diff options
Diffstat (limited to 'drivers/net/myri10ge')
-rw-r--r-- | drivers/net/myri10ge/Makefile | 5 | ||||
-rw-r--r-- | drivers/net/myri10ge/myri10ge.c | 2851 | ||||
-rw-r--r-- | drivers/net/myri10ge/myri10ge_mcp.h | 205 | ||||
-rw-r--r-- | drivers/net/myri10ge/myri10ge_mcp_gen_header.h | 58 |
4 files changed, 3119 insertions, 0 deletions
diff --git a/drivers/net/myri10ge/Makefile b/drivers/net/myri10ge/Makefile new file mode 100644 index 000000000000..5df891647aee --- /dev/null +++ b/drivers/net/myri10ge/Makefile | |||
@@ -0,0 +1,5 @@ | |||
1 | # | ||
2 | # Makefile for the Myricom Myri-10G ethernet driver | ||
3 | # | ||
4 | |||
5 | obj-$(CONFIG_MYRI10GE) += myri10ge.o | ||
diff --git a/drivers/net/myri10ge/myri10ge.c b/drivers/net/myri10ge/myri10ge.c new file mode 100644 index 000000000000..87933cba7e22 --- /dev/null +++ b/drivers/net/myri10ge/myri10ge.c | |||
@@ -0,0 +1,2851 @@ | |||
1 | /************************************************************************* | ||
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> | ||
47 | #include <linux/etherdevice.h> | ||
48 | #include <linux/if_ether.h> | ||
49 | #include <linux/if_vlan.h> | ||
50 | #include <linux/ip.h> | ||
51 | #include <linux/inet.h> | ||
52 | #include <linux/in.h> | ||
53 | #include <linux/ethtool.h> | ||
54 | #include <linux/firmware.h> | ||
55 | #include <linux/delay.h> | ||
56 | #include <linux/version.h> | ||
57 | #include <linux/timer.h> | ||
58 | #include <linux/vmalloc.h> | ||
59 | #include <linux/crc32.h> | ||
60 | #include <linux/moduleparam.h> | ||
61 | #include <linux/io.h> | ||
62 | #include <net/checksum.h> | ||
63 | #include <asm/byteorder.h> | ||
64 | #include <asm/io.h> | ||
65 | #include <asm/pci.h> | ||
66 | #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 | |||
74 | #define MYRI10GE_VERSION_STR "0.9.0" | ||
75 | |||
76 | MODULE_DESCRIPTION("Myricom 10G driver (10GbE)"); | ||
77 | MODULE_AUTHOR("Maintainer: help@myri.com"); | ||
78 | MODULE_VERSION(MYRI10GE_VERSION_STR); | ||
79 | MODULE_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 | |||
95 | struct myri10ge_rx_buffer_state { | ||
96 | struct sk_buff *skb; | ||
97 | DECLARE_PCI_UNMAP_ADDR(bus) | ||
98 | DECLARE_PCI_UNMAP_LEN(len) | ||
99 | }; | ||
100 | |||
101 | struct 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 | |||
108 | struct myri10ge_cmd { | ||
109 | u32 data0; | ||
110 | u32 data1; | ||
111 | u32 data2; | ||
112 | }; | ||
113 | |||
114 | struct 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 | |||
124 | struct 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 | |||
138 | struct myri10ge_rx_done { | ||
139 | struct mcp_slot *entry; | ||
140 | dma_addr_t bus; | ||
141 | int cnt; | ||
142 | int idx; | ||
143 | }; | ||
144 | |||
145 | struct 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; | ||
180 | int watchdog_resets; | ||
181 | int tx_linearized; | ||
182 | int pause; | ||
183 | char *fw_name; | ||
184 | char eeprom_strings[MYRI10GE_EEPROM_STRINGS_SIZE]; | ||
185 | char fw_version[128]; | ||
186 | u8 mac_addr[6]; /* eeprom mac address */ | ||
187 | unsigned long serial_number; | ||
188 | int vendor_specific_offset; | ||
189 | u32 devctl; | ||
190 | u16 msi_flags; | ||
191 | u32 pm_state[16]; | ||
192 | u32 read_dma; | ||
