1 /* starfire.c: Linux device driver for the Adaptec Starfire network adapter. */
2 /*
3 	Written 1998-2000 by Donald Becker.
4 
5 	Current maintainer is Ion Badulescu <ionut ta badula tod org>. Please
6 	send all bug reports to me, and not to Donald Becker, as this code
7 	has been heavily modified from Donald's original version.
8 
9 	This software may be used and distributed according to the terms of
10 	the GNU General Public License (GPL), incorporated herein by reference.
11 	Drivers based on or derived from this code fall under the GPL and must
12 	retain the authorship, copyright and license notice.  This file is not
13 	a complete program and may only be used when the entire operating
14 	system is licensed under the GPL.
15 
16 	The information below comes from Donald Becker's original driver:
17 
18 	The author may be reached as becker@scyld.com, or C/O
19 	Scyld Computing Corporation
20 	410 Severn Ave., Suite 210
21 	Annapolis MD 21403
22 
23 	Support and updates available at
24 	http://www.scyld.com/network/starfire.html
25 	[link no longer provides useful info -jgarzik]
26 
27 */
28 
29 #define DRV_NAME	"starfire"
30 #define DRV_VERSION	"2.1"
31 #define DRV_RELDATE	"July  6, 2008"
32 
33 #include <linux/interrupt.h>
34 #include <linux/module.h>
35 #include <linux/kernel.h>
36 #include <linux/pci.h>
37 #include <linux/netdevice.h>
38 #include <linux/etherdevice.h>
39 #include <linux/init.h>
40 #include <linux/delay.h>
41 #include <linux/crc32.h>
42 #include <linux/ethtool.h>
43 #include <linux/mii.h>
44 #include <linux/if_vlan.h>
45 #include <linux/mm.h>
46 #include <linux/firmware.h>
47 #include <asm/processor.h>		/* Processor type for cache alignment. */
48 #include <linux/uaccess.h>
49 #include <asm/io.h>
50 
51 /*
52  * The current frame processor firmware fails to checksum a fragment
53  * of length 1. If and when this is fixed, the #define below can be removed.
54  */
55 #define HAS_BROKEN_FIRMWARE
56 
57 /*
58  * If using the broken firmware, data must be padded to the next 32-bit boundary.
59  */
60 #ifdef HAS_BROKEN_FIRMWARE
61 #define PADDING_MASK 3
62 #endif
63 
64 /*
65  * Define this if using the driver with the zero-copy patch
66  */
67 #define ZEROCOPY
68 
69 #if IS_ENABLED(CONFIG_VLAN_8021Q)
70 #define VLAN_SUPPORT
71 #endif
72 
73 /* The user-configurable values.
74    These may be modified when a driver module is loaded.*/
75 
76 /* Used for tuning interrupt latency vs. overhead. */
77 static int intr_latency;
78 static int small_frames;
79 
80 static int debug = 1;			/* 1 normal messages, 0 quiet .. 7 verbose. */
81 static int max_interrupt_work = 20;
82 static int mtu;
83 /* Maximum number of multicast addresses to filter (vs. rx-all-multicast).
84    The Starfire has a 512 element hash table based on the Ethernet CRC. */
85 static const int multicast_filter_limit = 512;
86 /* Whether to do TCP/UDP checksums in hardware */
87 static int enable_hw_cksum = 1;
88 
89 #define PKT_BUF_SZ	1536		/* Size of each temporary Rx buffer.*/
90 /*
91  * Set the copy breakpoint for the copy-only-tiny-frames scheme.
92  * Setting to > 1518 effectively disables this feature.
93  *
94  * NOTE:
95  * The ia64 doesn't allow for unaligned loads even of integers being
96  * misaligned on a 2 byte boundary. Thus always force copying of
97  * packets as the starfire doesn't allow for misaligned DMAs ;-(
98  * 23/10/2000 - Jes
99  *
100  * The Alpha and the Sparc don't like unaligned loads, either. On Sparc64,
101  * at least, having unaligned frames leads to a rather serious performance
102  * penalty. -Ion
103  */
104 #if defined(__ia64__) || defined(__alpha__) || defined(__sparc__)
105 static int rx_copybreak = PKT_BUF_SZ;
106 #else
107 static int rx_copybreak /* = 0 */;
108 #endif
109 
110 /* PCI DMA burst size -- on sparc64 we want to force it to 64 bytes, on the others the default of 128 is fine. */
111 #ifdef __sparc__
112 #define DMA_BURST_SIZE 64
113 #else
114 #define DMA_BURST_SIZE 128
115 #endif
116 
117 /* Operational parameters that are set at compile time. */
118 
119 /* The "native" ring sizes are either 256 or 2048.
120    However in some modes a descriptor may be marked to wrap the ring earlier.
121 */
122 #define RX_RING_SIZE	256
123 #define TX_RING_SIZE	32
124 /* The completion queues are fixed at 1024 entries i.e. 4K or 8KB. */
125 #define DONE_Q_SIZE	1024
126 /* All queues must be aligned on a 256-byte boundary */
127 #define QUEUE_ALIGN	256
128 
129 #if RX_RING_SIZE > 256
130 #define RX_Q_ENTRIES Rx2048QEntries
131 #else
132 #define RX_Q_ENTRIES Rx256QEntries
133 #endif
134 
135 /* Operational parameters that usually are not changed. */
136 /* Time in jiffies before concluding the transmitter is hung. */
137 #define TX_TIMEOUT	(2 * HZ)
138 
139 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
140 /* 64-bit dma_addr_t */
141 #define ADDR_64BITS	/* This chip uses 64 bit addresses. */
142 #define netdrv_addr_t __le64
143 #define cpu_to_dma(x) cpu_to_le64(x)
144 #define dma_to_cpu(x) le64_to_cpu(x)
145 #define RX_DESC_Q_ADDR_SIZE RxDescQAddr64bit
146 #define TX_DESC_Q_ADDR_SIZE TxDescQAddr64bit
147 #define RX_COMPL_Q_ADDR_SIZE RxComplQAddr64bit
148 #define TX_COMPL_Q_ADDR_SIZE TxComplQAddr64bit
149 #define RX_DESC_ADDR_SIZE RxDescAddr64bit
150 #else  /* 32-bit dma_addr_t */
151 #define netdrv_addr_t __le32
152 #define cpu_to_dma(x) cpu_to_le32(x)
153 #define dma_to_cpu(x) le32_to_cpu(x)
154 #define RX_DESC_Q_ADDR_SIZE RxDescQAddr32bit
155 #define TX_DESC_Q_ADDR_SIZE TxDescQAddr32bit
156 #define RX_COMPL_Q_ADDR_SIZE RxComplQAddr32bit
157 #define TX_COMPL_Q_ADDR_SIZE TxComplQAddr32bit
158 #define RX_DESC_ADDR_SIZE RxDescAddr32bit
159 #endif
160 
161 #define skb_first_frag_len(skb)	skb_headlen(skb)
162 #define skb_num_frags(skb) (skb_shinfo(skb)->nr_frags + 1)
163 
164 /* Firmware names */
165 #define FIRMWARE_RX	"adaptec/starfire_rx.bin"
166 #define FIRMWARE_TX	"adaptec/starfire_tx.bin"
167 
168 /* These identify the driver base version and may not be removed. */
169 static const char version[] =
170 KERN_INFO "starfire.c:v1.03 7/26/2000  Written by Donald Becker <becker@scyld.com>\n"
171 " (unofficial 2.2/2.4 kernel port, version " DRV_VERSION ", " DRV_RELDATE ")\n";
172 
173 MODULE_AUTHOR("Donald Becker <becker@scyld.com>");
174 MODULE_DESCRIPTION("Adaptec Starfire Ethernet driver");
175 MODULE_LICENSE("GPL");
176 MODULE_VERSION(DRV_VERSION);
177 MODULE_FIRMWARE(FIRMWARE_RX);
178 MODULE_FIRMWARE(FIRMWARE_TX);
179 
180 module_param(max_interrupt_work, int, 0);
181 module_param(mtu, int, 0);
182 module_param(debug, int, 0);
183 module_param(rx_copybreak, int, 0);
184 module_param(intr_latency, int, 0);
185 module_param(small_frames, int, 0);
186 module_param(enable_hw_cksum, int, 0);
187 MODULE_PARM_DESC(max_interrupt_work, "Maximum events handled per interrupt");
188 MODULE_PARM_DESC(mtu, "MTU (all boards)");
189 MODULE_PARM_DESC(debug, "Debug level (0-6)");
190 MODULE_PARM_DESC(rx_copybreak, "Copy breakpoint for copy-only-tiny-frames");
191 MODULE_PARM_DESC(intr_latency, "Maximum interrupt latency, in microseconds");
192 MODULE_PARM_DESC(small_frames, "Maximum size of receive frames that bypass interrupt latency (0,64,128,256,512)");
193 MODULE_PARM_DESC(enable_hw_cksum, "Enable/disable hardware cksum support (0/1)");
194 
195 /*
196 				Theory of Operation
197 
198 I. Board Compatibility
199 
200 This driver is for the Adaptec 6915 "Starfire" 64 bit PCI Ethernet adapter.
201 
202 II. Board-specific settings
203 
204 III. Driver operation
205 
206 IIIa. Ring buffers
207 
208 The Starfire hardware uses multiple fixed-size descriptor queues/rings.  The
209 ring sizes are set fixed by the hardware, but may optionally be wrapped
210 earlier by the END bit in the descriptor.
211 This driver uses that hardware queue size for the Rx ring, where a large
212 number of entries has no ill effect beyond increases the potential backlog.
213 The Tx ring is wrapped with the END bit, since a large hardware Tx queue
214 disables the queue layer priority ordering and we have no mechanism to
215 utilize the hardware two-level priority queue.  When modifying the
216 RX/TX_RING_SIZE pay close attention to page sizes and the ring-empty warning
217 levels.
218 
219 IIIb/c. Transmit/Receive Structure
220 
221 See the Adaptec manual for the many possible structures, and options for
222 each structure.  There are far too many to document all of them here.
223 
224 For transmit this driver uses type 0/1 transmit descriptors (depending
225 on the 32/64 bitness of the architecture), and relies on automatic
226 minimum-length padding.  It does not use the completion queue
227 consumer index, but instead checks for non-zero status entries.
228 
229 For receive this driver uses type 2/3 receive descriptors.  The driver
230 allocates full frame size skbuffs for the Rx ring buffers, so all frames
231 should fit in a single descriptor.  The driver does not use the completion
232 queue consumer index, but instead checks for non-zero status entries.
233 
234 When an incoming frame is less than RX_COPYBREAK bytes long, a fresh skbuff
235 is allocated and the frame is copied to the new skbuff.  When the incoming
236 frame is larger, the skbuff is passed directly up the protocol stack.
237 Buffers consumed this way are replaced by newly allocated skbuffs in a later
238 phase of receive.
239 
240 A notable aspect of operation is that unaligned buffers are not permitted by
241 the Starfire hardware.  Thus the IP header at offset 14 in an ethernet frame
242 isn't longword aligned, which may cause problems on some machine
243 e.g. Alphas and IA64. For these architectures, the driver is forced to copy
244 the frame into a new skbuff unconditionally. Copied frames are put into the
245 skbuff at an offset of "+2", thus 16-byte aligning the IP header.
246 
247 IIId. Synchronization
248 
249 The driver runs as two independent, single-threaded flows of control.  One
250 is the send-packet routine, which enforces single-threaded use by the
251 dev->tbusy flag.  The other thread is the interrupt handler, which is single
252 threaded by the hardware and interrupt handling software.
253 
254 The send packet thread has partial control over the Tx ring and the netif_queue
255 status. If the number of free Tx slots in the ring falls below a certain number
256 (currently hardcoded to 4), it signals the upper layer to stop the queue.
257 
258 The interrupt handler has exclusive control over the Rx ring and records stats
259 from the Tx ring.  After reaping the stats, it marks the Tx queue entry as
260 empty by incrementing the dirty_tx mark. Iff the netif_queue is stopped and the
261 number of free Tx slow is above the threshold, it signals the upper layer to
262 restart the queue.
