Root/drivers/net/hamachi.c

1/* hamachi.c: A Packet Engines GNIC-II Gigabit Ethernet driver for Linux. */
2/*
3    Written 1998-2000 by Donald Becker.
4    Updates 2000 by Keith Underwood.
5
6    This software may be used and distributed according to the terms of
7    the GNU General Public License (GPL), incorporated herein by reference.
8    Drivers based on or derived from this code fall under the GPL and must
9    retain the authorship, copyright and license notice. This file is not
10    a complete program and may only be used when the entire operating
11    system is licensed under the GPL.
12
13    The author may be reached as becker@scyld.com, or C/O
14    Scyld Computing Corporation
15    410 Severn Ave., Suite 210
16    Annapolis MD 21403
17
18    This driver is for the Packet Engines GNIC-II PCI Gigabit Ethernet
19    adapter.
20
21    Support and updates available at
22    http://www.scyld.com/network/hamachi.html
23    [link no longer provides useful info -jgarzik]
24    or
25    http://www.parl.clemson.edu/~keithu/hamachi.html
26
27*/
28
29#define DRV_NAME "hamachi"
30#define DRV_VERSION "2.1"
31#define DRV_RELDATE "Sept 11, 2006"
32
33
34/* A few user-configurable values. */
35
36static int debug = 1; /* 1 normal messages, 0 quiet .. 7 verbose. */
37#define final_version
38#define hamachi_debug debug
39/* Maximum events (Rx packets, etc.) to handle at each interrupt. */
40static int max_interrupt_work = 40;
41static int mtu;
42/* Default values selected by testing on a dual processor PIII-450 */
43/* These six interrupt control parameters may be set directly when loading the
44 * module, or through the rx_params and tx_params variables
45 */
46static int max_rx_latency = 0x11;
47static int max_rx_gap = 0x05;
48static int min_rx_pkt = 0x18;
49static int max_tx_latency = 0x00;
50static int max_tx_gap = 0x00;
51static int min_tx_pkt = 0x30;
52
53/* Set the copy breakpoint for the copy-only-tiny-frames scheme.
54   -Setting to > 1518 causes all frames to be copied
55    -Setting to 0 disables copies
56*/
57static int rx_copybreak;
58
59/* An override for the hardware detection of bus width.
60    Set to 1 to force 32 bit PCI bus detection. Set to 4 to force 64 bit.
61    Add 2 to disable parity detection.
62*/
63static int force32;
64
65
66/* Used to pass the media type, etc.
67   These exist for driver interoperability.
68   No media types are currently defined.
69        - The lower 4 bits are reserved for the media type.
70        - The next three bits may be set to one of the following:
71            0x00000000 : Autodetect PCI bus
72            0x00000010 : Force 32 bit PCI bus
73            0x00000020 : Disable parity detection
74            0x00000040 : Force 64 bit PCI bus
75            Default is autodetect
76        - The next bit can be used to force half-duplex. This is a bad
77          idea since no known implementations implement half-duplex, and,
78          in general, half-duplex for gigabit ethernet is a bad idea.
79            0x00000080 : Force half-duplex
80            Default is full-duplex.
81        - In the original driver, the ninth bit could be used to force
82          full-duplex. Maintain that for compatibility
83           0x00000200 : Force full-duplex
84*/
85#define MAX_UNITS 8 /* More are supported, limit only on options */
86static int options[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1};
87static int full_duplex[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1};
88/* The Hamachi chipset supports 3 parameters each for Rx and Tx
89 * interruput management. Parameters will be loaded as specified into
90 * the TxIntControl and RxIntControl registers.
91 *
92 * The registers are arranged as follows:
93 * 23 - 16 15 - 8 7 - 0
94 * _________________________________
95 * | min_pkt | max_gap | max_latency |
96 * ---------------------------------
97 * min_pkt : The minimum number of packets processed between
98 * interrupts.
99 * max_gap : The maximum inter-packet gap in units of 8.192 us
100 * max_latency : The absolute time between interrupts in units of 8.192 us
101 *
102 */
103static int rx_params[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1};
104static int tx_params[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1};
105
106/* Operational parameters that are set at compile time. */
107
108/* Keep the ring sizes a power of two for compile efficiency.
109    The compiler will convert <unsigned>'%'<2^N> into a bit mask.
110   Making the Tx ring too large decreases the effectiveness of channel
111   bonding and packet priority.
112   There are no ill effects from too-large receive rings, except for
113    excessive memory usage */
114/* Empirically it appears that the Tx ring needs to be a little bigger
115   for these Gbit adapters or you get into an overrun condition really
116   easily. Also, things appear to work a bit better in back-to-back
117   configurations if the Rx ring is 8 times the size of the Tx ring
118*/
119#define TX_RING_SIZE 64
120#define RX_RING_SIZE 512
121#define TX_TOTAL_SIZE TX_RING_SIZE*sizeof(struct hamachi_desc)
122#define RX_TOTAL_SIZE RX_RING_SIZE*sizeof(struct hamachi_desc)
123
124/*
125 * Enable netdev_ioctl. Added interrupt coalescing parameter adjustment.
126 * 2/19/99 Pete Wyckoff <wyckoff@ca.sandia.gov>
127 */
128
129/* play with 64-bit addrlen; seems to be a teensy bit slower --pw */
130/* #define ADDRLEN 64 */
131
132/*
133 * RX_CHECKSUM turns on card-generated receive checksum generation for
134 * TCP and UDP packets. Otherwise the upper layers do the calculation.
135 * TX_CHECKSUM won't do anything too useful, even if it works. There's no
136 * easy mechanism by which to tell the TCP/UDP stack that it need not
137 * generate checksums for this device. But if somebody can find a way
138 * to get that to work, most of the card work is in here already.
139 * 3/10/1999 Pete Wyckoff <wyckoff@ca.sandia.gov>
140 */
141#undef TX_CHECKSUM
142#define RX_CHECKSUM
143
144/* Operational parameters that usually are not changed. */
145/* Time in jiffies before concluding the transmitter is hung. */
146#define TX_TIMEOUT (5*HZ)
147
148#include <linux/module.h>
149#include <linux/kernel.h>
150#include <linux/string.h>
151#include <linux/timer.h>
152#include <linux/time.h>
153#include <linux/errno.h>
154#include <linux/ioport.h>
155#include <linux/slab.h>
156#include <linux/interrupt.h>
157#include <linux/pci.h>
158#include <linux/init.h>
159#include <linux/ethtool.h>
160#include <linux/mii.h>
161#include <linux/netdevice.h>
162#include <linux/etherdevice.h>
163#include <linux/skbuff.h>
164#include <linux/ip.h>
165#include <linux/delay.h>
166#include <linux/bitops.h>
167
168#include <asm/uaccess.h>
169#include <asm/processor.h> /* Processor type for cache alignment. */
170#include <asm/io.h>
171#include <asm/unaligned.h>
172#include <asm/cache.h>
173
174static const char version[] __devinitconst =
175KERN_INFO DRV_NAME ".c:v" DRV_VERSION " " DRV_RELDATE " Written by Donald Becker\n"
176" Some modifications by Eric kasten <kasten@nscl.msu.edu>\n"
177" Further modifications by Keith Underwood <keithu@parl.clemson.edu>\n";
178
179
180/* IP_MF appears to be only defined in <netinet/ip.h>, however,
181   we need it for hardware checksumming support. FYI... some of
182   the definitions in <netinet/ip.h> conflict/duplicate those in
183   other linux headers causing many compiler warnings.
184*/
185#ifndef IP_MF
186  #define IP_MF 0x2000 /* IP more frags from <netinet/ip.h> */
187#endif
188
189/* Define IP_OFFSET to be IPOPT_OFFSET */
190#ifndef IP_OFFSET
191  #ifdef IPOPT_OFFSET
192    #define IP_OFFSET IPOPT_OFFSET
193  #else
194    #define IP_OFFSET 2
195  #endif
196#endif
197
198#define RUN_AT(x) (jiffies + (x))
199
200#ifndef ADDRLEN
201#define ADDRLEN 32
202#endif
203
204/* Condensed bus+endian portability operations. */
205#if ADDRLEN == 64
206#define cpu_to_leXX(addr) cpu_to_le64(addr)
207#define leXX_to_cpu(addr) le64_to_cpu(addr)
208#else
209#define cpu_to_leXX(addr) cpu_to_le32(addr)
210#define leXX_to_cpu(addr) le32_to_cpu(addr)
211#endif
212
213
214/*
215                Theory of Operation
216
217I. Board Compatibility
218
219This device driver is designed for the Packet Engines "Hamachi"
220Gigabit Ethernet chip. The only PCA currently supported is the GNIC-II 64-bit
22166Mhz PCI card.
222
223II. Board-specific settings
224
225No jumpers exist on the board. The chip supports software correction of
226various motherboard wiring errors, however this driver does not support
227that feature.
228
229III. Driver operation
230
231IIIa. Ring buffers
232
233The Hamachi uses a typical descriptor based bus-master architecture.
234The descriptor list is similar to that used by the Digital Tulip.
235This driver uses two statically allocated fixed-size descriptor lists
236formed into rings by a branch from the final descriptor to the beginning of
237the list. The ring sizes are set at compile time by RX/TX_RING_SIZE.
238
239This driver uses a zero-copy receive and transmit scheme similar my other
240network drivers.
241The driver allocates full frame size skbuffs for the Rx ring buffers at
242open() time and passes the skb->data field to the Hamachi as receive data
243buffers. When an incoming frame is less than RX_COPYBREAK bytes long,
244a fresh skbuff is allocated and the frame is copied to the new skbuff.
245When the incoming frame is larger, the skbuff is passed directly up the
246protocol stack and replaced by a newly allocated skbuff.
247
248The RX_COPYBREAK value is chosen to trade-off the memory wasted by
249using a full-sized skbuff for small frames vs. the copying costs of larger
250frames. Gigabit cards are typically used on generously configured machines
251and the underfilled buffers have negligible impact compared to the benefit of
252a single allocation size, so the default value of zero results in never
253copying packets.
254
255IIIb/c. Transmit/Receive Structure
256
257The Rx and Tx descriptor structure are straight-forward, with no historical
258baggage that must be explained. Unlike the awkward DBDMA structure, there
259are no unused fields or option bits that had only one allowable setting.
260
261Two details should be noted about the descriptors: The chip supports both 32
262bit and 64 bit address structures, and the length field is overwritten on
263the receive descriptors. The descriptor length is set in the control word
264for each channel. The development driver uses 32 bit addresses only, however
26564 bit addresses may be enabled for 64 bit architectures e.g. the Alpha.
266
267IIId. Synchronization
268
269This driver is very similar to my other network drivers.
270The driver runs as two independent, single-threaded flows of control. One
271is the send-packet routine, which enforces single-threaded use by the
272dev->tbusy flag. The other thread is the interrupt handler, which is single
273threaded by the hardware and other software.
274
275The send packet thread has partial control over the Tx ring and 'dev->tbusy'
276flag. It sets the tbusy flag whenever it's queuing a Tx packet. If the next
277queue slot is empty, it clears the tbusy flag when finished otherwise it sets
278the 'hmp->tx_full' flag.
279
280The interrupt handler has exclusive control over the Rx ring and records stats
281from the Tx ring. After reaping the stats, it marks the Tx queue entry as
282empty by incrementing the dirty_tx mark. Iff the 'hmp->tx_full' flag is set, it
283clears both the tx_full and tbusy flags.