193 | u32 write_dma; | ||
194 | u32 read_write_dma; | ||
195 | }; | ||
196 | |||
197 | static char *myri10ge_fw_unaligned = "myri10ge_ethp_z8e.dat"; | ||
198 | static char *myri10ge_fw_aligned = "myri10ge_eth_z8e.dat"; | ||
199 | |||
200 | static char *myri10ge_fw_name = NULL; | ||
201 | module_param(myri10ge_fw_name, charp, S_IRUGO | S_IWUSR); | ||
202 | MODULE_PARM_DESC(myri10ge_fw_name, "Firmware image name\n"); | ||
203 | |||
204 | static int myri10ge_ecrc_enable = 1; | ||
205 | module_param(myri10ge_ecrc_enable, int, S_IRUGO); | ||
206 | MODULE_PARM_DESC(myri10ge_ecrc_enable, "Enable Extended CRC on PCI-E\n"); | ||
207 | |||
208 | static int myri10ge_max_intr_slots = 1024; | ||
209 | module_param(myri10ge_max_intr_slots, int, S_IRUGO); | ||
210 | MODULE_PARM_DESC(myri10ge_max_intr_slots, "Interrupt queue slots\n"); | ||
211 | |||
212 | static int myri10ge_small_bytes = -1; /* -1 == auto */ | ||
213 | module_param(myri10ge_small_bytes, int, S_IRUGO | S_IWUSR); | ||
214 | MODULE_PARM_DESC(myri10ge_small_bytes, "Threshold of small packets\n"); | ||
215 | |||
216 | static int myri10ge_msi = 1; /* enable msi by default */ | ||
217 | module_param(myri10ge_msi, int, S_IRUGO); | ||
218 | MODULE_PARM_DESC(myri10ge_msi, "Enable Message Signalled Interrupts\n"); | ||
219 | |||
220 | static int myri10ge_intr_coal_delay = 25; | ||
221 | module_param(myri10ge_intr_coal_delay, int, S_IRUGO); | ||
222 | MODULE_PARM_DESC(myri10ge_intr_coal_delay, "Interrupt coalescing delay\n"); | ||
223 | |||
224 | static int myri10ge_flow_control = 1; | ||
225 | module_param(myri10ge_flow_control, int, S_IRUGO); | ||
226 | MODULE_PARM_DESC(myri10ge_flow_control, "Pause parameter\n"); | ||
227 | |||
228 | static int myri10ge_deassert_wait = 1; | ||
229 | module_param(myri10ge_deassert_wait, int, S_IRUGO | S_IWUSR); | ||
230 | MODULE_PARM_DESC(myri10ge_deassert_wait, | ||
231 | "Wait when deasserting legacy interrupts\n"); | ||
232 | |||
233 | static int myri10ge_force_firmware = 0; | ||
234 | module_param(myri10ge_force_firmware, int, S_IRUGO); | ||
235 | MODULE_PARM_DESC(myri10ge_force_firmware, | ||
236 | "Force firmware to assume aligned completions\n"); | ||
237 | |||
238 | static int myri10ge_skb_cross_4k = 0; | ||
239 | module_param(myri10ge_skb_cross_4k, int, S_IRUGO | S_IWUSR); | ||
240 | MODULE_PARM_DESC(myri10ge_skb_cross_4k, | ||
241 | "Can a small skb cross a 4KB boundary?\n"); | ||
242 | |||
243 | static int myri10ge_initial_mtu = MYRI10GE_MAX_ETHER_MTU - ETH_HLEN; | ||
244 | module_param(myri10ge_initial_mtu, int, S_IRUGO); | ||
245 | MODULE_PARM_DESC(myri10ge_initial_mtu, "Initial MTU\n"); | ||
246 | |||
247 | static int myri10ge_napi_weight = 64; | ||
248 | module_param(myri10ge_napi_weight, int, S_IRUGO); | ||
249 | MODULE_PARM_DESC(myri10ge_napi_weight, "Set NAPI weight\n"); | ||
250 | |||
251 | static int myri10ge_watchdog_timeout = 1; | ||
252 | module_param(myri10ge_watchdog_timeout, int, S_IRUGO); | ||
253 | MODULE_PARM_DESC(myri10ge_watchdog_timeout, "Set watchdog timeout\n"); | ||
254 | |||
255 | static int myri10ge_max_irq_loops = 1048576; | ||
256 | module_param(myri10ge_max_irq_loops, int, S_IRUGO); | ||
257 | MODULE_PARM_DESC(myri10ge_max_irq_loops, | ||
258 | "Set stuck legacy IRQ detection threshold\n"); | ||
259 | |||