263 
264 IV. Notes
265 
266 IVb. References
267 
268 The Adaptec Starfire manuals, available only from Adaptec.
269 http://www.scyld.com/expert/100mbps.html
270 http://www.scyld.com/expert/NWay.html
271 
272 IVc. Errata
273 
274 - StopOnPerr is broken, don't enable
275 - Hardware ethernet padding exposes random data, perform software padding
276   instead (unverified -- works correctly for all the hardware I have)
277 
278 */
279 
280 
281 
282 enum chip_capability_flags {CanHaveMII=1, };
283 
284 enum chipset {
285 	CH_6915 = 0,
286 };
287 
288 static const struct pci_device_id starfire_pci_tbl[] = {
289 	{ PCI_VDEVICE(ADAPTEC, 0x6915), CH_6915 },
290 	{ 0, }
291 };
292 MODULE_DEVICE_TABLE(pci, starfire_pci_tbl);
293 
294 /* A chip capabilities table, matching the CH_xxx entries in xxx_pci_tbl[] above. */
295 static const struct chip_info {
296 	const char *name;
297 	int drv_flags;
298 } netdrv_tbl[] = {
299 	{ "Adaptec Starfire 6915", CanHaveMII },
300 };
301 
302 
303 /* Offsets to the device registers.
304    Unlike software-only systems, device drivers interact with complex hardware.
305    It's not useful to define symbolic names for every register bit in the
306    device.  The name can only partially document the semantics and make
307    the driver longer and more difficult to read.
308    In general, only the important configuration values or bits changed
309    multiple times should be defined symbolically.
310 */
311 enum register_offsets {
312 	PCIDeviceConfig=0x50040, GenCtrl=0x50070, IntrTimerCtrl=0x50074,
313 	IntrClear=0x50080, IntrStatus=0x50084, IntrEnable=0x50088,
314 	MIICtrl=0x52000, TxStationAddr=0x50120, EEPROMCtrl=0x51000,
315 	GPIOCtrl=0x5008C, TxDescCtrl=0x50090,
316 	TxRingPtr=0x50098, HiPriTxRingPtr=0x50094, /* Low and High priority. */
317 	TxRingHiAddr=0x5009C,		/* 64 bit address extension. */
318 	TxProducerIdx=0x500A0, TxConsumerIdx=0x500A4,
319 	TxThreshold=0x500B0,
320 	CompletionHiAddr=0x500B4, TxCompletionAddr=0x500B8,
321 	RxCompletionAddr=0x500BC, RxCompletionQ2Addr=0x500C0,
322 	CompletionQConsumerIdx=0x500C4, RxDMACtrl=0x500D0,
323 	RxDescQCtrl=0x500D4, RxDescQHiAddr=0x500DC, RxDescQAddr=0x500E0,
324 	RxDescQIdx=0x500E8, RxDMAStatus=0x500F0, RxFilterMode=0x500F4,
325 	TxMode=0x55000, VlanType=0x55064,
326 	PerfFilterTable=0x56000, HashTable=0x56100,
327 	TxGfpMem=0x58000, RxGfpMem=0x5a000,
328 };
329 
330 /*
331  * Bits in the interrupt status/mask registers.
332  * Warning: setting Intr[Ab]NormalSummary in the IntrEnable register
333  * enables all the interrupt sources that are or'ed into those status bits.
334  */
335 enum intr_status_bits {
336 	IntrLinkChange=0xf0000000, IntrStatsMax=0x08000000,
337 	IntrAbnormalSummary=0x02000000, IntrGeneralTimer=0x01000000,
338 	IntrSoftware=0x800000, IntrRxComplQ1Low=0x400000,
339 	IntrTxComplQLow=0x200000, IntrPCI=0x100000,
340 	IntrDMAErr=0x080000, IntrTxDataLow=0x040000,
341 	IntrRxComplQ2Low=0x020000, IntrRxDescQ1Low=0x010000,
342 	IntrNormalSummary=0x8000, IntrTxDone=0x4000,
343 	IntrTxDMADone=0x2000, IntrTxEmpty=0x1000,
344 	IntrEarlyRxQ2=0x0800, IntrEarlyRxQ1=0x0400,
345 	IntrRxQ2Done=0x0200, IntrRxQ1Done=0x0100,
346 	IntrRxGFPDead=0x80, IntrRxDescQ2Low=0x40,
347 	IntrNoTxCsum=0x20, IntrTxBadID=0x10,
348 	IntrHiPriTxBadID=0x08, IntrRxGfp=0x04,
349 	IntrTxGfp=0x02, IntrPCIPad=0x01,
350 	/* not quite bits */
351 	IntrRxDone=IntrRxQ2Done | IntrRxQ1Done,
352 	IntrRxEmpty=IntrRxDescQ1Low | IntrRxDescQ2Low,
353 	IntrNormalMask=0xff00, IntrAbnormalMask=0x3ff00fe,
354 };
355 
356 /* Bits in the RxFilterMode register. */
357 enum rx_mode_bits {
358 	AcceptBroadcast=0x04, AcceptAllMulticast=0x02, AcceptAll=0x01,
359 	AcceptMulticast=0x10, PerfectFilter=0x40, HashFilter=0x30,
360 	PerfectFilterVlan=0x80, MinVLANPrio=0xE000, VlanMode=0x0200,
361 	WakeupOnGFP=0x0800,
362 };
363 
364 /* Bits in the TxMode register */
365 enum tx_mode_bits {
366 	MiiSoftReset=0x8000, MIILoopback=0x4000,
367 	TxFlowEnable=0x0800, RxFlowEnable=0x0400,
368 	PadEnable=0x04, FullDuplex=0x02, HugeFrame=0x01,
369 };
370 
371 /* Bits in the TxDescCtrl register. */
372 enum tx_ctrl_bits {
373 	TxDescSpaceUnlim=0x00, TxDescSpace32=0x10, TxDescSpace64=0x20,
374 	TxDescSpace128=0x30, TxDescSpace256=0x40,
375 	TxDescType0=0x00, TxDescType1=0x01, TxDescType2=0x02,
376 	TxDescType3=0x03, TxDescType4=0x04,
377 	TxNoDMACompletion=0x08,
378 	TxDescQAddr64bit=0x80, TxDescQAddr32bit=0,
379 	TxHiPriFIFOThreshShift=24, TxPadLenShift=16,
380 	TxDMABurstSizeShift=8,
381 };
382 
383 /* Bits in the RxDescQCtrl register. */
384 enum rx_ctrl_bits {
385 	RxBufferLenShift=16, RxMinDescrThreshShift=0,
386 	RxPrefetchMode=0x8000, RxVariableQ=0x2000,
387 	Rx2048QEntries=0x4000, Rx256QEntries=0,
388 	RxDescAddr64bit=0x1000, RxDescAddr32bit=0,
389 	RxDescQAddr64bit=0x0100, RxDescQAddr32bit=0,
390 	RxDescSpace4=0x000, RxDescSpace8=0x100,
391 	RxDescSpace16=0x200, RxDescSpace32=0x300,
392 	RxDescSpace64=0x400, RxDescSpace128=0x500,
393 	RxConsumerWrEn=0x80,
394 };
395 
396 /* Bits in the RxDMACtrl register. */
397 enum rx_dmactrl_bits {
398 	RxReportBadFrames=0x80000000, RxDMAShortFrames=0x40000000,
399 	RxDMABadFrames=0x20000000, RxDMACrcErrorFrames=0x10000000,
400 	RxDMAControlFrame=0x08000000, RxDMAPauseFrame=0x04000000,
401 	RxChecksumIgnore=0, RxChecksumRejectTCPUDP=0x02000000,
402 	RxChecksumRejectTCPOnly=0x01000000,
403 	RxCompletionQ2Enable=0x800000,
404 	RxDMAQ2Disable=0, RxDMAQ2FPOnly=0x100000,
405 	RxDMAQ2SmallPkt=0x200000, RxDMAQ2HighPrio=0x300000,
406 	RxDMAQ2NonIP=0x400000,
407 	RxUseBackupQueue=0x080000, RxDMACRC=0x040000,
408 	RxEarlyIntThreshShift=12, RxHighPrioThreshShift=8,
409 	RxBurstSizeShift=0,
410 };
411 
412 /* Bits in the RxCompletionAddr register */
413 enum rx_compl_bits {
414 	RxComplQAddr64bit=0x80, RxComplQAddr32bit=0,
415 	RxComplProducerWrEn=0x40,
416 	RxComplType0=0x00, RxComplType1=0x10,
417 	RxComplType2=0x20, RxComplType3=0x30,
418 	RxComplThreshShift=0,
419 };
420 
421 /* Bits in the TxCompletionAddr register */
422 enum tx_compl_bits {
423 	TxComplQAddr64bit=0x80, TxComplQAddr32bit=0,
424 	TxComplProducerWrEn=0x40,
425 	TxComplIntrStatus=0x20,
426 	CommonQueueMode=0x10,
427 	TxComplThreshShift=0,
428 };
429 
430 /* Bits in the GenCtrl register */
431 enum gen_ctrl_bits {
432 	RxEnable=0x05, TxEnable=0x0a,
433 	RxGFPEnable=0x10, TxGFPEnable=0x20,
434 };
435 
436 /* Bits in the IntrTimerCtrl register */
437 enum intr_ctrl_bits {
438 	Timer10X=0x800, EnableIntrMasking=0x60, SmallFrameBypass=0x100,
439 	SmallFrame64=0, SmallFrame128=0x200, SmallFrame256=0x400, SmallFrame512=0x600,
440 	IntrLatencyMask=0x1f,
441 };
442 
443 /* The Rx and Tx buffer descriptors. */
444 struct starfire_rx_desc {
445 	netdrv_addr_t rxaddr;
446 };
447 enum rx_desc_bits {
448 	RxDescValid=1, RxDescEndRing=2,
449 };
450 
451 /* Completion queue entry. */
452 struct short_rx_done_desc {
453 	__le32 status;			/* Low 16 bits is length. */
454 };
455 struct basic_rx_done_desc {
456 	__le32 status;			/* Low 16 bits is length. */
457 	__le16 vlanid;
458 	__le16 status2;
459 };
460 struct csum_rx_done_desc {
461 	__le32 status;			/* Low 16 bits is length. */
462 	__le16 csum;			/* Partial checksum */
463 	__le16 status2;
464 };
465 struct full_rx_done_desc {
466 	__le32 status;			/* Low 16 bits is length. */
467 	__le16 status3;
468 	__le16 status2;
469 	__le16 vlanid;
470 	__le16 csum;			/* partial checksum */
471 	__le32 timestamp;
472 };
473 /* XXX: this is ugly and I'm not sure it's worth the trouble -Ion */
474 #ifdef VLAN_SUPPORT
475 typedef struct full_rx_done_desc rx_done_desc;
476 #define RxComplType RxComplType3
477 #else  /* not VLAN_SUPPORT */
478 typedef struct csum_rx_done_desc rx_done_desc;
479 #define RxComplType RxComplType2
480 #endif /* not VLAN_SUPPORT */
481 
482 enum rx_done_bits {
483 	RxOK=0x20000000, RxFIFOErr=0x10000000, RxBufQ2=0x08000000,
484 };
485 
486 /* Type 1 Tx descriptor. */
487 struct starfire_tx_desc_1 {
488 	__le32 status;			/* Upper bits are status, lower 16 length. */
489 	__le32 addr;
490 };
491 
492 /* Type 2 Tx descriptor. */
493 struct starfire_tx_desc_2 {
494 	__le32 status;			/* Upper bits are status, lower 16 length. */
495 	__le32 reserved;
496 	__le64 addr;
497 };
498 
499 #ifdef ADDR_64BITS
500 typedef struct starfire_tx_desc_2 starfire_tx_desc;
501 #define TX_DESC_TYPE TxDescType2
502 #else  /* not ADDR_64BITS */
503 typedef struct starfire_tx_desc_1 starfire_tx_desc;
504 #define TX_DESC_TYPE TxDescType1
505 #endif /* not ADDR_64BITS */
506 #define TX_DESC_SPACING TxDescSpaceUnlim
507 
508 enum tx_desc_bits {
509 	TxDescID=0xB0000000,
510 	TxCRCEn=0x01000000, TxDescIntr=0x08000000,
511 	TxRingWrap=0x04000000, TxCalTCP=0x02000000,
512 };
513 struct tx_done_desc {
514 	__le32 status;			/* timestamp, index. */
515 #if 0
516 	__le32 intrstatus;		/* interrupt status */
517 #endif
518 };
519 
520 struct rx_ring_info {
521 	struct sk_buff *skb;
522 	dma_addr_t mapping;
523 };
524 struct tx_ring_info {
525 	struct sk_buff *skb;
526 	dma_addr_t mapping;
527 	unsigned int used_slots;
528 };
529 
530 #define PHY_CNT		2
531 struct netdev_private {
532 	/* Descriptor rings first for alignment. */
533 	struct starfire_rx_desc *rx_ring;
534 	starfire_tx_desc *tx_ring;
535 	dma_addr_t rx_ring_dma;
536 	dma_addr_t tx_ring_dma;
537 	/* The addresses of rx/tx-in-place skbuffs. */
538 	struct rx_ring_info rx_info[RX_RING_SIZE];
539 	struct tx_ring_info tx_info[TX_RING_SIZE];
540 	/* Pointers to completion queues (full pages). */
541 	rx_done_desc *rx_done_q;
542 	dma_addr_t rx_done_q_dma;
543 	unsigned int rx_done;
544 	struct tx_done_desc *tx_done_q;
545 	dma_addr_t tx_done_q_dma;
546 	unsigned int tx_done;
547 	struct napi_struct napi;
548 	struct net_device *dev;
549 	struct pci_dev *pci_dev;
550 #ifdef VLAN_SUPPORT
551 	unsigned long active_vlans[BITS_TO_LONGS(VLAN_N_VID)];
552 #endif
553 	void *queue_mem;
554 	dma_addr_t queue_mem_dma;
