diff options
author | Jeff Kirsher <jeffrey.t.kirsher@intel.com> | 2011-06-14 17:02:57 -0400 |
---|---|---|
committer | Jeff Kirsher <jeffrey.t.kirsher@intel.com> | 2011-08-12 06:41:14 -0400 |
commit | 554f4ffd3b2cd88e42007d069bc519563d0b459f (patch) | |
tree | 3701f6d17551bc275d5622721c149521327faaea /drivers/net/ethernet/packetengines | |
parent | b544dbac41218fd015ac79455cfc1e57736e9b0c (diff) |
hamachi/yellowfin: Move the packet engine drivers
Move the packet engine drivers to drivers/net/ethernet/packetengines/
and the necessary Kconfig and Makefile changes.
CC: Donald Becker <becker@scyld.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Diffstat (limited to 'drivers/net/ethernet/packetengines')
-rw-r--r-- | drivers/net/ethernet/packetengines/Kconfig | 46 | ||||
-rw-r--r-- | drivers/net/ethernet/packetengines/Makefile | 6 | ||||
-rw-r--r-- | drivers/net/ethernet/packetengines/hamachi.c | 1952 | ||||
-rw-r--r-- | drivers/net/ethernet/packetengines/yellowfin.c | 1430 |
4 files changed, 3434 insertions, 0 deletions
diff --git a/drivers/net/ethernet/packetengines/Kconfig b/drivers/net/ethernet/packetengines/Kconfig new file mode 100644 index 000000000000..4add1db20f1e --- /dev/null +++ b/drivers/net/ethernet/packetengines/Kconfig | |||
@@ -0,0 +1,46 @@ | |||
1 | # | ||
2 | # Packet engine device configuration | ||
3 | # | ||
4 | |||
5 | config NET_PACKET_ENGINE | ||
6 | bool "Packet Engine devices" | ||
7 | depends on PCI | ||
8 | ---help--- | ||
9 | If you have a network (Ethernet) card belonging to this class, say Y | ||
10 | and read the Ethernet-HOWTO, available from | ||
11 | <http://www.tldp.org/docs.html#howto>. | ||
12 | |||
13 | Note that the answer to this question doesn't directly affect the | ||
14 | kernel: saying N will just cause the configurator to skip all | ||
15 | the questions about packet engine devices. If you say Y, you will | ||
16 | be asked for your specific card in the following questions. | ||
17 | |||
18 | if NET_PACKET_ENGINE | ||
19 | |||
20 | config HAMACHI | ||
21 | tristate "Packet Engines Hamachi GNIC-II support" | ||
22 | depends on PCI | ||
23 | select MII | ||
24 | ---help--- | ||
25 | If you have a Gigabit Ethernet card of this type, say Y and read | ||
26 | the Ethernet-HOWTO, available from | ||
27 | <http://www.tldp.org/docs.html#howto>. | ||
28 | |||
29 | To compile this driver as a module, choose M here. The module will be | ||
30 | called hamachi. | ||
31 | |||
32 | config YELLOWFIN | ||
33 | tristate "Packet Engines Yellowfin Gigabit-NIC support (EXPERIMENTAL)" | ||
34 | depends on PCI && EXPERIMENTAL | ||
35 | select CRC32 | ||
36 | ---help--- | ||
37 | Say Y here if you have a Packet Engines G-NIC PCI Gigabit Ethernet | ||
38 | adapter or the SYM53C885 Ethernet controller. The Gigabit adapter is | ||
39 | used by the Beowulf Linux cluster project. See | ||
40 | <http://cesdis.gsfc.nasa.gov/linux/drivers/yellowfin.html> for more | ||
41 | information about this driver in particular and Beowulf in general. | ||
42 | |||
43 | To compile this driver as a module, choose M here: the module | ||
44 | will be called yellowfin. This is recommended. | ||
45 | |||
46 | endif # NET_PACKET_ENGINE | ||
diff --git a/drivers/net/ethernet/packetengines/Makefile b/drivers/net/ethernet/packetengines/Makefile new file mode 100644 index 000000000000..995ccd077d0c --- /dev/null +++ b/drivers/net/ethernet/packetengines/Makefile | |||
@@ -0,0 +1,6 @@ | |||
1 | # | ||
2 | # Makefile for the Packet Engine network device drivers. | ||
3 | # | ||
4 | |||
5 | obj-$(CONFIG_HAMACHI) += hamachi.o | ||
6 | obj-$(CONFIG_YELLOWFIN) += yellowfin.o | ||
diff --git a/drivers/net/ethernet/packetengines/hamachi.c b/drivers/net/ethernet/packetengines/hamachi.c new file mode 100644 index 000000000000..c274b3d77eb5 --- /dev/null +++ b/drivers/net/ethernet/packetengines/hamachi.c | |||
@@ -0,0 +1,1952 @@ | |||
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 | |||
36 | static 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. */ | ||
40 | static int max_interrupt_work = 40; | ||
41 | static 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 | */ | ||
46 | static int max_rx_latency = 0x11; | ||
47 | static int max_rx_gap = 0x05; | ||
48 | static int min_rx_pkt = 0x18; | ||
49 | static int max_tx_latency = 0x00; | ||
50 | static int max_tx_gap = 0x00; | ||
51 | static 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 | */ | ||
57 | static 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 | */ | ||
63 | static 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 */ | ||
86 | static int options[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1}; | ||
87 | static 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 | */ | ||
103 | static int rx_params[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1}; | ||
104 | static 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 | * 3/10/1999 Pete Wyckoff <wyckoff@ca.sandia.gov> | ||
136 | */ | ||
137 | #define RX_CHECKSUM | ||
138 | |||
139 | /* Operational parameters that usually are not changed. */ | ||
140 | /* Time in jiffies before concluding the transmitter is hung. */ | ||
141 | #define TX_TIMEOUT (5*HZ) | ||
142 | |||
143 | #include <linux/capability.h> | ||
144 | #include <linux/module.h> | ||
145 | #include <linux/kernel.h> | ||
146 | #include <linux/string.h> | ||
147 | #include <linux/timer.h> | ||
148 | #include <linux/time.h> | ||
149 | #include <linux/errno.h> | ||
150 | #include <linux/ioport.h> | ||
151 | #include <linux/interrupt.h> | ||
152 | #include <linux/pci.h> | ||
153 | #include <linux/init.h> | ||
154 | #include <linux/ethtool.h> | ||
155 | #include <linux/mii.h> | ||
156 | #include <linux/netdevice.h> | ||
157 | #include <linux/etherdevice.h> | ||
158 | #include <linux/skbuff.h> | ||
159 | #include <linux/ip.h> | ||
160 | #include <linux/delay.h> | ||
161 | #include <linux/bitops.h> | ||
162 | |||
163 | #include <asm/uaccess.h> | ||
164 | #include <asm/processor.h> /* Processor type for cache alignment. */ | ||
165 | #include <asm/io.h> | ||
166 | #include <asm/unaligned.h> | ||
167 | #include <asm/cache.h> | ||
168 | |||
169 | static const char version[] __devinitconst = | ||
170 | KERN_INFO DRV_NAME ".c:v" DRV_VERSION " " DRV_RELDATE " Written by Donald Becker\n" | ||
171 | " Some modifications by Eric kasten <kasten@nscl.msu.edu>\n" | ||
172 | " Further modifications by Keith Underwood <keithu@parl.clemson.edu>\n"; | ||
173 | |||
174 | |||
175 | /* IP_MF appears to be only defined in <netinet/ip.h>, however, | ||
176 | we need it for hardware checksumming support. FYI... some of | ||
177 | the definitions in <netinet/ip.h> conflict/duplicate those in | ||
178 | other linux headers causing many compiler warnings. | ||
179 | */ | ||
180 | #ifndef IP_MF | ||
181 | #define IP_MF 0x2000 /* IP more frags from <netinet/ip.h> */ | ||
182 | #endif | ||
183 | |||
184 | /* Define IP_OFFSET to be IPOPT_OFFSET */ | ||
185 | #ifndef IP_OFFSET | ||
186 | #ifdef IPOPT_OFFSET | ||
187 | #define IP_OFFSET IPOPT_OFFSET | ||
188 | #else | ||
189 | #define IP_OFFSET 2 | ||
190 | #endif | ||
191 | #endif | ||
192 | |||
193 | #define RUN_AT(x) (jiffies + (x)) | ||
194 | |||
195 | #ifndef ADDRLEN | ||
196 | #define ADDRLEN 32 | ||
197 | #endif | ||
198 | |||
199 | /* Condensed bus+endian portability operations. */ | ||
200 | #if ADDRLEN == 64 | ||
201 | #define cpu_to_leXX(addr) cpu_to_le64(addr) | ||
202 | #define leXX_to_cpu(addr) le64_to_cpu(addr) | ||
203 | #else | ||
204 | #define cpu_to_leXX(addr) cpu_to_le32(addr) | ||
205 | #define leXX_to_cpu(addr) le32_to_cpu(addr) | ||
206 | #endif | ||
207 | |||
208 | |||
209 | /* | ||
210 | Theory of Operation | ||
211 | |||
212 | I. Board Compatibility | ||
213 | |||
214 | This device driver is designed for the Packet Engines "Hamachi" | ||
215 | Gigabit Ethernet chip. The only PCA currently supported is the GNIC-II 64-bit | ||
216 | 66Mhz PCI card. | ||
217 | |||
218 | II. Board-specific settings | ||
219 | |||
220 | No jumpers exist on the board. The chip supports software correction of | ||
221 | various motherboard wiring errors, however this driver does not support | ||
222 | that feature. | ||
223 | |||
224 | III. Driver operation | ||
225 | |||
226 | IIIa. Ring buffers | ||
227 | |||
228 | The Hamachi uses a typical descriptor based bus-master architecture. | ||
229 | The descriptor list is similar to that used by the Digital Tulip. | ||
230 | This driver uses two statically allocated fixed-size descriptor lists | ||
231 | formed into rings by a branch from the final descriptor to the beginning of | ||
232 | the list. The ring sizes are set at compile time by RX/TX_RING_SIZE. | ||
233 | |||
234 | This driver uses a zero-copy receive and transmit scheme similar my other | ||
235 | network drivers. | ||
236 | The driver allocates full frame size skbuffs for the Rx ring buffers at | ||
237 | open() time and passes the skb->data field to the Hamachi as receive data | ||
238 | buffers. When an incoming frame is less than RX_COPYBREAK bytes long, | ||
239 | a fresh skbuff is allocated and the frame is copied to the new skbuff. | ||
240 | When the incoming frame is larger, the skbuff is passed directly up the | ||
241 | protocol stack and replaced by a newly allocated skbuff. | ||
242 | |||
243 | The RX_COPYBREAK value is chosen to trade-off the memory wasted by | ||
244 | using a full-sized skbuff for small frames vs. the copying costs of larger | ||
245 | frames. Gigabit cards are typically used on generously configured machines | ||
246 | and the underfilled buffers have negligible impact compared to the benefit of | ||
247 | a single allocation size, so the default value of zero results in never | ||
248 | copying packets. | ||
249 | |||
250 | IIIb/c. Transmit/Receive Structure | ||
251 | |||
252 | The Rx and Tx descriptor structure are straight-forward, with no historical | ||
253 | baggage that must be explained. Unlike the awkward DBDMA structure, there | ||
254 | are no unused fields or option bits that had only one allowable setting. | ||
255 | |||
256 | Two details should be noted about the descriptors: The chip supports both 32 | ||
257 | bit and 64 bit address structures, and the length field is overwritten on | ||
258 | the receive descriptors. The descriptor length is set in the control word | ||
259 | for each channel. The development driver uses 32 bit addresses only, however | ||
260 | 64 bit addresses may be enabled for 64 bit architectures e.g. the Alpha. | ||
261 | |||
262 | IIId. Synchronization | ||
263 | |||
264 | This driver is very similar to my other network drivers. | ||
265 | The driver runs as two independent, single-threaded flows of control. One | ||
266 | is the send-packet routine, which enforces single-threaded use by the | ||
267 | dev->tbusy flag. The other thread is the interrupt handler, which is single | ||
268 | threaded by the hardware and other software. | ||
269 | |||
270 | The send packet thread has partial control over the Tx ring and 'dev->tbusy' | ||
271 | flag. It sets the tbusy flag whenever it's queuing a Tx packet. If the next | ||
272 | queue slot is empty, it clears the tbusy flag when finished otherwise it sets | ||
273 | the 'hmp->tx_full' flag. | ||
274 | |||
275 | The interrupt handler has exclusive control over the Rx ring and records stats | ||
276 | from the Tx ring. After reaping the stats, it marks the Tx queue entry as | ||
277 | empty by incrementing the dirty_tx mark. Iff the 'hmp->tx_full' flag is set, it | ||
278 | clears both the tx_full and tbusy flags. | ||
279 | |||
280 | IV. Notes | ||
281 | |||
282 | Thanks to Kim Stearns of Packet Engines for providing a pair of GNIC-II boards. | ||
283 | |||
284 | IVb. References | ||
285 | |||
286 | Hamachi Engineering Design Specification, 5/15/97 | ||
287 | (Note: This version was marked "Confidential".) | ||
288 | |||
289 | IVc. Errata | ||
290 | |||
291 | None noted. | ||
292 | |||
293 | V. Recent Changes | ||
294 | |||
295 | 01/15/1999 EPK Enlargement of the TX and RX ring sizes. This appears | ||
296 | to help avoid some stall conditions -- this needs further research. | ||
297 | |||
298 | 01/15/1999 EPK Creation of the hamachi_tx function. This function cleans | ||
299 | the Tx ring and is called from hamachi_start_xmit (this used to be | ||
300 | called from hamachi_interrupt but it tends to delay execution of the | ||
301 | interrupt handler and thus reduce bandwidth by reducing the latency | ||
302 | between hamachi_rx()'s). Notably, some modification has been made so | ||
303 | that the cleaning loop checks only to make sure that the DescOwn bit | ||
304 | isn't set in the status flag since the card is not required | ||
305 | to set the entire flag to zero after processing. | ||
306 | |||
307 | 01/15/1999 EPK In the hamachi_start_tx function, the Tx ring full flag is | ||
308 | checked before attempting to add a buffer to the ring. If the ring is full | ||
309 | an attempt is made to free any dirty buffers and thus find space for | ||
310 | the new buffer or the function returns non-zero which should case the | ||
311 | scheduler to reschedule the buffer later. | ||
312 | |||
313 | 01/15/1999 EPK Some adjustments were made to the chip initialization. | ||
314 | End-to-end flow control should now be fully active and the interrupt | ||
315 | algorithm vars have been changed. These could probably use further tuning. | ||
316 | |||
317 | 01/15/1999 EPK Added the max_{rx,tx}_latency options. These are used to | ||
318 | set the rx and tx latencies for the Hamachi interrupts. If you're having | ||
319 | problems with network stalls, try setting these to higher values. | ||
320 | Valid values are 0x00 through 0xff. | ||
321 | |||
322 | 01/15/1999 EPK In general, the overall bandwidth has increased and | ||
323 | latencies are better (sometimes by a factor of 2). Stalls are rare at | ||
324 | this point, however there still appears to be a bug somewhere between the | ||
325 | hardware and driver. TCP checksum errors under load also appear to be | ||
326 | eliminated at this point. | ||
327 | |||
328 | 01/18/1999 EPK Ensured that the DescEndRing bit was being set on both the | ||
329 | Rx and Tx rings. This appears to have been affecting whether a particular | ||
330 | peer-to-peer connection would hang under high load. I believe the Rx | ||
331 | rings was typically getting set correctly, but the Tx ring wasn't getting | ||
332 | the DescEndRing bit set during initialization. ??? Does this mean the | ||
333 | hamachi card is using the DescEndRing in processing even if a particular | ||
334 | slot isn't in use -- hypothetically, the card might be searching the | ||
335 | entire Tx ring for slots with the DescOwn bit set and then processing | ||
336 | them. If the DescEndRing bit isn't set, then it might just wander off | ||
337 | through memory until it hits a chunk of data with that bit set | ||
338 | and then looping back. | ||
339 | |||
340 | 02/09/1999 EPK Added Michel Mueller's TxDMA Interrupt and Tx-timeout | ||
341 | problem (TxCmd and RxCmd need only to be set when idle or stopped. | ||
342 | |||
343 | 02/09/1999 EPK Added code to check/reset dev->tbusy in hamachi_interrupt. | ||
344 | (Michel Mueller pointed out the ``permanently busy'' potential | ||
345 | problem here). | ||
346 | |||
347 | 02/22/1999 EPK Added Pete Wyckoff's ioctl to control the Tx/Rx latencies. | ||
348 | |||
349 | 02/23/1999 EPK Verified that the interrupt status field bits for Tx were | ||
350 | incorrectly defined and corrected (as per Michel Mueller). | ||
351 | |||
352 | 02/23/1999 EPK Corrected the Tx full check to check that at least 4 slots | ||
353 | were available before reseting the tbusy and tx_full flags | ||
354 | (as per Michel Mueller). | ||
355 | |||
356 | 03/11/1999 EPK Added Pete Wyckoff's hardware checksumming support. | ||
357 | |||
358 | 12/31/1999 KDU Cleaned up assorted things and added Don's code to force | ||
359 | 32 bit. | ||
360 | |||
361 | 02/20/2000 KDU Some of the control was just plain odd. Cleaned up the | ||
362 | hamachi_start_xmit() and hamachi_interrupt() code. There is still some | ||
363 | re-structuring I would like to do. | ||
364 | |||
365 | 03/01/2000 KDU Experimenting with a WIDE range of interrupt mitigation | ||
366 | parameters on a dual P3-450 setup yielded the new default interrupt | ||
367 | mitigation parameters. Tx should interrupt VERY infrequently due to | ||
368 | Eric's scheme. Rx should be more often... | ||
369 | |||
370 | 03/13/2000 KDU Added a patch to make the Rx Checksum code interact | ||
371 | nicely with non-linux machines. | ||
372 | |||
373 | 03/13/2000 KDU Experimented with some of the configuration values: | ||
374 | |||
375 | -It seems that enabling PCI performance commands for descriptors | ||
376 | (changing RxDMACtrl and TxDMACtrl lower nibble from 5 to D) has minimal | ||
377 | performance impact for any of my tests. (ttcp, netpipe, netperf) I will | ||
378 | leave them that way until I hear further feedback. | ||
379 | |||
380 | -Increasing the PCI_LATENCY_TIMER to 130 | ||
381 | (2 + (burst size of 128 * (0 wait states + 1))) seems to slightly | ||
382 | degrade performance. Leaving default at 64 pending further information. | ||
383 | |||
384 | 03/14/2000 KDU Further tuning: | ||
385 | |||
386 | -adjusted boguscnt in hamachi_rx() to depend on interrupt | ||
387 | mitigation parameters chosen. | ||
388 | |||
389 | -Selected a set of interrupt parameters based on some extensive testing. | ||
390 | These may change with more testing. | ||
391 | |||
392 | TO DO: | ||
