blob: fdbb0d7213b036ee6fa7ef895a307c9894d0b29e [file] [log] [blame]
Brice Goglin0da34b62006-05-23 06:10:15 -04001/*************************************************************************
2 * myri10ge.c: Myricom Myri-10G Ethernet driver.
3 *
4 * Copyright (C) 2005, 2006 Myricom, Inc.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of Myricom, Inc. nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 *
32 * If the eeprom on your board is not recent enough, you will need to get a
33 * newer firmware image at:
34 * http://www.myri.com/scs/download-Myri10GE.html
35 *
36 * Contact Information:
37 * <help@myri.com>
38 * Myricom, Inc., 325N Santa Anita Avenue, Arcadia, CA 91006
39 *************************************************************************/
40
41#include <linux/tcp.h>
42#include <linux/netdevice.h>
43#include <linux/skbuff.h>
44#include <linux/string.h>
45#include <linux/module.h>
46#include <linux/pci.h>
Brice Goglinb10c0662006-06-08 10:25:00 -040047#include <linux/dma-mapping.h>
Brice Goglin0da34b62006-05-23 06:10:15 -040048#include <linux/etherdevice.h>
49#include <linux/if_ether.h>
50#include <linux/if_vlan.h>
51#include <linux/ip.h>
52#include <linux/inet.h>
53#include <linux/in.h>
54#include <linux/ethtool.h>
55#include <linux/firmware.h>
56#include <linux/delay.h>
57#include <linux/version.h>
58#include <linux/timer.h>
59#include <linux/vmalloc.h>
60#include <linux/crc32.h>
61#include <linux/moduleparam.h>
62#include <linux/io.h>
63#include <net/checksum.h>
64#include <asm/byteorder.h>
65#include <asm/io.h>
Brice Goglin0da34b62006-05-23 06:10:15 -040066#include <asm/processor.h>
67#ifdef CONFIG_MTRR
68#include <asm/mtrr.h>
69#endif
70
71#include "myri10ge_mcp.h"
72#include "myri10ge_mcp_gen_header.h"
73
Brice Goglinb10c0662006-06-08 10:25:00 -040074#define MYRI10GE_VERSION_STR "1.0.0"
Brice Goglin0da34b62006-05-23 06:10:15 -040075
76MODULE_DESCRIPTION("Myricom 10G driver (10GbE)");
77MODULE_AUTHOR("Maintainer: help@myri.com");
78MODULE_VERSION(MYRI10GE_VERSION_STR);
79MODULE_LICENSE("Dual BSD/GPL");
80
81#define MYRI10GE_MAX_ETHER_MTU 9014
82
83#define MYRI10GE_ETH_STOPPED 0
84#define MYRI10GE_ETH_STOPPING 1
85#define MYRI10GE_ETH_STARTING 2
86#define MYRI10GE_ETH_RUNNING 3
87#define MYRI10GE_ETH_OPEN_FAILED 4
88
89#define MYRI10GE_EEPROM_STRINGS_SIZE 256
90#define MYRI10GE_MAX_SEND_DESC_TSO ((65536 / 2048) * 2)
91
92#define MYRI10GE_NO_CONFIRM_DATA 0xffffffff
93#define MYRI10GE_NO_RESPONSE_RESULT 0xffffffff
94
95struct myri10ge_rx_buffer_state {
96 struct sk_buff *skb;
97 DECLARE_PCI_UNMAP_ADDR(bus)
98 DECLARE_PCI_UNMAP_LEN(len)
99};
100
101struct myri10ge_tx_buffer_state {
102 struct sk_buff *skb;
103 int last;
104 DECLARE_PCI_UNMAP_ADDR(bus)
105 DECLARE_PCI_UNMAP_LEN(len)
106};
107
108struct myri10ge_cmd {
109 u32 data0;
110 u32 data1;
111 u32 data2;
112};
113
114struct myri10ge_rx_buf {
115 struct mcp_kreq_ether_recv __iomem *lanai; /* lanai ptr for recv ring */
116 u8 __iomem *wc_fifo; /* w/c rx dma addr fifo address */
117 struct mcp_kreq_ether_recv *shadow; /* host shadow of recv ring */
118 struct myri10ge_rx_buffer_state *info;
119 int cnt;
120 int alloc_fail;
121 int mask; /* number of rx slots -1 */
122};
123
124struct myri10ge_tx_buf {
125 struct mcp_kreq_ether_send __iomem *lanai; /* lanai ptr for sendq */
126 u8 __iomem *wc_fifo; /* w/c send fifo address */
127 struct mcp_kreq_ether_send *req_list; /* host shadow of sendq */
128 char *req_bytes;
129 struct myri10ge_tx_buffer_state *info;
130 int mask; /* number of transmit slots -1 */
131 int boundary; /* boundary transmits cannot cross */
132 int req ____cacheline_aligned; /* transmit slots submitted */
133 int pkt_start; /* packets started */
134 int done ____cacheline_aligned; /* transmit slots completed */
135 int pkt_done; /* packets completed */
136};
137
138struct myri10ge_rx_done {
139 struct mcp_slot *entry;
140 dma_addr_t bus;
141 int cnt;
142 int idx;
143};
144
145struct myri10ge_priv {
146 int running; /* running? */
147 int csum_flag; /* rx_csums? */
148 struct myri10ge_tx_buf tx; /* transmit ring */
149 struct myri10ge_rx_buf rx_small;
150 struct myri10ge_rx_buf rx_big;
151 struct myri10ge_rx_done rx_done;
152 int small_bytes;
153 struct net_device *dev;
154 struct net_device_stats stats;
155 u8 __iomem *sram;
156 int sram_size;
157 unsigned long board_span;
158 unsigned long iomem_base;
159 u32 __iomem *irq_claim;
160 u32 __iomem *irq_deassert;
161 char *mac_addr_string;
162 struct mcp_cmd_response *cmd;
163 dma_addr_t cmd_bus;
164 struct mcp_irq_data *fw_stats;
165 dma_addr_t fw_stats_bus;
166 struct pci_dev *pdev;
167 int msi_enabled;
168 unsigned int link_state;
169 unsigned int rdma_tags_available;
170 int intr_coal_delay;
171 u32 __iomem *intr_coal_delay_ptr;
172 int mtrr;
173 int wake_queue;
174 int stop_queue;
175 int down_cnt;
176 wait_queue_head_t down_wq;
177 struct work_struct watchdog_work;
178 struct timer_list watchdog_timer;
179 int watchdog_tx_done;
Brice Goglinc54772e2006-07-30 00:14:15 -0400180 int watchdog_tx_req;
Brice Goglin0da34b62006-05-23 06:10:15 -0400181 int watchdog_resets;
182 int tx_linearized;
183 int pause;
184 char *fw_name;
185 char eeprom_strings[MYRI10GE_EEPROM_STRINGS_SIZE];
186 char fw_version[128];
187 u8 mac_addr[6]; /* eeprom mac address */
188 unsigned long serial_number;
189 int vendor_specific_offset;
Brice Goglin85a7ea12006-08-21 17:36:56 -0400190 int fw_multicast_support;
Brice Goglin0da34b62006-05-23 06:10:15 -0400191 u32 devctl;
192 u16 msi_flags;
Brice Goglin0da34b62006-05-23 06:10:15 -0400193 u32 read_dma;
194 u32 write_dma;
195 u32 read_write_dma;
Brice Goglinc58ac5c2006-08-21 17:36:49 -0400196 u32 link_changes;
197 u32 msg_enable;
Brice Goglin0da34b62006-05-23 06:10:15 -0400198};
199
200static char *myri10ge_fw_unaligned = "myri10ge_ethp_z8e.dat";
201static char *myri10ge_fw_aligned = "myri10ge_eth_z8e.dat";
202
203static char *myri10ge_fw_name = NULL;
204module_param(myri10ge_fw_name, charp, S_IRUGO | S_IWUSR);
205MODULE_PARM_DESC(myri10ge_fw_name, "Firmware image name\n");
206
207static int myri10ge_ecrc_enable = 1;
208module_param(myri10ge_ecrc_enable, int, S_IRUGO);
209MODULE_PARM_DESC(myri10ge_ecrc_enable, "Enable Extended CRC on PCI-E\n");
210
211static int myri10ge_max_intr_slots = 1024;
212module_param(myri10ge_max_intr_slots, int, S_IRUGO);
213MODULE_PARM_DESC(myri10ge_max_intr_slots, "Interrupt queue slots\n");
214
215static int myri10ge_small_bytes = -1; /* -1 == auto */
216module_param(myri10ge_small_bytes, int, S_IRUGO | S_IWUSR);
217MODULE_PARM_DESC(myri10ge_small_bytes, "Threshold of small packets\n");
218
219static int myri10ge_msi = 1; /* enable msi by default */
220module_param(myri10ge_msi, int, S_IRUGO);
221MODULE_PARM_DESC(myri10ge_msi, "Enable Message Signalled Interrupts\n");
222
223static int myri10ge_intr_coal_delay = 25;
224module_param(myri10ge_intr_coal_delay, int, S_IRUGO);
225MODULE_PARM_DESC(myri10ge_intr_coal_delay, "Interrupt coalescing delay\n");
226
227static int myri10ge_flow_control = 1;
228module_param(myri10ge_flow_control, int, S_IRUGO);
229MODULE_PARM_DESC(myri10ge_flow_control, "Pause parameter\n");
230
231static int myri10ge_deassert_wait = 1;
232module_param(myri10ge_deassert_wait, int, S_IRUGO | S_IWUSR);
233MODULE_PARM_DESC(myri10ge_deassert_wait,
234 "Wait when deasserting legacy interrupts\n");
235
236static int myri10ge_force_firmware = 0;
237module_param(myri10ge_force_firmware, int, S_IRUGO);
238MODULE_PARM_DESC(myri10ge_force_firmware,
239 "Force firmware to assume aligned completions\n");
240
241static int myri10ge_skb_cross_4k = 0;
242module_param(myri10ge_skb_cross_4k, int, S_IRUGO | S_IWUSR);
243MODULE_PARM_DESC(myri10ge_skb_cross_4k,
244 "Can a small skb cross a 4KB boundary?\n");
245
246static int myri10ge_initial_mtu = MYRI10GE_MAX_ETHER_MTU - ETH_HLEN;
247module_param(myri10ge_initial_mtu, int, S_IRUGO);
248MODULE_PARM_DESC(myri10ge_initial_mtu, "Initial MTU\n");
249
250static int myri10ge_napi_weight = 64;
251module_param(myri10ge_napi_weight, int, S_IRUGO);
252MODULE_PARM_DESC(myri10ge_napi_weight, "Set NAPI weight\n");
253
254static int myri10ge_watchdog_timeout = 1;
255module_param(myri10ge_watchdog_timeout, int, S_IRUGO);
256MODULE_PARM_DESC(myri10ge_watchdog_timeout, "Set watchdog timeout\n");
257
258static int myri10ge_max_irq_loops = 1048576;
259module_param(myri10ge_max_irq_loops, int, S_IRUGO);
260MODULE_PARM_DESC(myri10ge_max_irq_loops,
261 "Set stuck legacy IRQ detection threshold\n");
262
Brice Goglinc58ac5c2006-08-21 17:36:49 -0400263#define MYRI10GE_MSG_DEFAULT NETIF_MSG_LINK
264
265static int myri10ge_debug = -1; /* defaults above */
266module_param(myri10ge_debug, int, 0);
267MODULE_PARM_DESC(myri10ge_debug, "Debug level (0=none,...,16=all)");
268
Brice Goglin0da34b62006-05-23 06:10:15 -0400269#define MYRI10GE_FW_OFFSET 1024*1024
270#define MYRI10GE_HIGHPART_TO_U32(X) \
271(sizeof (X) == 8) ? ((u32)((u64)(X) >> 32)) : (0)
272#define MYRI10GE_LOWPART_TO_U32(X) ((u32)(X))
273
274#define myri10ge_pio_copy(to,from,size) __iowrite64_copy(to,from,size/8)
275
276static int
277myri10ge_send_cmd(struct myri10ge_priv *mgp, u32 cmd,
278 struct myri10ge_cmd *data, int atomic)
279{
280 struct mcp_cmd *buf;
281 char buf_bytes[sizeof(*buf) + 8];
282 struct mcp_cmd_response *response = mgp->cmd;
Brice Gogline700f9f2006-08-14 17:52:54 -0400283 char __iomem *cmd_addr = mgp->sram + MXGEFW_ETH_CMD;
Brice Goglin0da34b62006-05-23 06:10:15 -0400284 u32 dma_low, dma_high, result, value;
285 int sleep_total = 0;
286
287 /* ensure buf is aligned to 8 bytes */
288 buf = (struct mcp_cmd *)ALIGN((unsigned long)buf_bytes, 8);
289
290 buf->data0 = htonl(data->data0);
291 buf->data1 = htonl(data->data1);
292 buf->data2 = htonl(data->data2);
293 buf->cmd = htonl(cmd);
294 dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
295 dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
296
297 buf->response_addr.low = htonl(dma_low);
298 buf->response_addr.high = htonl(dma_high);
299 response->result = MYRI10GE_NO_RESPONSE_RESULT;
300 mb();
301 myri10ge_pio_copy(cmd_addr, buf, sizeof(*buf));
302
303 /* wait up to 15ms. Longest command is the DMA benchmark,
304 * which is capped at 5ms, but runs from a timeout handler
305 * that runs every 7.8ms. So a 15ms timeout leaves us with
306 * a 2.2ms margin
307 */
308 if (atomic) {
309 /* if atomic is set, do not sleep,
310 * and try to get the completion quickly
311 * (1ms will be enough for those commands) */
312 for (sleep_total = 0;
313 sleep_total < 1000
314 && response->result == MYRI10GE_NO_RESPONSE_RESULT;
315 sleep_total += 10)
316 udelay(10);
317 } else {
318 /* use msleep for most command */
319 for (sleep_total = 0;
320 sleep_total < 15
321 && response->result == MYRI10GE_NO_RESPONSE_RESULT;
322 sleep_total++)
323 msleep(1);
324 }
325
326 result = ntohl(response->result);
327 value = ntohl(response->data);
328 if (result != MYRI10GE_NO_RESPONSE_RESULT) {
329 if (result == 0) {
330 data->data0 = value;
331 return 0;
Brice Goglin85a7ea12006-08-21 17:36:56 -0400332 } else if (result == MXGEFW_CMD_UNKNOWN) {
333 return -ENOSYS;
Brice Goglin0da34b62006-05-23 06:10:15 -0400334 } else {
335 dev_err(&mgp->pdev->dev,
336 "command %d failed, result = %d\n",
337 cmd, result);
338 return -ENXIO;
339 }
340 }
341
342 dev_err(&mgp->pdev->dev, "command %d timed out, result = %d\n",
343 cmd, result);
344 return -EAGAIN;
345}
346
347/*
348 * The eeprom strings on the lanaiX have the format
349 * SN=x\0
350 * MAC=x:x:x:x:x:x\0
351 * PT:ddd mmm xx xx:xx:xx xx\0
352 * PV:ddd mmm xx xx:xx:xx xx\0
353 */
354static int myri10ge_read_mac_addr(struct myri10ge_priv *mgp)
355{
356 char *ptr, *limit;
357 int i;
358
359 ptr = mgp->eeprom_strings;
360 limit = mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE;
361
362 while (*ptr != '\0' && ptr < limit) {
363 if (memcmp(ptr, "MAC=", 4) == 0) {
364 ptr += 4;
365 mgp->mac_addr_string = ptr;
366 for (i = 0; i < 6; i++) {
367 if ((ptr + 2) > limit)
368 goto abort;
369 mgp->mac_addr[i] =
370 simple_strtoul(ptr, &ptr, 16);
371 ptr += 1;
372 }
373 }
374 if (memcmp((const void *)ptr, "SN=", 3) == 0) {
375 ptr += 3;
376 mgp->serial_number = simple_strtoul(ptr, &ptr, 10);
377 }
378 while (ptr < limit && *ptr++) ;
379 }
380
381 return 0;
382
383abort:
384 dev_err(&mgp->pdev->dev, "failed to parse eeprom_strings\n");
385 return -ENXIO;
386}
387
388/*
389 * Enable or disable periodic RDMAs from the host to make certain
390 * chipsets resend dropped PCIe messages
391 */
392
393static void myri10ge_dummy_rdma(struct myri10ge_priv *mgp, int enable)
394{
395 char __iomem *submit;
396 u32 buf[16];
397 u32 dma_low, dma_high;
398 int i;
399
400 /* clear confirmation addr */
401 mgp->cmd->data = 0;
402 mb();
403
404 /* send a rdma command to the PCIe engine, and wait for the
405 * response in the confirmation address. The firmware should
406 * write a -1 there to indicate it is alive and well
407 */
408 dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
409 dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
410
411 buf[0] = htonl(dma_high); /* confirm addr MSW */
412 buf[1] = htonl(dma_low); /* confirm addr LSW */
413 buf[2] = htonl(MYRI10GE_NO_CONFIRM_DATA); /* confirm data */
414 buf[3] = htonl(dma_high); /* dummy addr MSW */