260 | #define MYRI10GE_FW_OFFSET 1024*1024 | ||
261 | #define MYRI10GE_HIGHPART_TO_U32(X) \ | ||
262 | (sizeof (X) == 8) ? ((u32)((u64)(X) >> 32)) : (0) | ||
263 | #define MYRI10GE_LOWPART_TO_U32(X) ((u32)(X)) | ||
264 | |||
265 | #define myri10ge_pio_copy(to,from,size) __iowrite64_copy(to,from,size/8) | ||
266 | |||
267 | static int | ||
268 | myri10ge_send_cmd(struct myri10ge_priv *mgp, u32 cmd, | ||
269 | struct myri10ge_cmd *data, int atomic) | ||
270 | { | ||
271 | struct mcp_cmd *buf; | ||
272 | char buf_bytes[sizeof(*buf) + 8]; | ||
273 | struct mcp_cmd_response *response = mgp->cmd; | ||
274 | char __iomem *cmd_addr = mgp->sram + MXGEFW_CMD_OFFSET; | ||
275 | u32 dma_low, dma_high, result, value; | ||
276 | int sleep_total = 0; | ||
277 | |||
278 | /* ensure buf is aligned to 8 bytes */ | ||
279 | buf = (struct mcp_cmd *)ALIGN((unsigned long)buf_bytes, 8); | ||
280 | |||
281 | buf->data0 = htonl(data->data0); | ||
282 | buf->data1 = htonl(data->data1); | ||
283 | buf->data2 = htonl(data->data2); | ||
284 | buf->cmd = htonl(cmd); | ||
285 | dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus); | ||
286 | dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus); | ||
287 | |||
288 | buf->response_addr.low = htonl(dma_low); | ||
289 | buf->response_addr.high = htonl(dma_high); | ||
290 | response->result = MYRI10GE_NO_RESPONSE_RESULT; | ||
291 | mb(); | ||
292 | myri10ge_pio_copy(cmd_addr, buf, sizeof(*buf)); | ||
293 | |||
294 | /* wait up to 15ms. Longest command is the DMA benchmark, | ||
295 | * which is capped at 5ms, but runs from a timeout handler | ||
296 | * that runs every 7.8ms. So a 15ms timeout leaves us with | ||
297 | * a 2.2ms margin | ||
298 | */ | ||
299 | if (atomic) { | ||
300 | /* if atomic is set, do not sleep, | ||
301 | * and try to get the completion quickly | ||
302 | * (1ms will be enough for those commands) */ | ||
303 | for (sleep_total = 0; | ||
304 | sleep_total < 1000 | ||
305 | && response->result == MYRI10GE_NO_RESPONSE_RESULT; | ||
306 | sleep_total += 10) | ||
307 | udelay(10); | ||
308 | } else { | ||
309 | /* use msleep for most command */ | ||
310 | for (sleep_total = 0; | ||
311 | sleep_total < 15 | ||
312 | && response->result == MYRI10GE_NO_RESPONSE_RESULT; | ||
313 | sleep_total++) | ||
314 | msleep(1); | ||
315 | } | ||
316 | |||
317 | result = ntohl(response->result); | ||
318 | value = ntohl(response->data); | ||
319 | if (result != MYRI10GE_NO_RESPONSE_RESULT) { | ||
320 | if (result == 0) { | ||
321 | data->data0 = value; | ||
322 | return 0; | ||
323 | } else { | ||
324 | dev_err(&mgp->pdev->dev, | ||
325 | "command %d failed, result = %d\n", | ||
326 | cmd, result); | ||
327 | return -ENXIO; | ||
328 | } | ||
329 | } | ||
330 | |||
331 | dev_err(&mgp->pdev->dev, "command %d timed out, result = %d\n", | ||
332 | cmd, result); | ||
333 | return -EAGAIN; | ||
334 | } | ||
335 | |||
336 | /* | ||
337 | * The eeprom strings on the lanaiX have the format | ||
338 | * SN=x\0 | ||
339 | * MAC=x:x:x:x:x:x\0 | ||
340 | * PT:ddd mmm xx xx:xx:xx xx\0 | ||
341 | * PV:ddd mmm xx xx:xx:xx xx\0 | ||
342 | */ | ||
343 | static int myri10ge_read_mac_addr(struct myri10ge_priv *mgp) | ||
344 | { | ||
345 | char *ptr, *limit; | ||
346 | int i; | ||
347 | |||
348 | ptr = mgp->eeprom_strings; | ||