284
285IV. Notes
286
287Thanks to Kim Stearns of Packet Engines for providing a pair of GNIC-II boards.
288
289IVb. References
290
291Hamachi Engineering Design Specification, 5/15/97
292(Note: This version was marked "Confidential".)
293
294IVc. Errata
295
296None noted.
297
298V. Recent Changes
299
30001/15/1999 EPK Enlargement of the TX and RX ring sizes. This appears
301    to help avoid some stall conditions -- this needs further research.
302
30301/15/1999 EPK Creation of the hamachi_tx function. This function cleans
304    the Tx ring and is called from hamachi_start_xmit (this used to be
305    called from hamachi_interrupt but it tends to delay execution of the
306    interrupt handler and thus reduce bandwidth by reducing the latency
307    between hamachi_rx()'s). Notably, some modification has been made so
308    that the cleaning loop checks only to make sure that the DescOwn bit
309    isn't set in the status flag since the card is not required
310    to set the entire flag to zero after processing.
311
31201/15/1999 EPK In the hamachi_start_tx function, the Tx ring full flag is
313    checked before attempting to add a buffer to the ring. If the ring is full
314    an attempt is made to free any dirty buffers and thus find space for
315    the new buffer or the function returns non-zero which should case the
316    scheduler to reschedule the buffer later.
317
31801/15/1999 EPK Some adjustments were made to the chip initialization.
319    End-to-end flow control should now be fully active and the interrupt
320    algorithm vars have been changed. These could probably use further tuning.
321
32201/15/1999 EPK Added the max_{rx,tx}_latency options. These are used to
323    set the rx and tx latencies for the Hamachi interrupts. If you're having
324    problems with network stalls, try setting these to higher values.
325    Valid values are 0x00 through 0xff.
326
32701/15/1999 EPK In general, the overall bandwidth has increased and
328    latencies are better (sometimes by a factor of 2). Stalls are rare at
329    this point, however there still appears to be a bug somewhere between the
330    hardware and driver. TCP checksum errors under load also appear to be
331    eliminated at this point.
332
33301/18/1999 EPK Ensured that the DescEndRing bit was being set on both the
334    Rx and Tx rings. This appears to have been affecting whether a particular
335    peer-to-peer connection would hang under high load. I believe the Rx
336    rings was typically getting set correctly, but the Tx ring wasn't getting
337    the DescEndRing bit set during initialization. ??? Does this mean the
338    hamachi card is using the DescEndRing in processing even if a particular
339    slot isn't in use -- hypothetically, the card might be searching the
340    entire Tx ring for slots with the DescOwn bit set and then processing
341    them. If the DescEndRing bit isn't set, then it might just wander off
342    through memory until it hits a chunk of data with that bit set
343    and then looping back.
344
34502/09/1999 EPK Added Michel Mueller's TxDMA Interrupt and Tx-timeout
346    problem (TxCmd and RxCmd need only to be set when idle or stopped.
347
34802/09/1999 EPK Added code to check/reset dev->tbusy in hamachi_interrupt.
349    (Michel Mueller pointed out the ``permanently busy'' potential
350    problem here).
351
35202/22/1999 EPK Added Pete Wyckoff's ioctl to control the Tx/Rx latencies.
353
35402/23/1999 EPK Verified that the interrupt status field bits for Tx were
355    incorrectly defined and corrected (as per Michel Mueller).
356
35702/23/1999 EPK Corrected the Tx full check to check that at least 4 slots
358    were available before reseting the tbusy and tx_full flags
359    (as per Michel Mueller).
360
36103/11/1999 EPK Added Pete Wyckoff's hardware checksumming support.
362
36312/31/1999 KDU Cleaned up assorted things and added Don's code to force
36432 bit.
365
36602/20/2000 KDU Some of the control was just plain odd. Cleaned up the
367hamachi_start_xmit() and hamachi_interrupt() code. There is still some
368re-structuring I would like to do.
369
37003/01/2000 KDU Experimenting with a WIDE range of interrupt mitigation
371parameters on a dual P3-450 setup yielded the new default interrupt
372mitigation parameters. Tx should interrupt VERY infrequently due to
373Eric's scheme. Rx should be more often...
374
37503/13/2000 KDU Added a patch to make the Rx Checksum code interact
376nicely with non-linux machines.
377
37803/13/2000 KDU Experimented with some of the configuration values:
379
380    -It seems that enabling PCI performance commands for descriptors
381    (changing RxDMACtrl and TxDMACtrl lower nibble from 5 to D) has minimal
382    performance impact for any of my tests. (ttcp, netpipe, netperf) I will
383    leave them that way until I hear further feedback.
384
385    -Increasing the PCI_LATENCY_TIMER to 130
386    (2 + (burst size of 128 * (0 wait states + 1))) seems to slightly
387    degrade performance. Leaving default at 64 pending further information.
388
38903/14/2000 KDU Further tuning:
390
391    -adjusted boguscnt in hamachi_rx() to depend on interrupt
392    mitigation parameters chosen.
393
394    -Selected a set of interrupt parameters based on some extensive testing.
395    These may change with more testing.
396
397TO DO:
398
399-Consider borrowing from the acenic driver code to check PCI_COMMAND for
400PCI_COMMAND_INVALIDATE. Set maximum burst size to cache line size in
401that case.
402
403-fix the reset procedure. It doesn't quite work.
404*/
405
406/* A few values that may be tweaked. */
407/* Size of each temporary Rx buffer, calculated as:
408 * 1518 bytes (ethernet packet) + 2 bytes (to get 8 byte alignment for
409 * the card) + 8 bytes of status info + 8 bytes for the Rx Checksum +
410 * 2 more because we use skb_reserve.
411 */
412#define PKT_BUF_SZ 1538
413
414/* For now, this is going to be set to the maximum size of an ethernet
415 * packet. Eventually, we may want to make it a variable that is
416 * related to the MTU
417 */
418#define MAX_FRAME_SIZE 1518
419
420/* The rest of these values should never change. */
421
422static void hamachi_timer(unsigned long data);
423
424enum capability_flags {CanHaveMII=1, };
425static const struct chip_info {
426    u16 vendor_id, device_id, device_id_mask, pad;
427    const char *name;
428    void (*media_timer)(unsigned long data);
429    int flags;
430} chip_tbl[] = {
431    {0x1318, 0x0911, 0xffff, 0, "Hamachi GNIC-II", hamachi_timer, 0},
432    {0,},
433};
434
435/* Offsets to the Hamachi registers. Various sizes. */
436enum hamachi_offsets {
437    TxDMACtrl=0x00, TxCmd=0x04, TxStatus=0x06, TxPtr=0x08, TxCurPtr=0x10,
438    RxDMACtrl=0x20, RxCmd=0x24, RxStatus=0x26, RxPtr=0x28, RxCurPtr=0x30,
439    PCIClkMeas=0x060, MiscStatus=0x066, ChipRev=0x68, ChipReset=0x06B,
440    LEDCtrl=0x06C, VirtualJumpers=0x06D, GPIO=0x6E,
441    TxChecksum=0x074, RxChecksum=0x076,
442    TxIntrCtrl=0x078, RxIntrCtrl=0x07C,
443    InterruptEnable=0x080, InterruptClear=0x084, IntrStatus=0x088,
444    EventStatus=0x08C,
445    MACCnfg=0x0A0, FrameGap0=0x0A2, FrameGap1=0x0A4,
446    /* See enum MII_offsets below. */
447    MACCnfg2=0x0B0, RxDepth=0x0B8, FlowCtrl=0x0BC, MaxFrameSize=0x0CE,
448    AddrMode=0x0D0, StationAddr=0x0D2,
449    /* Gigabit AutoNegotiation. */
450    ANCtrl=0x0E0, ANStatus=0x0E2, ANXchngCtrl=0x0E4, ANAdvertise=0x0E8,
451    ANLinkPartnerAbility=0x0EA,
452    EECmdStatus=0x0F0, EEData=0x0F1, EEAddr=0x0F2,
453    FIFOcfg=0x0F8,
454};
455
456/* Offsets to the MII-mode registers. */
457enum MII_offsets {
458    MII_Cmd=0xA6, MII_Addr=0xA8, MII_Wr_Data=0xAA, MII_Rd_Data=0xAC,
459    MII_Status=0xAE,
460};
461
462/* Bits in the interrupt status/mask registers. */
463enum intr_status_bits {
464    IntrRxDone=0x01, IntrRxPCIFault=0x02, IntrRxPCIErr=0x04,
465    IntrTxDone=0x100, IntrTxPCIFault=0x200, IntrTxPCIErr=0x400,
466    LinkChange=0x10000, NegotiationChange=0x20000, StatsMax=0x40000, };
467
468/* The Hamachi Rx and Tx buffer descriptors. */
469struct hamachi_desc {
470    __le32 status_n_length;
471#if ADDRLEN == 64
472    u32 pad;
473    __le64 addr;
474#else
475    __le32 addr;
476#endif
477};
478
479/* Bits in hamachi_desc.status_n_length */
480enum desc_status_bits {
481    DescOwn=0x80000000, DescEndPacket=0x40000000, DescEndRing=0x20000000,
482    DescIntr=0x10000000,
483};
484
485#define PRIV_ALIGN 15 /* Required alignment mask */
486#define MII_CNT 4
487struct hamachi_private {
488    /* Descriptor rings first for alignment. Tx requires a second descriptor
489       for status. */
490    struct hamachi_desc *rx_ring;
491    struct hamachi_desc *tx_ring;
492    struct sk_buff* rx_skbuff[RX_RING_SIZE];
493    struct sk_buff* tx_skbuff[TX_RING_SIZE];
494    dma_addr_t tx_ring_dma;
495    dma_addr_t rx_ring_dma;
496    struct net_device_stats stats;
497    struct timer_list timer; /* Media selection timer. */
498    /* Frequently used and paired value: keep adjacent for cache effect. */
499    spinlock_t lock;
500    int chip_id;
501    unsigned int cur_rx, dirty_rx; /* Producer/consumer ring indices */
502    unsigned int cur_tx, dirty_tx;
503    unsigned int rx_buf_sz; /* Based on MTU+slack. */
504    unsigned int tx_full:1; /* The Tx queue is full. */
505    unsigned int duplex_lock:1;
506    unsigned int default_port:4; /* Last dev->if_port value. */
507    /* MII transceiver section. */
508    int mii_cnt; /* MII device addresses. */
509    struct mii_if_info mii_if; /* MII lib hooks/info */
510    unsigned char phys[MII_CNT]; /* MII device addresses, only first one used. */
511    u32 rx_int_var, tx_int_var; /* interrupt control variables */
512    u32 option; /* Hold on to a copy of the options */
513    struct pci_dev *pci_dev;
514    void __iomem *base;
515};
516
517MODULE_AUTHOR("Donald Becker <becker@scyld.com>, Eric Kasten <kasten@nscl.msu.edu>, Keith Underwood <keithu@parl.clemson.edu>");
518MODULE_DESCRIPTION("Packet Engines 'Hamachi' GNIC-II Gigabit Ethernet driver");
519MODULE_LICENSE("GPL");
520
521module_param(max_interrupt_work, int, 0);
522module_param(mtu, int, 0);
523module_param(debug, int, 0);
524module_param(min_rx_pkt, int, 0);
525module_param(max_rx_gap, int, 0);
526module_param(max_rx_latency, int, 0);
527module_param(min_tx_pkt, int, 0);
528module_param(max_tx_gap, int, 0);
529module_param(max_tx_latency, int, 0);
530module_param(rx_copybreak, int, 0);
531module_param_array(rx_params, int, NULL, 0);
532module_param_array(tx_params, int, NULL, 0);
533module_param_array(options, int, NULL, 0);
534module_param_array(full_duplex, int, NULL, 0);
535module_param(force32, int, 0);
536MODULE_PARM_DESC(max_interrupt_work, "GNIC-II maximum events handled per interrupt");
537MODULE_PARM_DESC(mtu, "GNIC-II MTU (all boards)");
538MODULE_PARM_DESC(debug, "GNIC-II debug level (0-7)");
539MODULE_PARM_DESC(min_rx_pkt, "GNIC-II minimum Rx packets processed between interrupts");