393 | |||
394 | -Consider borrowing from the acenic driver code to check PCI_COMMAND for | ||
395 | PCI_COMMAND_INVALIDATE. Set maximum burst size to cache line size in | ||
396 | that case. | ||
397 | |||
398 | -fix the reset procedure. It doesn't quite work. | ||
399 | */ | ||
400 | |||
401 | /* A few values that may be tweaked. */ | ||
402 | /* Size of each temporary Rx buffer, calculated as: | ||
403 | * 1518 bytes (ethernet packet) + 2 bytes (to get 8 byte alignment for | ||
404 | * the card) + 8 bytes of status info + 8 bytes for the Rx Checksum | ||
405 | */ | ||
406 | #define PKT_BUF_SZ 1536 | ||
407 | |||
408 | /* For now, this is going to be set to the maximum size of an ethernet | ||
409 | * packet. Eventually, we may want to make it a variable that is | ||
410 | * related to the MTU | ||
411 | */ | ||
412 | #define MAX_FRAME_SIZE 1518 | ||
413 | |||
414 | /* The rest of these values should never change. */ | ||
415 | |||
416 | static void hamachi_timer(unsigned long data); | ||
417 | |||
418 | enum capability_flags {CanHaveMII=1, }; | ||
419 | static const struct chip_info { | ||
420 | u16 vendor_id, device_id, device_id_mask, pad; | ||
421 | const char *name; | ||
422 | void (*media_timer)(unsigned long data); | ||
423 | int flags; | ||
424 | } chip_tbl[] = { | ||
425 | {0x1318, 0x0911, 0xffff, 0, "Hamachi GNIC-II", hamachi_timer, 0}, | ||
426 | {0,}, | ||
427 | }; | ||
428 | |||
429 | /* Offsets to the Hamachi registers. Various sizes. */ | ||
430 | enum hamachi_offsets { | ||
431 | TxDMACtrl=0x00, TxCmd=0x04, TxStatus=0x06, TxPtr=0x08, TxCurPtr=0x10, | ||
432 | RxDMACtrl=0x20, RxCmd=0x24, RxStatus=0x26, RxPtr=0x28, RxCurPtr=0x30, | ||
433 | PCIClkMeas=0x060, MiscStatus=0x066, ChipRev=0x68, ChipReset=0x06B, | ||
434 | LEDCtrl=0x06C, VirtualJumpers=0x06D, GPIO=0x6E, | ||
435 | TxChecksum=0x074, RxChecksum=0x076, | ||
436 | TxIntrCtrl=0x078, RxIntrCtrl=0x07C, | ||
437 | InterruptEnable=0x080, InterruptClear=0x084, IntrStatus=0x088, | ||
438 | EventStatus=0x08C, | ||
439 | MACCnfg=0x0A0, FrameGap0=0x0A2, FrameGap1=0x0A4, | ||
440 | /* See enum MII_offsets below. */ | ||
441 | MACCnfg2=0x0B0, RxDepth=0x0B8, FlowCtrl=0x0BC, MaxFrameSize=0x0CE, | ||
442 | AddrMode=0x0D0, StationAddr=0x0D2, | ||
443 | /* Gigabit AutoNegotiation. */ | ||
444 | ANCtrl=0x0E0, ANStatus=0x0E2, ANXchngCtrl=0x0E4, ANAdvertise=0x0E8, | ||
445 | ANLinkPartnerAbility=0x0EA, | ||
446 | EECmdStatus=0x0F0, EEData=0x0F1, EEAddr=0x0F2, | ||
447 | FIFOcfg=0x0F8, | ||
448 | }; | ||
449 | |||
450 | /* Offsets to the MII-mode registers. */ | ||
451 | enum MII_offsets { | ||
452 | MII_Cmd=0xA6, MII_Addr=0xA8, MII_Wr_Data=0xAA, MII_Rd_Data=0xAC, | ||
453 | MII_Status=0xAE, | ||
454 | }; | ||
455 | |||
456 | /* Bits in the interrupt status/mask registers. */ | ||
457 | enum intr_status_bits { | ||
458 | IntrRxDone=0x01, IntrRxPCIFault=0x02, IntrRxPCIErr=0x04, | ||
459 | IntrTxDone=0x100, IntrTxPCIFault=0x200, IntrTxPCIErr=0x400, | ||
460 | LinkChange=0x10000, NegotiationChange=0x20000, StatsMax=0x40000, }; | ||
461 | |||
462 | /* The Hamachi Rx and Tx buffer descriptors. */ | ||
463 | struct hamachi_desc { | ||
464 | __le32 status_n_length; | ||
465 | #if ADDRLEN == 64 | ||
466 | u32 pad; | ||
467 | __le64 addr; | ||
468 | #else | ||
469 | __le32 addr; | ||
470 | #endif | ||
471 | }; | ||
472 | |||
473 | /* Bits in hamachi_desc.status_n_length */ | ||
474 | enum desc_status_bits { | ||
475 | DescOwn=0x80000000, DescEndPacket=0x40000000, DescEndRing=0x20000000, | ||
476 | DescIntr=0x10000000, | ||
477 | }; | ||
478 | |||
479 | #define PRIV_ALIGN 15 /* Required alignment mask */ | ||
480 | #define MII_CNT 4 | ||
481 | struct hamachi_private { | ||
482 | /* Descriptor rings first for alignment. Tx requires a second descriptor | ||
483 | for status. */ | ||
484 | struct hamachi_desc *rx_ring; | ||
485 | struct hamachi_desc *tx_ring; | ||
486 | struct sk_buff* rx_skbuff[RX_RING_SIZE]; | ||
487 | struct sk_buff* tx_skbuff[TX_RING_SIZE]; | ||
488 | dma_addr_t tx_ring_dma; | ||
489 | dma_addr_t rx_ring_dma; | ||
490 | struct timer_list timer; /* Media selection timer. */ | ||
491 | /* Frequently used and paired value: keep adjacent for cache effect. */ | ||
492 | spinlock_t lock; | ||
493 | int chip_id; | ||
494 | unsigned int cur_rx, dirty_rx; /* Producer/consumer ring indices */ | ||
495 | unsigned int cur_tx, dirty_tx; | ||
496 | unsigned int rx_buf_sz; /* Based on MTU+slack. */ | ||
497 | unsigned int tx_full:1; /* The Tx queue is full. */ | ||
498 | unsigned int duplex_lock:1; | ||
499 | unsigned int default_port:4; /* Last dev->if_port value. */ | ||
500 | /* MII transceiver section. */ | ||
501 | int mii_cnt; /* MII device addresses. */ | ||
502 | struct mii_if_info mii_if; /* MII lib hooks/info */ | ||
503 | unsigned char phys[MII_CNT]; /* MII device addresses, only first one used. */ | ||
504 | u32 rx_int_var, tx_int_var; /* interrupt control variables */ | ||
505 | u32 option; /* Hold on to a copy of the options */ | ||
506 | struct pci_dev *pci_dev; | ||
507 | void __iomem *base; | ||
508 | }; | ||
509 | |||
510 | MODULE_AUTHOR("Donald Becker <becker@scyld.com>, Eric Kasten <kasten@nscl.msu.edu>, Keith Underwood <keithu@parl.clemson.edu>"); | ||
511 | MODULE_DESCRIPTION("Packet Engines 'Hamachi' GNIC-II Gigabit Ethernet driver"); | ||
512 | MODULE_LICENSE("GPL"); | ||
513 | |||
514 | module_param(max_interrupt_work, int, 0); | ||
515 | module_param(mtu, int, 0); | ||
516 | module_param(debug, int, 0); | ||
517 | module_param(min_rx_pkt, int, 0); | ||
518 | module_param(max_rx_gap, int, 0); | ||
519 | module_param(max_rx_latency, int, 0); | ||
520 | module_param(min_tx_pkt, int, 0); | ||
521 | module_param(max_tx_gap, int, 0); | ||
522 | module_param(max_tx_latency, int, 0); | ||
523 | module_param(rx_copybreak, int, 0); | ||
524 | module_param_array(rx_params, int, NULL, 0); | ||
525 | module_param_array(tx_params, int, NULL, 0); | ||
526 | module_param_array(options, int, NULL, 0); | ||
527 | module_param_array(full_duplex, int, NULL, 0); | ||
528 | module_param(force32, int, 0); | ||
529 | MODULE_PARM_DESC(max_interrupt_work, "GNIC-II maximum events handled per interrupt"); | ||
530 | MODULE_PARM_DESC(mtu, "GNIC-II MTU (all boards)"); | ||
531 | MODULE_PARM_DESC(debug, "GNIC-II debug level (0-7)"); | ||
532 | MODULE_PARM_DESC(min_rx_pkt, "GNIC-II minimum Rx packets processed between interrupts"); | ||
533 | MODULE_PARM_DESC(max_rx_gap, "GNIC-II maximum Rx inter-packet gap in 8.192 microsecond units"); | ||
534 | MODULE_PARM_DESC(max_rx_latency, "GNIC-II time between Rx interrupts in 8.192 microsecond units"); | ||
535 | MODULE_PARM_DESC(min_tx_pkt, "GNIC-II minimum Tx packets processed between interrupts"); | ||
536 | MODULE_PARM_DESC(max_tx_gap, "GNIC-II maximum Tx inter-packet gap in 8.192 microsecond units"); | ||
537 | MODULE_PARM_DESC(max_tx_latency, "GNIC-II time between Tx interrupts in 8.192 microsecond units"); | ||
538 | MODULE_PARM_DESC(rx_copybreak, "GNIC-II copy breakpoint for copy-only-tiny-frames"); | ||
539 | MODULE_PARM_DESC(rx_params, "GNIC-II min_rx_pkt+max_rx_gap+max_rx_latency"); | ||
540 | MODULE_PARM_DESC(tx_params, "GNIC-II min_tx_pkt+max_tx_gap+max_tx_latency"); | ||
541 | MODULE_PARM_DESC(options, "GNIC-II Bits 0-3: media type, bits 4-6: as force32, bit 7: half duplex, bit 9 full duplex"); | ||
542 | MODULE_PARM_DESC(full_duplex, "GNIC-II full duplex setting(s) (1)"); | ||
543 | MODULE_PARM_DESC(force32, "GNIC-II: Bit 0: 32 bit PCI, bit 1: disable parity, bit 2: 64 bit PCI (all boards)"); | ||
544 | |||
545 | static int read_eeprom(void __iomem *ioaddr, int location); | ||
546 | static int mdio_read(struct net_device *dev, int phy_id, int location); | ||
547 | static void mdio_write(struct net_device *dev, int phy_id, int location, int value); | ||
548 | static int hamachi_open(struct net_device *dev); | ||
549 | static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd); | ||
550 | static void hamachi_timer(unsigned long data); | ||
551 | static void hamachi_tx_timeout(struct net_device *dev); | ||
552 | static void hamachi_init_ring(struct net_device *dev); | ||
553 | static netdev_tx_t hamachi_start_xmit(struct sk_buff *skb, | ||
554 | struct net_device *dev); | ||
555 | static irqreturn_t hamachi_interrupt(int irq, void *dev_instance); | ||
556 | static int hamachi_rx(struct net_device *dev); | ||
557 | static inline int hamachi_tx(struct net_device *dev); | ||
558 | static void hamachi_error(struct net_device *dev, int intr_status); | ||
559 | static int hamachi_close(struct net_device *dev); | ||
560 | static struct net_device_stats *hamachi_get_stats(struct net_device *dev); | ||
561 | static void set_rx_mode(struct net_device *dev); | ||
562 | static const struct ethtool_ops ethtool_ops; | ||
563 | static const struct ethtool_ops ethtool_ops_no_mii; | ||
564 | |||
565 | static const struct net_device_ops hamachi_netdev_ops = { | ||
566 | .ndo_open = hamachi_open, | ||
567 | .ndo_stop = hamachi_close, | ||
568 | .ndo_start_xmit = hamachi_start_xmit, | ||
569 | .ndo_get_stats = hamachi_get_stats, | ||
570 | .ndo_set_multicast_list = set_rx_mode, | ||
571 | .ndo_change_mtu = eth_change_mtu, | ||
572 | .ndo_validate_addr = eth_validate_addr, | ||
573 | .ndo_set_mac_address = eth_mac_addr, | ||
574 | .ndo_tx_timeout = hamachi_tx_timeout, | ||
575 | .ndo_do_ioctl = netdev_ioctl, | ||
576 | }; | ||
577 | |||
578 | |||
579 | static int __devinit hamachi_init_one (struct pci_dev *pdev, | ||
580 | const struct pci_device_id *ent) | ||
581 | { | ||
582 | struct hamachi_private *hmp; | ||
583 | int option, i, rx_int_var, tx_int_var, boguscnt; | ||
584 | int chip_id = ent->driver_data; | ||
585 | int irq; | ||
586 | void __iomem *ioaddr; | ||
587 | unsigned long base; | ||
588 | static int card_idx; | ||
589 | struct net_device *dev; | ||
590 | void *ring_space; | ||
591 | dma_addr_t ring_dma; | ||
592 | int ret = -ENOMEM; | ||
593 | |||
594 | /* when built into the kernel, we only print version if device is found */ | ||
595 | #ifndef MODULE | ||
596 | static int printed_version; | ||
597 | if (!printed_version++) | ||
598 | printk(version); | ||
599 | #endif | ||
600 | |||
601 | if (pci_enable_device(pdev)) { | ||
602 | ret = -EIO; | ||
603 | goto err_out; | ||
604 | } | ||
605 | |||
606 | base = pci_resource_start(pdev, 0); | ||
607 | #ifdef __alpha__ /* Really "64 bit addrs" */ | ||
608 | base |= (pci_resource_start(pdev, 1) << 32); | ||
609 | #endif | ||
610 | |||
611 | pci_set_master(pdev); | ||
612 | |||
613 | i = pci_request_regions(pdev, DRV_NAME); | ||
614 | if (i) | ||
615 | return i; | ||
616 | |||
617 | irq = pdev->irq; | ||
618 | ioaddr = ioremap(base, 0x400); | ||
619 | if (!ioaddr) | ||
620 | goto err_out_release; | ||
621 | |||
622 | dev = alloc_etherdev(sizeof(struct hamachi_private)); | ||
623 | if (!dev) | ||
624 | goto err_out_iounmap; | ||
625 | |||
626 | SET_NETDEV_DEV(dev, &pdev->dev); | ||
627 | |||
628 | for (i = 0; i < 6; i++) | ||
629 | dev->dev_addr[i] = 1 ? read_eeprom(ioaddr, 4 + i) | ||
630 | : readb(ioaddr + StationAddr + i); | ||
631 | |||
632 | #if ! defined(final_version) | ||
633 | if (hamachi_debug > 4) | ||
634 | for (i = 0; i < 0x10; i++) | ||
635 | printk("%2.2x%s", | ||
636 | read_eeprom(ioaddr, i), i % 16 != 15 ? " " : "\n"); | ||
637 | #endif | ||
638 | |||
639 | hmp = netdev_priv(dev); | ||
640 | spin_lock_init(&hmp->lock); | ||
641 | |||
642 | hmp->mii_if.dev = dev; | ||
643 | hmp->mii_if.mdio_read = mdio_read; | ||
644 | hmp->mii_if.mdio_write = mdio_write; | ||
645 | hmp->mii_if.phy_id_mask = 0x1f; | ||
646 | hmp->mii_if.reg_num_mask = 0x1f; | ||
647 | |||
648 | ring_space = pci_alloc_consistent(pdev, TX_TOTAL_SIZE, &ring_dma); | ||
649 | if (!ring_space) | ||
650 | goto err_out_cleardev; | ||
651 | hmp->tx_ring = ring_space; | ||
652 | hmp->tx_ring_dma = ring_dma; | ||
653 | |||
654 | ring_space = pci_alloc_consistent(pdev, RX_TOTAL_SIZE, &ring_dma); | ||
655 | if (!ring_space) | ||
656 | goto err_out_unmap_tx; | ||
657 | hmp->rx_ring = ring_space; | ||
658 | hmp->rx_ring_dma = ring_dma; | ||
659 | |||
660 | /* Check for options being passed in */ | ||
661 | option = card_idx < MAX_UNITS ? options[card_idx] : 0; | ||
662 | if (dev->mem_start) | ||
663 | option = dev->mem_start; | ||
664 | |||
665 | /* If the bus size is misidentified, do the following. */ | ||
666 | force32 = force32 ? force32 : | ||
667 | ((option >= 0) ? ((option & 0x00000070) >> 4) : 0 ); | ||
668 | if (force32) | ||
669 | writeb(force32, ioaddr + VirtualJumpers); | ||
670 | |||
671 | /* Hmmm, do we really need to reset the chip???. */ | ||
672 | writeb(0x01, ioaddr + ChipReset); | ||
673 | |||
674 | /* After a reset, the clock speed measurement of the PCI bus will not | ||
675 | * be valid for a moment. Wait for a little while until it is. If | ||
676 | * it takes more than 10ms, forget it. | ||
677 | */ | ||
678 | udelay(10); | ||
679 | i = readb(ioaddr + PCIClkMeas); | ||
680 | for (boguscnt = 0; (!(i & 0x080)) && boguscnt < 1000; boguscnt++){ | ||
681 | udelay(10); | ||
682 | i = readb(ioaddr + PCIClkMeas); | ||
683 | } | ||
684 | |||
685 | hmp->base = ioaddr; | ||
686 | dev->base_addr = (unsigned long)ioaddr; | ||
687 | dev->irq = irq; | ||
688 | pci_set_drvdata(pdev, dev); | ||
689 | |||
690 | hmp->chip_id = chip_id; | ||
691 | hmp->pci_dev = pdev; | ||
692 | |||
693 | /* The lower four bits are the media type. */ | ||
694 | if (option > 0) { | ||
695 | hmp->option = option; | ||
696 | if (option & 0x200) | ||
697 | hmp->mii_if.full_duplex = 1; | ||
698 | else if (option & 0x080) | ||
699 | hmp->mii_if.full_duplex = 0; | ||
700 | hmp->default_port = option & 15; | ||
701 | if (hmp->default_port) | ||
702 | hmp->mii_if.force_media = 1; | ||
703 | } | ||
704 | if (card_idx < MAX_UNITS && full_duplex[card_idx] > 0) | ||
705 | hmp->mii_if.full_duplex = 1; | ||
706 | |||
707 | /* lock the duplex mode if someone specified a value */ | ||
708 | if (hmp->mii_if.full_duplex || (option & 0x080)) | ||
709 | hmp->duplex_lock = 1; | ||
710 | |||
711 | /* Set interrupt tuning parameters */ | ||
712 | max_rx_latency = max_rx_latency & 0x00ff; | ||
713 | max_rx_gap = max_rx_gap & 0x00ff; | ||
714 | min_rx_pkt = min_rx_pkt & 0x00ff; | ||
715 | max_tx_latency = max_tx_latency & 0x00ff; | ||
716 | max_tx_gap = max_tx_gap & 0x00ff; | ||
717 | min_tx_pkt = min_tx_pkt & 0x00ff; | ||
718 | |||
719 | rx_int_var = card_idx < MAX_UNITS ? rx_params[card_idx] : -1; | ||
720 | tx_int_var = card_idx < MAX_UNITS ? tx_params[card_idx] : -1; | ||
721 | hmp->rx_int_var = rx_int_var >= 0 ? rx_int_var : | ||
722 | (min_rx_pkt << 16 | max_rx_gap << 8 | max_rx_latency); | ||
723 | hmp->tx_int_var = tx_int_var >= 0 ? tx_int_var : | ||
724 | (min_tx_pkt << 16 | max_tx_gap << 8 | max_tx_latency); | ||
725 | |||
726 | |||
727 | /* The Hamachi-specific entries in the device structure. */ | ||
728 | dev->netdev_ops = &hamachi_netdev_ops; | ||
729 | if (chip_tbl[hmp->chip_id].flags & CanHaveMII) | ||
730 | SET_ETHTOOL_OPS(dev, ðtool_ops); | ||
731 | else | ||
732 | SET_ETHTOOL_OPS(dev, ðtool_ops_no_mii); | ||
733 | dev->watchdog_timeo = TX_TIMEOUT; | ||
734 | if (mtu) | ||
735 | dev->mtu = mtu; | ||
736 | |||
737 | i = register_netdev(dev); | ||
738 | if (i) { | ||
739 | ret = i; | ||
740 | goto err_out_unmap_rx; | ||
741 | } | ||
742 | |||
743 | printk(KERN_INFO "%s: %s type %x at %p, %pM, IRQ %d.\n", | ||
744 | dev->name, chip_tbl[chip_id].name, readl(ioaddr + ChipRev), | ||
745 | ioaddr, dev->dev_addr, irq); | ||
746 | i = readb(ioaddr + PCIClkMeas); | ||
747 | printk(KERN_INFO "%s: %d-bit %d Mhz PCI bus (%d), Virtual Jumpers " | ||
748 | "%2.2x, LPA %4.4x.\n", | ||
749 | dev->name, readw(ioaddr + MiscStatus) & 1 ? 64 : 32, | ||
750 | i ? 2000/(i&0x7f) : 0, i&0x7f, (int)readb(ioaddr + VirtualJumpers), | ||
751 | readw(ioaddr + ANLinkPartnerAbility)); | ||
752 | |||
753 | if (chip_tbl[hmp->chip_id].flags & CanHaveMII) { | ||
754 | int phy, phy_idx = 0; | ||
755 | for (phy = 0; phy < 32 && phy_idx < MII_CNT; phy++) { | ||
756 | int mii_status = mdio_read(dev, phy, MII_BMSR); | ||
757 | if (mii_status != 0xffff && | ||
758 | mii_status != 0x0000) { | ||
759 | hmp->phys[phy_idx++] = phy; | ||
760 | hmp->mii_if.advertising = mdio_read(dev, phy, MII_ADVERTISE); | ||
761 | printk(KERN_INFO "%s: MII PHY found at address %d, status " | ||
762 | "0x%4.4x advertising %4.4x.\n", | ||
763 | dev->name, phy, mii_status, hmp->mii_if.advertising); | ||
764 | } | ||
765 | } | ||
766 | hmp->mii_cnt = phy_idx; | ||
767 | if (hmp->mii_cnt > 0) | ||
768 | hmp->mii_if.phy_id = hmp->phys[0]; | ||
769 | else | ||
770 | memset(&hmp->mii_if, 0, sizeof(hmp->mii_if)); | ||
771 | } | ||
772 | /* Configure gigabit autonegotiation. */ | ||
773 | writew(0x0400, ioaddr + ANXchngCtrl); /* Enable legacy links. */ | ||
774 | writew(0x08e0, ioaddr + ANAdvertise); /* Set our advertise word. */ | ||
775 | writew(0x1000, ioaddr + ANCtrl); /* Enable negotiation */ | ||
776 | |||
777 | card_idx++; | ||
778 | return 0; | ||
779 | |||
780 | err_out_unmap_rx: | ||
781 | pci_free_consistent(pdev, RX_TOTAL_SIZE, hmp->rx_ring, | ||
782 | hmp->rx_ring_dma); | ||
783 | err_out_unmap_tx: | ||
784 | pci_free_consistent(pdev, TX_TOTAL_SIZE, hmp->tx_ring, | ||
785 | hmp->tx_ring_dma); | ||
786 | err_out_cleardev: | ||
787 | free_netdev (dev); | ||
788 | err_out_iounmap: | ||
789 | iounmap(ioaddr); | ||
790 | err_out_release: | ||
791 | pci_release_regions(pdev); | ||
792 | err_out: | ||
793 | return ret; | ||
794 | } | ||
795 | |||
796 | static int __devinit read_eeprom(void __iomem *ioaddr, int location) | ||
797 | { | ||