555 	size_t queue_mem_size;
556 
557 	/* Frequently used values: keep some adjacent for cache effect. */
558 	spinlock_t lock;
559 	unsigned int cur_rx, dirty_rx;	/* Producer/consumer ring indices */
560 	unsigned int cur_tx, dirty_tx, reap_tx;
561 	unsigned int rx_buf_sz;		/* Based on MTU+slack. */
562 	/* These values keep track of the transceiver/media in use. */
563 	int speed100;			/* Set if speed == 100MBit. */
564 	u32 tx_mode;
565 	u32 intr_timer_ctrl;
566 	u8 tx_threshold;
567 	/* MII transceiver section. */
568 	struct mii_if_info mii_if;		/* MII lib hooks/info */
569 	int phy_cnt;			/* MII device addresses. */
570 	unsigned char phys[PHY_CNT];	/* MII device addresses. */
571 	void __iomem *base;
572 };
573 
574 
575 static int	mdio_read(struct net_device *dev, int phy_id, int location);
576 static void	mdio_write(struct net_device *dev, int phy_id, int location, int value);
577 static int	netdev_open(struct net_device *dev);
578 static void	check_duplex(struct net_device *dev);
579 static void	tx_timeout(struct net_device *dev);
580 static void	init_ring(struct net_device *dev);
581 static netdev_tx_t start_tx(struct sk_buff *skb, struct net_device *dev);
582 static irqreturn_t intr_handler(int irq, void *dev_instance);
583 static void	netdev_error(struct net_device *dev, int intr_status);
584 static int	__netdev_rx(struct net_device *dev, int *quota);
585 static int	netdev_poll(struct napi_struct *napi, int budget);
586 static void	refill_rx_ring(struct net_device *dev);
587 static void	netdev_error(struct net_device *dev, int intr_status);
588 static void	set_rx_mode(struct net_device *dev);
589 static struct net_device_stats *get_stats(struct net_device *dev);
590 static int	netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd);
591 static int	netdev_close(struct net_device *dev);
592 static void	netdev_media_change(struct net_device *dev);
593 static const struct ethtool_ops ethtool_ops;
594 
595 
596 #ifdef VLAN_SUPPORT
netdev_vlan_rx_add_vid(struct net_device * dev,__be16 proto,u16 vid)597 static int netdev_vlan_rx_add_vid(struct net_device *dev,
598 				  __be16 proto, u16 vid)
599 {
600 	struct netdev_private *np = netdev_priv(dev);
601 
602 	spin_lock(&np->lock);
603 	if (debug > 1)
604 		printk("%s: Adding vlanid %d to vlan filter\n", dev->name, vid);
605 	set_bit(vid, np->active_vlans);
606 	set_rx_mode(dev);
607 	spin_unlock(&np->lock);
608 
609 	return 0;
610 }
611 
netdev_vlan_rx_kill_vid(struct net_device * dev,__be16 proto,u16 vid)612 static int netdev_vlan_rx_kill_vid(struct net_device *dev,
613 				   __be16 proto, u16 vid)
614 {
615 	struct netdev_private *np = netdev_priv(dev);
616 
617 	spin_lock(&np->lock);
618 	if (debug > 1)
619 		printk("%s: removing vlanid %d from vlan filter\n", dev->name, vid);
620 	clear_bit(vid, np->active_vlans);
621 	set_rx_mode(dev);
622 	spin_unlock(&np->lock);
623 
624 	return 0;
625 }
626 #endif /* VLAN_SUPPORT */
627 
628 
629 static const struct net_device_ops netdev_ops = {
630 	.ndo_open		= netdev_open,
631 	.ndo_stop		= netdev_close,
632 	.ndo_start_xmit		= start_tx,
633 	.ndo_tx_timeout		= tx_timeout,
634 	.ndo_get_stats		= get_stats,
635 	.ndo_set_rx_mode	= set_rx_mode,
636 	.ndo_do_ioctl		= netdev_ioctl,
637 	.ndo_set_mac_address	= eth_mac_addr,
638 	.ndo_validate_addr	= eth_validate_addr,
639 #ifdef VLAN_SUPPORT
640 	.ndo_vlan_rx_add_vid	= netdev_vlan_rx_add_vid,
641 	.ndo_vlan_rx_kill_vid	= netdev_vlan_rx_kill_vid,
642 #endif
643 };
644 
starfire_init_one(struct pci_dev * pdev,const struct pci_device_id * ent)645 static int starfire_init_one(struct pci_dev *pdev,
646 			     const struct pci_device_id *ent)
647 {
648 	struct device *d = &pdev->dev;
649 	struct netdev_private *np;
650 	int i, irq, chip_idx = ent->driver_data;
651 	struct net_device *dev;
652 	long ioaddr;
653 	void __iomem *base;
654 	int drv_flags, io_size;
655 	int boguscnt;
656 
657 /* when built into the kernel, we only print version if device is found */
658 #ifndef MODULE
659 	static int printed_version;
660 	if (!printed_version++)
661 		printk(version);
662 #endif
663 
664 	if (pci_enable_device (pdev))
665 		return -EIO;
666 
667 	ioaddr = pci_resource_start(pdev, 0);
668 	io_size = pci_resource_len(pdev, 0);
669 	if (!ioaddr || ((pci_resource_flags(pdev, 0) & IORESOURCE_MEM) == 0)) {
670 		dev_err(d, "no PCI MEM resources, aborting\n");
671 		return -ENODEV;
672 	}
673 
674 	dev = alloc_etherdev(sizeof(*np));
675 	if (!dev)
676 		return -ENOMEM;
677 
678 	SET_NETDEV_DEV(dev, &pdev->dev);
679 
680 	irq = pdev->irq;
681 
682 	if (pci_request_regions (pdev, DRV_NAME)) {
683 		dev_err(d, "cannot reserve PCI resources, aborting\n");
684 		goto err_out_free_netdev;
685 	}
686 
687 	base = ioremap(ioaddr, io_size);
688 	if (!base) {
689 		dev_err(d, "cannot remap %#x @ %#lx, aborting\n",
690 			io_size, ioaddr);
691 		goto err_out_free_res;
692 	}
693 
694 	pci_set_master(pdev);
695 
696 	/* enable MWI -- it vastly improves Rx performance on sparc64 */
697 	pci_try_set_mwi(pdev);
698 
699 #ifdef ZEROCOPY
700 	/* Starfire can do TCP/UDP checksumming */
701 	if (enable_hw_cksum)
702 		dev->features |= NETIF_F_IP_CSUM | NETIF_F_SG;
703 #endif /* ZEROCOPY */
704 
705 #ifdef VLAN_SUPPORT
706 	dev->features |= NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
707 #endif /* VLAN_RX_KILL_VID */
708 #ifdef ADDR_64BITS
709 	dev->features |= NETIF_F_HIGHDMA;
710 #endif /* ADDR_64BITS */
711 
712 	/* Serial EEPROM reads are hidden by the hardware. */
713 	for (i = 0; i < 6; i++)
714 		dev->dev_addr[i] = readb(base + EEPROMCtrl + 20 - i);
715 
716 #if ! defined(final_version) /* Dump the EEPROM contents during development. */
717 	if (debug > 4)
718 		for (i = 0; i < 0x20; i++)
719 			printk("%2.2x%s",
720 			       (unsigned int)readb(base + EEPROMCtrl + i),
721 			       i % 16 != 15 ? " " : "\n");
722 #endif
723 
724 	/* Issue soft reset */
725 	writel(MiiSoftReset, base + TxMode);
726 	udelay(1000);
727 	writel(0, base + TxMode);
728 
729 	/* Reset the chip to erase previous misconfiguration. */
730 	writel(1, base + PCIDeviceConfig);
731 	boguscnt = 1000;
732 	while (--boguscnt > 0) {
733 		udelay(10);
734 		if ((readl(base + PCIDeviceConfig) & 1) == 0)
735 			break;
736 	}
737 	if (boguscnt == 0)
738 		printk("%s: chipset reset never completed!\n", dev->name);
739 	/* wait a little longer */
740 	udelay(1000);
741 
742 	np = netdev_priv(dev);
743 	np->dev = dev;
744 	np->base = base;
745 	spin_lock_init(&np->lock);
746 	pci_set_drvdata(pdev, dev);
747 
748 	np->pci_dev = pdev;
749 
750 	np->mii_if.dev = dev;
751 	np->mii_if.mdio_read = mdio_read;
752 	np->mii_if.mdio_write = mdio_write;
753 	np->mii_if.phy_id_mask = 0x1f;
754 	np->mii_if.reg_num_mask = 0x1f;
755 
756 	drv_flags = netdrv_tbl[chip_idx].drv_flags;
757 
758 	np->speed100 = 1;
759 
760 	/* timer resolution is 128 * 0.8us */
761 	np->intr_timer_ctrl = (((intr_latency * 10) / 1024) & IntrLatencyMask) |
762 		Timer10X | EnableIntrMasking;
763 
764 	if (small_frames > 0) {
765 		np->intr_timer_ctrl |= SmallFrameBypass;
766 		switch (small_frames) {
767 		case 1 ... 64:
768 			np->intr_timer_ctrl |= SmallFrame64;
769 			break;
770 		case 65 ... 128:
771 			np->intr_timer_ctrl |= SmallFrame128;
772 			break;
773 		case 129 ... 256:
774 			np->intr_timer_ctrl |= SmallFrame256;
775 			break;
776 		default:
777 			np->intr_timer_ctrl |= SmallFrame512;
778 			if (small_frames > 512)
779 				printk("Adjusting small_frames down to 512\n");
780 			break;
781 		}
782 	}
783 
784 	dev->netdev_ops = &netdev_ops;
785 	dev->watchdog_timeo = TX_TIMEOUT;
786 	dev->ethtool_ops = &ethtool_ops;
787 
788 	netif_napi_add(dev, &np->napi, netdev_poll, max_interrupt_work);
789 
790 	if (mtu)
791 		dev->mtu = mtu;
792 
793 	if (register_netdev(dev))
794 		goto err_out_cleardev;
795 
796 	printk(KERN_INFO "%s: %s at %p, %pM, IRQ %d.\n",
797 	       dev->name, netdrv_tbl[chip_idx].name, base,
798 	       dev->dev_addr, irq);
799 
800 	if (drv_flags & CanHaveMII) {
801 		int phy, phy_idx = 0;
802 		int mii_status;
803 		for (phy = 0; phy < 32 && phy_idx < PHY_CNT; phy++) {
804 			mdio_write(dev, phy, MII_BMCR, BMCR_RESET);
805 			msleep(100);
806 			boguscnt = 1000;
807 			while (--boguscnt > 0)
808 				if ((mdio_read(dev, phy, MII_BMCR) & BMCR_RESET) == 0)
809 					break;
810 			if (boguscnt == 0) {
811 				printk("%s: PHY#%d reset never completed!\n", dev->name, phy);
812 				continue;
813 			}
814 			mii_status = mdio_read(dev, phy, MII_BMSR);
815 			if (mii_status != 0) {
816 				np->phys[phy_idx++] = phy;
817 				np->mii_if.advertising = mdio_read(dev, phy, MII_ADVERTISE);
818 				printk(KERN_INFO "%s: MII PHY found at address %d, status "
819 					   "%#4.4x advertising %#4.4x.\n",
820 					   dev->name, phy, mii_status, np->mii_if.advertising);
821 				/* there can be only one PHY on-board */
822 				break;
823 			}
824 		}
825 		np->phy_cnt = phy_idx;
826 		if (np->phy_cnt > 0)
827 			np->mii_if.phy_id = np->phys[0];
828 		else
829 			memset(&np->mii_if, 0, sizeof(np->mii_if));
830 	}
831 
832 	printk(KERN_INFO "%s: scatter-gather and hardware TCP cksumming %s.\n",
833 	       dev->name, enable_hw_cksum ? "enabled" : "disabled");
834 	return 0;
835 
836 err_out_cleardev:
837 	iounmap(base);
838 err_out_free_res:
839 	pci_release_regions (pdev);
840 err_out_free_netdev:
841 	free_netdev(dev);
842 	return -ENODEV;
843 }
844 
845 
846 /* Read the MII Management Data I/O (MDIO) interfaces. */
mdio_read(struct net_device * dev,int phy_id,int location)847 static int mdio_read(struct net_device *dev, int phy_id, int location)
848 {
849 	struct netdev_private *np = netdev_priv(dev);
850 	void __iomem *mdio_addr = np->base + MIICtrl + (phy_id<<7) + (location<<2);
851 	int result, boguscnt=1000;
852 	/* ??? Should we add a busy-wait here? */
853 	do {
854 		result = readl(mdio_addr);