415 buf[4] = htonl(dma_low); /* dummy addr LSW */
416 buf[5] = htonl(enable); /* enable? */
417
Brice Gogline700f9f2006-08-14 17:52:54 -0400418 submit = mgp->sram + MXGEFW_BOOT_DUMMY_RDMA;
Brice Goglin0da34b62006-05-23 06:10:15 -0400419
420 myri10ge_pio_copy(submit, &buf, sizeof(buf));
421 for (i = 0; mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 20; i++)
422 msleep(1);
423 if (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA)
424 dev_err(&mgp->pdev->dev, "dummy rdma %s failed\n",
425 (enable ? "enable" : "disable"));
426}
427
428static int
429myri10ge_validate_firmware(struct myri10ge_priv *mgp,
430 struct mcp_gen_header *hdr)
431{
432 struct device *dev = &mgp->pdev->dev;
433 int major, minor;
434
435 /* check firmware type */
436 if (ntohl(hdr->mcp_type) != MCP_TYPE_ETH) {
437 dev_err(dev, "Bad firmware type: 0x%x\n", ntohl(hdr->mcp_type));
438 return -EINVAL;
439 }
440
441 /* save firmware version for ethtool */
442 strncpy(mgp->fw_version, hdr->version, sizeof(mgp->fw_version));
443
444 sscanf(mgp->fw_version, "%d.%d", &major, &minor);
445
446 if (!(major == MXGEFW_VERSION_MAJOR && minor == MXGEFW_VERSION_MINOR)) {
447 dev_err(dev, "Found firmware version %s\n", mgp->fw_version);
448 dev_err(dev, "Driver needs %d.%d\n", MXGEFW_VERSION_MAJOR,
449 MXGEFW_VERSION_MINOR);
450 return -EINVAL;
451 }
452 return 0;
453}
454
455static int myri10ge_load_hotplug_firmware(struct myri10ge_priv *mgp, u32 * size)
456{
457 unsigned crc, reread_crc;
458 const struct firmware *fw;
459 struct device *dev = &mgp->pdev->dev;
460 struct mcp_gen_header *hdr;
461 size_t hdr_offset;
462 int status;
Brice Gogline4543582006-07-30 00:14:09 -0400463 unsigned i;
Brice Goglin0da34b62006-05-23 06:10:15 -0400464
465 if ((status = request_firmware(&fw, mgp->fw_name, dev)) < 0) {
466 dev_err(dev, "Unable to load %s firmware image via hotplug\n",
467 mgp->fw_name);
468 status = -EINVAL;
469 goto abort_with_nothing;
470 }
471
472 /* check size */
473
474 if (fw->size >= mgp->sram_size - MYRI10GE_FW_OFFSET ||
475 fw->size < MCP_HEADER_PTR_OFFSET + 4) {
476 dev_err(dev, "Firmware size invalid:%d\n", (int)fw->size);
477 status = -EINVAL;
478 goto abort_with_fw;
479 }
480
481 /* check id */
482 hdr_offset = ntohl(*(u32 *) (fw->data + MCP_HEADER_PTR_OFFSET));
483 if ((hdr_offset & 3) || hdr_offset + sizeof(*hdr) > fw->size) {
484 dev_err(dev, "Bad firmware file\n");
485 status = -EINVAL;
486 goto abort_with_fw;
487 }
488 hdr = (void *)(fw->data + hdr_offset);
489
490 status = myri10ge_validate_firmware(mgp, hdr);
491 if (status != 0)
492 goto abort_with_fw;
493
494 crc = crc32(~0, fw->data, fw->size);
Brice Gogline4543582006-07-30 00:14:09 -0400495 for (i = 0; i < fw->size; i += 256) {
496 myri10ge_pio_copy(mgp->sram + MYRI10GE_FW_OFFSET + i,
497 fw->data + i,
498 min(256U, (unsigned)(fw->size - i)));
499 mb();
500 readb(mgp->sram);
Brice Goglinb10c0662006-06-08 10:25:00 -0400501 }
Brice Goglin0da34b62006-05-23 06:10:15 -0400502 /* corruption checking is good for parity recovery and buggy chipset */
503 memcpy_fromio(fw->data, mgp->sram + MYRI10GE_FW_OFFSET, fw->size);
504 reread_crc = crc32(~0, fw->data, fw->size);
505 if (crc != reread_crc) {
506 dev_err(dev, "CRC failed(fw-len=%u), got 0x%x (expect 0x%x)\n",
507 (unsigned)fw->size, reread_crc, crc);
508 status = -EIO;
509 goto abort_with_fw;
510 }
511 *size = (u32) fw->size;
512
513abort_with_fw:
514 release_firmware(fw);
515
516abort_with_nothing:
517 return status;
518}
519
520static int myri10ge_adopt_running_firmware(struct myri10ge_priv *mgp)
521{
522 struct mcp_gen_header *hdr;
523 struct device *dev = &mgp->pdev->dev;
524 const size_t bytes = sizeof(struct mcp_gen_header);
525 size_t hdr_offset;
526 int status;
527
528 /* find running firmware header */
529 hdr_offset = ntohl(__raw_readl(mgp->sram + MCP_HEADER_PTR_OFFSET));
530
531 if ((hdr_offset & 3) || hdr_offset + sizeof(*hdr) > mgp->sram_size) {
532 dev_err(dev, "Running firmware has bad header offset (%d)\n",
533 (int)hdr_offset);
534 return -EIO;
535 }
536
537 /* copy header of running firmware from SRAM to host memory to
538 * validate firmware */
539 hdr = kmalloc(bytes, GFP_KERNEL);
540 if (hdr == NULL) {
541 dev_err(dev, "could not malloc firmware hdr\n");
542 return -ENOMEM;
543 }
544 memcpy_fromio(hdr, mgp->sram + hdr_offset, bytes);
545 status = myri10ge_validate_firmware(mgp, hdr);
546 kfree(hdr);
547 return status;
548}
549
550static int myri10ge_load_firmware(struct myri10ge_priv *mgp)
551{
552 char __iomem *submit;
553 u32 buf[16];
554 u32 dma_low, dma_high, size;
555 int status, i;
556
Brice Goglinb10c0662006-06-08 10:25:00 -0400557 size = 0;
Brice Goglin0da34b62006-05-23 06:10:15 -0400558 status = myri10ge_load_hotplug_firmware(mgp, &size);
559 if (status) {
560 dev_warn(&mgp->pdev->dev, "hotplug firmware loading failed\n");
561
562 /* Do not attempt to adopt firmware if there
563 * was a bad crc */
564 if (status == -EIO)
565 return status;
566
567 status = myri10ge_adopt_running_firmware(mgp);
568 if (status != 0) {
569 dev_err(&mgp->pdev->dev,
570 "failed to adopt running firmware\n");
571 return status;
572 }
573 dev_info(&mgp->pdev->dev,
574 "Successfully adopted running firmware\n");
575 if (mgp->tx.boundary == 4096) {
576 dev_warn(&mgp->pdev->dev,
577 "Using firmware currently running on NIC"
578 ". For optimal\n");
579 dev_warn(&mgp->pdev->dev,
580 "performance consider loading optimized "
581 "firmware\n");
582 dev_warn(&mgp->pdev->dev, "via hotplug\n");
583 }
584
585 mgp->fw_name = "adopted";
586 mgp->tx.boundary = 2048;
587 return status;
588 }
589
590 /* clear confirmation addr */
591 mgp->cmd->data = 0;
592 mb();
593
594 /* send a reload command to the bootstrap MCP, and wait for the
595 * response in the confirmation address. The firmware should
596 * write a -1 there to indicate it is alive and well
597 */
598 dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
599 dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
600
601 buf[0] = htonl(dma_high); /* confirm addr MSW */
602 buf[1] = htonl(dma_low); /* confirm addr LSW */
603 buf[2] = htonl(MYRI10GE_NO_CONFIRM_DATA); /* confirm data */
604
605 /* FIX: All newest firmware should un-protect the bottom of
606 * the sram before handoff. However, the very first interfaces
607 * do not. Therefore the handoff copy must skip the first 8 bytes
608 */
609 buf[3] = htonl(MYRI10GE_FW_OFFSET + 8); /* where the code starts */
610 buf[4] = htonl(size - 8); /* length of code */
611 buf[5] = htonl(8); /* where to copy to */
612 buf[6] = htonl(0); /* where to jump to */
613
Brice Gogline700f9f2006-08-14 17:52:54 -0400614 submit = mgp->sram + MXGEFW_BOOT_HANDOFF;
Brice Goglin0da34b62006-05-23 06:10:15 -0400615
616 myri10ge_pio_copy(submit, &buf, sizeof(buf));
617 mb();
618 msleep(1);
619 mb();
620 i = 0;
621 while (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 20) {
622 msleep(1);
623 i++;
624 }
625 if (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA) {
626 dev_err(&mgp->pdev->dev, "handoff failed\n");
627 return -ENXIO;
628 }
629 dev_info(&mgp->pdev->dev, "handoff confirmed\n");
Brice Goglin9a71db72006-07-21 15:49:32 -0400630 myri10ge_dummy_rdma(mgp, 1);
Brice Goglin0da34b62006-05-23 06:10:15 -0400631
632 return 0;
633}
634
635static int myri10ge_update_mac_address(struct myri10ge_priv *mgp, u8 * addr)
636{
637 struct myri10ge_cmd cmd;
638 int status;
639
640 cmd.data0 = ((addr[0] << 24) | (addr[1] << 16)
641 | (addr[2] << 8) | addr[3]);
642
643 cmd.data1 = ((addr[4] << 8) | (addr[5]));
644
645 status = myri10ge_send_cmd(mgp, MXGEFW_SET_MAC_ADDRESS, &cmd, 0);
646 return status;
647}
648
649static int myri10ge_change_pause(struct myri10ge_priv *mgp, int pause)
650{
651 struct myri10ge_cmd cmd;
652 int status, ctl;
653
654 ctl = pause ? MXGEFW_ENABLE_FLOW_CONTROL : MXGEFW_DISABLE_FLOW_CONTROL;
655 status = myri10ge_send_cmd(mgp, ctl, &cmd, 0);
656
657 if (status) {
658 printk(KERN_ERR
659 "myri10ge: %s: Failed to set flow control mode\n",
660 mgp->dev->name);
661 return status;
662 }
663 mgp->pause = pause;
664 return 0;
665}
666
667static void
668myri10ge_change_promisc(struct myri10ge_priv *mgp, int promisc, int atomic)
669{
670 struct myri10ge_cmd cmd;
671 int status, ctl;
672
673 ctl = promisc ? MXGEFW_ENABLE_PROMISC : MXGEFW_DISABLE_PROMISC;
674 status = myri10ge_send_cmd(mgp, ctl, &cmd, atomic);
675 if (status)
676 printk(KERN_ERR "myri10ge: %s: Failed to set promisc mode\n",
677 mgp->dev->name);
678}
679
680static int myri10ge_reset(struct myri10ge_priv *mgp)
681{
682 struct myri10ge_cmd cmd;
683 int status;
684 size_t bytes;
685 u32 len;
686
687 /* try to send a reset command to the card to see if it
688 * is alive */
689 memset(&cmd, 0, sizeof(cmd));
690 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_RESET, &cmd, 0);
691 if (status != 0) {
692 dev_err(&mgp->pdev->dev, "failed reset\n");
693 return -ENXIO;
694 }
695
696 /* Now exchange information about interrupts */
697
698 bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
699 memset(mgp->rx_done.entry, 0, bytes);
700 cmd.data0 = (u32) bytes;
701 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_SIZE, &cmd, 0);
702 cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus);
703 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus);
704 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_DMA, &cmd, 0);
705
706 status |=
707 myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_IRQ_ACK_OFFSET, &cmd, 0);
708 mgp->irq_claim = (__iomem u32 *) (mgp->sram + cmd.data0);
709 if (!mgp->msi_enabled) {
710 status |= myri10ge_send_cmd
711 (mgp, MXGEFW_CMD_GET_IRQ_DEASSERT_OFFSET, &cmd, 0);
712 mgp->irq_deassert = (__iomem u32 *) (mgp->sram + cmd.data0);
713
714 }
715 status |= myri10ge_send_cmd
716 (mgp, MXGEFW_CMD_GET_INTR_COAL_DELAY_OFFSET, &cmd, 0);
717 mgp->intr_coal_delay_ptr = (__iomem u32 *) (mgp->sram + cmd.data0);
718 if (status != 0) {
719 dev_err(&mgp->pdev->dev, "failed set interrupt parameters\n");
720 return status;
721 }
722 __raw_writel(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
723
724 /* Run a small DMA test.
725 * The magic multipliers to the length tell the firmware
726 * to do DMA read, write, or read+write tests. The
727 * results are returned in cmd.data0. The upper 16
728 * bits or the return is the number of transfers completed.
729 * The lower 16 bits is the time in 0.5us ticks that the
730 * transfers took to complete.
731 */
732
733 len = mgp->tx.boundary;
734
735 cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus);
736 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus);
737 cmd.data2 = len * 0x10000;
738 status = myri10ge_send_cmd(mgp, MXGEFW_DMA_TEST, &cmd, 0);
739 if (status == 0)
740 mgp->read_dma = ((cmd.data0 >> 16) * len * 2) /
741 (cmd.data0 & 0xffff);
742 else
743 dev_warn(&mgp->pdev->dev, "DMA read benchmark failed: %d\n",
744 status);
745 cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus);
746 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus);
747 cmd.data2 = len * 0x1;
748 status = myri10ge_send_cmd(mgp, MXGEFW_DMA_TEST, &cmd, 0);
749 if (status == 0)
750 mgp->write_dma = ((cmd.data0 >> 16) * len * 2) /
751 (cmd.data0 & 0xffff);
752 else
753 dev_warn(&mgp->pdev->dev, "DMA write benchmark failed: %d\n",
754 status);
755
756 cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus);
757 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus);
758 cmd.data2 = len * 0x10001;
759 status = myri10ge_send_cmd(mgp, MXGEFW_DMA_TEST, &cmd, 0);
760 if (status == 0)
761 mgp->read_write_dma = ((cmd.data0 >> 16) * len * 2 * 2) /
762 (cmd.data0 & 0xffff);
763 else
764 dev_warn(&mgp->pdev->dev,
765 "DMA read/write benchmark failed: %d\n", status);
766
767 memset(mgp->rx_done.entry, 0, bytes);
768
769 /* reset mcp/driver shared state back to 0 */
770 mgp->tx.req = 0;
771 mgp->tx.done = 0;
772 mgp->tx.pkt_start = 0;
773 mgp->tx.pkt_done = 0;
774 mgp->rx_big.cnt = 0;
775 mgp->rx_small.cnt = 0;
776 mgp->rx_done.idx = 0;
777 mgp->rx_done.cnt = 0;
Brice Goglinc58ac5c2006-08-21 17:36:49 -0400778 mgp->link_changes = 0;
Brice Goglin0da34b62006-05-23 06:10:15 -0400779 status = myri10ge_update_mac_address(mgp, mgp->dev->dev_addr);
780 myri10ge_change_promisc(mgp, 0, 0);
781 myri10ge_change_pause(mgp, mgp->pause);
782 return status;
783}
784
785static inline void
786myri10ge_submit_8rx(struct mcp_kreq_ether_recv __iomem * dst,
787 struct mcp_kreq_ether_recv *src)
788{
789 u32 low;
790
791 low = src->addr_low;
792 src->addr_low = DMA_32BIT_MASK;
793 myri10ge_pio_copy(dst, src, 8 * sizeof(*src));
794 mb();
795 src->addr_low = low;
796 __raw_writel(low, &dst->addr_low);
797 mb();
798}
799
800/*
Brice Goglinb10c0662006-06-08 10:25:00 -0400801 * Set of routines to get a new receive buffer. Any buffer which
Brice Goglin0da34b62006-05-23 06:10:15 -0400802 * crosses a 4KB boundary must start on a 4KB boundary due to PCIe
803 * wdma restrictions. We also try to align any smaller allocation to
804 * at least a 16 byte boundary for efficiency. We assume the linux
805 * memory allocator works by powers of 2, and will not return memory
806 * smaller than 2KB which crosses a 4KB boundary. If it does, we fall
807 * back to allocating 2x as much space as required.
