Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * acenic.c: Linux driver for the Alteon AceNIC Gigabit Ethernet card |
| 3 | * and other Tigon based cards. |
| 4 | * |
| 5 | * Copyright 1998-2002 by Jes Sorensen, <jes@trained-monkey.org>. |
| 6 | * |
| 7 | * Thanks to Alteon and 3Com for providing hardware and documentation |
| 8 | * enabling me to write this driver. |
| 9 | * |
| 10 | * A mailing list for discussing the use of this driver has been |
| 11 | * setup, please subscribe to the lists if you have any questions |
| 12 | * about the driver. Send mail to linux-acenic-help@sunsite.auc.dk to |
| 13 | * see how to subscribe. |
| 14 | * |
| 15 | * This program is free software; you can redistribute it and/or modify |
| 16 | * it under the terms of the GNU General Public License as published by |
| 17 | * the Free Software Foundation; either version 2 of the License, or |
| 18 | * (at your option) any later version. |
| 19 | * |
| 20 | * Additional credits: |
| 21 | * Pete Wyckoff <wyckoff@ca.sandia.gov>: Initial Linux/Alpha and trace |
| 22 | * dump support. The trace dump support has not been |
| 23 | * integrated yet however. |
| 24 | * Troy Benjegerdes: Big Endian (PPC) patches. |
| 25 | * Nate Stahl: Better out of memory handling and stats support. |
| 26 | * Aman Singla: Nasty race between interrupt handler and tx code dealing |
| 27 | * with 'testing the tx_ret_csm and setting tx_full' |
| 28 | * David S. Miller <davem@redhat.com>: conversion to new PCI dma mapping |
| 29 | * infrastructure and Sparc support |
| 30 | * Pierrick Pinasseau (CERN): For lending me an Ultra 5 to test the |
| 31 | * driver under Linux/Sparc64 |
| 32 | * Matt Domsch <Matt_Domsch@dell.com>: Detect Alteon 1000baseT cards |
| 33 | * ETHTOOL_GDRVINFO support |
| 34 | * Chip Salzenberg <chip@valinux.com>: Fix race condition between tx |
| 35 | * handler and close() cleanup. |
| 36 | * Ken Aaker <kdaaker@rchland.vnet.ibm.com>: Correct check for whether |
| 37 | * memory mapped IO is enabled to |
| 38 | * make the driver work on RS/6000. |
| 39 | * Takayoshi Kouchi <kouchi@hpc.bs1.fc.nec.co.jp>: Identifying problem |
| 40 | * where the driver would disable |
| 41 | * bus master mode if it had to disable |
| 42 | * write and invalidate. |
| 43 | * Stephen Hack <stephen_hack@hp.com>: Fixed ace_set_mac_addr for little |
| 44 | * endian systems. |
| 45 | * Val Henson <vhenson@esscom.com>: Reset Jumbo skb producer and |
| 46 | * rx producer index when |
| 47 | * flushing the Jumbo ring. |
| 48 | * Hans Grobler <grobh@sun.ac.za>: Memory leak fixes in the |
| 49 | * driver init path. |
| 50 | * Grant Grundler <grundler@cup.hp.com>: PCI write posting fixes. |
| 51 | */ |
| 52 | |
| 53 | #include <linux/config.h> |
| 54 | #include <linux/module.h> |
| 55 | #include <linux/moduleparam.h> |
| 56 | #include <linux/version.h> |
| 57 | #include <linux/types.h> |
| 58 | #include <linux/errno.h> |
| 59 | #include <linux/ioport.h> |
| 60 | #include <linux/pci.h> |
| 61 | #include <linux/kernel.h> |
| 62 | #include <linux/netdevice.h> |
| 63 | #include <linux/etherdevice.h> |
| 64 | #include <linux/skbuff.h> |
| 65 | #include <linux/init.h> |
| 66 | #include <linux/delay.h> |
| 67 | #include <linux/mm.h> |
| 68 | #include <linux/highmem.h> |
| 69 | #include <linux/sockios.h> |
| 70 | |
| 71 | #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE) |
| 72 | #include <linux/if_vlan.h> |
| 73 | #endif |
| 74 | |
| 75 | #ifdef SIOCETHTOOL |
| 76 | #include <linux/ethtool.h> |
| 77 | #endif |
| 78 | |
| 79 | #include <net/sock.h> |
| 80 | #include <net/ip.h> |
| 81 | |
| 82 | #include <asm/system.h> |
| 83 | #include <asm/io.h> |
| 84 | #include <asm/irq.h> |
| 85 | #include <asm/byteorder.h> |
| 86 | #include <asm/uaccess.h> |
| 87 | |
| 88 | |
| 89 | #define DRV_NAME "acenic" |
| 90 | |
| 91 | #undef INDEX_DEBUG |
| 92 | |
| 93 | #ifdef CONFIG_ACENIC_OMIT_TIGON_I |
| 94 | #define ACE_IS_TIGON_I(ap) 0 |
| 95 | #define ACE_TX_RING_ENTRIES(ap) MAX_TX_RING_ENTRIES |
| 96 | #else |
| 97 | #define ACE_IS_TIGON_I(ap) (ap->version == 1) |
| 98 | #define ACE_TX_RING_ENTRIES(ap) ap->tx_ring_entries |
| 99 | #endif |
| 100 | |
| 101 | #ifndef PCI_VENDOR_ID_ALTEON |
| 102 | #define PCI_VENDOR_ID_ALTEON 0x12ae |
| 103 | #endif |
| 104 | #ifndef PCI_DEVICE_ID_ALTEON_ACENIC_FIBRE |
| 105 | #define PCI_DEVICE_ID_ALTEON_ACENIC_FIBRE 0x0001 |
| 106 | #define PCI_DEVICE_ID_ALTEON_ACENIC_COPPER 0x0002 |
| 107 | #endif |
| 108 | #ifndef PCI_DEVICE_ID_3COM_3C985 |
| 109 | #define PCI_DEVICE_ID_3COM_3C985 0x0001 |
| 110 | #endif |
| 111 | #ifndef PCI_VENDOR_ID_NETGEAR |
| 112 | #define PCI_VENDOR_ID_NETGEAR 0x1385 |
| 113 | #define PCI_DEVICE_ID_NETGEAR_GA620 0x620a |
| 114 | #endif |
| 115 | #ifndef PCI_DEVICE_ID_NETGEAR_GA620T |
| 116 | #define PCI_DEVICE_ID_NETGEAR_GA620T 0x630a |
| 117 | #endif |
| 118 | |
| 119 | |
| 120 | /* |
| 121 | * Farallon used the DEC vendor ID by mistake and they seem not |
| 122 | * to care - stinky! |
| 123 | */ |
| 124 | #ifndef PCI_DEVICE_ID_FARALLON_PN9000SX |
| 125 | #define PCI_DEVICE_ID_FARALLON_PN9000SX 0x1a |
| 126 | #endif |
| 127 | #ifndef PCI_DEVICE_ID_FARALLON_PN9100T |
| 128 | #define PCI_DEVICE_ID_FARALLON_PN9100T 0xfa |
| 129 | #endif |
| 130 | #ifndef PCI_VENDOR_ID_SGI |
| 131 | #define PCI_VENDOR_ID_SGI 0x10a9 |
| 132 | #endif |
| 133 | #ifndef PCI_DEVICE_ID_SGI_ACENIC |
| 134 | #define PCI_DEVICE_ID_SGI_ACENIC 0x0009 |
| 135 | #endif |
| 136 | |
| 137 | static struct pci_device_id acenic_pci_tbl[] = { |
| 138 | { PCI_VENDOR_ID_ALTEON, PCI_DEVICE_ID_ALTEON_ACENIC_FIBRE, |
| 139 | PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, }, |
| 140 | { PCI_VENDOR_ID_ALTEON, PCI_DEVICE_ID_ALTEON_ACENIC_COPPER, |
| 141 | PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, }, |
| 142 | { PCI_VENDOR_ID_3COM, PCI_DEVICE_ID_3COM_3C985, |
| 143 | PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, }, |
| 144 | { PCI_VENDOR_ID_NETGEAR, PCI_DEVICE_ID_NETGEAR_GA620, |
| 145 | PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, }, |
| 146 | { PCI_VENDOR_ID_NETGEAR, PCI_DEVICE_ID_NETGEAR_GA620T, |
| 147 | PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, }, |
| 148 | /* |
| 149 | * Farallon used the DEC vendor ID on their cards incorrectly, |
| 150 | * then later Alteon's ID. |
| 151 | */ |
| 152 | { PCI_VENDOR_ID_DEC, PCI_DEVICE_ID_FARALLON_PN9000SX, |
| 153 | PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, }, |
| 154 | { PCI_VENDOR_ID_ALTEON, PCI_DEVICE_ID_FARALLON_PN9100T, |
| 155 | PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, }, |
| 156 | { PCI_VENDOR_ID_SGI, PCI_DEVICE_ID_SGI_ACENIC, |
| 157 | PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, }, |
| 158 | { } |
| 159 | }; |
| 160 | MODULE_DEVICE_TABLE(pci, acenic_pci_tbl); |
| 161 | |
| 162 | #ifndef SET_NETDEV_DEV |
| 163 | #define SET_NETDEV_DEV(net, pdev) do{} while(0) |
| 164 | #endif |
| 165 | |
| 166 | #if LINUX_VERSION_CODE >= 0x2051c |
| 167 | #define ace_sync_irq(irq) synchronize_irq(irq) |
| 168 | #else |
| 169 | #define ace_sync_irq(irq) synchronize_irq() |
| 170 | #endif |
| 171 | |
| 172 | #ifndef offset_in_page |
| 173 | #define offset_in_page(ptr) ((unsigned long)(ptr) & ~PAGE_MASK) |
| 174 | #endif |
| 175 | |
| 176 | #define ACE_MAX_MOD_PARMS 8 |
| 177 | #define BOARD_IDX_STATIC 0 |
| 178 | #define BOARD_IDX_OVERFLOW -1 |
| 179 | |
| 180 | #if (defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)) && \ |
| 181 | defined(NETIF_F_HW_VLAN_RX) |
| 182 | #define ACENIC_DO_VLAN 1 |
| 183 | #define ACE_RCB_VLAN_FLAG RCB_FLG_VLAN_ASSIST |
| 184 | #else |
| 185 | #define ACENIC_DO_VLAN 0 |
| 186 | #define ACE_RCB_VLAN_FLAG 0 |
| 187 | #endif |
| 188 | |
| 189 | #include "acenic.h" |
| 190 | |
| 191 | /* |
| 192 | * These must be defined before the firmware is included. |
| 193 | */ |
| 194 | #define MAX_TEXT_LEN 96*1024 |
| 195 | #define MAX_RODATA_LEN 8*1024 |
| 196 | #define MAX_DATA_LEN 2*1024 |
| 197 | |
| 198 | #include "acenic_firmware.h" |
| 199 | |
| 200 | #ifndef tigon2FwReleaseLocal |
| 201 | #define tigon2FwReleaseLocal 0 |
| 202 | #endif |
| 203 | |
| 204 | /* |
| 205 | * This driver currently supports Tigon I and Tigon II based cards |
| 206 | * including the Alteon AceNIC, the 3Com 3C985[B] and NetGear |
| 207 | * GA620. The driver should also work on the SGI, DEC and Farallon |
| 208 | * versions of the card, however I have not been able to test that |
| 209 | * myself. |
| 210 | * |
| 211 | * This card is really neat, it supports receive hardware checksumming |
| 212 | * and jumbo frames (up to 9000 bytes) and does a lot of work in the |
| 213 | * firmware. Also the programming interface is quite neat, except for |
| 214 | * the parts dealing with the i2c eeprom on the card ;-) |
| 215 | * |
| 216 | * Using jumbo frames: |
| 217 | * |
| 218 | * To enable jumbo frames, simply specify an mtu between 1500 and 9000 |
| 219 | * bytes to ifconfig. Jumbo frames can be enabled or disabled at any time |
| 220 | * by running `ifconfig eth<X> mtu <MTU>' with <X> being the Ethernet |
| 221 | * interface number and <MTU> being the MTU value. |
| 222 | * |
| 223 | * Module parameters: |
| 224 | * |
| 225 | * When compiled as a loadable module, the driver allows for a number |
| 226 | * of module parameters to be specified. The driver supports the |
| 227 | * following module parameters: |
| 228 | * |
| 229 | * trace=<val> - Firmware trace level. This requires special traced |
| 230 | * firmware to replace the firmware supplied with |
| 231 | * the driver - for debugging purposes only. |
| 232 | * |
| 233 | * link=<val> - Link state. Normally you want to use the default link |
| 234 | * parameters set by the driver. This can be used to |
| 235 | * override these in case your switch doesn't negotiate |
| 236 | * the link properly. Valid values are: |
| 237 | * 0x0001 - Force half duplex link. |
| 238 | * 0x0002 - Do not negotiate line speed with the other end. |
| 239 | * 0x0010 - 10Mbit/sec link. |
| 240 | * 0x0020 - 100Mbit/sec link. |
| 241 | * 0x0040 - 1000Mbit/sec link. |
| 242 | * 0x0100 - Do not negotiate flow control. |
| 243 | * 0x0200 - Enable RX flow control Y |
| 244 | * 0x0400 - Enable TX flow control Y (Tigon II NICs only). |
| 245 | * Default value is 0x0270, ie. enable link+flow |
| 246 | * control negotiation. Negotiating the highest |
| 247 | * possible link speed with RX flow control enabled. |
| 248 | * |
| 249 | * When disabling link speed negotiation, only one link |
| 250 | * speed is allowed to be specified! |
| 251 | * |
| 252 | * tx_coal_tick=<val> - number of coalescing clock ticks (us) allowed |
| 253 | * to wait for more packets to arive before |
| 254 | * interrupting the host, from the time the first |
| 255 | * packet arrives. |
| 256 | * |
| 257 | * rx_coal_tick=<val> - number of coalescing clock ticks (us) allowed |
| 258 | * to wait for more packets to arive in the transmit ring, |
| 259 | * before interrupting the host, after transmitting the |
| 260 | * first packet in the ring. |
| 261 | * |
| 262 | * max_tx_desc=<val> - maximum number of transmit descriptors |
| 263 | * (packets) transmitted before interrupting the host. |
| 264 | * |
| 265 | * max_rx_desc=<val> - maximum number of receive descriptors |
| 266 | * (packets) received before interrupting the host. |
| 267 | * |
| 268 | * tx_ratio=<val> - 7 bit value (0 - 63) specifying the split in 64th |
| 269 | * increments of the NIC's on board memory to be used for |
| 270 | * transmit and receive buffers. For the 1MB NIC app. 800KB |
| 271 | * is available, on the 1/2MB NIC app. 300KB is available. |
| 272 | * 68KB will always be available as a minimum for both |
| 273 | * directions. The default value is a 50/50 split. |
| 274 | * dis_pci_mem_inval=<val> - disable PCI memory write and invalidate |
| 275 | * operations, default (1) is to always disable this as |
| 276 | * that is what Alteon does on NT. I have not been able |
| 277 | * to measure any real performance differences with |
| 278 | * this on my systems. Set <val>=0 if you want to |
| 279 | * enable these operations. |
| 280 | * |
| 281 | * If you use more than one NIC, specify the parameters for the |
| 282 | * individual NICs with a comma, ie. trace=0,0x00001fff,0 you want to |
| 283 | * run tracing on NIC #2 but not on NIC #1 and #3. |
| 284 | * |
| 285 | * TODO: |
| 286 | * |
| 287 | * - Proper multicast support. |
| 288 | * - NIC dump support. |
| 289 | * - More tuning parameters. |
| 290 | * |
| 291 | * The mini ring is not used under Linux and I am not sure it makes sense |
| 292 | * to actually use it. |
| 293 | * |
| 294 | * New interrupt handler strategy: |
| 295 | * |
| 296 | * The old interrupt handler worked using the traditional method of |
| 297 | * replacing an skbuff with a new one when a packet arrives. However |
| 298 | * the rx rings do not need to contain a static number of buffer |
| 299 | * descriptors, thus it makes sense to move the memory allocation out |
| 300 | * of the main interrupt handler and do it in a bottom half handler |
| 301 | * and only allocate new buffers when the number of buffers in the |
| 302 | * ring is below a certain threshold. In order to avoid starving the |
| 303 | * NIC under heavy load it is however necessary to force allocation |
| 304 | * when hitting a minimum threshold. The strategy for alloction is as |
| 305 | * follows: |
| 306 | * |
| 307 | * RX_LOW_BUF_THRES - allocate buffers in the bottom half |
| 308 | * RX_PANIC_LOW_THRES - we are very low on buffers, allocate |
| 309 | * the buffers in the interrupt handler |
| 310 | * RX_RING_THRES - maximum number of buffers in the rx ring |
| 311 | * RX_MINI_THRES - maximum number of buffers in the mini ring |
| 312 | * RX_JUMBO_THRES - maximum number of buffers in the jumbo ring |
| 313 | * |
| 314 | * One advantagous side effect of this allocation approach is that the |
| 315 | * entire rx processing can be done without holding any spin lock |
| 316 | * since the rx rings and registers are totally independent of the tx |
| 317 | * ring and its registers. This of course includes the kmalloc's of |
| 318 | * new skb's. Thus start_xmit can run in parallel with rx processing |
| 319 | * and the memory allocation on SMP systems. |
| 320 | * |
| 321 | * Note that running the skb reallocation in a bottom half opens up |
| 322 | * another can of races which needs to be handled properly. In |
| 323 | * particular it can happen that the interrupt handler tries to run |