540MODULE_PARM_DESC(max_rx_gap, "GNIC-II maximum Rx inter-packet gap in 8.192 microsecond units");
541MODULE_PARM_DESC(max_rx_latency, "GNIC-II time between Rx interrupts in 8.192 microsecond units");
542MODULE_PARM_DESC(min_tx_pkt, "GNIC-II minimum Tx packets processed between interrupts");
543MODULE_PARM_DESC(max_tx_gap, "GNIC-II maximum Tx inter-packet gap in 8.192 microsecond units");
544MODULE_PARM_DESC(max_tx_latency, "GNIC-II time between Tx interrupts in 8.192 microsecond units");
545MODULE_PARM_DESC(rx_copybreak, "GNIC-II copy breakpoint for copy-only-tiny-frames");
546MODULE_PARM_DESC(rx_params, "GNIC-II min_rx_pkt+max_rx_gap+max_rx_latency");
547MODULE_PARM_DESC(tx_params, "GNIC-II min_tx_pkt+max_tx_gap+max_tx_latency");
548MODULE_PARM_DESC(options, "GNIC-II Bits 0-3: media type, bits 4-6: as force32, bit 7: half duplex, bit 9 full duplex");
549MODULE_PARM_DESC(full_duplex, "GNIC-II full duplex setting(s) (1)");
550MODULE_PARM_DESC(force32, "GNIC-II: Bit 0: 32 bit PCI, bit 1: disable parity, bit 2: 64 bit PCI (all boards)");
551
552static int read_eeprom(void __iomem *ioaddr, int location);
553static int mdio_read(struct net_device *dev, int phy_id, int location);
554static void mdio_write(struct net_device *dev, int phy_id, int location, int value);
555static int hamachi_open(struct net_device *dev);
556static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd);
557static void hamachi_timer(unsigned long data);
558static void hamachi_tx_timeout(struct net_device *dev);
559static void hamachi_init_ring(struct net_device *dev);
560static int hamachi_start_xmit(struct sk_buff *skb, struct net_device *dev);
561static irqreturn_t hamachi_interrupt(int irq, void *dev_instance);
562static int hamachi_rx(struct net_device *dev);
563static inline int hamachi_tx(struct net_device *dev);
564static void hamachi_error(struct net_device *dev, int intr_status);
565static int hamachi_close(struct net_device *dev);
566static struct net_device_stats *hamachi_get_stats(struct net_device *dev);
567static void set_rx_mode(struct net_device *dev);
568static const struct ethtool_ops ethtool_ops;
569static const struct ethtool_ops ethtool_ops_no_mii;
570
571static const struct net_device_ops hamachi_netdev_ops = {
572    .ndo_open = hamachi_open,
573    .ndo_stop = hamachi_close,
574    .ndo_start_xmit = hamachi_start_xmit,
575    .ndo_get_stats = hamachi_get_stats,
576    .ndo_set_multicast_list = set_rx_mode,
577    .ndo_change_mtu = eth_change_mtu,
578    .ndo_validate_addr = eth_validate_addr,
579    .ndo_set_mac_address = eth_mac_addr,
580    .ndo_tx_timeout = hamachi_tx_timeout,
581    .ndo_do_ioctl = netdev_ioctl,
582};
583
584
585static int __devinit hamachi_init_one (struct pci_dev *pdev,
586                    const struct pci_device_id *ent)
587{
588    struct hamachi_private *hmp;
589    int option, i, rx_int_var, tx_int_var, boguscnt;
590    int chip_id = ent->driver_data;
591    int irq;
592    void __iomem *ioaddr;
593    unsigned long base;
594    static int card_idx;
595    struct net_device *dev;
596    void *ring_space;
597    dma_addr_t ring_dma;
598    int ret = -ENOMEM;
599
600/* when built into the kernel, we only print version if device is found */
601#ifndef MODULE
602    static int printed_version;
603    if (!printed_version++)
604        printk(version);
605#endif
606
607    if (pci_enable_device(pdev)) {
608        ret = -EIO;
609        goto err_out;
610    }
611
612    base = pci_resource_start(pdev, 0);
613#ifdef __alpha__ /* Really "64 bit addrs" */
614    base |= (pci_resource_start(pdev, 1) << 32);
615#endif
616
617    pci_set_master(pdev);
618
619    i = pci_request_regions(pdev, DRV_NAME);
620    if (i)
621        return i;
622
623    irq = pdev->irq;
624    ioaddr = ioremap(base, 0x400);
625    if (!ioaddr)
626        goto err_out_release;
627
628    dev = alloc_etherdev(sizeof(struct hamachi_private));
629    if (!dev)
630        goto err_out_iounmap;
631
632    SET_NETDEV_DEV(dev, &pdev->dev);
633
634#ifdef TX_CHECKSUM
635    printk("check that skbcopy in ip_queue_xmit isn't happening\n");
636    dev->hard_header_len += 8; /* for cksum tag */
637#endif
638
639    for (i = 0; i < 6; i++)
640        dev->dev_addr[i] = 1 ? read_eeprom(ioaddr, 4 + i)
641            : readb(ioaddr + StationAddr + i);
642
643#if ! defined(final_version)
644    if (hamachi_debug > 4)
645        for (i = 0; i < 0x10; i++)
646            printk("%2.2x%s",
647                   read_eeprom(ioaddr, i), i % 16 != 15 ? " " : "\n");
648#endif
649
650    hmp = netdev_priv(dev);
651    spin_lock_init(&hmp->lock);
652
653    hmp->mii_if.dev = dev;
654    hmp->mii_if.mdio_read = mdio_read;
655    hmp->mii_if.mdio_write = mdio_write;
656    hmp->mii_if.phy_id_mask = 0x1f;
657    hmp->mii_if.reg_num_mask = 0x1f;
658
659    ring_space = pci_alloc_consistent(pdev, TX_TOTAL_SIZE, &ring_dma);
660    if (!ring_space)
661        goto err_out_cleardev;
662    hmp->tx_ring = (struct hamachi_desc *)ring_space;
663    hmp->tx_ring_dma = ring_dma;
664
665    ring_space = pci_alloc_consistent(pdev, RX_TOTAL_SIZE, &ring_dma);
666    if (!ring_space)
667        goto err_out_unmap_tx;
668    hmp->rx_ring = (struct hamachi_desc *)ring_space;
669    hmp->rx_ring_dma = ring_dma;
670
671    /* Check for options being passed in */
672    option = card_idx < MAX_UNITS ? options[card_idx] : 0;
673    if (dev->mem_start)
674        option = dev->mem_start;
675
676    /* If the bus size is misidentified, do the following. */
677    force32 = force32 ? force32 :
678        ((option >= 0) ? ((option & 0x00000070) >> 4) : 0 );
679    if (force32)
680        writeb(force32, ioaddr + VirtualJumpers);
681
682    /* Hmmm, do we really need to reset the chip???. */
683    writeb(0x01, ioaddr + ChipReset);
684
685    /* After a reset, the clock speed measurement of the PCI bus will not
686     * be valid for a moment. Wait for a little while until it is. If
687     * it takes more than 10ms, forget it.
688     */
689    udelay(10);
690    i = readb(ioaddr + PCIClkMeas);
691    for (boguscnt = 0; (!(i & 0x080)) && boguscnt < 1000; boguscnt++){
692        udelay(10);
693        i = readb(ioaddr + PCIClkMeas);
694    }
695
696    hmp->base = ioaddr;
697    dev->base_addr = (unsigned long)ioaddr;
698    dev->irq = irq;
699    pci_set_drvdata(pdev, dev);
700
701    hmp->chip_id = chip_id;
702    hmp->pci_dev = pdev;
703
704    /* The lower four bits are the media type. */
705    if (option > 0) {
706        hmp->option = option;
707        if (option & 0x200)
708            hmp->mii_if.full_duplex = 1;
709        else if (option & 0x080)
710            hmp->mii_if.full_duplex = 0;
711        hmp->default_port = option & 15;
712        if (hmp->default_port)
713            hmp->mii_if.force_media = 1;
714    }
715    if (card_idx < MAX_UNITS && full_duplex[card_idx] > 0)
716        hmp->mii_if.full_duplex = 1;
717
718    /* lock the duplex mode if someone specified a value */
719    if (hmp->mii_if.full_duplex || (option & 0x080))
720        hmp->duplex_lock = 1;
721
722    /* Set interrupt tuning parameters */
723    max_rx_latency = max_rx_latency & 0x00ff;
724    max_rx_gap = max_rx_gap & 0x00ff;
725    min_rx_pkt = min_rx_pkt & 0x00ff;
726    max_tx_latency = max_tx_latency & 0x00ff;
727    max_tx_gap = max_tx_gap & 0x00ff;
728    min_tx_pkt = min_tx_pkt & 0x00ff;
729
730    rx_int_var = card_idx < MAX_UNITS ? rx_params[card_idx] : -1;
731    tx_int_var = card_idx < MAX_UNITS ? tx_params[card_idx] : -1;
732    hmp->rx_int_var = rx_int_var >= 0 ? rx_int_var :
733        (min_rx_pkt << 16 | max_rx_gap << 8 | max_rx_latency);
734    hmp->tx_int_var = tx_int_var >= 0 ? tx_int_var :
735        (min_tx_pkt << 16 | max_tx_gap << 8 | max_tx_latency);
736
737
738    /* The Hamachi-specific entries in the device structure. */
739    dev->netdev_ops = &hamachi_netdev_ops;
740    if (chip_tbl[hmp->chip_id].flags & CanHaveMII)
741        SET_ETHTOOL_OPS(dev, &ethtool_ops);
742    else
743        SET_ETHTOOL_OPS(dev, &ethtool_ops_no_mii);
744    dev->watchdog_timeo = TX_TIMEOUT;
745    if (mtu)
746        dev->mtu = mtu;
747
748    i = register_netdev(dev);
749    if (i) {
750        ret = i;
751        goto err_out_unmap_rx;
752    }
753
754    printk(KERN_INFO "%s: %s type %x at %p, %pM, IRQ %d.\n",
755           dev->name, chip_tbl[chip_id].name, readl(ioaddr + ChipRev),
756           ioaddr, dev->dev_addr, irq);
757    i = readb(ioaddr + PCIClkMeas);
758    printk(KERN_INFO "%s: %d-bit %d Mhz PCI bus (%d), Virtual Jumpers "
759           "%2.2x, LPA %4.4x.\n",
760           dev->name, readw(ioaddr + MiscStatus) & 1 ? 64 : 32,
761           i ? 2000/(i&0x7f) : 0, i&0x7f, (int)readb(ioaddr + VirtualJumpers),
762           readw(ioaddr + ANLinkPartnerAbility));
763
764    if (chip_tbl[hmp->chip_id].flags & CanHaveMII) {
765        int phy, phy_idx = 0;
766        for (phy = 0; phy < 32 && phy_idx < MII_CNT; phy++) {
767            int mii_status = mdio_read(dev, phy, MII_BMSR);
768            if (mii_status != 0xffff &&
769                mii_status != 0x0000) {
770                hmp->phys[phy_idx++] = phy;
771                hmp->mii_if.advertising = mdio_read(dev, phy, MII_ADVERTISE);
772                printk(KERN_INFO "%s: MII PHY found at address %d, status "
773                       "0x%4.4x advertising %4.4x.\n",
774                       dev->name, phy, mii_status, hmp->mii_if.advertising);
775            }
776        }
777        hmp->mii_cnt = phy_idx;
778        if (hmp->mii_cnt > 0)
779            hmp->mii_if.phy_id = hmp->phys[0];
780        else
781            memset(&hmp->mii_if, 0, sizeof(hmp->mii_if));
782    }
783    /* Configure gigabit autonegotiation. */
784    writew(0x0400, ioaddr + ANXchngCtrl); /* Enable legacy links. */
785    writew(0x08e0, ioaddr + ANAdvertise); /* Set our advertise word. */
786    writew(0x1000, ioaddr + ANCtrl); /* Enable negotiation */
787
788    card_idx++;
789    return 0;
790
791err_out_unmap_rx:
792    pci_free_consistent(pdev, RX_TOTAL_SIZE, hmp->rx_ring,
793        hmp->rx_ring_dma);
794err_out_unmap_tx:
795    pci_free_consistent(pdev, TX_TOTAL_SIZE, hmp->tx_ring,
796        hmp->tx_ring_dma);
797err_out_cleardev:
798    free_netdev (dev);
799err_out_iounmap:
800    iounmap(ioaddr);
801err_out_release:
802    pci_release_regions(pdev);
803err_out:
804    return ret;
805}
806
807static int __devinit read_eeprom(void __iomem *ioaddr, int location)
808{
809    int bogus_cnt = 1000;
810
811    /* We should check busy first - per docs -KDU */
812    while ((readb(ioaddr + EECmdStatus) & 0x40) && --bogus_cnt > 0);
813    writew(location, ioaddr + EEAddr);
814    writeb(0x02, ioaddr + EECmdStatus);
815    bogus_cnt = 1000;
816    while ((readb(ioaddr + EECmdStatus) & 0x40) && --bogus_cnt > 0);
817    if (hamachi_debug > 5)
818        printk(" EEPROM status is %2.2x after %d ticks.\n",
819               (int)readb(ioaddr + EECmdStatus), 1000- bogus_cnt);
820    return readb(ioaddr + EEData);
821}
822
823/* MII Managemen Data I/O accesses.