798 | int bogus_cnt = 1000; | ||
799 | |||
800 | /* We should check busy first - per docs -KDU */ | ||
801 | while ((readb(ioaddr + EECmdStatus) & 0x40) && --bogus_cnt > 0); | ||
802 | writew(location, ioaddr + EEAddr); | ||
803 | writeb(0x02, ioaddr + EECmdStatus); | ||
804 | bogus_cnt = 1000; | ||
805 | while ((readb(ioaddr + EECmdStatus) & 0x40) && --bogus_cnt > 0); | ||
806 | if (hamachi_debug > 5) | ||
807 | printk(" EEPROM status is %2.2x after %d ticks.\n", | ||
808 | (int)readb(ioaddr + EECmdStatus), 1000- bogus_cnt); | ||
809 | return readb(ioaddr + EEData); | ||
810 | } | ||
811 | |||
812 | /* MII Managemen Data I/O accesses. | ||
813 | These routines assume the MDIO controller is idle, and do not exit until | ||
814 | the command is finished. */ | ||
815 | |||
816 | static int mdio_read(struct net_device *dev, int phy_id, int location) | ||
817 | { | ||
818 | struct hamachi_private *hmp = netdev_priv(dev); | ||
819 | void __iomem *ioaddr = hmp->base; | ||
820 | int i; | ||
821 | |||
822 | /* We should check busy first - per docs -KDU */ | ||
823 | for (i = 10000; i >= 0; i--) | ||
824 | if ((readw(ioaddr + MII_Status) & 1) == 0) | ||
825 | break; | ||
826 | writew((phy_id<<8) + location, ioaddr + MII_Addr); | ||
827 | writew(0x0001, ioaddr + MII_Cmd); | ||
828 | for (i = 10000; i >= 0; i--) | ||
829 | if ((readw(ioaddr + MII_Status) & 1) == 0) | ||
830 | break; | ||
831 | return readw(ioaddr + MII_Rd_Data); | ||
832 | } | ||
833 | |||
834 | static void mdio_write(struct net_device *dev, int phy_id, int location, int value) | ||
835 | { | ||
836 | struct hamachi_private *hmp = netdev_priv(dev); | ||
837 | void __iomem *ioaddr = hmp->base; | ||
838 | int i; | ||
839 | |||
840 | /* We should check busy first - per docs -KDU */ | ||
841 | for (i = 10000; i >= 0; i--) | ||
842 | if ((readw(ioaddr + MII_Status) & 1) == 0) | ||
843 | break; | ||
844 | writew((phy_id<<8) + location, ioaddr + MII_Addr); | ||
845 | writew(value, ioaddr + MII_Wr_Data); | ||
846 | |||
847 | /* Wait for the command to finish. */ | ||
848 | for (i = 10000; i >= 0; i--) | ||
849 | if ((readw(ioaddr + MII_Status) & 1) == 0) | ||
850 | break; | ||
851 | } | ||
852 | |||
853 | |||
854 | static int hamachi_open(struct net_device *dev) | ||
855 | { | ||
856 | struct hamachi_private *hmp = netdev_priv(dev); | ||
857 | void __iomem *ioaddr = hmp->base; | ||
858 | int i; | ||
859 | u32 rx_int_var, tx_int_var; | ||
860 | u16 fifo_info; | ||
861 | |||
862 | i = request_irq(dev->irq, hamachi_interrupt, IRQF_SHARED, dev->name, dev); | ||
863 | if (i) | ||
864 | return i; | ||
865 | |||
866 | if (hamachi_debug > 1) | ||
867 | printk(KERN_DEBUG "%s: hamachi_open() irq %d.\n", | ||
868 | dev->name, dev->irq); | ||
869 | |||
870 | hamachi_init_ring(dev); | ||
871 | |||
872 | #if ADDRLEN == 64 | ||
873 | /* writellll anyone ? */ | ||
874 | writel(hmp->rx_ring_dma, ioaddr + RxPtr); | ||
875 | writel(hmp->rx_ring_dma >> 32, ioaddr + RxPtr + 4); | ||
876 | writel(hmp->tx_ring_dma, ioaddr + TxPtr); | ||
877 | writel(hmp->tx_ring_dma >> 32, ioaddr + TxPtr + 4); | ||
878 | #else | ||
879 | writel(hmp->rx_ring_dma, ioaddr + RxPtr); | ||
880 | writel(hmp->tx_ring_dma, ioaddr + TxPtr); | ||
881 | #endif | ||
882 | |||
883 | /* TODO: It would make sense to organize this as words since the card | ||
884 | * documentation does. -KDU | ||
885 | */ | ||
886 | for (i = 0; i < 6; i++) | ||
887 | writeb(dev->dev_addr[i], ioaddr + StationAddr + i); | ||
888 | |||
889 | /* Initialize other registers: with so many this eventually this will | ||
890 | converted to an offset/value list. */ | ||
891 | |||
892 | /* Configure the FIFO */ | ||
893 | fifo_info = (readw(ioaddr + GPIO) & 0x00C0) >> 6; | ||
894 | switch (fifo_info){ | ||
895 | case 0 : | ||
896 | /* No FIFO */ | ||
897 | writew(0x0000, ioaddr + FIFOcfg); | ||
898 | break; | ||
899 | case 1 : | ||
900 | /* Configure the FIFO for 512K external, 16K used for Tx. */ | ||
901 | writew(0x0028, ioaddr + FIFOcfg); | ||
902 | break; | ||
903 | case 2 : | ||
904 | /* Configure the FIFO for 1024 external, 32K used for Tx. */ | ||
905 | writew(0x004C, ioaddr + FIFOcfg); | ||
906 | break; | ||
907 | case 3 : | ||
908 | /* Configure the FIFO for 2048 external, 32K used for Tx. */ | ||
909 | writew(0x006C, ioaddr + FIFOcfg); | ||
910 | break; | ||
911 | default : | ||
912 | printk(KERN_WARNING "%s: Unsupported external memory config!\n", | ||
913 | dev->name); | ||
914 | /* Default to no FIFO */ | ||
915 | writew(0x0000, ioaddr + FIFOcfg); | ||
916 | break; | ||
917 | } | ||
918 | |||
919 | if (dev->if_port == 0) | ||
920 | dev->if_port = hmp->default_port; | ||
921 | |||
922 | |||
923 | /* Setting the Rx mode will start the Rx process. */ | ||
924 | /* If someone didn't choose a duplex, default to full-duplex */ | ||
925 | if (hmp->duplex_lock != 1) | ||
926 | hmp->mii_if.full_duplex = 1; | ||
927 | |||
928 | /* always 1, takes no more time to do it */ | ||
929 | writew(0x0001, ioaddr + RxChecksum); | ||
930 | writew(0x0000, ioaddr + TxChecksum); | ||
931 | writew(0x8000, ioaddr + MACCnfg); /* Soft reset the MAC */ | ||
932 | writew(0x215F, ioaddr + MACCnfg); | ||
933 | writew(0x000C, ioaddr + FrameGap0); | ||
934 | /* WHAT?!?!? Why isn't this documented somewhere? -KDU */ | ||
935 | writew(0x1018, ioaddr + FrameGap1); | ||
936 | /* Why do we enable receives/transmits here? -KDU */ | ||
937 | writew(0x0780, ioaddr + MACCnfg2); /* Upper 16 bits control LEDs. */ | ||
938 | /* Enable automatic generation of flow control frames, period 0xffff. */ | ||
939 | writel(0x0030FFFF, ioaddr + FlowCtrl); | ||
940 | writew(MAX_FRAME_SIZE, ioaddr + MaxFrameSize); /* dev->mtu+14 ??? */ | ||
941 | |||
942 | /* Enable legacy links. */ | ||
943 | writew(0x0400, ioaddr + ANXchngCtrl); /* Enable legacy links. */ | ||
944 | /* Initial Link LED to blinking red. */ | ||
945 | writeb(0x03, ioaddr + LEDCtrl); | ||
946 | |||
947 | /* Configure interrupt mitigation. This has a great effect on | ||
948 | performance, so systems tuning should start here!. */ | ||
949 | |||
950 | rx_int_var = hmp->rx_int_var; | ||
951 | tx_int_var = hmp->tx_int_var; | ||
952 | |||
953 | if (hamachi_debug > 1) { | ||
954 | printk("max_tx_latency: %d, max_tx_gap: %d, min_tx_pkt: %d\n", | ||
955 | tx_int_var & 0x00ff, (tx_int_var & 0x00ff00) >> 8, | ||
956 | (tx_int_var & 0x00ff0000) >> 16); | ||
957 | printk("max_rx_latency: %d, max_rx_gap: %d, min_rx_pkt: %d\n", | ||
958 | rx_int_var & 0x00ff, (rx_int_var & 0x00ff00) >> 8, | ||
959 | (rx_int_var & 0x00ff0000) >> 16); | ||
960 | printk("rx_int_var: %x, tx_int_var: %x\n", rx_int_var, tx_int_var); | ||
961 | } | ||
962 | |||
963 | writel(tx_int_var, ioaddr + TxIntrCtrl); | ||
964 | writel(rx_int_var, ioaddr + RxIntrCtrl); | ||
965 | |||
966 | set_rx_mode(dev); | ||
967 | |||
968 | netif_start_queue(dev); | ||
969 | |||
970 | /* Enable interrupts by setting the interrupt mask. */ | ||
971 | writel(0x80878787, ioaddr + InterruptEnable); | ||
972 | writew(0x0000, ioaddr + EventStatus); /* Clear non-interrupting events */ | ||
973 | |||
974 | /* Configure and start the DMA channels. */ | ||
975 | /* Burst sizes are in the low three bits: size = 4<<(val&7) */ | ||
976 | #if ADDRLEN == 64 | ||
977 | writew(0x005D, ioaddr + RxDMACtrl); /* 128 dword bursts */ | ||
978 | writew(0x005D, ioaddr + TxDMACtrl); | ||
979 | #else | ||
980 | writew(0x001D, ioaddr + RxDMACtrl); | ||
981 | writew(0x001D, ioaddr + TxDMACtrl); | ||
982 | #endif | ||
983 | writew(0x0001, ioaddr + RxCmd); | ||
984 | |||
985 | if (hamachi_debug > 2) { | ||
986 | printk(KERN_DEBUG "%s: Done hamachi_open(), status: Rx %x Tx %x.\n", | ||
987 | dev->name, readw(ioaddr + RxStatus), readw(ioaddr + TxStatus)); | ||
988 | } | ||
989 | /* Set the timer to check for link beat. */ | ||
990 | init_timer(&hmp->timer); | ||
991 | hmp->timer.expires = RUN_AT((24*HZ)/10); /* 2.4 sec. */ | ||
992 | hmp->timer.data = (unsigned long)dev; | ||
993 | hmp->timer.function = hamachi_timer; /* timer handler */ | ||
994 | add_timer(&hmp->timer); | ||
995 | |||
996 | return 0; | ||
997 | } | ||
998 | |||
999 | static inline int hamachi_tx(struct net_device *dev) | ||
1000 | { | ||
1001 | struct hamachi_private *hmp = netdev_priv(dev); | ||
1002 | |||
1003 | /* Update the dirty pointer until we find an entry that is | ||
1004 | still owned by the card */ | ||
1005 | for (; hmp->cur_tx - hmp->dirty_tx > 0; hmp->dirty_tx++) { | ||
1006 | int entry = hmp->dirty_tx % TX_RING_SIZE; | ||
1007 | struct sk_buff *skb; | ||
1008 | |||
1009 | if (hmp->tx_ring[entry].status_n_length & cpu_to_le32(DescOwn)) | ||
1010 | break; | ||
1011 | /* Free the original skb. */ | ||
1012 | skb = hmp->tx_skbuff[entry]; | ||
1013 | if (skb) { | ||
1014 | pci_unmap_single(hmp->pci_dev, | ||
1015 | leXX_to_cpu(hmp->tx_ring[entry].addr), | ||
1016 | skb->len, PCI_DMA_TODEVICE); | ||
1017 | dev_kfree_skb(skb); | ||
1018 | hmp->tx_skbuff[entry] = NULL; | ||
1019 | } | ||
1020 | hmp->tx_ring[entry].status_n_length = 0; | ||
1021 | if (entry >= TX_RING_SIZE-1) | ||
1022 | hmp->tx_ring[TX_RING_SIZE-1].status_n_length |= | ||
1023 | cpu_to_le32(DescEndRing); | ||
1024 | dev->stats.tx_packets++; | ||
1025 | } | ||
1026 | |||
1027 | return 0; | ||
1028 | } | ||
1029 | |||
1030 | static void hamachi_timer(unsigned long data) | ||
1031 | { | ||
1032 | struct net_device *dev = (struct net_device *)data; | ||
1033 | struct hamachi_private *hmp = netdev_priv(dev); | ||
1034 | void __iomem *ioaddr = hmp->base; | ||
1035 | int next_tick = 10*HZ; | ||
1036 | |||
1037 | if (hamachi_debug > 2) { | ||
1038 | printk(KERN_INFO "%s: Hamachi Autonegotiation status %4.4x, LPA " | ||
1039 | "%4.4x.\n", dev->name, readw(ioaddr + ANStatus), | ||
1040 | readw(ioaddr + ANLinkPartnerAbility)); | ||
1041 | printk(KERN_INFO "%s: Autonegotiation regs %4.4x %4.4x %4.4x " | ||
1042 | "%4.4x %4.4x %4.4x.\n", dev->name, | ||
1043 | readw(ioaddr + 0x0e0), | ||
1044 | readw(ioaddr + 0x0e2), | ||
1045 | readw(ioaddr + 0x0e4), | ||
1046 | readw(ioaddr + 0x0e6), | ||
1047 | readw(ioaddr + 0x0e8), | ||
1048 | readw(ioaddr + 0x0eA)); | ||
1049 | } | ||
1050 | /* We could do something here... nah. */ | ||
1051 | hmp->timer.expires = RUN_AT(next_tick); | ||
1052 | add_timer(&hmp->timer); | ||
1053 | } | ||
1054 | |||
1055 | static void hamachi_tx_timeout(struct net_device *dev) | ||
1056 | { | ||
1057 | int i; | ||
1058 | struct hamachi_private *hmp = netdev_priv(dev); | ||
1059 | void __iomem *ioaddr = hmp->base; | ||
1060 | |||
1061 | printk(KERN_WARNING "%s: Hamachi transmit timed out, status %8.8x," | ||
1062 | " resetting...\n", dev->name, (int)readw(ioaddr + TxStatus)); | ||
1063 | |||
1064 | { | ||
1065 | printk(KERN_DEBUG " Rx ring %p: ", hmp->rx_ring); | ||
1066 | for (i = 0; i < RX_RING_SIZE; i++) | ||
1067 | printk(KERN_CONT " %8.8x", | ||
1068 | le32_to_cpu(hmp->rx_ring[i].status_n_length)); | ||
1069 | printk(KERN_CONT "\n"); | ||
1070 | printk(KERN_DEBUG" Tx ring %p: ", hmp->tx_ring); | ||
1071 | for (i = 0; i < TX_RING_SIZE; i++) | ||
1072 | printk(KERN_CONT " %4.4x", | ||
1073 | le32_to_cpu(hmp->tx_ring[i].status_n_length)); | ||
1074 | printk(KERN_CONT "\n"); | ||
1075 | } | ||
1076 | |||
1077 | /* Reinit the hardware and make sure the Rx and Tx processes | ||
1078 | are up and running. | ||
1079 | */ | ||
1080 | dev->if_port = 0; | ||
1081 | /* The right way to do Reset. -KDU | ||
1082 | * -Clear OWN bit in all Rx/Tx descriptors | ||
1083 | * -Wait 50 uS for channels to go idle | ||
1084 | * -Turn off MAC receiver | ||
1085 | * -Issue Reset | ||
1086 | */ | ||
1087 | |||
1088 | for (i = 0; i < RX_RING_SIZE; i++) | ||
1089 | hmp->rx_ring[i].status_n_length &= cpu_to_le32(~DescOwn); | ||
1090 | |||
1091 | /* Presume that all packets in the Tx queue are gone if we have to | ||
1092 | * re-init the hardware. | ||
1093 | */ | ||
1094 | for (i = 0; i < TX_RING_SIZE; i++){ | ||
1095 | struct sk_buff *skb; | ||
1096 | |||
1097 | if (i >= TX_RING_SIZE - 1) | ||
1098 | hmp->tx_ring[i].status_n_length = | ||
1099 | cpu_to_le32(DescEndRing) | | ||
1100 | (hmp->tx_ring[i].status_n_length & | ||
1101 | cpu_to_le32(0x0000ffff)); | ||
1102 | else | ||
1103 | hmp->tx_ring[i].status_n_length &= cpu_to_le32(0x0000ffff); | ||
1104 | skb = hmp->tx_skbuff[i]; | ||
1105 | if (skb){ | ||
1106 | pci_unmap_single(hmp->pci_dev, leXX_to_cpu(hmp->tx_ring[i].addr), | ||
1107 | skb->len, PCI_DMA_TODEVICE); | ||
1108 | dev_kfree_skb(skb); | ||
1109 | hmp->tx_skbuff[i] = NULL; | ||
1110 | } | ||
1111 | } | ||
1112 | |||
1113 | udelay(60); /* Sleep 60 us just for safety sake */ | ||
1114 | writew(0x0002, ioaddr + RxCmd); /* STOP Rx */ | ||
1115 | |||
1116 | writeb(0x01, ioaddr + ChipReset); /* Reinit the hardware */ | ||
1117 | |||
1118 | hmp->tx_full = 0; | ||
1119 | hmp->cur_rx = hmp->cur_tx = 0; | ||
1120 | hmp->dirty_rx = hmp->dirty_tx = 0; | ||
1121 | /* Rx packets are also presumed lost; however, we need to make sure a | ||
1122 | * ring of buffers is in tact. -KDU | ||
1123 | */ | ||
1124 | for (i = 0; i < RX_RING_SIZE; i++){ | ||
1125 | struct sk_buff *skb = hmp->rx_skbuff[i]; | ||
1126 | |||
1127 | if (skb){ | ||
1128 | pci_unmap_single(hmp->pci_dev, | ||
1129 | leXX_to_cpu(hmp->rx_ring[i].addr), | ||
1130 | hmp->rx_buf_sz, PCI_DMA_FROMDEVICE); | ||
1131 | dev_kfree_skb(skb); | ||
1132 | hmp->rx_skbuff[i] = NULL; | ||
1133 | } | ||
1134 | } | ||
1135 | /* Fill in the Rx buffers. Handle allocation failure gracefully. */ | ||
1136 | for (i = 0; i < RX_RING_SIZE; i++) { | ||
1137 | struct sk_buff *skb; | ||
1138 | |||
1139 | skb = netdev_alloc_skb_ip_align(dev, hmp->rx_buf_sz); | ||
1140 | hmp->rx_skbuff[i] = skb; | ||
1141 | if (skb == NULL) | ||
1142 | break; | ||
1143 | |||
1144 | hmp->rx_ring[i].addr = cpu_to_leXX(pci_map_single(hmp->pci_dev, | ||
1145 | skb->data, hmp->rx_buf_sz, PCI_DMA_FROMDEVICE)); | ||
1146 | hmp->rx_ring[i].status_n_length = cpu_to_le32(DescOwn | | ||
1147 | DescEndPacket | DescIntr | (hmp->rx_buf_sz - 2)); | ||
1148 | } | ||
1149 | hmp->dirty_rx = (unsigned int)(i - RX_RING_SIZE); | ||
1150 | /* Mark the last entry as wrapping the ring. */ | ||
1151 | hmp->rx_ring[RX_RING_SIZE-1].status_n_length |= cpu_to_le32(DescEndRing); | ||
1152 | |||
1153 | /* Trigger an immediate transmit demand. */ | ||
1154 | dev->trans_start = jiffies; /* prevent tx timeout */ | ||
1155 | dev->stats.tx_errors++; | ||
1156 | |||
1157 | /* Restart the chip's Tx/Rx processes . */ | ||
1158 | writew(0x0002, ioaddr + TxCmd); /* STOP Tx */ | ||
1159 | writew(0x0001, ioaddr + TxCmd); /* START Tx */ | ||
1160 | writew(0x0001, ioaddr + RxCmd); /* START Rx */ | ||
1161 | |||
1162 | netif_wake_queue(dev); | ||
1163 | } | ||
1164 | |||
1165 | |||
1166 | /* Initialize the Rx and Tx rings, along with various 'dev' bits. */ | ||
1167 | static void hamachi_init_ring(struct net_device *dev) | ||
1168 | { | ||
1169 | struct hamachi_private *hmp = netdev_priv(dev); | ||
1170 | int i; | ||
1171 | |||
1172 | hmp->tx_full = 0; | ||
1173 | hmp->cur_rx = hmp->cur_tx = 0; | ||
1174 | hmp->dirty_rx = hmp->dirty_tx = 0; | ||
1175 | |||
1176 | /* +26 gets the maximum ethernet encapsulation, +7 & ~7 because the | ||
1177 | * card needs room to do 8 byte alignment, +2 so we can reserve | ||
1178 | * the first 2 bytes, and +16 gets room for the status word from the | ||
1179 | * card. -KDU | ||
1180 | */ | ||
1181 | hmp->rx_buf_sz = (dev->mtu <= 1492 ? PKT_BUF_SZ : | ||
1182 | (((dev->mtu+26+7) & ~7) + 16)); | ||
1183 | |||
1184 | /* Initialize all Rx descriptors. */ | ||
1185 | for (i = 0; i < RX_RING_SIZE; i++) { | ||
1186 | hmp->rx_ring[i].status_n_length = 0; | ||
1187 | hmp->rx_skbuff[i] = NULL; | ||
1188 | } | ||
1189 | /* Fill in the Rx buffers. Handle allocation failure gracefully. */ | ||
1190 | for (i = 0; i < RX_RING_SIZE; i++) { | ||
1191 | struct sk_buff *skb = dev_alloc_skb(hmp->rx_buf_sz + 2); | ||
1192 | hmp->rx_skbuff[i] = skb; | ||
1193 | if (skb == NULL) | ||
1194 | break; | ||
1195 | skb->dev = dev; /* Mark as being used by this device. */ | ||
1196 | skb_reserve(skb, 2); /* 16 byte align the IP header. */ | ||
1197 | hmp->rx_ring[i].addr = cpu_to_leXX(pci_map_single(hmp->pci_dev, | ||
1198 | skb->data, hmp->rx_buf_sz, PCI_DMA_FROMDEVICE)); | ||
1199 | /* -2 because it doesn't REALLY have that first 2 bytes -KDU */ | ||
1200 | hmp->rx_ring[i].status_n_length = cpu_to_le32(DescOwn | | ||
1201 | DescEndPacket | DescIntr | (hmp->rx_buf_sz -2)); | ||
1202 | } | ||
1203 | hmp->dirty_rx = (unsigned int)(i - RX_RING_SIZE); | ||
1204 | hmp->rx_ring[RX_RING_SIZE-1].status_n_length |= cpu_to_le32(DescEndRing); | ||
1205 | |||
1206 | for (i = 0; i < TX_RING_SIZE; i++) { | ||
1207 | hmp->tx_skbuff[i] = NULL; | ||
1208 | hmp->tx_ring[i].status_n_length = 0; | ||
1209 | } | ||
1210 | /* Mark the last entry of the ring */ | ||
1211 | hmp->tx_ring[TX_RING_SIZE-1].status_n_length |= cpu_to_le32(DescEndRing); | ||
1212 | } | ||
1213 | |||
1214 | |||
1215 | static netdev_tx_t hamachi_start_xmit(struct sk_buff *skb, | ||
1216 | struct net_device *dev) | ||
1217 | { | ||
1218 | struct hamachi_private *hmp = netdev_priv(dev); | ||
1219 | unsigned entry; | ||
1220 | u16 status; | ||
1221 | |||
1222 | /* Ok, now make sure that the queue has space before trying to | ||
1223 | add another skbuff. if we return non-zero the scheduler | ||
1224 | should interpret this as a queue full and requeue the buffer | ||
1225 | for later. | ||
1226 | */ | ||
1227 | if (hmp->tx_full) { | ||
1228 | /* We should NEVER reach this point -KDU */ | ||
1229 | printk(KERN_WARNING "%s: Hamachi transmit queue full at slot %d.\n",dev->name, hmp->cur_tx); | ||
1230 | |||