855 	} while ((result & 0xC0000000) != 0x80000000 && --boguscnt > 0);
856 	if (boguscnt == 0)
857 		return 0;
858 	if ((result & 0xffff) == 0xffff)
859 		return 0;
860 	return result & 0xffff;
861 }
862 
863 
mdio_write(struct net_device * dev,int phy_id,int location,int value)864 static void mdio_write(struct net_device *dev, int phy_id, int location, int value)
865 {
866 	struct netdev_private *np = netdev_priv(dev);
867 	void __iomem *mdio_addr = np->base + MIICtrl + (phy_id<<7) + (location<<2);
868 	writel(value, mdio_addr);
869 	/* The busy-wait will occur before a read. */
870 }
871 
872 
netdev_open(struct net_device * dev)873 static int netdev_open(struct net_device *dev)
874 {
875 	const struct firmware *fw_rx, *fw_tx;
876 	const __be32 *fw_rx_data, *fw_tx_data;
877 	struct netdev_private *np = netdev_priv(dev);
878 	void __iomem *ioaddr = np->base;
879 	const int irq = np->pci_dev->irq;
880 	int i, retval;
881 	size_t tx_size, rx_size;
882 	size_t tx_done_q_size, rx_done_q_size, tx_ring_size, rx_ring_size;
883 
884 	/* Do we ever need to reset the chip??? */
885 
886 	retval = request_irq(irq, intr_handler, IRQF_SHARED, dev->name, dev);
887 	if (retval)
888 		return retval;
889 
890 	/* Disable the Rx and Tx, and reset the chip. */
891 	writel(0, ioaddr + GenCtrl);
892 	writel(1, ioaddr + PCIDeviceConfig);
893 	if (debug > 1)
894 		printk(KERN_DEBUG "%s: netdev_open() irq %d.\n",
895 		       dev->name, irq);
896 
897 	/* Allocate the various queues. */
898 	if (!np->queue_mem) {
899 		tx_done_q_size = ((sizeof(struct tx_done_desc) * DONE_Q_SIZE + QUEUE_ALIGN - 1) / QUEUE_ALIGN) * QUEUE_ALIGN;
900 		rx_done_q_size = ((sizeof(rx_done_desc) * DONE_Q_SIZE + QUEUE_ALIGN - 1) / QUEUE_ALIGN) * QUEUE_ALIGN;
901 		tx_ring_size = ((sizeof(starfire_tx_desc) * TX_RING_SIZE + QUEUE_ALIGN - 1) / QUEUE_ALIGN) * QUEUE_ALIGN;
902 		rx_ring_size = sizeof(struct starfire_rx_desc) * RX_RING_SIZE;
903 		np->queue_mem_size = tx_done_q_size + rx_done_q_size + tx_ring_size + rx_ring_size;
904 		np->queue_mem = pci_alloc_consistent(np->pci_dev, np->queue_mem_size, &np->queue_mem_dma);
905 		if (np->queue_mem == NULL) {
906 			free_irq(irq, dev);
907 			return -ENOMEM;
908 		}
909 
910 		np->tx_done_q     = np->queue_mem;
911 		np->tx_done_q_dma = np->queue_mem_dma;
912 		np->rx_done_q     = (void *) np->tx_done_q + tx_done_q_size;
913 		np->rx_done_q_dma = np->tx_done_q_dma + tx_done_q_size;
914 		np->tx_ring       = (void *) np->rx_done_q + rx_done_q_size;
915 		np->tx_ring_dma   = np->rx_done_q_dma + rx_done_q_size;
916 		np->rx_ring       = (void *) np->tx_ring + tx_ring_size;
917 		np->rx_ring_dma   = np->tx_ring_dma + tx_ring_size;
918 	}
919 
920 	/* Start with no carrier, it gets adjusted later */
921 	netif_carrier_off(dev);
922 	init_ring(dev);
923 	/* Set the size of the Rx buffers. */
924 	writel((np->rx_buf_sz << RxBufferLenShift) |
925 	       (0 << RxMinDescrThreshShift) |
926 	       RxPrefetchMode | RxVariableQ |
927 	       RX_Q_ENTRIES |
928 	       RX_DESC_Q_ADDR_SIZE | RX_DESC_ADDR_SIZE |
929 	       RxDescSpace4,
930 	       ioaddr + RxDescQCtrl);
931 
932 	/* Set up the Rx DMA controller. */
933 	writel(RxChecksumIgnore |
934 	       (0 << RxEarlyIntThreshShift) |
935 	       (6 << RxHighPrioThreshShift) |
936 	       ((DMA_BURST_SIZE / 32) << RxBurstSizeShift),
937 	       ioaddr + RxDMACtrl);
938 
939 	/* Set Tx descriptor */
940 	writel((2 << TxHiPriFIFOThreshShift) |
941 	       (0 << TxPadLenShift) |
942 	       ((DMA_BURST_SIZE / 32) << TxDMABurstSizeShift) |
943 	       TX_DESC_Q_ADDR_SIZE |
944 	       TX_DESC_SPACING | TX_DESC_TYPE,
945 	       ioaddr + TxDescCtrl);
946 
947 	writel( (np->queue_mem_dma >> 16) >> 16, ioaddr + RxDescQHiAddr);
948 	writel( (np->queue_mem_dma >> 16) >> 16, ioaddr + TxRingHiAddr);
949 	writel( (np->queue_mem_dma >> 16) >> 16, ioaddr + CompletionHiAddr);
950 	writel(np->rx_ring_dma, ioaddr + RxDescQAddr);
951 	writel(np->tx_ring_dma, ioaddr + TxRingPtr);
952 
953 	writel(np->tx_done_q_dma, ioaddr + TxCompletionAddr);
954 	writel(np->rx_done_q_dma |
955 	       RxComplType |
956 	       (0 << RxComplThreshShift),
957 	       ioaddr + RxCompletionAddr);
958 
959 	if (debug > 1)
960 		printk(KERN_DEBUG "%s: Filling in the station address.\n", dev->name);
961 
962 	/* Fill both the Tx SA register and the Rx perfect filter. */
963 	for (i = 0; i < 6; i++)
964 		writeb(dev->dev_addr[i], ioaddr + TxStationAddr + 5 - i);
965 	/* The first entry is special because it bypasses the VLAN filter.
966 	   Don't use it. */
967 	writew(0, ioaddr + PerfFilterTable);
968 	writew(0, ioaddr + PerfFilterTable + 4);
969 	writew(0, ioaddr + PerfFilterTable + 8);
970 	for (i = 1; i < 16; i++) {
971 		__be16 *eaddrs = (__be16 *)dev->dev_addr;
972 		void __iomem *setup_frm = ioaddr + PerfFilterTable + i * 16;
973 		writew(be16_to_cpu(eaddrs[2]), setup_frm); setup_frm += 4;
974 		writew(be16_to_cpu(eaddrs[1]), setup_frm); setup_frm += 4;
975 		writew(be16_to_cpu(eaddrs[0]), setup_frm); setup_frm += 8;
976 	}
977 
978 	/* Initialize other registers. */
979 	/* Configure the PCI bus bursts and FIFO thresholds. */
980 	np->tx_mode = TxFlowEnable|RxFlowEnable|PadEnable;	/* modified when link is up. */
981 	writel(MiiSoftReset | np->tx_mode, ioaddr + TxMode);
982 	udelay(1000);
983 	writel(np->tx_mode, ioaddr + TxMode);
984 	np->tx_threshold = 4;
985 	writel(np->tx_threshold, ioaddr + TxThreshold);
986 
987 	writel(np->intr_timer_ctrl, ioaddr + IntrTimerCtrl);
988 
989 	napi_enable(&np->napi);
990 
991 	netif_start_queue(dev);
992 
993 	if (debug > 1)
994 		printk(KERN_DEBUG "%s: Setting the Rx and Tx modes.\n", dev->name);
995 	set_rx_mode(dev);
996 
997 	np->mii_if.advertising = mdio_read(dev, np->phys[0], MII_ADVERTISE);
998 	check_duplex(dev);
999 
1000 	/* Enable GPIO interrupts on link change */
1001 	writel(0x0f00ff00, ioaddr + GPIOCtrl);
1002 
1003 	/* Set the interrupt mask */
1004 	writel(IntrRxDone | IntrRxEmpty | IntrDMAErr |
1005 	       IntrTxDMADone | IntrStatsMax | IntrLinkChange |
1006 	       IntrRxGFPDead | IntrNoTxCsum | IntrTxBadID,
1007 	       ioaddr + IntrEnable);
1008 	/* Enable PCI interrupts. */
1009 	writel(0x00800000 | readl(ioaddr + PCIDeviceConfig),
1010 	       ioaddr + PCIDeviceConfig);
1011 
1012 #ifdef VLAN_SUPPORT
1013 	/* Set VLAN type to 802.1q */
1014 	writel(ETH_P_8021Q, ioaddr + VlanType);
1015 #endif /* VLAN_SUPPORT */
1016 
1017 	retval = request_firmware(&fw_rx, FIRMWARE_RX, &np->pci_dev->dev);
1018 	if (retval) {
1019 		printk(KERN_ERR "starfire: Failed to load firmware \"%s\"\n",
1020 		       FIRMWARE_RX);
1021 		goto out_init;
1022 	}
1023 	if (fw_rx->size % 4) {
1024 		printk(KERN_ERR "starfire: bogus length %zu in \"%s\"\n",
1025 		       fw_rx->size, FIRMWARE_RX);
1026 		retval = -EINVAL;
1027 		goto out_rx;
1028 	}
1029 	retval = request_firmware(&fw_tx, FIRMWARE_TX, &np->pci_dev->dev);
1030 	if (retval) {
1031 		printk(KERN_ERR "starfire: Failed to load firmware \"%s\"\n",
1032 		       FIRMWARE_TX);
1033 		goto out_rx;
1034 	}
1035 	if (fw_tx->size % 4) {
1036 		printk(KERN_ERR "starfire: bogus length %zu in \"%s\"\n",
1037 		       fw_tx->size, FIRMWARE_TX);
1038 		retval = -EINVAL;
1039 		goto out_tx;
1040 	}
1041 	fw_rx_data = (const __be32 *)&fw_rx->data[0];
1042 	fw_tx_data = (const __be32 *)&fw_tx->data[0];
1043 	rx_size = fw_rx->size / 4;
1044 	tx_size = fw_tx->size / 4;
1045 
1046 	/* Load Rx/Tx firmware into the frame processors */
1047 	for (i = 0; i < rx_size; i++)
1048 		writel(be32_to_cpup(&fw_rx_data[i]), ioaddr + RxGfpMem + i * 4);
1049 	for (i = 0; i < tx_size; i++)
1050 		writel(be32_to_cpup(&fw_tx_data[i]), ioaddr + TxGfpMem + i * 4);
1051 	if (enable_hw_cksum)
1052 		/* Enable the Rx and Tx units, and the Rx/Tx frame processors. */
1053 		writel(TxEnable|TxGFPEnable|RxEnable|RxGFPEnable, ioaddr + GenCtrl);
1054 	else
1055 		/* Enable the Rx and Tx units only. */
1056 		writel(TxEnable|RxEnable, ioaddr + GenCtrl);
1057 
1058 	if (debug > 1)
1059 		printk(KERN_DEBUG "%s: Done netdev_open().\n",
1060 		       dev->name);
1061 
1062 out_tx:
1063 	release_firmware(fw_tx);
1064 out_rx:
1065 	release_firmware(fw_rx);
1066 out_init:
1067 	if (retval)
1068 		netdev_close(dev);
1069 	return retval;
1070 }
1071 
1072 
check_duplex(struct net_device * dev)1073 static void check_duplex(struct net_device *dev)
1074 {
1075 	struct netdev_private *np = netdev_priv(dev);
1076 	u16 reg0;
1077 	int silly_count = 1000;
1078 
1079 	mdio_write(dev, np->phys[0], MII_ADVERTISE, np->mii_if.advertising);
1080 	mdio_write(dev, np->phys[0], MII_BMCR, BMCR_RESET);
1081 	udelay(500);
1082 	while (--silly_count && mdio_read(dev, np->phys[0], MII_BMCR) & BMCR_RESET)
1083 		/* do nothing */;
1084 	if (!silly_count) {
1085 		printk("%s: MII reset failed!\n", dev->name);
1086 		return;
1087 	}
1088 
1089 	reg0 = mdio_read(dev, np->phys[0], MII_BMCR);
1090 
1091 	if (!np->mii_if.force_media) {
1092 		reg0 |= BMCR_ANENABLE | BMCR_ANRESTART;
1093 	} else {
1094 		reg0 &= ~(BMCR_ANENABLE | BMCR_ANRESTART);
1095 		if (np->speed100)
1096 			reg0 |= BMCR_SPEED100;
1097 		if (np->mii_if.full_duplex)
1098 			reg0 |= BMCR_FULLDPLX;
1099 		printk(KERN_DEBUG "%s: Link forced to %sMbit %s-duplex\n",
1100 		       dev->name,
1101 		       np->speed100 ? "100" : "10",
1102 		       np->mii_if.full_duplex ? "full" : "half");
1103 	}
1104 	mdio_write(dev, np->phys[0], MII_BMCR, reg0);
1105 }
1106 
1107 
tx_timeout(struct net_device * dev)1108 static void tx_timeout(struct net_device *dev)
1109 {
1110 	struct netdev_private *np = netdev_priv(dev);
1111 	void __iomem *ioaddr = np->base;
1112 	int old_debug;
1113 
1114 	printk(KERN_WARNING "%s: Transmit timed out, status %#8.8x, "
1115 	       "resetting...\n", dev->name, (int) readl(ioaddr + IntrStatus));
1116 
1117 	/* Perhaps we should reinitialize the hardware here. */
1118 
1119 	/*
1120 	 * Stop and restart the interface.