808 *
809 * We intend to replace large (>4KB) skb allocations by using
810 * pages directly and building a fraglist in the near future.
811 */
812
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400813static inline struct sk_buff *myri10ge_alloc_big(struct net_device *dev,
814 int bytes)
Brice Goglin0da34b62006-05-23 06:10:15 -0400815{
816 struct sk_buff *skb;
817 unsigned long data, roundup;
818
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400819 skb = netdev_alloc_skb(dev, bytes + 4096 + MXGEFW_PAD);
Brice Goglin0da34b62006-05-23 06:10:15 -0400820 if (skb == NULL)
821 return NULL;
822
823 /* Correct skb->truesize so that socket buffer
824 * accounting is not confused the rounding we must
825 * do to satisfy alignment constraints.
826 */
827 skb->truesize -= 4096;
828
829 data = (unsigned long)(skb->data);
830 roundup = (-data) & (4095);
831 skb_reserve(skb, roundup);
832 return skb;
833}
834
835/* Allocate 2x as much space as required and use whichever portion
836 * does not cross a 4KB boundary */
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400837static inline struct sk_buff *myri10ge_alloc_small_safe(struct net_device *dev,
838 unsigned int bytes)
Brice Goglin0da34b62006-05-23 06:10:15 -0400839{
840 struct sk_buff *skb;
841 unsigned long data, boundary;
842
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400843 skb = netdev_alloc_skb(dev, 2 * (bytes + MXGEFW_PAD) - 1);
Brice Goglin0da34b62006-05-23 06:10:15 -0400844 if (unlikely(skb == NULL))
845 return NULL;
846
847 /* Correct skb->truesize so that socket buffer
848 * accounting is not confused the rounding we must
849 * do to satisfy alignment constraints.
850 */
851 skb->truesize -= bytes + MXGEFW_PAD;
852
853 data = (unsigned long)(skb->data);
854 boundary = (data + 4095UL) & ~4095UL;
855 if ((boundary - data) >= (bytes + MXGEFW_PAD))
856 return skb;
857
858 skb_reserve(skb, boundary - data);
859 return skb;
860}
861
862/* Allocate just enough space, and verify that the allocated
863 * space does not cross a 4KB boundary */
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400864static inline struct sk_buff *myri10ge_alloc_small(struct net_device *dev,
865 int bytes)
Brice Goglin0da34b62006-05-23 06:10:15 -0400866{
867 struct sk_buff *skb;
868 unsigned long roundup, data, end;
869
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400870 skb = netdev_alloc_skb(dev, bytes + 16 + MXGEFW_PAD);
Brice Goglin0da34b62006-05-23 06:10:15 -0400871 if (unlikely(skb == NULL))
872 return NULL;
873
874 /* Round allocated buffer to 16 byte boundary */
875 data = (unsigned long)(skb->data);
876 roundup = (-data) & 15UL;
877 skb_reserve(skb, roundup);
878 /* Verify that the data buffer does not cross a page boundary */
879 data = (unsigned long)(skb->data);
880 end = data + bytes + MXGEFW_PAD - 1;
881 if (unlikely(((end >> 12) != (data >> 12)) && (data & 4095UL))) {
882 printk(KERN_NOTICE
883 "myri10ge_alloc_small: small skb crossed 4KB boundary\n");
884 myri10ge_skb_cross_4k = 1;
885 dev_kfree_skb_any(skb);
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400886 skb = myri10ge_alloc_small_safe(dev, bytes);
Brice Goglin0da34b62006-05-23 06:10:15 -0400887 }
888 return skb;
889}
890
891static inline int
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400892myri10ge_getbuf(struct myri10ge_rx_buf *rx, struct myri10ge_priv *mgp,
893 int bytes, int idx)
Brice Goglin0da34b62006-05-23 06:10:15 -0400894{
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400895 struct net_device *dev = mgp->dev;
896 struct pci_dev *pdev = mgp->pdev;
Brice Goglin0da34b62006-05-23 06:10:15 -0400897 struct sk_buff *skb;
898 dma_addr_t bus;
899 int len, retval = 0;
900
901 bytes += VLAN_HLEN; /* account for 802.1q vlan tag */
902
903 if ((bytes + MXGEFW_PAD) > (4096 - 16) /* linux overhead */ )
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400904 skb = myri10ge_alloc_big(dev, bytes);
Brice Goglin0da34b62006-05-23 06:10:15 -0400905 else if (myri10ge_skb_cross_4k)
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400906 skb = myri10ge_alloc_small_safe(dev, bytes);
Brice Goglin0da34b62006-05-23 06:10:15 -0400907 else
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400908 skb = myri10ge_alloc_small(dev, bytes);
Brice Goglin0da34b62006-05-23 06:10:15 -0400909
910 if (unlikely(skb == NULL)) {
911 rx->alloc_fail++;
912 retval = -ENOBUFS;
913 goto done;
914 }
915
916 /* set len so that it only covers the area we
917 * need mapped for DMA */
918 len = bytes + MXGEFW_PAD;
919
920 bus = pci_map_single(pdev, skb->data, len, PCI_DMA_FROMDEVICE);
921 rx->info[idx].skb = skb;
922 pci_unmap_addr_set(&rx->info[idx], bus, bus);
923 pci_unmap_len_set(&rx->info[idx], len, len);
924 rx->shadow[idx].addr_low = htonl(MYRI10GE_LOWPART_TO_U32(bus));
925 rx->shadow[idx].addr_high = htonl(MYRI10GE_HIGHPART_TO_U32(bus));
926
927done:
928 /* copy 8 descriptors (64-bytes) to the mcp at a time */
929 if ((idx & 7) == 7) {
930 if (rx->wc_fifo == NULL)
931 myri10ge_submit_8rx(&rx->lanai[idx - 7],
932 &rx->shadow[idx - 7]);
933 else {
934 mb();
935 myri10ge_pio_copy(rx->wc_fifo,
936 &rx->shadow[idx - 7], 64);
937 }
938 }
939 return retval;
940}
941
942static inline void myri10ge_vlan_ip_csum(struct sk_buff *skb, u16 hw_csum)
943{
944 struct vlan_hdr *vh = (struct vlan_hdr *)(skb->data);
945
946 if ((skb->protocol == ntohs(ETH_P_8021Q)) &&
947 (vh->h_vlan_encapsulated_proto == htons(ETH_P_IP) ||
948 vh->h_vlan_encapsulated_proto == htons(ETH_P_IPV6))) {
949 skb->csum = hw_csum;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700950 skb->ip_summed = CHECKSUM_COMPLETE;
Brice Goglin0da34b62006-05-23 06:10:15 -0400951 }
952}
953
954static inline unsigned long
955myri10ge_rx_done(struct myri10ge_priv *mgp, struct myri10ge_rx_buf *rx,
956 int bytes, int len, int csum)
957{
958 dma_addr_t bus;
959 struct sk_buff *skb;
960 int idx, unmap_len;
961
962 idx = rx->cnt & rx->mask;
963 rx->cnt++;
964
965 /* save a pointer to the received skb */
966 skb = rx->info[idx].skb;
967 bus = pci_unmap_addr(&rx->info[idx], bus);
968 unmap_len = pci_unmap_len(&rx->info[idx], len);
969
970 /* try to replace the received skb */
Brice Goglinfd6746daa2006-08-14 17:53:15 -0400971 if (myri10ge_getbuf(rx, mgp, bytes, idx)) {
Brice Goglin0da34b62006-05-23 06:10:15 -0400972 /* drop the frame -- the old skbuf is re-cycled */
973 mgp->stats.rx_dropped += 1;
974 return 0;
975 }
976
977 /* unmap the recvd skb */
978 pci_unmap_single(mgp->pdev, bus, unmap_len, PCI_DMA_FROMDEVICE);
979
980 /* mcp implicitly skips 1st bytes so that packet is properly
981 * aligned */
982 skb_reserve(skb, MXGEFW_PAD);
983
984 /* set the length of the frame */
985 skb_put(skb, len);
986
987 skb->protocol = eth_type_trans(skb, mgp->dev);
Brice Goglin0da34b62006-05-23 06:10:15 -0400988 if (mgp->csum_flag) {
989 if ((skb->protocol == ntohs(ETH_P_IP)) ||
990 (skb->protocol == ntohs(ETH_P_IPV6))) {
991 skb->csum = ntohs((u16) csum);
Patrick McHardy84fa7932006-08-29 16:44:56 -0700992 skb->ip_summed = CHECKSUM_COMPLETE;
Brice Goglin0da34b62006-05-23 06:10:15 -0400993 } else
994 myri10ge_vlan_ip_csum(skb, ntohs((u16) csum));
995 }
996
997 netif_receive_skb(skb);
998 mgp->dev->last_rx = jiffies;
999 return 1;
1000}
1001
1002static inline void myri10ge_tx_done(struct myri10ge_priv *mgp, int mcp_index)
1003{
1004 struct pci_dev *pdev = mgp->pdev;
1005 struct myri10ge_tx_buf *tx = &mgp->tx;
1006 struct sk_buff *skb;
1007 int idx, len;
1008 int limit = 0;
1009
1010 while (tx->pkt_done != mcp_index) {
1011 idx = tx->done & tx->mask;
1012 skb = tx->info[idx].skb;
1013
1014 /* Mark as free */
1015 tx->info[idx].skb = NULL;
1016 if (tx->info[idx].last) {
1017 tx->pkt_done++;
1018 tx->info[idx].last = 0;
1019 }
1020 tx->done++;
1021 len = pci_unmap_len(&tx->info[idx], len);
1022 pci_unmap_len_set(&tx->info[idx], len, 0);
1023 if (skb) {
1024 mgp->stats.tx_bytes += skb->len;
1025 mgp->stats.tx_packets++;
1026 dev_kfree_skb_irq(skb);
1027 if (len)
1028 pci_unmap_single(pdev,
1029 pci_unmap_addr(&tx->info[idx],
1030 bus), len,
1031 PCI_DMA_TODEVICE);
1032 } else {
1033 if (len)
1034 pci_unmap_page(pdev,
1035 pci_unmap_addr(&tx->info[idx],
1036 bus), len,
1037 PCI_DMA_TODEVICE);
1038 }
1039
1040 /* limit potential for livelock by only handling
1041 * 2 full tx rings per call */
1042 if (unlikely(++limit > 2 * tx->mask))
1043 break;
1044 }
1045 /* start the queue if we've stopped it */
1046 if (netif_queue_stopped(mgp->dev)
1047 && tx->req - tx->done < (tx->mask >> 1)) {
1048 mgp->wake_queue++;
1049 netif_wake_queue(mgp->dev);
1050 }
1051}
1052
1053static inline void myri10ge_clean_rx_done(struct myri10ge_priv *mgp, int *limit)
1054{
1055 struct myri10ge_rx_done *rx_done = &mgp->rx_done;
1056 unsigned long rx_bytes = 0;
1057 unsigned long rx_packets = 0;
1058 unsigned long rx_ok;
1059
1060 int idx = rx_done->idx;
1061 int cnt = rx_done->cnt;
1062 u16 length;
1063 u16 checksum;
1064
1065 while (rx_done->entry[idx].length != 0 && *limit != 0) {
1066 length = ntohs(rx_done->entry[idx].length);
1067 rx_done->entry[idx].length = 0;
1068 checksum = ntohs(rx_done->entry[idx].checksum);
1069 if (length <= mgp->small_bytes)
1070 rx_ok = myri10ge_rx_done(mgp, &mgp->rx_small,
1071 mgp->small_bytes,
1072 length, checksum);
1073 else
1074 rx_ok = myri10ge_rx_done(mgp, &mgp->rx_big,
1075 mgp->dev->mtu + ETH_HLEN,
1076 length, checksum);
1077 rx_packets += rx_ok;
1078 rx_bytes += rx_ok * (unsigned long)length;
1079 cnt++;
1080 idx = cnt & (myri10ge_max_intr_slots - 1);
1081
1082 /* limit potential for livelock by only handling a
1083 * limited number of frames. */
1084 (*limit)--;
1085 }
1086 rx_done->idx = idx;
1087 rx_done->cnt = cnt;
1088 mgp->stats.rx_packets += rx_packets;
1089 mgp->stats.rx_bytes += rx_bytes;
1090}
1091
1092static inline void myri10ge_check_statblock(struct myri10ge_priv *mgp)
1093{
1094 struct mcp_irq_data *stats = mgp->fw_stats;
1095
1096 if (unlikely(stats->stats_updated)) {
1097 if (mgp->link_state != stats->link_up) {
1098 mgp->link_state = stats->link_up;
1099 if (mgp->link_state) {
Brice Goglinc58ac5c2006-08-21 17:36:49 -04001100 if (netif_msg_link(mgp))
1101 printk(KERN_INFO
1102 "myri10ge: %s: link up\n",
1103 mgp->dev->name);
Brice Goglin0da34b62006-05-23 06:10:15 -04001104 netif_carrier_on(mgp->dev);
Brice Goglinc58ac5c2006-08-21 17:36:49 -04001105 mgp->link_changes++;
Brice Goglin0da34b62006-05-23 06:10:15 -04001106 } else {
Brice Goglinc58ac5c2006-08-21 17:36:49 -04001107 if (netif_msg_link(mgp))
1108 printk(KERN_INFO
1109 "myri10ge: %s: link down\n",
1110 mgp->dev->name);
Brice Goglin0da34b62006-05-23 06:10:15 -04001111 netif_carrier_off(mgp->dev);
Brice Goglinc58ac5c2006-08-21 17:36:49 -04001112 mgp->link_changes++;
Brice Goglin0da34b62006-05-23 06:10:15 -04001113 }
1114 }
1115 if (mgp->rdma_tags_available !=
1116 ntohl(mgp->fw_stats->rdma_tags_available)) {
1117 mgp->rdma_tags_available =
1118 ntohl(mgp->fw_stats->rdma_tags_available);
1119 printk(KERN_WARNING "myri10ge: %s: RDMA timed out! "
1120 "%d tags left\n", mgp->dev->name,
1121 mgp->rdma_tags_available);
1122 }
1123 mgp->down_cnt += stats->link_down;
1124 if (stats->link_down)
1125 wake_up(&mgp->down_wq);
1126 }
1127}
1128
1129static int myri10ge_poll(struct net_device *netdev, int *budget)
1130{
1131 struct myri10ge_priv *mgp = netdev_priv(netdev);
1132 struct myri10ge_rx_done *rx_done = &mgp->rx_done;
1133 int limit, orig_limit, work_done;
1134
1135 /* process as many rx events as NAPI will allow */
1136 limit = min(*budget, netdev->quota);
1137 orig_limit = limit;
1138 myri10ge_clean_rx_done(mgp, &limit);
1139 work_done = orig_limit - limit;
1140 *budget -= work_done;
1141 netdev->quota -= work_done;
1142
1143 if (rx_done->entry[rx_done->idx].length == 0 || !netif_running(netdev)) {
1144 netif_rx_complete(netdev);
1145 __raw_writel(htonl(3), mgp->irq_claim);
1146 return 0;
1147 }
1148 return 1;
1149}
1150
David Howells7d12e782006-10-05 14:55:46 +01001151static irqreturn_t myri10ge_intr(int irq, void *arg)
Brice Goglin0da34b62006-05-23 06:10:15 -04001152{
1153 struct myri10ge_priv *mgp = arg;
1154 struct mcp_irq_data *stats = mgp->fw_stats;
1155 struct myri10ge_tx_buf *tx = &mgp->tx;
1156 u32 send_done_count;
1157 int i;
1158
1159 /* make sure it is our IRQ, and that the DMA has finished */
1160 if (unlikely(!stats->valid))
1161 return (IRQ_NONE);
1162
1163 /* low bit indicates receives are present, so schedule
1164 * napi poll handler */
1165 if (stats->valid & 1)
1166 netif_rx_schedule(mgp->dev);