1231 | /* Wake the potentially-idle transmit channel. */ | ||
1232 | /* If we don't need to read status, DON'T -KDU */ | ||
1233 | status=readw(hmp->base + TxStatus); | ||
1234 | if( !(status & 0x0001) || (status & 0x0002)) | ||
1235 | writew(0x0001, hmp->base + TxCmd); | ||
1236 | return NETDEV_TX_BUSY; | ||
1237 | } | ||
1238 | |||
1239 | /* Caution: the write order is important here, set the field | ||
1240 | with the "ownership" bits last. */ | ||
1241 | |||
1242 | /* Calculate the next Tx descriptor entry. */ | ||
1243 | entry = hmp->cur_tx % TX_RING_SIZE; | ||
1244 | |||
1245 | hmp->tx_skbuff[entry] = skb; | ||
1246 | |||
1247 | hmp->tx_ring[entry].addr = cpu_to_leXX(pci_map_single(hmp->pci_dev, | ||
1248 | skb->data, skb->len, PCI_DMA_TODEVICE)); | ||
1249 | |||
1250 | /* Hmmmm, could probably put a DescIntr on these, but the way | ||
1251 | the driver is currently coded makes Tx interrupts unnecessary | ||
1252 | since the clearing of the Tx ring is handled by the start_xmit | ||
1253 | routine. This organization helps mitigate the interrupts a | ||
1254 | bit and probably renders the max_tx_latency param useless. | ||
1255 | |||
1256 | Update: Putting a DescIntr bit on all of the descriptors and | ||
1257 | mitigating interrupt frequency with the tx_min_pkt parameter. -KDU | ||
1258 | */ | ||
1259 | if (entry >= TX_RING_SIZE-1) /* Wrap ring */ | ||
1260 | hmp->tx_ring[entry].status_n_length = cpu_to_le32(DescOwn | | ||
1261 | DescEndPacket | DescEndRing | DescIntr | skb->len); | ||
1262 | else | ||
1263 | hmp->tx_ring[entry].status_n_length = cpu_to_le32(DescOwn | | ||
1264 | DescEndPacket | DescIntr | skb->len); | ||
1265 | hmp->cur_tx++; | ||
1266 | |||
1267 | /* Non-x86 Todo: explicitly flush cache lines here. */ | ||
1268 | |||
1269 | /* Wake the potentially-idle transmit channel. */ | ||
1270 | /* If we don't need to read status, DON'T -KDU */ | ||
1271 | status=readw(hmp->base + TxStatus); | ||
1272 | if( !(status & 0x0001) || (status & 0x0002)) | ||
1273 | writew(0x0001, hmp->base + TxCmd); | ||
1274 | |||
1275 | /* Immediately before returning, let's clear as many entries as we can. */ | ||
1276 | hamachi_tx(dev); | ||
1277 | |||
1278 | /* We should kick the bottom half here, since we are not accepting | ||
1279 | * interrupts with every packet. i.e. realize that Gigabit ethernet | ||
1280 | * can transmit faster than ordinary machines can load packets; | ||
1281 | * hence, any packet that got put off because we were in the transmit | ||
1282 | * routine should IMMEDIATELY get a chance to be re-queued. -KDU | ||
1283 | */ | ||
1284 | if ((hmp->cur_tx - hmp->dirty_tx) < (TX_RING_SIZE - 4)) | ||
1285 | netif_wake_queue(dev); /* Typical path */ | ||
1286 | else { | ||
1287 | hmp->tx_full = 1; | ||
1288 | netif_stop_queue(dev); | ||
1289 | } | ||
1290 | |||
1291 | if (hamachi_debug > 4) { | ||
1292 | printk(KERN_DEBUG "%s: Hamachi transmit frame #%d queued in slot %d.\n", | ||
1293 | dev->name, hmp->cur_tx, entry); | ||
1294 | } | ||
1295 | return NETDEV_TX_OK; | ||
1296 | } | ||
1297 | |||
1298 | /* The interrupt handler does all of the Rx thread work and cleans up | ||
1299 | after the Tx thread. */ | ||
1300 | static irqreturn_t hamachi_interrupt(int irq, void *dev_instance) | ||
1301 | { | ||
1302 | struct net_device *dev = dev_instance; | ||
1303 | struct hamachi_private *hmp = netdev_priv(dev); | ||
1304 | void __iomem *ioaddr = hmp->base; | ||
1305 | long boguscnt = max_interrupt_work; | ||
1306 | int handled = 0; | ||
1307 | |||
1308 | #ifndef final_version /* Can never occur. */ | ||
1309 | if (dev == NULL) { | ||
1310 | printk (KERN_ERR "hamachi_interrupt(): irq %d for unknown device.\n", irq); | ||
1311 | return IRQ_NONE; | ||
1312 | } | ||
1313 | #endif | ||
1314 | |||
1315 | spin_lock(&hmp->lock); | ||
1316 | |||
1317 | do { | ||
1318 | u32 intr_status = readl(ioaddr + InterruptClear); | ||
1319 | |||
1320 | if (hamachi_debug > 4) | ||
1321 | printk(KERN_DEBUG "%s: Hamachi interrupt, status %4.4x.\n", | ||
1322 | dev->name, intr_status); | ||
1323 | |||
1324 | if (intr_status == 0) | ||
1325 | break; | ||
1326 | |||
1327 | handled = 1; | ||
1328 | |||
1329 | if (intr_status & IntrRxDone) | ||
1330 | hamachi_rx(dev); | ||
1331 | |||
1332 | if (intr_status & IntrTxDone){ | ||
1333 | /* This code should RARELY need to execute. After all, this is | ||
1334 | * a gigabit link, it should consume packets as fast as we put | ||
1335 | * them in AND we clear the Tx ring in hamachi_start_xmit(). | ||
1336 | */ | ||
1337 | if (hmp->tx_full){ | ||
1338 | for (; hmp->cur_tx - hmp->dirty_tx > 0; hmp->dirty_tx++){ | ||
1339 | int entry = hmp->dirty_tx % TX_RING_SIZE; | ||
1340 | struct sk_buff *skb; | ||
1341 | |||
1342 | if (hmp->tx_ring[entry].status_n_length & cpu_to_le32(DescOwn)) | ||
1343 | break; | ||
1344 | skb = hmp->tx_skbuff[entry]; | ||
1345 | /* Free the original skb. */ | ||
1346 | if (skb){ | ||
1347 | pci_unmap_single(hmp->pci_dev, | ||
1348 | leXX_to_cpu(hmp->tx_ring[entry].addr), | ||
1349 | skb->len, | ||
1350 | PCI_DMA_TODEVICE); | ||
1351 | dev_kfree_skb_irq(skb); | ||
1352 | hmp->tx_skbuff[entry] = NULL; | ||
1353 | } | ||
1354 | hmp->tx_ring[entry].status_n_length = 0; | ||
1355 | if (entry >= TX_RING_SIZE-1) | ||
1356 | hmp->tx_ring[TX_RING_SIZE-1].status_n_length |= | ||
1357 | cpu_to_le32(DescEndRing); | ||
1358 | dev->stats.tx_packets++; | ||
1359 | } | ||
1360 | if (hmp->cur_tx - hmp->dirty_tx < TX_RING_SIZE - 4){ | ||
1361 | /* The ring is no longer full */ | ||
1362 | hmp->tx_full = 0; | ||
1363 | netif_wake_queue(dev); | ||
1364 | } | ||
1365 | } else { | ||
1366 | netif_wake_queue(dev); | ||
1367 | } | ||
1368 | } | ||
1369 | |||
1370 | |||
1371 | /* Abnormal error summary/uncommon events handlers. */ | ||
1372 | if (intr_status & | ||
1373 | (IntrTxPCIFault | IntrTxPCIErr | IntrRxPCIFault | IntrRxPCIErr | | ||
1374 | LinkChange | NegotiationChange | StatsMax)) | ||
1375 | hamachi_error(dev, intr_status); | ||
1376 | |||
1377 | if (--boguscnt < 0) { | ||
1378 | printk(KERN_WARNING "%s: Too much work at interrupt, status=0x%4.4x.\n", | ||
1379 | dev->name, intr_status); | ||
1380 | break; | ||
1381 | } | ||
1382 | } while (1); | ||
1383 | |||
1384 | if (hamachi_debug > 3) | ||
1385 | printk(KERN_DEBUG "%s: exiting interrupt, status=%#4.4x.\n", | ||
1386 | dev->name, readl(ioaddr + IntrStatus)); | ||
1387 | |||
1388 | #ifndef final_version | ||
1389 | /* Code that should never be run! Perhaps remove after testing.. */ | ||
1390 | { | ||
1391 | static int stopit = 10; | ||
1392 | if (dev->start == 0 && --stopit < 0) { | ||
1393 | printk(KERN_ERR "%s: Emergency stop, looping startup interrupt.\n", | ||
1394 | dev->name); | ||
1395 | free_irq(irq, dev); | ||
1396 | } | ||
1397 | } | ||
1398 | #endif | ||
1399 | |||
1400 | spin_unlock(&hmp->lock); | ||
1401 | return IRQ_RETVAL(handled); | ||
1402 | } | ||
1403 | |||
1404 | /* This routine is logically part of the interrupt handler, but separated | ||
1405 | for clarity and better register allocation. */ | ||
1406 | static int hamachi_rx(struct net_device *dev) | ||
1407 | { | ||
1408 | struct hamachi_private *hmp = netdev_priv(dev); | ||
1409 | int entry = hmp->cur_rx % RX_RING_SIZE; | ||
1410 | int boguscnt = (hmp->dirty_rx + RX_RING_SIZE) - hmp->cur_rx; | ||
1411 | |||
1412 | if (hamachi_debug > 4) { | ||
1413 | printk(KERN_DEBUG " In hamachi_rx(), entry %d status %4.4x.\n", | ||
1414 | entry, hmp->rx_ring[entry].status_n_length); | ||
1415 | } | ||
1416 | |||
1417 | /* If EOP is set on the next entry, it's a new packet. Send it up. */ | ||
1418 | while (1) { | ||
1419 | struct hamachi_desc *desc = &(hmp->rx_ring[entry]); | ||
1420 | u32 desc_status = le32_to_cpu(desc->status_n_length); | ||
1421 | u16 data_size = desc_status; /* Implicit truncate */ | ||
1422 | u8 *buf_addr; | ||
1423 | s32 frame_status; | ||
1424 | |||
1425 | if (desc_status & DescOwn) | ||
1426 | break; | ||
1427 | pci_dma_sync_single_for_cpu(hmp->pci_dev, | ||
1428 | leXX_to_cpu(desc->addr), | ||
1429 | hmp->rx_buf_sz, | ||
1430 | PCI_DMA_FROMDEVICE); | ||
1431 | buf_addr = (u8 *) hmp->rx_skbuff[entry]->data; | ||
1432 | frame_status = get_unaligned_le32(&(buf_addr[data_size - 12])); | ||
1433 | if (hamachi_debug > 4) | ||
1434 | printk(KERN_DEBUG " hamachi_rx() status was %8.8x.\n", | ||
1435 | frame_status); | ||
1436 | if (--boguscnt < 0) | ||
1437 | break; | ||
1438 | if ( ! (desc_status & DescEndPacket)) { | ||
1439 | printk(KERN_WARNING "%s: Oversized Ethernet frame spanned " | ||
1440 | "multiple buffers, entry %#x length %d status %4.4x!\n", | ||
1441 | dev->name, hmp->cur_rx, data_size, desc_status); | ||
1442 | printk(KERN_WARNING "%s: Oversized Ethernet frame %p vs %p.\n", | ||
1443 | dev->name, desc, &hmp->rx_ring[hmp->cur_rx % RX_RING_SIZE]); | ||
1444 | printk(KERN_WARNING "%s: Oversized Ethernet frame -- next status %x/%x last status %x.\n", | ||
1445 | dev->name, | ||
1446 | le32_to_cpu(hmp->rx_ring[(hmp->cur_rx+1) % RX_RING_SIZE].status_n_length) & 0xffff0000, | ||
1447 | le32_to_cpu(hmp->rx_ring[(hmp->cur_rx+1) % RX_RING_SIZE].status_n_length) & 0x0000ffff, | ||
1448 | le32_to_cpu(hmp->rx_ring[(hmp->cur_rx-1) % RX_RING_SIZE].status_n_length)); | ||
1449 | dev->stats.rx_length_errors++; | ||
1450 | } /* else Omit for prototype errata??? */ | ||
1451 | if (frame_status & 0x00380000) { | ||
1452 | /* There was an error. */ | ||
1453 | if (hamachi_debug > 2) | ||
1454 | printk(KERN_DEBUG " hamachi_rx() Rx error was %8.8x.\n", | ||
1455 | frame_status); | ||
1456 | dev->stats.rx_errors++; | ||
1457 | if (frame_status & 0x00600000) | ||
1458 | dev->stats.rx_length_errors++; | ||
1459 | if (frame_status & 0x00080000) | ||
1460 | dev->stats.rx_frame_errors++; | ||
1461 | if (frame_status & 0x00100000) | ||
1462 | dev->stats.rx_crc_errors++; | ||
1463 | if (frame_status < 0) | ||
1464 | dev->stats.rx_dropped++; | ||
1465 | } else { | ||
1466 | struct sk_buff *skb; | ||
1467 | /* Omit CRC */ | ||
1468 | u16 pkt_len = (frame_status & 0x07ff) - 4; | ||
1469 | #ifdef RX_CHECKSUM | ||
1470 | u32 pfck = *(u32 *) &buf_addr[data_size - 8]; | ||
1471 | #endif | ||
1472 | |||
1473 | |||
1474 | #ifndef final_version | ||
1475 | if (hamachi_debug > 4) | ||
1476 | printk(KERN_DEBUG " hamachi_rx() normal Rx pkt length %d" | ||
1477 | " of %d, bogus_cnt %d.\n", | ||
1478 | pkt_len, data_size, boguscnt); | ||
1479 | if (hamachi_debug > 5) | ||
1480 | printk(KERN_DEBUG"%s: rx status %8.8x %8.8x %8.8x %8.8x %8.8x.\n", | ||
1481 | dev->name, | ||
1482 | *(s32*)&(buf_addr[data_size - 20]), | ||
1483 | *(s32*)&(buf_addr[data_size - 16]), | ||
1484 | *(s32*)&(buf_addr[data_size - 12]), | ||
1485 | *(s32*)&(buf_addr[data_size - 8]), | ||
1486 | *(s32*)&(buf_addr[data_size - 4])); | ||
1487 | #endif | ||
1488 | /* Check if the packet is long enough to accept without copying | ||
1489 | to a minimally-sized skbuff. */ | ||
1490 | if (pkt_len < rx_copybreak && | ||
1491 | (skb = dev_alloc_skb(pkt_len + 2)) != NULL) { | ||
1492 | #ifdef RX_CHECKSUM | ||
1493 | printk(KERN_ERR "%s: rx_copybreak non-zero " | ||
1494 | "not good with RX_CHECKSUM\n", dev->name); | ||
1495 | #endif | ||
1496 | skb_reserve(skb, 2); /* 16 byte align the IP header */ | ||
1497 | pci_dma_sync_single_for_cpu(hmp->pci_dev, | ||
1498 | leXX_to_cpu(hmp->rx_ring[entry].addr), | ||
1499 | hmp->rx_buf_sz, | ||
1500 | PCI_DMA_FROMDEVICE); | ||
1501 | /* Call copy + cksum if available. */ | ||
1502 | #if 1 || USE_IP_COPYSUM | ||
1503 | skb_copy_to_linear_data(skb, | ||
1504 | hmp->rx_skbuff[entry]->data, pkt_len); | ||
1505 | skb_put(skb, pkt_len); | ||
1506 | #else | ||
1507 | memcpy(skb_put(skb, pkt_len), hmp->rx_ring_dma | ||
1508 | + entry*sizeof(*desc), pkt_len); | ||
1509 | #endif | ||
1510 | pci_dma_sync_single_for_device(hmp->pci_dev, | ||
1511 | leXX_to_cpu(hmp->rx_ring[entry].addr), | ||
1512 | hmp->rx_buf_sz, | ||
1513 | PCI_DMA_FROMDEVICE); | ||
1514 | } else { | ||
1515 | pci_unmap_single(hmp->pci_dev, | ||
1516 | leXX_to_cpu(hmp->rx_ring[entry].addr), | ||
1517 | hmp->rx_buf_sz, PCI_DMA_FROMDEVICE); | ||
1518 | skb_put(skb = hmp->rx_skbuff[entry], pkt_len); | ||
1519 | hmp->rx_skbuff[entry] = NULL; | ||
1520 | } | ||
1521 | skb->protocol = eth_type_trans(skb, dev); | ||
1522 | |||
1523 | |||
1524 | #ifdef RX_CHECKSUM | ||
1525 | /* TCP or UDP on ipv4, DIX encoding */ | ||
1526 | if (pfck>>24 == 0x91 || pfck>>24 == 0x51) { | ||
1527 | struct iphdr *ih = (struct iphdr *) skb->data; | ||
1528 | /* Check that IP packet is at least 46 bytes, otherwise, | ||
1529 | * there may be pad bytes included in the hardware checksum. | ||
1530 | * This wouldn't happen if everyone padded with 0. | ||
1531 | */ | ||
1532 | if (ntohs(ih->tot_len) >= 46){ | ||
1533 | /* don't worry about frags */ | ||
1534 | if (!(ih->frag_off & cpu_to_be16(IP_MF|IP_OFFSET))) { | ||
1535 | u32 inv = *(u32 *) &buf_addr[data_size - 16]; | ||
1536 | u32 *p = (u32 *) &buf_addr[data_size - 20]; | ||
1537 | register u32 crc, p_r, p_r1; | ||
1538 | |||
1539 | if (inv & 4) { | ||
1540 | inv &= ~4; | ||
1541 | --p; | ||
1542 | } | ||
1543 | p_r = *p; | ||
1544 | p_r1 = *(p-1); | ||
1545 | switch (inv) { | ||
1546 | case 0: | ||
1547 | crc = (p_r & 0xffff) + (p_r >> 16); | ||
1548 | break; | ||
1549 | case 1: | ||
1550 | crc = (p_r >> 16) + (p_r & 0xffff) | ||
1551 | + (p_r1 >> 16 & 0xff00); | ||
1552 | break; | ||
1553 | case 2: | ||
1554 | crc = p_r + (p_r1 >> 16); | ||
1555 | break; | ||
1556 | case 3: | ||
1557 | crc = p_r + (p_r1 & 0xff00) + (p_r1 >> 16); | ||
1558 | break; | ||
1559 | default: /*NOTREACHED*/ crc = 0; | ||
1560 | } | ||
1561 | if (crc & 0xffff0000) { | ||
1562 | crc &= 0xffff; | ||
1563 | ++crc; | ||
1564 | } | ||
1565 | /* tcp/udp will add in pseudo */ | ||
1566 | skb->csum = ntohs(pfck & 0xffff); | ||
1567 | if (skb->csum > crc) | ||
1568 | skb->csum -= crc; | ||
1569 | else | ||
1570 | skb->csum += (~crc & 0xffff); | ||
1571 | /* | ||
1572 | * could do the pseudo myself and return | ||
1573 | * CHECKSUM_UNNECESSARY | ||
1574 | */ | ||
1575 | skb->ip_summed = CHECKSUM_COMPLETE; | ||
1576 | } | ||
1577 | } | ||
1578 | } | ||
1579 | #endif /* RX_CHECKSUM */ | ||
1580 | |||
1581 | netif_rx(skb); | ||
1582 | dev->stats.rx_packets++; | ||
1583 | } | ||
1584 | entry = (++hmp->cur_rx) % RX_RING_SIZE; | ||
1585 | } | ||
1586 | |||
1587 | /* Refill the Rx ring buffers. */ | ||
1588 | for (; hmp->cur_rx - hmp->dirty_rx > 0; hmp->dirty_rx++) { | ||
1589 | struct hamachi_desc *desc; | ||
1590 | |||
1591 | entry = hmp->dirty_rx % RX_RING_SIZE; | ||
1592 | desc = &(hmp->rx_ring[entry]); | ||
1593 | if (hmp->rx_skbuff[entry] == NULL) { | ||
1594 | struct sk_buff *skb = dev_alloc_skb(hmp->rx_buf_sz + 2); | ||
1595 | |||
1596 | hmp->rx_skbuff[entry] = skb; | ||
1597 | if (skb == NULL) | ||
1598 | break; /* Better luck next round. */ | ||
1599 | skb->dev = dev; /* Mark as being used by this device. */ | ||
1600 | skb_reserve(skb, 2); /* Align IP on 16 byte boundaries */ | ||
1601 | desc->addr = cpu_to_leXX(pci_map_single(hmp->pci_dev, | ||
1602 | skb->data, hmp->rx_buf_sz, PCI_DMA_FROMDEVICE)); | ||
1603 | } | ||
1604 | desc->status_n_length = cpu_to_le32(hmp->rx_buf_sz); | ||
1605 | if (entry >= RX_RING_SIZE-1) | ||
1606 | desc->status_n_length |= cpu_to_le32(DescOwn | | ||
1607 | DescEndPacket | DescEndRing | DescIntr); | ||
1608 | else | ||
1609 | desc->status_n_length |= cpu_to_le32(DescOwn | | ||
1610 | DescEndPacket | DescIntr); | ||
1611 | } | ||
1612 | |||
1613 | /* Restart Rx engine if stopped. */ | ||
1614 | /* If we don't need to check status, don't. -KDU */ | ||
1615 | if (readw(hmp->base + RxStatus) & 0x0002) | ||
1616 | writew(0x0001, hmp->base + RxCmd); | ||
1617 | |||
1618 | return 0; | ||
1619 | } | ||
1620 | |||
1621 | /* This is more properly named "uncommon interrupt events", as it covers more | ||
1622 | than just errors. */ | ||
1623 | static void hamachi_error(struct net_device *dev, int intr_status) | ||
1624 | { | ||
1625 | struct hamachi_private *hmp = netdev_priv(dev); | ||
1626 | void __iomem *ioaddr = hmp->base; | ||
1627 | |||
1628 | if (intr_status & (LinkChange|NegotiationChange)) { | ||
1629 | if (hamachi_debug > 1) | ||
1630 | printk(KERN_INFO "%s: Link changed: AutoNegotiation Ctrl" | ||
1631 | " %4.4x, Status %4.4x %4.4x Intr status %4.4x.\n", | ||
1632 | dev->name, readw(ioaddr + 0x0E0), readw(ioaddr + 0x0E2), | ||
1633 | readw(ioaddr + ANLinkPartnerAbility), | ||
1634 | readl(ioaddr + IntrStatus)); | ||
1635 | if (readw(ioaddr + ANStatus) & 0x20) | ||
1636 | writeb(0x01, ioaddr + LEDCtrl); | ||
1637 | else | ||
1638 | writeb(0x03, ioaddr + LEDCtrl); | ||
1639 | } | ||
1640 | if (intr_status & StatsMax) { | ||
1641 | hamachi_get_stats(dev); | ||
1642 | /* Read the overflow bits to clear. */ | ||
1643 | readl(ioaddr + 0x370); | ||
1644 | readl(ioaddr + 0x3F0); | ||
1645 | } | ||
1646 | if ((intr_status & ~(LinkChange|StatsMax|NegotiationChange|IntrRxDone|IntrTxDone)) && | ||
1647 | hamachi_debug) | ||
1648 | printk(KERN_ERR "%s: Something Wicked happened! %4.4x.\n", | ||
1649 | dev->name, intr_status); | ||
1650 | /* Hmmmmm, it's not clear how to recover from PCI faults. */ | ||
1651 | if (intr_status & (IntrTxPCIErr | IntrTxPCIFault)) | ||
1652 | dev->stats.tx_fifo_errors++; | ||
1653 | if (intr_status & (IntrRxPCIErr | IntrRxPCIFault)) | ||
1654 | dev->stats.rx_fifo_errors++; | ||
1655 | } | ||
1656 | |||
1657 | static int hamachi_close(struct net_device *dev) | ||
1658 | { | ||
1659 | struct hamachi_private *hmp = netdev_priv(dev); | ||
1660 | void __iomem *ioaddr = hmp->base; | ||
1661 | struct sk_buff *skb; | ||
1662 | int i; | ||
1663 | |||