| 324 | * the reallocation while the bottom half is either running on another |
| 325 | * CPU or was interrupted on the same CPU. To get around this the |
| 326 | * driver uses bitops to prevent the reallocation routines from being |
| 327 | * reentered. |
| 328 | * |
| 329 | * TX handling can also be done without holding any spin lock, wheee |
| 330 | * this is fun! since tx_ret_csm is only written to by the interrupt |
| 331 | * handler. The case to be aware of is when shutting down the device |
| 332 | * and cleaning up where it is necessary to make sure that |
| 333 | * start_xmit() is not running while this is happening. Well DaveM |
| 334 | * informs me that this case is already protected against ... bye bye |
| 335 | * Mr. Spin Lock, it was nice to know you. |
| 336 | * |
| 337 | * TX interrupts are now partly disabled so the NIC will only generate |
| 338 | * TX interrupts for the number of coal ticks, not for the number of |
| 339 | * TX packets in the queue. This should reduce the number of TX only, |
| 340 | * ie. when no RX processing is done, interrupts seen. |
| 341 | */ |
| 342 | |
| 343 | /* |
| 344 | * Threshold values for RX buffer allocation - the low water marks for |
| 345 | * when to start refilling the rings are set to 75% of the ring |
| 346 | * sizes. It seems to make sense to refill the rings entirely from the |
| 347 | * intrrupt handler once it gets below the panic threshold, that way |
| 348 | * we don't risk that the refilling is moved to another CPU when the |
| 349 | * one running the interrupt handler just got the slab code hot in its |
| 350 | * cache. |
| 351 | */ |
| 352 | #define RX_RING_SIZE 72 |
| 353 | #define RX_MINI_SIZE 64 |
| 354 | #define RX_JUMBO_SIZE 48 |
| 355 | |
| 356 | #define RX_PANIC_STD_THRES 16 |
| 357 | #define RX_PANIC_STD_REFILL (3*RX_PANIC_STD_THRES)/2 |
| 358 | #define RX_LOW_STD_THRES (3*RX_RING_SIZE)/4 |
| 359 | #define RX_PANIC_MINI_THRES 12 |
| 360 | #define RX_PANIC_MINI_REFILL (3*RX_PANIC_MINI_THRES)/2 |
| 361 | #define RX_LOW_MINI_THRES (3*RX_MINI_SIZE)/4 |
| 362 | #define RX_PANIC_JUMBO_THRES 6 |
| 363 | #define RX_PANIC_JUMBO_REFILL (3*RX_PANIC_JUMBO_THRES)/2 |
| 364 | #define RX_LOW_JUMBO_THRES (3*RX_JUMBO_SIZE)/4 |
| 365 | |
| 366 | |
| 367 | /* |
| 368 | * Size of the mini ring entries, basically these just should be big |
| 369 | * enough to take TCP ACKs |
| 370 | */ |
| 371 | #define ACE_MINI_SIZE 100 |
| 372 | |
| 373 | #define ACE_MINI_BUFSIZE ACE_MINI_SIZE |
| 374 | #define ACE_STD_BUFSIZE (ACE_STD_MTU + ETH_HLEN + 4) |
| 375 | #define ACE_JUMBO_BUFSIZE (ACE_JUMBO_MTU + ETH_HLEN + 4) |
| 376 | |
| 377 | /* |
| 378 | * There seems to be a magic difference in the effect between 995 and 996 |
| 379 | * but little difference between 900 and 995 ... no idea why. |
| 380 | * |
| 381 | * There is now a default set of tuning parameters which is set, depending |
| 382 | * on whether or not the user enables Jumbo frames. It's assumed that if |
| 383 | * Jumbo frames are enabled, the user wants optimal tuning for that case. |
| 384 | */ |
| 385 | #define DEF_TX_COAL 400 /* 996 */ |
| 386 | #define DEF_TX_MAX_DESC 60 /* was 40 */ |
| 387 | #define DEF_RX_COAL 120 /* 1000 */ |
| 388 | #define DEF_RX_MAX_DESC 25 |
| 389 | #define DEF_TX_RATIO 21 /* 24 */ |
| 390 | |
| 391 | #define DEF_JUMBO_TX_COAL 20 |
| 392 | #define DEF_JUMBO_TX_MAX_DESC 60 |
| 393 | #define DEF_JUMBO_RX_COAL 30 |
| 394 | #define DEF_JUMBO_RX_MAX_DESC 6 |
| 395 | #define DEF_JUMBO_TX_RATIO 21 |
| 396 | |
| 397 | #if tigon2FwReleaseLocal < 20001118 |
| 398 | /* |
| 399 | * Standard firmware and early modifications duplicate |
| 400 | * IRQ load without this flag (coal timer is never reset). |
| 401 | * Note that with this flag tx_coal should be less than |
| 402 | * time to xmit full tx ring. |
| 403 | * 400usec is not so bad for tx ring size of 128. |
| 404 | */ |
| 405 | #define TX_COAL_INTS_ONLY 1 /* worth it */ |
| 406 | #else |
| 407 | /* |
| 408 | * With modified firmware, this is not necessary, but still useful. |
| 409 | */ |
| 410 | #define TX_COAL_INTS_ONLY 1 |
| 411 | #endif |
| 412 | |
| 413 | #define DEF_TRACE 0 |
| 414 | #define DEF_STAT (2 * TICKS_PER_SEC) |
| 415 | |
| 416 | |
| 417 | static int link[ACE_MAX_MOD_PARMS]; |
| 418 | static int trace[ACE_MAX_MOD_PARMS]; |
| 419 | static int tx_coal_tick[ACE_MAX_MOD_PARMS]; |
| 420 | static int rx_coal_tick[ACE_MAX_MOD_PARMS]; |
| 421 | static int max_tx_desc[ACE_MAX_MOD_PARMS]; |
| 422 | static int max_rx_desc[ACE_MAX_MOD_PARMS]; |
| 423 | static int tx_ratio[ACE_MAX_MOD_PARMS]; |
| 424 | static int dis_pci_mem_inval[ACE_MAX_MOD_PARMS] = {1, 1, 1, 1, 1, 1, 1, 1}; |
| 425 | |
| 426 | MODULE_AUTHOR("Jes Sorensen <jes@trained-monkey.org>"); |
| 427 | MODULE_LICENSE("GPL"); |
| 428 | MODULE_DESCRIPTION("AceNIC/3C985/GA620 Gigabit Ethernet driver"); |
| 429 | |
| 430 | module_param_array(link, int, NULL, 0); |
| 431 | module_param_array(trace, int, NULL, 0); |
| 432 | module_param_array(tx_coal_tick, int, NULL, 0); |
| 433 | module_param_array(max_tx_desc, int, NULL, 0); |
| 434 | module_param_array(rx_coal_tick, int, NULL, 0); |
| 435 | module_param_array(max_rx_desc, int, NULL, 0); |
| 436 | module_param_array(tx_ratio, int, NULL, 0); |
| 437 | MODULE_PARM_DESC(link, "AceNIC/3C985/NetGear link state"); |
| 438 | MODULE_PARM_DESC(trace, "AceNIC/3C985/NetGear firmware trace level"); |
| 439 | MODULE_PARM_DESC(tx_coal_tick, "AceNIC/3C985/GA620 max clock ticks to wait from first tx descriptor arrives"); |
| 440 | MODULE_PARM_DESC(max_tx_desc, "AceNIC/3C985/GA620 max number of transmit descriptors to wait"); |
| 441 | MODULE_PARM_DESC(rx_coal_tick, "AceNIC/3C985/GA620 max clock ticks to wait from first rx descriptor arrives"); |
| 442 | MODULE_PARM_DESC(max_rx_desc, "AceNIC/3C985/GA620 max number of receive descriptors to wait"); |
| 443 | MODULE_PARM_DESC(tx_ratio, "AceNIC/3C985/GA620 ratio of NIC memory used for TX/RX descriptors (range 0-63)"); |
| 444 | |
| 445 | |
| 446 | static char version[] __devinitdata = |
| 447 | "acenic.c: v0.92 08/05/2002 Jes Sorensen, linux-acenic@SunSITE.dk\n" |
| 448 | " http://home.cern.ch/~jes/gige/acenic.html\n"; |
| 449 | |
| 450 | static int ace_get_settings(struct net_device *, struct ethtool_cmd *); |
| 451 | static int ace_set_settings(struct net_device *, struct ethtool_cmd *); |
| 452 | static void ace_get_drvinfo(struct net_device *, struct ethtool_drvinfo *); |
| 453 | |
| 454 | static struct ethtool_ops ace_ethtool_ops = { |
| 455 | .get_settings = ace_get_settings, |
| 456 | .set_settings = ace_set_settings, |
| 457 | .get_drvinfo = ace_get_drvinfo, |
| 458 | }; |
| 459 | |
| 460 | static void ace_watchdog(struct net_device *dev); |
| 461 | |
| 462 | static int __devinit acenic_probe_one(struct pci_dev *pdev, |
| 463 | const struct pci_device_id *id) |
| 464 | { |
| 465 | struct net_device *dev; |
| 466 | struct ace_private *ap; |
| 467 | static int boards_found; |
| 468 | |
| 469 | dev = alloc_etherdev(sizeof(struct ace_private)); |
| 470 | if (dev == NULL) { |
| 471 | printk(KERN_ERR "acenic: Unable to allocate " |
| 472 | "net_device structure!\n"); |
| 473 | return -ENOMEM; |
| 474 | } |
| 475 | |
| 476 | SET_MODULE_OWNER(dev); |
| 477 | SET_NETDEV_DEV(dev, &pdev->dev); |
| 478 | |
| 479 | ap = dev->priv; |
| 480 | ap->pdev = pdev; |
| 481 | ap->name = pci_name(pdev); |
| 482 | |
| 483 | dev->features |= NETIF_F_SG | NETIF_F_IP_CSUM; |
| 484 | #if ACENIC_DO_VLAN |
| 485 | dev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX; |
| 486 | dev->vlan_rx_register = ace_vlan_rx_register; |
| 487 | dev->vlan_rx_kill_vid = ace_vlan_rx_kill_vid; |
| 488 | #endif |
| 489 | if (1) { |
| 490 | dev->tx_timeout = &ace_watchdog; |
| 491 | dev->watchdog_timeo = 5*HZ; |
| 492 | } |
| 493 | |
| 494 | dev->open = &ace_open; |
| 495 | dev->stop = &ace_close; |
| 496 | dev->hard_start_xmit = &ace_start_xmit; |
| 497 | dev->get_stats = &ace_get_stats; |
| 498 | dev->set_multicast_list = &ace_set_multicast_list; |
| 499 | SET_ETHTOOL_OPS(dev, &ace_ethtool_ops); |
| 500 | dev->set_mac_address = &ace_set_mac_addr; |
| 501 | dev->change_mtu = &ace_change_mtu; |
| 502 | |
| 503 | /* we only display this string ONCE */ |
| 504 | if (!boards_found) |
| 505 | printk(version); |
| 506 | |
| 507 | if (pci_enable_device(pdev)) |
| 508 | goto fail_free_netdev; |
| 509 | |
| 510 | /* |
| 511 | * Enable master mode before we start playing with the |
| 512 | * pci_command word since pci_set_master() will modify |
| 513 | * it. |
| 514 | */ |
| 515 | pci_set_master(pdev); |
| 516 | |
| 517 | pci_read_config_word(pdev, PCI_COMMAND, &ap->pci_command); |
| 518 | |
| 519 | /* OpenFirmware on Mac's does not set this - DOH.. */ |
| 520 | if (!(ap->pci_command & PCI_COMMAND_MEMORY)) { |
| 521 | printk(KERN_INFO "%s: Enabling PCI Memory Mapped " |
| 522 | "access - was not enabled by BIOS/Firmware\n", |
| 523 | ap->name); |
| 524 | ap->pci_command = ap->pci_command | PCI_COMMAND_MEMORY; |
| 525 | pci_write_config_word(ap->pdev, PCI_COMMAND, |
| 526 | ap->pci_command); |
| 527 | wmb(); |
| 528 | } |
| 529 | |
| 530 | pci_read_config_byte(pdev, PCI_LATENCY_TIMER, &ap->pci_latency); |
| 531 | if (ap->pci_latency <= 0x40) { |
| 532 | ap->pci_latency = 0x40; |
| 533 | pci_write_config_byte(pdev, PCI_LATENCY_TIMER, ap->pci_latency); |
| 534 | } |
| 535 | |
| 536 | /* |
| 537 | * Remap the regs into kernel space - this is abuse of |
| 538 | * dev->base_addr since it was means for I/O port |
| 539 | * addresses but who gives a damn. |
| 540 | */ |
| 541 | dev->base_addr = pci_resource_start(pdev, 0); |
| 542 | ap->regs = ioremap(dev->base_addr, 0x4000); |
| 543 | if (!ap->regs) { |
| 544 | printk(KERN_ERR "%s: Unable to map I/O register, " |
| 545 | "AceNIC %i will be disabled.\n", |
| 546 | ap->name, boards_found); |
| 547 | goto fail_free_netdev; |
| 548 | } |
| 549 | |
| 550 | switch(pdev->vendor) { |
| 551 | case PCI_VENDOR_ID_ALTEON: |
| 552 | if (pdev->device == PCI_DEVICE_ID_FARALLON_PN9100T) { |
| 553 | printk(KERN_INFO "%s: Farallon PN9100-T ", |
| 554 | ap->name); |
| 555 | } else { |
| 556 | printk(KERN_INFO "%s: Alteon AceNIC ", |
| 557 | ap->name); |
| 558 | } |
| 559 | break; |
| 560 | case PCI_VENDOR_ID_3COM: |
| 561 | printk(KERN_INFO "%s: 3Com 3C985 ", ap->name); |
| 562 | break; |
| 563 | case PCI_VENDOR_ID_NETGEAR: |
| 564 | printk(KERN_INFO "%s: NetGear GA620 ", ap->name); |
| 565 | break; |
| 566 | case PCI_VENDOR_ID_DEC: |
| 567 | if (pdev->device == PCI_DEVICE_ID_FARALLON_PN9000SX) { |
| 568 | printk(KERN_INFO "%s: Farallon PN9000-SX ", |
| 569 | ap->name); |
| 570 | break; |
| 571 | } |
| 572 | case PCI_VENDOR_ID_SGI: |
| 573 | printk(KERN_INFO "%s: SGI AceNIC ", ap->name); |
| 574 | break; |
| 575 | default: |
| 576 | printk(KERN_INFO "%s: Unknown AceNIC ", ap->name); |
| 577 | break; |
| 578 | } |
| 579 | |
| 580 | printk("Gigabit Ethernet at 0x%08lx, ", dev->base_addr); |
| 581 | #ifdef __sparc__ |
| 582 | printk("irq %s\n", __irq_itoa(pdev->irq)); |
| 583 | #else |
| 584 | printk("irq %i\n", pdev->irq); |
| 585 | #endif |
| 586 | |
| 587 | #ifdef CONFIG_ACENIC_OMIT_TIGON_I |
| 588 | if ((readl(&ap->regs->HostCtrl) >> 28) == 4) { |
| 589 | printk(KERN_ERR "%s: Driver compiled without Tigon I" |
| 590 | " support - NIC disabled\n", dev->name); |
| 591 | goto fail_uninit; |
| 592 | } |
| 593 | #endif |
| 594 | |
| 595 | if (ace_allocate_descriptors(dev)) |
| 596 | goto fail_free_netdev; |
| 597 | |
| 598 | #ifdef MODULE |
| 599 | if (boards_found >= ACE_MAX_MOD_PARMS) |
| 600 | ap->board_idx = BOARD_IDX_OVERFLOW; |
| 601 | else |
| 602 | ap->board_idx = boards_found; |
| 603 | #else |
| 604 | ap->board_idx = BOARD_IDX_STATIC; |
| 605 | #endif |
| 606 | |
| 607 | if (ace_init(dev)) |
| 608 | goto fail_free_netdev; |
| 609 | |
| 610 | if (register_netdev(dev)) { |
| 611 | printk(KERN_ERR "acenic: device registration failed\n"); |
| 612 | goto fail_uninit; |
| 613 | } |
| 614 | ap->name = dev->name; |
| 615 | |
| 616 | if (ap->pci_using_dac) |
| 617 | dev->features |= NETIF_F_HIGHDMA; |
| 618 | |
| 619 | pci_set_drvdata(pdev, dev); |
| 620 | |
| 621 | boards_found++; |
| 622 | return 0; |
| 623 | |
| 624 | fail_uninit: |
| 625 | ace_init_cleanup(dev); |
| 626 | fail_free_netdev: |
| 627 | free_netdev(dev); |
| 628 | return -ENODEV; |
| 629 | } |
| 630 | |
| 631 | static void __devexit acenic_remove_one(struct pci_dev *pdev) |
| 632 | { |
| 633 | struct net_device *dev = pci_get_drvdata(pdev); |
| 634 | struct ace_private *ap = netdev_priv(dev); |
| 635 | struct ace_regs __iomem *regs = ap->regs; |
| 636 | short i; |
| 637 | |
| 638 | unregister_netdev(dev); |
| 639 | |
| 640 | writel(readl(®s->CpuCtrl) | CPU_HALT, ®s->CpuCtrl); |
| 641 | if (ap->version >= 2) |
| 642 | writel(readl(®s->CpuBCtrl) | CPU_HALT, ®s->CpuBCtrl); |
| 643 | |
| 644 | /* |
| 645 | * This clears any pending interrupts |
| 646 | */ |
| 647 | writel(1, ®s->Mb0Lo); |
| 648 | readl(®s->CpuCtrl); /* flush */ |
| 649 | |
| 650 | /* |
| 651 | * Make sure no other CPUs are processing interrupts |
| 652 | * on the card before the buffers are being released. |
| 653 | * Otherwise one might experience some `interesting' |
| 654 | * effects. |
| 655 | * |
| 656 | * Then release the RX buffers - jumbo buffers were |
| 657 | * already released in ace_close(). |
| 658 | */ |
| 659 | ace_sync_irq(dev->irq); |
| 660 | |
| 661 | for (i = 0; i < RX_STD_RING_ENTRIES; i++) { |
| 662 | struct sk_buff *skb = ap->skb->rx_std_skbuff[i].skb; |
| 663 | |
| 664 | if (skb) { |
| 665 | struct ring_info *ringp; |
| 666 | dma_addr_t mapping; |
| 667 | |
| 668 | ringp = &ap->skb->rx_std_skbuff[i]; |
| 669 | mapping = pci_unmap_addr(ringp, mapping); |
| 670 | pci_unmap_page(ap->pdev, mapping, |
| 671 | ACE_STD_BUFSIZE, |
| 672 | PCI_DMA_FROMDEVICE); |
| 673 | |
| 674 | ap->rx_std_ring[i].size = 0; |
| 675 | ap->skb->rx_std_skbuff[i].skb = NULL; |
| 676 | dev_kfree_skb(skb); |
| 677 | } |
| 678 | } |
| 679 | |
| 680 | if (ap->version >= 2) { |
| 681 | for (i = 0; i < RX_MINI_RING_ENTRIES; i++) { |
| 682 | struct sk_buff *skb = ap->skb->rx_mini_skbuff[i].skb; |
| 683 | |
| 684 | if (skb) { |
| 685 | struct ring_info *ringp; |
| 686 | dma_addr_t mapping; |
| 687 | |
| 688 | ringp = &ap->skb->rx_mini_skbuff[i]; |
| 689 | mapping = pci_unmap_addr(ringp,mapping); |
| 690 | pci_unmap_page(ap->pdev, mapping, |
| 691 | ACE_MINI_BUFSIZE, |
| 692 | PCI_DMA_FROMDEVICE); |
| 693 | |
| 694 | ap->rx_mini_ring[i].size = 0; |
| 695 | ap->skb->rx_mini_skbuff[i].skb = NULL; |