349 | limit = mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE; | ||
350 | |||
351 | while (*ptr != '\0' && ptr < limit) { | ||
352 | if (memcmp(ptr, "MAC=", 4) == 0) { | ||
353 | ptr += 4; | ||
354 | mgp->mac_addr_string = ptr; | ||
355 | for (i = 0; i < 6; i++) { | ||
356 | if ((ptr + 2) > limit) | ||
357 | goto abort; | ||
358 | mgp->mac_addr[i] = | ||
359 | simple_strtoul(ptr, &ptr, 16); | ||
360 | ptr += 1; | ||
361 | } | ||
362 | } | ||
363 | if (memcmp((const void *)ptr, "SN=", 3) == 0) { | ||
364 | ptr += 3; | ||
365 | mgp->serial_number = simple_strtoul(ptr, &ptr, 10); | ||
366 | } | ||
367 | while (ptr < limit && *ptr++) ; | ||
368 | } | ||
369 | |||
370 | return 0; | ||
371 | |||
372 | abort: | ||
373 | dev_err(&mgp->pdev->dev, "failed to parse eeprom_strings\n"); | ||
374 | return -ENXIO; | ||
375 | } | ||
376 | |||
377 | /* | ||
378 | * Enable or disable periodic RDMAs from the host to make certain | ||
379 | * chipsets resend dropped PCIe messages | ||
380 | */ | ||
381 | |||
382 | static void myri10ge_dummy_rdma(struct myri10ge_priv *mgp, int enable) | ||
383 | { | ||
384 | char __iomem *submit; | ||
385 | u32 buf[16]; | ||
386 | u32 dma_low, dma_high; | ||
387 | int i; | ||
388 | |||
389 | /* clear confirmation addr */ | ||
390 | mgp->cmd->data = 0; | ||
391 | mb(); | ||
392 | |||
393 | /* send a rdma command to the PCIe engine, and wait for the | ||
394 | * response in the confirmation address. The firmware should | ||
395 | * write a -1 there to indicate it is alive and well | ||
396 | */ | ||
397 | dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus); | ||
398 | dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus); | ||
399 | |||
400 | buf[0] = htonl(dma_high); /* confirm addr MSW */ | ||
401 | buf[1] = htonl(dma_low); /* confirm addr LSW */ | ||
402 | buf[2] = htonl(MYRI10GE_NO_CONFIRM_DATA); /* confirm data */ | ||
403 | buf[3] = htonl(dma_high); /* dummy addr MSW */ | ||
404 | buf[4] = htonl(dma_low); /* dummy addr LSW */ | ||
405 | buf[5] = htonl(enable); /* enable? */ | ||
406 | |||
407 | submit = mgp->sram + 0xfc01c0; | ||
408 | |||
409 | myri10ge_pio_copy(submit, &buf, sizeof(buf)); | ||
410 | for (i = 0; mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 20; i++) | ||
411 | msleep(1); | ||
412 | if (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA) | ||
413 | dev_err(&mgp->pdev->dev, "dummy rdma %s failed\n", | ||
414 | (enable ? "enable" : "disable")); | ||
415 | } | ||
416 | |||
417 | static int | ||
418 | myri10ge_validate_firmware(struct myri10ge_priv *mgp, | ||
419 | struct mcp_gen_header *hdr) | ||
420 | { | ||
421 | struct device *dev = &mgp->pdev->dev; | ||
422 | int major, minor; | ||
423 | |||
424 | /* check firmware type */ | ||
425 | if (ntohl(hdr->mcp_type) != MCP_TYPE_ETH) { | ||
426 | dev_err(dev, "Bad firmware type: 0x%x\n", ntohl(hdr->mcp_type)); | ||
427 | return -EINVAL; | ||
428 | } | ||
429 | |||
430 | /* save firmware version for ethtool */ | ||
431 | strncpy(mgp->fw_version, hdr->version, sizeof(mgp->fw_version)); | ||
432 | |||
433 | sscanf(mgp->fw_version, "%d.%d", &major, &minor); | ||
434 | |||
435 | if (!(major == MXGEFW_VERSION_MAJOR && minor == MXGEFW_VERSION_MINOR)) { | ||
436 | dev_err(dev, "Found firmware version %s\n", mgp->fw_version); | ||