824   These routines assume the MDIO controller is idle, and do not exit until
825   the command is finished. */
826
827static int mdio_read(struct net_device *dev, int phy_id, int location)
828{
829    struct hamachi_private *hmp = netdev_priv(dev);
830    void __iomem *ioaddr = hmp->base;
831    int i;
832
833    /* We should check busy first - per docs -KDU */
834    for (i = 10000; i >= 0; i--)
835        if ((readw(ioaddr + MII_Status) & 1) == 0)
836            break;
837    writew((phy_id<<8) + location, ioaddr + MII_Addr);
838    writew(0x0001, ioaddr + MII_Cmd);
839    for (i = 10000; i >= 0; i--)
840        if ((readw(ioaddr + MII_Status) & 1) == 0)
841            break;
842    return readw(ioaddr + MII_Rd_Data);
843}
844
845static void mdio_write(struct net_device *dev, int phy_id, int location, int value)
846{
847    struct hamachi_private *hmp = netdev_priv(dev);
848    void __iomem *ioaddr = hmp->base;
849    int i;
850
851    /* We should check busy first - per docs -KDU */
852    for (i = 10000; i >= 0; i--)
853        if ((readw(ioaddr + MII_Status) & 1) == 0)
854            break;
855    writew((phy_id<<8) + location, ioaddr + MII_Addr);
856    writew(value, ioaddr + MII_Wr_Data);
857
858    /* Wait for the command to finish. */
859    for (i = 10000; i >= 0; i--)
860        if ((readw(ioaddr + MII_Status) & 1) == 0)
861            break;
862    return;
863}
864
865
866static int hamachi_open(struct net_device *dev)
867{
868    struct hamachi_private *hmp = netdev_priv(dev);
869    void __iomem *ioaddr = hmp->base;
870    int i;
871    u32 rx_int_var, tx_int_var;
872    u16 fifo_info;
873
874    i = request_irq(dev->irq, &hamachi_interrupt, IRQF_SHARED, dev->name, dev);
875    if (i)
876        return i;
877
878    if (hamachi_debug > 1)
879        printk(KERN_DEBUG "%s: hamachi_open() irq %d.\n",
880               dev->name, dev->irq);
881
882    hamachi_init_ring(dev);
883
884#if ADDRLEN == 64
885    /* writellll anyone ? */
886    writel(hmp->rx_ring_dma, ioaddr + RxPtr);
887    writel(hmp->rx_ring_dma >> 32, ioaddr + RxPtr + 4);
888    writel(hmp->tx_ring_dma, ioaddr + TxPtr);
889    writel(hmp->tx_ring_dma >> 32, ioaddr + TxPtr + 4);
890#else
891    writel(hmp->rx_ring_dma, ioaddr + RxPtr);
892    writel(hmp->tx_ring_dma, ioaddr + TxPtr);
893#endif
894
895    /* TODO: It would make sense to organize this as words since the card
896     * documentation does. -KDU
897     */
898    for (i = 0; i < 6; i++)
899        writeb(dev->dev_addr[i], ioaddr + StationAddr + i);
900
901    /* Initialize other registers: with so many this eventually this will
902       converted to an offset/value list. */
903
904    /* Configure the FIFO */
905    fifo_info = (readw(ioaddr + GPIO) & 0x00C0) >> 6;
906    switch (fifo_info){
907        case 0 :
908            /* No FIFO */
909            writew(0x0000, ioaddr + FIFOcfg);
910            break;
911        case 1 :
912            /* Configure the FIFO for 512K external, 16K used for Tx. */
913            writew(0x0028, ioaddr + FIFOcfg);
914            break;
915        case 2 :
916            /* Configure the FIFO for 1024 external, 32K used for Tx. */
917            writew(0x004C, ioaddr + FIFOcfg);
918            break;
919        case 3 :
920            /* Configure the FIFO for 2048 external, 32K used for Tx. */
921            writew(0x006C, ioaddr + FIFOcfg);
922            break;
923        default :
924            printk(KERN_WARNING "%s: Unsupported external memory config!\n",
925                dev->name);
926            /* Default to no FIFO */
927            writew(0x0000, ioaddr + FIFOcfg);
928            break;
929    }
930
931    if (dev->if_port == 0)
932        dev->if_port = hmp->default_port;
933
934
935    /* Setting the Rx mode will start the Rx process. */
936    /* If someone didn't choose a duplex, default to full-duplex */
937    if (hmp->duplex_lock != 1)
938        hmp->mii_if.full_duplex = 1;
939
940    /* always 1, takes no more time to do it */
941    writew(0x0001, ioaddr + RxChecksum);
942#ifdef TX_CHECKSUM
943    writew(0x0001, ioaddr + TxChecksum);
944#else
945    writew(0x0000, ioaddr + TxChecksum);
946#endif
947    writew(0x8000, ioaddr + MACCnfg); /* Soft reset the MAC */
948    writew(0x215F, ioaddr + MACCnfg);
949    writew(0x000C, ioaddr + FrameGap0);
950    /* WHAT?!?!? Why isn't this documented somewhere? -KDU */
951    writew(0x1018, ioaddr + FrameGap1);
952    /* Why do we enable receives/transmits here? -KDU */
953    writew(0x0780, ioaddr + MACCnfg2); /* Upper 16 bits control LEDs. */
954    /* Enable automatic generation of flow control frames, period 0xffff. */
955    writel(0x0030FFFF, ioaddr + FlowCtrl);
956    writew(MAX_FRAME_SIZE, ioaddr + MaxFrameSize); /* dev->mtu+14 ??? */
957
958    /* Enable legacy links. */
959    writew(0x0400, ioaddr + ANXchngCtrl); /* Enable legacy links. */
960    /* Initial Link LED to blinking red. */
961    writeb(0x03, ioaddr + LEDCtrl);
962
963    /* Configure interrupt mitigation. This has a great effect on
964       performance, so systems tuning should start here!. */
965
966    rx_int_var = hmp->rx_int_var;
967    tx_int_var = hmp->tx_int_var;
968
969    if (hamachi_debug > 1) {
970        printk("max_tx_latency: %d, max_tx_gap: %d, min_tx_pkt: %d\n",
971            tx_int_var & 0x00ff, (tx_int_var & 0x00ff00) >> 8,
972            (tx_int_var & 0x00ff0000) >> 16);
973        printk("max_rx_latency: %d, max_rx_gap: %d, min_rx_pkt: %d\n",
974            rx_int_var & 0x00ff, (rx_int_var & 0x00ff00) >> 8,
975            (rx_int_var & 0x00ff0000) >> 16);
976        printk("rx_int_var: %x, tx_int_var: %x\n", rx_int_var, tx_int_var);
977    }
978
979    writel(tx_int_var, ioaddr + TxIntrCtrl);
980    writel(rx_int_var, ioaddr + RxIntrCtrl);
981
982    set_rx_mode(dev);
983
984    netif_start_queue(dev);
985
986    /* Enable interrupts by setting the interrupt mask. */
987    writel(0x80878787, ioaddr + InterruptEnable);
988    writew(0x0000, ioaddr + EventStatus); /* Clear non-interrupting events */
989
990    /* Configure and start the DMA channels. */
991    /* Burst sizes are in the low three bits: size = 4<<(val&7) */
992#if ADDRLEN == 64
993    writew(0x005D, ioaddr + RxDMACtrl); /* 128 dword bursts */
994    writew(0x005D, ioaddr + TxDMACtrl);
995#else
996    writew(0x001D, ioaddr + RxDMACtrl);
997    writew(0x001D, ioaddr + TxDMACtrl);
998#endif
999    writew(0x0001, ioaddr + RxCmd);
1000
1001    if (hamachi_debug > 2) {
1002        printk(KERN_DEBUG "%s: Done hamachi_open(), status: Rx %x Tx %x.\n",
1003               dev->name, readw(ioaddr + RxStatus), readw(ioaddr + TxStatus));
1004    }
1005    /* Set the timer to check for link beat. */
1006    init_timer(&hmp->timer);
1007    hmp->timer.expires = RUN_AT((24*HZ)/10); /* 2.4 sec. */
1008    hmp->timer.data = (unsigned long)dev;
1009    hmp->timer.function = &hamachi_timer; /* timer handler */
1010    add_timer(&hmp->timer);
1011
1012    return 0;
1013}
1014
1015static inline int hamachi_tx(struct net_device *dev)
1016{
1017    struct hamachi_private *hmp = netdev_priv(dev);
1018
1019    /* Update the dirty pointer until we find an entry that is
1020        still owned by the card */
1021    for (; hmp->cur_tx - hmp->dirty_tx > 0; hmp->dirty_tx++) {
1022        int entry = hmp->dirty_tx % TX_RING_SIZE;
1023        struct sk_buff *skb;
1024
1025        if (hmp->tx_ring[entry].status_n_length & cpu_to_le32(DescOwn))
1026            break;
1027        /* Free the original skb. */
1028        skb = hmp->tx_skbuff[entry];
1029        if (skb) {
1030            pci_unmap_single(hmp->pci_dev,
1031                leXX_to_cpu(hmp->tx_ring[entry].addr),
1032                skb->len, PCI_DMA_TODEVICE);
1033            dev_kfree_skb(skb);
1034            hmp->tx_skbuff[entry] = NULL;
1035        }
1036        hmp->tx_ring[entry].status_n_length = 0;
1037        if (entry >= TX_RING_SIZE-1)
1038            hmp->tx_ring[TX_RING_SIZE-1].status_n_length |=
1039                cpu_to_le32(DescEndRing);
1040        hmp->stats.tx_packets++;
1041    }
1042
1043    return 0;
1044}
1045
1046static void hamachi_timer(unsigned long data)
1047{
1048    struct net_device *dev = (struct net_device *)data;
1049    struct hamachi_private *hmp = netdev_priv(dev);
1050    void __iomem *ioaddr = hmp->base;
1051    int next_tick = 10*HZ;
1052
1053    if (hamachi_debug > 2) {
1054        printk(KERN_INFO "%s: Hamachi Autonegotiation status %4.4x, LPA "
1055               "%4.4x.\n", dev->name, readw(ioaddr + ANStatus),
1056               readw(ioaddr + ANLinkPartnerAbility));
1057        printk(KERN_INFO "%s: Autonegotiation regs %4.4x %4.4x %4.4x "
1058               "%4.4x %4.4x %4.4x.\n", dev->name,
1059               readw(ioaddr + 0x0e0),
1060               readw(ioaddr + 0x0e2),
1061               readw(ioaddr + 0x0e4),
1062               readw(ioaddr + 0x0e6),
1063               readw(ioaddr + 0x0e8),
1064               readw(ioaddr + 0x0eA));
1065    }
1066    /* We could do something here... nah. */
1067    hmp->timer.expires = RUN_AT(next_tick);
1068    add_timer(&hmp->timer);
1069}
1070
1071static void hamachi_tx_timeout(struct net_device *dev)
1072{
1073    int i;
1074    struct hamachi_private *hmp = netdev_priv(dev);
1075    void __iomem *ioaddr = hmp->base;
1076
1077    printk(KERN_WARNING "%s: Hamachi transmit timed out, status %8.8x,"
1078           " resetting...\n", dev->name, (int)readw(ioaddr + TxStatus));
1079
1080    {
1081        printk(KERN_DEBUG " Rx ring %p: ", hmp->rx_ring);
1082        for (i = 0; i < RX_RING_SIZE; i++)
1083            printk(KERN_CONT " %8.8x",
1084                   le32_to_cpu(hmp->rx_ring[i].status_n_length));
1085        printk(KERN_CONT "\n");
1086        printk(KERN_DEBUG" Tx ring %p: ", hmp->tx_ring);
1087        for (i = 0; i < TX_RING_SIZE; i++)
1088            printk(KERN_CONT " %4.4x",
1089                   le32_to_cpu(hmp->tx_ring[i].status_n_length));
1090        printk(KERN_CONT "\n");
1091    }
1092
1093    /* Reinit the hardware and make sure the Rx and Tx processes
1094        are up and running.