1167
1168 if (!mgp->msi_enabled) {
1169 __raw_writel(0, mgp->irq_deassert);
1170 if (!myri10ge_deassert_wait)
1171 stats->valid = 0;
1172 mb();
1173 } else
1174 stats->valid = 0;
1175
1176 /* Wait for IRQ line to go low, if using INTx */
1177 i = 0;
1178 while (1) {
1179 i++;
1180 /* check for transmit completes and receives */
1181 send_done_count = ntohl(stats->send_done_count);
1182 if (send_done_count != tx->pkt_done)
1183 myri10ge_tx_done(mgp, (int)send_done_count);
1184 if (unlikely(i > myri10ge_max_irq_loops)) {
1185 printk(KERN_WARNING "myri10ge: %s: irq stuck?\n",
1186 mgp->dev->name);
1187 stats->valid = 0;
1188 schedule_work(&mgp->watchdog_work);
1189 }
1190 if (likely(stats->valid == 0))
1191 break;
1192 cpu_relax();
1193 barrier();
1194 }
1195
1196 myri10ge_check_statblock(mgp);
1197
1198 __raw_writel(htonl(3), mgp->irq_claim + 1);
1199 return (IRQ_HANDLED);
1200}
1201
1202static int
1203myri10ge_get_settings(struct net_device *netdev, struct ethtool_cmd *cmd)
1204{
1205 cmd->autoneg = AUTONEG_DISABLE;
1206 cmd->speed = SPEED_10000;
1207 cmd->duplex = DUPLEX_FULL;
1208 return 0;
1209}
1210
1211static void
1212myri10ge_get_drvinfo(struct net_device *netdev, struct ethtool_drvinfo *info)
1213{
1214 struct myri10ge_priv *mgp = netdev_priv(netdev);
1215
1216 strlcpy(info->driver, "myri10ge", sizeof(info->driver));
1217 strlcpy(info->version, MYRI10GE_VERSION_STR, sizeof(info->version));
1218 strlcpy(info->fw_version, mgp->fw_version, sizeof(info->fw_version));
1219 strlcpy(info->bus_info, pci_name(mgp->pdev), sizeof(info->bus_info));
1220}
1221
1222static int
1223myri10ge_get_coalesce(struct net_device *netdev, struct ethtool_coalesce *coal)
1224{
1225 struct myri10ge_priv *mgp = netdev_priv(netdev);
1226 coal->rx_coalesce_usecs = mgp->intr_coal_delay;
1227 return 0;
1228}
1229
1230static int
1231myri10ge_set_coalesce(struct net_device *netdev, struct ethtool_coalesce *coal)
1232{
1233 struct myri10ge_priv *mgp = netdev_priv(netdev);
1234
1235 mgp->intr_coal_delay = coal->rx_coalesce_usecs;
1236 __raw_writel(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
1237 return 0;
1238}
1239
1240static void
1241myri10ge_get_pauseparam(struct net_device *netdev,
1242 struct ethtool_pauseparam *pause)
1243{
1244 struct myri10ge_priv *mgp = netdev_priv(netdev);
1245
1246 pause->autoneg = 0;
1247 pause->rx_pause = mgp->pause;
1248 pause->tx_pause = mgp->pause;
1249}
1250
1251static int
1252myri10ge_set_pauseparam(struct net_device *netdev,
1253 struct ethtool_pauseparam *pause)
1254{
1255 struct myri10ge_priv *mgp = netdev_priv(netdev);
1256
1257 if (pause->tx_pause != mgp->pause)
1258 return myri10ge_change_pause(mgp, pause->tx_pause);
1259 if (pause->rx_pause != mgp->pause)
1260 return myri10ge_change_pause(mgp, pause->tx_pause);
1261 if (pause->autoneg != 0)
1262 return -EINVAL;
1263 return 0;
1264}
1265
1266static void
1267myri10ge_get_ringparam(struct net_device *netdev,
1268 struct ethtool_ringparam *ring)
1269{
1270 struct myri10ge_priv *mgp = netdev_priv(netdev);
1271
1272 ring->rx_mini_max_pending = mgp->rx_small.mask + 1;
1273 ring->rx_max_pending = mgp->rx_big.mask + 1;
1274 ring->rx_jumbo_max_pending = 0;
1275 ring->tx_max_pending = mgp->rx_small.mask + 1;
1276 ring->rx_mini_pending = ring->rx_mini_max_pending;
1277 ring->rx_pending = ring->rx_max_pending;
1278 ring->rx_jumbo_pending = ring->rx_jumbo_max_pending;
1279 ring->tx_pending = ring->tx_max_pending;
1280}
1281
1282static u32 myri10ge_get_rx_csum(struct net_device *netdev)
1283{
1284 struct myri10ge_priv *mgp = netdev_priv(netdev);
1285 if (mgp->csum_flag)
1286 return 1;
1287 else
1288 return 0;
1289}
1290
1291static int myri10ge_set_rx_csum(struct net_device *netdev, u32 csum_enabled)
1292{
1293 struct myri10ge_priv *mgp = netdev_priv(netdev);
1294 if (csum_enabled)
1295 mgp->csum_flag = MXGEFW_FLAGS_CKSUM;
1296 else
1297 mgp->csum_flag = 0;
1298 return 0;
1299}
1300
1301static const char myri10ge_gstrings_stats[][ETH_GSTRING_LEN] = {
1302 "rx_packets", "tx_packets", "rx_bytes", "tx_bytes", "rx_errors",
1303 "tx_errors", "rx_dropped", "tx_dropped", "multicast", "collisions",
1304 "rx_length_errors", "rx_over_errors", "rx_crc_errors",
1305 "rx_frame_errors", "rx_fifo_errors", "rx_missed_errors",
1306 "tx_aborted_errors", "tx_carrier_errors", "tx_fifo_errors",
1307 "tx_heartbeat_errors", "tx_window_errors",
1308 /* device-specific stats */
Brice Goglin2c1a1082006-07-03 18:16:46 -04001309 "tx_boundary", "WC", "irq", "MSI",
Brice Goglin0da34b62006-05-23 06:10:15 -04001310 "read_dma_bw_MBs", "write_dma_bw_MBs", "read_write_dma_bw_MBs",
1311 "serial_number", "tx_pkt_start", "tx_pkt_done",
1312 "tx_req", "tx_done", "rx_small_cnt", "rx_big_cnt",
1313 "wake_queue", "stop_queue", "watchdog_resets", "tx_linearized",
Brice Goglinc58ac5c2006-08-21 17:36:49 -04001314 "link_changes", "link_up", "dropped_link_overflow",
Brice Goglin85a7ea12006-08-21 17:36:56 -04001315 "dropped_link_error_or_filtered", "dropped_multicast_filtered",
Brice Goglin0da34b62006-05-23 06:10:15 -04001316 "dropped_runt", "dropped_overrun", "dropped_no_small_buffer",
1317 "dropped_no_big_buffer"
1318};
1319
1320#define MYRI10GE_NET_STATS_LEN 21
1321#define MYRI10GE_STATS_LEN sizeof(myri10ge_gstrings_stats) / ETH_GSTRING_LEN
1322
1323static void
1324myri10ge_get_strings(struct net_device *netdev, u32 stringset, u8 * data)
1325{
1326 switch (stringset) {
1327 case ETH_SS_STATS:
1328 memcpy(data, *myri10ge_gstrings_stats,
1329 sizeof(myri10ge_gstrings_stats));
1330 break;
1331 }
1332}
1333
1334static int myri10ge_get_stats_count(struct net_device *netdev)
1335{
1336 return MYRI10GE_STATS_LEN;
1337}
1338
1339static void
1340myri10ge_get_ethtool_stats(struct net_device *netdev,
1341 struct ethtool_stats *stats, u64 * data)
1342{
1343 struct myri10ge_priv *mgp = netdev_priv(netdev);
1344 int i;
1345
1346 for (i = 0; i < MYRI10GE_NET_STATS_LEN; i++)
1347 data[i] = ((unsigned long *)&mgp->stats)[i];
1348
Brice Goglin2c1a1082006-07-03 18:16:46 -04001349 data[i++] = (unsigned int)mgp->tx.boundary;
1350 data[i++] = (unsigned int)(mgp->mtrr >= 0);
1351 data[i++] = (unsigned int)mgp->pdev->irq;
1352 data[i++] = (unsigned int)mgp->msi_enabled;
Brice Goglin0da34b62006-05-23 06:10:15 -04001353 data[i++] = (unsigned int)mgp->read_dma;
1354 data[i++] = (unsigned int)mgp->write_dma;
1355 data[i++] = (unsigned int)mgp->read_write_dma;
1356 data[i++] = (unsigned int)mgp->serial_number;
1357 data[i++] = (unsigned int)mgp->tx.pkt_start;
1358 data[i++] = (unsigned int)mgp->tx.pkt_done;
1359 data[i++] = (unsigned int)mgp->tx.req;
1360 data[i++] = (unsigned int)mgp->tx.done;
1361 data[i++] = (unsigned int)mgp->rx_small.cnt;
1362 data[i++] = (unsigned int)mgp->rx_big.cnt;
1363 data[i++] = (unsigned int)mgp->wake_queue;
1364 data[i++] = (unsigned int)mgp->stop_queue;
1365 data[i++] = (unsigned int)mgp->watchdog_resets;
1366 data[i++] = (unsigned int)mgp->tx_linearized;
Brice Goglinc58ac5c2006-08-21 17:36:49 -04001367 data[i++] = (unsigned int)mgp->link_changes;
Brice Goglin0da34b62006-05-23 06:10:15 -04001368 data[i++] = (unsigned int)ntohl(mgp->fw_stats->link_up);
1369 data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_link_overflow);
1370 data[i++] =
1371 (unsigned int)ntohl(mgp->fw_stats->dropped_link_error_or_filtered);
Brice Goglin85a7ea12006-08-21 17:36:56 -04001372 data[i++] =
1373 (unsigned int)ntohl(mgp->fw_stats->dropped_multicast_filtered);
Brice Goglin0da34b62006-05-23 06:10:15 -04001374 data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_runt);
1375 data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_overrun);
1376 data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_no_small_buffer);
1377 data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_no_big_buffer);
1378}
1379
Brice Goglinc58ac5c2006-08-21 17:36:49 -04001380static void myri10ge_set_msglevel(struct net_device *netdev, u32 value)
1381{
1382 struct myri10ge_priv *mgp = netdev_priv(netdev);
1383 mgp->msg_enable = value;
1384}
1385
1386static u32 myri10ge_get_msglevel(struct net_device *netdev)
1387{
1388 struct myri10ge_priv *mgp = netdev_priv(netdev);
1389 return mgp->msg_enable;
1390}
1391
Jeff Garzik7282d492006-09-13 14:30:00 -04001392static const struct ethtool_ops myri10ge_ethtool_ops = {
Brice Goglin0da34b62006-05-23 06:10:15 -04001393 .get_settings = myri10ge_get_settings,
1394 .get_drvinfo = myri10ge_get_drvinfo,
1395 .get_coalesce = myri10ge_get_coalesce,
1396 .set_coalesce = myri10ge_set_coalesce,
1397 .get_pauseparam = myri10ge_get_pauseparam,
1398 .set_pauseparam = myri10ge_set_pauseparam,
1399 .get_ringparam = myri10ge_get_ringparam,
1400 .get_rx_csum = myri10ge_get_rx_csum,
1401 .set_rx_csum = myri10ge_set_rx_csum,
1402 .get_tx_csum = ethtool_op_get_tx_csum,
Brice Goglinb10c0662006-06-08 10:25:00 -04001403 .set_tx_csum = ethtool_op_set_tx_hw_csum,
Brice Goglin0da34b62006-05-23 06:10:15 -04001404 .get_sg = ethtool_op_get_sg,
1405 .set_sg = ethtool_op_set_sg,
1406#ifdef NETIF_F_TSO
1407 .get_tso = ethtool_op_get_tso,
1408 .set_tso = ethtool_op_set_tso,
1409#endif
1410 .get_strings = myri10ge_get_strings,
1411 .get_stats_count = myri10ge_get_stats_count,
Brice Goglinc58ac5c2006-08-21 17:36:49 -04001412 .get_ethtool_stats = myri10ge_get_ethtool_stats,
1413 .set_msglevel = myri10ge_set_msglevel,
1414 .get_msglevel = myri10ge_get_msglevel
Brice Goglin0da34b62006-05-23 06:10:15 -04001415};
1416
1417static int myri10ge_allocate_rings(struct net_device *dev)
1418{
1419 struct myri10ge_priv *mgp;
1420 struct myri10ge_cmd cmd;
1421 int tx_ring_size, rx_ring_size;
1422 int tx_ring_entries, rx_ring_entries;
1423 int i, status;
1424 size_t bytes;
1425
1426 mgp = netdev_priv(dev);
1427
1428 /* get ring sizes */
1429
1430 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_RING_SIZE, &cmd, 0);
1431 tx_ring_size = cmd.data0;
1432 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_RX_RING_SIZE, &cmd, 0);
1433 rx_ring_size = cmd.data0;
1434
1435 tx_ring_entries = tx_ring_size / sizeof(struct mcp_kreq_ether_send);
1436 rx_ring_entries = rx_ring_size / sizeof(struct mcp_dma_addr);
1437 mgp->tx.mask = tx_ring_entries - 1;
1438 mgp->rx_small.mask = mgp->rx_big.mask = rx_ring_entries - 1;
1439
1440 /* allocate the host shadow rings */
1441
1442 bytes = 8 + (MYRI10GE_MAX_SEND_DESC_TSO + 4)
1443 * sizeof(*mgp->tx.req_list);
1444 mgp->tx.req_bytes = kzalloc(bytes, GFP_KERNEL);
1445 if (mgp->tx.req_bytes == NULL)
1446 goto abort_with_nothing;
1447
1448 /* ensure req_list entries are aligned to 8 bytes */
1449 mgp->tx.req_list = (struct mcp_kreq_ether_send *)
1450 ALIGN((unsigned long)mgp->tx.req_bytes, 8);
1451
1452 bytes = rx_ring_entries * sizeof(*mgp->rx_small.shadow);
1453 mgp->rx_small.shadow = kzalloc(bytes, GFP_KERNEL);
1454 if (mgp->rx_small.shadow == NULL)
1455 goto abort_with_tx_req_bytes;
1456
1457 bytes = rx_ring_entries * sizeof(*mgp->rx_big.shadow);
1458 mgp->rx_big.shadow = kzalloc(bytes, GFP_KERNEL);
1459 if (mgp->rx_big.shadow == NULL)
1460 goto abort_with_rx_small_shadow;
1461
1462 /* allocate the host info rings */
1463
1464 bytes = tx_ring_entries * sizeof(*mgp->tx.info);
1465 mgp->tx.info = kzalloc(bytes, GFP_KERNEL);
1466 if (mgp->tx.info == NULL)
1467 goto abort_with_rx_big_shadow;
1468
1469 bytes = rx_ring_entries * sizeof(*mgp->rx_small.info);
1470 mgp->rx_small.info = kzalloc(bytes, GFP_KERNEL);
1471 if (mgp->rx_small.info == NULL)
1472 goto abort_with_tx_info;
1473
1474 bytes = rx_ring_entries * sizeof(*mgp->rx_big.info);
1475 mgp->rx_big.info = kzalloc(bytes, GFP_KERNEL);
1476 if (mgp->rx_big.info == NULL)
1477 goto abort_with_rx_small_info;
1478
1479 /* Fill the receive rings */
1480
1481 for (i = 0; i <= mgp->rx_small.mask; i++) {
Brice Goglinfd6746daa2006-08-14 17:53:15 -04001482 status = myri10ge_getbuf(&mgp->rx_small, mgp,
Brice Goglin0da34b62006-05-23 06:10:15 -04001483 mgp->small_bytes, i);
1484 if (status) {
1485 printk(KERN_ERR
1486 "myri10ge: %s: alloced only %d small bufs\n",
1487 dev->name, i);
1488 goto abort_with_rx_small_ring;
1489 }
1490 }
1491
1492 for (i = 0; i <= mgp->rx_big.mask; i++) {
1493 status =
Brice Goglinfd6746daa2006-08-14 17:53:15 -04001494 myri10ge_getbuf(&mgp->rx_big, mgp, dev->mtu + ETH_HLEN, i);
Brice Goglin0da34b62006-05-23 06:10:15 -04001495 if (status) {
1496 printk(KERN_ERR
1497 "myri10ge: %s: alloced only %d big bufs\n",
1498 dev->name, i);
1499 goto abort_with_rx_big_ring;
1500 }
1501 }
1502
1503 return 0;
1504
1505abort_with_rx_big_ring:
1506 for (i = 0; i <= mgp->rx_big.mask; i++) {