| 696 | dev_kfree_skb(skb); |
| 697 | } |
| 698 | } |
| 699 | } |
| 700 | |
| 701 | for (i = 0; i < RX_JUMBO_RING_ENTRIES; i++) { |
| 702 | struct sk_buff *skb = ap->skb->rx_jumbo_skbuff[i].skb; |
| 703 | if (skb) { |
| 704 | struct ring_info *ringp; |
| 705 | dma_addr_t mapping; |
| 706 | |
| 707 | ringp = &ap->skb->rx_jumbo_skbuff[i]; |
| 708 | mapping = pci_unmap_addr(ringp, mapping); |
| 709 | pci_unmap_page(ap->pdev, mapping, |
| 710 | ACE_JUMBO_BUFSIZE, |
| 711 | PCI_DMA_FROMDEVICE); |
| 712 | |
| 713 | ap->rx_jumbo_ring[i].size = 0; |
| 714 | ap->skb->rx_jumbo_skbuff[i].skb = NULL; |
| 715 | dev_kfree_skb(skb); |
| 716 | } |
| 717 | } |
| 718 | |
| 719 | ace_init_cleanup(dev); |
| 720 | free_netdev(dev); |
| 721 | } |
| 722 | |
| 723 | static struct pci_driver acenic_pci_driver = { |
| 724 | .name = "acenic", |
| 725 | .id_table = acenic_pci_tbl, |
| 726 | .probe = acenic_probe_one, |
| 727 | .remove = __devexit_p(acenic_remove_one), |
| 728 | }; |
| 729 | |
| 730 | static int __init acenic_init(void) |
| 731 | { |
| 732 | return pci_module_init(&acenic_pci_driver); |
| 733 | } |
| 734 | |
| 735 | static void __exit acenic_exit(void) |
| 736 | { |
| 737 | pci_unregister_driver(&acenic_pci_driver); |
| 738 | } |
| 739 | |
| 740 | module_init(acenic_init); |
| 741 | module_exit(acenic_exit); |
| 742 | |
| 743 | static void ace_free_descriptors(struct net_device *dev) |
| 744 | { |
| 745 | struct ace_private *ap = netdev_priv(dev); |
| 746 | int size; |
| 747 | |
| 748 | if (ap->rx_std_ring != NULL) { |
| 749 | size = (sizeof(struct rx_desc) * |
| 750 | (RX_STD_RING_ENTRIES + |
| 751 | RX_JUMBO_RING_ENTRIES + |
| 752 | RX_MINI_RING_ENTRIES + |
| 753 | RX_RETURN_RING_ENTRIES)); |
| 754 | pci_free_consistent(ap->pdev, size, ap->rx_std_ring, |
| 755 | ap->rx_ring_base_dma); |
| 756 | ap->rx_std_ring = NULL; |
| 757 | ap->rx_jumbo_ring = NULL; |
| 758 | ap->rx_mini_ring = NULL; |
| 759 | ap->rx_return_ring = NULL; |
| 760 | } |
| 761 | if (ap->evt_ring != NULL) { |
| 762 | size = (sizeof(struct event) * EVT_RING_ENTRIES); |
| 763 | pci_free_consistent(ap->pdev, size, ap->evt_ring, |
| 764 | ap->evt_ring_dma); |
| 765 | ap->evt_ring = NULL; |
| 766 | } |
| 767 | if (ap->tx_ring != NULL && !ACE_IS_TIGON_I(ap)) { |
| 768 | size = (sizeof(struct tx_desc) * MAX_TX_RING_ENTRIES); |
| 769 | pci_free_consistent(ap->pdev, size, ap->tx_ring, |
| 770 | ap->tx_ring_dma); |
| 771 | } |
| 772 | ap->tx_ring = NULL; |
| 773 | |
| 774 | if (ap->evt_prd != NULL) { |
| 775 | pci_free_consistent(ap->pdev, sizeof(u32), |
| 776 | (void *)ap->evt_prd, ap->evt_prd_dma); |
| 777 | ap->evt_prd = NULL; |
| 778 | } |
| 779 | if (ap->rx_ret_prd != NULL) { |
| 780 | pci_free_consistent(ap->pdev, sizeof(u32), |
| 781 | (void *)ap->rx_ret_prd, |
| 782 | ap->rx_ret_prd_dma); |
| 783 | ap->rx_ret_prd = NULL; |
| 784 | } |
| 785 | if (ap->tx_csm != NULL) { |
| 786 | pci_free_consistent(ap->pdev, sizeof(u32), |
| 787 | (void *)ap->tx_csm, ap->tx_csm_dma); |
| 788 | ap->tx_csm = NULL; |
| 789 | } |
| 790 | } |
| 791 | |
| 792 | |
| 793 | static int ace_allocate_descriptors(struct net_device *dev) |
| 794 | { |
| 795 | struct ace_private *ap = netdev_priv(dev); |
| 796 | int size; |
| 797 | |
| 798 | size = (sizeof(struct rx_desc) * |
| 799 | (RX_STD_RING_ENTRIES + |
| 800 | RX_JUMBO_RING_ENTRIES + |
| 801 | RX_MINI_RING_ENTRIES + |
| 802 | RX_RETURN_RING_ENTRIES)); |
| 803 | |
| 804 | ap->rx_std_ring = pci_alloc_consistent(ap->pdev, size, |
| 805 | &ap->rx_ring_base_dma); |
| 806 | if (ap->rx_std_ring == NULL) |
| 807 | goto fail; |
| 808 | |
| 809 | ap->rx_jumbo_ring = ap->rx_std_ring + RX_STD_RING_ENTRIES; |
| 810 | ap->rx_mini_ring = ap->rx_jumbo_ring + RX_JUMBO_RING_ENTRIES; |
| 811 | ap->rx_return_ring = ap->rx_mini_ring + RX_MINI_RING_ENTRIES; |
| 812 | |
| 813 | size = (sizeof(struct event) * EVT_RING_ENTRIES); |
| 814 | |
| 815 | ap->evt_ring = pci_alloc_consistent(ap->pdev, size, &ap->evt_ring_dma); |
| 816 | |
| 817 | if (ap->evt_ring == NULL) |
| 818 | goto fail; |
| 819 | |
| 820 | /* |
| 821 | * Only allocate a host TX ring for the Tigon II, the Tigon I |
| 822 | * has to use PCI registers for this ;-( |
| 823 | */ |
| 824 | if (!ACE_IS_TIGON_I(ap)) { |
| 825 | size = (sizeof(struct tx_desc) * MAX_TX_RING_ENTRIES); |
| 826 | |
| 827 | ap->tx_ring = pci_alloc_consistent(ap->pdev, size, |
| 828 | &ap->tx_ring_dma); |
| 829 | |
| 830 | if (ap->tx_ring == NULL) |
| 831 | goto fail; |
| 832 | } |
| 833 | |
| 834 | ap->evt_prd = pci_alloc_consistent(ap->pdev, sizeof(u32), |
| 835 | &ap->evt_prd_dma); |
| 836 | if (ap->evt_prd == NULL) |
| 837 | goto fail; |
| 838 | |
| 839 | ap->rx_ret_prd = pci_alloc_consistent(ap->pdev, sizeof(u32), |
| 840 | &ap->rx_ret_prd_dma); |
| 841 | if (ap->rx_ret_prd == NULL) |
| 842 | goto fail; |
| 843 | |
| 844 | ap->tx_csm = pci_alloc_consistent(ap->pdev, sizeof(u32), |
| 845 | &ap->tx_csm_dma); |
| 846 | if (ap->tx_csm == NULL) |
| 847 | goto fail; |
| 848 | |
| 849 | return 0; |
| 850 | |
| 851 | fail: |
| 852 | /* Clean up. */ |
| 853 | ace_init_cleanup(dev); |
| 854 | return 1; |
| 855 | } |
| 856 | |
| 857 | |
| 858 | /* |
| 859 | * Generic cleanup handling data allocated during init. Used when the |
| 860 | * module is unloaded or if an error occurs during initialization |
| 861 | */ |
| 862 | static void ace_init_cleanup(struct net_device *dev) |
| 863 | { |
| 864 | struct ace_private *ap; |
| 865 | |
| 866 | ap = netdev_priv(dev); |
| 867 | |
| 868 | ace_free_descriptors(dev); |
| 869 | |
| 870 | if (ap->info) |
| 871 | pci_free_consistent(ap->pdev, sizeof(struct ace_info), |
| 872 | ap->info, ap->info_dma); |
| 873 | if (ap->skb) |
| 874 | kfree(ap->skb); |
| 875 | if (ap->trace_buf) |
| 876 | kfree(ap->trace_buf); |
| 877 | |
| 878 | if (dev->irq) |
| 879 | free_irq(dev->irq, dev); |
| 880 | |
| 881 | iounmap(ap->regs); |
| 882 | } |
| 883 | |
| 884 | |
| 885 | /* |
| 886 | * Commands are considered to be slow. |
| 887 | */ |
| 888 | static inline void ace_issue_cmd(struct ace_regs __iomem *regs, struct cmd *cmd) |
| 889 | { |
| 890 | u32 idx; |
| 891 | |
| 892 | idx = readl(®s->CmdPrd); |
| 893 | |
| 894 | writel(*(u32 *)(cmd), ®s->CmdRng[idx]); |
| 895 | idx = (idx + 1) % CMD_RING_ENTRIES; |
| 896 | |
| 897 | writel(idx, ®s->CmdPrd); |
| 898 | } |
| 899 | |
| 900 | |
| 901 | static int __devinit ace_init(struct net_device *dev) |
| 902 | { |
| 903 | struct ace_private *ap; |
| 904 | struct ace_regs __iomem *regs; |
| 905 | struct ace_info *info = NULL; |
| 906 | struct pci_dev *pdev; |
| 907 | unsigned long myjif; |
| 908 | u64 tmp_ptr; |
| 909 | u32 tig_ver, mac1, mac2, tmp, pci_state; |
| 910 | int board_idx, ecode = 0; |
| 911 | short i; |
| 912 | unsigned char cache_size; |
| 913 | |
| 914 | ap = netdev_priv(dev); |
| 915 | regs = ap->regs; |
| 916 | |
| 917 | board_idx = ap->board_idx; |
| 918 | |
| 919 | /* |
| 920 | * aman@sgi.com - its useful to do a NIC reset here to |
| 921 | * address the `Firmware not running' problem subsequent |
| 922 | * to any crashes involving the NIC |
| 923 | */ |
| 924 | writel(HW_RESET | (HW_RESET << 24), ®s->HostCtrl); |
| 925 | readl(®s->HostCtrl); /* PCI write posting */ |
| 926 | udelay(5); |
| 927 | |
| 928 | /* |
| 929 | * Don't access any other registers before this point! |
| 930 | */ |
| 931 | #ifdef __BIG_ENDIAN |
| 932 | /* |
| 933 | * This will most likely need BYTE_SWAP once we switch |
| 934 | * to using __raw_writel() |
| 935 | */ |
| 936 | writel((WORD_SWAP | CLR_INT | ((WORD_SWAP | CLR_INT) << 24)), |
| 937 | ®s->HostCtrl); |
| 938 | #else |
| 939 | writel((CLR_INT | WORD_SWAP | ((CLR_INT | WORD_SWAP) << 24)), |
| 940 | ®s->HostCtrl); |
| 941 | #endif |
| 942 | readl(®s->HostCtrl); /* PCI write posting */ |
| 943 | |
| 944 | /* |
| 945 | * Stop the NIC CPU and clear pending interrupts |
| 946 | */ |
| 947 | writel(readl(®s->CpuCtrl) | CPU_HALT, ®s->CpuCtrl); |
| 948 | readl(®s->CpuCtrl); /* PCI write posting */ |
| 949 | writel(0, ®s->Mb0Lo); |
| 950 | |
| 951 | tig_ver = readl(®s->HostCtrl) >> 28; |
| 952 | |
| 953 | switch(tig_ver){ |
| 954 | #ifndef CONFIG_ACENIC_OMIT_TIGON_I |
| 955 | case 4: |
| 956 | case 5: |
| 957 | printk(KERN_INFO " Tigon I (Rev. %i), Firmware: %i.%i.%i, ", |
| 958 | tig_ver, tigonFwReleaseMajor, tigonFwReleaseMinor, |
| 959 | tigonFwReleaseFix); |
| 960 | writel(0, ®s->LocalCtrl); |
| 961 | ap->version = 1; |
| 962 | ap->tx_ring_entries = TIGON_I_TX_RING_ENTRIES; |
| 963 | break; |
| 964 | #endif |
| 965 | case 6: |
| 966 | printk(KERN_INFO " Tigon II (Rev. %i), Firmware: %i.%i.%i, ", |
| 967 | tig_ver, tigon2FwReleaseMajor, tigon2FwReleaseMinor, |
| 968 | tigon2FwReleaseFix); |
| 969 | writel(readl(®s->CpuBCtrl) | CPU_HALT, ®s->CpuBCtrl); |
| 970 | readl(®s->CpuBCtrl); /* PCI write posting */ |
| 971 | /* |
| 972 | * The SRAM bank size does _not_ indicate the amount |
| 973 | * of memory on the card, it controls the _bank_ size! |
| 974 | * Ie. a 1MB AceNIC will have two banks of 512KB. |
| 975 | */ |
| 976 | writel(SRAM_BANK_512K, ®s->LocalCtrl); |
| 977 | writel(SYNC_SRAM_TIMING, ®s->MiscCfg); |
| 978 | ap->version = 2; |
| 979 | ap->tx_ring_entries = MAX_TX_RING_ENTRIES; |
| 980 | break; |
| 981 | default: |
| 982 | printk(KERN_WARNING " Unsupported Tigon version detected " |
| 983 | "(%i)\n", tig_ver); |
| 984 | ecode = -ENODEV; |
| 985 | goto init_error; |
| 986 | } |
| 987 | |
| 988 | /* |
| 989 | * ModeStat _must_ be set after the SRAM settings as this change |
| 990 | * seems to corrupt the ModeStat and possible other registers. |
| 991 | * The SRAM settings survive resets and setting it to the same |
| 992 | * value a second time works as well. This is what caused the |
| 993 | * `Firmware not running' problem on the Tigon II. |
| 994 | */ |
| 995 | #ifdef __BIG_ENDIAN |
| 996 | writel(ACE_BYTE_SWAP_DMA | ACE_WARN | ACE_FATAL | ACE_BYTE_SWAP_BD | |
| 997 | ACE_WORD_SWAP_BD | ACE_NO_JUMBO_FRAG, ®s->ModeStat); |
| 998 | #else |
| 999 | writel(ACE_BYTE_SWAP_DMA | ACE_WARN | ACE_FATAL | |
| 1000 | ACE_WORD_SWAP_BD | ACE_NO_JUMBO_FRAG, ®s->ModeStat); |
| 1001 | #endif |
| 1002 | readl(®s->ModeStat); /* PCI write posting */ |
| 1003 | |
| 1004 | mac1 = 0; |
| 1005 | for(i = 0; i < 4; i++) { |
| 1006 | mac1 = mac1 << 8; |
| 1007 | tmp = read_eeprom_byte(dev, 0x8c+i); |
| 1008 | if (tmp < 0) { |
| 1009 | ecode = -EIO; |
| 1010 | goto init_error; |
| 1011 | } else |
| 1012 | mac1 |= (tmp & 0xff); |
| 1013 | } |
| 1014 | mac2 = 0; |
| 1015 | for(i = 4; i < 8; i++) { |
| 1016 | mac2 = mac2 << 8; |
| 1017 | tmp = read_eeprom_byte(dev, 0x8c+i); |
| 1018 | if (tmp < 0) { |
| 1019 | ecode = -EIO; |
| 1020 | goto init_error; |
| 1021 | } else |
| 1022 | mac2 |= (tmp & 0xff); |
| 1023 | } |
| 1024 | |
| 1025 | writel(mac1, ®s->MacAddrHi); |
| 1026 | writel(mac2, ®s->MacAddrLo); |
| 1027 | |
| 1028 | printk("MAC: %02x:%02x:%02x:%02x:%02x:%02x\n", |
| 1029 | (mac1 >> 8) & 0xff, mac1 & 0xff, (mac2 >> 24) &0xff, |
| 1030 | (mac2 >> 16) & 0xff, (mac2 >> 8) & 0xff, mac2 & 0xff); |
| 1031 | |
| 1032 | dev->dev_addr[0] = (mac1 >> 8) & 0xff; |
| 1033 | dev->dev_addr[1] = mac1 & 0xff; |
| 1034 | dev->dev_addr[2] = (mac2 >> 24) & 0xff; |
| 1035 | dev->dev_addr[3] = (mac2 >> 16) & 0xff; |
| 1036 | dev->dev_addr[4] = (mac2 >> 8) & 0xff; |
| 1037 | dev->dev_addr[5] = mac2 & 0xff; |
| 1038 | |
| 1039 | /* |
| 1040 | * Looks like this is necessary to deal with on all architectures, |
| 1041 | * even this %$#%$# N440BX Intel based thing doesn't get it right. |
| 1042 | * Ie. having two NICs in the machine, one will have the cache |
| 1043 | * line set at boot time, the other will not. |
| 1044 | */ |
| 1045 | pdev = ap->pdev; |
| 1046 | pci_read_config_byte(pdev, PCI_CACHE_LINE_SIZE, &cache_size); |
| 1047 | cache_size <<= 2; |
| 1048 | if (cache_size != SMP_CACHE_BYTES) { |
| 1049 | printk(KERN_INFO " PCI cache line size set incorrectly " |
| 1050 | "(%i bytes) by BIOS/FW, ", cache_size); |
| 1051 | if (cache_size > SMP_CACHE_BYTES) |
| 1052 | printk("expecting %i\n", SMP_CACHE_BYTES); |
| 1053 | else { |
| 1054 | printk("correcting to %i\n", SMP_CACHE_BYTES); |
| 1055 | pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, |
| 1056 | SMP_CACHE_BYTES >> 2); |
| 1057 | } |
| 1058 | } |
| 1059 | |
| 1060 | pci_state = readl(®s->PciState); |
| 1061 | printk(KERN_INFO " PCI bus width: %i bits, speed: %iMHz, " |
| 1062 | "latency: %i clks\n", |
| 1063 | (pci_state & PCI_32BIT) ? 32 : 64, |
| 1064 | (pci_state & PCI_66MHZ) ? 66 : 33, |
| 1065 | ap->pci_latency); |
| 1066 | |
| 1067 | /* |
| 1068 | * Set the max DMA transfer size. Seems that for most systems |
| 1069 | * the performance is better when no MAX parameter is |
| 1070 | * set. However for systems enabling PCI write and invalidate, |
| 1071 | * DMA writes must be set to the L1 cache line size to get |
| 1072 | * optimal performance. |
| 1073 | * |
| 1074 | * The default is now to turn the PCI write and invalidate off |
| 1075 | * - that is what Alteon does for NT. |
| 1076 | */ |
| 1077 | tmp = READ_CMD_MEM | WRITE_CMD_MEM; |
| 1078 | if (ap->version >= 2) { |
| 1079 | tmp |= (MEM_READ_MULTIPLE | (pci_state & PCI_66MHZ)); |
| 1080 | /* |
| 1081 | * Tuning parameters only supported for 8 cards |
| 1082 | */ |
| 1083 | if (board_idx == BOARD_IDX_OVERFLOW || |
| 1084 | dis_pci_mem_inval[board_idx]) { |
| 1085 | if (ap->pci_command & PCI_COMMAND_INVALIDATE) { |
| 1086 | ap->pci_command &= ~PCI_COMMAND_INVALIDATE; |
| 1087 | pci_write_config_word(pdev, PCI_COMMAND, |
| 1088 | ap->pci_command); |
| 1089 | printk(KERN_INFO " Disabling PCI memory " |
| 1090 | "write and invalidate\n"); |
| 1091 | } |
| 1092 | } else if (ap->pci_command & PCI_COMMAND_INVALIDATE) { |
| 1093 | printk(KERN_INFO " PCI memory write & invalidate " |
| 1094 | "enabled by BIOS, enabling counter measures\n"); |
| 1095 | |
| 1096 | switch(SMP_CACHE_BYTES) { |
| 1097 | case 16: |
| 1098 | tmp |= DMA_WRITE_MAX_16; |
| 1099 | break; |
| 1100 | case 32: |
| 1101 | tmp |= DMA_WRITE_MAX_32; |
| 1102 | break; |
| 1103 | case 64: |
| 1104 | tmp |= DMA_WRITE_MAX_64; |
| 1105 | break; |
| 1106 | case 128: |
| 1107 | tmp |= DMA_WRITE_MAX_128; |
| 1108 | break; |
| 1109 | default: |
| 1110 | printk(KERN_INFO " Cache line size %i not " |
| 1111 | "supported, PCI write and invalidate " |
| 1112 | "disabled\n", SMP_CACHE_BYTES); |
| 1113 | ap->pci_command &= ~PCI_COMMAND_INVALIDATE; |
| 1114 | pci_write_config_word(pdev, PCI_COMMAND, |
| 1115 | ap->pci_command); |
| 1116 | } |
| 1117 | } |
| 1118 | } |