1121 	 * Cheat and increase the debug level temporarily.
1122 	 */
1123 	old_debug = debug;
1124 	debug = 2;
1125 	netdev_close(dev);
1126 	netdev_open(dev);
1127 	debug = old_debug;
1128 
1129 	/* Trigger an immediate transmit demand. */
1130 
1131 	netif_trans_update(dev); /* prevent tx timeout */
1132 	dev->stats.tx_errors++;
1133 	netif_wake_queue(dev);
1134 }
1135 
1136 
1137 /* Initialize the Rx and Tx rings, along with various 'dev' bits. */
init_ring(struct net_device * dev)1138 static void init_ring(struct net_device *dev)
1139 {
1140 	struct netdev_private *np = netdev_priv(dev);
1141 	int i;
1142 
1143 	np->cur_rx = np->cur_tx = np->reap_tx = 0;
1144 	np->dirty_rx = np->dirty_tx = np->rx_done = np->tx_done = 0;
1145 
1146 	np->rx_buf_sz = (dev->mtu <= 1500 ? PKT_BUF_SZ : dev->mtu + 32);
1147 
1148 	/* Fill in the Rx buffers.  Handle allocation failure gracefully. */
1149 	for (i = 0; i < RX_RING_SIZE; i++) {
1150 		struct sk_buff *skb = netdev_alloc_skb(dev, np->rx_buf_sz);
1151 		np->rx_info[i].skb = skb;
1152 		if (skb == NULL)
1153 			break;
1154 		np->rx_info[i].mapping = pci_map_single(np->pci_dev, skb->data, np->rx_buf_sz, PCI_DMA_FROMDEVICE);
1155 		if (pci_dma_mapping_error(np->pci_dev,
1156 					  np->rx_info[i].mapping)) {
1157 			dev_kfree_skb(skb);
1158 			np->rx_info[i].skb = NULL;
1159 			break;
1160 		}
1161 		/* Grrr, we cannot offset to correctly align the IP header. */
1162 		np->rx_ring[i].rxaddr = cpu_to_dma(np->rx_info[i].mapping | RxDescValid);
1163 	}
1164 	writew(i - 1, np->base + RxDescQIdx);
1165 	np->dirty_rx = (unsigned int)(i - RX_RING_SIZE);
1166 
1167 	/* Clear the remainder of the Rx buffer ring. */
1168 	for (  ; i < RX_RING_SIZE; i++) {
1169 		np->rx_ring[i].rxaddr = 0;
1170 		np->rx_info[i].skb = NULL;
1171 		np->rx_info[i].mapping = 0;
1172 	}
1173 	/* Mark the last entry as wrapping the ring. */
1174 	np->rx_ring[RX_RING_SIZE - 1].rxaddr |= cpu_to_dma(RxDescEndRing);
1175 
1176 	/* Clear the completion rings. */
1177 	for (i = 0; i < DONE_Q_SIZE; i++) {
1178 		np->rx_done_q[i].status = 0;
1179 		np->tx_done_q[i].status = 0;
1180 	}
1181 
1182 	for (i = 0; i < TX_RING_SIZE; i++)
1183 		memset(&np->tx_info[i], 0, sizeof(np->tx_info[i]));
1184 }
1185 
1186 
start_tx(struct sk_buff * skb,struct net_device * dev)1187 static netdev_tx_t start_tx(struct sk_buff *skb, struct net_device *dev)
1188 {
1189 	struct netdev_private *np = netdev_priv(dev);
1190 	unsigned int entry;
1191 	unsigned int prev_tx;
1192 	u32 status;
1193 	int i, j;
1194 
1195 	/*
1196 	 * be cautious here, wrapping the queue has weird semantics
1197 	 * and we may not have enough slots even when it seems we do.
1198 	 */
1199 	if ((np->cur_tx - np->dirty_tx) + skb_num_frags(skb) * 2 > TX_RING_SIZE) {
1200 		netif_stop_queue(dev);
1201 		return NETDEV_TX_BUSY;
1202 	}
1203 
1204 #if defined(ZEROCOPY) && defined(HAS_BROKEN_FIRMWARE)
1205 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1206 		if (skb_padto(skb, (skb->len + PADDING_MASK) & ~PADDING_MASK))
1207 			return NETDEV_TX_OK;
1208 	}
1209 #endif /* ZEROCOPY && HAS_BROKEN_FIRMWARE */
1210 
1211 	prev_tx = np->cur_tx;
1212 	entry = np->cur_tx % TX_RING_SIZE;
1213 	for (i = 0; i < skb_num_frags(skb); i++) {
1214 		int wrap_ring = 0;
1215 		status = TxDescID;
1216 
1217 		if (i == 0) {
1218 			np->tx_info[entry].skb = skb;
1219 			status |= TxCRCEn;
1220 			if (entry >= TX_RING_SIZE - skb_num_frags(skb)) {
1221 				status |= TxRingWrap;
1222 				wrap_ring = 1;
1223 			}
1224 			if (np->reap_tx) {
1225 				status |= TxDescIntr;
1226 				np->reap_tx = 0;
1227 			}
1228 			if (skb->ip_summed == CHECKSUM_PARTIAL) {
1229 				status |= TxCalTCP;
1230 				dev->stats.tx_compressed++;
1231 			}
1232 			status |= skb_first_frag_len(skb) | (skb_num_frags(skb) << 16);
1233 
1234 			np->tx_info[entry].mapping =
1235 				pci_map_single(np->pci_dev, skb->data, skb_first_frag_len(skb), PCI_DMA_TODEVICE);
1236 		} else {
1237 			const skb_frag_t *this_frag = &skb_shinfo(skb)->frags[i - 1];
1238 			status |= skb_frag_size(this_frag);
1239 			np->tx_info[entry].mapping =
1240 				pci_map_single(np->pci_dev,
1241 					       skb_frag_address(this_frag),
1242 					       skb_frag_size(this_frag),
1243 					       PCI_DMA_TODEVICE);
1244 		}
1245 		if (pci_dma_mapping_error(np->pci_dev,
1246 					  np->tx_info[entry].mapping)) {
1247 			dev->stats.tx_dropped++;
1248 			goto err_out;
1249 		}
1250 
1251 		np->tx_ring[entry].addr = cpu_to_dma(np->tx_info[entry].mapping);
1252 		np->tx_ring[entry].status = cpu_to_le32(status);
1253 		if (debug > 3)
1254 			printk(KERN_DEBUG "%s: Tx #%d/#%d slot %d status %#8.8x.\n",
1255 			       dev->name, np->cur_tx, np->dirty_tx,
1256 			       entry, status);
1257 		if (wrap_ring) {
1258 			np->tx_info[entry].used_slots = TX_RING_SIZE - entry;
1259 			np->cur_tx += np->tx_info[entry].used_slots;
1260 			entry = 0;
1261 		} else {
1262 			np->tx_info[entry].used_slots = 1;
1263 			np->cur_tx += np->tx_info[entry].used_slots;
1264 			entry++;
1265 		}
1266 		/* scavenge the tx descriptors twice per TX_RING_SIZE */
1267 		if (np->cur_tx % (TX_RING_SIZE / 2) == 0)
1268 			np->reap_tx = 1;
1269 	}
1270 
1271 	/* Non-x86: explicitly flush descriptor cache lines here. */
1272 	/* Ensure all descriptors are written back before the transmit is
1273 	   initiated. - Jes */
1274 	wmb();
1275 
1276 	/* Update the producer index. */
1277 	writel(entry * (sizeof(starfire_tx_desc) / 8), np->base + TxProducerIdx);
1278 
1279 	/* 4 is arbitrary, but should be ok */
1280 	if ((np->cur_tx - np->dirty_tx) + 4 > TX_RING_SIZE)
1281 		netif_stop_queue(dev);
1282 
1283 	return NETDEV_TX_OK;
1284 
1285 err_out:
1286 	entry = prev_tx % TX_RING_SIZE;
1287 	np->tx_info[entry].skb = NULL;
1288 	if (i > 0) {
1289 		pci_unmap_single(np->pci_dev,
1290 				 np->tx_info[entry].mapping,
1291 				 skb_first_frag_len(skb),
1292 				 PCI_DMA_TODEVICE);
1293 		np->tx_info[entry].mapping = 0;
1294 		entry = (entry + np->tx_info[entry].used_slots) % TX_RING_SIZE;
1295 		for (j = 1; j < i; j++) {
1296 			pci_unmap_single(np->pci_dev,
1297 					 np->tx_info[entry].mapping,
1298 					 skb_frag_size(
1299 						&skb_shinfo(skb)->frags[j-1]),
1300 					 PCI_DMA_TODEVICE);
1301 			entry++;
1302 		}
1303 	}
1304 	dev_kfree_skb_any(skb);
1305 	np->cur_tx = prev_tx;
1306 	return NETDEV_TX_OK;
1307 }
1308 
1309 /* The interrupt handler does all of the Rx thread work and cleans up
1310    after the Tx thread. */
intr_handler(int irq,void * dev_instance)1311 static irqreturn_t intr_handler(int irq, void *dev_instance)
1312 {
1313 	struct net_device *dev = dev_instance;
1314 	struct netdev_private *np = netdev_priv(dev);
1315 	void __iomem *ioaddr = np->base;
1316 	int boguscnt = max_interrupt_work;
1317 	int consumer;
1318 	int tx_status;
1319 	int handled = 0;
1320 
1321 	do {
1322 		u32 intr_status = readl(ioaddr + IntrClear);
1323 
1324 		if (debug > 4)
1325 			printk(KERN_DEBUG "%s: Interrupt status %#8.8x.\n",
1326 			       dev->name, intr_status);
1327 
1328 		if (intr_status == 0 || intr_status == (u32) -1)
1329 			break;
1330 
1331 		handled = 1;
1332 
1333 		if (intr_status & (IntrRxDone | IntrRxEmpty)) {
1334 			u32 enable;
1335 
1336 			if (likely(napi_schedule_prep(&np->napi))) {
1337 				__napi_schedule(&np->napi);
1338 				enable = readl(ioaddr + IntrEnable);
1339 				enable &= ~(IntrRxDone | IntrRxEmpty);
1340 				writel(enable, ioaddr + IntrEnable);
1341 				/* flush PCI posting buffers */
1342 				readl(ioaddr + IntrEnable);
1343 			} else {
1344 				/* Paranoia check */
1345 				enable = readl(ioaddr + IntrEnable);
1346 				if (enable & (IntrRxDone | IntrRxEmpty)) {
1347 					printk(KERN_INFO
1348 					       "%s: interrupt while in poll!\n",
1349 					       dev->name);
1350 					enable &= ~(IntrRxDone | IntrRxEmpty);
1351 					writel(enable, ioaddr + IntrEnable);
1352 				}
1353 			}
1354 		}
1355 
1356 		/* Scavenge the skbuff list based on the Tx-done queue.
