AUTHOR("Benjamin LaHaise <bcrl@kvack.org>");
2251MODULE_DESCRIPTION("National Semiconductor DP83820 10/100/1000 driver");
2252MODULE_LICENSE("GPL");
2253
2254MODULE_DEVICE_TABLE(pci, ns83820_pci_tbl);
2255
2256module_param(lnksts, int, 0);
2257MODULE_PARM_DESC(lnksts, "Polarity of LNKSTS bit");
2258
2259module_param(ihr, int, 0);
2260MODULE_PARM_DESC(ihr, "Time in 100 us increments to delay interrupts (range 0-127)");
2261
2262module_param(reset_phy, int, 0);
2263MODULE_PARM_DESC(reset_phy, "Set to 1 to reset the PHY on startup");
2264
2265module_init(ns83820_init);
2266module_exit(ns83820_exit);