| 1119 | |
| 1120 | #ifdef __sparc__ |
| 1121 | /* |
| 1122 | * On this platform, we know what the best dma settings |
| 1123 | * are. We use 64-byte maximum bursts, because if we |
| 1124 | * burst larger than the cache line size (or even cross |
| 1125 | * a 64byte boundary in a single burst) the UltraSparc |
| 1126 | * PCI controller will disconnect at 64-byte multiples. |
| 1127 | * |
| 1128 | * Read-multiple will be properly enabled above, and when |
| 1129 | * set will give the PCI controller proper hints about |
| 1130 | * prefetching. |
| 1131 | */ |
| 1132 | tmp &= ~DMA_READ_WRITE_MASK; |
| 1133 | tmp |= DMA_READ_MAX_64; |
| 1134 | tmp |= DMA_WRITE_MAX_64; |
| 1135 | #endif |
| 1136 | #ifdef __alpha__ |
| 1137 | tmp &= ~DMA_READ_WRITE_MASK; |
| 1138 | tmp |= DMA_READ_MAX_128; |
| 1139 | /* |
| 1140 | * All the docs say MUST NOT. Well, I did. |
| 1141 | * Nothing terrible happens, if we load wrong size. |
| 1142 | * Bit w&i still works better! |
| 1143 | */ |
| 1144 | tmp |= DMA_WRITE_MAX_128; |
| 1145 | #endif |
| 1146 | writel(tmp, ®s->PciState); |
| 1147 | |
| 1148 | #if 0 |
| 1149 | /* |
| 1150 | * The Host PCI bus controller driver has to set FBB. |
| 1151 | * If all devices on that PCI bus support FBB, then the controller |
| 1152 | * can enable FBB support in the Host PCI Bus controller (or on |
| 1153 | * the PCI-PCI bridge if that applies). |
| 1154 | * -ggg |
| 1155 | */ |
| 1156 | /* |
| 1157 | * I have received reports from people having problems when this |
| 1158 | * bit is enabled. |
| 1159 | */ |
| 1160 | if (!(ap->pci_command & PCI_COMMAND_FAST_BACK)) { |
| 1161 | printk(KERN_INFO " Enabling PCI Fast Back to Back\n"); |
| 1162 | ap->pci_command |= PCI_COMMAND_FAST_BACK; |
| 1163 | pci_write_config_word(pdev, PCI_COMMAND, ap->pci_command); |
| 1164 | } |
| 1165 | #endif |
| 1166 | |
| 1167 | /* |
| 1168 | * Configure DMA attributes. |
| 1169 | */ |
| 1170 | if (!pci_set_dma_mask(pdev, 0xffffffffffffffffULL)) { |
| 1171 | ap->pci_using_dac = 1; |
| 1172 | } else if (!pci_set_dma_mask(pdev, 0xffffffffULL)) { |
| 1173 | ap->pci_using_dac = 0; |
| 1174 | } else { |
| 1175 | ecode = -ENODEV; |
| 1176 | goto init_error; |
| 1177 | } |
| 1178 | |
| 1179 | /* |
| 1180 | * Initialize the generic info block and the command+event rings |
| 1181 | * and the control blocks for the transmit and receive rings |
| 1182 | * as they need to be setup once and for all. |
| 1183 | */ |
| 1184 | if (!(info = pci_alloc_consistent(ap->pdev, sizeof(struct ace_info), |
| 1185 | &ap->info_dma))) { |
| 1186 | ecode = -EAGAIN; |
| 1187 | goto init_error; |
| 1188 | } |
| 1189 | ap->info = info; |
| 1190 | |
| 1191 | /* |
| 1192 | * Get the memory for the skb rings. |
| 1193 | */ |
| 1194 | if (!(ap->skb = kmalloc(sizeof(struct ace_skb), GFP_KERNEL))) { |
| 1195 | ecode = -EAGAIN; |
| 1196 | goto init_error; |
| 1197 | } |
| 1198 | |
| 1199 | ecode = request_irq(pdev->irq, ace_interrupt, SA_SHIRQ, |
| 1200 | DRV_NAME, dev); |
| 1201 | if (ecode) { |
| 1202 | printk(KERN_WARNING "%s: Requested IRQ %d is busy\n", |
| 1203 | DRV_NAME, pdev->irq); |
| 1204 | goto init_error; |
| 1205 | } else |
| 1206 | dev->irq = pdev->irq; |
| 1207 | |
| 1208 | #ifdef INDEX_DEBUG |
| 1209 | spin_lock_init(&ap->debug_lock); |
| 1210 | ap->last_tx = ACE_TX_RING_ENTRIES(ap) - 1; |
| 1211 | ap->last_std_rx = 0; |
| 1212 | ap->last_mini_rx = 0; |
| 1213 | #endif |
| 1214 | |
| 1215 | memset(ap->info, 0, sizeof(struct ace_info)); |
| 1216 | memset(ap->skb, 0, sizeof(struct ace_skb)); |
| 1217 | |
| 1218 | ace_load_firmware(dev); |
| 1219 | ap->fw_running = 0; |
| 1220 | |
| 1221 | tmp_ptr = ap->info_dma; |
| 1222 | writel(tmp_ptr >> 32, ®s->InfoPtrHi); |
| 1223 | writel(tmp_ptr & 0xffffffff, ®s->InfoPtrLo); |
| 1224 | |
| 1225 | memset(ap->evt_ring, 0, EVT_RING_ENTRIES * sizeof(struct event)); |
| 1226 | |
| 1227 | set_aceaddr(&info->evt_ctrl.rngptr, ap->evt_ring_dma); |
| 1228 | info->evt_ctrl.flags = 0; |
| 1229 | |
| 1230 | *(ap->evt_prd) = 0; |
| 1231 | wmb(); |
| 1232 | set_aceaddr(&info->evt_prd_ptr, ap->evt_prd_dma); |
| 1233 | writel(0, ®s->EvtCsm); |
| 1234 | |
| 1235 | set_aceaddr(&info->cmd_ctrl.rngptr, 0x100); |
| 1236 | info->cmd_ctrl.flags = 0; |
| 1237 | info->cmd_ctrl.max_len = 0; |
| 1238 | |
| 1239 | for (i = 0; i < CMD_RING_ENTRIES; i++) |
| 1240 | writel(0, ®s->CmdRng[i]); |
| 1241 | |
| 1242 | writel(0, ®s->CmdPrd); |
| 1243 | writel(0, ®s->CmdCsm); |
| 1244 | |
| 1245 | tmp_ptr = ap->info_dma; |
| 1246 | tmp_ptr += (unsigned long) &(((struct ace_info *)0)->s.stats); |
| 1247 | set_aceaddr(&info->stats2_ptr, (dma_addr_t) tmp_ptr); |
| 1248 | |
| 1249 | set_aceaddr(&info->rx_std_ctrl.rngptr, ap->rx_ring_base_dma); |
| 1250 | info->rx_std_ctrl.max_len = ACE_STD_BUFSIZE; |
| 1251 | info->rx_std_ctrl.flags = |
| 1252 | RCB_FLG_TCP_UDP_SUM | RCB_FLG_NO_PSEUDO_HDR | ACE_RCB_VLAN_FLAG; |
| 1253 | |
| 1254 | memset(ap->rx_std_ring, 0, |
| 1255 | RX_STD_RING_ENTRIES * sizeof(struct rx_desc)); |
| 1256 | |
| 1257 | for (i = 0; i < RX_STD_RING_ENTRIES; i++) |
| 1258 | ap->rx_std_ring[i].flags = BD_FLG_TCP_UDP_SUM; |
| 1259 | |
| 1260 | ap->rx_std_skbprd = 0; |
| 1261 | atomic_set(&ap->cur_rx_bufs, 0); |
| 1262 | |
| 1263 | set_aceaddr(&info->rx_jumbo_ctrl.rngptr, |
| 1264 | (ap->rx_ring_base_dma + |
| 1265 | (sizeof(struct rx_desc) * RX_STD_RING_ENTRIES))); |
| 1266 | info->rx_jumbo_ctrl.max_len = 0; |
| 1267 | info->rx_jumbo_ctrl.flags = |
| 1268 | RCB_FLG_TCP_UDP_SUM | RCB_FLG_NO_PSEUDO_HDR | ACE_RCB_VLAN_FLAG; |
| 1269 | |
| 1270 | memset(ap->rx_jumbo_ring, 0, |
| 1271 | RX_JUMBO_RING_ENTRIES * sizeof(struct rx_desc)); |
| 1272 | |
| 1273 | for (i = 0; i < RX_JUMBO_RING_ENTRIES; i++) |
| 1274 | ap->rx_jumbo_ring[i].flags = BD_FLG_TCP_UDP_SUM | BD_FLG_JUMBO; |
| 1275 | |
| 1276 | ap->rx_jumbo_skbprd = 0; |
| 1277 | atomic_set(&ap->cur_jumbo_bufs, 0); |
| 1278 | |
| 1279 | memset(ap->rx_mini_ring, 0, |
| 1280 | RX_MINI_RING_ENTRIES * sizeof(struct rx_desc)); |
| 1281 | |
| 1282 | if (ap->version >= 2) { |
| 1283 | set_aceaddr(&info->rx_mini_ctrl.rngptr, |
| 1284 | (ap->rx_ring_base_dma + |
| 1285 | (sizeof(struct rx_desc) * |
| 1286 | (RX_STD_RING_ENTRIES + |
| 1287 | RX_JUMBO_RING_ENTRIES)))); |
| 1288 | info->rx_mini_ctrl.max_len = ACE_MINI_SIZE; |
| 1289 | info->rx_mini_ctrl.flags = |
| 1290 | RCB_FLG_TCP_UDP_SUM|RCB_FLG_NO_PSEUDO_HDR|ACE_RCB_VLAN_FLAG; |
| 1291 | |
| 1292 | for (i = 0; i < RX_MINI_RING_ENTRIES; i++) |
| 1293 | ap->rx_mini_ring[i].flags = |
| 1294 | BD_FLG_TCP_UDP_SUM | BD_FLG_MINI; |
| 1295 | } else { |
| 1296 | set_aceaddr(&info->rx_mini_ctrl.rngptr, 0); |
| 1297 | info->rx_mini_ctrl.flags = RCB_FLG_RNG_DISABLE; |
| 1298 | info->rx_mini_ctrl.max_len = 0; |
| 1299 | } |
| 1300 | |
| 1301 | ap->rx_mini_skbprd = 0; |
| 1302 | atomic_set(&ap->cur_mini_bufs, 0); |
| 1303 | |
| 1304 | set_aceaddr(&info->rx_return_ctrl.rngptr, |
| 1305 | (ap->rx_ring_base_dma + |
| 1306 | (sizeof(struct rx_desc) * |
| 1307 | (RX_STD_RING_ENTRIES + |
| 1308 | RX_JUMBO_RING_ENTRIES + |
| 1309 | RX_MINI_RING_ENTRIES)))); |
| 1310 | info->rx_return_ctrl.flags = 0; |
| 1311 | info->rx_return_ctrl.max_len = RX_RETURN_RING_ENTRIES; |
| 1312 | |
| 1313 | memset(ap->rx_return_ring, 0, |
| 1314 | RX_RETURN_RING_ENTRIES * sizeof(struct rx_desc)); |
| 1315 | |
| 1316 | set_aceaddr(&info->rx_ret_prd_ptr, ap->rx_ret_prd_dma); |
| 1317 | *(ap->rx_ret_prd) = 0; |
| 1318 | |
| 1319 | writel(TX_RING_BASE, ®s->WinBase); |
| 1320 | |
| 1321 | if (ACE_IS_TIGON_I(ap)) { |
| 1322 | ap->tx_ring = (struct tx_desc *) regs->Window; |
| 1323 | for (i = 0; i < (TIGON_I_TX_RING_ENTRIES |
| 1324 | * sizeof(struct tx_desc)) / sizeof(u32); i++) |
| 1325 | writel(0, (void __iomem *)ap->tx_ring + i * 4); |
| 1326 | |
| 1327 | set_aceaddr(&info->tx_ctrl.rngptr, TX_RING_BASE); |
| 1328 | } else { |
| 1329 | memset(ap->tx_ring, 0, |
| 1330 | MAX_TX_RING_ENTRIES * sizeof(struct tx_desc)); |
| 1331 | |
| 1332 | set_aceaddr(&info->tx_ctrl.rngptr, ap->tx_ring_dma); |
| 1333 | } |
| 1334 | |
| 1335 | info->tx_ctrl.max_len = ACE_TX_RING_ENTRIES(ap); |
| 1336 | tmp = RCB_FLG_TCP_UDP_SUM | RCB_FLG_NO_PSEUDO_HDR | ACE_RCB_VLAN_FLAG; |
| 1337 | |
| 1338 | /* |
| 1339 | * The Tigon I does not like having the TX ring in host memory ;-( |
| 1340 | */ |
| 1341 | if (!ACE_IS_TIGON_I(ap)) |
| 1342 | tmp |= RCB_FLG_TX_HOST_RING; |
| 1343 | #if TX_COAL_INTS_ONLY |
| 1344 | tmp |= RCB_FLG_COAL_INT_ONLY; |
| 1345 | #endif |
| 1346 | info->tx_ctrl.flags = tmp; |
| 1347 | |
| 1348 | set_aceaddr(&info->tx_csm_ptr, ap->tx_csm_dma); |
| 1349 | |
| 1350 | /* |
| 1351 | * Potential item for tuning parameter |
| 1352 | */ |
| 1353 | #if 0 /* NO */ |
| 1354 | writel(DMA_THRESH_16W, ®s->DmaReadCfg); |
| 1355 | writel(DMA_THRESH_16W, ®s->DmaWriteCfg); |
| 1356 | #else |
| 1357 | writel(DMA_THRESH_8W, ®s->DmaReadCfg); |
| 1358 | writel(DMA_THRESH_8W, ®s->DmaWriteCfg); |
| 1359 | #endif |
| 1360 | |
| 1361 | writel(0, ®s->MaskInt); |
| 1362 | writel(1, ®s->IfIdx); |
| 1363 | #if 0 |
| 1364 | /* |
| 1365 | * McKinley boxes do not like us fiddling with AssistState |
| 1366 | * this early |
| 1367 | */ |
| 1368 | writel(1, ®s->AssistState); |
| 1369 | #endif |
| 1370 | |
| 1371 | writel(DEF_STAT, ®s->TuneStatTicks); |
| 1372 | writel(DEF_TRACE, ®s->TuneTrace); |
| 1373 | |
| 1374 | ace_set_rxtx_parms(dev, 0); |
| 1375 | |
| 1376 | if (board_idx == BOARD_IDX_OVERFLOW) { |
| 1377 | printk(KERN_WARNING "%s: more than %i NICs detected, " |
| 1378 | "ignoring module parameters!\n", |
| 1379 | ap->name, ACE_MAX_MOD_PARMS); |
| 1380 | } else if (board_idx >= 0) { |
| 1381 | if (tx_coal_tick[board_idx]) |
| 1382 | writel(tx_coal_tick[board_idx], |
| 1383 | ®s->TuneTxCoalTicks); |
| 1384 | if (max_tx_desc[board_idx]) |
| 1385 | writel(max_tx_desc[board_idx], ®s->TuneMaxTxDesc); |
| 1386 | |
| 1387 | if (rx_coal_tick[board_idx]) |
| 1388 | writel(rx_coal_tick[board_idx], |
| 1389 | ®s->TuneRxCoalTicks); |
| 1390 | if (max_rx_desc[board_idx]) |
| 1391 | writel(max_rx_desc[board_idx], ®s->TuneMaxRxDesc); |
| 1392 | |
| 1393 | if (trace[board_idx]) |
| 1394 | writel(trace[board_idx], ®s->TuneTrace); |
| 1395 | |
| 1396 | if ((tx_ratio[board_idx] > 0) && (tx_ratio[board_idx] < 64)) |
| 1397 | writel(tx_ratio[board_idx], ®s->TxBufRat); |
| 1398 | } |
| 1399 | |
| 1400 | /* |
| 1401 | * Default link parameters |
| 1402 | */ |
| 1403 | tmp = LNK_ENABLE | LNK_FULL_DUPLEX | LNK_1000MB | LNK_100MB | |
| 1404 | LNK_10MB | LNK_RX_FLOW_CTL_Y | LNK_NEG_FCTL | LNK_NEGOTIATE; |
| 1405 | if(ap->version >= 2) |
| 1406 | tmp |= LNK_TX_FLOW_CTL_Y; |
| 1407 | |
| 1408 | /* |
| 1409 | * Override link default parameters |
| 1410 | */ |
| 1411 | if ((board_idx >= 0) && link[board_idx]) { |
| 1412 | int option = link[board_idx]; |
| 1413 | |
| 1414 | tmp = LNK_ENABLE; |
| 1415 | |
| 1416 | if (option & 0x01) { |
| 1417 | printk(KERN_INFO "%s: Setting half duplex link\n", |
| 1418 | ap->name); |
| 1419 | tmp &= ~LNK_FULL_DUPLEX; |
| 1420 | } |
| 1421 | if (option & 0x02) |
| 1422 | tmp &= ~LNK_NEGOTIATE; |
| 1423 | if (option & 0x10) |
| 1424 | tmp |= LNK_10MB; |
| 1425 | if (option & 0x20) |
| 1426 | tmp |= LNK_100MB; |
| 1427 | if (option & 0x40) |
| 1428 | tmp |= LNK_1000MB; |
| 1429 | if ((option & 0x70) == 0) { |
| 1430 | printk(KERN_WARNING "%s: No media speed specified, " |
| 1431 | "forcing auto negotiation\n", ap->name); |
| 1432 | tmp |= LNK_NEGOTIATE | LNK_1000MB | |
| 1433 | LNK_100MB | LNK_10MB; |
| 1434 | } |
| 1435 | if ((option & 0x100) == 0) |
| 1436 | tmp |= LNK_NEG_FCTL; |
| 1437 | else |
| 1438 | printk(KERN_INFO "%s: Disabling flow control " |
| 1439 | "negotiation\n", ap->name); |
| 1440 | if (option & 0x200) |
| 1441 | tmp |= LNK_RX_FLOW_CTL_Y; |
| 1442 | if ((option & 0x400) && (ap->version >= 2)) { |
| 1443 | printk(KERN_INFO "%s: Enabling TX flow control\n", |
| 1444 | ap->name); |
| 1445 | tmp |= LNK_TX_FLOW_CTL_Y; |
| 1446 | } |
| 1447 | } |
| 1448 | |
| 1449 | ap->link = tmp; |
| 1450 | writel(tmp, ®s->TuneLink); |
| 1451 | if (ap->version >= 2) |
| 1452 | writel(tmp, ®s->TuneFastLink); |
| 1453 | |
| 1454 | if (ACE_IS_TIGON_I(ap)) |
| 1455 | writel(tigonFwStartAddr, ®s->Pc); |
| 1456 | if (ap->version == 2) |
| 1457 | writel(tigon2FwStartAddr, ®s->Pc); |
| 1458 | |
| 1459 | writel(0, ®s->Mb0Lo); |
| 1460 | |
| 1461 | /* |
| 1462 | * Set tx_csm before we start receiving interrupts, otherwise |
| 1463 | * the interrupt handler might think it is supposed to process |
| 1464 | * tx ints before we are up and running, which may cause a null |
| 1465 | * pointer access in the int handler. |
| 1466 | */ |
| 1467 | ap->cur_rx = 0; |
| 1468 | ap->tx_prd = *(ap->tx_csm) = ap->tx_ret_csm = 0; |
| 1469 | |
| 1470 | wmb(); |
| 1471 | ace_set_txprd(regs, ap, 0); |
| 1472 | writel(0, ®s->RxRetCsm); |
| 1473 | |
| 1474 | /* |
| 1475 | * Zero the stats before starting the interface |
| 1476 | */ |
| 1477 | memset(&ap->stats, 0, sizeof(ap->stats)); |
| 1478 | |
| 1479 | /* |
| 1480 | * Enable DMA engine now. |
| 1481 | * If we do this sooner, Mckinley box pukes. |
| 1482 | * I assume it's because Tigon II DMA engine wants to check |
| 1483 | * *something* even before the CPU is started. |
| 1484 | */ |
| 1485 | writel(1, ®s->AssistState); /* enable DMA */ |
| 1486 | |
| 1487 | /* |
| 1488 | * Start the NIC CPU |
| 1489 | */ |
| 1490 | writel(readl(®s->CpuCtrl) & ~(CPU_HALT|CPU_TRACE), ®s->CpuCtrl); |
| 1491 | readl(®s->CpuCtrl); |
| 1492 | |
| 1493 | /* |
| 1494 | * Wait for the firmware to spin up - max 3 seconds. |
| 1495 | */ |
| 1496 | myjif = jiffies + 3 * HZ; |
| 1497 | while (time_before(jiffies, myjif) && !ap->fw_running) |
| 1498 | cpu_relax(); |
| 1499 | |
| 1500 | if (!ap->fw_running) { |
| 1501 | printk(KERN_ERR "%s: Firmware NOT running!\n", ap->name); |
| 1502 | |
| 1503 | ace_dump_trace(ap); |
| 1504 | writel(readl(®s->CpuCtrl) | CPU_HALT, ®s->CpuCtrl); |
| 1505 | readl(®s->CpuCtrl); |
| 1506 | |
| 1507 | /* aman@sgi.com - account for badly behaving firmware/NIC: |
| 1508 | * - have observed that the NIC may continue to generate |
| 1509 | * interrupts for some reason; attempt to stop it - halt |
| 1510 | * second CPU for Tigon II cards, and also clear Mb0 |
| 1511 | * - if we're a module, we'll fail to load if this was |
| 1512 | * the only GbE card in the system => if the kernel does |
| 1513 | * see an interrupt from the NIC, code to handle it is |
| 1514 | * gone and OOps! - so free_irq also |
| 1515 | */ |
| 1516 | if (ap->version >= 2) |
| 1517 | writel(readl(®s->CpuBCtrl) | CPU_HALT, |