1357 		   There are redundant checks here that may be cleaned up
1358 		   after the driver has proven to be reliable. */
1359 		consumer = readl(ioaddr + TxConsumerIdx);
1360 		if (debug > 3)
1361 			printk(KERN_DEBUG "%s: Tx Consumer index is %d.\n",
1362 			       dev->name, consumer);
1363 
1364 		while ((tx_status = le32_to_cpu(np->tx_done_q[np->tx_done].status)) != 0) {
1365 			if (debug > 3)
1366 				printk(KERN_DEBUG "%s: Tx completion #%d entry %d is %#8.8x.\n",
1367 				       dev->name, np->dirty_tx, np->tx_done, tx_status);
1368 			if ((tx_status & 0xe0000000) == 0xa0000000) {
1369 				dev->stats.tx_packets++;
1370 			} else if ((tx_status & 0xe0000000) == 0x80000000) {
1371 				u16 entry = (tx_status & 0x7fff) / sizeof(starfire_tx_desc);
1372 				struct sk_buff *skb = np->tx_info[entry].skb;
1373 				np->tx_info[entry].skb = NULL;
1374 				pci_unmap_single(np->pci_dev,
1375 						 np->tx_info[entry].mapping,
1376 						 skb_first_frag_len(skb),
1377 						 PCI_DMA_TODEVICE);
1378 				np->tx_info[entry].mapping = 0;
1379 				np->dirty_tx += np->tx_info[entry].used_slots;
1380 				entry = (entry + np->tx_info[entry].used_slots) % TX_RING_SIZE;
1381 				{
1382 					int i;
1383 					for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1384 						pci_unmap_single(np->pci_dev,
1385 								 np->tx_info[entry].mapping,
1386 								 skb_frag_size(&skb_shinfo(skb)->frags[i]),
1387 								 PCI_DMA_TODEVICE);
1388 						np->dirty_tx++;
1389 						entry++;
1390 					}
1391 				}
1392 
1393 				dev_kfree_skb_irq(skb);
1394 			}
1395 			np->tx_done_q[np->tx_done].status = 0;
1396 			np->tx_done = (np->tx_done + 1) % DONE_Q_SIZE;
1397 		}
1398 		writew(np->tx_done, ioaddr + CompletionQConsumerIdx + 2);
1399 
1400 		if (netif_queue_stopped(dev) &&
1401 		    (np->cur_tx - np->dirty_tx + 4 < TX_RING_SIZE)) {
1402 			/* The ring is no longer full, wake the queue. */
1403 			netif_wake_queue(dev);
1404 		}
1405 
1406 		/* Stats overflow */
1407 		if (intr_status & IntrStatsMax)
1408 			get_stats(dev);
1409 
1410 		/* Media change interrupt. */
1411 		if (intr_status & IntrLinkChange)
1412 			netdev_media_change(dev);
1413 
1414 		/* Abnormal error summary/uncommon events handlers. */
1415 		if (intr_status & IntrAbnormalSummary)
1416 			netdev_error(dev, intr_status);
1417 
1418 		if (--boguscnt < 0) {
1419 			if (debug > 1)
1420 				printk(KERN_WARNING "%s: Too much work at interrupt, "
1421 				       "status=%#8.8x.\n",
1422 				       dev->name, intr_status);
1423 			break;
1424 		}
1425 	} while (1);
1426 
1427 	if (debug > 4)
1428 		printk(KERN_DEBUG "%s: exiting interrupt, status=%#8.8x.\n",
1429 		       dev->name, (int) readl(ioaddr + IntrStatus));
1430 	return IRQ_RETVAL(handled);
1431 }
1432 
1433 
1434 /*
1435  * This routine is logically part of the interrupt/poll handler, but separated
1436  * for clarity and better register allocation.
1437  */
__netdev_rx(struct net_device * dev,int * quota)1438 static int __netdev_rx(struct net_device *dev, int *quota)
1439 {
1440 	struct netdev_private *np = netdev_priv(dev);
1441 	u32 desc_status;
1442 	int retcode = 0;
1443 
1444 	/* If EOP is set on the next entry, it's a new packet. Send it up. */
1445 	while ((desc_status = le32_to_cpu(np->rx_done_q[np->rx_done].status)) != 0) {
1446 		struct sk_buff *skb;
1447 		u16 pkt_len;
1448 		int entry;
1449 		rx_done_desc *desc = &np->rx_done_q[np->rx_done];
1450 
1451 		if (debug > 4)
1452 			printk(KERN_DEBUG "  netdev_rx() status of %d was %#8.8x.\n", np->rx_done, desc_status);
1453 		if (!(desc_status & RxOK)) {
1454 			/* There was an error. */
1455 			if (debug > 2)
1456 				printk(KERN_DEBUG "  netdev_rx() Rx error was %#8.8x.\n", desc_status);
1457 			dev->stats.rx_errors++;
1458 			if (desc_status & RxFIFOErr)
1459 				dev->stats.rx_fifo_errors++;
1460 			goto next_rx;
1461 		}
1462 
1463 		if (*quota <= 0) {	/* out of rx quota */
1464 			retcode = 1;
1465 			goto out;
1466 		}
1467 		(*quota)--;
1468 
1469 		pkt_len = desc_status;	/* Implicitly Truncate */
1470 		entry = (desc_status >> 16) & 0x7ff;
1471 
1472 		if (debug > 4)
1473 			printk(KERN_DEBUG "  netdev_rx() normal Rx pkt length %d, quota %d.\n", pkt_len, *quota);
1474 		/* Check if the packet is long enough to accept without copying
1475 		   to a minimally-sized skbuff. */
1476 		if (pkt_len < rx_copybreak &&
1477 		    (skb = netdev_alloc_skb(dev, pkt_len + 2)) != NULL) {
1478 			skb_reserve(skb, 2);	/* 16 byte align the IP header */
1479 			pci_dma_sync_single_for_cpu(np->pci_dev,
1480 						    np->rx_info[entry].mapping,
1481 						    pkt_len, PCI_DMA_FROMDEVICE);
1482 			skb_copy_to_linear_data(skb, np->rx_info[entry].skb->data, pkt_len);
1483 			pci_dma_sync_single_for_device(np->pci_dev,
1484 						       np->rx_info[entry].mapping,
1485 						       pkt_len, PCI_DMA_FROMDEVICE);
1486 			skb_put(skb, pkt_len);
1487 		} else {
1488 			pci_unmap_single(np->pci_dev, np->rx_info[entry].mapping, np->rx_buf_sz, PCI_DMA_FROMDEVICE);
1489 			skb = np->rx_info[entry].skb;
1490 			skb_put(skb, pkt_len);
1491 			np->rx_info[entry].skb = NULL;
1492 			np->rx_info[entry].mapping = 0;
1493 		}
1494 #ifndef final_version			/* Remove after testing. */
1495 		/* You will want this info for the initial debug. */
1496 		if (debug > 5) {
1497 			printk(KERN_DEBUG "  Rx data %pM %pM %2.2x%2.2x.\n",
1498 			       skb->data, skb->data + 6,
1499 			       skb->data[12], skb->data[13]);
1500 		}
1501 #endif
1502 
1503 		skb->protocol = eth_type_trans(skb, dev);
1504 #ifdef VLAN_SUPPORT
1505 		if (debug > 4)
1506 			printk(KERN_DEBUG "  netdev_rx() status2 of %d was %#4.4x.\n", np->rx_done, le16_to_cpu(desc->status2));
1507 #endif
1508 		if (le16_to_cpu(desc->status2) & 0x0100) {
1509 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1510 			dev->stats.rx_compressed++;
1511 		}
1512 		/*
1513 		 * This feature doesn't seem to be working, at least
1514 		 * with the two firmware versions I have. If the GFP sees
1515 		 * an IP fragment, it either ignores it completely, or reports
1516 		 * "bad checksum" on it.
1517 		 *
1518 		 * Maybe I missed something -- corrections are welcome.