1507 if (mgp->rx_big.info[i].skb != NULL)
1508 dev_kfree_skb_any(mgp->rx_big.info[i].skb);
1509 if (pci_unmap_len(&mgp->rx_big.info[i], len))
1510 pci_unmap_single(mgp->pdev,
1511 pci_unmap_addr(&mgp->rx_big.info[i],
1512 bus),
1513 pci_unmap_len(&mgp->rx_big.info[i],
1514 len),
1515 PCI_DMA_FROMDEVICE);
1516 }
1517
1518abort_with_rx_small_ring:
1519 for (i = 0; i <= mgp->rx_small.mask; i++) {
1520 if (mgp->rx_small.info[i].skb != NULL)
1521 dev_kfree_skb_any(mgp->rx_small.info[i].skb);
1522 if (pci_unmap_len(&mgp->rx_small.info[i], len))
1523 pci_unmap_single(mgp->pdev,
1524 pci_unmap_addr(&mgp->rx_small.info[i],
1525 bus),
1526 pci_unmap_len(&mgp->rx_small.info[i],
1527 len),
1528 PCI_DMA_FROMDEVICE);
1529 }
1530 kfree(mgp->rx_big.info);
1531
1532abort_with_rx_small_info:
1533 kfree(mgp->rx_small.info);
1534
1535abort_with_tx_info:
1536 kfree(mgp->tx.info);
1537
1538abort_with_rx_big_shadow:
1539 kfree(mgp->rx_big.shadow);
1540
1541abort_with_rx_small_shadow:
1542 kfree(mgp->rx_small.shadow);
1543
1544abort_with_tx_req_bytes:
1545 kfree(mgp->tx.req_bytes);
1546 mgp->tx.req_bytes = NULL;
1547 mgp->tx.req_list = NULL;
1548
1549abort_with_nothing:
1550 return status;
1551}
1552
1553static void myri10ge_free_rings(struct net_device *dev)
1554{
1555 struct myri10ge_priv *mgp;
1556 struct sk_buff *skb;
1557 struct myri10ge_tx_buf *tx;
1558 int i, len, idx;
1559
1560 mgp = netdev_priv(dev);
1561
1562 for (i = 0; i <= mgp->rx_big.mask; i++) {
1563 if (mgp->rx_big.info[i].skb != NULL)
1564 dev_kfree_skb_any(mgp->rx_big.info[i].skb);
1565 if (pci_unmap_len(&mgp->rx_big.info[i], len))
1566 pci_unmap_single(mgp->pdev,
1567 pci_unmap_addr(&mgp->rx_big.info[i],
1568 bus),
1569 pci_unmap_len(&mgp->rx_big.info[i],
1570 len),
1571 PCI_DMA_FROMDEVICE);
1572 }
1573
1574 for (i = 0; i <= mgp->rx_small.mask; i++) {
1575 if (mgp->rx_small.info[i].skb != NULL)
1576 dev_kfree_skb_any(mgp->rx_small.info[i].skb);
1577 if (pci_unmap_len(&mgp->rx_small.info[i], len))
1578 pci_unmap_single(mgp->pdev,
1579 pci_unmap_addr(&mgp->rx_small.info[i],
1580 bus),
1581 pci_unmap_len(&mgp->rx_small.info[i],
1582 len),
1583 PCI_DMA_FROMDEVICE);
1584 }
1585
1586 tx = &mgp->tx;
1587 while (tx->done != tx->req) {
1588 idx = tx->done & tx->mask;
1589 skb = tx->info[idx].skb;
1590
1591 /* Mark as free */
1592 tx->info[idx].skb = NULL;
1593 tx->done++;
1594 len = pci_unmap_len(&tx->info[idx], len);
1595 pci_unmap_len_set(&tx->info[idx], len, 0);
1596 if (skb) {
1597 mgp->stats.tx_dropped++;
1598 dev_kfree_skb_any(skb);
1599 if (len)
1600 pci_unmap_single(mgp->pdev,
1601 pci_unmap_addr(&tx->info[idx],
1602 bus), len,
1603 PCI_DMA_TODEVICE);
1604 } else {
1605 if (len)
1606 pci_unmap_page(mgp->pdev,
1607 pci_unmap_addr(&tx->info[idx],
1608 bus), len,
1609 PCI_DMA_TODEVICE);
1610 }
1611 }
1612 kfree(mgp->rx_big.info);
1613
1614 kfree(mgp->rx_small.info);
1615
1616 kfree(mgp->tx.info);
1617
1618 kfree(mgp->rx_big.shadow);
1619
1620 kfree(mgp->rx_small.shadow);
1621
1622 kfree(mgp->tx.req_bytes);
1623 mgp->tx.req_bytes = NULL;
1624 mgp->tx.req_list = NULL;
1625}
1626
1627static int myri10ge_open(struct net_device *dev)
1628{
1629 struct myri10ge_priv *mgp;
1630 struct myri10ge_cmd cmd;
1631 int status, big_pow2;
1632
1633 mgp = netdev_priv(dev);
1634
1635 if (mgp->running != MYRI10GE_ETH_STOPPED)
1636 return -EBUSY;
1637
1638 mgp->running = MYRI10GE_ETH_STARTING;
1639 status = myri10ge_reset(mgp);
1640 if (status != 0) {
1641 printk(KERN_ERR "myri10ge: %s: failed reset\n", dev->name);
1642 mgp->running = MYRI10GE_ETH_STOPPED;
1643 return -ENXIO;
1644 }
1645
1646 /* decide what small buffer size to use. For good TCP rx
1647 * performance, it is important to not receive 1514 byte
1648 * frames into jumbo buffers, as it confuses the socket buffer
1649 * accounting code, leading to drops and erratic performance.
1650 */
1651
1652 if (dev->mtu <= ETH_DATA_LEN)
1653 mgp->small_bytes = 128; /* enough for a TCP header */
1654 else
1655 mgp->small_bytes = ETH_FRAME_LEN; /* enough for an ETH_DATA_LEN frame */
1656
1657 /* Override the small buffer size? */
1658 if (myri10ge_small_bytes > 0)
1659 mgp->small_bytes = myri10ge_small_bytes;
1660
1661 /* If the user sets an obscenely small MTU, adjust the small
1662 * bytes down to nearly nothing */
1663 if (mgp->small_bytes >= (dev->mtu + ETH_HLEN))
1664 mgp->small_bytes = 64;
1665
1666 /* get the lanai pointers to the send and receive rings */
1667
1668 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_OFFSET, &cmd, 0);
1669 mgp->tx.lanai =
1670 (struct mcp_kreq_ether_send __iomem *)(mgp->sram + cmd.data0);
1671
1672 status |=
1673 myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SMALL_RX_OFFSET, &cmd, 0);
1674 mgp->rx_small.lanai =
1675 (struct mcp_kreq_ether_recv __iomem *)(mgp->sram + cmd.data0);
1676
1677 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_BIG_RX_OFFSET, &cmd, 0);
1678 mgp->rx_big.lanai =
1679 (struct mcp_kreq_ether_recv __iomem *)(mgp->sram + cmd.data0);
1680
1681 if (status != 0) {
1682 printk(KERN_ERR
1683 "myri10ge: %s: failed to get ring sizes or locations\n",
1684 dev->name);
1685 mgp->running = MYRI10GE_ETH_STOPPED;
1686 return -ENXIO;
1687 }
1688
1689 if (mgp->mtrr >= 0) {
Brice Gogline700f9f2006-08-14 17:52:54 -04001690 mgp->tx.wc_fifo = (u8 __iomem *) mgp->sram + MXGEFW_ETH_SEND_4;
1691 mgp->rx_small.wc_fifo =
1692 (u8 __iomem *) mgp->sram + MXGEFW_ETH_RECV_SMALL;
1693 mgp->rx_big.wc_fifo =
1694 (u8 __iomem *) mgp->sram + MXGEFW_ETH_RECV_BIG;
Brice Goglin0da34b62006-05-23 06:10:15 -04001695 } else {
1696 mgp->tx.wc_fifo = NULL;
1697 mgp->rx_small.wc_fifo = NULL;
1698 mgp->rx_big.wc_fifo = NULL;
1699 }
1700
1701 status = myri10ge_allocate_rings(dev);
1702 if (status != 0)
1703 goto abort_with_nothing;
1704
1705 /* Firmware needs the big buff size as a power of 2. Lie and
1706 * tell him the buffer is larger, because we only use 1
1707 * buffer/pkt, and the mtu will prevent overruns.
1708 */
1709 big_pow2 = dev->mtu + ETH_HLEN + MXGEFW_PAD;
1710 while ((big_pow2 & (big_pow2 - 1)) != 0)
1711 big_pow2++;
1712
1713 /* now give firmware buffers sizes, and MTU */
1714 cmd.data0 = dev->mtu + ETH_HLEN + VLAN_HLEN;
1715 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_MTU, &cmd, 0);
1716 cmd.data0 = mgp->small_bytes;
1717 status |=
1718 myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_SMALL_BUFFER_SIZE, &cmd, 0);
1719 cmd.data0 = big_pow2;
1720 status |=
1721 myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_BIG_BUFFER_SIZE, &cmd, 0);
1722 if (status) {
1723 printk(KERN_ERR "myri10ge: %s: Couldn't set buffer sizes\n",
1724 dev->name);
1725 goto abort_with_rings;
1726 }
1727
1728 cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->fw_stats_bus);
1729 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->fw_stats_bus);
Brice Goglin85a7ea12006-08-21 17:36:56 -04001730 cmd.data2 = sizeof(struct mcp_irq_data);
1731 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_STATS_DMA_V2, &cmd, 0);
1732 if (status == -ENOSYS) {
1733 dma_addr_t bus = mgp->fw_stats_bus;
1734 bus += offsetof(struct mcp_irq_data, send_done_count);
1735 cmd.data0 = MYRI10GE_LOWPART_TO_U32(bus);
1736 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(bus);
1737 status = myri10ge_send_cmd(mgp,
1738 MXGEFW_CMD_SET_STATS_DMA_OBSOLETE,
1739 &cmd, 0);
1740 /* Firmware cannot support multicast without STATS_DMA_V2 */
1741 mgp->fw_multicast_support = 0;
1742 } else {
1743 mgp->fw_multicast_support = 1;
1744 }
Brice Goglin0da34b62006-05-23 06:10:15 -04001745 if (status) {
1746 printk(KERN_ERR "myri10ge: %s: Couldn't set stats DMA\n",
1747 dev->name);
1748 goto abort_with_rings;
1749 }
1750
1751 mgp->link_state = -1;
1752 mgp->rdma_tags_available = 15;
1753
1754 netif_poll_enable(mgp->dev); /* must happen prior to any irq */
1755
1756 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_UP, &cmd, 0);
1757 if (status) {
1758 printk(KERN_ERR "myri10ge: %s: Couldn't bring up link\n",
1759 dev->name);
1760 goto abort_with_rings;
1761 }
1762
1763 mgp->wake_queue = 0;
1764 mgp->stop_queue = 0;
1765 mgp->running = MYRI10GE_ETH_RUNNING;
1766 mgp->watchdog_timer.expires = jiffies + myri10ge_watchdog_timeout * HZ;
1767 add_timer(&mgp->watchdog_timer);
1768 netif_wake_queue(dev);
1769 return 0;
1770
1771abort_with_rings:
1772 myri10ge_free_rings(dev);
1773
1774abort_with_nothing:
1775 mgp->running = MYRI10GE_ETH_STOPPED;
1776 return -ENOMEM;
1777}
1778
1779static int myri10ge_close(struct net_device *dev)
1780{
1781 struct myri10ge_priv *mgp;
1782 struct myri10ge_cmd cmd;
1783 int status, old_down_cnt;
1784
1785 mgp = netdev_priv(dev);
1786
1787 if (mgp->running != MYRI10GE_ETH_RUNNING)
1788 return 0;
1789
1790 if (mgp->tx.req_bytes == NULL)
1791 return 0;
1792
1793 del_timer_sync(&mgp->watchdog_timer);
1794 mgp->running = MYRI10GE_ETH_STOPPING;
1795 netif_poll_disable(mgp->dev);
1796 netif_carrier_off(dev);
1797 netif_stop_queue(dev);
1798 old_down_cnt = mgp->down_cnt;
1799 mb();
1800 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_DOWN, &cmd, 0);
1801 if (status)
1802 printk(KERN_ERR "myri10ge: %s: Couldn't bring down link\n",
1803 dev->name);
1804
1805 wait_event_timeout(mgp->down_wq, old_down_cnt != mgp->down_cnt, HZ);
1806 if (old_down_cnt == mgp->down_cnt)
1807 printk(KERN_ERR "myri10ge: %s never got down irq\n", dev->name);
1808
1809 netif_tx_disable(dev);
1810
1811 myri10ge_free_rings(dev);
1812
1813 mgp->running = MYRI10GE_ETH_STOPPED;
1814 return 0;
1815}
1816
1817/* copy an array of struct mcp_kreq_ether_send's to the mcp. Copy
1818 * backwards one at a time and handle ring wraps */
1819
1820static inline void
1821myri10ge_submit_req_backwards(struct myri10ge_tx_buf *tx,
1822 struct mcp_kreq_ether_send *src, int cnt)
1823{
1824 int idx, starting_slot;
1825 starting_slot = tx->req;
1826 while (cnt > 1) {
1827 cnt--;
1828 idx = (starting_slot + cnt) & tx->mask;
1829 myri10ge_pio_copy(&tx->lanai[idx], &src[cnt], sizeof(*src));
1830 mb();
1831 }
1832}
1833
1834/*
1835 * copy an array of struct mcp_kreq_ether_send's to the mcp. Copy
1836 * at most 32 bytes at a time, so as to avoid involving the software
1837 * pio handler in the nic. We re-write the first segment's flags
1838 * to mark them valid only after writing the entire chain.
1839 */
1840
1841static inline void
1842myri10ge_submit_req(struct myri10ge_tx_buf *tx, struct mcp_kreq_ether_send *src,
1843 int cnt)
1844{
1845 int idx, i;
1846 struct mcp_kreq_ether_send __iomem *dstp, *dst;
1847 struct mcp_kreq_ether_send *srcp;
1848 u8 last_flags;
1849
1850 idx = tx->req & tx->mask;
1851
1852 last_flags = src->flags;
1853 src->flags = 0;
1854 mb();
1855 dst = dstp = &tx->lanai[idx];
1856 srcp = src;
1857
1858 if ((idx + cnt) < tx->mask) {
1859 for (i = 0; i < (cnt - 1); i += 2) {
1860 myri10ge_pio_copy(dstp, srcp, 2 * sizeof(*src));
1861 mb(); /* force write every 32 bytes */
1862 srcp += 2;
1863 dstp += 2;
1864 }
1865 } else {
1866 /* submit all but the first request, and ensure
1867 * that it is submitted below */
1868 myri10ge_submit_req_backwards(tx, src, cnt);
1869 i = 0;
1870 }
1871 if (i < cnt) {
1872 /* submit the first request */
1873 myri10ge_pio_copy(dstp, srcp, sizeof(*src));
1874 mb(); /* barrier before setting valid flag */
1875 }
1876
1877 /* re-write the last 32-bits with the valid flags */
1878 src->flags = last_flags;
1879 __raw_writel(*((u32 *) src + 3), (u32 __iomem *) dst + 3);
1880 tx->req += cnt;
1881 mb();
1882}
1883
1884static inline void
1885myri10ge_submit_req_wc(struct myri10ge_tx_buf *tx,
1886 struct mcp_kreq_ether_send *src, int cnt)
1887{
1888 tx->req += cnt;
1889 mb();
1890 while (cnt >= 4) {
1891 myri10ge_pio_copy(tx->wc_fifo, src, 64);
1892 mb();
1893 src += 4;
1894 cnt -= 4;
1895 }
1896 if (cnt > 0) {
1897 /* pad it to 64 bytes. The src is 64 bytes bigger than it
1898 * needs to be so that we don't overrun it */
Brice Gogline700f9f2006-08-14 17:52:54 -04001899 myri10ge_pio_copy(tx->wc_fifo + MXGEFW_ETH_SEND_OFFSET(cnt),
1900 src, 64);
Brice Goglin0da34b62006-05-23 06:10:15 -04001901 mb();
1902 }
1903}
1904
1905/*
1906 * Transmit a packet. We need to split the packet so that a single
1907 * segment does not cross myri10ge->tx.boundary, so this makes segment
1908 * counting tricky. So rather than try to count segments up front, we
1909 * just give up if there are too few segments to hold a reasonably
1910 * fragmented packet currently available. If we run
1911 * out of segments while preparing a packet for DMA, we just linearize
1912 * it and try again.