437 | dev_err(dev, "Driver needs %d.%d\n", MXGEFW_VERSION_MAJOR, | ||
438 | MXGEFW_VERSION_MINOR); | ||
439 | return -EINVAL; | ||
440 | } | ||
441 | return 0; | ||
442 | } | ||
443 | |||
444 | static int myri10ge_load_hotplug_firmware(struct myri10ge_priv *mgp, u32 * size) | ||
445 | { | ||
446 | unsigned crc, reread_crc; | ||
447 | const struct firmware *fw; | ||
448 | struct device *dev = &mgp->pdev->dev; | ||
449 | struct mcp_gen_header *hdr; | ||
450 | size_t hdr_offset; | ||
451 | int status; | ||
452 | |||
453 | if ((status = request_firmware(&fw, mgp->fw_name, dev)) < 0) { | ||
454 | dev_err(dev, "Unable to load %s firmware image via hotplug\n", | ||
455 | mgp->fw_name); | ||
456 | status = -EINVAL; | ||
457 | goto abort_with_nothing; | ||
458 | } | ||
459 | |||
460 | /* check size */ | ||
461 | |||
462 | if (fw->size >= mgp->sram_size - MYRI10GE_FW_OFFSET || | ||
463 | fw->size < MCP_HEADER_PTR_OFFSET + 4) { | ||
464 | dev_err(dev, "Firmware size invalid:%d\n", (int)fw->size); | ||
465 | status = -EINVAL; | ||
466 | goto abort_with_fw; | ||
467 | } | ||
468 | |||
469 | /* check id */ | ||
470 | hdr_offset = ntohl(*(u32 *) (fw->data + MCP_HEADER_PTR_OFFSET)); | ||
471 | if ((hdr_offset & 3) || hdr_offset + sizeof(*hdr) > fw->size) { | ||
472 | dev_err(dev, "Bad firmware file\n"); | ||
473 | status = -EINVAL; | ||
474 | goto abort_with_fw; | ||
475 | } | ||
476 | hdr = (void *)(fw->data + hdr_offset); | ||
477 | |||
478 | status = myri10ge_validate_firmware(mgp, hdr); | ||
479 | if (status != 0) | ||
480 | goto abort_with_fw; | ||
481 | |||
482 | crc = crc32(~0, fw->data, fw->size); | ||
483 | memcpy_toio(mgp->sram + MYRI10GE_FW_OFFSET, fw->data, fw->size); | ||
484 | /* corruption checking is good for parity recovery and buggy chipset */ | ||
485 | memcpy_fromio(fw->data, mgp->sram + MYRI10GE_FW_OFFSET, fw->size); | ||
486 | reread_crc = crc32(~0, fw->data, fw->size); | ||
487 | if (crc != reread_crc) { | ||
488 | dev_err(dev, "CRC failed(fw-len=%u), got 0x%x (expect 0x%x)\n", | ||
489 | (unsigned)fw->size, reread_crc, crc); | ||
490 | status = -EIO; | ||
491 | goto abort_with_fw; | ||
492 | } | ||
493 | *size = (u32) fw->size; | ||
494 | |||
495 | abort_with_fw: | ||
496 | release_firmware(fw); | ||
497 | |||
498 | abort_with_nothing: | ||
499 | return status; | ||
500 | } | ||
501 | |||
502 | static int myri10ge_adopt_running_firmware(struct myri10ge_priv *mgp) | ||
503 | { | ||
504 | struct mcp_gen_header *hdr; | ||
505 | struct device *dev = &mgp->pdev->dev; | ||
506 | const size_t bytes = sizeof(struct mcp_gen_header); | ||
507 | size_t hdr_offset; | ||
508 | int status; | ||
509 | |||
510 | /* find running firmware header */ | ||
511 | hdr_offset = ntohl(__raw_readl(mgp->sram + MCP_HEADER_PTR_OFFSET)); | ||
512 | |||
513 | if ((hdr_offset & 3) || hdr_offset + sizeof(*hdr) > mgp->sram_size) { | ||
514 | dev_err(dev, "Running firmware has bad header offset (%d)\n", | ||
515 | (int)hdr_offset); | ||
516 | return -EIO; | ||
517 | } | ||
518 | |||
519 | /* copy header of running firmware from SRAM to host memory to | ||
520 | * validate firmware */ | ||
521 | hdr = kmalloc(bytes, GFP_KERNEL); | ||
522 | if (hdr == NULL) { | ||
523 | dev_err(dev, "could not malloc firmware hdr\n"); | ||
524 | return -ENOMEM; | ||