| 1518 | ®s->CpuBCtrl); |
| 1519 | writel(0, ®s->Mb0Lo); |
| 1520 | readl(®s->Mb0Lo); |
| 1521 | |
| 1522 | ecode = -EBUSY; |
| 1523 | goto init_error; |
| 1524 | } |
| 1525 | |
| 1526 | /* |
| 1527 | * We load the ring here as there seem to be no way to tell the |
| 1528 | * firmware to wipe the ring without re-initializing it. |
| 1529 | */ |
| 1530 | if (!test_and_set_bit(0, &ap->std_refill_busy)) |
| 1531 | ace_load_std_rx_ring(ap, RX_RING_SIZE); |
| 1532 | else |
| 1533 | printk(KERN_ERR "%s: Someone is busy refilling the RX ring\n", |
| 1534 | ap->name); |
| 1535 | if (ap->version >= 2) { |
| 1536 | if (!test_and_set_bit(0, &ap->mini_refill_busy)) |
| 1537 | ace_load_mini_rx_ring(ap, RX_MINI_SIZE); |
| 1538 | else |
| 1539 | printk(KERN_ERR "%s: Someone is busy refilling " |
| 1540 | "the RX mini ring\n", ap->name); |
| 1541 | } |
| 1542 | return 0; |
| 1543 | |
| 1544 | init_error: |
| 1545 | ace_init_cleanup(dev); |
| 1546 | return ecode; |
| 1547 | } |
| 1548 | |
| 1549 | |
| 1550 | static void ace_set_rxtx_parms(struct net_device *dev, int jumbo) |
| 1551 | { |
| 1552 | struct ace_private *ap = netdev_priv(dev); |
| 1553 | struct ace_regs __iomem *regs = ap->regs; |
| 1554 | int board_idx = ap->board_idx; |
| 1555 | |
| 1556 | if (board_idx >= 0) { |
| 1557 | if (!jumbo) { |
| 1558 | if (!tx_coal_tick[board_idx]) |
| 1559 | writel(DEF_TX_COAL, ®s->TuneTxCoalTicks); |
| 1560 | if (!max_tx_desc[board_idx]) |
| 1561 | writel(DEF_TX_MAX_DESC, ®s->TuneMaxTxDesc); |
| 1562 | if (!rx_coal_tick[board_idx]) |
| 1563 | writel(DEF_RX_COAL, ®s->TuneRxCoalTicks); |
| 1564 | if (!max_rx_desc[board_idx]) |
| 1565 | writel(DEF_RX_MAX_DESC, ®s->TuneMaxRxDesc); |
| 1566 | if (!tx_ratio[board_idx]) |
| 1567 | writel(DEF_TX_RATIO, ®s->TxBufRat); |
| 1568 | } else { |
| 1569 | if (!tx_coal_tick[board_idx]) |
| 1570 | writel(DEF_JUMBO_TX_COAL, |
| 1571 | ®s->TuneTxCoalTicks); |
| 1572 | if (!max_tx_desc[board_idx]) |
| 1573 | writel(DEF_JUMBO_TX_MAX_DESC, |
| 1574 | ®s->TuneMaxTxDesc); |
| 1575 | if (!rx_coal_tick[board_idx]) |
| 1576 | writel(DEF_JUMBO_RX_COAL, |
| 1577 | ®s->TuneRxCoalTicks); |
| 1578 | if (!max_rx_desc[board_idx]) |
| 1579 | writel(DEF_JUMBO_RX_MAX_DESC, |
| 1580 | ®s->TuneMaxRxDesc); |
| 1581 | if (!tx_ratio[board_idx]) |
| 1582 | writel(DEF_JUMBO_TX_RATIO, ®s->TxBufRat); |
| 1583 | } |
| 1584 | } |
| 1585 | } |
| 1586 | |
| 1587 | |
| 1588 | static void ace_watchdog(struct net_device *data) |
| 1589 | { |
| 1590 | struct net_device *dev = data; |
| 1591 | struct ace_private *ap = netdev_priv(dev); |
| 1592 | struct ace_regs __iomem *regs = ap->regs; |
| 1593 | |
| 1594 | /* |
| 1595 | * We haven't received a stats update event for more than 2.5 |
| 1596 | * seconds and there is data in the transmit queue, thus we |
| 1597 | * asume the card is stuck. |
| 1598 | */ |
| 1599 | if (*ap->tx_csm != ap->tx_ret_csm) { |
| 1600 | printk(KERN_WARNING "%s: Transmitter is stuck, %08x\n", |
| 1601 | dev->name, (unsigned int)readl(®s->HostCtrl)); |
| 1602 | /* This can happen due to ieee flow control. */ |
| 1603 | } else { |
| 1604 | printk(KERN_DEBUG "%s: BUG... transmitter died. Kicking it.\n", |
| 1605 | dev->name); |
| 1606 | #if 0 |
| 1607 | netif_wake_queue(dev); |
| 1608 | #endif |
| 1609 | } |
| 1610 | } |
| 1611 | |
| 1612 | |
| 1613 | static void ace_tasklet(unsigned long dev) |
| 1614 | { |
| 1615 | struct ace_private *ap = netdev_priv((struct net_device *)dev); |
| 1616 | int cur_size; |
| 1617 | |
| 1618 | cur_size = atomic_read(&ap->cur_rx_bufs); |
| 1619 | if ((cur_size < RX_LOW_STD_THRES) && |
| 1620 | !test_and_set_bit(0, &ap->std_refill_busy)) { |
| 1621 | #ifdef DEBUG |
| 1622 | printk("refilling buffers (current %i)\n", cur_size); |
| 1623 | #endif |
| 1624 | ace_load_std_rx_ring(ap, RX_RING_SIZE - cur_size); |
| 1625 | } |
| 1626 | |
| 1627 | if (ap->version >= 2) { |
| 1628 | cur_size = atomic_read(&ap->cur_mini_bufs); |
| 1629 | if ((cur_size < RX_LOW_MINI_THRES) && |
| 1630 | !test_and_set_bit(0, &ap->mini_refill_busy)) { |
| 1631 | #ifdef DEBUG |
| 1632 | printk("refilling mini buffers (current %i)\n", |
| 1633 | cur_size); |
| 1634 | #endif |
| 1635 | ace_load_mini_rx_ring(ap, RX_MINI_SIZE - cur_size); |
| 1636 | } |
| 1637 | } |
| 1638 | |
| 1639 | cur_size = atomic_read(&ap->cur_jumbo_bufs); |
| 1640 | if (ap->jumbo && (cur_size < RX_LOW_JUMBO_THRES) && |
| 1641 | !test_and_set_bit(0, &ap->jumbo_refill_busy)) { |
| 1642 | #ifdef DEBUG |
| 1643 | printk("refilling jumbo buffers (current %i)\n", cur_size); |
| 1644 | #endif |
| 1645 | ace_load_jumbo_rx_ring(ap, RX_JUMBO_SIZE - cur_size); |
| 1646 | } |
| 1647 | ap->tasklet_pending = 0; |
| 1648 | } |
| 1649 | |
| 1650 | |
| 1651 | /* |
| 1652 | * Copy the contents of the NIC's trace buffer to kernel memory. |
| 1653 | */ |
| 1654 | static void ace_dump_trace(struct ace_private *ap) |
| 1655 | { |
| 1656 | #if 0 |
| 1657 | if (!ap->trace_buf) |
| 1658 | if (!(ap->trace_buf = kmalloc(ACE_TRACE_SIZE, GFP_KERNEL))) |
| 1659 | return; |
| 1660 | #endif |
| 1661 | } |
| 1662 | |
| 1663 | |
| 1664 | /* |
| 1665 | * Load the standard rx ring. |
| 1666 | * |
| 1667 | * Loading rings is safe without holding the spin lock since this is |
| 1668 | * done only before the device is enabled, thus no interrupts are |
| 1669 | * generated and by the interrupt handler/tasklet handler. |
| 1670 | */ |
| 1671 | static void ace_load_std_rx_ring(struct ace_private *ap, int nr_bufs) |
| 1672 | { |
| 1673 | struct ace_regs __iomem *regs = ap->regs; |
| 1674 | short i, idx; |
| 1675 | |
| 1676 | |
| 1677 | prefetchw(&ap->cur_rx_bufs); |
| 1678 | |
| 1679 | idx = ap->rx_std_skbprd; |
| 1680 | |
| 1681 | for (i = 0; i < nr_bufs; i++) { |
| 1682 | struct sk_buff *skb; |
| 1683 | struct rx_desc *rd; |
| 1684 | dma_addr_t mapping; |
| 1685 | |
| 1686 | skb = alloc_skb(ACE_STD_BUFSIZE + NET_IP_ALIGN, GFP_ATOMIC); |
| 1687 | if (!skb) |
| 1688 | break; |
| 1689 | |
| 1690 | skb_reserve(skb, NET_IP_ALIGN); |
| 1691 | mapping = pci_map_page(ap->pdev, virt_to_page(skb->data), |
| 1692 | offset_in_page(skb->data), |
| 1693 | ACE_STD_BUFSIZE, |
| 1694 | PCI_DMA_FROMDEVICE); |
| 1695 | ap->skb->rx_std_skbuff[idx].skb = skb; |
| 1696 | pci_unmap_addr_set(&ap->skb->rx_std_skbuff[idx], |
| 1697 | mapping, mapping); |
| 1698 | |
| 1699 | rd = &ap->rx_std_ring[idx]; |
| 1700 | set_aceaddr(&rd->addr, mapping); |
| 1701 | rd->size = ACE_STD_BUFSIZE; |
| 1702 | rd->idx = idx; |
| 1703 | idx = (idx + 1) % RX_STD_RING_ENTRIES; |
| 1704 | } |
| 1705 | |
| 1706 | if (!i) |
| 1707 | goto error_out; |
| 1708 | |
| 1709 | atomic_add(i, &ap->cur_rx_bufs); |
| 1710 | ap->rx_std_skbprd = idx; |
| 1711 | |
| 1712 | if (ACE_IS_TIGON_I(ap)) { |
| 1713 | struct cmd cmd; |
| 1714 | cmd.evt = C_SET_RX_PRD_IDX; |
| 1715 | cmd.code = 0; |
| 1716 | cmd.idx = ap->rx_std_skbprd; |
| 1717 | ace_issue_cmd(regs, &cmd); |
| 1718 | } else { |
| 1719 | writel(idx, ®s->RxStdPrd); |
| 1720 | wmb(); |
| 1721 | } |
| 1722 | |
| 1723 | out: |
| 1724 | clear_bit(0, &ap->std_refill_busy); |
| 1725 | return; |
| 1726 | |
| 1727 | error_out: |
| 1728 | printk(KERN_INFO "Out of memory when allocating " |
| 1729 | "standard receive buffers\n"); |
| 1730 | goto out; |
| 1731 | } |
| 1732 | |
| 1733 | |
| 1734 | static void ace_load_mini_rx_ring(struct ace_private *ap, int nr_bufs) |
| 1735 | { |
| 1736 | struct ace_regs __iomem *regs = ap->regs; |
| 1737 | short i, idx; |
| 1738 | |
| 1739 | prefetchw(&ap->cur_mini_bufs); |
| 1740 | |
| 1741 | idx = ap->rx_mini_skbprd; |
| 1742 | for (i = 0; i < nr_bufs; i++) { |
| 1743 | struct sk_buff *skb; |
| 1744 | struct rx_desc *rd; |
| 1745 | dma_addr_t mapping; |
| 1746 | |
| 1747 | skb = alloc_skb(ACE_MINI_BUFSIZE + NET_IP_ALIGN, GFP_ATOMIC); |
| 1748 | if (!skb) |
| 1749 | break; |
| 1750 | |
| 1751 | skb_reserve(skb, NET_IP_ALIGN); |
| 1752 | mapping = pci_map_page(ap->pdev, virt_to_page(skb->data), |
| 1753 | offset_in_page(skb->data), |
| 1754 | ACE_MINI_BUFSIZE, |
| 1755 | PCI_DMA_FROMDEVICE); |
| 1756 | ap->skb->rx_mini_skbuff[idx].skb = skb; |
| 1757 | pci_unmap_addr_set(&ap->skb->rx_mini_skbuff[idx], |
| 1758 | mapping, mapping); |
| 1759 | |
| 1760 | rd = &ap->rx_mini_ring[idx]; |
| 1761 | set_aceaddr(&rd->addr, mapping); |
| 1762 | rd->size = ACE_MINI_BUFSIZE; |
| 1763 | rd->idx = idx; |
| 1764 | idx = (idx + 1) % RX_MINI_RING_ENTRIES; |
| 1765 | } |
| 1766 | |
| 1767 | if (!i) |
| 1768 | goto error_out; |
| 1769 | |
| 1770 | atomic_add(i, &ap->cur_mini_bufs); |
| 1771 | |
| 1772 | ap->rx_mini_skbprd = idx; |
| 1773 | |
| 1774 | writel(idx, ®s->RxMiniPrd); |
| 1775 | wmb(); |
| 1776 | |
| 1777 | out: |
| 1778 | clear_bit(0, &ap->mini_refill_busy); |
| 1779 | return; |
| 1780 | error_out: |
| 1781 | printk(KERN_INFO "Out of memory when allocating " |
| 1782 | "mini receive buffers\n"); |
| 1783 | goto out; |
| 1784 | } |
| 1785 | |
| 1786 | |
| 1787 | /* |
| 1788 | * Load the jumbo rx ring, this may happen at any time if the MTU |
| 1789 | * is changed to a value > 1500. |
| 1790 | */ |
| 1791 | static void ace_load_jumbo_rx_ring(struct ace_private *ap, int nr_bufs) |
| 1792 | { |
| 1793 | struct ace_regs __iomem *regs = ap->regs; |
| 1794 | short i, idx; |
| 1795 | |
| 1796 | idx = ap->rx_jumbo_skbprd; |
| 1797 | |
| 1798 | for (i = 0; i < nr_bufs; i++) { |
| 1799 | struct sk_buff *skb; |
| 1800 | struct rx_desc *rd; |
| 1801 | dma_addr_t mapping; |
| 1802 | |
| 1803 | skb = alloc_skb(ACE_JUMBO_BUFSIZE + NET_IP_ALIGN, GFP_ATOMIC); |
| 1804 | if (!skb) |
| 1805 | break; |
| 1806 | |
| 1807 | skb_reserve(skb, NET_IP_ALIGN); |
| 1808 | mapping = pci_map_page(ap->pdev, virt_to_page(skb->data), |
| 1809 | offset_in_page(skb->data), |
| 1810 | ACE_JUMBO_BUFSIZE, |
| 1811 | PCI_DMA_FROMDEVICE); |
| 1812 | ap->skb->rx_jumbo_skbuff[idx].skb = skb; |
| 1813 | pci_unmap_addr_set(&ap->skb->rx_jumbo_skbuff[idx], |
| 1814 | mapping, mapping); |
| 1815 | |
| 1816 | rd = &ap->rx_jumbo_ring[idx]; |
| 1817 | set_aceaddr(&rd->addr, mapping); |
| 1818 | rd->size = ACE_JUMBO_BUFSIZE; |
| 1819 | rd->idx = idx; |
| 1820 | idx = (idx + 1) % RX_JUMBO_RING_ENTRIES; |
| 1821 | } |
| 1822 | |
| 1823 | if (!i) |
| 1824 | goto error_out; |
| 1825 | |
| 1826 | atomic_add(i, &ap->cur_jumbo_bufs); |
| 1827 | ap->rx_jumbo_skbprd = idx; |
| 1828 | |
| 1829 | if (ACE_IS_TIGON_I(ap)) { |
| 1830 | struct cmd cmd; |
| 1831 | cmd.evt = C_SET_RX_JUMBO_PRD_IDX; |
| 1832 | cmd.code = 0; |
| 1833 | cmd.idx = ap->rx_jumbo_skbprd; |
| 1834 | ace_issue_cmd(regs, &cmd); |
| 1835 | } else { |
| 1836 | writel(idx, ®s->RxJumboPrd); |
| 1837 | wmb(); |
| 1838 | } |
| 1839 | |
| 1840 | out: |
| 1841 | clear_bit(0, &ap->jumbo_refill_busy); |
| 1842 | return; |
| 1843 | error_out: |
| 1844 | if (net_ratelimit()) |
| 1845 | printk(KERN_INFO "Out of memory when allocating " |
| 1846 | "jumbo receive buffers\n"); |
| 1847 | goto out; |
| 1848 | } |
| 1849 | |
| 1850 | |
| 1851 | /* |
| 1852 | * All events are considered to be slow (RX/TX ints do not generate |
| 1853 | * events) and are handled here, outside the main interrupt handler, |
| 1854 | * to reduce the size of the handler. |
| 1855 | */ |
| 1856 | static u32 ace_handle_event(struct net_device *dev, u32 evtcsm, u32 evtprd) |
| 1857 | { |
| 1858 | struct ace_private *ap; |
| 1859 | |
| 1860 | ap = netdev_priv(dev); |
| 1861 | |
| 1862 | while (evtcsm != evtprd) { |
| 1863 | switch (ap->evt_ring[evtcsm].evt) { |
| 1864 | case E_FW_RUNNING: |
| 1865 | printk(KERN_INFO "%s: Firmware up and running\n", |
| 1866 | ap->name); |
| 1867 | ap->fw_running = 1; |
| 1868 | wmb(); |
| 1869 | break; |
| 1870 | case E_STATS_UPDATED: |
| 1871 | break; |
| 1872 | case E_LNK_STATE: |
| 1873 | { |
| 1874 | u16 code = ap->evt_ring[evtcsm].code; |
| 1875 | switch (code) { |
| 1876 | case E_C_LINK_UP: |
| 1877 | { |
| 1878 | u32 state = readl(&ap->regs->GigLnkState); |
| 1879 | printk(KERN_WARNING "%s: Optical link UP " |
| 1880 | "(%s Duplex, Flow Control: %s%s)\n", |
| 1881 | ap->name, |
| 1882 | state & LNK_FULL_DUPLEX ? "Full":"Half", |
| 1883 | state & LNK_TX_FLOW_CTL_Y ? "TX " : "", |
| 1884 | state & LNK_RX_FLOW_CTL_Y ? "RX" : ""); |
| 1885 | break; |
| 1886 | } |
| 1887 | case E_C_LINK_DOWN: |
| 1888 | printk(KERN_WARNING "%s: Optical link DOWN\n", |
| 1889 | ap->name); |
| 1890 | break; |
| 1891 | case E_C_LINK_10_100: |
| 1892 | printk(KERN_WARNING "%s: 10/100BaseT link " |
| 1893 | "UP\n", ap->name); |
| 1894 | break; |
| 1895 | default: |
| 1896 | printk(KERN_ERR "%s: Unknown optical link " |
| 1897 | "state %02x\n", ap->name, code); |
| 1898 | } |
| 1899 | break; |
| 1900 | } |
| 1901 | case E_ERROR: |
| 1902 | switch(ap->evt_ring[evtcsm].code) { |
| 1903 | case E_C_ERR_INVAL_CMD: |
| 1904 | printk(KERN_ERR "%s: invalid command error\n", |
| 1905 | ap->name); |
| 1906 | break; |
| 1907 | case E_C_ERR_UNIMP_CMD: |
| 1908 | printk(KERN_ERR "%s: unimplemented command " |
| 1909 | "error\n", ap->name); |
| 1910 | break; |
| 1911 | case E_C_ERR_BAD_CFG: |
| 1912 | printk(KERN_ERR "%s: bad config error\n", |
| 1913 | ap->name); |
| 1914 | break; |
| 1915 | default: |
| 1916 | printk(KERN_ERR "%s: unknown error %02x\n", |
| 1917 | ap->name, ap->evt_ring[evtcsm].code); |
| 1918 | } |
| 1919 | break; |
| 1920 | case E_RESET_JUMBO_RNG: |
| 1921 | { |
| 1922 | int i; |
| 1923 | for (i = 0; i < RX_JUMBO_RING_ENTRIES; i++) { |
| 1924 | if (ap->skb->rx_jumbo_skbuff[i].skb) { |
| 1925 | ap->rx_jumbo_ring[i].size = 0; |
| 1926 | set_aceaddr(&ap->rx_jumbo_ring[i].addr, 0); |
| 1927 | dev_kfree_skb(ap->skb->rx_jumbo_skbuff[i].skb); |
| 1928 | ap->skb->rx_jumbo_skbuff[i].skb = NULL; |
| 1929 | } |
| 1930 | } |
| 1931 | |
| 1932 | if (ACE_IS_TIGON_I(ap)) { |
| 1933 | struct cmd cmd; |
| 1934 | cmd.evt = C_SET_RX_JUMBO_PRD_IDX; |
| 1935 | cmd.code = 0; |
| 1936 | cmd.idx = 0; |
| 1937 | ace_issue_cmd(ap->regs, &cmd); |
| 1938 | } else { |
| 1939 | writel(0, &((ap->regs)->RxJumboPrd)); |
| 1940 | wmb(); |
| 1941 | } |
| 1942 | |
| 1943 | ap->jumbo = 0; |
| 1944 | ap->rx_jumbo_skbprd = 0; |
| 1945 | printk(KERN_INFO "%s: Jumbo ring flushed\n", |
| 1946 | ap->name); |
| 1947 | clear_bit(0, &ap->jumbo_refill_busy); |
| 1948 | break; |
| 1949 | } |
| 1950 | default: |
| 1951 | printk(KERN_ERR "%s: Unhandled event 0x%02x\n", |
| 1952 | ap->name, ap->evt_ring[evtcsm].evt); |
| 1953 | } |
| 1954 | evtcsm = (evtcsm + 1) % EVT_RING_ENTRIES; |
| 1955 | } |
| 1956 | |
| 1957 | return evtcsm; |
| 1958 | } |
| 1959 | |
| 1960 | |
| 1961 | static void ace_rx_int(struct net_device *dev, u32 rxretprd, u32 rxretcsm) |