1913 */
1914
1915static int myri10ge_xmit(struct sk_buff *skb, struct net_device *dev)
1916{
1917 struct myri10ge_priv *mgp = netdev_priv(dev);
1918 struct mcp_kreq_ether_send *req;
1919 struct myri10ge_tx_buf *tx = &mgp->tx;
1920 struct skb_frag_struct *frag;
1921 dma_addr_t bus;
1922 u32 low, high_swapped;
1923 unsigned int len;
1924 int idx, last_idx, avail, frag_cnt, frag_idx, count, mss, max_segments;
1925 u16 pseudo_hdr_offset, cksum_offset;
1926 int cum_len, seglen, boundary, rdma_count;
1927 u8 flags, odd_flag;
1928
1929again:
1930 req = tx->req_list;
1931 avail = tx->mask - 1 - (tx->req - tx->done);
1932
1933 mss = 0;
1934 max_segments = MXGEFW_MAX_SEND_DESC;
1935
1936#ifdef NETIF_F_TSO
1937 if (skb->len > (dev->mtu + ETH_HLEN)) {
Herbert Xu79671682006-06-22 02:40:14 -07001938 mss = skb_shinfo(skb)->gso_size;
Brice Goglin0da34b62006-05-23 06:10:15 -04001939 if (mss != 0)
1940 max_segments = MYRI10GE_MAX_SEND_DESC_TSO;
1941 }
1942#endif /*NETIF_F_TSO */
1943
1944 if ((unlikely(avail < max_segments))) {
1945 /* we are out of transmit resources */
1946 mgp->stop_queue++;
1947 netif_stop_queue(dev);
1948 return 1;
1949 }
1950
1951 /* Setup checksum offloading, if needed */
1952 cksum_offset = 0;
1953 pseudo_hdr_offset = 0;
1954 odd_flag = 0;
1955 flags = (MXGEFW_FLAGS_NO_TSO | MXGEFW_FLAGS_FIRST);
Patrick McHardy84fa7932006-08-29 16:44:56 -07001956 if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
Brice Goglin0da34b62006-05-23 06:10:15 -04001957 cksum_offset = (skb->h.raw - skb->data);
1958 pseudo_hdr_offset = (skb->h.raw + skb->csum) - skb->data;
1959 /* If the headers are excessively large, then we must
1960 * fall back to a software checksum */
1961 if (unlikely(cksum_offset > 255 || pseudo_hdr_offset > 127)) {
Patrick McHardy84fa7932006-08-29 16:44:56 -07001962 if (skb_checksum_help(skb))
Brice Goglin0da34b62006-05-23 06:10:15 -04001963 goto drop;
1964 cksum_offset = 0;
1965 pseudo_hdr_offset = 0;
1966 } else {
1967 pseudo_hdr_offset = htons(pseudo_hdr_offset);
1968 odd_flag = MXGEFW_FLAGS_ALIGN_ODD;
1969 flags |= MXGEFW_FLAGS_CKSUM;
1970 }
1971 }
1972
1973 cum_len = 0;
1974
1975#ifdef NETIF_F_TSO
1976 if (mss) { /* TSO */
1977 /* this removes any CKSUM flag from before */
1978 flags = (MXGEFW_FLAGS_TSO_HDR | MXGEFW_FLAGS_FIRST);
1979
1980 /* negative cum_len signifies to the
1981 * send loop that we are still in the
1982 * header portion of the TSO packet.
1983 * TSO header must be at most 134 bytes long */
1984 cum_len = -((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
1985
1986 /* for TSO, pseudo_hdr_offset holds mss.
1987 * The firmware figures out where to put
1988 * the checksum by parsing the header. */
1989 pseudo_hdr_offset = htons(mss);
1990 } else
1991#endif /*NETIF_F_TSO */
1992 /* Mark small packets, and pad out tiny packets */
1993 if (skb->len <= MXGEFW_SEND_SMALL_SIZE) {
1994 flags |= MXGEFW_FLAGS_SMALL;
1995
1996 /* pad frames to at least ETH_ZLEN bytes */
1997 if (unlikely(skb->len < ETH_ZLEN)) {
Herbert Xu5b057c62006-06-23 02:06:41 -07001998 if (skb_padto(skb, ETH_ZLEN)) {
Brice Goglin0da34b62006-05-23 06:10:15 -04001999 /* The packet is gone, so we must
2000 * return 0 */
2001 mgp->stats.tx_dropped += 1;
2002 return 0;
2003 }
2004 /* adjust the len to account for the zero pad
2005 * so that the nic can know how long it is */
2006 skb->len = ETH_ZLEN;
2007 }
2008 }
2009
2010 /* map the skb for DMA */
2011 len = skb->len - skb->data_len;
2012 idx = tx->req & tx->mask;
2013 tx->info[idx].skb = skb;
2014 bus = pci_map_single(mgp->pdev, skb->data, len, PCI_DMA_TODEVICE);
2015 pci_unmap_addr_set(&tx->info[idx], bus, bus);
2016 pci_unmap_len_set(&tx->info[idx], len, len);
2017
2018 frag_cnt = skb_shinfo(skb)->nr_frags;
2019 frag_idx = 0;
2020 count = 0;
2021 rdma_count = 0;
2022
2023 /* "rdma_count" is the number of RDMAs belonging to the
2024 * current packet BEFORE the current send request. For
2025 * non-TSO packets, this is equal to "count".
2026 * For TSO packets, rdma_count needs to be reset
2027 * to 0 after a segment cut.
2028 *
2029 * The rdma_count field of the send request is
2030 * the number of RDMAs of the packet starting at
2031 * that request. For TSO send requests with one ore more cuts
2032 * in the middle, this is the number of RDMAs starting
2033 * after the last cut in the request. All previous
2034 * segments before the last cut implicitly have 1 RDMA.
2035 *
2036 * Since the number of RDMAs is not known beforehand,
2037 * it must be filled-in retroactively - after each
2038 * segmentation cut or at the end of the entire packet.
2039 */
2040
2041 while (1) {
2042 /* Break the SKB or Fragment up into pieces which
2043 * do not cross mgp->tx.boundary */
2044 low = MYRI10GE_LOWPART_TO_U32(bus);
2045 high_swapped = htonl(MYRI10GE_HIGHPART_TO_U32(bus));
2046 while (len) {
2047 u8 flags_next;
2048 int cum_len_next;
2049
2050 if (unlikely(count == max_segments))
2051 goto abort_linearize;
2052
2053 boundary = (low + tx->boundary) & ~(tx->boundary - 1);
2054 seglen = boundary - low;
2055 if (seglen > len)
2056 seglen = len;
2057 flags_next = flags & ~MXGEFW_FLAGS_FIRST;
2058 cum_len_next = cum_len + seglen;
2059#ifdef NETIF_F_TSO
2060 if (mss) { /* TSO */
2061 (req - rdma_count)->rdma_count = rdma_count + 1;
2062
2063 if (likely(cum_len >= 0)) { /* payload */
2064 int next_is_first, chop;
2065
2066 chop = (cum_len_next > mss);
2067 cum_len_next = cum_len_next % mss;
2068 next_is_first = (cum_len_next == 0);
2069 flags |= chop * MXGEFW_FLAGS_TSO_CHOP;
2070 flags_next |= next_is_first *
2071 MXGEFW_FLAGS_FIRST;
2072 rdma_count |= -(chop | next_is_first);
2073 rdma_count += chop & !next_is_first;
2074 } else if (likely(cum_len_next >= 0)) { /* header ends */
2075 int small;
2076
2077 rdma_count = -1;
2078 cum_len_next = 0;
2079 seglen = -cum_len;
2080 small = (mss <= MXGEFW_SEND_SMALL_SIZE);
2081 flags_next = MXGEFW_FLAGS_TSO_PLD |
2082 MXGEFW_FLAGS_FIRST |
2083 (small * MXGEFW_FLAGS_SMALL);
2084 }
2085 }
2086#endif /* NETIF_F_TSO */
2087 req->addr_high = high_swapped;
2088 req->addr_low = htonl(low);
2089 req->pseudo_hdr_offset = pseudo_hdr_offset;
2090 req->pad = 0; /* complete solid 16-byte block; does this matter? */
2091 req->rdma_count = 1;
2092 req->length = htons(seglen);
2093 req->cksum_offset = cksum_offset;
2094 req->flags = flags | ((cum_len & 1) * odd_flag);
2095
2096 low += seglen;
2097 len -= seglen;
2098 cum_len = cum_len_next;
2099 flags = flags_next;
2100 req++;
2101 count++;
2102 rdma_count++;
2103 if (unlikely(cksum_offset > seglen))
2104 cksum_offset -= seglen;
2105 else
2106 cksum_offset = 0;
2107 }
2108 if (frag_idx == frag_cnt)
2109 break;
2110
2111 /* map next fragment for DMA */
2112 idx = (count + tx->req) & tx->mask;
2113 frag = &skb_shinfo(skb)->frags[frag_idx];
2114 frag_idx++;
2115 len = frag->size;
2116 bus = pci_map_page(mgp->pdev, frag->page, frag->page_offset,
2117 len, PCI_DMA_TODEVICE);
2118 pci_unmap_addr_set(&tx->info[idx], bus, bus);
2119 pci_unmap_len_set(&tx->info[idx], len, len);
2120 }
2121
2122 (req - rdma_count)->rdma_count = rdma_count;
2123#ifdef NETIF_F_TSO
2124 if (mss)
2125 do {
2126 req--;
2127 req->flags |= MXGEFW_FLAGS_TSO_LAST;
2128 } while (!(req->flags & (MXGEFW_FLAGS_TSO_CHOP |
2129 MXGEFW_FLAGS_FIRST)));
2130#endif
2131 idx = ((count - 1) + tx->req) & tx->mask;
2132 tx->info[idx].last = 1;
2133 if (tx->wc_fifo == NULL)
2134 myri10ge_submit_req(tx, tx->req_list, count);
2135 else
2136 myri10ge_submit_req_wc(tx, tx->req_list, count);
2137 tx->pkt_start++;
2138 if ((avail - count) < MXGEFW_MAX_SEND_DESC) {
2139 mgp->stop_queue++;
2140 netif_stop_queue(dev);
2141 }
2142 dev->trans_start = jiffies;
2143 return 0;
2144
2145abort_linearize:
2146 /* Free any DMA resources we've alloced and clear out the skb
2147 * slot so as to not trip up assertions, and to avoid a
2148 * double-free if linearizing fails */
2149
2150 last_idx = (idx + 1) & tx->mask;
2151 idx = tx->req & tx->mask;
2152 tx->info[idx].skb = NULL;
2153 do {
2154 len = pci_unmap_len(&tx->info[idx], len);
2155 if (len) {
2156 if (tx->info[idx].skb != NULL)
2157 pci_unmap_single(mgp->pdev,
2158 pci_unmap_addr(&tx->info[idx],
2159 bus), len,
2160 PCI_DMA_TODEVICE);
2161 else
2162 pci_unmap_page(mgp->pdev,
2163 pci_unmap_addr(&tx->info[idx],
2164 bus), len,
2165 PCI_DMA_TODEVICE);
2166 pci_unmap_len_set(&tx->info[idx], len, 0);
2167 tx->info[idx].skb = NULL;
2168 }
2169 idx = (idx + 1) & tx->mask;
2170 } while (idx != last_idx);
Herbert Xu89114af2006-07-08 13:34:32 -07002171 if (skb_is_gso(skb)) {
Brice Goglin0da34b62006-05-23 06:10:15 -04002172 printk(KERN_ERR
2173 "myri10ge: %s: TSO but wanted to linearize?!?!?\n",
2174 mgp->dev->name);
2175 goto drop;
2176 }
2177
Andrew Mortonbec0e852006-06-22 14:47:19 -07002178 if (skb_linearize(skb))
Brice Goglin0da34b62006-05-23 06:10:15 -04002179 goto drop;
2180
2181 mgp->tx_linearized++;
2182 goto again;
2183
2184drop:
2185 dev_kfree_skb_any(skb);
2186 mgp->stats.tx_dropped += 1;
2187 return 0;
2188
2189}
2190
2191static struct net_device_stats *myri10ge_get_stats(struct net_device *dev)
2192{
2193 struct myri10ge_priv *mgp = netdev_priv(dev);
2194 return &mgp->stats;
2195}
2196
2197static void myri10ge_set_multicast_list(struct net_device *dev)
2198{
Brice Goglin85a7ea12006-08-21 17:36:56 -04002199 struct myri10ge_cmd cmd;
2200 struct myri10ge_priv *mgp;
2201 struct dev_mc_list *mc_list;
2202 int err;
2203
2204 mgp = netdev_priv(dev);
Brice Goglin0da34b62006-05-23 06:10:15 -04002205 /* can be called from atomic contexts,
2206 * pass 1 to force atomicity in myri10ge_send_cmd() */
Brice Goglin85a7ea12006-08-21 17:36:56 -04002207 myri10ge_change_promisc(mgp, dev->flags & IFF_PROMISC, 1);
2208
2209 /* This firmware is known to not support multicast */
2210 if (!mgp->fw_multicast_support)
2211 return;
2212
2213 /* Disable multicast filtering */
2214
2215 err = myri10ge_send_cmd(mgp, MXGEFW_ENABLE_ALLMULTI, &cmd, 1);
2216 if (err != 0) {
2217 printk(KERN_ERR "myri10ge: %s: Failed MXGEFW_ENABLE_ALLMULTI,"
2218 " error status: %d\n", dev->name, err);
2219 goto abort;
2220 }
2221
2222 if (dev->flags & IFF_ALLMULTI) {
2223 /* request to disable multicast filtering, so quit here */
2224 return;
2225 }
2226
2227 /* Flush the filters */
2228
2229 err = myri10ge_send_cmd(mgp, MXGEFW_LEAVE_ALL_MULTICAST_GROUPS,
2230 &cmd, 1);
2231 if (err != 0) {
2232 printk(KERN_ERR
2233 "myri10ge: %s: Failed MXGEFW_LEAVE_ALL_MULTICAST_GROUPS"
2234 ", error status: %d\n", dev->name, err);
2235 goto abort;
2236 }
2237
2238 /* Walk the multicast list, and add each address */
2239 for (mc_list = dev->mc_list; mc_list != NULL; mc_list = mc_list->next) {
2240 memcpy(&cmd.data0, &mc_list->dmi_addr, 4);
2241 memcpy(&cmd.data1, ((char *)&mc_list->dmi_addr) + 4, 2);
2242 cmd.data0 = htonl(cmd.data0);
2243 cmd.data1 = htonl(cmd.data1);
2244 err = myri10ge_send_cmd(mgp, MXGEFW_JOIN_MULTICAST_GROUP,
2245 &cmd, 1);
2246
2247 if (err != 0) {
2248 printk(KERN_ERR "myri10ge: %s: Failed "
2249 "MXGEFW_JOIN_MULTICAST_GROUP, error status:"
2250 "%d\t", dev->name, err);
2251 printk(KERN_ERR "MAC %02x:%02x:%02x:%02x:%02x:%02x\n",
2252 ((unsigned char *)&mc_list->dmi_addr)[0],
2253 ((unsigned char *)&mc_list->dmi_addr)[1],
2254 ((unsigned char *)&mc_list->dmi_addr)[2],
2255 ((unsigned char *)&mc_list->dmi_addr)[3],
2256 ((unsigned char *)&mc_list->dmi_addr)[4],
2257 ((unsigned char *)&mc_list->dmi_addr)[5]
2258 );
2259 goto abort;
2260 }
2261 }
2262 /* Enable multicast filtering */
2263 err = myri10ge_send_cmd(mgp, MXGEFW_DISABLE_ALLMULTI, &cmd, 1);
2264 if (err != 0) {
2265 printk(KERN_ERR "myri10ge: %s: Failed MXGEFW_DISABLE_ALLMULTI,"
2266 "error status: %d\n", dev->name, err);
2267 goto abort;
2268 }
2269
2270 return;
2271
2272abort:
2273 return;
Brice Goglin0da34b62006-05-23 06:10:15 -04002274}
2275
2276static int myri10ge_set_mac_address(struct net_device *dev, void *addr)
2277{
2278 struct sockaddr *sa = addr;
2279 struct myri10ge_priv *mgp = netdev_priv(dev);
2280 int status;
2281
2282 if (!is_valid_ether_addr(sa->sa_data))
2283 return -EADDRNOTAVAIL;
2284
2285 status = myri10ge_update_mac_address(mgp, sa->sa_data);
2286 if (status != 0) {
2287 printk(KERN_ERR
2288 "myri10ge: %s: changing mac address failed with %d\n",
2289 dev->name, status);
2290 return status;
2291 }
2292
2293 /* change the dev structure */
2294 memcpy(dev->dev_addr, sa->sa_data, 6);
2295 return 0;
2296}
2297
2298static int myri10ge_change_mtu(struct net_device *dev, int new_mtu)
2299{
2300 struct myri10ge_priv *mgp = netdev_priv(dev);
2301 int error = 0;
2302
2303 if ((new_mtu < 68) || (ETH_HLEN + new_mtu > MYRI10GE_MAX_ETHER_MTU)) {
2304 printk(KERN_ERR "myri10ge: %s: new mtu (%d) is not valid\n",
2305 dev->name, new_mtu);
2306 return -EINVAL;
2307 }
2308 printk(KERN_INFO "%s: changing mtu from %d to %d\n",
2309 dev->name, dev->mtu, new_mtu);
2310 if (mgp->running) {
2311 /* if we change the mtu on an active device, we must
2312 * reset the device so the firmware sees the change */
2313 myri10ge_close(dev);
2314 dev->mtu = new_mtu;
2315 myri10ge_open(dev);
2316 } else
2317 dev->mtu = new_mtu;
2318
2319 return error;
2320}
2321
2322/*
2323 * Enable ECRC to align PCI-E Completion packets on an 8-byte boundary.