1664 | netif_stop_queue(dev); | ||
1665 | |||
1666 | if (hamachi_debug > 1) { | ||
1667 | printk(KERN_DEBUG "%s: Shutting down ethercard, status was Tx %4.4x Rx %4.4x Int %2.2x.\n", | ||
1668 | dev->name, readw(ioaddr + TxStatus), | ||
1669 | readw(ioaddr + RxStatus), readl(ioaddr + IntrStatus)); | ||
1670 | printk(KERN_DEBUG "%s: Queue pointers were Tx %d / %d, Rx %d / %d.\n", | ||
1671 | dev->name, hmp->cur_tx, hmp->dirty_tx, hmp->cur_rx, hmp->dirty_rx); | ||
1672 | } | ||
1673 | |||
1674 | /* Disable interrupts by clearing the interrupt mask. */ | ||
1675 | writel(0x0000, ioaddr + InterruptEnable); | ||
1676 | |||
1677 | /* Stop the chip's Tx and Rx processes. */ | ||
1678 | writel(2, ioaddr + RxCmd); | ||
1679 | writew(2, ioaddr + TxCmd); | ||
1680 | |||
1681 | #ifdef __i386__ | ||
1682 | if (hamachi_debug > 2) { | ||
1683 | printk(KERN_DEBUG " Tx ring at %8.8x:\n", | ||
1684 | (int)hmp->tx_ring_dma); | ||
1685 | for (i = 0; i < TX_RING_SIZE; i++) | ||
1686 | printk(KERN_DEBUG " %c #%d desc. %8.8x %8.8x.\n", | ||
1687 | readl(ioaddr + TxCurPtr) == (long)&hmp->tx_ring[i] ? '>' : ' ', | ||
1688 | i, hmp->tx_ring[i].status_n_length, hmp->tx_ring[i].addr); | ||
1689 | printk(KERN_DEBUG " Rx ring %8.8x:\n", | ||
1690 | (int)hmp->rx_ring_dma); | ||
1691 | for (i = 0; i < RX_RING_SIZE; i++) { | ||
1692 | printk(KERN_DEBUG " %c #%d desc. %4.4x %8.8x\n", | ||
1693 | readl(ioaddr + RxCurPtr) == (long)&hmp->rx_ring[i] ? '>' : ' ', | ||
1694 | i, hmp->rx_ring[i].status_n_length, hmp->rx_ring[i].addr); | ||
1695 | if (hamachi_debug > 6) { | ||
1696 | if (*(u8*)hmp->rx_skbuff[i]->data != 0x69) { | ||
1697 | u16 *addr = (u16 *) | ||
1698 | hmp->rx_skbuff[i]->data; | ||
1699 | int j; | ||
1700 | printk(KERN_DEBUG "Addr: "); | ||
1701 | for (j = 0; j < 0x50; j++) | ||
1702 | printk(" %4.4x", addr[j]); | ||
1703 | printk("\n"); | ||
1704 | } | ||
1705 | } | ||
1706 | } | ||
1707 | } | ||
1708 | #endif /* __i386__ debugging only */ | ||
1709 | |||
1710 | free_irq(dev->irq, dev); | ||
1711 | |||
1712 | del_timer_sync(&hmp->timer); | ||
1713 | |||
1714 | /* Free all the skbuffs in the Rx queue. */ | ||
1715 | for (i = 0; i < RX_RING_SIZE; i++) { | ||
1716 | skb = hmp->rx_skbuff[i]; | ||
1717 | hmp->rx_ring[i].status_n_length = 0; | ||
1718 | if (skb) { | ||
1719 | pci_unmap_single(hmp->pci_dev, | ||
1720 | leXX_to_cpu(hmp->rx_ring[i].addr), | ||
1721 | hmp->rx_buf_sz, PCI_DMA_FROMDEVICE); | ||
1722 | dev_kfree_skb(skb); | ||
1723 | hmp->rx_skbuff[i] = NULL; | ||
1724 | } | ||
1725 | hmp->rx_ring[i].addr = cpu_to_leXX(0xBADF00D0); /* An invalid address. */ | ||
1726 | } | ||
1727 | for (i = 0; i < TX_RING_SIZE; i++) { | ||
1728 | skb = hmp->tx_skbuff[i]; | ||
1729 | if (skb) { | ||
1730 | pci_unmap_single(hmp->pci_dev, | ||
1731 | leXX_to_cpu(hmp->tx_ring[i].addr), | ||
1732 | skb->len, PCI_DMA_TODEVICE); | ||
1733 | dev_kfree_skb(skb); | ||
1734 | hmp->tx_skbuff[i] = NULL; | ||
1735 | } | ||
1736 | } | ||
1737 | |||
1738 | writeb(0x00, ioaddr + LEDCtrl); | ||
1739 | |||
1740 | return 0; | ||
1741 | } | ||
1742 | |||
1743 | static struct net_device_stats *hamachi_get_stats(struct net_device *dev) | ||
1744 | { | ||
1745 | struct hamachi_private *hmp = netdev_priv(dev); | ||
1746 | void __iomem *ioaddr = hmp->base; | ||
1747 | |||
1748 | /* We should lock this segment of code for SMP eventually, although | ||
1749 | the vulnerability window is very small and statistics are | ||
1750 | non-critical. */ | ||
1751 | /* Ok, what goes here? This appears to be stuck at 21 packets | ||
1752 | according to ifconfig. It does get incremented in hamachi_tx(), | ||
1753 | so I think I'll comment it out here and see if better things | ||
1754 | happen. | ||
1755 | */ | ||
1756 | /* dev->stats.tx_packets = readl(ioaddr + 0x000); */ | ||
1757 | |||
1758 | /* Total Uni+Brd+Multi */ | ||
1759 | dev->stats.rx_bytes = readl(ioaddr + 0x330); | ||
1760 | /* Total Uni+Brd+Multi */ | ||
1761 | dev->stats.tx_bytes = readl(ioaddr + 0x3B0); | ||
1762 | /* Multicast Rx */ | ||
1763 | dev->stats.multicast = readl(ioaddr + 0x320); | ||
1764 | |||
1765 | /* Over+Undersized */ | ||
1766 | dev->stats.rx_length_errors = readl(ioaddr + 0x368); | ||
1767 | /* Jabber */ | ||
1768 | dev->stats.rx_over_errors = readl(ioaddr + 0x35C); | ||
1769 | /* Jabber */ | ||
1770 | dev->stats.rx_crc_errors = readl(ioaddr + 0x360); | ||
1771 | /* Symbol Errs */ | ||
1772 | dev->stats.rx_frame_errors = readl(ioaddr + 0x364); | ||
1773 | /* Dropped */ | ||
1774 | dev->stats.rx_missed_errors = readl(ioaddr + 0x36C); | ||
1775 | |||
1776 | return &dev->stats; | ||
1777 | } | ||
1778 | |||
1779 | static void set_rx_mode(struct net_device *dev) | ||
1780 | { | ||
1781 | struct hamachi_private *hmp = netdev_priv(dev); | ||
1782 | void __iomem *ioaddr = hmp->base; | ||
1783 | |||
1784 | if (dev->flags & IFF_PROMISC) { /* Set promiscuous. */ | ||
1785 | writew(0x000F, ioaddr + AddrMode); | ||
1786 | } else if ((netdev_mc_count(dev) > 63) || (dev->flags & IFF_ALLMULTI)) { | ||
1787 | /* Too many to match, or accept all multicasts. */ | ||
1788 | writew(0x000B, ioaddr + AddrMode); | ||
1789 | } else if (!netdev_mc_empty(dev)) { /* Must use the CAM filter. */ | ||
1790 | struct netdev_hw_addr *ha; | ||
1791 | int i = 0; | ||
1792 | |||
1793 | netdev_for_each_mc_addr(ha, dev) { | ||
1794 | writel(*(u32 *)(ha->addr), ioaddr + 0x100 + i*8); | ||
1795 | writel(0x20000 | (*(u16 *)&ha->addr[4]), | ||
1796 | ioaddr + 0x104 + i*8); | ||
1797 | i++; | ||
1798 | } | ||
1799 | /* Clear remaining entries. */ | ||
1800 | for (; i < 64; i++) | ||
1801 | writel(0, ioaddr + 0x104 + i*8); | ||
1802 | writew(0x0003, ioaddr + AddrMode); | ||
1803 | } else { /* Normal, unicast/broadcast-only mode. */ | ||
1804 | writew(0x0001, ioaddr + AddrMode); | ||
1805 | } | ||
1806 | } | ||
1807 | |||
1808 | static int check_if_running(struct net_device *dev) | ||
1809 | { | ||
1810 | if (!netif_running(dev)) | ||
1811 | return -EINVAL; | ||
1812 | return 0; | ||
1813 | } | ||
1814 | |||
1815 | static void hamachi_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) | ||
1816 | { | ||
1817 | struct hamachi_private *np = netdev_priv(dev); | ||
1818 | strcpy(info->driver, DRV_NAME); | ||
1819 | strcpy(info->version, DRV_VERSION); | ||
1820 | strcpy(info->bus_info, pci_name(np->pci_dev)); | ||
1821 | } | ||
1822 | |||
1823 | static int hamachi_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd) | ||
1824 | { | ||
1825 | struct hamachi_private *np = netdev_priv(dev); | ||
1826 | spin_lock_irq(&np->lock); | ||
1827 | mii_ethtool_gset(&np->mii_if, ecmd); | ||
1828 | spin_unlock_irq(&np->lock); | ||
1829 | return 0; | ||
1830 | } | ||
1831 | |||
1832 | static int hamachi_set_settings(struct net_device *dev, struct ethtool_cmd *ecmd) | ||
1833 | { | ||
1834 | struct hamachi_private *np = netdev_priv(dev); | ||
1835 | int res; | ||
1836 | spin_lock_irq(&np->lock); | ||
1837 | res = mii_ethtool_sset(&np->mii_if, ecmd); | ||
1838 | spin_unlock_irq(&np->lock); | ||
1839 | return res; | ||
1840 | } | ||
1841 | |||
1842 | static int hamachi_nway_reset(struct net_device *dev) | ||
1843 | { | ||
1844 | struct hamachi_private *np = netdev_priv(dev); | ||
1845 | return mii_nway_restart(&np->mii_if); | ||
1846 | } | ||
1847 | |||
1848 | static u32 hamachi_get_link(struct net_device *dev) | ||
1849 | { | ||
1850 | struct hamachi_private *np = netdev_priv(dev); | ||
1851 | return mii_link_ok(&np->mii_if); | ||
1852 | } | ||
1853 | |||
1854 | static const struct ethtool_ops ethtool_ops = { | ||
1855 | .begin = check_if_running, | ||
1856 | .get_drvinfo = hamachi_get_drvinfo, | ||
1857 | .get_settings = hamachi_get_settings, | ||
1858 | .set_settings = hamachi_set_settings, | ||
1859 | .nway_reset = hamachi_nway_reset, | ||
1860 | .get_link = hamachi_get_link, | ||
1861 | }; | ||
1862 | |||
1863 | static const struct ethtool_ops ethtool_ops_no_mii = { | ||
1864 | .begin = check_if_running, | ||
1865 | .get_drvinfo = hamachi_get_drvinfo, | ||
1866 | }; | ||
1867 | |||
1868 | static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd) | ||
1869 | { | ||
1870 | struct hamachi_private *np = netdev_priv(dev); | ||
1871 | struct mii_ioctl_data *data = if_mii(rq); | ||
1872 | int rc; | ||
1873 | |||
1874 | if (!netif_running(dev)) | ||
1875 | return -EINVAL; | ||
1876 | |||
1877 | if (cmd == (SIOCDEVPRIVATE+3)) { /* set rx,tx intr params */ | ||
1878 | u32 *d = (u32 *)&rq->ifr_ifru; | ||
1879 | /* Should add this check here or an ordinary user can do nasty | ||
1880 | * things. -KDU | ||
1881 | * | ||
1882 | * TODO: Shut down the Rx and Tx engines while doing this. | ||
1883 | */ | ||
1884 | if (!capable(CAP_NET_ADMIN)) | ||
1885 | return -EPERM; | ||
1886 | writel(d[0], np->base + TxIntrCtrl); | ||
1887 | writel(d[1], np->base + RxIntrCtrl); | ||
1888 | printk(KERN_NOTICE "%s: tx %08x, rx %08x intr\n", dev->name, | ||
1889 | (u32) readl(np->base + TxIntrCtrl), | ||
1890 | (u32) readl(np->base + RxIntrCtrl)); | ||
1891 | rc = 0; | ||
1892 | } | ||
1893 | |||
1894 | else { | ||
1895 | spin_lock_irq(&np->lock); | ||
1896 | rc = generic_mii_ioctl(&np->mii_if, data, cmd, NULL); | ||
1897 | spin_unlock_irq(&np->lock); | ||
1898 | } | ||
1899 | |||
1900 | return rc; | ||
1901 | } | ||
1902 | |||
1903 | |||
1904 | static void __devexit hamachi_remove_one (struct pci_dev *pdev) | ||
1905 | { | ||
1906 | struct net_device *dev = pci_get_drvdata(pdev); | ||
1907 | |||
1908 | if (dev) { | ||
1909 | struct hamachi_private *hmp = netdev_priv(dev); | ||
1910 | |||
1911 | pci_free_consistent(pdev, RX_TOTAL_SIZE, hmp->rx_ring, | ||
1912 | hmp->rx_ring_dma); | ||
1913 | pci_free_consistent(pdev, TX_TOTAL_SIZE, hmp->tx_ring, | ||
1914 | hmp->tx_ring_dma); | ||
1915 | unregister_netdev(dev); | ||
1916 | iounmap(hmp->base); | ||
1917 | free_netdev(dev); | ||
1918 | pci_release_regions(pdev); | ||
1919 | pci_set_drvdata(pdev, NULL); | ||
1920 | } | ||
1921 | } | ||
1922 | |||
1923 | static DEFINE_PCI_DEVICE_TABLE(hamachi_pci_tbl) = { | ||
1924 | { 0x1318, 0x0911, PCI_ANY_ID, PCI_ANY_ID, }, | ||
1925 | { 0, } | ||
1926 | }; | ||
1927 | MODULE_DEVICE_TABLE(pci, hamachi_pci_tbl); | ||
1928 | |||
1929 | static struct pci_driver hamachi_driver = { | ||
1930 | .name = DRV_NAME, | ||
1931 | .id_table = hamachi_pci_tbl, | ||
1932 | .probe = hamachi_init_one, | ||
1933 | .remove = __devexit_p(hamachi_remove_one), | ||
1934 | }; | ||
1935 | |||
1936 | static int __init hamachi_init (void) | ||
1937 | { | ||
1938 | /* when a module, this is printed whether or not devices are found in probe */ | ||
1939 | #ifdef MODULE | ||
1940 | printk(version); | ||
1941 | #endif | ||
1942 | return pci_register_driver(&hamachi_driver); | ||
1943 | } | ||
1944 | |||
1945 | static void __exit hamachi_exit (void) | ||
1946 | { | ||
1947 | pci_unregister_driver(&hamachi_driver); | ||
1948 | } | ||
1949 | |||
1950 | |||
1951 | module_init(hamachi_init); | ||
1952 | module_exit(hamachi_exit); | ||
diff --git a/drivers/net/ethernet/packetengines/yellowfin.c b/drivers/net/ethernet/packetengines/yellowfin.c new file mode 100644 index 000000000000..3e5ac60b89ac --- /dev/null +++ b/drivers/net/ethernet/packetengines/yellowfin.c | |||
@@ -0,0 +1,1430 @@ | |||
1 | /* yellowfin.c: A Packet Engines G-NIC ethernet driver for linux. */ | ||
2 | /* | ||
3 | Written 1997-2001 by Donald Becker. | ||
4 | |||
5 | This software may be used and distributed according to the terms of | ||
6 | the GNU General Public License (GPL), incorporated herein by reference. | ||
7 | Drivers based on or derived from this code fall under the GPL and must | ||
8 | retain the authorship, copyright and license notice. This file is not | ||
9 | a complete program and may only be used when the entire operating | ||
10 | system is licensed under the GPL. | ||
11 | |||
12 | This driver is for the Packet Engines G-NIC PCI Gigabit Ethernet adapter. | ||
13 | It also supports the Symbios Logic version of the same chip core. | ||
14 | |||
15 | The author may be reached as becker@scyld.com, or C/O | ||
16 | Scyld Computing Corporation | ||
17 | 410 Severn Ave., Suite 210 | ||
18 | Annapolis MD 21403 | ||
19 | |||
20 | Support and updates available at | ||
21 | http://www.scyld.com/network/yellowfin.html | ||
22 | [link no longer provides useful info -jgarzik] | ||
23 | |||
24 | */ | ||
25 | |||
26 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | ||
27 | |||
28 | #define DRV_NAME "yellowfin" | ||
29 | #define DRV_VERSION "2.1" | ||
30 | #define DRV_RELDATE "Sep 11, 2006" | ||
31 | |||
32 | /* The user-configurable values. | ||
33 | These may be modified when a driver module is loaded.*/ | ||
34 | |||
35 | static int debug = 1; /* 1 normal messages, 0 quiet .. 7 verbose. */ | ||
36 | /* Maximum events (Rx packets, etc.) to handle at each interrupt. */ | ||
37 | static int max_interrupt_work = 20; | ||
38 | static int mtu; | ||
39 | #ifdef YF_PROTOTYPE /* Support for prototype hardware errata. */ | ||
40 | /* System-wide count of bogus-rx frames. */ | ||
41 | static int bogus_rx; | ||
42 | static int dma_ctrl = 0x004A0263; /* Constrained by errata */ | ||
43 | static int fifo_cfg = 0x0020; /* Bypass external Tx FIFO. */ | ||
44 | #elif defined(YF_NEW) /* A future perfect board :->. */ | ||
45 | static int dma_ctrl = 0x00CAC277; /* Override when loading module! */ | ||
46 | static int fifo_cfg = 0x0028; | ||
47 | #else | ||
48 | static const int dma_ctrl = 0x004A0263; /* Constrained by errata */ | ||
49 | static const int fifo_cfg = 0x0020; /* Bypass external Tx FIFO. */ | ||
50 | #endif | ||
51 | |||
52 | /* Set the copy breakpoint for the copy-only-tiny-frames scheme. | ||
53 | Setting to > 1514 effectively disables this feature. */ | ||
54 | static int rx_copybreak; | ||
55 | |||
56 | /* Used to pass the media type, etc. | ||
57 | No media types are currently defined. These exist for driver | ||
58 | interoperability. | ||
59 | */ | ||
60 | #define MAX_UNITS 8 /* More are supported, limit only on options */ | ||
61 | static int options[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1}; | ||
62 | static int full_duplex[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1}; | ||
63 | |||
64 | /* Do ugly workaround for GX server chipset errata. */ | ||
65 | static int gx_fix; | ||
66 | |||
67 | /* Operational parameters that are set at compile time. */ | ||
68 | |||
69 | /* Keep the ring sizes a power of two for efficiency. | ||
70 | Making the Tx ring too long decreases the effectiveness of channel | ||
71 | bonding and packet priority. | ||
72 | There are no ill effects from too-large receive rings. */ | ||
73 | #define TX_RING_SIZE 16 | ||
74 | #define TX_QUEUE_SIZE 12 /* Must be > 4 && <= TX_RING_SIZE */ | ||
75 | #define RX_RING_SIZE 64 | ||
76 | #define STATUS_TOTAL_SIZE TX_RING_SIZE*sizeof(struct tx_status_words) | ||
77 | #define TX_TOTAL_SIZE 2*TX_RING_SIZE*sizeof(struct yellowfin_desc) | ||
78 | #define RX_TOTAL_SIZE RX_RING_SIZE*sizeof(struct yellowfin_desc) | ||
79 | |||
80 | /* Operational parameters that usually are not changed. */ | ||
81 | /* Time in jiffies before concluding the transmitter is hung. */ | ||
82 | #define TX_TIMEOUT (2*HZ) | ||
83 | #define PKT_BUF_SZ 1536 /* Size of each temporary Rx buffer.*/ | ||
84 | |||
85 | #define yellowfin_debug debug | ||
86 | |||
87 | #include <linux/module.h> | ||
88 | #include <linux/kernel.h> | ||
89 | #include <linux/string.h> | ||
90 | #include <linux/timer.h> | ||
91 | #include <linux/errno.h> | ||
92 | #include <linux/ioport.h> | ||
93 | #include <linux/interrupt.h> | ||
94 | #include <linux/pci.h> | ||
95 | #include <linux/init.h> | ||
96 | #include <linux/mii.h> | ||
97 | #include <linux/netdevice.h> | ||
98 | #include <linux/etherdevice.h> | ||
99 | #include <linux/skbuff.h> | ||
100 | #include <linux/ethtool.h> | ||
101 | #include <linux/crc32.h> | ||
102 | #include <linux/bitops.h> | ||
103 | #include <asm/uaccess.h> | ||
104 | #include <asm/processor.h> /* Processor type for cache alignment. */ | ||
105 | #include <asm/unaligned.h> | ||
106 | #include <asm/io.h> | ||
107 | |||
108 | /* These identify the driver base version and may not be removed. */ | ||
109 | static const char version[] __devinitconst = | ||
110 | KERN_INFO DRV_NAME ".c:v1.05 1/09/2001 Written by Donald Becker <becker@scyld.com>\n" | ||
111 | " (unofficial 2.4.x port, " DRV_VERSION ", " DRV_RELDATE ")\n"; | ||
112 | |||
113 | MODULE_AUTHOR("Donald Becker <becker@scyld.com>"); | ||
114 | MODULE_DESCRIPTION("Packet Engines Yellowfin G-NIC Gigabit Ethernet driver"); | ||
115 | MODULE_LICENSE("GPL"); | ||
116 | |||
117 | module_param(max_interrupt_work, int, 0); | ||
118 | module_param(mtu, int, 0); | ||
119 | module_param(debug, int, 0); | ||
120 | module_param(rx_copybreak, int, 0); | ||
121 | module_param_array(options, int, NULL, 0); | ||
122 | module_param_array(full_duplex, int, NULL, 0); | ||
123 | module_param(gx_fix, int, 0); | ||
124 | MODULE_PARM_DESC(max_interrupt_work, "G-NIC maximum events handled per interrupt"); | ||
125 | MODULE_PARM_DESC(mtu, "G-NIC MTU (all boards)"); | ||
126 | MODULE_PARM_DESC(debug, "G-NIC debug level (0-7)"); | ||
127 | MODULE_PARM_DESC(rx_copybreak, "G-NIC copy breakpoint for copy-only-tiny-frames"); | ||
128 | MODULE_PARM_DESC(options, "G-NIC: Bits 0-3: media type, bit 17: full duplex"); | ||
129 | MODULE_PARM_DESC(full_duplex, "G-NIC full duplex setting(s) (1)"); | ||
130 | MODULE_PARM_DESC(gx_fix, "G-NIC: enable GX server chipset bug workaround (0-1)"); | ||
131 | |||
132 | /* | ||
133 | Theory of Operation | ||
134 | |||
135 | I. Board Compatibility | ||
136 | |||