1519 		 * Until then, the printk stays. :-) -Ion
1520 		 */
1521 		else if (le16_to_cpu(desc->status2) & 0x0040) {
1522 			skb->ip_summed = CHECKSUM_COMPLETE;
1523 			skb->csum = le16_to_cpu(desc->csum);
1524 			printk(KERN_DEBUG "%s: checksum_hw, status2 = %#x\n", dev->name, le16_to_cpu(desc->status2));
1525 		}
1526 #ifdef VLAN_SUPPORT
1527 		if (le16_to_cpu(desc->status2) & 0x0200) {
1528 			u16 vlid = le16_to_cpu(desc->vlanid);
1529 
1530 			if (debug > 4) {
1531 				printk(KERN_DEBUG "  netdev_rx() vlanid = %d\n",
1532 				       vlid);
1533 			}
1534 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlid);
1535 		}
1536 #endif /* VLAN_SUPPORT */
1537 		netif_receive_skb(skb);
1538 		dev->stats.rx_packets++;
1539 
1540 	next_rx:
1541 		np->cur_rx++;
1542 		desc->status = 0;
1543 		np->rx_done = (np->rx_done + 1) % DONE_Q_SIZE;
1544 	}
1545 
1546 	if (*quota == 0) {	/* out of rx quota */
1547 		retcode = 1;
1548 		goto out;
1549 	}
1550 	writew(np->rx_done, np->base + CompletionQConsumerIdx);
1551 
1552  out:
1553 	refill_rx_ring(dev);
1554 	if (debug > 5)
1555 		printk(KERN_DEBUG "  exiting netdev_rx(): %d, status of %d was %#8.8x.\n",
1556 		       retcode, np->rx_done, desc_status);
1557 	return retcode;
1558 }
1559 
netdev_poll(struct napi_struct * napi,int budget)1560 static int netdev_poll(struct napi_struct *napi, int budget)
1561 {
1562 	struct netdev_private *np = container_of(napi, struct netdev_private, napi);
1563 	struct net_device *dev = np->dev;
1564 	u32 intr_status;
1565 	void __iomem *ioaddr = np->base;
1566 	int quota = budget;
1567 
1568 	do {
1569 		writel(IntrRxDone | IntrRxEmpty, ioaddr + IntrClear);
1570 
1571 		if (__netdev_rx(dev, &quota))
1572 			goto out;
1573 
1574 		intr_status = readl(ioaddr + IntrStatus);
1575 	} while (intr_status & (IntrRxDone | IntrRxEmpty));
1576 
1577 	napi_complete(napi);
1578 	intr_status = readl(ioaddr + IntrEnable);
1579 	intr_status |= IntrRxDone | IntrRxEmpty;
1580 	writel(intr_status, ioaddr + IntrEnable);
1581 
1582  out:
1583 	if (debug > 5)
1584 		printk(KERN_DEBUG "  exiting netdev_poll(): %d.\n",
1585 		       budget - quota);
1586 
1587 	/* Restart Rx engine if stopped. */
1588 	return budget - quota;
1589 }
1590 
refill_rx_ring(struct net_device * dev)1591 static void refill_rx_ring(struct net_device *dev)
1592 {
1593 	struct netdev_private *np = netdev_priv(dev);
1594 	struct sk_buff *skb;
1595 	int entry = -1;
1596 
1597 	/* Refill the Rx ring buffers. */
1598 	for (; np->cur_rx - np->dirty_rx > 0; np->dirty_rx++) {
1599 		entry = np->dirty_rx % RX_RING_SIZE;
1600 		if (np->rx_info[entry].skb == NULL) {
1601 			skb = netdev_alloc_skb(dev, np->rx_buf_sz);
1602 			np->rx_info[entry].skb = skb;
1603 			if (skb == NULL)
1604 				break;	/* Better luck next round. */
1605 			np->rx_info[entry].mapping =
1606 				pci_map_single(np->pci_dev, skb->data, np->rx_buf_sz, PCI_DMA_FROMDEVICE);
1607 			if (pci_dma_mapping_error(np->pci_dev,
1608 						np->rx_info[entry].mapping)) {
1609 				dev_kfree_skb(skb);
1610 				np->rx_info[entry].skb = NULL;
1611 				break;
1612 			}
1613 			np->rx_ring[entry].rxaddr =
1614 				cpu_to_dma(np->rx_info[entry].mapping | RxDescValid);
1615 		}
1616 		if (entry == RX_RING_SIZE - 1)
1617 			np->rx_ring[entry].rxaddr |= cpu_to_dma(RxDescEndRing);
1618 	}
1619 	if (entry >= 0)
1620 		writew(entry, np->base + RxDescQIdx);
1621 }
1622 
1623 
netdev_media_change(struct net_device * dev)1624 static void netdev_media_change(struct net_device *dev)
1625 {
1626 	struct netdev_private *np = netdev_priv(dev);
1627 	void __iomem *ioaddr = np->base;
1628 	u16 reg0, reg1, reg4, reg5;
1629 	u32 new_tx_mode;
1630 	u32 new_intr_timer_ctrl;
1631 
1632 	/* reset status first */
1633 	mdio_read(dev, np->phys[0], MII_BMCR);
1634 	mdio_read(dev, np->phys[0], MII_BMSR);
1635 
1636 	reg0 = mdio_read(dev, np->phys[0], MII_BMCR);
1637 	reg1 = mdio_read(dev, np->phys[0], MII_BMSR);
1638 
1639 	if (reg1 & BMSR_LSTATUS) {
1640 		/* link is up */
1641 		if (reg0 & BMCR_ANENABLE) {
1642 			/* autonegotiation is enabled */
1643 			reg4 = mdio_read(dev, np->phys[0], MII_ADVERTISE);
1644 			reg5 = mdio_read(dev, np->phys[0], MII_LPA);
1645 			if (reg4 & ADVERTISE_100FULL && reg5 & LPA_100FULL) {
1646 				np->speed100 = 1;
1647 				np->mii_if.full_duplex = 1;
1648 			} else if (reg4 & ADVERTISE_100HALF && reg5 & LPA_100HALF) {
1649 				np->speed100 = 1;
1650 				np->mii_if.full_duplex = 0;
1651 			} else if (reg4 & ADVERTISE_10FULL && reg5 & LPA_10FULL) {
1652 				np->speed100 = 0;
1653 				np->mii_if.full_duplex = 1;
1654 			} else {
1655 				np->speed100 = 0;
1656 				np->mii_if.full_duplex = 0;
1657 			}
1658 		} else {
1659 			/* autonegotiation is disabled */
1660 			if (reg0 & BMCR_SPEED100)
1661 				np->speed100 = 1;
1662 			else
1663 				np->speed100 = 0;
1664 			if (reg0 & BMCR_FULLDPLX)
1665 				np->mii_if.full_duplex = 1;
1666 			else
1667 				np->mii_if.full_duplex = 0;
1668 		}
1669 		netif_carrier_on(dev);
1670 		printk(KERN_DEBUG "%s: Link is up, running at %sMbit %s-duplex\n",
1671 		       dev->name,
1672 		       np->speed100 ? "100" : "10",
1673 		       np->mii_if.full_duplex ? "full" : "half");
1674 
1675 		new_tx_mode = np->tx_mode & ~FullDuplex;	/* duplex setting */
1676 		if (np->mii_if.full_duplex)
1677 			new_tx_mode |= FullDuplex;
1678 		if (np->tx_mode != new_tx_mode) {
1679 			np->tx_mode = new_tx_mode;
1680 			writel(np->tx_mode | MiiSoftReset, ioaddr + TxMode);
1681 			udelay(1000);
1682 			writel(np->tx_mode, ioaddr + TxMode);
1683 		}
1684 
1685 		new_intr_timer_ctrl = np->intr_timer_ctrl & ~Timer10X;
1686 		if (np->speed100)
1687 			new_intr_timer_ctrl |= Timer10X;
1688 		if (np->intr_timer_ctrl != new_intr_timer_ctrl) {
1689 			np->intr_timer_ctrl = new_intr_timer_ctrl;
1690 			writel(new_intr_timer_ctrl, ioaddr + IntrTimerCtrl);
1691 		}
1692 	} else {
1693 		netif_carrier_off(dev);
1694 		printk(KERN_DEBUG "%s: Link is down\n", dev->name);
1695 	}
1696 }
1697 
1698 
netdev_error(struct net_device * dev,int intr_status)1699 static void netdev_error(struct net_device *dev, int intr_status)
1700 {
1701 	struct netdev_private *np = netdev_priv(dev);
1702 
1703 	/* Came close to underrunning the Tx FIFO, increase threshold. */
1704 	if (intr_status & IntrTxDataLow) {
1705 		if (np->tx_threshold <= PKT_BUF_SZ / 16) {
1706 			writel(++np->tx_threshold, np->base + TxThreshold);
1707 			printk(KERN_NOTICE "%s: PCI bus congestion, increasing Tx FIFO threshold to %d bytes\n",
1708 			       dev->name, np->tx_threshold * 16);
1709 		} else
1710 			printk(KERN_WARNING "%s: PCI Tx underflow -- adapter is probably malfunctioning\n", dev->name);
1711 	}
1712 	if (intr_status & IntrRxGFPDead) {
1713 		dev->stats.rx_fifo_errors++;
1714 		dev->stats.rx_errors++;
1715 	}
1716 	if (intr_status & (IntrNoTxCsum | IntrDMAErr)) {
1717 		dev->stats.tx_fifo_errors++;
1718 		dev->stats.tx_errors++;
1719 	}
1720 	if ((intr_status & ~(IntrNormalMask | IntrAbnormalSummary | IntrLinkChange | IntrStatsMax | IntrTxDataLow | IntrRxGFPDead | IntrNoTxCsum | IntrPCIPad)) && debug)
1721 		printk(KERN_ERR "%s: Something Wicked happened! %#8.8x.\n",
1722 		       dev->name, intr_status);
1723 }
1724 
1725 
get_stats(struct net_device * dev)1726 static struct net_device_stats *get_stats(struct net_device *dev)
1727 {
1728 	struct netdev_private *np = netdev_priv(dev);
1729 	void __iomem *ioaddr = np->base;
1730 
1731 	/* This adapter architecture needs no SMP locks. */
1732 	dev->stats.tx_bytes = readl(ioaddr + 0x57010);
1733 	dev->stats.rx_bytes = readl(ioaddr + 0x57044);
1734 	dev->stats.tx_packets = readl(ioaddr + 0x57000);
1735 	dev->stats.tx_aborted_errors =
1736 		readl(ioaddr + 0x57024) + readl(ioaddr + 0x57028);
1737 	dev->stats.tx_window_errors = readl(ioaddr + 0x57018);
1738 	dev->stats.collisions =
1739 		readl(ioaddr + 0x57004) + readl(ioaddr + 0x57008);
1740 
1741 	/* The chip only need report frame silently dropped. */
1742 	dev->stats.rx_dropped += readw(ioaddr + RxDMAStatus);
1743 	writew(0, ioaddr + RxDMAStatus);
1744 	dev->stats.rx_crc_errors = readl(ioaddr + 0x5703C);
1745 	dev->stats.rx_frame_errors = readl(ioaddr + 0x57040);
1746 	dev->stats.rx_length_errors = readl(ioaddr + 0x57058);
1747 	dev->stats.rx_missed_errors = readl(ioaddr + 0x5707C);
1748 
1749 	return &dev->stats;
1750 }
1751 
1752 #ifdef VLAN_SUPPORT
set_vlan_mode(struct netdev_private * np)1753 static u32 set_vlan_mode(struct netdev_private *np)
1754 {
1755 	u32 ret = VlanMode;
1756 	u16 vid;
1757 	void __iomem *filter_addr = np->base + HashTable + 8;
1758 	int vlan_count = 0;
1759 
1760 	for_each_set_bit(vid, np->active_vlans, VLAN_N_VID) {
1761 		if (vlan_count == 32)
1762 			break;
1763 		writew(vid, filter_addr);
1764 		filter_addr += 16;
1765 		vlan_count++;
1766 	}
1767 	if (vlan_count == 32) {
1768 		ret |= PerfectFilterVlan;
1769 		while (vlan_count < 32) {
1770 			writew(0, filter_addr);
1771 			filter_addr += 16;
1772 			vlan_count++;
1773 		}
1774 	}
1775 	return ret;
1776 }
1777 #endif /* VLAN_SUPPORT */
1778 
set_rx_mode(struct net_device * dev)1779 static void set_rx_mode(struct net_device *dev)
1780 {
1781 	struct netdev_private *np = netdev_priv(dev);
1782 	void __iomem *ioaddr = np->base;
1783 	u32 rx_mode = MinVLANPrio;
1784 	struct netdev_hw_addr *ha;
1785 	int i;
1786 
1787 #ifdef VLAN_SUPPORT
1788 	rx_mode |= set_vlan_mode(np);
1789 #endif /* VLAN_SUPPORT */
1790 
1791 	if (dev->flags & IFF_PROMISC) {	/* Set promiscuous. */
1792 		rx_mode |= AcceptAll;
1793 	} else if ((netdev_mc_count(dev) > multicast_filter_limit) ||
1794 		   (dev->flags & IFF_ALLMULTI)) {
1795 		/* Too many to match, or accept all multicasts. */
1796 		rx_mode |= AcceptBroadcast|AcceptAllMulticast|PerfectFilter;
1797 	} else if (netdev_mc_count(dev) <= 14) {
1798 		/* Use the 16 element perfect filter, skip first two entries. */
1799 		void __iomem *filter_addr = ioaddr + PerfFilterTable + 2 * 16;
1800 		__be16 *eaddrs;
1801 		netdev_for_each_mc_addr(ha, dev) {
1802 			eaddrs = (__be16 *) ha->addr;