2324 * Only do it if the bridge is a root port since we don't want to disturb
2325 * any other device, except if forced with myri10ge_ecrc_enable > 1.
2326 */
2327
Brice Goglin0da34b62006-05-23 06:10:15 -04002328static void myri10ge_enable_ecrc(struct myri10ge_priv *mgp)
2329{
2330 struct pci_dev *bridge = mgp->pdev->bus->self;
2331 struct device *dev = &mgp->pdev->dev;
2332 unsigned cap;
2333 unsigned err_cap;
2334 u16 val;
2335 u8 ext_type;
2336 int ret;
2337
2338 if (!myri10ge_ecrc_enable || !bridge)
2339 return;
2340
2341 /* check that the bridge is a root port */
2342 cap = pci_find_capability(bridge, PCI_CAP_ID_EXP);
2343 pci_read_config_word(bridge, cap + PCI_CAP_FLAGS, &val);
2344 ext_type = (val & PCI_EXP_FLAGS_TYPE) >> 4;
2345 if (ext_type != PCI_EXP_TYPE_ROOT_PORT) {
2346 if (myri10ge_ecrc_enable > 1) {
2347 struct pci_dev *old_bridge = bridge;
2348
2349 /* Walk the hierarchy up to the root port
2350 * where ECRC has to be enabled */
2351 do {
2352 bridge = bridge->bus->self;
2353 if (!bridge) {
2354 dev_err(dev,
2355 "Failed to find root port"
2356 " to force ECRC\n");
2357 return;
2358 }
2359 cap =
2360 pci_find_capability(bridge, PCI_CAP_ID_EXP);
2361 pci_read_config_word(bridge,
2362 cap + PCI_CAP_FLAGS, &val);
2363 ext_type = (val & PCI_EXP_FLAGS_TYPE) >> 4;
2364 } while (ext_type != PCI_EXP_TYPE_ROOT_PORT);
2365
2366 dev_info(dev,
2367 "Forcing ECRC on non-root port %s"
2368 " (enabling on root port %s)\n",
2369 pci_name(old_bridge), pci_name(bridge));
2370 } else {
2371 dev_err(dev,
2372 "Not enabling ECRC on non-root port %s\n",
2373 pci_name(bridge));
2374 return;
2375 }
2376 }
2377
2378 cap = pci_find_ext_capability(bridge, PCI_EXT_CAP_ID_ERR);
Brice Goglin0da34b62006-05-23 06:10:15 -04002379 if (!cap)
2380 return;
2381
2382 ret = pci_read_config_dword(bridge, cap + PCI_ERR_CAP, &err_cap);
2383 if (ret) {
2384 dev_err(dev, "failed reading ext-conf-space of %s\n",
2385 pci_name(bridge));
2386 dev_err(dev, "\t pci=nommconf in use? "
2387 "or buggy/incomplete/absent ACPI MCFG attr?\n");
2388 return;
2389 }
2390 if (!(err_cap & PCI_ERR_CAP_ECRC_GENC))
2391 return;
2392
2393 err_cap |= PCI_ERR_CAP_ECRC_GENE;
2394 pci_write_config_dword(bridge, cap + PCI_ERR_CAP, err_cap);
2395 dev_info(dev, "Enabled ECRC on upstream bridge %s\n", pci_name(bridge));
2396 mgp->tx.boundary = 4096;
2397 mgp->fw_name = myri10ge_fw_aligned;
2398}
2399
2400/*
2401 * The Lanai Z8E PCI-E interface achieves higher Read-DMA throughput
2402 * when the PCI-E Completion packets are aligned on an 8-byte
2403 * boundary. Some PCI-E chip sets always align Completion packets; on
2404 * the ones that do not, the alignment can be enforced by enabling
2405 * ECRC generation (if supported).
2406 *
2407 * When PCI-E Completion packets are not aligned, it is actually more
2408 * efficient to limit Read-DMA transactions to 2KB, rather than 4KB.
2409 *
2410 * If the driver can neither enable ECRC nor verify that it has
2411 * already been enabled, then it must use a firmware image which works
2412 * around unaligned completion packets (myri10ge_ethp_z8e.dat), and it
2413 * should also ensure that it never gives the device a Read-DMA which is
2414 * larger than 2KB by setting the tx.boundary to 2KB. If ECRC is
2415 * enabled, then the driver should use the aligned (myri10ge_eth_z8e.dat)
2416 * firmware image, and set tx.boundary to 4KB.
2417 */
2418
2419#define PCI_DEVICE_ID_SERVERWORKS_HT2000_PCIE 0x0132
Brice Goglince7f9362006-08-31 01:32:59 -04002420#define PCI_DEVICE_ID_INTEL_E5000_PCIE23 0x25f7
2421#define PCI_DEVICE_ID_INTEL_E5000_PCIE47 0x25fa
Brice Goglin0da34b62006-05-23 06:10:15 -04002422
2423static void myri10ge_select_firmware(struct myri10ge_priv *mgp)
2424{
2425 struct pci_dev *bridge = mgp->pdev->bus->self;
2426
2427 mgp->tx.boundary = 2048;
2428 mgp->fw_name = myri10ge_fw_unaligned;
2429
2430 if (myri10ge_force_firmware == 0) {
Brice Goglince7f9362006-08-31 01:32:59 -04002431 int link_width, exp_cap;
2432 u16 lnk;
2433
2434 exp_cap = pci_find_capability(mgp->pdev, PCI_CAP_ID_EXP);
2435 pci_read_config_word(mgp->pdev, exp_cap + PCI_EXP_LNKSTA, &lnk);
2436 link_width = (lnk >> 4) & 0x3f;
2437
Brice Goglin0da34b62006-05-23 06:10:15 -04002438 myri10ge_enable_ecrc(mgp);
2439
Brice Goglince7f9362006-08-31 01:32:59 -04002440 /* Check to see if Link is less than 8 or if the
2441 * upstream bridge is known to provide aligned
2442 * completions */
2443 if (link_width < 8) {
2444 dev_info(&mgp->pdev->dev, "PCIE x%d Link\n",
2445 link_width);
2446 mgp->tx.boundary = 4096;
2447 mgp->fw_name = myri10ge_fw_aligned;
2448 } else if (bridge &&
2449 /* ServerWorks HT2000/HT1000 */
2450 ((bridge->vendor == PCI_VENDOR_ID_SERVERWORKS
2451 && bridge->device ==
2452 PCI_DEVICE_ID_SERVERWORKS_HT2000_PCIE)
2453 /* All Intel E5000 PCIE ports */
2454 || (bridge->vendor == PCI_VENDOR_ID_INTEL
2455 && bridge->device >=
2456 PCI_DEVICE_ID_INTEL_E5000_PCIE23
2457 && bridge->device <=
2458 PCI_DEVICE_ID_INTEL_E5000_PCIE47))) {
Brice Goglin0da34b62006-05-23 06:10:15 -04002459 dev_info(&mgp->pdev->dev,
2460 "Assuming aligned completions (0x%x:0x%x)\n",
2461 bridge->vendor, bridge->device);
2462 mgp->tx.boundary = 4096;
2463 mgp->fw_name = myri10ge_fw_aligned;
2464 }
2465 } else {
2466 if (myri10ge_force_firmware == 1) {
2467 dev_info(&mgp->pdev->dev,
2468 "Assuming aligned completions (forced)\n");
2469 mgp->tx.boundary = 4096;
2470 mgp->fw_name = myri10ge_fw_aligned;
2471 } else {
2472 dev_info(&mgp->pdev->dev,
2473 "Assuming unaligned completions (forced)\n");
2474 mgp->tx.boundary = 2048;
2475 mgp->fw_name = myri10ge_fw_unaligned;
2476 }
2477 }
2478 if (myri10ge_fw_name != NULL) {
2479 dev_info(&mgp->pdev->dev, "overriding firmware to %s\n",
2480 myri10ge_fw_name);
2481 mgp->fw_name = myri10ge_fw_name;
2482 }
2483}
2484
2485static void myri10ge_save_state(struct myri10ge_priv *mgp)
2486{
2487 struct pci_dev *pdev = mgp->pdev;
2488 int cap;
2489
2490 pci_save_state(pdev);
2491 /* now save PCIe and MSI state that Linux will not
2492 * save for us */
2493 cap = pci_find_capability(pdev, PCI_CAP_ID_EXP);
2494 pci_read_config_dword(pdev, cap + PCI_EXP_DEVCTL, &mgp->devctl);
2495 cap = pci_find_capability(pdev, PCI_CAP_ID_MSI);
2496 pci_read_config_word(pdev, cap + PCI_MSI_FLAGS, &mgp->msi_flags);
2497}
2498
2499static void myri10ge_restore_state(struct myri10ge_priv *mgp)
2500{
2501 struct pci_dev *pdev = mgp->pdev;
2502 int cap;
2503
2504 /* restore PCIe and MSI state that linux will not */
2505 cap = pci_find_capability(pdev, PCI_CAP_ID_EXP);
2506 pci_write_config_dword(pdev, cap + PCI_CAP_ID_EXP, mgp->devctl);
2507 cap = pci_find_capability(pdev, PCI_CAP_ID_MSI);
2508 pci_write_config_word(pdev, cap + PCI_MSI_FLAGS, mgp->msi_flags);
2509
2510 pci_restore_state(pdev);
2511}
2512
2513#ifdef CONFIG_PM
2514
2515static int myri10ge_suspend(struct pci_dev *pdev, pm_message_t state)
2516{
2517 struct myri10ge_priv *mgp;
2518 struct net_device *netdev;
2519
2520 mgp = pci_get_drvdata(pdev);
2521 if (mgp == NULL)
2522 return -EINVAL;
2523 netdev = mgp->dev;
2524
2525 netif_device_detach(netdev);
2526 if (netif_running(netdev)) {
2527 printk(KERN_INFO "myri10ge: closing %s\n", netdev->name);
2528 rtnl_lock();
2529 myri10ge_close(netdev);
2530 rtnl_unlock();
2531 }
2532 myri10ge_dummy_rdma(mgp, 0);
2533 free_irq(pdev->irq, mgp);
2534 myri10ge_save_state(mgp);
2535 pci_disable_device(pdev);
2536 pci_set_power_state(pdev, pci_choose_state(pdev, state));
2537 return 0;
2538}
2539
2540static int myri10ge_resume(struct pci_dev *pdev)
2541{
2542 struct myri10ge_priv *mgp;
2543 struct net_device *netdev;
2544 int status;
2545 u16 vendor;
2546
2547 mgp = pci_get_drvdata(pdev);
2548 if (mgp == NULL)
2549 return -EINVAL;
2550 netdev = mgp->dev;
2551 pci_set_power_state(pdev, 0); /* zeros conf space as a side effect */
2552 msleep(5); /* give card time to respond */
2553 pci_read_config_word(mgp->pdev, PCI_VENDOR_ID, &vendor);
2554 if (vendor == 0xffff) {
2555 printk(KERN_ERR "myri10ge: %s: device disappeared!\n",
2556 mgp->dev->name);
2557 return -EIO;
2558 }
2559 myri10ge_restore_state(mgp);
Brice Goglin4c2248c2006-07-09 21:10:18 -04002560
2561 status = pci_enable_device(pdev);
2562 if (status < 0) {
2563 dev_err(&pdev->dev, "failed to enable device\n");
2564 return -EIO;
2565 }
2566
Brice Goglin0da34b62006-05-23 06:10:15 -04002567 pci_set_master(pdev);
2568
Thomas Gleixner1fb9df52006-07-01 19:29:39 -07002569 status = request_irq(pdev->irq, myri10ge_intr, IRQF_SHARED,
Brice Goglin0da34b62006-05-23 06:10:15 -04002570 netdev->name, mgp);
2571 if (status != 0) {
2572 dev_err(&pdev->dev, "failed to allocate IRQ\n");
Brice Goglin4c2248c2006-07-09 21:10:18 -04002573 goto abort_with_enabled;
Brice Goglin0da34b62006-05-23 06:10:15 -04002574 }
2575
2576 myri10ge_reset(mgp);
Brice Goglin013b68b2006-08-09 00:07:53 -04002577 myri10ge_dummy_rdma(mgp, 1);
Brice Goglin0da34b62006-05-23 06:10:15 -04002578
2579 /* Save configuration space to be restored if the
2580 * nic resets due to a parity error */
2581 myri10ge_save_state(mgp);
2582
2583 if (netif_running(netdev)) {
2584 rtnl_lock();
2585 myri10ge_open(netdev);
2586 rtnl_unlock();
2587 }
2588 netif_device_attach(netdev);
2589
2590 return 0;
2591
Brice Goglin4c2248c2006-07-09 21:10:18 -04002592abort_with_enabled:
2593 pci_disable_device(pdev);
Brice Goglin0da34b62006-05-23 06:10:15 -04002594 return -EIO;
2595
2596}
2597
2598#endif /* CONFIG_PM */
2599
2600static u32 myri10ge_read_reboot(struct myri10ge_priv *mgp)
2601{
2602 struct pci_dev *pdev = mgp->pdev;
2603 int vs = mgp->vendor_specific_offset;
2604 u32 reboot;
2605
2606 /*enter read32 mode */
2607 pci_write_config_byte(pdev, vs + 0x10, 0x3);
2608
2609 /*read REBOOT_STATUS (0xfffffff0) */
2610 pci_write_config_dword(pdev, vs + 0x18, 0xfffffff0);
2611 pci_read_config_dword(pdev, vs + 0x14, &reboot);
2612 return reboot;
2613}
2614
2615/*
2616 * This watchdog is used to check whether the board has suffered
2617 * from a parity error and needs to be recovered.
2618 */
2619static void myri10ge_watchdog(void *arg)
2620{
2621 struct myri10ge_priv *mgp = arg;
2622 u32 reboot;
2623 int status;
2624 u16 cmd, vendor;
2625
2626 mgp->watchdog_resets++;
2627 pci_read_config_word(mgp->pdev, PCI_COMMAND, &cmd);
2628 if ((cmd & PCI_COMMAND_MASTER) == 0) {
2629 /* Bus master DMA disabled? Check to see
2630 * if the card rebooted due to a parity error
2631 * For now, just report it */
2632 reboot = myri10ge_read_reboot(mgp);
2633 printk(KERN_ERR
2634 "myri10ge: %s: NIC rebooted (0x%x), resetting\n",
2635 mgp->dev->name, reboot);
2636 /*
2637 * A rebooted nic will come back with config space as
2638 * it was after power was applied to PCIe bus.
2639 * Attempt to restore config space which was saved
2640 * when the driver was loaded, or the last time the
2641 * nic was resumed from power saving mode.
2642 */
2643 myri10ge_restore_state(mgp);
2644 } else {
2645 /* if we get back -1's from our slot, perhaps somebody
2646 * powered off our card. Don't try to reset it in
2647 * this case */
2648 if (cmd == 0xffff) {
2649 pci_read_config_word(mgp->pdev, PCI_VENDOR_ID, &vendor);
2650 if (vendor == 0xffff) {
2651 printk(KERN_ERR
2652 "myri10ge: %s: device disappeared!\n",
2653 mgp->dev->name);
2654 return;
2655 }
2656 }
2657 /* Perhaps it is a software error. Try to reset */
2658
2659 printk(KERN_ERR "myri10ge: %s: device timeout, resetting\n",
2660 mgp->dev->name);
2661 printk(KERN_INFO "myri10ge: %s: %d %d %d %d %d\n",
2662 mgp->dev->name, mgp->tx.req, mgp->tx.done,
2663 mgp->tx.pkt_start, mgp->tx.pkt_done,
2664 (int)ntohl(mgp->fw_stats->send_done_count));
2665 msleep(2000);
2666 printk(KERN_INFO "myri10ge: %s: %d %d %d %d %d\n",
2667 mgp->dev->name, mgp->tx.req, mgp->tx.done,
2668 mgp->tx.pkt_start, mgp->tx.pkt_done,
2669 (int)ntohl(mgp->fw_stats->send_done_count));
2670 }
2671 rtnl_lock();
2672 myri10ge_close(mgp->dev);
2673 status = myri10ge_load_firmware(mgp);
2674 if (status != 0)
2675 printk(KERN_ERR "myri10ge: %s: failed to load firmware\n",
2676 mgp->dev->name);
2677 else
2678 myri10ge_open(mgp->dev);
2679 rtnl_unlock();
2680}
2681
2682/*
2683 * We use our own timer routine rather than relying upon
2684 * netdev->tx_timeout because we have a very large hardware transmit
2685 * queue. Due to the large queue, the netdev->tx_timeout function
2686 * cannot detect a NIC with a parity error in a timely fashion if the
2687 * NIC is lightly loaded.