525 | } | ||
526 | memcpy_fromio(hdr, mgp->sram + hdr_offset, bytes); | ||
527 | status = myri10ge_validate_firmware(mgp, hdr); | ||
528 | kfree(hdr); | ||
529 | return status; | ||
530 | } | ||
531 | |||
532 | static int myri10ge_load_firmware(struct myri10ge_priv *mgp) | ||
533 | { | ||
534 | char __iomem *submit; | ||
535 | u32 buf[16]; | ||
536 | u32 dma_low, dma_high, size; | ||
537 | int status, i; | ||
538 | |||
539 | status = myri10ge_load_hotplug_firmware(mgp, &size); | ||
540 | if (status) { | ||
541 | dev_warn(&mgp->pdev->dev, "hotplug firmware loading failed\n"); | ||
542 | |||
543 | /* Do not attempt to adopt firmware if there | ||
544 | * was a bad crc */ | ||
545 | if (status == -EIO) | ||
546 | return status; | ||
547 | |||
548 | status = myri10ge_adopt_running_firmware(mgp); | ||
549 | if (status != 0) { | ||
550 | dev_err(&mgp->pdev->dev, | ||
551 | "failed to adopt running firmware\n"); | ||
552 | return status; | ||
553 | } | ||
554 | dev_info(&mgp->pdev->dev, | ||
555 | "Successfully adopted running firmware\n"); | ||
556 | if (mgp->tx.boundary == 4096) { | ||
557 | dev_warn(&mgp->pdev->dev, | ||
558 | "Using firmware currently running on NIC" | ||
559 | ". For optimal\n"); | ||
560 | dev_warn(&mgp->pdev->dev, | ||
561 | "performance consider loading optimized " | ||
562 | "firmware\n"); | ||
563 | dev_warn(&mgp->pdev->dev, "via hotplug\n"); | ||
564 | } | ||
565 | |||
566 | mgp->fw_name = "adopted"; | ||
567 | mgp->tx.boundary = 2048; | ||
568 | return status; | ||
569 | } | ||
570 | |||
571 | /* clear confirmation addr */ | ||
572 | mgp->cmd->data = 0; | ||
573 | mb(); | ||
574 | |||
575 | /* send a reload command to the bootstrap MCP, and wait for the | ||
576 | * response in the confirmation address. The firmware should | ||
577 | * write a -1 there to indicate it is alive and well | ||
578 | */ | ||
579 | dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus); | ||
580 | dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus); | ||
581 | |||
582 | buf[0] = htonl(dma_high); /* confirm addr MSW */ | ||
583 | buf[1] = htonl(dma_low); /* confirm addr LSW */ | ||
584 | buf[2] = htonl(MYRI10GE_NO_CONFIRM_DATA); /* confirm data */ | ||
585 | |||
586 | /* FIX: All newest firmware should un-protect the bottom of | ||
587 | * the sram before handoff. However, the very first interfaces | ||
588 | * do not. Therefore the handoff copy must skip the first 8 bytes | ||
589 | */ | ||
590 | buf[3] = htonl(MYRI10GE_FW_OFFSET + 8); /* where the code starts */ | ||
591 | buf[4] = htonl(size - 8); /* length of code */ | ||
592 | buf[5] = htonl(8); /* where to copy to */ | ||
593 | buf[6] = htonl(0); /* where to jump to */ | ||
594 | |||
595 | submit = mgp->sram + 0xfc0000; | ||
596 | |||
597 | myri10ge_pio_copy(submit, &buf, sizeof(buf)); | ||
598 | mb(); | ||
599 | msleep(1); | ||
600 | mb(); | ||
601 | i = 0; | ||
602 | while (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 20) { | ||
603 | msleep(1); | ||
604 | i++; | ||
605 | } | ||
606 | if (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA) { | ||
607 | dev_err(&mgp->pdev->dev, "handoff failed\n"); | ||
608 | return -ENXIO; | ||
609 | } | ||
610 | dev_info(&mgp->pdev->dev, "handoff confirmed\n"); | ||
611 | myri10ge_dummy_rdma(mgp, mgp->tx.boundary != 4096); | ||