1095     */
1096    dev->if_port = 0;
1097    /* The right way to do Reset. -KDU
1098     * -Clear OWN bit in all Rx/Tx descriptors
1099     * -Wait 50 uS for channels to go idle
1100     * -Turn off MAC receiver
1101     * -Issue Reset
1102     */
1103
1104    for (i = 0; i < RX_RING_SIZE; i++)
1105        hmp->rx_ring[i].status_n_length &= cpu_to_le32(~DescOwn);
1106
1107    /* Presume that all packets in the Tx queue are gone if we have to
1108     * re-init the hardware.
1109     */
1110    for (i = 0; i < TX_RING_SIZE; i++){
1111        struct sk_buff *skb;
1112
1113        if (i >= TX_RING_SIZE - 1)
1114            hmp->tx_ring[i].status_n_length =
1115                cpu_to_le32(DescEndRing) |
1116                (hmp->tx_ring[i].status_n_length &
1117                 cpu_to_le32(0x0000ffff));
1118        else
1119            hmp->tx_ring[i].status_n_length &= cpu_to_le32(0x0000ffff);
1120        skb = hmp->tx_skbuff[i];
1121        if (skb){
1122            pci_unmap_single(hmp->pci_dev, leXX_to_cpu(hmp->tx_ring[i].addr),
1123                skb->len, PCI_DMA_TODEVICE);
1124            dev_kfree_skb(skb);
1125            hmp->tx_skbuff[i] = NULL;
1126        }
1127    }
1128
1129    udelay(60); /* Sleep 60 us just for safety sake */
1130    writew(0x0002, ioaddr + RxCmd); /* STOP Rx */
1131
1132    writeb(0x01, ioaddr + ChipReset); /* Reinit the hardware */
1133
1134    hmp->tx_full = 0;
1135    hmp->cur_rx = hmp->cur_tx = 0;
1136    hmp->dirty_rx = hmp->dirty_tx = 0;
1137    /* Rx packets are also presumed lost; however, we need to make sure a
1138     * ring of buffers is in tact. -KDU
1139     */
1140    for (i = 0; i < RX_RING_SIZE; i++){
1141        struct sk_buff *skb = hmp->rx_skbuff[i];
1142
1143        if (skb){
1144            pci_unmap_single(hmp->pci_dev,
1145                leXX_to_cpu(hmp->rx_ring[i].addr),
1146                hmp->rx_buf_sz, PCI_DMA_FROMDEVICE);
1147            dev_kfree_skb(skb);
1148            hmp->rx_skbuff[i] = NULL;
1149        }
1150    }
1151    /* Fill in the Rx buffers. Handle allocation failure gracefully. */
1152    for (i = 0; i < RX_RING_SIZE; i++) {
1153        struct sk_buff *skb = netdev_alloc_skb(dev, hmp->rx_buf_sz);
1154        hmp->rx_skbuff[i] = skb;
1155        if (skb == NULL)
1156            break;
1157
1158        skb_reserve(skb, 2); /* 16 byte align the IP header. */
1159                hmp->rx_ring[i].addr = cpu_to_leXX(pci_map_single(hmp->pci_dev,
1160            skb->data, hmp->rx_buf_sz, PCI_DMA_FROMDEVICE));
1161        hmp->rx_ring[i].status_n_length = cpu_to_le32(DescOwn |
1162            DescEndPacket | DescIntr | (hmp->rx_buf_sz - 2));
1163    }
1164    hmp->dirty_rx = (unsigned int)(i - RX_RING_SIZE);
1165    /* Mark the last entry as wrapping the ring. */
1166    hmp->rx_ring[RX_RING_SIZE-1].status_n_length |= cpu_to_le32(DescEndRing);
1167
1168    /* Trigger an immediate transmit demand. */
1169    dev->trans_start = jiffies; /* prevent tx timeout */
1170    hmp->stats.tx_errors++;
1171
1172    /* Restart the chip's Tx/Rx processes . */
1173    writew(0x0002, ioaddr + TxCmd); /* STOP Tx */
1174    writew(0x0001, ioaddr + TxCmd); /* START Tx */
1175    writew(0x0001, ioaddr + RxCmd); /* START Rx */
1176
1177    netif_wake_queue(dev);
1178}
1179
1180
1181/* Initialize the Rx and Tx rings, along with various 'dev' bits. */
1182static void hamachi_init_ring(struct net_device *dev)
1183{
1184    struct hamachi_private *hmp = netdev_priv(dev);
1185    int i;
1186
1187    hmp->tx_full = 0;
1188    hmp->cur_rx = hmp->cur_tx = 0;
1189    hmp->dirty_rx = hmp->dirty_tx = 0;
1190
1191    /* +26 gets the maximum ethernet encapsulation, +7 & ~7 because the
1192     * card needs room to do 8 byte alignment, +2 so we can reserve
1193     * the first 2 bytes, and +16 gets room for the status word from the
1194     * card. -KDU
1195     */
1196    hmp->rx_buf_sz = (dev->mtu <= 1492 ? PKT_BUF_SZ :
1197        (((dev->mtu+26+7) & ~7) + 2 + 16));
1198
1199    /* Initialize all Rx descriptors. */
1200    for (i = 0; i < RX_RING_SIZE; i++) {
1201        hmp->rx_ring[i].status_n_length = 0;
1202        hmp->rx_skbuff[i] = NULL;
1203    }
1204    /* Fill in the Rx buffers. Handle allocation failure gracefully. */
1205    for (i = 0; i < RX_RING_SIZE; i++) {
1206        struct sk_buff *skb = dev_alloc_skb(hmp->rx_buf_sz);
1207        hmp->rx_skbuff[i] = skb;
1208        if (skb == NULL)
1209            break;
1210        skb->dev = dev; /* Mark as being used by this device. */
1211        skb_reserve(skb, 2); /* 16 byte align the IP header. */
1212                hmp->rx_ring[i].addr = cpu_to_leXX(pci_map_single(hmp->pci_dev,
1213            skb->data, hmp->rx_buf_sz, PCI_DMA_FROMDEVICE));
1214        /* -2 because it doesn't REALLY have that first 2 bytes -KDU */
1215        hmp->rx_ring[i].status_n_length = cpu_to_le32(DescOwn |
1216            DescEndPacket | DescIntr | (hmp->rx_buf_sz -2));
1217    }
1218    hmp->dirty_rx = (unsigned int)(i - RX_RING_SIZE);
1219    hmp->rx_ring[RX_RING_SIZE-1].status_n_length |= cpu_to_le32(DescEndRing);
1220
1221    for (i = 0; i < TX_RING_SIZE; i++) {
1222        hmp->tx_skbuff[i] = NULL;
1223        hmp->tx_ring[i].status_n_length = 0;
1224    }
1225    /* Mark the last entry of the ring */
1226    hmp->tx_ring[TX_RING_SIZE-1].status_n_length |= cpu_to_le32(DescEndRing);
1227
1228    return;
1229}
1230
1231
1232#ifdef TX_CHECKSUM
1233#define csum_add(it, val) \
1234do { \
1235    it += (u16) (val); \
1236    if (it & 0xffff0000) { \
1237    it &= 0xffff; \
1238    ++it; \
1239    } \
1240} while (0)
1241    /* printk("add %04x --> %04x\n", val, it); \ */
1242
1243/* uh->len already network format, do not swap */
1244#define pseudo_csum_udp(sum,ih,uh) do { \
1245    sum = 0; \
1246    csum_add(sum, (ih)->saddr >> 16); \
1247    csum_add(sum, (ih)->saddr & 0xffff); \
1248    csum_add(sum, (ih)->daddr >> 16); \
1249    csum_add(sum, (ih)->daddr & 0xffff); \
1250    csum_add(sum, cpu_to_be16(IPPROTO_UDP)); \
1251    csum_add(sum, (uh)->len); \
1252} while (0)
1253
1254/* swap len */
1255#define pseudo_csum_tcp(sum,ih,len) do { \
1256    sum = 0; \
1257    csum_add(sum, (ih)->saddr >> 16); \
1258    csum_add(sum, (ih)->saddr & 0xffff); \
1259    csum_add(sum, (ih)->daddr >> 16); \
1260    csum_add(sum, (ih)->daddr & 0xffff); \
1261    csum_add(sum, cpu_to_be16(IPPROTO_TCP)); \
1262    csum_add(sum, htons(len)); \
1263} while (0)
1264#endif
1265
1266static int hamachi_start_xmit(struct sk_buff *skb, struct net_device *dev)
1267{
1268    struct hamachi_private *hmp = netdev_priv(dev);
1269    unsigned entry;
1270    u16 status;
1271
1272    /* Ok, now make sure that the queue has space before trying to
1273        add another skbuff. if we return non-zero the scheduler
1274        should interpret this as a queue full and requeue the buffer
1275        for later.