2688 */
2689static void myri10ge_watchdog_timer(unsigned long arg)
2690{
2691 struct myri10ge_priv *mgp;
2692
2693 mgp = (struct myri10ge_priv *)arg;
2694 if (mgp->tx.req != mgp->tx.done &&
Brice Goglinc54772e2006-07-30 00:14:15 -04002695 mgp->tx.done == mgp->watchdog_tx_done &&
2696 mgp->watchdog_tx_req != mgp->watchdog_tx_done)
Brice Goglin0da34b62006-05-23 06:10:15 -04002697 /* nic seems like it might be stuck.. */
2698 schedule_work(&mgp->watchdog_work);
2699 else
2700 /* rearm timer */
2701 mod_timer(&mgp->watchdog_timer,
2702 jiffies + myri10ge_watchdog_timeout * HZ);
2703
2704 mgp->watchdog_tx_done = mgp->tx.done;
Brice Goglinc54772e2006-07-30 00:14:15 -04002705 mgp->watchdog_tx_req = mgp->tx.req;
Brice Goglin0da34b62006-05-23 06:10:15 -04002706}
2707
2708static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2709{
2710 struct net_device *netdev;
2711 struct myri10ge_priv *mgp;
2712 struct device *dev = &pdev->dev;
2713 size_t bytes;
2714 int i;
2715 int status = -ENXIO;
2716 int cap;
2717 int dac_enabled;
2718 u16 val;
2719
2720 netdev = alloc_etherdev(sizeof(*mgp));
2721 if (netdev == NULL) {
2722 dev_err(dev, "Could not allocate ethernet device\n");
2723 return -ENOMEM;
2724 }
2725
2726 mgp = netdev_priv(netdev);
2727 memset(mgp, 0, sizeof(*mgp));
2728 mgp->dev = netdev;
2729 mgp->pdev = pdev;
2730 mgp->csum_flag = MXGEFW_FLAGS_CKSUM;
2731 mgp->pause = myri10ge_flow_control;
2732 mgp->intr_coal_delay = myri10ge_intr_coal_delay;
Brice Goglinc58ac5c2006-08-21 17:36:49 -04002733 mgp->msg_enable = netif_msg_init(myri10ge_debug, MYRI10GE_MSG_DEFAULT);
Brice Goglin0da34b62006-05-23 06:10:15 -04002734 init_waitqueue_head(&mgp->down_wq);
2735
2736 if (pci_enable_device(pdev)) {
2737 dev_err(&pdev->dev, "pci_enable_device call failed\n");
2738 status = -ENODEV;
2739 goto abort_with_netdev;
2740 }
2741 myri10ge_select_firmware(mgp);
2742
2743 /* Find the vendor-specific cap so we can check
2744 * the reboot register later on */
2745 mgp->vendor_specific_offset
2746 = pci_find_capability(pdev, PCI_CAP_ID_VNDR);
2747
2748 /* Set our max read request to 4KB */
2749 cap = pci_find_capability(pdev, PCI_CAP_ID_EXP);
2750 if (cap < 64) {
2751 dev_err(&pdev->dev, "Bad PCI_CAP_ID_EXP location %d\n", cap);
2752 goto abort_with_netdev;
2753 }
2754 status = pci_read_config_word(pdev, cap + PCI_EXP_DEVCTL, &val);
2755 if (status != 0) {
2756 dev_err(&pdev->dev, "Error %d reading PCI_EXP_DEVCTL\n",
2757 status);
2758 goto abort_with_netdev;
2759 }
2760 val = (val & ~PCI_EXP_DEVCTL_READRQ) | (5 << 12);
2761 status = pci_write_config_word(pdev, cap + PCI_EXP_DEVCTL, val);
2762 if (status != 0) {
2763 dev_err(&pdev->dev, "Error %d writing PCI_EXP_DEVCTL\n",
2764 status);
2765 goto abort_with_netdev;
2766 }
2767
2768 pci_set_master(pdev);
2769 dac_enabled = 1;
2770 status = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
2771 if (status != 0) {
2772 dac_enabled = 0;
2773 dev_err(&pdev->dev,
2774 "64-bit pci address mask was refused, trying 32-bit");
2775 status = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
2776 }
2777 if (status != 0) {
2778 dev_err(&pdev->dev, "Error %d setting DMA mask\n", status);
2779 goto abort_with_netdev;
2780 }
Brice Goglinb10c0662006-06-08 10:25:00 -04002781 mgp->cmd = dma_alloc_coherent(&pdev->dev, sizeof(*mgp->cmd),
2782 &mgp->cmd_bus, GFP_KERNEL);
Brice Goglin0da34b62006-05-23 06:10:15 -04002783 if (mgp->cmd == NULL)
2784 goto abort_with_netdev;
2785
Brice Goglinb10c0662006-06-08 10:25:00 -04002786 mgp->fw_stats = dma_alloc_coherent(&pdev->dev, sizeof(*mgp->fw_stats),
2787 &mgp->fw_stats_bus, GFP_KERNEL);
Brice Goglin0da34b62006-05-23 06:10:15 -04002788 if (mgp->fw_stats == NULL)
2789 goto abort_with_cmd;
2790
2791 mgp->board_span = pci_resource_len(pdev, 0);
2792 mgp->iomem_base = pci_resource_start(pdev, 0);
2793 mgp->mtrr = -1;
2794#ifdef CONFIG_MTRR
2795 mgp->mtrr = mtrr_add(mgp->iomem_base, mgp->board_span,
2796 MTRR_TYPE_WRCOMB, 1);
2797#endif
2798 /* Hack. need to get rid of these magic numbers */
2799 mgp->sram_size =
2800 2 * 1024 * 1024 - (2 * (48 * 1024) + (32 * 1024)) - 0x100;
2801 if (mgp->sram_size > mgp->board_span) {
2802 dev_err(&pdev->dev, "board span %ld bytes too small\n",
2803 mgp->board_span);
2804 goto abort_with_wc;
2805 }
2806 mgp->sram = ioremap(mgp->iomem_base, mgp->board_span);
2807 if (mgp->sram == NULL) {
2808 dev_err(&pdev->dev, "ioremap failed for %ld bytes at 0x%lx\n",
2809 mgp->board_span, mgp->iomem_base);
2810 status = -ENXIO;
2811 goto abort_with_wc;
2812 }
2813 memcpy_fromio(mgp->eeprom_strings,
2814 mgp->sram + mgp->sram_size - MYRI10GE_EEPROM_STRINGS_SIZE,
2815 MYRI10GE_EEPROM_STRINGS_SIZE);
2816 memset(mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE - 2, 0, 2);
2817 status = myri10ge_read_mac_addr(mgp);
2818 if (status)
2819 goto abort_with_ioremap;
2820
2821 for (i = 0; i < ETH_ALEN; i++)
2822 netdev->dev_addr[i] = mgp->mac_addr[i];
2823
2824 /* allocate rx done ring */
2825 bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
Brice Goglinb10c0662006-06-08 10:25:00 -04002826 mgp->rx_done.entry = dma_alloc_coherent(&pdev->dev, bytes,
2827 &mgp->rx_done.bus, GFP_KERNEL);
Brice Goglin0da34b62006-05-23 06:10:15 -04002828 if (mgp->rx_done.entry == NULL)
2829 goto abort_with_ioremap;
2830 memset(mgp->rx_done.entry, 0, bytes);
2831
2832 status = myri10ge_load_firmware(mgp);
2833 if (status != 0) {
2834 dev_err(&pdev->dev, "failed to load firmware\n");
2835 goto abort_with_rx_done;
2836 }
2837
2838 status = myri10ge_reset(mgp);
2839 if (status != 0) {
2840 dev_err(&pdev->dev, "failed reset\n");
2841 goto abort_with_firmware;
2842 }
2843
2844 if (myri10ge_msi) {
2845 status = pci_enable_msi(pdev);
2846 if (status != 0)
2847 dev_err(&pdev->dev,
2848 "Error %d setting up MSI; falling back to xPIC\n",
2849 status);
2850 else
2851 mgp->msi_enabled = 1;
2852 }
2853
Thomas Gleixner1fb9df52006-07-01 19:29:39 -07002854 status = request_irq(pdev->irq, myri10ge_intr, IRQF_SHARED,
Brice Goglin0da34b62006-05-23 06:10:15 -04002855 netdev->name, mgp);
2856 if (status != 0) {
2857 dev_err(&pdev->dev, "failed to allocate IRQ\n");
2858 goto abort_with_firmware;
2859 }
2860
2861 pci_set_drvdata(pdev, mgp);
2862 if ((myri10ge_initial_mtu + ETH_HLEN) > MYRI10GE_MAX_ETHER_MTU)
2863 myri10ge_initial_mtu = MYRI10GE_MAX_ETHER_MTU - ETH_HLEN;
2864 if ((myri10ge_initial_mtu + ETH_HLEN) < 68)
2865 myri10ge_initial_mtu = 68;
2866 netdev->mtu = myri10ge_initial_mtu;
2867 netdev->open = myri10ge_open;
2868 netdev->stop = myri10ge_close;
2869 netdev->hard_start_xmit = myri10ge_xmit;
2870 netdev->get_stats = myri10ge_get_stats;
2871 netdev->base_addr = mgp->iomem_base;
2872 netdev->irq = pdev->irq;
2873 netdev->change_mtu = myri10ge_change_mtu;
2874 netdev->set_multicast_list = myri10ge_set_multicast_list;
2875 netdev->set_mac_address = myri10ge_set_mac_address;
2876 netdev->features = NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_TSO;
2877 if (dac_enabled)
2878 netdev->features |= NETIF_F_HIGHDMA;
2879 netdev->poll = myri10ge_poll;
2880 netdev->weight = myri10ge_napi_weight;
2881
2882 /* Save configuration space to be restored if the
2883 * nic resets due to a parity error */
2884 myri10ge_save_state(mgp);
Brice Goglin0da34b62006-05-23 06:10:15 -04002885
2886 /* Setup the watchdog timer */
2887 setup_timer(&mgp->watchdog_timer, myri10ge_watchdog_timer,
2888 (unsigned long)mgp);
2889
2890 SET_ETHTOOL_OPS(netdev, &myri10ge_ethtool_ops);
2891 INIT_WORK(&mgp->watchdog_work, myri10ge_watchdog, mgp);
2892 status = register_netdev(netdev);
2893 if (status != 0) {
2894 dev_err(&pdev->dev, "register_netdev failed: %d\n", status);
2895 goto abort_with_irq;
2896 }
Brice Goglind6020782006-07-03 18:41:30 -04002897 dev_info(dev, "%s IRQ %d, tx bndry %d, fw %s, WC %s\n",
2898 (mgp->msi_enabled ? "MSI" : "xPIC"),
2899 pdev->irq, mgp->tx.boundary, mgp->fw_name,
2900 (mgp->mtrr >= 0 ? "Enabled" : "Disabled"));
Brice Goglin0da34b62006-05-23 06:10:15 -04002901
2902 return 0;
2903
2904abort_with_irq:
2905 free_irq(pdev->irq, mgp);
2906 if (mgp->msi_enabled)
2907 pci_disable_msi(pdev);
2908
2909abort_with_firmware:
2910 myri10ge_dummy_rdma(mgp, 0);
2911
2912abort_with_rx_done:
2913 bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
Brice Goglinb10c0662006-06-08 10:25:00 -04002914 dma_free_coherent(&pdev->dev, bytes,
2915 mgp->rx_done.entry, mgp->rx_done.bus);
Brice Goglin0da34b62006-05-23 06:10:15 -04002916
2917abort_with_ioremap:
2918 iounmap(mgp->sram);
2919
2920abort_with_wc:
2921#ifdef CONFIG_MTRR
2922 if (mgp->mtrr >= 0)
2923 mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span);
2924#endif
Brice Goglinb10c0662006-06-08 10:25:00 -04002925 dma_free_coherent(&pdev->dev, sizeof(*mgp->fw_stats),
2926 mgp->fw_stats, mgp->fw_stats_bus);
Brice Goglin0da34b62006-05-23 06:10:15 -04002927
2928abort_with_cmd:
Brice Goglinb10c0662006-06-08 10:25:00 -04002929 dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
2930 mgp->cmd, mgp->cmd_bus);
Brice Goglin0da34b62006-05-23 06:10:15 -04002931
2932abort_with_netdev:
2933
2934 free_netdev(netdev);
2935 return status;
2936}
2937
2938/*
2939 * myri10ge_remove
2940 *
2941 * Does what is necessary to shutdown one Myrinet device. Called
2942 * once for each Myrinet card by the kernel when a module is
2943 * unloaded.
2944 */
2945static void myri10ge_remove(struct pci_dev *pdev)
2946{
2947 struct myri10ge_priv *mgp;
2948 struct net_device *netdev;
2949 size_t bytes;
2950
2951 mgp = pci_get_drvdata(pdev);
2952 if (mgp == NULL)
2953 return;
2954
2955 flush_scheduled_work();
2956 netdev = mgp->dev;
2957 unregister_netdev(netdev);
2958 free_irq(pdev->irq, mgp);
2959 if (mgp->msi_enabled)
2960 pci_disable_msi(pdev);
2961
2962 myri10ge_dummy_rdma(mgp, 0);
2963
2964 bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
Brice Goglinb10c0662006-06-08 10:25:00 -04002965 dma_free_coherent(&pdev->dev, bytes,
2966 mgp->rx_done.entry, mgp->rx_done.bus);
Brice Goglin0da34b62006-05-23 06:10:15 -04002967
2968 iounmap(mgp->sram);
2969
2970#ifdef CONFIG_MTRR
2971 if (mgp->mtrr >= 0)
2972 mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span);
2973#endif
Brice Goglinb10c0662006-06-08 10:25:00 -04002974 dma_free_coherent(&pdev->dev, sizeof(*mgp->fw_stats),
2975 mgp->fw_stats, mgp->fw_stats_bus);
Brice Goglin0da34b62006-05-23 06:10:15 -04002976
Brice Goglinb10c0662006-06-08 10:25:00 -04002977 dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
2978 mgp->cmd, mgp->cmd_bus);
Brice Goglin0da34b62006-05-23 06:10:15 -04002979
2980 free_netdev(netdev);
2981 pci_set_drvdata(pdev, NULL);
2982}
2983
Brice Goglinb10c0662006-06-08 10:25:00 -04002984#define PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E 0x0008
Brice Goglin0da34b62006-05-23 06:10:15 -04002985
2986static struct pci_device_id myri10ge_pci_tbl[] = {
Brice Goglinb10c0662006-06-08 10:25:00 -04002987 {PCI_DEVICE(PCI_VENDOR_ID_MYRICOM, PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E)},
Brice Goglin0da34b62006-05-23 06:10:15 -04002988 {0},
2989};
2990
2991static struct pci_driver myri10ge_driver = {
2992 .name = "myri10ge",
2993 .probe = myri10ge_probe,
2994 .remove = myri10ge_remove,
2995 .id_table = myri10ge_pci_tbl,
2996#ifdef CONFIG_PM
2997 .suspend = myri10ge_suspend,
2998 .resume = myri10ge_resume,
2999#endif
3000};
3001
3002static __init int myri10ge_init_module(void)
3003{
3004 printk(KERN_INFO "%s: Version %s\n", myri10ge_driver.name,
3005 MYRI10GE_VERSION_STR);
3006 return pci_register_driver(&myri10ge_driver);
3007}
3008
3009module_init(myri10ge_init_module);
3010
3011static __exit void myri10ge_cleanup_module(void)
3012{
3013 pci_unregister_driver(&myri10ge_driver);
3014}
3015
3016module_exit(myri10ge_cleanup_module);