137 | This device driver is designed for the Packet Engines "Yellowfin" Gigabit | ||
138 | Ethernet adapter. The G-NIC 64-bit PCI card is supported, as well as the | ||
139 | Symbios 53C885E dual function chip. | ||
140 | |||
141 | II. Board-specific settings | ||
142 | |||
143 | PCI bus devices are configured by the system at boot time, so no jumpers | ||
144 | need to be set on the board. The system BIOS preferably should assign the | ||
145 | PCI INTA signal to an otherwise unused system IRQ line. | ||
146 | Note: Kernel versions earlier than 1.3.73 do not support shared PCI | ||
147 | interrupt lines. | ||
148 | |||
149 | III. Driver operation | ||
150 | |||
151 | IIIa. Ring buffers | ||
152 | |||
153 | The Yellowfin uses the Descriptor Based DMA Architecture specified by Apple. | ||
154 | This is a descriptor list scheme similar to that used by the EEPro100 and | ||
155 | Tulip. This driver uses two statically allocated fixed-size descriptor lists | ||
156 | formed into rings by a branch from the final descriptor to the beginning of | ||
157 | the list. The ring sizes are set at compile time by RX/TX_RING_SIZE. | ||
158 | |||
159 | The driver allocates full frame size skbuffs for the Rx ring buffers at | ||
160 | open() time and passes the skb->data field to the Yellowfin as receive data | ||
161 | buffers. When an incoming frame is less than RX_COPYBREAK bytes long, | ||
162 | a fresh skbuff is allocated and the frame is copied to the new skbuff. | ||
163 | When the incoming frame is larger, the skbuff is passed directly up the | ||
164 | protocol stack and replaced by a newly allocated skbuff. | ||
165 | |||
166 | The RX_COPYBREAK value is chosen to trade-off the memory wasted by | ||
167 | using a full-sized skbuff for small frames vs. the copying costs of larger | ||
168 | frames. For small frames the copying cost is negligible (esp. considering | ||
169 | that we are pre-loading the cache with immediately useful header | ||
170 | information). For large frames the copying cost is non-trivial, and the | ||
171 | larger copy might flush the cache of useful data. | ||
172 | |||
173 | IIIC. Synchronization | ||
174 | |||
175 | The driver runs as two independent, single-threaded flows of control. One | ||
176 | is the send-packet routine, which enforces single-threaded use by the | ||
177 | dev->tbusy flag. The other thread is the interrupt handler, which is single | ||
178 | threaded by the hardware and other software. | ||
179 | |||
180 | The send packet thread has partial control over the Tx ring and 'dev->tbusy' | ||
181 | flag. It sets the tbusy flag whenever it's queuing a Tx packet. If the next | ||
182 | queue slot is empty, it clears the tbusy flag when finished otherwise it sets | ||
183 | the 'yp->tx_full' flag. | ||
184 | |||
185 | The interrupt handler has exclusive control over the Rx ring and records stats | ||
186 | from the Tx ring. After reaping the stats, it marks the Tx queue entry as | ||
187 | empty by incrementing the dirty_tx mark. Iff the 'yp->tx_full' flag is set, it | ||
188 | clears both the tx_full and tbusy flags. | ||
189 | |||
190 | IV. Notes | ||
191 | |||
192 | Thanks to Kim Stearns of Packet Engines for providing a pair of G-NIC boards. | ||
193 | Thanks to Bruce Faust of Digitalscape for providing both their SYM53C885 board | ||
194 | and an AlphaStation to verifty the Alpha port! | ||
195 | |||
196 | IVb. References | ||
197 | |||
198 | Yellowfin Engineering Design Specification, 4/23/97 Preliminary/Confidential | ||
199 | Symbios SYM53C885 PCI-SCSI/Fast Ethernet Multifunction Controller Preliminary | ||
200 | Data Manual v3.0 | ||
201 | http://cesdis.gsfc.nasa.gov/linux/misc/NWay.html | ||
202 | http://cesdis.gsfc.nasa.gov/linux/misc/100mbps.html | ||
203 | |||
204 | IVc. Errata | ||
205 | |||
206 | See Packet Engines confidential appendix (prototype chips only). | ||
207 | */ | ||
208 | |||
209 | |||
210 | |||
211 | enum capability_flags { | ||
212 | HasMII=1, FullTxStatus=2, IsGigabit=4, HasMulticastBug=8, FullRxStatus=16, | ||
213 | HasMACAddrBug=32, /* Only on early revs. */ | ||
214 | DontUseEeprom=64, /* Don't read the MAC from the EEPROm. */ | ||
215 | }; | ||
216 | |||
217 | /* The PCI I/O space extent. */ | ||
218 | enum { | ||
219 | YELLOWFIN_SIZE = 0x100, | ||
220 | }; | ||
221 | |||
222 | struct pci_id_info { | ||
223 | const char *name; | ||
224 | struct match_info { | ||
225 | int pci, pci_mask, subsystem, subsystem_mask; | ||
226 | int revision, revision_mask; /* Only 8 bits. */ | ||
227 | } id; | ||
228 | int drv_flags; /* Driver use, intended as capability flags. */ | ||
229 | }; | ||
230 | |||
231 | static const struct pci_id_info pci_id_tbl[] = { | ||
232 | {"Yellowfin G-NIC Gigabit Ethernet", { 0x07021000, 0xffffffff}, | ||
233 | FullTxStatus | IsGigabit | HasMulticastBug | HasMACAddrBug | DontUseEeprom}, | ||
234 | {"Symbios SYM83C885", { 0x07011000, 0xffffffff}, | ||
235 | HasMII | DontUseEeprom }, | ||
236 | { } | ||
237 | }; | ||
238 | |||
239 | static DEFINE_PCI_DEVICE_TABLE(yellowfin_pci_tbl) = { | ||
240 | { 0x1000, 0x0702, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 }, | ||
241 | { 0x1000, 0x0701, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 1 }, | ||
242 | { } | ||
243 | }; | ||
244 | MODULE_DEVICE_TABLE (pci, yellowfin_pci_tbl); | ||
245 | |||
246 | |||
247 | /* Offsets to the Yellowfin registers. Various sizes and alignments. */ | ||
248 | enum yellowfin_offsets { | ||
249 | TxCtrl=0x00, TxStatus=0x04, TxPtr=0x0C, | ||
250 | TxIntrSel=0x10, TxBranchSel=0x14, TxWaitSel=0x18, | ||
251 | RxCtrl=0x40, RxStatus=0x44, RxPtr=0x4C, | ||
252 | RxIntrSel=0x50, RxBranchSel=0x54, RxWaitSel=0x58, | ||
253 | EventStatus=0x80, IntrEnb=0x82, IntrClear=0x84, IntrStatus=0x86, | ||
254 | ChipRev=0x8C, DMACtrl=0x90, TxThreshold=0x94, | ||
255 | Cnfg=0xA0, FrameGap0=0xA2, FrameGap1=0xA4, | ||
256 | MII_Cmd=0xA6, MII_Addr=0xA8, MII_Wr_Data=0xAA, MII_Rd_Data=0xAC, | ||
257 | MII_Status=0xAE, | ||
258 | RxDepth=0xB8, FlowCtrl=0xBC, | ||
259 | AddrMode=0xD0, StnAddr=0xD2, HashTbl=0xD8, FIFOcfg=0xF8, | ||
260 | EEStatus=0xF0, EECtrl=0xF1, EEAddr=0xF2, EERead=0xF3, EEWrite=0xF4, | ||
261 | EEFeature=0xF5, | ||
262 | }; | ||
263 | |||
264 | /* The Yellowfin Rx and Tx buffer descriptors. | ||
265 | Elements are written as 32 bit for endian portability. */ | ||
266 | struct yellowfin_desc { | ||
267 | __le32 dbdma_cmd; | ||
268 | __le32 addr; | ||
269 | __le32 branch_addr; | ||
270 | __le32 result_status; | ||
271 | }; | ||
272 | |||
273 | struct tx_status_words { | ||
274 | #ifdef __BIG_ENDIAN | ||
275 | u16 tx_errs; | ||
276 | u16 tx_cnt; | ||
277 | u16 paused; | ||
278 | u16 total_tx_cnt; | ||
279 | #else /* Little endian chips. */ | ||
280 | u16 tx_cnt; | ||
281 | u16 tx_errs; | ||
282 | u16 total_tx_cnt; | ||
283 | u16 paused; | ||
284 | #endif /* __BIG_ENDIAN */ | ||
285 | }; | ||
286 | |||
287 | /* Bits in yellowfin_desc.cmd */ | ||
288 | enum desc_cmd_bits { | ||
289 | CMD_TX_PKT=0x10000000, CMD_RX_BUF=0x20000000, CMD_TXSTATUS=0x30000000, | ||
290 | CMD_NOP=0x60000000, CMD_STOP=0x70000000, | ||
291 | BRANCH_ALWAYS=0x0C0000, INTR_ALWAYS=0x300000, WAIT_ALWAYS=0x030000, | ||
292 | BRANCH_IFTRUE=0x040000, | ||
293 | }; | ||
294 | |||
295 | /* Bits in yellowfin_desc.status */ | ||
296 | enum desc_status_bits { RX_EOP=0x0040, }; | ||
297 | |||
298 | /* Bits in the interrupt status/mask registers. */ | ||
299 | enum intr_status_bits { | ||
300 | IntrRxDone=0x01, IntrRxInvalid=0x02, IntrRxPCIFault=0x04,IntrRxPCIErr=0x08, | ||
301 | IntrTxDone=0x10, IntrTxInvalid=0x20, IntrTxPCIFault=0x40,IntrTxPCIErr=0x80, | ||
302 | IntrEarlyRx=0x100, IntrWakeup=0x200, }; | ||
303 | |||
304 | #define PRIV_ALIGN 31 /* Required alignment mask */ | ||
305 | #define MII_CNT 4 | ||
306 | struct yellowfin_private { | ||
307 | /* Descriptor rings first for alignment. | ||
308 | Tx requires a second descriptor for status. */ | ||
309 | struct yellowfin_desc *rx_ring; | ||
310 | struct yellowfin_desc *tx_ring; | ||
311 | struct sk_buff* rx_skbuff[RX_RING_SIZE]; | ||
312 | struct sk_buff* tx_skbuff[TX_RING_SIZE]; | ||
313 | dma_addr_t rx_ring_dma; | ||
314 | dma_addr_t tx_ring_dma; | ||
315 | |||
316 | struct tx_status_words *tx_status; | ||
317 | dma_addr_t tx_status_dma; | ||
318 | |||
319 | struct timer_list timer; /* Media selection timer. */ | ||
320 | /* Frequently used and paired value: keep adjacent for cache effect. */ | ||
321 | int chip_id, drv_flags; | ||
322 | struct pci_dev *pci_dev; | ||
323 | unsigned int cur_rx, dirty_rx; /* Producer/consumer ring indices */ | ||
324 | unsigned int rx_buf_sz; /* Based on MTU+slack. */ | ||
325 | struct tx_status_words *tx_tail_desc; | ||
326 | unsigned int cur_tx, dirty_tx; | ||
327 | int tx_threshold; | ||
328 | unsigned int tx_full:1; /* The Tx queue is full. */ | ||
329 | unsigned int full_duplex:1; /* Full-duplex operation requested. */ | ||
330 | unsigned int duplex_lock:1; | ||
331 | unsigned int medialock:1; /* Do not sense media. */ | ||
332 | unsigned int default_port:4; /* Last dev->if_port value. */ | ||
333 | /* MII transceiver section. */ | ||
334 | int mii_cnt; /* MII device addresses. */ | ||
335 | u16 advertising; /* NWay media advertisement */ | ||
336 | unsigned char phys[MII_CNT]; /* MII device addresses, only first one used */ | ||
337 | spinlock_t lock; | ||
338 | void __iomem *base; | ||
339 | }; | ||
340 | |||
341 | static int read_eeprom(void __iomem *ioaddr, int location); | ||
342 | static int mdio_read(void __iomem *ioaddr, int phy_id, int location); | ||
343 | static void mdio_write(void __iomem *ioaddr, int phy_id, int location, int value); | ||
344 | static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd); | ||
345 | static int yellowfin_open(struct net_device *dev); | ||
346 | static void yellowfin_timer(unsigned long data); | ||
347 | static void yellowfin_tx_timeout(struct net_device *dev); | ||
348 | static int yellowfin_init_ring(struct net_device *dev); | ||
349 | static netdev_tx_t yellowfin_start_xmit(struct sk_buff *skb, | ||
350 | struct net_device *dev); | ||
351 | static irqreturn_t yellowfin_interrupt(int irq, void *dev_instance); | ||
352 | static int yellowfin_rx(struct net_device *dev); | ||
353 | static void yellowfin_error(struct net_device *dev, int intr_status); | ||
354 | static int yellowfin_close(struct net_device *dev); | ||
355 | static void set_rx_mode(struct net_device *dev); | ||
356 | static const struct ethtool_ops ethtool_ops; | ||
357 | |||
358 | static const struct net_device_ops netdev_ops = { | ||
359 | .ndo_open = yellowfin_open, | ||
360 | .ndo_stop = yellowfin_close, | ||
361 | .ndo_start_xmit = yellowfin_start_xmit, | ||
362 | .ndo_set_multicast_list = set_rx_mode, | ||
363 | .ndo_change_mtu = eth_change_mtu, | ||
364 | .ndo_validate_addr = eth_validate_addr, | ||
365 | .ndo_set_mac_address = eth_mac_addr, | ||
366 | .ndo_do_ioctl = netdev_ioctl, | ||
367 | .ndo_tx_timeout = yellowfin_tx_timeout, | ||
368 | }; | ||
369 | |||
370 | static int __devinit yellowfin_init_one(struct pci_dev *pdev, | ||
371 | const struct pci_device_id *ent) | ||
372 | { | ||
373 | struct net_device *dev; | ||
374 | struct yellowfin_private *np; | ||
375 | int irq; | ||
376 | int chip_idx = ent->driver_data; | ||
377 | static int find_cnt; | ||
378 | void __iomem *ioaddr; | ||
379 | int i, option = find_cnt < MAX_UNITS ? options[find_cnt] : 0; | ||
380 | int drv_flags = pci_id_tbl[chip_idx].drv_flags; | ||
381 | void *ring_space; | ||
382 | dma_addr_t ring_dma; | ||
383 | #ifdef USE_IO_OPS | ||
384 | int bar = 0; | ||
385 | #else | ||
386 | int bar = 1; | ||
387 | #endif | ||
388 | |||
389 | /* when built into the kernel, we only print version if device is found */ | ||
390 | #ifndef MODULE | ||
391 | static int printed_version; | ||
392 | if (!printed_version++) | ||
393 | printk(version); | ||
394 | #endif | ||
395 | |||
396 | i = pci_enable_device(pdev); | ||
397 | if (i) return i; | ||
398 | |||
399 | dev = alloc_etherdev(sizeof(*np)); | ||
400 | if (!dev) { | ||
401 | pr_err("cannot allocate ethernet device\n"); | ||
402 | return -ENOMEM; | ||
403 | } | ||
404 | SET_NETDEV_DEV(dev, &pdev->dev); | ||
405 | |||
406 | np = netdev_priv(dev); | ||
407 | |||
408 | if (pci_request_regions(pdev, DRV_NAME)) | ||
409 | goto err_out_free_netdev; | ||
410 | |||
411 | pci_set_master (pdev); | ||
412 | |||
413 | ioaddr = pci_iomap(pdev, bar, YELLOWFIN_SIZE); | ||
414 | if (!ioaddr) | ||
415 | goto err_out_free_res; | ||
416 | |||
417 | irq = pdev->irq; | ||
418 | |||
419 | if (drv_flags & DontUseEeprom) | ||
420 | for (i = 0; i < 6; i++) | ||
421 | dev->dev_addr[i] = ioread8(ioaddr + StnAddr + i); | ||
422 | else { | ||
423 | int ee_offset = (read_eeprom(ioaddr, 6) == 0xff ? 0x100 : 0); | ||
424 | for (i = 0; i < 6; i++) | ||
425 | dev->dev_addr[i] = read_eeprom(ioaddr, ee_offset + i); | ||
426 | } | ||
427 | |||
428 | /* Reset the chip. */ | ||
429 | iowrite32(0x80000000, ioaddr + DMACtrl); | ||
430 | |||
431 | dev->base_addr = (unsigned long)ioaddr; | ||
432 | dev->irq = irq; | ||
433 | |||
434 | pci_set_drvdata(pdev, dev); | ||
435 | spin_lock_init(&np->lock); | ||
436 | |||
437 | np->pci_dev = pdev; | ||
438 | np->chip_id = chip_idx; | ||
439 | np->drv_flags = drv_flags; | ||
440 | np->base = ioaddr; | ||
441 | |||
442 | ring_space = pci_alloc_consistent(pdev, TX_TOTAL_SIZE, &ring_dma); | ||
443 | if (!ring_space) | ||
444 | goto err_out_cleardev; | ||
445 | np->tx_ring = ring_space; | ||
446 | np->tx_ring_dma = ring_dma; | ||
447 | |||
448 | ring_space = pci_alloc_consistent(pdev, RX_TOTAL_SIZE, &ring_dma); | ||
449 | if (!ring_space) | ||
450 | goto err_out_unmap_tx; | ||
451 | np->rx_ring = ring_space; | ||
452 | np->rx_ring_dma = ring_dma; | ||
453 | |||
454 | ring_space = pci_alloc_consistent(pdev, STATUS_TOTAL_SIZE, &ring_dma); | ||
455 | if (!ring_space) | ||
456 | goto err_out_unmap_rx; | ||
457 | np->tx_status = ring_space; | ||
458 | np->tx_status_dma = ring_dma; | ||
459 | |||
460 | if (dev->mem_start) | ||
461 | option = dev->mem_start; | ||
462 | |||
463 | /* The lower four bits are the media type. */ | ||
464 | if (option > 0) { | ||
465 | if (option & 0x200) | ||
466 | np->full_duplex = 1; | ||
467 | np->default_port = option & 15; | ||
468 | if (np->default_port) | ||
469 | np->medialock = 1; | ||
470 | } | ||
471 | if (find_cnt < MAX_UNITS && full_duplex[find_cnt] > 0) | ||
472 | np->full_duplex = 1; | ||
473 | |||
474 | if (np->full_duplex) | ||
475 | np->duplex_lock = 1; | ||
476 | |||
477 | /* The Yellowfin-specific entries in the device structure. */ | ||
478 | dev->netdev_ops = &netdev_ops; | ||
479 | SET_ETHTOOL_OPS(dev, ðtool_ops); | ||
480 | dev->watchdog_timeo = TX_TIMEOUT; | ||
481 | |||
482 | if (mtu) | ||
483 | dev->mtu = mtu; | ||
484 | |||
485 | i = register_netdev(dev); | ||
486 | if (i) | ||
487 | goto err_out_unmap_status; | ||
488 | |||
489 | netdev_info(dev, "%s type %8x at %p, %pM, IRQ %d\n", | ||
490 | pci_id_tbl[chip_idx].name, | ||
491 | ioread32(ioaddr + ChipRev), ioaddr, | ||
492 | dev->dev_addr, irq); | ||
493 | |||
494 | if (np->drv_flags & HasMII) { | ||
495 | int phy, phy_idx = 0; | ||
496 | for (phy = 0; phy < 32 && phy_idx < MII_CNT; phy++) { | ||
497 | int mii_status = mdio_read(ioaddr, phy, 1); | ||
498 | if (mii_status != 0xffff && mii_status != 0x0000) { | ||
499 | np->phys[phy_idx++] = phy; | ||
500 | np->advertising = mdio_read(ioaddr, phy, 4); | ||
501 | netdev_info(dev, "MII PHY found at address %d, status 0x%04x advertising %04x\n", | ||
502 | phy, mii_status, np->advertising); | ||
503 | } | ||
504 | } | ||
505 | np->mii_cnt = phy_idx; | ||
506 | } | ||
507 | |||
508 | find_cnt++; | ||
509 | |||
510 | return 0; | ||
511 | |||
512 | err_out_unmap_status: | ||
513 | pci_free_consistent(pdev, STATUS_TOTAL_SIZE, np->tx_status, | ||
514 | np->tx_status_dma); | ||
515 | err_out_unmap_rx: | ||
516 | pci_free_consistent(pdev, RX_TOTAL_SIZE, np->rx_ring, np->rx_ring_dma); | ||
517 | err_out_unmap_tx: | ||
518 | pci_free_consistent(pdev, TX_TOTAL_SIZE, np->tx_ring, np->tx_ring_dma); | ||
519 | err_out_cleardev: | ||
520 | pci_set_drvdata(pdev, NULL); | ||
521 | pci_iounmap(pdev, ioaddr); | ||
522 | err_out_free_res: | ||
523 | pci_release_regions(pdev); | ||
524 | err_out_free_netdev: | ||
525 | free_netdev (dev); | ||
526 | return -ENODEV; | ||
527 | } | ||
528 | |||
529 | static int __devinit read_eeprom(void __iomem *ioaddr, int location) | ||
530 | { | ||
531 | int bogus_cnt = 10000; /* Typical 33Mhz: 1050 ticks */ | ||
532 | |||
533 | iowrite8(location, ioaddr + EEAddr); | ||
534 | iowrite8(0x30 | ((location >> 8) & 7), ioaddr + EECtrl); | ||
535 | while ((ioread8(ioaddr + EEStatus) & 0x80) && --bogus_cnt > 0) | ||
536 | ; | ||
537 | return ioread8(ioaddr + EERead); | ||
538 | } | ||
539 | |||
540 | /* MII Managemen Data I/O accesses. | ||
541 | These routines assume the MDIO controller is idle, and do not exit until | ||
542 | the command is finished. */ | ||
543 | |||
544 | static int mdio_read(void __iomem *ioaddr, int phy_id, int location) | ||
545 | { | ||
546 | int i; | ||
547 | |||
548 | iowrite16((phy_id<<8) + location, ioaddr + MII_Addr); | ||
549 | iowrite16(1, ioaddr + MII_Cmd); | ||
550 | for (i = 10000; i >= 0; i--) | ||
551 | if ((ioread16(ioaddr + MII_Status) & 1) == 0) | ||
552 | break; | ||
553 | return ioread16(ioaddr + MII_Rd_Data); | ||
554 | } | ||
555 | |||
556 | static void mdio_write(void __iomem *ioaddr, int phy_id, int location, int value) | ||
557 | { | ||
558 | int i; | ||
559 | |||
560 | iowrite16((phy_id<<8) + location, ioaddr + MII_Addr); | ||