1276     */
1277    if (hmp->tx_full) {
1278        /* We should NEVER reach this point -KDU */
1279        printk(KERN_WARNING "%s: Hamachi transmit queue full at slot %d.\n",dev->name, hmp->cur_tx);
1280
1281        /* Wake the potentially-idle transmit channel. */
1282        /* If we don't need to read status, DON'T -KDU */
1283        status=readw(hmp->base + TxStatus);
1284        if( !(status & 0x0001) || (status & 0x0002))
1285            writew(0x0001, hmp->base + TxCmd);
1286        return NETDEV_TX_BUSY;
1287    }
1288
1289    /* Caution: the write order is important here, set the field
1290       with the "ownership" bits last. */
1291
1292    /* Calculate the next Tx descriptor entry. */
1293    entry = hmp->cur_tx % TX_RING_SIZE;
1294
1295    hmp->tx_skbuff[entry] = skb;
1296
1297#ifdef TX_CHECKSUM
1298    {
1299        /* tack on checksum tag */
1300        u32 tagval = 0;
1301        struct ethhdr *eh = (struct ethhdr *)skb->data;
1302        if (eh->h_proto == cpu_to_be16(ETH_P_IP)) {
1303        struct iphdr *ih = (struct iphdr *)((char *)eh + ETH_HLEN);
1304        if (ih->protocol == IPPROTO_UDP) {
1305            struct udphdr *uh
1306              = (struct udphdr *)((char *)ih + ih->ihl*4);
1307            u32 offset = ((unsigned char *)uh + 6) - skb->data;
1308            u32 pseudo;
1309            pseudo_csum_udp(pseudo, ih, uh);
1310            pseudo = htons(pseudo);
1311            printk("udp cksum was %04x, sending pseudo %04x\n",
1312              uh->check, pseudo);
1313            uh->check = 0; /* zero out uh->check before card calc */
1314            /*
1315             * start at 14 (skip ethhdr), store at offset (uh->check),
1316             * use pseudo value given.
1317             */
1318            tagval = (14 << 24) | (offset << 16) | pseudo;
1319        } else if (ih->protocol == IPPROTO_TCP) {
1320            printk("tcp, no auto cksum\n");
1321        }
1322        }
1323        *(u32 *)skb_push(skb, 8) = tagval;
1324    }
1325#endif
1326
1327        hmp->tx_ring[entry].addr = cpu_to_leXX(pci_map_single(hmp->pci_dev,
1328        skb->data, skb->len, PCI_DMA_TODEVICE));
1329
1330    /* Hmmmm, could probably put a DescIntr on these, but the way
1331        the driver is currently coded makes Tx interrupts unnecessary
1332        since the clearing of the Tx ring is handled by the start_xmit
1333        routine. This organization helps mitigate the interrupts a
1334        bit and probably renders the max_tx_latency param useless.
1335
1336        Update: Putting a DescIntr bit on all of the descriptors and
1337        mitigating interrupt frequency with the tx_min_pkt parameter. -KDU
1338    */
1339    if (entry >= TX_RING_SIZE-1) /* Wrap ring */
1340        hmp->tx_ring[entry].status_n_length = cpu_to_le32(DescOwn |
1341            DescEndPacket | DescEndRing | DescIntr | skb->len);
1342    else
1343        hmp->tx_ring[entry].status_n_length = cpu_to_le32(DescOwn |
1344            DescEndPacket | DescIntr | skb->len);
1345    hmp->cur_tx++;
1346
1347    /* Non-x86 Todo: explicitly flush cache lines here. */
1348
1349    /* Wake the potentially-idle transmit channel. */
1350    /* If we don't need to read status, DON'T -KDU */
1351    status=readw(hmp->base + TxStatus);
1352    if( !(status & 0x0001) || (status & 0x0002))
1353        writew(0x0001, hmp->base + TxCmd);
1354
1355    /* Immediately before returning, let's clear as many entries as we can. */
1356    hamachi_tx(dev);
1357
1358    /* We should kick the bottom half here, since we are not accepting
1359     * interrupts with every packet. i.e. realize that Gigabit ethernet
1360     * can transmit faster than ordinary machines can load packets;
1361     * hence, any packet that got put off because we were in the transmit
1362     * routine should IMMEDIATELY get a chance to be re-queued. -KDU
1363     */
1364    if ((hmp->cur_tx - hmp->dirty_tx) < (TX_RING_SIZE - 4))
1365        netif_wake_queue(dev); /* Typical path */
1366    else {
1367        hmp->tx_full = 1;
1368        netif_stop_queue(dev);
1369    }
1370
1371    if (hamachi_debug > 4) {
1372        printk(KERN_DEBUG "%s: Hamachi transmit frame #%d queued in slot %d.\n",
1373               dev->name, hmp->cur_tx, entry);
1374    }
1375    return 0;
1376}
1377
1378/* The interrupt handler does all of the Rx thread work and cleans up
1379   after the Tx thread. */
1380static irqreturn_t hamachi_interrupt(int irq, void *dev_instance)
1381{
1382    struct net_device *dev = dev_instance;
1383    struct hamachi_private *hmp = netdev_priv(dev);
1384    void __iomem *ioaddr = hmp->base;
1385    long boguscnt = max_interrupt_work;
1386    int handled = 0;
1387
1388#ifndef final_version /* Can never occur. */
1389    if (dev == NULL) {
1390        printk (KERN_ERR "hamachi_interrupt(): irq %d for unknown device.\n", irq);
1391        return IRQ_NONE;
1392    }
1393#endif
1394
1395    spin_lock(&hmp->lock);
1396
1397    do {
1398        u32 intr_status = readl(ioaddr + InterruptClear);
1399
1400        if (hamachi_debug > 4)
1401            printk(KERN_DEBUG "%s: Hamachi interrupt, status %4.4x.\n",
1402                   dev->name, intr_status);
1403
1404        if (intr_status == 0)
1405            break;
1406
1407        handled = 1;
1408
1409        if (intr_status & IntrRxDone)
1410            hamachi_rx(dev);
1411
1412        if (intr_status & IntrTxDone){
1413            /* This code should RARELY need to execute. After all, this is
1414             * a gigabit link, it should consume packets as fast as we put
1415             * them in AND we clear the Tx ring in hamachi_start_xmit().
1416             */
1417            if (hmp->tx_full){
1418                for (; hmp->cur_tx - hmp->dirty_tx > 0; hmp->dirty_tx++){
1419                    int entry = hmp->dirty_tx % TX_RING_SIZE;
1420                    struct sk_buff *skb;
1421
1422                    if (hmp->tx_ring[entry].status_n_length & cpu_to_le32(DescOwn))
1423                        break;
1424                    skb = hmp->tx_skbuff[entry];
1425                    /* Free the original skb. */
1426                    if (skb){
1427                        pci_unmap_single(hmp->pci_dev,
1428                            leXX_to_cpu(hmp->tx_ring[entry].addr),
1429                            skb->len,
1430                            PCI_DMA_TODEVICE);
1431                        dev_kfree_skb_irq(skb);
1432                        hmp->tx_skbuff[entry] = NULL;
1433                    }
1434                    hmp->tx_ring[entry].status_n_length = 0;
1435                    if (entry >= TX_RING_SIZE-1)
1436                        hmp->tx_ring[TX_RING_SIZE-1].status_n_length |=
1437                            cpu_to_le32(DescEndRing);
1438                    hmp->stats.tx_packets++;
1439                }
1440                if (hmp->cur_tx - hmp->dirty_tx < TX_RING_SIZE - 4){
1441                    /* The ring is no longer full */
1442                    hmp->tx_full = 0;
1443                    netif_wake_queue(dev);
1444                }
1445            } else {
1446                netif_wake_queue(dev);
1447            }
1448        }
1449
1450
1451        /* Abnormal error summary/uncommon events handlers. */
1452        if (intr_status &
1453            (IntrTxPCIFault | IntrTxPCIErr | IntrRxPCIFault | IntrRxPCIErr |
1454             LinkChange | NegotiationChange | StatsMax))
1455            hamachi_error(dev, intr_status);
1456
1457        if (--boguscnt < 0) {
1458            printk(KERN_WARNING "%s: Too much work at interrupt, status=0x%4.4x.\n",
1459                   dev->name, intr_status);
1460            break;
1461        }
1462    } while (1);
1463
1464    if (hamachi_debug > 3)
1465        printk(KERN_DEBUG "%s: exiting interrupt, status=%#4.4x.\n",
1466               dev->name, readl(ioaddr + IntrStatus));
1467
1468#ifndef final_version
1469    /* Code that should never be run! Perhaps remove after testing.. */
1470    {
1471        static int stopit = 10;
1472        if (dev->start == 0 && --stopit < 0) {
1473            printk(KERN_ERR "%s: Emergency stop, looping startup interrupt.\n",
1474                   dev->name);
1475            free_irq(irq, dev);
1476        }
1477    }
1478#endif
1479
1480    spin_unlock(&hmp->lock);
1481    return IRQ_RETVAL(handled);
1482}
1483
1484/* This routine is logically part of the interrupt handler, but separated
1485   for clarity and better register allocation. */
1486static int hamachi_rx(struct net_device *dev)
1487{
1488    struct hamachi_private *hmp = netdev_priv(dev);
1489    int entry = hmp->cur_rx % RX_RING_SIZE;
1490    int boguscnt = (hmp->dirty_rx + RX_RING_SIZE) - hmp->cur_rx;
1491
1492    if (hamachi_debug > 4) {
1493        printk(KERN_DEBUG " In hamachi_rx(), entry %d status %4.4x.\n",
1494               entry, hmp->rx_ring[entry].status_n_length);
1495    }
1496
1497    /* If EOP is set on the next entry, it's a new packet. Send it up. */
1498    while (1) {
1499        struct hamachi_desc *desc = &(hmp->rx_ring[entry]);
1500        u32 desc_status = le32_to_cpu(desc->status_n_length);
1501        u16 data_size = desc_status; /* Implicit truncate */
1502        u8 *buf_addr;
1503        s32 frame_status;
1504
1505        if (desc_status & DescOwn)
1506            break;
1507        pci_dma_sync_single_for_cpu(hmp->pci_dev,
1508                        leXX_to_cpu(desc->addr),
1509                        hmp->rx_buf_sz,
1510                        PCI_DMA_FROMDEVICE);
1511        buf_addr = (u8 *) hmp->rx_skbuff[entry]->data;
1512        frame_status = get_unaligned_le32(&(buf_addr[data_size - 12]));
1513        if (hamachi_debug > 4)
1514            printk(KERN_DEBUG " hamachi_rx() status was %8.8x.\n",
1515                frame_status);
1516        if (--boguscnt < 0)
1517            break;
1518        if ( ! (desc_status & DescEndPacket)) {
1519            printk(KERN_WARNING "%s: Oversized Ethernet frame spanned "
1520                   "multiple buffers, entry %#x length %d status %4.4x!\n",
1521                   dev->name, hmp->cur_rx, data_size, desc_status);
1522            printk(KERN_WARNING "%s: Oversized Ethernet frame %p vs %p.\n",