| 1962 | { |
| 1963 | struct ace_private *ap = netdev_priv(dev); |
| 1964 | u32 idx; |
| 1965 | int mini_count = 0, std_count = 0; |
| 1966 | |
| 1967 | idx = rxretcsm; |
| 1968 | |
| 1969 | prefetchw(&ap->cur_rx_bufs); |
| 1970 | prefetchw(&ap->cur_mini_bufs); |
| 1971 | |
| 1972 | while (idx != rxretprd) { |
| 1973 | struct ring_info *rip; |
| 1974 | struct sk_buff *skb; |
| 1975 | struct rx_desc *rxdesc, *retdesc; |
| 1976 | u32 skbidx; |
| 1977 | int bd_flags, desc_type, mapsize; |
| 1978 | u16 csum; |
| 1979 | |
| 1980 | |
| 1981 | /* make sure the rx descriptor isn't read before rxretprd */ |
| 1982 | if (idx == rxretcsm) |
| 1983 | rmb(); |
| 1984 | |
| 1985 | retdesc = &ap->rx_return_ring[idx]; |
| 1986 | skbidx = retdesc->idx; |
| 1987 | bd_flags = retdesc->flags; |
| 1988 | desc_type = bd_flags & (BD_FLG_JUMBO | BD_FLG_MINI); |
| 1989 | |
| 1990 | switch(desc_type) { |
| 1991 | /* |
| 1992 | * Normal frames do not have any flags set |
| 1993 | * |
| 1994 | * Mini and normal frames arrive frequently, |
| 1995 | * so use a local counter to avoid doing |
| 1996 | * atomic operations for each packet arriving. |
| 1997 | */ |
| 1998 | case 0: |
| 1999 | rip = &ap->skb->rx_std_skbuff[skbidx]; |
| 2000 | mapsize = ACE_STD_BUFSIZE; |
| 2001 | rxdesc = &ap->rx_std_ring[skbidx]; |
| 2002 | std_count++; |
| 2003 | break; |
| 2004 | case BD_FLG_JUMBO: |
| 2005 | rip = &ap->skb->rx_jumbo_skbuff[skbidx]; |
| 2006 | mapsize = ACE_JUMBO_BUFSIZE; |
| 2007 | rxdesc = &ap->rx_jumbo_ring[skbidx]; |
| 2008 | atomic_dec(&ap->cur_jumbo_bufs); |
| 2009 | break; |
| 2010 | case BD_FLG_MINI: |
| 2011 | rip = &ap->skb->rx_mini_skbuff[skbidx]; |
| 2012 | mapsize = ACE_MINI_BUFSIZE; |
| 2013 | rxdesc = &ap->rx_mini_ring[skbidx]; |
| 2014 | mini_count++; |
| 2015 | break; |
| 2016 | default: |
| 2017 | printk(KERN_INFO "%s: unknown frame type (0x%02x) " |
| 2018 | "returned by NIC\n", dev->name, |
| 2019 | retdesc->flags); |
| 2020 | goto error; |
| 2021 | } |
| 2022 | |
| 2023 | skb = rip->skb; |
| 2024 | rip->skb = NULL; |
| 2025 | pci_unmap_page(ap->pdev, |
| 2026 | pci_unmap_addr(rip, mapping), |
| 2027 | mapsize, |
| 2028 | PCI_DMA_FROMDEVICE); |
| 2029 | skb_put(skb, retdesc->size); |
| 2030 | |
| 2031 | /* |
| 2032 | * Fly baby, fly! |
| 2033 | */ |
| 2034 | csum = retdesc->tcp_udp_csum; |
| 2035 | |
| 2036 | skb->dev = dev; |
| 2037 | skb->protocol = eth_type_trans(skb, dev); |
| 2038 | |
| 2039 | /* |
| 2040 | * Instead of forcing the poor tigon mips cpu to calculate |
| 2041 | * pseudo hdr checksum, we do this ourselves. |
| 2042 | */ |
| 2043 | if (bd_flags & BD_FLG_TCP_UDP_SUM) { |
| 2044 | skb->csum = htons(csum); |
| 2045 | skb->ip_summed = CHECKSUM_HW; |
| 2046 | } else { |
| 2047 | skb->ip_summed = CHECKSUM_NONE; |
| 2048 | } |
| 2049 | |
| 2050 | /* send it up */ |
| 2051 | #if ACENIC_DO_VLAN |
| 2052 | if (ap->vlgrp && (bd_flags & BD_FLG_VLAN_TAG)) { |
| 2053 | vlan_hwaccel_rx(skb, ap->vlgrp, retdesc->vlan); |
| 2054 | } else |
| 2055 | #endif |
| 2056 | netif_rx(skb); |
| 2057 | |
| 2058 | dev->last_rx = jiffies; |
| 2059 | ap->stats.rx_packets++; |
| 2060 | ap->stats.rx_bytes += retdesc->size; |
| 2061 | |
| 2062 | idx = (idx + 1) % RX_RETURN_RING_ENTRIES; |
| 2063 | } |
| 2064 | |
| 2065 | atomic_sub(std_count, &ap->cur_rx_bufs); |
| 2066 | if (!ACE_IS_TIGON_I(ap)) |
| 2067 | atomic_sub(mini_count, &ap->cur_mini_bufs); |
| 2068 | |
| 2069 | out: |
| 2070 | /* |
| 2071 | * According to the documentation RxRetCsm is obsolete with |
| 2072 | * the 12.3.x Firmware - my Tigon I NICs seem to disagree! |
| 2073 | */ |
| 2074 | if (ACE_IS_TIGON_I(ap)) { |
| 2075 | writel(idx, &ap->regs->RxRetCsm); |
| 2076 | } |
| 2077 | ap->cur_rx = idx; |
| 2078 | |
| 2079 | return; |
| 2080 | error: |
| 2081 | idx = rxretprd; |
| 2082 | goto out; |
| 2083 | } |
| 2084 | |
| 2085 | |
| 2086 | static inline void ace_tx_int(struct net_device *dev, |
| 2087 | u32 txcsm, u32 idx) |
| 2088 | { |
| 2089 | struct ace_private *ap = netdev_priv(dev); |
| 2090 | |
| 2091 | do { |
| 2092 | struct sk_buff *skb; |
| 2093 | dma_addr_t mapping; |
| 2094 | struct tx_ring_info *info; |
| 2095 | |
| 2096 | info = ap->skb->tx_skbuff + idx; |
| 2097 | skb = info->skb; |
| 2098 | mapping = pci_unmap_addr(info, mapping); |
| 2099 | |
| 2100 | if (mapping) { |
| 2101 | pci_unmap_page(ap->pdev, mapping, |
| 2102 | pci_unmap_len(info, maplen), |
| 2103 | PCI_DMA_TODEVICE); |
| 2104 | pci_unmap_addr_set(info, mapping, 0); |
| 2105 | } |
| 2106 | |
| 2107 | if (skb) { |
| 2108 | ap->stats.tx_packets++; |
| 2109 | ap->stats.tx_bytes += skb->len; |
| 2110 | dev_kfree_skb_irq(skb); |
| 2111 | info->skb = NULL; |
| 2112 | } |
| 2113 | |
| 2114 | idx = (idx + 1) % ACE_TX_RING_ENTRIES(ap); |
| 2115 | } while (idx != txcsm); |
| 2116 | |
| 2117 | if (netif_queue_stopped(dev)) |
| 2118 | netif_wake_queue(dev); |
| 2119 | |
| 2120 | wmb(); |
| 2121 | ap->tx_ret_csm = txcsm; |
| 2122 | |
| 2123 | /* So... tx_ret_csm is advanced _after_ check for device wakeup. |
| 2124 | * |
| 2125 | * We could try to make it before. In this case we would get |
| 2126 | * the following race condition: hard_start_xmit on other cpu |
| 2127 | * enters after we advanced tx_ret_csm and fills space, |
| 2128 | * which we have just freed, so that we make illegal device wakeup. |
| 2129 | * There is no good way to workaround this (at entry |
| 2130 | * to ace_start_xmit detects this condition and prevents |
| 2131 | * ring corruption, but it is not a good workaround.) |
| 2132 | * |
| 2133 | * When tx_ret_csm is advanced after, we wake up device _only_ |
| 2134 | * if we really have some space in ring (though the core doing |
| 2135 | * hard_start_xmit can see full ring for some period and has to |
| 2136 | * synchronize.) Superb. |
| 2137 | * BUT! We get another subtle race condition. hard_start_xmit |
| 2138 | * may think that ring is full between wakeup and advancing |
| 2139 | * tx_ret_csm and will stop device instantly! It is not so bad. |
| 2140 | * We are guaranteed that there is something in ring, so that |
| 2141 | * the next irq will resume transmission. To speedup this we could |
| 2142 | * mark descriptor, which closes ring with BD_FLG_COAL_NOW |
| 2143 | * (see ace_start_xmit). |
| 2144 | * |
| 2145 | * Well, this dilemma exists in all lock-free devices. |
| 2146 | * We, following scheme used in drivers by Donald Becker, |
| 2147 | * select the least dangerous. |
| 2148 | * --ANK |
| 2149 | */ |
| 2150 | } |
| 2151 | |
| 2152 | |
| 2153 | static irqreturn_t ace_interrupt(int irq, void *dev_id, struct pt_regs *ptregs) |
| 2154 | { |
| 2155 | struct net_device *dev = (struct net_device *)dev_id; |
| 2156 | struct ace_private *ap = netdev_priv(dev); |
| 2157 | struct ace_regs __iomem *regs = ap->regs; |
| 2158 | u32 idx; |
| 2159 | u32 txcsm, rxretcsm, rxretprd; |
| 2160 | u32 evtcsm, evtprd; |
| 2161 | |
| 2162 | /* |
| 2163 | * In case of PCI shared interrupts or spurious interrupts, |
| 2164 | * we want to make sure it is actually our interrupt before |
| 2165 | * spending any time in here. |
| 2166 | */ |
| 2167 | if (!(readl(®s->HostCtrl) & IN_INT)) |
| 2168 | return IRQ_NONE; |
| 2169 | |
| 2170 | /* |
| 2171 | * ACK intr now. Otherwise we will lose updates to rx_ret_prd, |
| 2172 | * which happened _after_ rxretprd = *ap->rx_ret_prd; but before |
| 2173 | * writel(0, ®s->Mb0Lo). |
| 2174 | * |
| 2175 | * "IRQ avoidance" recommended in docs applies to IRQs served |
| 2176 | * threads and it is wrong even for that case. |
| 2177 | */ |
| 2178 | writel(0, ®s->Mb0Lo); |
| 2179 | readl(®s->Mb0Lo); |
| 2180 | |
| 2181 | /* |
| 2182 | * There is no conflict between transmit handling in |
| 2183 | * start_xmit and receive processing, thus there is no reason |
| 2184 | * to take a spin lock for RX handling. Wait until we start |
| 2185 | * working on the other stuff - hey we don't need a spin lock |
| 2186 | * anymore. |
| 2187 | */ |
| 2188 | rxretprd = *ap->rx_ret_prd; |
| 2189 | rxretcsm = ap->cur_rx; |
| 2190 | |
| 2191 | if (rxretprd != rxretcsm) |
| 2192 | ace_rx_int(dev, rxretprd, rxretcsm); |
| 2193 | |
| 2194 | txcsm = *ap->tx_csm; |
| 2195 | idx = ap->tx_ret_csm; |
| 2196 | |
| 2197 | if (txcsm != idx) { |
| 2198 | /* |
| 2199 | * If each skb takes only one descriptor this check degenerates |
| 2200 | * to identity, because new space has just been opened. |
| 2201 | * But if skbs are fragmented we must check that this index |
| 2202 | * update releases enough of space, otherwise we just |
| 2203 | * wait for device to make more work. |
| 2204 | */ |
| 2205 | if (!tx_ring_full(ap, txcsm, ap->tx_prd)) |
| 2206 | ace_tx_int(dev, txcsm, idx); |
| 2207 | } |
| 2208 | |
| 2209 | evtcsm = readl(®s->EvtCsm); |
| 2210 | evtprd = *ap->evt_prd; |
| 2211 | |
| 2212 | if (evtcsm != evtprd) { |
| 2213 | evtcsm = ace_handle_event(dev, evtcsm, evtprd); |
| 2214 | writel(evtcsm, ®s->EvtCsm); |
| 2215 | } |
| 2216 | |
| 2217 | /* |
| 2218 | * This has to go last in the interrupt handler and run with |
| 2219 | * the spin lock released ... what lock? |
| 2220 | */ |
| 2221 | if (netif_running(dev)) { |
| 2222 | int cur_size; |
| 2223 | int run_tasklet = 0; |
| 2224 | |
| 2225 | cur_size = atomic_read(&ap->cur_rx_bufs); |
| 2226 | if (cur_size < RX_LOW_STD_THRES) { |
| 2227 | if ((cur_size < RX_PANIC_STD_THRES) && |
| 2228 | !test_and_set_bit(0, &ap->std_refill_busy)) { |
| 2229 | #ifdef DEBUG |
| 2230 | printk("low on std buffers %i\n", cur_size); |
| 2231 | #endif |
| 2232 | ace_load_std_rx_ring(ap, |
| 2233 | RX_RING_SIZE - cur_size); |
| 2234 | } else |
| 2235 | run_tasklet = 1; |
| 2236 | } |
| 2237 | |
| 2238 | if (!ACE_IS_TIGON_I(ap)) { |
| 2239 | cur_size = atomic_read(&ap->cur_mini_bufs); |
| 2240 | if (cur_size < RX_LOW_MINI_THRES) { |
| 2241 | if ((cur_size < RX_PANIC_MINI_THRES) && |
| 2242 | !test_and_set_bit(0, |
| 2243 | &ap->mini_refill_busy)) { |
| 2244 | #ifdef DEBUG |
| 2245 | printk("low on mini buffers %i\n", |
| 2246 | cur_size); |
| 2247 | #endif |
| 2248 | ace_load_mini_rx_ring(ap, RX_MINI_SIZE - cur_size); |
| 2249 | } else |
| 2250 | run_tasklet = 1; |
| 2251 | } |
| 2252 | } |
| 2253 | |
| 2254 | if (ap->jumbo) { |
| 2255 | cur_size = atomic_read(&ap->cur_jumbo_bufs); |
| 2256 | if (cur_size < RX_LOW_JUMBO_THRES) { |
| 2257 | if ((cur_size < RX_PANIC_JUMBO_THRES) && |
| 2258 | !test_and_set_bit(0, |
| 2259 | &ap->jumbo_refill_busy)){ |
| 2260 | #ifdef DEBUG |
| 2261 | printk("low on jumbo buffers %i\n", |
| 2262 | cur_size); |
| 2263 | #endif |
| 2264 | ace_load_jumbo_rx_ring(ap, RX_JUMBO_SIZE - cur_size); |
| 2265 | } else |
| 2266 | run_tasklet = 1; |
| 2267 | } |
| 2268 | } |
| 2269 | if (run_tasklet && !ap->tasklet_pending) { |
| 2270 | ap->tasklet_pending = 1; |
| 2271 | tasklet_schedule(&ap->ace_tasklet); |
| 2272 | } |
| 2273 | } |
| 2274 | |
| 2275 | return IRQ_HANDLED; |
| 2276 | } |
| 2277 | |
| 2278 | |
| 2279 | #if ACENIC_DO_VLAN |
| 2280 | static void ace_vlan_rx_register(struct net_device *dev, struct vlan_group *grp) |
| 2281 | { |
| 2282 | struct ace_private *ap = netdev_priv(dev); |
| 2283 | unsigned long flags; |
| 2284 | |
| 2285 | local_irq_save(flags); |
| 2286 | ace_mask_irq(dev); |
| 2287 | |
| 2288 | ap->vlgrp = grp; |
| 2289 | |
| 2290 | ace_unmask_irq(dev); |
| 2291 | local_irq_restore(flags); |
| 2292 | } |
| 2293 | |
| 2294 | |
| 2295 | static void ace_vlan_rx_kill_vid(struct net_device *dev, unsigned short vid) |
| 2296 | { |
| 2297 | struct ace_private *ap = netdev_priv(dev); |
| 2298 | unsigned long flags; |
| 2299 | |
| 2300 | local_irq_save(flags); |
| 2301 | ace_mask_irq(dev); |
| 2302 | |
| 2303 | if (ap->vlgrp) |
| 2304 | ap->vlgrp->vlan_devices[vid] = NULL; |
| 2305 | |
| 2306 | ace_unmask_irq(dev); |
| 2307 | local_irq_restore(flags); |
| 2308 | } |
| 2309 | #endif /* ACENIC_DO_VLAN */ |
| 2310 | |
| 2311 | |
| 2312 | static int ace_open(struct net_device *dev) |
| 2313 | { |
| 2314 | struct ace_private *ap = netdev_priv(dev); |
| 2315 | struct ace_regs __iomem *regs = ap->regs; |
| 2316 | struct cmd cmd; |
| 2317 | |
| 2318 | if (!(ap->fw_running)) { |
| 2319 | printk(KERN_WARNING "%s: Firmware not running!\n", dev->name); |
| 2320 | return -EBUSY; |
| 2321 | } |
| 2322 | |
| 2323 | writel(dev->mtu + ETH_HLEN + 4, ®s->IfMtu); |
| 2324 | |
| 2325 | cmd.evt = C_CLEAR_STATS; |
| 2326 | cmd.code = 0; |
| 2327 | cmd.idx = 0; |
| 2328 | ace_issue_cmd(regs, &cmd); |
| 2329 | |
| 2330 | cmd.evt = C_HOST_STATE; |
| 2331 | cmd.code = C_C_STACK_UP; |
| 2332 | cmd.idx = 0; |
| 2333 | ace_issue_cmd(regs, &cmd); |
| 2334 | |
| 2335 | if (ap->jumbo && |
| 2336 | !test_and_set_bit(0, &ap->jumbo_refill_busy)) |
| 2337 | ace_load_jumbo_rx_ring(ap, RX_JUMBO_SIZE); |
| 2338 | |
| 2339 | if (dev->flags & IFF_PROMISC) { |
| 2340 | cmd.evt = C_SET_PROMISC_MODE; |
| 2341 | cmd.code = C_C_PROMISC_ENABLE; |
| 2342 | cmd.idx = 0; |
| 2343 | ace_issue_cmd(regs, &cmd); |
| 2344 | |
| 2345 | ap->promisc = 1; |
| 2346 | }else |
| 2347 | ap->promisc = 0; |
| 2348 | ap->mcast_all = 0; |
| 2349 | |
| 2350 | #if 0 |
| 2351 | cmd.evt = C_LNK_NEGOTIATION; |
| 2352 | cmd.code = 0; |
| 2353 | cmd.idx = 0; |
| 2354 | ace_issue_cmd(regs, &cmd); |
| 2355 | #endif |
| 2356 | |
| 2357 | netif_start_queue(dev); |
| 2358 | |
| 2359 | /* |
| 2360 | * Setup the bottom half rx ring refill handler |
| 2361 | */ |
| 2362 | tasklet_init(&ap->ace_tasklet, ace_tasklet, (unsigned long)dev); |
| 2363 | return 0; |
| 2364 | } |
| 2365 | |
| 2366 | |
| 2367 | static int ace_close(struct net_device *dev) |
| 2368 | { |
| 2369 | struct ace_private *ap = netdev_priv(dev); |
| 2370 | struct ace_regs __iomem *regs = ap->regs; |
| 2371 | struct cmd cmd; |
| 2372 | unsigned long flags; |
| 2373 | short i; |
| 2374 | |
| 2375 | /* |
| 2376 | * Without (or before) releasing irq and stopping hardware, this |
| 2377 | * is an absolute non-sense, by the way. It will be reset instantly |
| 2378 | * by the first irq. |
| 2379 | */ |
| 2380 | netif_stop_queue(dev); |
| 2381 | |
| 2382 | |
| 2383 | if (ap->promisc) { |
| 2384 | cmd.evt = C_SET_PROMISC_MODE; |
| 2385 | cmd.code = C_C_PROMISC_DISABLE; |
| 2386 | cmd.idx = 0; |
| 2387 | ace_issue_cmd(regs, &cmd); |
| 2388 | ap->promisc = 0; |
| 2389 | } |
| 2390 | |
| 2391 | cmd.evt = C_HOST_STATE; |
| 2392 | cmd.code = C_C_STACK_DOWN; |
| 2393 | cmd.idx = 0; |
| 2394 | ace_issue_cmd(regs, &cmd); |
| 2395 | |
| 2396 | tasklet_kill(&ap->ace_tasklet); |
| 2397 | |
| 2398 | /* |
| 2399 | * Make sure one CPU is not processing packets while |