561 | iowrite16(value, ioaddr + MII_Wr_Data); | ||
562 | |||
563 | /* Wait for the command to finish. */ | ||
564 | for (i = 10000; i >= 0; i--) | ||
565 | if ((ioread16(ioaddr + MII_Status) & 1) == 0) | ||
566 | break; | ||
567 | } | ||
568 | |||
569 | |||
570 | static int yellowfin_open(struct net_device *dev) | ||
571 | { | ||
572 | struct yellowfin_private *yp = netdev_priv(dev); | ||
573 | void __iomem *ioaddr = yp->base; | ||
574 | int i, ret; | ||
575 | |||
576 | /* Reset the chip. */ | ||
577 | iowrite32(0x80000000, ioaddr + DMACtrl); | ||
578 | |||
579 | ret = request_irq(dev->irq, yellowfin_interrupt, IRQF_SHARED, dev->name, dev); | ||
580 | if (ret) | ||
581 | return ret; | ||
582 | |||
583 | if (yellowfin_debug > 1) | ||
584 | netdev_printk(KERN_DEBUG, dev, "%s() irq %d\n", | ||
585 | __func__, dev->irq); | ||
586 | |||
587 | ret = yellowfin_init_ring(dev); | ||
588 | if (ret) { | ||
589 | free_irq(dev->irq, dev); | ||
590 | return ret; | ||
591 | } | ||
592 | |||
593 | iowrite32(yp->rx_ring_dma, ioaddr + RxPtr); | ||
594 | iowrite32(yp->tx_ring_dma, ioaddr + TxPtr); | ||
595 | |||
596 | for (i = 0; i < 6; i++) | ||
597 | iowrite8(dev->dev_addr[i], ioaddr + StnAddr + i); | ||
598 | |||
599 | /* Set up various condition 'select' registers. | ||
600 | There are no options here. */ | ||
601 | iowrite32(0x00800080, ioaddr + TxIntrSel); /* Interrupt on Tx abort */ | ||
602 | iowrite32(0x00800080, ioaddr + TxBranchSel); /* Branch on Tx abort */ | ||
603 | iowrite32(0x00400040, ioaddr + TxWaitSel); /* Wait on Tx status */ | ||
604 | iowrite32(0x00400040, ioaddr + RxIntrSel); /* Interrupt on Rx done */ | ||
605 | iowrite32(0x00400040, ioaddr + RxBranchSel); /* Branch on Rx error */ | ||
606 | iowrite32(0x00400040, ioaddr + RxWaitSel); /* Wait on Rx done */ | ||
607 | |||
608 | /* Initialize other registers: with so many this eventually this will | ||
609 | converted to an offset/value list. */ | ||
610 | iowrite32(dma_ctrl, ioaddr + DMACtrl); | ||
611 | iowrite16(fifo_cfg, ioaddr + FIFOcfg); | ||
612 | /* Enable automatic generation of flow control frames, period 0xffff. */ | ||
613 | iowrite32(0x0030FFFF, ioaddr + FlowCtrl); | ||
614 | |||
615 | yp->tx_threshold = 32; | ||
616 | iowrite32(yp->tx_threshold, ioaddr + TxThreshold); | ||
617 | |||
618 | if (dev->if_port == 0) | ||
619 | dev->if_port = yp->default_port; | ||
620 | |||
621 | netif_start_queue(dev); | ||
622 | |||
623 | /* Setting the Rx mode will start the Rx process. */ | ||
624 | if (yp->drv_flags & IsGigabit) { | ||
625 | /* We are always in full-duplex mode with gigabit! */ | ||
626 | yp->full_duplex = 1; | ||
627 | iowrite16(0x01CF, ioaddr + Cnfg); | ||
628 | } else { | ||
629 | iowrite16(0x0018, ioaddr + FrameGap0); /* 0060/4060 for non-MII 10baseT */ | ||
630 | iowrite16(0x1018, ioaddr + FrameGap1); | ||
631 | iowrite16(0x101C | (yp->full_duplex ? 2 : 0), ioaddr + Cnfg); | ||
632 | } | ||
633 | set_rx_mode(dev); | ||
634 | |||
635 | /* Enable interrupts by setting the interrupt mask. */ | ||
636 | iowrite16(0x81ff, ioaddr + IntrEnb); /* See enum intr_status_bits */ | ||
637 | iowrite16(0x0000, ioaddr + EventStatus); /* Clear non-interrupting events */ | ||
638 | iowrite32(0x80008000, ioaddr + RxCtrl); /* Start Rx and Tx channels. */ | ||
639 | iowrite32(0x80008000, ioaddr + TxCtrl); | ||
640 | |||
641 | if (yellowfin_debug > 2) { | ||
642 | netdev_printk(KERN_DEBUG, dev, "Done %s()\n", __func__); | ||
643 | } | ||
644 | |||
645 | /* Set the timer to check for link beat. */ | ||
646 | init_timer(&yp->timer); | ||
647 | yp->timer.expires = jiffies + 3*HZ; | ||
648 | yp->timer.data = (unsigned long)dev; | ||
649 | yp->timer.function = yellowfin_timer; /* timer handler */ | ||
650 | add_timer(&yp->timer); | ||
651 | |||
652 | return 0; | ||
653 | } | ||
654 | |||
655 | static void yellowfin_timer(unsigned long data) | ||
656 | { | ||
657 | struct net_device *dev = (struct net_device *)data; | ||
658 | struct yellowfin_private *yp = netdev_priv(dev); | ||
659 | void __iomem *ioaddr = yp->base; | ||
660 | int next_tick = 60*HZ; | ||
661 | |||
662 | if (yellowfin_debug > 3) { | ||
663 | netdev_printk(KERN_DEBUG, dev, "Yellowfin timer tick, status %08x\n", | ||
664 | ioread16(ioaddr + IntrStatus)); | ||
665 | } | ||
666 | |||
667 | if (yp->mii_cnt) { | ||
668 | int bmsr = mdio_read(ioaddr, yp->phys[0], MII_BMSR); | ||
669 | int lpa = mdio_read(ioaddr, yp->phys[0], MII_LPA); | ||
670 | int negotiated = lpa & yp->advertising; | ||
671 | if (yellowfin_debug > 1) | ||
672 | netdev_printk(KERN_DEBUG, dev, "MII #%d status register is %04x, link partner capability %04x\n", | ||
673 | yp->phys[0], bmsr, lpa); | ||
674 | |||
675 | yp->full_duplex = mii_duplex(yp->duplex_lock, negotiated); | ||
676 | |||
677 | iowrite16(0x101C | (yp->full_duplex ? 2 : 0), ioaddr + Cnfg); | ||
678 | |||
679 | if (bmsr & BMSR_LSTATUS) | ||
680 | next_tick = 60*HZ; | ||
681 | else | ||
682 | next_tick = 3*HZ; | ||
683 | } | ||
684 | |||
685 | yp->timer.expires = jiffies + next_tick; | ||
686 | add_timer(&yp->timer); | ||
687 | } | ||
688 | |||
689 | static void yellowfin_tx_timeout(struct net_device *dev) | ||
690 | { | ||
691 | struct yellowfin_private *yp = netdev_priv(dev); | ||
692 | void __iomem *ioaddr = yp->base; | ||
693 | |||
694 | netdev_warn(dev, "Yellowfin transmit timed out at %d/%d Tx status %04x, Rx status %04x, resetting...\n", | ||
695 | yp->cur_tx, yp->dirty_tx, | ||
696 | ioread32(ioaddr + TxStatus), | ||
697 | ioread32(ioaddr + RxStatus)); | ||
698 | |||
699 | /* Note: these should be KERN_DEBUG. */ | ||
700 | if (yellowfin_debug) { | ||
701 | int i; | ||
702 | pr_warning(" Rx ring %p: ", yp->rx_ring); | ||
703 | for (i = 0; i < RX_RING_SIZE; i++) | ||
704 | pr_cont(" %08x", yp->rx_ring[i].result_status); | ||
705 | pr_cont("\n"); | ||
706 | pr_warning(" Tx ring %p: ", yp->tx_ring); | ||
707 | for (i = 0; i < TX_RING_SIZE; i++) | ||
708 | pr_cont(" %04x /%08x", | ||
709 | yp->tx_status[i].tx_errs, | ||
710 | yp->tx_ring[i].result_status); | ||
711 | pr_cont("\n"); | ||
712 | } | ||
713 | |||
714 | /* If the hardware is found to hang regularly, we will update the code | ||
715 | to reinitialize the chip here. */ | ||
716 | dev->if_port = 0; | ||
717 | |||
718 | /* Wake the potentially-idle transmit channel. */ | ||
719 | iowrite32(0x10001000, yp->base + TxCtrl); | ||
720 | if (yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE) | ||
721 | netif_wake_queue (dev); /* Typical path */ | ||
722 | |||
723 | dev->trans_start = jiffies; /* prevent tx timeout */ | ||
724 | dev->stats.tx_errors++; | ||
725 | } | ||
726 | |||
727 | /* Initialize the Rx and Tx rings, along with various 'dev' bits. */ | ||
728 | static int yellowfin_init_ring(struct net_device *dev) | ||
729 | { | ||
730 | struct yellowfin_private *yp = netdev_priv(dev); | ||
731 | int i, j; | ||
732 | |||
733 | yp->tx_full = 0; | ||
734 | yp->cur_rx = yp->cur_tx = 0; | ||
735 | yp->dirty_tx = 0; | ||
736 | |||
737 | yp->rx_buf_sz = (dev->mtu <= 1500 ? PKT_BUF_SZ : dev->mtu + 32); | ||
738 | |||
739 | for (i = 0; i < RX_RING_SIZE; i++) { | ||
740 | yp->rx_ring[i].dbdma_cmd = | ||
741 | cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | yp->rx_buf_sz); | ||
742 | yp->rx_ring[i].branch_addr = cpu_to_le32(yp->rx_ring_dma + | ||
743 | ((i+1)%RX_RING_SIZE)*sizeof(struct yellowfin_desc)); | ||
744 | } | ||
745 | |||
746 | for (i = 0; i < RX_RING_SIZE; i++) { | ||
747 | struct sk_buff *skb = dev_alloc_skb(yp->rx_buf_sz + 2); | ||
748 | yp->rx_skbuff[i] = skb; | ||
749 | if (skb == NULL) | ||
750 | break; | ||
751 | skb->dev = dev; /* Mark as being used by this device. */ | ||
752 | skb_reserve(skb, 2); /* 16 byte align the IP header. */ | ||
753 | yp->rx_ring[i].addr = cpu_to_le32(pci_map_single(yp->pci_dev, | ||
754 | skb->data, yp->rx_buf_sz, PCI_DMA_FROMDEVICE)); | ||
755 | } | ||
756 | if (i != RX_RING_SIZE) { | ||
757 | for (j = 0; j < i; j++) | ||
758 | dev_kfree_skb(yp->rx_skbuff[j]); | ||
759 | return -ENOMEM; | ||
760 | } | ||
761 | yp->rx_ring[i-1].dbdma_cmd = cpu_to_le32(CMD_STOP); | ||
762 | yp->dirty_rx = (unsigned int)(i - RX_RING_SIZE); | ||
763 | |||
764 | #define NO_TXSTATS | ||
765 | #ifdef NO_TXSTATS | ||
766 | /* In this mode the Tx ring needs only a single descriptor. */ | ||
767 | for (i = 0; i < TX_RING_SIZE; i++) { | ||
768 | yp->tx_skbuff[i] = NULL; | ||
769 | yp->tx_ring[i].dbdma_cmd = cpu_to_le32(CMD_STOP); | ||
770 | yp->tx_ring[i].branch_addr = cpu_to_le32(yp->tx_ring_dma + | ||
771 | ((i+1)%TX_RING_SIZE)*sizeof(struct yellowfin_desc)); | ||
772 | } | ||
773 | /* Wrap ring */ | ||
774 | yp->tx_ring[--i].dbdma_cmd = cpu_to_le32(CMD_STOP | BRANCH_ALWAYS); | ||
775 | #else | ||
776 | { | ||
777 | /* Tx ring needs a pair of descriptors, the second for the status. */ | ||
778 | for (i = 0; i < TX_RING_SIZE; i++) { | ||
779 | j = 2*i; | ||
780 | yp->tx_skbuff[i] = 0; | ||
781 | /* Branch on Tx error. */ | ||
782 | yp->tx_ring[j].dbdma_cmd = cpu_to_le32(CMD_STOP); | ||
783 | yp->tx_ring[j].branch_addr = cpu_to_le32(yp->tx_ring_dma + | ||
784 | (j+1)*sizeof(struct yellowfin_desc)); | ||
785 | j++; | ||
786 | if (yp->flags & FullTxStatus) { | ||
787 | yp->tx_ring[j].dbdma_cmd = | ||
788 | cpu_to_le32(CMD_TXSTATUS | sizeof(*yp->tx_status)); | ||
789 | yp->tx_ring[j].request_cnt = sizeof(*yp->tx_status); | ||
790 | yp->tx_ring[j].addr = cpu_to_le32(yp->tx_status_dma + | ||
791 | i*sizeof(struct tx_status_words)); | ||
792 | } else { | ||
793 | /* Symbios chips write only tx_errs word. */ | ||
794 | yp->tx_ring[j].dbdma_cmd = | ||
795 | cpu_to_le32(CMD_TXSTATUS | INTR_ALWAYS | 2); | ||
796 | yp->tx_ring[j].request_cnt = 2; | ||
797 | /* Om pade ummmmm... */ | ||
798 | yp->tx_ring[j].addr = cpu_to_le32(yp->tx_status_dma + | ||
799 | i*sizeof(struct tx_status_words) + | ||
800 | &(yp->tx_status[0].tx_errs) - | ||
801 | &(yp->tx_status[0])); | ||
802 | } | ||
803 | yp->tx_ring[j].branch_addr = cpu_to_le32(yp->tx_ring_dma + | ||
804 | ((j+1)%(2*TX_RING_SIZE))*sizeof(struct yellowfin_desc)); | ||
805 | } | ||
806 | /* Wrap ring */ | ||
807 | yp->tx_ring[++j].dbdma_cmd |= cpu_to_le32(BRANCH_ALWAYS | INTR_ALWAYS); | ||
808 | } | ||
809 | #endif | ||
810 | yp->tx_tail_desc = &yp->tx_status[0]; | ||
811 | return 0; | ||
812 | } | ||
813 | |||
814 | static netdev_tx_t yellowfin_start_xmit(struct sk_buff *skb, | ||
815 | struct net_device *dev) | ||
816 | { | ||
817 | struct yellowfin_private *yp = netdev_priv(dev); | ||
818 | unsigned entry; | ||
819 | int len = skb->len; | ||
820 | |||
821 | netif_stop_queue (dev); | ||
822 | |||
823 | /* Note: Ordering is important here, set the field with the | ||
824 | "ownership" bit last, and only then increment cur_tx. */ | ||
825 | |||
826 | /* Calculate the next Tx descriptor entry. */ | ||
827 | entry = yp->cur_tx % TX_RING_SIZE; | ||
828 | |||
829 | if (gx_fix) { /* Note: only works for paddable protocols e.g. IP. */ | ||
830 | int cacheline_end = ((unsigned long)skb->data + skb->len) % 32; | ||
831 | /* Fix GX chipset errata. */ | ||
832 | if (cacheline_end > 24 || cacheline_end == 0) { | ||
833 | len = skb->len + 32 - cacheline_end + 1; | ||
834 | if (skb_padto(skb, len)) { | ||
835 | yp->tx_skbuff[entry] = NULL; | ||
836 | netif_wake_queue(dev); | ||
837 | return NETDEV_TX_OK; | ||
838 | } | ||
839 | } | ||
840 | } | ||
841 | yp->tx_skbuff[entry] = skb; | ||
842 | |||
843 | #ifdef NO_TXSTATS | ||
844 | yp->tx_ring[entry].addr = cpu_to_le32(pci_map_single(yp->pci_dev, | ||
845 | skb->data, len, PCI_DMA_TODEVICE)); | ||
846 | yp->tx_ring[entry].result_status = 0; | ||
847 | if (entry >= TX_RING_SIZE-1) { | ||
848 | /* New stop command. */ | ||
849 | yp->tx_ring[0].dbdma_cmd = cpu_to_le32(CMD_STOP); | ||
850 | yp->tx_ring[TX_RING_SIZE-1].dbdma_cmd = | ||
851 | cpu_to_le32(CMD_TX_PKT|BRANCH_ALWAYS | len); | ||
852 | } else { | ||
853 | yp->tx_ring[entry+1].dbdma_cmd = cpu_to_le32(CMD_STOP); | ||
854 | yp->tx_ring[entry].dbdma_cmd = | ||
855 | cpu_to_le32(CMD_TX_PKT | BRANCH_IFTRUE | len); | ||
856 | } | ||
857 | yp->cur_tx++; | ||
858 | #else | ||
859 | yp->tx_ring[entry<<1].request_cnt = len; | ||
860 | yp->tx_ring[entry<<1].addr = cpu_to_le32(pci_map_single(yp->pci_dev, | ||
861 | skb->data, len, PCI_DMA_TODEVICE)); | ||
862 | /* The input_last (status-write) command is constant, but we must | ||
863 | rewrite the subsequent 'stop' command. */ | ||
864 | |||
865 | yp->cur_tx++; | ||
866 | { | ||
867 | unsigned next_entry = yp->cur_tx % TX_RING_SIZE; | ||
868 | yp->tx_ring[next_entry<<1].dbdma_cmd = cpu_to_le32(CMD_STOP); | ||
869 | } | ||
870 | /* Final step -- overwrite the old 'stop' command. */ | ||
871 | |||
872 | yp->tx_ring[entry<<1].dbdma_cmd = | ||
873 | cpu_to_le32( ((entry % 6) == 0 ? CMD_TX_PKT|INTR_ALWAYS|BRANCH_IFTRUE : | ||
874 | CMD_TX_PKT | BRANCH_IFTRUE) | len); | ||
875 | #endif | ||
876 | |||
877 | /* Non-x86 Todo: explicitly flush cache lines here. */ | ||
878 | |||
879 | /* Wake the potentially-idle transmit channel. */ | ||
880 | iowrite32(0x10001000, yp->base + TxCtrl); | ||
881 | |||
882 | if (yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE) | ||
883 | netif_start_queue (dev); /* Typical path */ | ||
884 | else | ||
885 | yp->tx_full = 1; | ||
886 | |||
887 | if (yellowfin_debug > 4) { | ||
888 | netdev_printk(KERN_DEBUG, dev, "Yellowfin transmit frame #%d queued in slot %d\n", | ||
889 | yp->cur_tx, entry); | ||
890 | } | ||
891 | return NETDEV_TX_OK; | ||
892 | } | ||
893 | |||
894 | /* The interrupt handler does all of the Rx thread work and cleans up | ||
895 | after the Tx thread. */ | ||
896 | static irqreturn_t yellowfin_interrupt(int irq, void *dev_instance) | ||
897 | { | ||
898 | struct net_device *dev = dev_instance; | ||
899 | struct yellowfin_private *yp; | ||
900 | void __iomem *ioaddr; | ||
901 | int boguscnt = max_interrupt_work; | ||
902 | unsigned int handled = 0; | ||
903 | |||
904 | yp = netdev_priv(dev); | ||
905 | ioaddr = yp->base; | ||
906 | |||
907 | spin_lock (&yp->lock); | ||
908 | |||
909 | do { | ||
910 | u16 intr_status = ioread16(ioaddr + IntrClear); | ||
911 | |||
912 | if (yellowfin_debug > 4) | ||
913 | netdev_printk(KERN_DEBUG, dev, "Yellowfin interrupt, status %04x\n", | ||
914 | intr_status); | ||
915 | |||
916 | if (intr_status == 0) | ||
917 | break; | ||
918 | handled = 1; | ||
919 | |||
920 | if (intr_status & (IntrRxDone | IntrEarlyRx)) { | ||
921 | yellowfin_rx(dev); | ||
922 | iowrite32(0x10001000, ioaddr + RxCtrl); /* Wake Rx engine. */ | ||
923 | } | ||
924 | |||
925 | #ifdef NO_TXSTATS | ||
926 | for (; yp->cur_tx - yp->dirty_tx > 0; yp->dirty_tx++) { | ||
927 | int entry = yp->dirty_tx % TX_RING_SIZE; | ||
928 | struct sk_buff *skb; | ||
929 | |||
930 | if (yp->tx_ring[entry].result_status == 0) | ||
931 | break; | ||
932 | skb = yp->tx_skbuff[entry]; | ||
933 | dev->stats.tx_packets++; | ||
934 | dev->stats.tx_bytes += skb->len; | ||
935 | /* Free the original skb. */ | ||
936 | pci_unmap_single(yp->pci_dev, le32_to_cpu(yp->tx_ring[entry].addr), | ||
937 | skb->len, PCI_DMA_TODEVICE); | ||
938 | dev_kfree_skb_irq(skb); | ||
939 | yp->tx_skbuff[entry] = NULL; | ||
940 | } | ||
941 | if (yp->tx_full && | ||
942 | yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE - 4) { | ||
943 | /* The ring is no longer full, clear tbusy. */ | ||
944 | yp->tx_full = 0; | ||
945 | netif_wake_queue(dev); | ||
946 | } | ||
947 | #else | ||
948 | if ((intr_status & IntrTxDone) || (yp->tx_tail_desc->tx_errs)) { | ||
949 | unsigned dirty_tx = yp->dirty_tx; | ||
950 | |||
951 | for (dirty_tx = yp->dirty_tx; yp->cur_tx - dirty_tx > 0; | ||
952 | dirty_tx++) { | ||
953 | /* Todo: optimize this. */ | ||
954 | int entry = dirty_tx % TX_RING_SIZE; | ||
955 | u16 tx_errs = yp->tx_status[entry].tx_errs; | ||
956 | struct sk_buff *skb; | ||
957 | |||
958 | #ifndef final_version | ||
959 | if (yellowfin_debug > 5) | ||
960 | netdev_printk(KERN_DEBUG, dev, "Tx queue %d check, Tx status %04x %04x %04x %04x\n", | ||
961 | entry, | ||
962 | yp->tx_status[entry].tx_cnt, | ||
963 | yp->tx_status[entry].tx_errs, | ||
964 | yp->tx_status[entry].total_tx_cnt, | ||
965 | yp->tx_status[entry].paused); | ||
966 | #endif | ||
967 | if (tx_errs == 0) | ||
968 | break; /* It still hasn't been Txed */ | ||
969 | skb = yp->tx_skbuff[entry]; | ||
970 | if (tx_errs & 0xF810) { | ||
971 | /* There was an major error, log it. */ | ||
972 | #ifndef final_version | ||
973 | if (yellowfin_debug > 1) | ||
974 | netdev_printk(KERN_DEBUG, dev, "Transmit error, Tx status %04x\n", | ||
975 | tx_errs); | ||
976 | #endif | ||
977 | dev->stats.tx_errors++; | ||
978 | if (tx_errs & 0xF800) dev->stats.tx_aborted_errors++; | ||
979 | if (tx_errs & 0x0800) dev->stats.tx_carrier_errors++; | ||
980 | if (tx_errs & 0x2000) dev->stats.tx_window_errors++; | ||
981 | if (tx_errs & 0x8000) dev->stats.tx_fifo_errors++; | ||
982 | } else { | ||
983 | #ifndef final_version | ||
984 | if (yellowfin_debug > 4) | ||
985 | netdev_printk(KERN_DEBUG, dev, "Normal transmit, Tx status %04x\n", | ||
986 | tx_errs); | ||
987 | #endif | ||
988 | dev->stats.tx_bytes += skb->len; | ||
989 | dev->stats.collisions += tx_errs & 15; | ||
990 | dev->stats.tx_packets++; | ||
991 | } | ||
992 | /* Free the original skb. */ | ||
993 | pci_unmap_single(yp->pci_dev, | ||
994 | yp->tx_ring[entry<<1].addr, skb->len, | ||
995 | PCI_DMA_TODEVICE); | ||
996 | dev_kfree_skb_irq(skb); | ||
997 | yp->tx_skbuff[entry] = 0; | ||
998 | /* Mark status as empty. */ | ||
999 | yp->tx_status[entry].tx_errs = 0; | ||
1000 | } | ||
1001 | |||
1002 | #ifndef final_version | ||
1003 | if (yp->cur_tx - dirty_tx > TX_RING_SIZE) { | ||