1523                   dev->name, desc, &hmp->rx_ring[hmp->cur_rx % RX_RING_SIZE]);
1524            printk(KERN_WARNING "%s: Oversized Ethernet frame -- next status %x/%x last status %x.\n",
1525                   dev->name,
1526                   le32_to_cpu(hmp->rx_ring[(hmp->cur_rx+1) % RX_RING_SIZE].status_n_length) & 0xffff0000,
1527                   le32_to_cpu(hmp->rx_ring[(hmp->cur_rx+1) % RX_RING_SIZE].status_n_length) & 0x0000ffff,
1528                   le32_to_cpu(hmp->rx_ring[(hmp->cur_rx-1) % RX_RING_SIZE].status_n_length));
1529            hmp->stats.rx_length_errors++;
1530        } /* else Omit for prototype errata??? */
1531        if (frame_status & 0x00380000) {
1532            /* There was an error. */
1533            if (hamachi_debug > 2)
1534                printk(KERN_DEBUG " hamachi_rx() Rx error was %8.8x.\n",
1535                       frame_status);
1536            hmp->stats.rx_errors++;
1537            if (frame_status & 0x00600000) hmp->stats.rx_length_errors++;
1538            if (frame_status & 0x00080000) hmp->stats.rx_frame_errors++;
1539            if (frame_status & 0x00100000) hmp->stats.rx_crc_errors++;
1540            if (frame_status < 0) hmp->stats.rx_dropped++;
1541        } else {
1542            struct sk_buff *skb;
1543            /* Omit CRC */
1544            u16 pkt_len = (frame_status & 0x07ff) - 4;
1545#ifdef RX_CHECKSUM
1546            u32 pfck = *(u32 *) &buf_addr[data_size - 8];
1547#endif
1548
1549
1550#ifndef final_version
1551            if (hamachi_debug > 4)
1552                printk(KERN_DEBUG " hamachi_rx() normal Rx pkt length %d"
1553                       " of %d, bogus_cnt %d.\n",
1554                       pkt_len, data_size, boguscnt);
1555            if (hamachi_debug > 5)
1556                printk(KERN_DEBUG"%s: rx status %8.8x %8.8x %8.8x %8.8x %8.8x.\n",
1557                       dev->name,
1558                       *(s32*)&(buf_addr[data_size - 20]),
1559                       *(s32*)&(buf_addr[data_size - 16]),
1560                       *(s32*)&(buf_addr[data_size - 12]),
1561                       *(s32*)&(buf_addr[data_size - 8]),
1562                       *(s32*)&(buf_addr[data_size - 4]));
1563#endif
1564            /* Check if the packet is long enough to accept without copying
1565               to a minimally-sized skbuff. */
1566            if (pkt_len < rx_copybreak
1567                && (skb = dev_alloc_skb(pkt_len + 2)) != NULL) {
1568#ifdef RX_CHECKSUM
1569                printk(KERN_ERR "%s: rx_copybreak non-zero "
1570                  "not good with RX_CHECKSUM\n", dev->name);
1571#endif
1572                skb_reserve(skb, 2); /* 16 byte align the IP header */
1573                pci_dma_sync_single_for_cpu(hmp->pci_dev,
1574                                leXX_to_cpu(hmp->rx_ring[entry].addr),
1575                                hmp->rx_buf_sz,
1576                                PCI_DMA_FROMDEVICE);
1577                /* Call copy + cksum if available. */
1578#if 1 || USE_IP_COPYSUM
1579                skb_copy_to_linear_data(skb,
1580                    hmp->rx_skbuff[entry]->data, pkt_len);
1581                skb_put(skb, pkt_len);
1582#else
1583                memcpy(skb_put(skb, pkt_len), hmp->rx_ring_dma
1584                    + entry*sizeof(*desc), pkt_len);
1585#endif
1586                pci_dma_sync_single_for_device(hmp->pci_dev,
1587                                   leXX_to_cpu(hmp->rx_ring[entry].addr),
1588                                   hmp->rx_buf_sz,
1589                                   PCI_DMA_FROMDEVICE);
1590            } else {
1591                pci_unmap_single(hmp->pci_dev,
1592                         leXX_to_cpu(hmp->rx_ring[entry].addr),
1593                         hmp->rx_buf_sz, PCI_DMA_FROMDEVICE);
1594                skb_put(skb = hmp->rx_skbuff[entry], pkt_len);
1595                hmp->rx_skbuff[entry] = NULL;
1596            }
1597            skb->protocol = eth_type_trans(skb, dev);
1598
1599
1600#ifdef RX_CHECKSUM
1601            /* TCP or UDP on ipv4, DIX encoding */
1602            if (pfck>>24 == 0x91 || pfck>>24 == 0x51) {
1603                struct iphdr *ih = (struct iphdr *) skb->data;
1604                /* Check that IP packet is at least 46 bytes, otherwise,
1605                 * there may be pad bytes included in the hardware checksum.
1606                 * This wouldn't happen if everyone padded with 0.
1607                 */
1608                if (ntohs(ih->tot_len) >= 46){
1609                    /* don't worry about frags */
1610                    if (!(ih->frag_off & cpu_to_be16(IP_MF|IP_OFFSET))) {
1611                        u32 inv = *(u32 *) &buf_addr[data_size - 16];
1612                        u32 *p = (u32 *) &buf_addr[data_size - 20];
1613                        register u32 crc, p_r, p_r1;
1614
1615                        if (inv & 4) {
1616                            inv &= ~4;
1617                            --p;
1618                        }
1619                        p_r = *p;
1620                        p_r1 = *(p-1);
1621                        switch (inv) {
1622                            case 0:
1623                                crc = (p_r & 0xffff) + (p_r >> 16);
1624                                break;
1625                            case 1:
1626                                crc = (p_r >> 16) + (p_r & 0xffff)
1627                                    + (p_r1 >> 16 & 0xff00);
1628                                break;
1629                            case 2:
1630                                crc = p_r + (p_r1 >> 16);
1631                                break;
1632                            case 3:
1633                                crc = p_r + (p_r1 & 0xff00) + (p_r1 >> 16);
1634                                break;
1635                            default: /*NOTREACHED*/ crc = 0;
1636                        }
1637                        if (crc & 0xffff0000) {
1638                            crc &= 0xffff;
1639                            ++crc;
1640                        }
1641                        /* tcp/udp will add in pseudo */
1642                        skb->csum = ntohs(pfck & 0xffff);
1643                        if (skb->csum > crc)
1644                            skb->csum -= crc;
1645                        else
1646                            skb->csum += (~crc & 0xffff);
1647                        /*
1648                        * could do the pseudo myself and return
1649                        * CHECKSUM_UNNECESSARY
1650                        */
1651                        skb->ip_summed = CHECKSUM_COMPLETE;
1652                    }
1653                }
1654            }
1655#endif /* RX_CHECKSUM */
1656
1657            netif_rx(skb);
1658            hmp->stats.rx_packets++;
1659        }
1660        entry = (++hmp->cur_rx) % RX_RING_SIZE;
1661    }
1662
1663    /* Refill the Rx ring buffers. */
1664    for (; hmp->cur_rx - hmp->dirty_rx > 0; hmp->dirty_rx++) {
1665        struct hamachi_desc *desc;
1666
1667        entry = hmp->dirty_rx % RX_RING_SIZE;
1668        desc = &(hmp->rx_ring[entry]);
1669        if (hmp->rx_skbuff[entry] == NULL) {
1670            struct sk_buff *skb = dev_alloc_skb(hmp->rx_buf_sz);
1671
1672            hmp->rx_skbuff[entry] = skb;
1673            if (skb == NULL)
1674                break; /* Better luck next round. */
1675            skb->dev = dev; /* Mark as being used by this device. */
1676            skb_reserve(skb, 2); /* Align IP on 16 byte boundaries */
1677                    desc->addr = cpu_to_leXX(pci_map_single(hmp->pci_dev,
1678                skb->data, hmp->rx_buf_sz, PCI_DMA_FROMDEVICE));
1679        }
1680        desc->status_n_length = cpu_to_le32(hmp->rx_buf_sz);
1681        if (entry >= RX_RING_SIZE-1)
1682            desc->status_n_length |= cpu_to_le32(DescOwn |
1683                DescEndPacket | DescEndRing | DescIntr);
1684        else
1685            desc->status_n_length |= cpu_to_le32(DescOwn |
1686                DescEndPacket | DescIntr);
1687    }
1688
1689    /* Restart Rx engine if stopped. */
1690    /* If we don't need to check status, don't. -KDU */
1691    if (readw(hmp->base + RxStatus) & 0x0002)
1692        writew(0x0001, hmp->base + RxCmd);
1693
1694    return 0;
1695}
1696
1697/* This is more properly named "uncommon interrupt events", as it covers more
1698   than just errors. */
1699static void hamachi_error(struct net_device *dev, int intr_status)
1700{
1701    struct hamachi_private *hmp = netdev_priv(dev);
1702    void __iomem *ioaddr = hmp->base;
1703
1704    if (intr_status & (LinkChange|NegotiationChange)) {
1705        if (hamachi_debug > 1)
1706            printk(KERN_INFO "%s: Link changed: AutoNegotiation Ctrl"
1707                   " %4.4x, Status %4.4x %4.4x Intr status %4.4x.\n",
1708                   dev->name, readw(ioaddr + 0x0E0), readw(ioaddr + 0x0E2),
1709                   readw(ioaddr + ANLinkPartnerAbility),
1710                   readl(ioaddr + IntrStatus));
1711        if (readw(ioaddr + ANStatus) & 0x20)
1712            writeb(0x01, ioaddr + LEDCtrl);
1713        else
1714            writeb(0x03, ioaddr + LEDCtrl);
1715    }
1716    if (intr_status & StatsMax) {
1717        hamachi_get_stats(dev);
1718        /* Read the overflow bits to clear. */
1719        readl(ioaddr + 0x370);
1720        readl(ioaddr + 0x3F0);
1721    }
1722    if ((intr_status & ~(LinkChange|StatsMax|NegotiationChange|IntrRxDone|IntrTxDone))
1723        && hamachi_debug)
1724        printk(KERN_ERR "%s: Something Wicked happened! %4.4x.\n",
1725               dev->name, intr_status);
1726    /* Hmmmmm, it's not clear how to recover from PCI faults. */
1727    if (intr_status & (IntrTxPCIErr | IntrTxPCIFault))
1728        hmp->stats.tx_fifo_errors++;
1729    if (intr_status & (IntrRxPCIErr | IntrRxPCIFault))
1730        hmp->stats.rx_fifo_errors++;
1731}
1732
1733static int hamachi_close(struct net_device *dev)
1734{
1735    struct hamachi_private *hmp = netdev_priv(dev);
1736    void __iomem *ioaddr = hmp->base;
1737    struct sk_buff *skb;
1738    int i;
1739