1803 			writew(be16_to_cpu(eaddrs[2]), filter_addr); filter_addr += 4;
1804 			writew(be16_to_cpu(eaddrs[1]), filter_addr); filter_addr += 4;
1805 			writew(be16_to_cpu(eaddrs[0]), filter_addr); filter_addr += 8;
1806 		}
1807 		eaddrs = (__be16 *)dev->dev_addr;
1808 		i = netdev_mc_count(dev) + 2;
1809 		while (i++ < 16) {
1810 			writew(be16_to_cpu(eaddrs[0]), filter_addr); filter_addr += 4;
1811 			writew(be16_to_cpu(eaddrs[1]), filter_addr); filter_addr += 4;
1812 			writew(be16_to_cpu(eaddrs[2]), filter_addr); filter_addr += 8;
1813 		}
1814 		rx_mode |= AcceptBroadcast|PerfectFilter;
1815 	} else {
1816 		/* Must use a multicast hash table. */
1817 		void __iomem *filter_addr;
1818 		__be16 *eaddrs;
1819 		__le16 mc_filter[32] __attribute__ ((aligned(sizeof(long))));	/* Multicast hash filter */
1820 
1821 		memset(mc_filter, 0, sizeof(mc_filter));
1822 		netdev_for_each_mc_addr(ha, dev) {
1823 			/* The chip uses the upper 9 CRC bits
1824 			   as index into the hash table */
1825 			int bit_nr = ether_crc_le(ETH_ALEN, ha->addr) >> 23;
1826 			__le32 *fptr = (__le32 *) &mc_filter[(bit_nr >> 4) & ~1];
1827 
1828 			*fptr |= cpu_to_le32(1 << (bit_nr & 31));
1829 		}
1830 		/* Clear the perfect filter list, skip first two entries. */
1831 		filter_addr = ioaddr + PerfFilterTable + 2 * 16;
1832 		eaddrs = (__be16 *)dev->dev_addr;
1833 		for (i = 2; i < 16; i++) {
1834 			writew(be16_to_cpu(eaddrs[0]), filter_addr); filter_addr += 4;
1835 			writew(be16_to_cpu(eaddrs[1]), filter_addr); filter_addr += 4;
1836 			writew(be16_to_cpu(eaddrs[2]), filter_addr); filter_addr += 8;
1837 		}
1838 		for (filter_addr = ioaddr + HashTable, i = 0; i < 32; filter_addr+= 16, i++)
1839 			writew(mc_filter[i], filter_addr);
1840 		rx_mode |= AcceptBroadcast|PerfectFilter|HashFilter;
1841 	}
1842 	writel(rx_mode, ioaddr + RxFilterMode);
1843 }
1844 
check_if_running(struct net_device * dev)1845 static int check_if_running(struct net_device *dev)
1846 {
1847 	if (!netif_running(dev))
1848 		return -EINVAL;
1849 	return 0;
1850 }
1851 
get_drvinfo(struct net_device * dev,struct ethtool_drvinfo * info)1852 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1853 {
1854 	struct netdev_private *np = netdev_priv(dev);
1855 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1856 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1857 	strlcpy(info->bus_info, pci_name(np->pci_dev), sizeof(info->bus_info));
1858 }
1859 
get_link_ksettings(struct net_device * dev,struct ethtool_link_ksettings * cmd)1860 static int get_link_ksettings(struct net_device *dev,
1861 			      struct ethtool_link_ksettings *cmd)
1862 {
1863 	struct netdev_private *np = netdev_priv(dev);
1864 	spin_lock_irq(&np->lock);
1865 	mii_ethtool_get_link_ksettings(&np->mii_if, cmd);
1866 	spin_unlock_irq(&np->lock);
1867 	return 0;
1868 }
1869 
set_link_ksettings(struct net_device * dev,const struct ethtool_link_ksettings * cmd)1870 static int set_link_ksettings(struct net_device *dev,
1871 			      const struct ethtool_link_ksettings *cmd)
1872 {
1873 	struct netdev_private *np = netdev_priv(dev);
1874 	int res;
1875 	spin_lock_irq(&np->lock);
1876 	res = mii_ethtool_set_link_ksettings(&np->mii_if, cmd);
1877 	spin_unlock_irq(&np->lock);
1878 	check_duplex(dev);
1879 	return res;
1880 }
1881 
nway_reset(struct net_device * dev)1882 static int nway_reset(struct net_device *dev)
1883 {
1884 	struct netdev_private *np = netdev_priv(dev);
1885 	return mii_nway_restart(&np->mii_if);
1886 }
1887 
get_link(struct net_device * dev)1888 static u32 get_link(struct net_device *dev)
1889 {
1890 	struct netdev_private *np = netdev_priv(dev);
1891 	return mii_link_ok(&np->mii_if);
1892 }
1893 
get_msglevel(struct net_device * dev)1894 static u32 get_msglevel(struct net_device *dev)
1895 {
1896 	return debug;
1897 }
1898 
set_msglevel(struct net_device * dev,u32 val)1899 static void set_msglevel(struct net_device *dev, u32 val)
1900 {
1901 	debug = val;
1902 }
1903 
1904 static const struct ethtool_ops ethtool_ops = {
1905 	.begin = check_if_running,
1906 	.get_drvinfo = get_drvinfo,
1907 	.nway_reset = nway_reset,
1908 	.get_link = get_link,
1909 	.get_msglevel = get_msglevel,
1910 	.set_msglevel = set_msglevel,
1911 	.get_link_ksettings = get_link_ksettings,
1912 	.set_link_ksettings = set_link_ksettings,
1913 };
1914 
netdev_ioctl(struct net_device * dev,struct ifreq * rq,int cmd)1915 static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
1916 {
1917 	struct netdev_private *np = netdev_priv(dev);
1918 	struct mii_ioctl_data *data = if_mii(rq);
1919 	int rc;
1920 
1921 	if (!netif_running(dev))
1922 		return -EINVAL;
1923 
1924 	spin_lock_irq(&np->lock);
1925 	rc = generic_mii_ioctl(&np->mii_if, data, cmd, NULL);
1926 	spin_unlock_irq(&np->lock);
1927 
1928 	if ((cmd == SIOCSMIIREG) && (data->phy_id == np->phys[0]))
1929 		check_duplex(dev);
1930 
1931 	return rc;
1932 }
1933 
netdev_close(struct net_device * dev)1934 static int netdev_close(struct net_device *dev)
1935 {
1936 	struct netdev_private *np = netdev_priv(dev);
1937 	void __iomem *ioaddr = np->base;
1938 	int i;
1939 
1940 	netif_stop_queue(dev);
1941 
1942 	napi_disable(&np->napi);
1943 
1944 	if (debug > 1) {
1945 		printk(KERN_DEBUG "%s: Shutting down ethercard, Intr status %#8.8x.\n",
1946 			   dev->name, (int) readl(ioaddr + IntrStatus));
1947 		printk(KERN_DEBUG "%s: Queue pointers were Tx %d / %d, Rx %d / %d.\n",
1948 		       dev->name, np->cur_tx, np->dirty_tx,
1949 		       np->cur_rx, np->dirty_rx);
1950 	}
1951 
1952 	/* Disable interrupts by clearing the interrupt mask. */
1953 	writel(0, ioaddr + IntrEnable);
1954 
1955 	/* Stop the chip's Tx and Rx processes. */
1956 	writel(0, ioaddr + GenCtrl);
1957 	readl(ioaddr + GenCtrl);
1958 
1959 	if (debug > 5) {
1960 		printk(KERN_DEBUG"  Tx ring at %#llx:\n",
1961 		       (long long) np->tx_ring_dma);
1962 		for (i = 0; i < 8 /* TX_RING_SIZE is huge! */; i++)
1963 			printk(KERN_DEBUG " #%d desc. %#8.8x %#llx -> %#8.8x.\n",
1964 			       i, le32_to_cpu(np->tx_ring[i].status),
1965 			       (long long) dma_to_cpu(np->tx_ring[i].addr),
1966 			       le32_to_cpu(np->tx_done_q[i].status));
1967 		printk(KERN_DEBUG "  Rx ring at %#llx -> %p:\n",
1968 		       (long long) np->rx_ring_dma, np->rx_done_q);
1969 		if (np->rx_done_q)
1970 			for (i = 0; i < 8 /* RX_RING_SIZE */; i++) {
1971 				printk(KERN_DEBUG " #%d desc. %#llx -> %#8.8x\n",
1972 				       i, (long long) dma_to_cpu(np->rx_ring[i].rxaddr), le32_to_cpu(np->rx_done_q[i].status));
1973 		}
1974 	}
1975 
1976 	free_irq(np->pci_dev->irq, dev);
1977 
1978 	/* Free all the skbuffs in the Rx queue. */
1979 	for (i = 0; i < RX_RING_SIZE; i++) {
1980 		np->rx_ring[i].rxaddr = cpu_to_dma(0xBADF00D0); /* An invalid address. */
1981 		if (np->rx_info[i].skb != NULL) {
1982 			pci_unmap_single(np->pci_dev, np->rx_info[i].mapping, np->rx_buf_sz, PCI_DMA_FROMDEVICE);
1983 			dev_kfree_skb(np->rx_info[i].skb);
1984 		}
1985 		np->rx_info[i].skb = NULL;
1986 		np->rx_info[i].mapping = 0;
1987 	}
1988 	for (i = 0; i < TX_RING_SIZE; i++) {
1989 		struct sk_buff *skb = np->tx_info[i].skb;
1990 		if (skb == NULL)
1991 			continue;
1992 		pci_unmap_single(np->pci_dev,
1993 				 np->tx_info[i].mapping,
1994 				 skb_first_frag_len(skb), PCI_DMA_TODEVICE);
1995 		np->tx_info[i].mapping = 0;
1996 		dev_kfree_skb(skb);
1997 		np->tx_info[i].skb = NULL;
1998 	}
1999 
2000 	return 0;
2001 }
2002 
2003 #ifdef CONFIG_PM
starfire_suspend(struct pci_dev * pdev,pm_message_t state)2004 static int starfire_suspend(struct pci_dev *pdev, pm_message_t state)
2005 {
2006 	struct net_device *dev = pci_get_drvdata(pdev);
2007 
2008 	if (netif_running(dev)) {
2009 		netif_device_detach(dev);
2010 		netdev_close(dev);
2011 	}
2012 
2013 	pci_save_state(pdev);
2014 	pci_set_power_state(pdev, pci_choose_state(pdev,state));
2015 
2016 	return 0;
2017 }
2018 
starfire_resume(struct pci_dev * pdev)2019 static int starfire_resume(struct pci_dev *pdev)
2020 {
2021 	struct net_device *dev = pci_get_drvdata(pdev);
2022 
2023 	pci_set_power_state(pdev, PCI_D0);
2024 	pci_restore_state(pdev);
2025 
2026 	if (netif_running(dev)) {
2027 		netdev_open(dev);
2028 		netif_device_attach(dev);
2029 	}
2030 
2031 	return 0;
2032 }
2033 #endif /* CONFIG_PM */
2034 
2035 
starfire_remove_one(struct pci_dev * pdev)2036 static void starfire_remove_one(struct pci_dev *pdev)
2037 {
2038 	struct net_device *dev = pci_get_drvdata(pdev);
2039 	struct netdev_private *np = netdev_priv(dev);
2040 
2041 	BUG_ON(!dev);
2042 
2043 	unregister_netdev(dev);
2044 
2045 	if (np->queue_mem)
2046 		pci_free_consistent(pdev, np->queue_mem_size, np->queue_mem, np->queue_mem_dma);
2047 
2048 
2049 	/* XXX: add wakeup code -- requires firmware for MagicPacket */
2050 	pci_set_power_state(pdev, PCI_D3hot);	/* go to sleep in D3 mode */
2051 	pci_disable_device(pdev);
2052 
2053 	iounmap(np->base);
2054 	pci_release_regions(pdev);
2055 
2056 	free_netdev(dev);			/* Will also free np!! */
2057 }
2058 
2059 
2060 static struct pci_driver starfire_driver = {
2061 	.name		= DRV_NAME,
2062 	.probe		= starfire_init_one,
2063 	.remove		= starfire_remove_one,
2064 #ifdef CONFIG_PM
2065 	.suspend	= starfire_suspend,
2066 	.resume		= starfire_resume,
2067 #endif /* CONFIG_PM */
2068 	.id_table	= starfire_pci_tbl,
2069 };
2070 
2071 
starfire_init(void)2072 static int __init starfire_init (void)
2073 {
2074 /* when a module, this is printed whether or not devices are found in probe */
2075 #ifdef MODULE
2076 	printk(version);
2077 
2078 	printk(KERN_INFO DRV_NAME ": polling (NAPI) enabled\n");
2079 #endif
2080 
2081 	BUILD_BUG_ON(sizeof(dma_addr_t) != sizeof(netdrv_addr_t));
2082 
2083 	return pci_register_driver(&starfire_driver);
2084 }
2085 
2086 
starfire_cleanup(void)2087 static void __exit starfire_cleanup (void)
2088 {
2089 	pci_unregister_driver (&starfire_driver);
2090 }
2091 
2092 
2093 module_init(starfire_init);
2094 module_exit(starfire_cleanup);
2095 
2096 
2097 /*
2098  * Local variables:
2099  *  c-basic-offset: 8
2100  *  tab-width: 8
2101  * End:
2102  */
2103