| 2400 | * buffers are being released by another. |
| 2401 | */ |
| 2402 | |
| 2403 | local_irq_save(flags); |
| 2404 | ace_mask_irq(dev); |
| 2405 | |
| 2406 | for (i = 0; i < ACE_TX_RING_ENTRIES(ap); i++) { |
| 2407 | struct sk_buff *skb; |
| 2408 | dma_addr_t mapping; |
| 2409 | struct tx_ring_info *info; |
| 2410 | |
| 2411 | info = ap->skb->tx_skbuff + i; |
| 2412 | skb = info->skb; |
| 2413 | mapping = pci_unmap_addr(info, mapping); |
| 2414 | |
| 2415 | if (mapping) { |
| 2416 | if (ACE_IS_TIGON_I(ap)) { |
| 2417 | struct tx_desc __iomem *tx |
| 2418 | = (struct tx_desc __iomem *) &ap->tx_ring[i]; |
| 2419 | writel(0, &tx->addr.addrhi); |
| 2420 | writel(0, &tx->addr.addrlo); |
| 2421 | writel(0, &tx->flagsize); |
| 2422 | } else |
| 2423 | memset(ap->tx_ring + i, 0, |
| 2424 | sizeof(struct tx_desc)); |
| 2425 | pci_unmap_page(ap->pdev, mapping, |
| 2426 | pci_unmap_len(info, maplen), |
| 2427 | PCI_DMA_TODEVICE); |
| 2428 | pci_unmap_addr_set(info, mapping, 0); |
| 2429 | } |
| 2430 | if (skb) { |
| 2431 | dev_kfree_skb(skb); |
| 2432 | info->skb = NULL; |
| 2433 | } |
| 2434 | } |
| 2435 | |
| 2436 | if (ap->jumbo) { |
| 2437 | cmd.evt = C_RESET_JUMBO_RNG; |
| 2438 | cmd.code = 0; |
| 2439 | cmd.idx = 0; |
| 2440 | ace_issue_cmd(regs, &cmd); |
| 2441 | } |
| 2442 | |
| 2443 | ace_unmask_irq(dev); |
| 2444 | local_irq_restore(flags); |
| 2445 | |
| 2446 | return 0; |
| 2447 | } |
| 2448 | |
| 2449 | |
| 2450 | static inline dma_addr_t |
| 2451 | ace_map_tx_skb(struct ace_private *ap, struct sk_buff *skb, |
| 2452 | struct sk_buff *tail, u32 idx) |
| 2453 | { |
| 2454 | dma_addr_t mapping; |
| 2455 | struct tx_ring_info *info; |
| 2456 | |
| 2457 | mapping = pci_map_page(ap->pdev, virt_to_page(skb->data), |
| 2458 | offset_in_page(skb->data), |
| 2459 | skb->len, PCI_DMA_TODEVICE); |
| 2460 | |
| 2461 | info = ap->skb->tx_skbuff + idx; |
| 2462 | info->skb = tail; |
| 2463 | pci_unmap_addr_set(info, mapping, mapping); |
| 2464 | pci_unmap_len_set(info, maplen, skb->len); |
| 2465 | return mapping; |
| 2466 | } |
| 2467 | |
| 2468 | |
| 2469 | static inline void |
| 2470 | ace_load_tx_bd(struct ace_private *ap, struct tx_desc *desc, u64 addr, |
| 2471 | u32 flagsize, u32 vlan_tag) |
| 2472 | { |
| 2473 | #if !USE_TX_COAL_NOW |
| 2474 | flagsize &= ~BD_FLG_COAL_NOW; |
| 2475 | #endif |
| 2476 | |
| 2477 | if (ACE_IS_TIGON_I(ap)) { |
| 2478 | struct tx_desc __iomem *io = (struct tx_desc __iomem *) desc; |
| 2479 | writel(addr >> 32, &io->addr.addrhi); |
| 2480 | writel(addr & 0xffffffff, &io->addr.addrlo); |
| 2481 | writel(flagsize, &io->flagsize); |
| 2482 | #if ACENIC_DO_VLAN |
| 2483 | writel(vlan_tag, &io->vlanres); |
| 2484 | #endif |
| 2485 | } else { |
| 2486 | desc->addr.addrhi = addr >> 32; |
| 2487 | desc->addr.addrlo = addr; |
| 2488 | desc->flagsize = flagsize; |
| 2489 | #if ACENIC_DO_VLAN |
| 2490 | desc->vlanres = vlan_tag; |
| 2491 | #endif |
| 2492 | } |
| 2493 | } |
| 2494 | |
| 2495 | |
| 2496 | static int ace_start_xmit(struct sk_buff *skb, struct net_device *dev) |
| 2497 | { |
| 2498 | struct ace_private *ap = netdev_priv(dev); |
| 2499 | struct ace_regs __iomem *regs = ap->regs; |
| 2500 | struct tx_desc *desc; |
| 2501 | u32 idx, flagsize; |
| 2502 | unsigned long maxjiff = jiffies + 3*HZ; |
| 2503 | |
| 2504 | restart: |
| 2505 | idx = ap->tx_prd; |
| 2506 | |
| 2507 | if (tx_ring_full(ap, ap->tx_ret_csm, idx)) |
| 2508 | goto overflow; |
| 2509 | |
| 2510 | if (!skb_shinfo(skb)->nr_frags) { |
| 2511 | dma_addr_t mapping; |
| 2512 | u32 vlan_tag = 0; |
| 2513 | |
| 2514 | mapping = ace_map_tx_skb(ap, skb, skb, idx); |
| 2515 | flagsize = (skb->len << 16) | (BD_FLG_END); |
| 2516 | if (skb->ip_summed == CHECKSUM_HW) |
| 2517 | flagsize |= BD_FLG_TCP_UDP_SUM; |
| 2518 | #if ACENIC_DO_VLAN |
| 2519 | if (vlan_tx_tag_present(skb)) { |
| 2520 | flagsize |= BD_FLG_VLAN_TAG; |
| 2521 | vlan_tag = vlan_tx_tag_get(skb); |
| 2522 | } |
| 2523 | #endif |
| 2524 | desc = ap->tx_ring + idx; |
| 2525 | idx = (idx + 1) % ACE_TX_RING_ENTRIES(ap); |
| 2526 | |
| 2527 | /* Look at ace_tx_int for explanations. */ |
| 2528 | if (tx_ring_full(ap, ap->tx_ret_csm, idx)) |
| 2529 | flagsize |= BD_FLG_COAL_NOW; |
| 2530 | |
| 2531 | ace_load_tx_bd(ap, desc, mapping, flagsize, vlan_tag); |
| 2532 | } else { |
| 2533 | dma_addr_t mapping; |
| 2534 | u32 vlan_tag = 0; |
| 2535 | int i, len = 0; |
| 2536 | |
| 2537 | mapping = ace_map_tx_skb(ap, skb, NULL, idx); |
| 2538 | flagsize = (skb_headlen(skb) << 16); |
| 2539 | if (skb->ip_summed == CHECKSUM_HW) |
| 2540 | flagsize |= BD_FLG_TCP_UDP_SUM; |
| 2541 | #if ACENIC_DO_VLAN |
| 2542 | if (vlan_tx_tag_present(skb)) { |
| 2543 | flagsize |= BD_FLG_VLAN_TAG; |
| 2544 | vlan_tag = vlan_tx_tag_get(skb); |
| 2545 | } |
| 2546 | #endif |
| 2547 | |
| 2548 | ace_load_tx_bd(ap, ap->tx_ring + idx, mapping, flagsize, vlan_tag); |
| 2549 | |
| 2550 | idx = (idx + 1) % ACE_TX_RING_ENTRIES(ap); |
| 2551 | |
| 2552 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
| 2553 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
| 2554 | struct tx_ring_info *info; |
| 2555 | |
| 2556 | len += frag->size; |
| 2557 | info = ap->skb->tx_skbuff + idx; |
| 2558 | desc = ap->tx_ring + idx; |
| 2559 | |
| 2560 | mapping = pci_map_page(ap->pdev, frag->page, |
| 2561 | frag->page_offset, frag->size, |
| 2562 | PCI_DMA_TODEVICE); |
| 2563 | |
| 2564 | flagsize = (frag->size << 16); |
| 2565 | if (skb->ip_summed == CHECKSUM_HW) |
| 2566 | flagsize |= BD_FLG_TCP_UDP_SUM; |
| 2567 | idx = (idx + 1) % ACE_TX_RING_ENTRIES(ap); |
| 2568 | |
| 2569 | if (i == skb_shinfo(skb)->nr_frags - 1) { |
| 2570 | flagsize |= BD_FLG_END; |
| 2571 | if (tx_ring_full(ap, ap->tx_ret_csm, idx)) |
| 2572 | flagsize |= BD_FLG_COAL_NOW; |
| 2573 | |
| 2574 | /* |
| 2575 | * Only the last fragment frees |
| 2576 | * the skb! |
| 2577 | */ |
| 2578 | info->skb = skb; |
| 2579 | } else { |
| 2580 | info->skb = NULL; |
| 2581 | } |
| 2582 | pci_unmap_addr_set(info, mapping, mapping); |
| 2583 | pci_unmap_len_set(info, maplen, frag->size); |
| 2584 | ace_load_tx_bd(ap, desc, mapping, flagsize, vlan_tag); |
| 2585 | } |
| 2586 | } |
| 2587 | |
| 2588 | wmb(); |
| 2589 | ap->tx_prd = idx; |
| 2590 | ace_set_txprd(regs, ap, idx); |
| 2591 | |
| 2592 | if (flagsize & BD_FLG_COAL_NOW) { |
| 2593 | netif_stop_queue(dev); |
| 2594 | |
| 2595 | /* |
| 2596 | * A TX-descriptor producer (an IRQ) might have gotten |
| 2597 | * inbetween, making the ring free again. Since xmit is |
| 2598 | * serialized, this is the only situation we have to |
| 2599 | * re-test. |
| 2600 | */ |
| 2601 | if (!tx_ring_full(ap, ap->tx_ret_csm, idx)) |
| 2602 | netif_wake_queue(dev); |
| 2603 | } |
| 2604 | |
| 2605 | dev->trans_start = jiffies; |
| 2606 | return NETDEV_TX_OK; |
| 2607 | |
| 2608 | overflow: |
| 2609 | /* |
| 2610 | * This race condition is unavoidable with lock-free drivers. |
| 2611 | * We wake up the queue _before_ tx_prd is advanced, so that we can |
| 2612 | * enter hard_start_xmit too early, while tx ring still looks closed. |
| 2613 | * This happens ~1-4 times per 100000 packets, so that we can allow |
| 2614 | * to loop syncing to other CPU. Probably, we need an additional |
| 2615 | * wmb() in ace_tx_intr as well. |
| 2616 | * |
| 2617 | * Note that this race is relieved by reserving one more entry |
| 2618 | * in tx ring than it is necessary (see original non-SG driver). |
| 2619 | * However, with SG we need to reserve 2*MAX_SKB_FRAGS+1, which |
| 2620 | * is already overkill. |
| 2621 | * |
| 2622 | * Alternative is to return with 1 not throttling queue. In this |
| 2623 | * case loop becomes longer, no more useful effects. |
| 2624 | */ |
| 2625 | if (time_before(jiffies, maxjiff)) { |
| 2626 | barrier(); |
| 2627 | cpu_relax(); |
| 2628 | goto restart; |
| 2629 | } |
| 2630 | |
| 2631 | /* The ring is stuck full. */ |
| 2632 | printk(KERN_WARNING "%s: Transmit ring stuck full\n", dev->name); |
| 2633 | return NETDEV_TX_BUSY; |
| 2634 | } |
| 2635 | |
| 2636 | |
| 2637 | static int ace_change_mtu(struct net_device *dev, int new_mtu) |
| 2638 | { |
| 2639 | struct ace_private *ap = netdev_priv(dev); |
| 2640 | struct ace_regs __iomem *regs = ap->regs; |
| 2641 | |
| 2642 | if (new_mtu > ACE_JUMBO_MTU) |
| 2643 | return -EINVAL; |
| 2644 | |
| 2645 | writel(new_mtu + ETH_HLEN + 4, ®s->IfMtu); |
| 2646 | dev->mtu = new_mtu; |
| 2647 | |
| 2648 | if (new_mtu > ACE_STD_MTU) { |
| 2649 | if (!(ap->jumbo)) { |
| 2650 | printk(KERN_INFO "%s: Enabling Jumbo frame " |
| 2651 | "support\n", dev->name); |
| 2652 | ap->jumbo = 1; |
| 2653 | if (!test_and_set_bit(0, &ap->jumbo_refill_busy)) |
| 2654 | ace_load_jumbo_rx_ring(ap, RX_JUMBO_SIZE); |
| 2655 | ace_set_rxtx_parms(dev, 1); |
| 2656 | } |
| 2657 | } else { |
| 2658 | while (test_and_set_bit(0, &ap->jumbo_refill_busy)); |
| 2659 | ace_sync_irq(dev->irq); |
| 2660 | ace_set_rxtx_parms(dev, 0); |
| 2661 | if (ap->jumbo) { |
| 2662 | struct cmd cmd; |
| 2663 | |
| 2664 | cmd.evt = C_RESET_JUMBO_RNG; |
| 2665 | cmd.code = 0; |
| 2666 | cmd.idx = 0; |
| 2667 | ace_issue_cmd(regs, &cmd); |
| 2668 | } |
| 2669 | } |
| 2670 | |
| 2671 | return 0; |
| 2672 | } |
| 2673 | |
| 2674 | static int ace_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd) |
| 2675 | { |
| 2676 | struct ace_private *ap = netdev_priv(dev); |
| 2677 | struct ace_regs __iomem *regs = ap->regs; |
| 2678 | u32 link; |
| 2679 | |
| 2680 | memset(ecmd, 0, sizeof(struct ethtool_cmd)); |
| 2681 | ecmd->supported = |
| 2682 | (SUPPORTED_10baseT_Half | SUPPORTED_10baseT_Full | |
| 2683 | SUPPORTED_100baseT_Half | SUPPORTED_100baseT_Full | |
| 2684 | SUPPORTED_1000baseT_Half | SUPPORTED_1000baseT_Full | |
| 2685 | SUPPORTED_Autoneg | SUPPORTED_FIBRE); |
| 2686 | |
| 2687 | ecmd->port = PORT_FIBRE; |
| 2688 | ecmd->transceiver = XCVR_INTERNAL; |
| 2689 | |
| 2690 | link = readl(®s->GigLnkState); |
| 2691 | if (link & LNK_1000MB) |
| 2692 | ecmd->speed = SPEED_1000; |
| 2693 | else { |
| 2694 | link = readl(®s->FastLnkState); |
| 2695 | if (link & LNK_100MB) |
| 2696 | ecmd->speed = SPEED_100; |
| 2697 | else if (link & LNK_10MB) |
| 2698 | ecmd->speed = SPEED_10; |
| 2699 | else |
| 2700 | ecmd->speed = 0; |
| 2701 | } |
| 2702 | if (link & LNK_FULL_DUPLEX) |
| 2703 | ecmd->duplex = DUPLEX_FULL; |
| 2704 | else |
| 2705 | ecmd->duplex = DUPLEX_HALF; |
| 2706 | |
| 2707 | if (link & LNK_NEGOTIATE) |
| 2708 | ecmd->autoneg = AUTONEG_ENABLE; |
| 2709 | else |
| 2710 | ecmd->autoneg = AUTONEG_DISABLE; |
| 2711 | |
| 2712 | #if 0 |
| 2713 | /* |
| 2714 | * Current struct ethtool_cmd is insufficient |
| 2715 | */ |
| 2716 | ecmd->trace = readl(®s->TuneTrace); |
| 2717 | |
| 2718 | ecmd->txcoal = readl(®s->TuneTxCoalTicks); |
| 2719 | ecmd->rxcoal = readl(®s->TuneRxCoalTicks); |
| 2720 | #endif |
| 2721 | ecmd->maxtxpkt = readl(®s->TuneMaxTxDesc); |
| 2722 | ecmd->maxrxpkt = readl(®s->TuneMaxRxDesc); |
| 2723 | |
| 2724 | return 0; |
| 2725 | } |
| 2726 | |
| 2727 | static int ace_set_settings(struct net_device *dev, struct ethtool_cmd *ecmd) |
| 2728 | { |
| 2729 | struct ace_private *ap = netdev_priv(dev); |
| 2730 | struct ace_regs __iomem *regs = ap->regs; |
| 2731 | u32 link, speed; |
| 2732 | |
| 2733 | link = readl(®s->GigLnkState); |
| 2734 | if (link & LNK_1000MB) |
| 2735 | speed = SPEED_1000; |
| 2736 | else { |
| 2737 | link = readl(®s->FastLnkState); |
| 2738 | if (link & LNK_100MB) |
| 2739 | speed = SPEED_100; |
| 2740 | else if (link & LNK_10MB) |
| 2741 | speed = SPEED_10; |
| 2742 | else |
| 2743 | speed = SPEED_100; |
| 2744 | } |
| 2745 | |
| 2746 | link = LNK_ENABLE | LNK_1000MB | LNK_100MB | LNK_10MB | |
| 2747 | LNK_RX_FLOW_CTL_Y | LNK_NEG_FCTL; |
| 2748 | if (!ACE_IS_TIGON_I(ap)) |
| 2749 | link |= LNK_TX_FLOW_CTL_Y; |
| 2750 | if (ecmd->autoneg == AUTONEG_ENABLE) |
| 2751 | link |= LNK_NEGOTIATE; |
| 2752 | if (ecmd->speed != speed) { |
| 2753 | link &= ~(LNK_1000MB | LNK_100MB | LNK_10MB); |
| 2754 | switch (speed) { |
| 2755 | case SPEED_1000: |
| 2756 | link |= LNK_1000MB; |
| 2757 | break; |
| 2758 | case SPEED_100: |
| 2759 | link |= LNK_100MB; |
| 2760 | break; |
| 2761 | case SPEED_10: |
| 2762 | link |= LNK_10MB; |
| 2763 | break; |
| 2764 | } |
| 2765 | } |
| 2766 | |
| 2767 | if (ecmd->duplex == DUPLEX_FULL) |
| 2768 | link |= LNK_FULL_DUPLEX; |
| 2769 | |
| 2770 | if (link != ap->link) { |
| 2771 | struct cmd cmd; |
| 2772 | printk(KERN_INFO "%s: Renegotiating link state\n", |
| 2773 | dev->name); |
| 2774 | |
| 2775 | ap->link = link; |
| 2776 | writel(link, ®s->TuneLink); |
| 2777 | if (!ACE_IS_TIGON_I(ap)) |
| 2778 | writel(link, ®s->TuneFastLink); |
| 2779 | wmb(); |
| 2780 | |
| 2781 | cmd.evt = C_LNK_NEGOTIATION; |
| 2782 | cmd.code = 0; |
| 2783 | cmd.idx = 0; |
| 2784 | ace_issue_cmd(regs, &cmd); |
| 2785 | } |
| 2786 | return 0; |
| 2787 | } |
| 2788 | |
| 2789 | static void ace_get_drvinfo(struct net_device *dev, |
| 2790 | struct ethtool_drvinfo *info) |
| 2791 | { |
| 2792 | struct ace_private *ap = netdev_priv(dev); |
| 2793 | |
| 2794 | strlcpy(info->driver, "acenic", sizeof(info->driver)); |
| 2795 | snprintf(info->version, sizeof(info->version), "%i.%i.%i", |
| 2796 | tigonFwReleaseMajor, tigonFwReleaseMinor, |
| 2797 | tigonFwReleaseFix); |
| 2798 | |
| 2799 | if (ap->pdev) |
| 2800 | strlcpy(info->bus_info, pci_name(ap->pdev), |
| 2801 | sizeof(info->bus_info)); |
| 2802 | |
| 2803 | } |
| 2804 | |
| 2805 | /* |
| 2806 | * Set the hardware MAC address. |
| 2807 | */ |
| 2808 | static int ace_set_mac_addr(struct net_device *dev, void *p) |
| 2809 | { |
| 2810 | struct ace_private *ap = netdev_priv(dev); |
| 2811 | struct ace_regs __iomem *regs = ap->regs; |
| 2812 | struct sockaddr *addr=p; |
| 2813 | u8 *da; |
| 2814 | struct cmd cmd; |
| 2815 | |
| 2816 | if(netif_running(dev)) |
| 2817 | return -EBUSY; |
| 2818 | |
| 2819 | memcpy(dev->dev_addr, addr->sa_data,dev->addr_len); |
| 2820 | |
| 2821 | da = (u8 *)dev->dev_addr; |
| 2822 | |
| 2823 | writel(da[0] << 8 | da[1], ®s->MacAddrHi); |
| 2824 | writel((da[2] << 24) | (da[3] << 16) | (da[4] << 8) | da[5], |
| 2825 | ®s->MacAddrLo); |
| 2826 | |
| 2827 | cmd.evt = C_SET_MAC_ADDR; |
| 2828 | cmd.code = 0; |
| 2829 | cmd.idx = 0; |
| 2830 | ace_issue_cmd(regs, &cmd); |
| 2831 | |
| 2832 | return 0; |
| 2833 | } |
| 2834 | |
| 2835 | |
| 2836 | static void ace_set_multicast_list(struct net_device *dev) |
| 2837 | { |
| 2838 | struct ace_private *ap = netdev_priv(dev); |