1740    netif_stop_queue(dev);
1741
1742    if (hamachi_debug > 1) {
1743        printk(KERN_DEBUG "%s: Shutting down ethercard, status was Tx %4.4x Rx %4.4x Int %2.2x.\n",
1744               dev->name, readw(ioaddr + TxStatus),
1745               readw(ioaddr + RxStatus), readl(ioaddr + IntrStatus));
1746        printk(KERN_DEBUG "%s: Queue pointers were Tx %d / %d, Rx %d / %d.\n",
1747               dev->name, hmp->cur_tx, hmp->dirty_tx, hmp->cur_rx, hmp->dirty_rx);
1748    }
1749
1750    /* Disable interrupts by clearing the interrupt mask. */
1751    writel(0x0000, ioaddr + InterruptEnable);
1752
1753    /* Stop the chip's Tx and Rx processes. */
1754    writel(2, ioaddr + RxCmd);
1755    writew(2, ioaddr + TxCmd);
1756
1757#ifdef __i386__
1758    if (hamachi_debug > 2) {
1759        printk(KERN_DEBUG " Tx ring at %8.8x:\n",
1760               (int)hmp->tx_ring_dma);
1761        for (i = 0; i < TX_RING_SIZE; i++)
1762            printk(KERN_DEBUG " %c #%d desc. %8.8x %8.8x.\n",
1763                   readl(ioaddr + TxCurPtr) == (long)&hmp->tx_ring[i] ? '>' : ' ',
1764                   i, hmp->tx_ring[i].status_n_length, hmp->tx_ring[i].addr);
1765        printk(KERN_DEBUG " Rx ring %8.8x:\n",
1766               (int)hmp->rx_ring_dma);
1767        for (i = 0; i < RX_RING_SIZE; i++) {
1768            printk(KERN_DEBUG " %c #%d desc. %4.4x %8.8x\n",
1769                   readl(ioaddr + RxCurPtr) == (long)&hmp->rx_ring[i] ? '>' : ' ',
1770                   i, hmp->rx_ring[i].status_n_length, hmp->rx_ring[i].addr);
1771            if (hamachi_debug > 6) {
1772                if (*(u8*)hmp->rx_skbuff[i]->data != 0x69) {
1773                    u16 *addr = (u16 *)
1774                        hmp->rx_skbuff[i]->data;
1775                    int j;
1776                    printk(KERN_DEBUG "Addr: ");
1777                    for (j = 0; j < 0x50; j++)
1778                        printk(" %4.4x", addr[j]);
1779                    printk("\n");
1780                }
1781            }
1782        }
1783    }
1784#endif /* __i386__ debugging only */
1785
1786    free_irq(dev->irq, dev);
1787
1788    del_timer_sync(&hmp->timer);
1789
1790    /* Free all the skbuffs in the Rx queue. */
1791    for (i = 0; i < RX_RING_SIZE; i++) {
1792        skb = hmp->rx_skbuff[i];
1793        hmp->rx_ring[i].status_n_length = 0;
1794        if (skb) {
1795            pci_unmap_single(hmp->pci_dev,
1796                leXX_to_cpu(hmp->rx_ring[i].addr),
1797                hmp->rx_buf_sz, PCI_DMA_FROMDEVICE);
1798            dev_kfree_skb(skb);
1799            hmp->rx_skbuff[i] = NULL;
1800        }
1801        hmp->rx_ring[i].addr = cpu_to_leXX(0xBADF00D0); /* An invalid address. */
1802    }
1803    for (i = 0; i < TX_RING_SIZE; i++) {
1804        skb = hmp->tx_skbuff[i];
1805        if (skb) {
1806            pci_unmap_single(hmp->pci_dev,
1807                leXX_to_cpu(hmp->tx_ring[i].addr),
1808                skb->len, PCI_DMA_TODEVICE);
1809            dev_kfree_skb(skb);
1810            hmp->tx_skbuff[i] = NULL;
1811        }
1812    }
1813
1814    writeb(0x00, ioaddr + LEDCtrl);
1815
1816    return 0;
1817}
1818
1819static struct net_device_stats *hamachi_get_stats(struct net_device *dev)
1820{
1821    struct hamachi_private *hmp = netdev_priv(dev);
1822    void __iomem *ioaddr = hmp->base;
1823
1824    /* We should lock this segment of code for SMP eventually, although
1825       the vulnerability window is very small and statistics are
1826       non-critical. */
1827        /* Ok, what goes here? This appears to be stuck at 21 packets
1828           according to ifconfig. It does get incremented in hamachi_tx(),
1829           so I think I'll comment it out here and see if better things
1830           happen.
1831        */
1832    /* hmp->stats.tx_packets = readl(ioaddr + 0x000); */
1833
1834    hmp->stats.rx_bytes = readl(ioaddr + 0x330); /* Total Uni+Brd+Multi */
1835    hmp->stats.tx_bytes = readl(ioaddr + 0x3B0); /* Total Uni+Brd+Multi */
1836    hmp->stats.multicast = readl(ioaddr + 0x320); /* Multicast Rx */
1837
1838    hmp->stats.rx_length_errors = readl(ioaddr + 0x368); /* Over+Undersized */
1839    hmp->stats.rx_over_errors = readl(ioaddr + 0x35C); /* Jabber */
1840    hmp->stats.rx_crc_errors = readl(ioaddr + 0x360); /* Jabber */
1841    hmp->stats.rx_frame_errors = readl(ioaddr + 0x364); /* Symbol Errs */
1842    hmp->stats.rx_missed_errors = readl(ioaddr + 0x36C); /* Dropped */
1843
1844    return &hmp->stats;
1845}
1846
1847static void set_rx_mode(struct net_device *dev)
1848{
1849    struct hamachi_private *hmp = netdev_priv(dev);
1850    void __iomem *ioaddr = hmp->base;
1851
1852    if (dev->flags & IFF_PROMISC) { /* Set promiscuous. */
1853        writew(0x000F, ioaddr + AddrMode);
1854    } else if ((dev->mc_count > 63) || (dev->flags & IFF_ALLMULTI)) {
1855        /* Too many to match, or accept all multicasts. */
1856        writew(0x000B, ioaddr + AddrMode);
1857    } else if (dev->mc_count > 0) { /* Must use the CAM filter. */
1858        struct dev_mc_list *mclist;
1859        int i;
1860        for (i = 0, mclist = dev->mc_list; mclist && i < dev->mc_count;
1861             i++, mclist = mclist->next) {
1862            writel(*(u32*)(mclist->dmi_addr), ioaddr + 0x100 + i*8);
1863            writel(0x20000 | (*(u16*)&mclist->dmi_addr[4]),
1864                   ioaddr + 0x104 + i*8);
1865        }
1866        /* Clear remaining entries. */
1867        for (; i < 64; i++)
1868            writel(0, ioaddr + 0x104 + i*8);
1869        writew(0x0003, ioaddr + AddrMode);
1870    } else { /* Normal, unicast/broadcast-only mode. */
1871        writew(0x0001, ioaddr + AddrMode);
1872    }
1873}
1874
1875static int check_if_running(struct net_device *dev)
1876{
1877    if (!netif_running(dev))
1878        return -EINVAL;
1879    return 0;
1880}
1881
1882static void hamachi_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1883{
1884    struct hamachi_private *np = netdev_priv(dev);
1885    strcpy(info->driver, DRV_NAME);
1886    strcpy(info->version, DRV_VERSION);
1887    strcpy(info->bus_info, pci_name(np->pci_dev));
1888}
1889
1890static int hamachi_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
1891{
1892    struct hamachi_private *np = netdev_priv(dev);
1893    spin_lock_irq(&np->lock);
1894    mii_ethtool_gset(&np->mii_if, ecmd);
1895    spin_unlock_irq(&np->lock);
1896    return 0;
1897}
1898
1899static int hamachi_set_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
1900{
1901    struct hamachi_private *np = netdev_priv(dev);
1902    int res;
1903    spin_lock_irq(&np->lock);
1904    res = mii_ethtool_sset(&np->mii_if, ecmd);
1905    spin_unlock_irq(&np->lock);
1906    return res;
1907}
1908
1909static int hamachi_nway_reset(struct net_device *dev)
1910{
1911    struct hamachi_private *np = netdev_priv(dev);
1912    return mii_nway_restart(&np->mii_if);
1913}
1914
1915static u32 hamachi_get_link(struct net_device *dev)
1916{
1917    struct hamachi_private *np = netdev_priv(dev);
1918    return mii_link_ok(&np->mii_if);
1919}
1920
1921static const struct ethtool_ops ethtool_ops = {
1922    .begin = check_if_running,
1923    .get_drvinfo = hamachi_get_drvinfo,
1924    .get_settings = hamachi_get_settings,
1925    .set_settings = hamachi_set_settings,
1926    .nway_reset = hamachi_nway_reset,
1927    .get_link = hamachi_get_link,
1928};
1929
1930static const struct ethtool_ops ethtool_ops_no_mii = {
1931    .begin = check_if_running,
1932    .get_drvinfo = hamachi_get_drvinfo,
1933};
1934
1935static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
1936{
1937    struct hamachi_private *np = netdev_priv(dev);
1938    struct mii_ioctl_data *data = if_mii(rq);
1939    int rc;
1940
1941    if (!netif_running(dev))
1942        return -EINVAL;
1943
1944    if (cmd == (SIOCDEVPRIVATE+3)) { /* set rx,tx intr params */
1945        u32 *d = (u32 *)&rq->ifr_ifru;
1946        /* Should add this check here or an ordinary user can do nasty
1947         * things. -KDU
1948         *
1949         * TODO: Shut down the Rx and Tx engines while doing this.
1950         */
1951        if (!capable(CAP_NET_ADMIN))
1952            return -EPERM;
1953        writel(d[0], np->base + TxIntrCtrl);
1954        writel(d[1], np->base + RxIntrCtrl);
1955        printk(KERN_NOTICE "%s: tx %08x, rx %08x intr\n", dev->name,
1956          (u32) readl(np->base + TxIntrCtrl),
1957          (u32) readl(np->base + RxIntrCtrl));
1958        rc = 0;
1959    }
1960
1961    else {
1962        spin_lock_irq(&np->lock);
1963        rc = generic_mii_ioctl(&np->mii_if, data, cmd, NULL);
1964        spin_unlock_irq(&np->lock);
1965    }
1966
1967    return rc;
1968}
1969
1970
1971static void __devexit hamachi_remove_one (struct pci_dev *pdev)
1972{
1973    struct net_device *dev = pci_get_drvdata(pdev);
1974
1975    if (dev) {
1976        struct hamachi_private *hmp = netdev_priv(dev);
1977
1978        pci_free_consistent(pdev, RX_TOTAL_SIZE, hmp->rx_ring,
1979            hmp->rx_ring_dma);
1980        pci_free_consistent(pdev, TX_TOTAL_SIZE, hmp->tx_ring,
1981            hmp->tx_ring_dma);
1982        unregister_netdev(dev);
1983        iounmap(hmp->base);
1984        free_netdev(dev);
1985        pci_release_regions(pdev);
1986        pci_set_drvdata(pdev, NULL);
1987    }
1988}
1989
1990static struct pci_device_id hamachi_pci_tbl[] = {
1991    { 0x1318, 0x0911, PCI_ANY_ID, PCI_ANY_ID, },
1992    { 0, }
1993};
1994MODULE_DEVICE_TABLE(pci, hamachi_pci_tbl);
1995
1996static struct pci_driver hamachi_driver = {
1997    .name = DRV_NAME,
1998    .id_table = hamachi_pci_tbl,
1999    .probe = hamachi_init_one,
2000    .remove = __devexit_p(hamachi_remove_one),
2001};
2002
2003static int __init hamachi_init (void)
2004{
2005/* when a module, this is printed whether or not devices are found in probe */
2006#ifdef MODULE
2007    printk(version);
2008#endif
2009    return pci_register_driver(&hamachi_driver);
2010}
2011
2012static void __exit hamachi_exit (void)
2013{
2014    pci_unregister_driver(&hamachi_driver);
2015}
2016
2017
2018module_init(hamachi_init);
2019module_exit(hamachi_exit);
2020

Archive Download this file



interactive