1004 | netdev_err(dev, "Out-of-sync dirty pointer, %d vs. %d, full=%d\n", | ||
1005 | dirty_tx, yp->cur_tx, yp->tx_full); | ||
1006 | dirty_tx += TX_RING_SIZE; | ||
1007 | } | ||
1008 | #endif | ||
1009 | |||
1010 | if (yp->tx_full && | ||
1011 | yp->cur_tx - dirty_tx < TX_QUEUE_SIZE - 2) { | ||
1012 | /* The ring is no longer full, clear tbusy. */ | ||
1013 | yp->tx_full = 0; | ||
1014 | netif_wake_queue(dev); | ||
1015 | } | ||
1016 | |||
1017 | yp->dirty_tx = dirty_tx; | ||
1018 | yp->tx_tail_desc = &yp->tx_status[dirty_tx % TX_RING_SIZE]; | ||
1019 | } | ||
1020 | #endif | ||
1021 | |||
1022 | /* Log errors and other uncommon events. */ | ||
1023 | if (intr_status & 0x2ee) /* Abnormal error summary. */ | ||
1024 | yellowfin_error(dev, intr_status); | ||
1025 | |||
1026 | if (--boguscnt < 0) { | ||
1027 | netdev_warn(dev, "Too much work at interrupt, status=%#04x\n", | ||
1028 | intr_status); | ||
1029 | break; | ||
1030 | } | ||
1031 | } while (1); | ||
1032 | |||
1033 | if (yellowfin_debug > 3) | ||
1034 | netdev_printk(KERN_DEBUG, dev, "exiting interrupt, status=%#04x\n", | ||
1035 | ioread16(ioaddr + IntrStatus)); | ||
1036 | |||
1037 | spin_unlock (&yp->lock); | ||
1038 | return IRQ_RETVAL(handled); | ||
1039 | } | ||
1040 | |||
1041 | /* This routine is logically part of the interrupt handler, but separated | ||
1042 | for clarity and better register allocation. */ | ||
1043 | static int yellowfin_rx(struct net_device *dev) | ||
1044 | { | ||
1045 | struct yellowfin_private *yp = netdev_priv(dev); | ||
1046 | int entry = yp->cur_rx % RX_RING_SIZE; | ||
1047 | int boguscnt = yp->dirty_rx + RX_RING_SIZE - yp->cur_rx; | ||
1048 | |||
1049 | if (yellowfin_debug > 4) { | ||
1050 | printk(KERN_DEBUG " In yellowfin_rx(), entry %d status %08x\n", | ||
1051 | entry, yp->rx_ring[entry].result_status); | ||
1052 | printk(KERN_DEBUG " #%d desc. %08x %08x %08x\n", | ||
1053 | entry, yp->rx_ring[entry].dbdma_cmd, yp->rx_ring[entry].addr, | ||
1054 | yp->rx_ring[entry].result_status); | ||
1055 | } | ||
1056 | |||
1057 | /* If EOP is set on the next entry, it's a new packet. Send it up. */ | ||
1058 | while (1) { | ||
1059 | struct yellowfin_desc *desc = &yp->rx_ring[entry]; | ||
1060 | struct sk_buff *rx_skb = yp->rx_skbuff[entry]; | ||
1061 | s16 frame_status; | ||
1062 | u16 desc_status; | ||
1063 | int data_size; | ||
1064 | u8 *buf_addr; | ||
1065 | |||
1066 | if(!desc->result_status) | ||
1067 | break; | ||
1068 | pci_dma_sync_single_for_cpu(yp->pci_dev, le32_to_cpu(desc->addr), | ||
1069 | yp->rx_buf_sz, PCI_DMA_FROMDEVICE); | ||
1070 | desc_status = le32_to_cpu(desc->result_status) >> 16; | ||
1071 | buf_addr = rx_skb->data; | ||
1072 | data_size = (le32_to_cpu(desc->dbdma_cmd) - | ||
1073 | le32_to_cpu(desc->result_status)) & 0xffff; | ||
1074 | frame_status = get_unaligned_le16(&(buf_addr[data_size - 2])); | ||
1075 | if (yellowfin_debug > 4) | ||
1076 | printk(KERN_DEBUG " %s() status was %04x\n", | ||
1077 | __func__, frame_status); | ||
1078 | if (--boguscnt < 0) | ||
1079 | break; | ||
1080 | if ( ! (desc_status & RX_EOP)) { | ||
1081 | if (data_size != 0) | ||
1082 | netdev_warn(dev, "Oversized Ethernet frame spanned multiple buffers, status %04x, data_size %d!\n", | ||
1083 | desc_status, data_size); | ||
1084 | dev->stats.rx_length_errors++; | ||
1085 | } else if ((yp->drv_flags & IsGigabit) && (frame_status & 0x0038)) { | ||
1086 | /* There was a error. */ | ||
1087 | if (yellowfin_debug > 3) | ||
1088 | printk(KERN_DEBUG " %s() Rx error was %04x\n", | ||
1089 | __func__, frame_status); | ||
1090 | dev->stats.rx_errors++; | ||
1091 | if (frame_status & 0x0060) dev->stats.rx_length_errors++; | ||
1092 | if (frame_status & 0x0008) dev->stats.rx_frame_errors++; | ||
1093 | if (frame_status & 0x0010) dev->stats.rx_crc_errors++; | ||
1094 | if (frame_status < 0) dev->stats.rx_dropped++; | ||
1095 | } else if ( !(yp->drv_flags & IsGigabit) && | ||
1096 | ((buf_addr[data_size-1] & 0x85) || buf_addr[data_size-2] & 0xC0)) { | ||
1097 | u8 status1 = buf_addr[data_size-2]; | ||
1098 | u8 status2 = buf_addr[data_size-1]; | ||
1099 | dev->stats.rx_errors++; | ||
1100 | if (status1 & 0xC0) dev->stats.rx_length_errors++; | ||
1101 | if (status2 & 0x03) dev->stats.rx_frame_errors++; | ||
1102 | if (status2 & 0x04) dev->stats.rx_crc_errors++; | ||
1103 | if (status2 & 0x80) dev->stats.rx_dropped++; | ||
1104 | #ifdef YF_PROTOTYPE /* Support for prototype hardware errata. */ | ||
1105 | } else if ((yp->flags & HasMACAddrBug) && | ||
1106 | memcmp(le32_to_cpu(yp->rx_ring_dma + | ||
1107 | entry*sizeof(struct yellowfin_desc)), | ||
1108 | dev->dev_addr, 6) != 0 && | ||
1109 | memcmp(le32_to_cpu(yp->rx_ring_dma + | ||
1110 | entry*sizeof(struct yellowfin_desc)), | ||
1111 | "\377\377\377\377\377\377", 6) != 0) { | ||
1112 | if (bogus_rx++ == 0) | ||
1113 | netdev_warn(dev, "Bad frame to %pM\n", | ||
1114 | buf_addr); | ||
1115 | #endif | ||
1116 | } else { | ||
1117 | struct sk_buff *skb; | ||
1118 | int pkt_len = data_size - | ||
1119 | (yp->chip_id ? 7 : 8 + buf_addr[data_size - 8]); | ||
1120 | /* To verify: Yellowfin Length should omit the CRC! */ | ||
1121 | |||
1122 | #ifndef final_version | ||
1123 | if (yellowfin_debug > 4) | ||
1124 | printk(KERN_DEBUG " %s() normal Rx pkt length %d of %d, bogus_cnt %d\n", | ||
1125 | __func__, pkt_len, data_size, boguscnt); | ||
1126 | #endif | ||
1127 | /* Check if the packet is long enough to just pass up the skbuff | ||
1128 | without copying to a properly sized skbuff. */ | ||
1129 | if (pkt_len > rx_copybreak) { | ||
1130 | skb_put(skb = rx_skb, pkt_len); | ||
1131 | pci_unmap_single(yp->pci_dev, | ||
1132 | le32_to_cpu(yp->rx_ring[entry].addr), | ||
1133 | yp->rx_buf_sz, | ||
1134 | PCI_DMA_FROMDEVICE); | ||
1135 | yp->rx_skbuff[entry] = NULL; | ||
1136 | } else { | ||
1137 | skb = dev_alloc_skb(pkt_len + 2); | ||
1138 | if (skb == NULL) | ||
1139 | break; | ||
1140 | skb_reserve(skb, 2); /* 16 byte align the IP header */ | ||
1141 | skb_copy_to_linear_data(skb, rx_skb->data, pkt_len); | ||
1142 | skb_put(skb, pkt_len); | ||
1143 | pci_dma_sync_single_for_device(yp->pci_dev, | ||
1144 | le32_to_cpu(desc->addr), | ||
1145 | yp->rx_buf_sz, | ||
1146 | PCI_DMA_FROMDEVICE); | ||
1147 | } | ||
1148 | skb->protocol = eth_type_trans(skb, dev); | ||
1149 | netif_rx(skb); | ||
1150 | dev->stats.rx_packets++; | ||
1151 | dev->stats.rx_bytes += pkt_len; | ||
1152 | } | ||
1153 | entry = (++yp->cur_rx) % RX_RING_SIZE; | ||
1154 | } | ||
1155 | |||
1156 | /* Refill the Rx ring buffers. */ | ||
1157 | for (; yp->cur_rx - yp->dirty_rx > 0; yp->dirty_rx++) { | ||
1158 | entry = yp->dirty_rx % RX_RING_SIZE; | ||
1159 | if (yp->rx_skbuff[entry] == NULL) { | ||
1160 | struct sk_buff *skb = dev_alloc_skb(yp->rx_buf_sz + 2); | ||
1161 | if (skb == NULL) | ||
1162 | break; /* Better luck next round. */ | ||
1163 | yp->rx_skbuff[entry] = skb; | ||
1164 | skb->dev = dev; /* Mark as being used by this device. */ | ||
1165 | skb_reserve(skb, 2); /* Align IP on 16 byte boundaries */ | ||
1166 | yp->rx_ring[entry].addr = cpu_to_le32(pci_map_single(yp->pci_dev, | ||
1167 | skb->data, yp->rx_buf_sz, PCI_DMA_FROMDEVICE)); | ||
1168 | } | ||
1169 | yp->rx_ring[entry].dbdma_cmd = cpu_to_le32(CMD_STOP); | ||
1170 | yp->rx_ring[entry].result_status = 0; /* Clear complete bit. */ | ||
1171 | if (entry != 0) | ||
1172 | yp->rx_ring[entry - 1].dbdma_cmd = | ||
1173 | cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | yp->rx_buf_sz); | ||
1174 | else | ||
1175 | yp->rx_ring[RX_RING_SIZE - 1].dbdma_cmd = | ||
1176 | cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | BRANCH_ALWAYS | ||
1177 | | yp->rx_buf_sz); | ||
1178 | } | ||
1179 | |||
1180 | return 0; | ||
1181 | } | ||
1182 | |||
1183 | static void yellowfin_error(struct net_device *dev, int intr_status) | ||
1184 | { | ||
1185 | netdev_err(dev, "Something Wicked happened! %04x\n", intr_status); | ||
1186 | /* Hmmmmm, it's not clear what to do here. */ | ||
1187 | if (intr_status & (IntrTxPCIErr | IntrTxPCIFault)) | ||
1188 | dev->stats.tx_errors++; | ||
1189 | if (intr_status & (IntrRxPCIErr | IntrRxPCIFault)) | ||
1190 | dev->stats.rx_errors++; | ||
1191 | } | ||
1192 | |||
1193 | static int yellowfin_close(struct net_device *dev) | ||
1194 | { | ||
1195 | struct yellowfin_private *yp = netdev_priv(dev); | ||
1196 | void __iomem *ioaddr = yp->base; | ||
1197 | int i; | ||
1198 | |||
1199 | netif_stop_queue (dev); | ||
1200 | |||
1201 | if (yellowfin_debug > 1) { | ||
1202 | netdev_printk(KERN_DEBUG, dev, "Shutting down ethercard, status was Tx %04x Rx %04x Int %02x\n", | ||
1203 | ioread16(ioaddr + TxStatus), | ||
1204 | ioread16(ioaddr + RxStatus), | ||
1205 | ioread16(ioaddr + IntrStatus)); | ||
1206 | netdev_printk(KERN_DEBUG, dev, "Queue pointers were Tx %d / %d, Rx %d / %d\n", | ||
1207 | yp->cur_tx, yp->dirty_tx, | ||
1208 | yp->cur_rx, yp->dirty_rx); | ||
1209 | } | ||
1210 | |||
1211 | /* Disable interrupts by clearing the interrupt mask. */ | ||
1212 | iowrite16(0x0000, ioaddr + IntrEnb); | ||
1213 | |||
1214 | /* Stop the chip's Tx and Rx processes. */ | ||
1215 | iowrite32(0x80000000, ioaddr + RxCtrl); | ||
1216 | iowrite32(0x80000000, ioaddr + TxCtrl); | ||
1217 | |||
1218 | del_timer(&yp->timer); | ||
1219 | |||
1220 | #if defined(__i386__) | ||
1221 | if (yellowfin_debug > 2) { | ||
1222 | printk(KERN_DEBUG " Tx ring at %08llx:\n", | ||
1223 | (unsigned long long)yp->tx_ring_dma); | ||
1224 | for (i = 0; i < TX_RING_SIZE*2; i++) | ||
1225 | printk(KERN_DEBUG " %c #%d desc. %08x %08x %08x %08x\n", | ||
1226 | ioread32(ioaddr + TxPtr) == (long)&yp->tx_ring[i] ? '>' : ' ', | ||
1227 | i, yp->tx_ring[i].dbdma_cmd, yp->tx_ring[i].addr, | ||
1228 | yp->tx_ring[i].branch_addr, yp->tx_ring[i].result_status); | ||
1229 | printk(KERN_DEBUG " Tx status %p:\n", yp->tx_status); | ||
1230 | for (i = 0; i < TX_RING_SIZE; i++) | ||
1231 | printk(KERN_DEBUG " #%d status %04x %04x %04x %04x\n", | ||
1232 | i, yp->tx_status[i].tx_cnt, yp->tx_status[i].tx_errs, | ||
1233 | yp->tx_status[i].total_tx_cnt, yp->tx_status[i].paused); | ||
1234 | |||
1235 | printk(KERN_DEBUG " Rx ring %08llx:\n", | ||
1236 | (unsigned long long)yp->rx_ring_dma); | ||
1237 | for (i = 0; i < RX_RING_SIZE; i++) { | ||
1238 | printk(KERN_DEBUG " %c #%d desc. %08x %08x %08x\n", | ||
1239 | ioread32(ioaddr + RxPtr) == (long)&yp->rx_ring[i] ? '>' : ' ', | ||
1240 | i, yp->rx_ring[i].dbdma_cmd, yp->rx_ring[i].addr, | ||
1241 | yp->rx_ring[i].result_status); | ||
1242 | if (yellowfin_debug > 6) { | ||
1243 | if (get_unaligned((u8*)yp->rx_ring[i].addr) != 0x69) { | ||
1244 | int j; | ||
1245 | |||
1246 | printk(KERN_DEBUG); | ||
1247 | for (j = 0; j < 0x50; j++) | ||
1248 | pr_cont(" %04x", | ||
1249 | get_unaligned(((u16*)yp->rx_ring[i].addr) + j)); | ||
1250 | pr_cont("\n"); | ||
1251 | } | ||
1252 | } | ||
1253 | } | ||
1254 | } | ||
1255 | #endif /* __i386__ debugging only */ | ||
1256 | |||
1257 | free_irq(dev->irq, dev); | ||
1258 | |||
1259 | /* Free all the skbuffs in the Rx queue. */ | ||
1260 | for (i = 0; i < RX_RING_SIZE; i++) { | ||
1261 | yp->rx_ring[i].dbdma_cmd = cpu_to_le32(CMD_STOP); | ||
1262 | yp->rx_ring[i].addr = cpu_to_le32(0xBADF00D0); /* An invalid address. */ | ||
1263 | if (yp->rx_skbuff[i]) { | ||
1264 | dev_kfree_skb(yp->rx_skbuff[i]); | ||
1265 | } | ||
1266 | yp->rx_skbuff[i] = NULL; | ||
1267 | } | ||
1268 | for (i = 0; i < TX_RING_SIZE; i++) { | ||
1269 | if (yp->tx_skbuff[i]) | ||
1270 | dev_kfree_skb(yp->tx_skbuff[i]); | ||
1271 | yp->tx_skbuff[i] = NULL; | ||
1272 | } | ||
1273 | |||
1274 | #ifdef YF_PROTOTYPE /* Support for prototype hardware errata. */ | ||
1275 | if (yellowfin_debug > 0) { | ||
1276 | netdev_printk(KERN_DEBUG, dev, "Received %d frames that we should not have\n", | ||
1277 | bogus_rx); | ||
1278 | } | ||
1279 | #endif | ||
1280 | |||
1281 | return 0; | ||
1282 | } | ||
1283 | |||
1284 | /* Set or clear the multicast filter for this adaptor. */ | ||
1285 | |||
1286 | static void set_rx_mode(struct net_device *dev) | ||
1287 | { | ||
1288 | struct yellowfin_private *yp = netdev_priv(dev); | ||
1289 | void __iomem *ioaddr = yp->base; | ||
1290 | u16 cfg_value = ioread16(ioaddr + Cnfg); | ||
1291 | |||
1292 | /* Stop the Rx process to change any value. */ | ||
1293 | iowrite16(cfg_value & ~0x1000, ioaddr + Cnfg); | ||
1294 | if (dev->flags & IFF_PROMISC) { /* Set promiscuous. */ | ||
1295 | iowrite16(0x000F, ioaddr + AddrMode); | ||
1296 | } else if ((netdev_mc_count(dev) > 64) || | ||
1297 | (dev->flags & IFF_ALLMULTI)) { | ||
1298 | /* Too many to filter well, or accept all multicasts. */ | ||
1299 | iowrite16(0x000B, ioaddr + AddrMode); | ||
1300 | } else if (!netdev_mc_empty(dev)) { /* Must use the multicast hash table. */ | ||
1301 | struct netdev_hw_addr *ha; | ||
1302 | u16 hash_table[4]; | ||
1303 | int i; | ||
1304 | |||
1305 | memset(hash_table, 0, sizeof(hash_table)); | ||
1306 | netdev_for_each_mc_addr(ha, dev) { | ||
1307 | unsigned int bit; | ||
1308 | |||
1309 | /* Due to a bug in the early chip versions, multiple filter | ||
1310 | slots must be set for each address. */ | ||
1311 | if (yp->drv_flags & HasMulticastBug) { | ||
1312 | bit = (ether_crc_le(3, ha->addr) >> 3) & 0x3f; | ||
1313 | hash_table[bit >> 4] |= (1 << bit); | ||
1314 | bit = (ether_crc_le(4, ha->addr) >> 3) & 0x3f; | ||
1315 | hash_table[bit >> 4] |= (1 << bit); | ||
1316 | bit = (ether_crc_le(5, ha->addr) >> 3) & 0x3f; | ||
1317 | hash_table[bit >> 4] |= (1 << bit); | ||
1318 | } | ||
1319 | bit = (ether_crc_le(6, ha->addr) >> 3) & 0x3f; | ||
1320 | hash_table[bit >> 4] |= (1 << bit); | ||
1321 | } | ||
1322 | /* Copy the hash table to the chip. */ | ||
1323 | for (i = 0; i < 4; i++) | ||
1324 | iowrite16(hash_table[i], ioaddr + HashTbl + i*2); | ||
1325 | iowrite16(0x0003, ioaddr + AddrMode); | ||
1326 | } else { /* Normal, unicast/broadcast-only mode. */ | ||
1327 | iowrite16(0x0001, ioaddr + AddrMode); | ||
1328 | } | ||
1329 | /* Restart the Rx process. */ | ||
1330 | iowrite16(cfg_value | 0x1000, ioaddr + Cnfg); | ||
1331 | } | ||
1332 | |||
1333 | static void yellowfin_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) | ||
1334 | { | ||
1335 | struct yellowfin_private *np = netdev_priv(dev); | ||
1336 | strcpy(info->driver, DRV_NAME); | ||
1337 | strcpy(info->version, DRV_VERSION); | ||
1338 | strcpy(info->bus_info, pci_name(np->pci_dev)); | ||
1339 | } | ||
1340 | |||
1341 | static const struct ethtool_ops ethtool_ops = { | ||
1342 | .get_drvinfo = yellowfin_get_drvinfo | ||
1343 | }; | ||
1344 | |||
1345 | static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd) | ||
1346 | { | ||
1347 | struct yellowfin_private *np = netdev_priv(dev); | ||
1348 | void __iomem *ioaddr = np->base; | ||
1349 | struct mii_ioctl_data *data = if_mii(rq); | ||
1350 | |||
1351 | switch(cmd) { | ||
1352 | case SIOCGMIIPHY: /* Get address of MII PHY in use. */ | ||
1353 | data->phy_id = np->phys[0] & 0x1f; | ||
1354 | /* Fall Through */ | ||
1355 | |||
1356 | case SIOCGMIIREG: /* Read MII PHY register. */ | ||
1357 | data->val_out = mdio_read(ioaddr, data->phy_id & 0x1f, data->reg_num & 0x1f); | ||
1358 | return 0; | ||
1359 | |||
1360 | case SIOCSMIIREG: /* Write MII PHY register. */ | ||
1361 | if (data->phy_id == np->phys[0]) { | ||
1362 | u16 value = data->val_in; | ||
1363 | switch (data->reg_num) { | ||
1364 | case 0: | ||
1365 | /* Check for autonegotiation on or reset. */ | ||
1366 | np->medialock = (value & 0x9000) ? 0 : 1; | ||
1367 | if (np->medialock) | ||
1368 | np->full_duplex = (value & 0x0100) ? 1 : 0; | ||
1369 | break; | ||
1370 | case 4: np->advertising = value; break; | ||
1371 | } | ||
1372 | /* Perhaps check_duplex(dev), depending on chip semantics. */ | ||
1373 | } | ||
1374 | mdio_write(ioaddr, data->phy_id & 0x1f, data->reg_num & 0x1f, data->val_in); | ||
1375 | return 0; | ||
1376 | default: | ||
1377 | return -EOPNOTSUPP; | ||
1378 | } | ||
1379 | } | ||
1380 | |||
1381 | |||
1382 | static void __devexit yellowfin_remove_one (struct pci_dev *pdev) | ||
1383 | { | ||
1384 | struct net_device *dev = pci_get_drvdata(pdev); | ||
1385 | struct yellowfin_private *np; | ||
1386 | |||
1387 | BUG_ON(!dev); | ||
1388 | np = netdev_priv(dev); | ||
1389 | |||
1390 | pci_free_consistent(pdev, STATUS_TOTAL_SIZE, np->tx_status, | ||
1391 | np->tx_status_dma); | ||
1392 | pci_free_consistent(pdev, RX_TOTAL_SIZE, np->rx_ring, np->rx_ring_dma); | ||
1393 | pci_free_consistent(pdev, TX_TOTAL_SIZE, np->tx_ring, np->tx_ring_dma); | ||
1394 | unregister_netdev (dev); | ||
1395 | |||
1396 | pci_iounmap(pdev, np->base); | ||
1397 | |||
1398 | pci_release_regions (pdev); | ||
1399 | |||
1400 | free_netdev (dev); | ||
1401 | pci_set_drvdata(pdev, NULL); | ||
1402 | } | ||
1403 | |||
1404 | |||
1405 | static struct pci_driver yellowfin_driver = { | ||
1406 | .name = DRV_NAME, | ||
1407 | .id_table = yellowfin_pci_tbl, | ||
1408 | .probe = yellowfin_init_one, | ||
1409 | .remove = __devexit_p(yellowfin_remove_one), | ||
1410 | }; | ||
1411 | |||
1412 | |||
1413 | static int __init yellowfin_init (void) | ||
1414 | { | ||
1415 | /* when a module, this is printed whether or not devices are found in probe */ | ||
1416 | #ifdef MODULE | ||
1417 | printk(version); | ||
1418 | #endif | ||
1419 | return pci_register_driver(&yellowfin_driver); | ||
1420 | } | ||
1421 | |||
1422 | |||
1423 | static void __exit yellowfin_cleanup (void) | ||
1424 | { | ||
1425 | pci_unregister_driver (&yellowfin_driver); | ||
1426 | } | ||
1427 | |||
1428 | |||
1429 | module_init(yellowfin_init); | ||
1430 | module_exit(yellowfin_cleanup); | ||