| 2839 | struct ace_regs __iomem *regs = ap->regs; |
| 2840 | struct cmd cmd; |
| 2841 | |
| 2842 | if ((dev->flags & IFF_ALLMULTI) && !(ap->mcast_all)) { |
| 2843 | cmd.evt = C_SET_MULTICAST_MODE; |
| 2844 | cmd.code = C_C_MCAST_ENABLE; |
| 2845 | cmd.idx = 0; |
| 2846 | ace_issue_cmd(regs, &cmd); |
| 2847 | ap->mcast_all = 1; |
| 2848 | } else if (ap->mcast_all) { |
| 2849 | cmd.evt = C_SET_MULTICAST_MODE; |
| 2850 | cmd.code = C_C_MCAST_DISABLE; |
| 2851 | cmd.idx = 0; |
| 2852 | ace_issue_cmd(regs, &cmd); |
| 2853 | ap->mcast_all = 0; |
| 2854 | } |
| 2855 | |
| 2856 | if ((dev->flags & IFF_PROMISC) && !(ap->promisc)) { |
| 2857 | cmd.evt = C_SET_PROMISC_MODE; |
| 2858 | cmd.code = C_C_PROMISC_ENABLE; |
| 2859 | cmd.idx = 0; |
| 2860 | ace_issue_cmd(regs, &cmd); |
| 2861 | ap->promisc = 1; |
| 2862 | }else if (!(dev->flags & IFF_PROMISC) && (ap->promisc)) { |
| 2863 | cmd.evt = C_SET_PROMISC_MODE; |
| 2864 | cmd.code = C_C_PROMISC_DISABLE; |
| 2865 | cmd.idx = 0; |
| 2866 | ace_issue_cmd(regs, &cmd); |
| 2867 | ap->promisc = 0; |
| 2868 | } |
| 2869 | |
| 2870 | /* |
| 2871 | * For the time being multicast relies on the upper layers |
| 2872 | * filtering it properly. The Firmware does not allow one to |
| 2873 | * set the entire multicast list at a time and keeping track of |
| 2874 | * it here is going to be messy. |
| 2875 | */ |
| 2876 | if ((dev->mc_count) && !(ap->mcast_all)) { |
| 2877 | cmd.evt = C_SET_MULTICAST_MODE; |
| 2878 | cmd.code = C_C_MCAST_ENABLE; |
| 2879 | cmd.idx = 0; |
| 2880 | ace_issue_cmd(regs, &cmd); |
| 2881 | }else if (!ap->mcast_all) { |
| 2882 | cmd.evt = C_SET_MULTICAST_MODE; |
| 2883 | cmd.code = C_C_MCAST_DISABLE; |
| 2884 | cmd.idx = 0; |
| 2885 | ace_issue_cmd(regs, &cmd); |
| 2886 | } |
| 2887 | } |
| 2888 | |
| 2889 | |
| 2890 | static struct net_device_stats *ace_get_stats(struct net_device *dev) |
| 2891 | { |
| 2892 | struct ace_private *ap = netdev_priv(dev); |
| 2893 | struct ace_mac_stats __iomem *mac_stats = |
| 2894 | (struct ace_mac_stats __iomem *)ap->regs->Stats; |
| 2895 | |
| 2896 | ap->stats.rx_missed_errors = readl(&mac_stats->drop_space); |
| 2897 | ap->stats.multicast = readl(&mac_stats->kept_mc); |
| 2898 | ap->stats.collisions = readl(&mac_stats->coll); |
| 2899 | |
| 2900 | return &ap->stats; |
| 2901 | } |
| 2902 | |
| 2903 | |
| 2904 | static void __devinit ace_copy(struct ace_regs __iomem *regs, void *src, |
| 2905 | u32 dest, int size) |
| 2906 | { |
| 2907 | void __iomem *tdest; |
| 2908 | u32 *wsrc; |
| 2909 | short tsize, i; |
| 2910 | |
| 2911 | if (size <= 0) |
| 2912 | return; |
| 2913 | |
| 2914 | while (size > 0) { |
| 2915 | tsize = min_t(u32, ((~dest & (ACE_WINDOW_SIZE - 1)) + 1), |
| 2916 | min_t(u32, size, ACE_WINDOW_SIZE)); |
| 2917 | tdest = (void __iomem *) ®s->Window + |
| 2918 | (dest & (ACE_WINDOW_SIZE - 1)); |
| 2919 | writel(dest & ~(ACE_WINDOW_SIZE - 1), ®s->WinBase); |
| 2920 | /* |
| 2921 | * This requires byte swapping on big endian, however |
| 2922 | * writel does that for us |
| 2923 | */ |
| 2924 | wsrc = src; |
| 2925 | for (i = 0; i < (tsize / 4); i++) { |
| 2926 | writel(wsrc[i], tdest + i*4); |
| 2927 | } |
| 2928 | dest += tsize; |
| 2929 | src += tsize; |
| 2930 | size -= tsize; |
| 2931 | } |
| 2932 | |
| 2933 | return; |
| 2934 | } |
| 2935 | |
| 2936 | |
| 2937 | static void __devinit ace_clear(struct ace_regs __iomem *regs, u32 dest, int size) |
| 2938 | { |
| 2939 | void __iomem *tdest; |
| 2940 | short tsize = 0, i; |
| 2941 | |
| 2942 | if (size <= 0) |
| 2943 | return; |
| 2944 | |
| 2945 | while (size > 0) { |
| 2946 | tsize = min_t(u32, ((~dest & (ACE_WINDOW_SIZE - 1)) + 1), |
| 2947 | min_t(u32, size, ACE_WINDOW_SIZE)); |
| 2948 | tdest = (void __iomem *) ®s->Window + |
| 2949 | (dest & (ACE_WINDOW_SIZE - 1)); |
| 2950 | writel(dest & ~(ACE_WINDOW_SIZE - 1), ®s->WinBase); |
| 2951 | |
| 2952 | for (i = 0; i < (tsize / 4); i++) { |
| 2953 | writel(0, tdest + i*4); |
| 2954 | } |
| 2955 | |
| 2956 | dest += tsize; |
| 2957 | size -= tsize; |
| 2958 | } |
| 2959 | |
| 2960 | return; |
| 2961 | } |
| 2962 | |
| 2963 | |
| 2964 | /* |
| 2965 | * Download the firmware into the SRAM on the NIC |
| 2966 | * |
| 2967 | * This operation requires the NIC to be halted and is performed with |
| 2968 | * interrupts disabled and with the spinlock hold. |
| 2969 | */ |
| 2970 | int __devinit ace_load_firmware(struct net_device *dev) |
| 2971 | { |
| 2972 | struct ace_private *ap = netdev_priv(dev); |
| 2973 | struct ace_regs __iomem *regs = ap->regs; |
| 2974 | |
| 2975 | if (!(readl(®s->CpuCtrl) & CPU_HALTED)) { |
| 2976 | printk(KERN_ERR "%s: trying to download firmware while the " |
| 2977 | "CPU is running!\n", ap->name); |
| 2978 | return -EFAULT; |
| 2979 | } |
| 2980 | |
| 2981 | /* |
| 2982 | * Do not try to clear more than 512KB or we end up seeing |
| 2983 | * funny things on NICs with only 512KB SRAM |
| 2984 | */ |
| 2985 | ace_clear(regs, 0x2000, 0x80000-0x2000); |
| 2986 | if (ACE_IS_TIGON_I(ap)) { |
| 2987 | ace_copy(regs, tigonFwText, tigonFwTextAddr, tigonFwTextLen); |
| 2988 | ace_copy(regs, tigonFwData, tigonFwDataAddr, tigonFwDataLen); |
| 2989 | ace_copy(regs, tigonFwRodata, tigonFwRodataAddr, |
| 2990 | tigonFwRodataLen); |
| 2991 | ace_clear(regs, tigonFwBssAddr, tigonFwBssLen); |
| 2992 | ace_clear(regs, tigonFwSbssAddr, tigonFwSbssLen); |
| 2993 | }else if (ap->version == 2) { |
| 2994 | ace_clear(regs, tigon2FwBssAddr, tigon2FwBssLen); |
| 2995 | ace_clear(regs, tigon2FwSbssAddr, tigon2FwSbssLen); |
| 2996 | ace_copy(regs, tigon2FwText, tigon2FwTextAddr,tigon2FwTextLen); |
| 2997 | ace_copy(regs, tigon2FwRodata, tigon2FwRodataAddr, |
| 2998 | tigon2FwRodataLen); |
| 2999 | ace_copy(regs, tigon2FwData, tigon2FwDataAddr,tigon2FwDataLen); |
| 3000 | } |
| 3001 | |
| 3002 | return 0; |
| 3003 | } |
| 3004 | |
| 3005 | |
| 3006 | /* |
| 3007 | * The eeprom on the AceNIC is an Atmel i2c EEPROM. |
| 3008 | * |
| 3009 | * Accessing the EEPROM is `interesting' to say the least - don't read |
| 3010 | * this code right after dinner. |
| 3011 | * |
| 3012 | * This is all about black magic and bit-banging the device .... I |
| 3013 | * wonder in what hospital they have put the guy who designed the i2c |
| 3014 | * specs. |
| 3015 | * |
| 3016 | * Oh yes, this is only the beginning! |
| 3017 | * |
| 3018 | * Thanks to Stevarino Webinski for helping tracking down the bugs in the |
| 3019 | * code i2c readout code by beta testing all my hacks. |
| 3020 | */ |
| 3021 | static void __devinit eeprom_start(struct ace_regs __iomem *regs) |
| 3022 | { |
| 3023 | u32 local; |
| 3024 | |
| 3025 | readl(®s->LocalCtrl); |
| 3026 | udelay(ACE_SHORT_DELAY); |
| 3027 | local = readl(®s->LocalCtrl); |
| 3028 | local |= EEPROM_DATA_OUT | EEPROM_WRITE_ENABLE; |
| 3029 | writel(local, ®s->LocalCtrl); |
| 3030 | readl(®s->LocalCtrl); |
| 3031 | mb(); |
| 3032 | udelay(ACE_SHORT_DELAY); |
| 3033 | local |= EEPROM_CLK_OUT; |
| 3034 | writel(local, ®s->LocalCtrl); |
| 3035 | readl(®s->LocalCtrl); |
| 3036 | mb(); |
| 3037 | udelay(ACE_SHORT_DELAY); |
| 3038 | local &= ~EEPROM_DATA_OUT; |
| 3039 | writel(local, ®s->LocalCtrl); |
| 3040 | readl(®s->LocalCtrl); |
| 3041 | mb(); |
| 3042 | udelay(ACE_SHORT_DELAY); |
| 3043 | local &= ~EEPROM_CLK_OUT; |
| 3044 | writel(local, ®s->LocalCtrl); |
| 3045 | readl(®s->LocalCtrl); |
| 3046 | mb(); |
| 3047 | } |
| 3048 | |
| 3049 | |
| 3050 | static void __devinit eeprom_prep(struct ace_regs __iomem *regs, u8 magic) |
| 3051 | { |
| 3052 | short i; |
| 3053 | u32 local; |
| 3054 | |
| 3055 | udelay(ACE_SHORT_DELAY); |
| 3056 | local = readl(®s->LocalCtrl); |
| 3057 | local &= ~EEPROM_DATA_OUT; |
| 3058 | local |= EEPROM_WRITE_ENABLE; |
| 3059 | writel(local, ®s->LocalCtrl); |
| 3060 | readl(®s->LocalCtrl); |
| 3061 | mb(); |
| 3062 | |
| 3063 | for (i = 0; i < 8; i++, magic <<= 1) { |
| 3064 | udelay(ACE_SHORT_DELAY); |
| 3065 | if (magic & 0x80) |
| 3066 | local |= EEPROM_DATA_OUT; |
| 3067 | else |
| 3068 | local &= ~EEPROM_DATA_OUT; |
| 3069 | writel(local, ®s->LocalCtrl); |
| 3070 | readl(®s->LocalCtrl); |
| 3071 | mb(); |
| 3072 | |
| 3073 | udelay(ACE_SHORT_DELAY); |
| 3074 | local |= EEPROM_CLK_OUT; |
| 3075 | writel(local, ®s->LocalCtrl); |
| 3076 | readl(®s->LocalCtrl); |
| 3077 | mb(); |
| 3078 | udelay(ACE_SHORT_DELAY); |
| 3079 | local &= ~(EEPROM_CLK_OUT | EEPROM_DATA_OUT); |
| 3080 | writel(local, ®s->LocalCtrl); |
| 3081 | readl(®s->LocalCtrl); |
| 3082 | mb(); |
| 3083 | } |
| 3084 | } |
| 3085 | |
| 3086 | |
| 3087 | static int __devinit eeprom_check_ack(struct ace_regs __iomem *regs) |
| 3088 | { |
| 3089 | int state; |
| 3090 | u32 local; |
| 3091 | |
| 3092 | local = readl(®s->LocalCtrl); |
| 3093 | local &= ~EEPROM_WRITE_ENABLE; |
| 3094 | writel(local, ®s->LocalCtrl); |
| 3095 | readl(®s->LocalCtrl); |
| 3096 | mb(); |
| 3097 | udelay(ACE_LONG_DELAY); |
| 3098 | local |= EEPROM_CLK_OUT; |
| 3099 | writel(local, ®s->LocalCtrl); |
| 3100 | readl(®s->LocalCtrl); |
| 3101 | mb(); |
| 3102 | udelay(ACE_SHORT_DELAY); |
| 3103 | /* sample data in middle of high clk */ |
| 3104 | state = (readl(®s->LocalCtrl) & EEPROM_DATA_IN) != 0; |
| 3105 | udelay(ACE_SHORT_DELAY); |
| 3106 | mb(); |
| 3107 | writel(readl(®s->LocalCtrl) & ~EEPROM_CLK_OUT, ®s->LocalCtrl); |
| 3108 | readl(®s->LocalCtrl); |
| 3109 | mb(); |
| 3110 | |
| 3111 | return state; |
| 3112 | } |
| 3113 | |
| 3114 | |
| 3115 | static void __devinit eeprom_stop(struct ace_regs __iomem *regs) |
| 3116 | { |
| 3117 | u32 local; |
| 3118 | |
| 3119 | udelay(ACE_SHORT_DELAY); |
| 3120 | local = readl(®s->LocalCtrl); |
| 3121 | local |= EEPROM_WRITE_ENABLE; |
| 3122 | writel(local, ®s->LocalCtrl); |
| 3123 | readl(®s->LocalCtrl); |
| 3124 | mb(); |
| 3125 | udelay(ACE_SHORT_DELAY); |
| 3126 | local &= ~EEPROM_DATA_OUT; |
| 3127 | writel(local, ®s->LocalCtrl); |
| 3128 | readl(®s->LocalCtrl); |
| 3129 | mb(); |
| 3130 | udelay(ACE_SHORT_DELAY); |
| 3131 | local |= EEPROM_CLK_OUT; |
| 3132 | writel(local, ®s->LocalCtrl); |
| 3133 | readl(®s->LocalCtrl); |
| 3134 | mb(); |
| 3135 | udelay(ACE_SHORT_DELAY); |
| 3136 | local |= EEPROM_DATA_OUT; |
| 3137 | writel(local, ®s->LocalCtrl); |
| 3138 | readl(®s->LocalCtrl); |
| 3139 | mb(); |
| 3140 | udelay(ACE_LONG_DELAY); |
| 3141 | local &= ~EEPROM_CLK_OUT; |
| 3142 | writel(local, ®s->LocalCtrl); |
| 3143 | mb(); |
| 3144 | } |
| 3145 | |
| 3146 | |
| 3147 | /* |
| 3148 | * Read a whole byte from the EEPROM. |
| 3149 | */ |
| 3150 | static int __devinit read_eeprom_byte(struct net_device *dev, |
| 3151 | unsigned long offset) |
| 3152 | { |
| 3153 | struct ace_private *ap = netdev_priv(dev); |
| 3154 | struct ace_regs __iomem *regs = ap->regs; |
| 3155 | unsigned long flags; |
| 3156 | u32 local; |
| 3157 | int result = 0; |
| 3158 | short i; |
| 3159 | |
| 3160 | if (!dev) { |
| 3161 | printk(KERN_ERR "No device!\n"); |
| 3162 | result = -ENODEV; |
| 3163 | goto out; |
| 3164 | } |
| 3165 | |
| 3166 | /* |
| 3167 | * Don't take interrupts on this CPU will bit banging |
| 3168 | * the %#%#@$ I2C device |
| 3169 | */ |
| 3170 | local_irq_save(flags); |
| 3171 | |
| 3172 | eeprom_start(regs); |
| 3173 | |
| 3174 | eeprom_prep(regs, EEPROM_WRITE_SELECT); |
| 3175 | if (eeprom_check_ack(regs)) { |
| 3176 | local_irq_restore(flags); |
| 3177 | printk(KERN_ERR "%s: Unable to sync eeprom\n", ap->name); |
| 3178 | result = -EIO; |
| 3179 | goto eeprom_read_error; |
| 3180 | } |
| 3181 | |
| 3182 | eeprom_prep(regs, (offset >> 8) & 0xff); |
| 3183 | if (eeprom_check_ack(regs)) { |
| 3184 | local_irq_restore(flags); |
| 3185 | printk(KERN_ERR "%s: Unable to set address byte 0\n", |
| 3186 | ap->name); |
| 3187 | result = -EIO; |
| 3188 | goto eeprom_read_error; |
| 3189 | } |
| 3190 | |
| 3191 | eeprom_prep(regs, offset & 0xff); |
| 3192 | if (eeprom_check_ack(regs)) { |
| 3193 | local_irq_restore(flags); |
| 3194 | printk(KERN_ERR "%s: Unable to set address byte 1\n", |
| 3195 | ap->name); |
| 3196 | result = -EIO; |
| 3197 | goto eeprom_read_error; |
| 3198 | } |
| 3199 | |
| 3200 | eeprom_start(regs); |
| 3201 | eeprom_prep(regs, EEPROM_READ_SELECT); |
| 3202 | if (eeprom_check_ack(regs)) { |
| 3203 | local_irq_restore(flags); |
| 3204 | printk(KERN_ERR "%s: Unable to set READ_SELECT\n", |
| 3205 | ap->name); |
| 3206 | result = -EIO; |
| 3207 | goto eeprom_read_error; |
| 3208 | } |
| 3209 | |
| 3210 | for (i = 0; i < 8; i++) { |
| 3211 | local = readl(®s->LocalCtrl); |
| 3212 | local &= ~EEPROM_WRITE_ENABLE; |
| 3213 | writel(local, ®s->LocalCtrl); |
| 3214 | readl(®s->LocalCtrl); |
| 3215 | udelay(ACE_LONG_DELAY); |
| 3216 | mb(); |
| 3217 | local |= EEPROM_CLK_OUT; |
| 3218 | writel(local, ®s->LocalCtrl); |
| 3219 | readl(®s->LocalCtrl); |
| 3220 | mb(); |
| 3221 | udelay(ACE_SHORT_DELAY); |
| 3222 | /* sample data mid high clk */ |
| 3223 | result = (result << 1) | |
| 3224 | ((readl(®s->LocalCtrl) & EEPROM_DATA_IN) != 0); |
| 3225 | udelay(ACE_SHORT_DELAY); |
| 3226 | mb(); |
| 3227 | local = readl(®s->LocalCtrl); |
| 3228 | local &= ~EEPROM_CLK_OUT; |
| 3229 | writel(local, ®s->LocalCtrl); |
| 3230 | readl(®s->LocalCtrl); |
| 3231 | udelay(ACE_SHORT_DELAY); |
| 3232 | mb(); |
| 3233 | if (i == 7) { |
| 3234 | local |= EEPROM_WRITE_ENABLE; |
| 3235 | writel(local, ®s->LocalCtrl); |
| 3236 | readl(®s->LocalCtrl); |
| 3237 | mb(); |
| 3238 | udelay(ACE_SHORT_DELAY); |
| 3239 | } |
| 3240 | } |
| 3241 | |
| 3242 | local |= EEPROM_DATA_OUT; |
| 3243 | writel(local, ®s->LocalCtrl); |
| 3244 | readl(®s->LocalCtrl); |
| 3245 | mb(); |
| 3246 | udelay(ACE_SHORT_DELAY); |
| 3247 | writel(readl(®s->LocalCtrl) | EEPROM_CLK_OUT, ®s->LocalCtrl); |
| 3248 | readl(®s->LocalCtrl); |
| 3249 | udelay(ACE_LONG_DELAY); |
| 3250 | writel(readl(®s->LocalCtrl) & ~EEPROM_CLK_OUT, ®s->LocalCtrl); |
| 3251 | readl(®s->LocalCtrl); |
| 3252 | mb(); |
| 3253 | udelay(ACE_SHORT_DELAY); |
| 3254 | eeprom_stop(regs); |
| 3255 | |
| 3256 | local_irq_restore(flags); |
| 3257 | out: |
| 3258 | return result; |
| 3259 | |
| 3260 | eeprom_read_error: |
| 3261 | printk(KERN_ERR "%s: Unable to read eeprom byte 0x%02lx\n", |
| 3262 | ap->name, offset); |
| 3263 | goto out; |
| 3264 | } |
| 3265 | |
| 3266 | |
| 3267 | /* |
| 3268 | * Local variables: |
| 3269 | * compile-command: "gcc -D__SMP__ -D__KERNEL__ -DMODULE -I../../include -Wall -Wstrict-prototypes -O2 -fomit-frame-pointer -pipe -fno-strength-reduce -DMODVERSIONS -include ../../include/linux/modversions.h -c -o acenic.o acenic.c" |
| 3270 | * End: |
| 3271 | */ |