612 | |||
613 | return 0; | ||
614 | } | ||
615 | |||
616 | static int myri10ge_update_mac_address(struct myri10ge_priv *mgp, u8 * addr) | ||
617 | { | ||
618 | struct myri10ge_cmd cmd; | ||
619 | int status; | ||
620 | |||
621 | cmd.data0 = ((addr[0] << 24) | (addr[1] << 16) | ||
622 | | (addr[2] << 8) | addr[3]); | ||
623 | |||
624 | cmd.data1 = ((addr[4] << 8) | (addr[5])); | ||
625 | |||
626 | status = myri10ge_send_cmd(mgp, MXGEFW_SET_MAC_ADDRESS, &cmd, 0); | ||
627 | return status; | ||
628 | } | ||
629 | |||
630 | static int myri10ge_change_pause(struct myri10ge_priv *mgp, int pause) | ||
631 | { | ||
632 | struct myri10ge_cmd cmd; | ||
633 | int status, ctl; | ||
634 | |||
635 | ctl = pause ? MXGEFW_ENABLE_FLOW_CONTROL : MXGEFW_DISABLE_FLOW_CONTROL; | ||
636 | status = myri10ge_send_cmd(mgp, ctl, &cmd, 0); | ||
637 | |||
638 | if (status) { | ||
639 | printk(KERN_ERR | ||
640 | "myri10ge: %s: Failed to set flow control mode\n", | ||
641 | mgp->dev->name); | ||
642 | return status; | ||
643 | } | ||
644 | mgp->pause = pause; | ||
645 | return 0; | ||
646 | } | ||
647 | |||
648 | static void | ||
649 | myri10ge_change_promisc(struct myri10ge_priv *mgp, int promisc, int atomic) | ||
650 | { | ||
651 | struct myri10ge_cmd cmd; | ||
652 | int status, ctl; | ||
653 | |||
654 | ctl = promisc ? MXGEFW_ENABLE_PROMISC : MXGEFW_DISABLE_PROMISC; | ||
655 | status = myri10ge_send_cmd(mgp, ctl, &cmd, atomic); | ||
656 | if (status) | ||
657 | printk(KERN_ERR "myri10ge: %s: Failed to set promisc mode\n", | ||
658 | mgp->dev->name); | ||
659 | } | ||
660 | |||
661 | static int myri10ge_reset(struct myri10ge_priv *mgp) | ||
662 | { | ||
663 | struct myri10ge_cmd cmd; | ||
664 | int status; | ||
665 | size_t bytes; | ||
666 | u32 len; | ||
667 | |||
668 | /* try to send a reset command to the card to see if it | ||
669 | * is alive */ | ||
670 | memset(&cmd, 0, sizeof(cmd)); | ||
671 | status = myri10ge_send_cmd(mgp, MXGEFW_CMD_RESET, &cmd, 0); | ||
672 | if (status != 0) { | ||
673 | dev_err(&mgp->pdev->dev, "failed reset\n"); | ||
674 | return -ENXIO; | ||
675 | } | ||
676 | |||
677 | /* Now exchange information about interrupts */ | ||
678 | |||
679 | bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry); | ||
680 | memset(mgp->rx_done.entry, 0, bytes); | ||
681 | cmd.data0 = (u32) bytes; | ||
682 | status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_SIZE, &cmd, 0); | ||
683 | cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus); | ||
684 | cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus); | ||
685 | status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_DMA, &cmd, 0); | ||
686 | |||
687 | status |= | ||
688 | myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_IRQ_ACK_OFFSET, &cmd, 0); | ||
689 | mgp->irq_claim = (__iomem u32 *) (mgp->sram + cmd.data0); | ||
690 | if (!mgp->msi_enabled) { | ||
691 | status |= myri10ge_send_cmd | ||
692 | (mgp, MXGEFW_CMD_GET_IRQ_DEASSERT_OFFSET, &cmd, 0); | ||
693 | mgp->irq_deassert = (__iomem u32 *) (mgp->sram + cmd.data0); | ||
694 | |||
695 | } | ||
696 | status |= myri10ge_send_cmd | ||
697 | (mgp, MXGEFW_CMD_GET_INTR_COAL_DELAY_OFFSET, &cmd, 0); | ||
698 | mgp->intr_coal_delay_ptr = (__iomem u32 *) (mgp->sram + cmd.data0); | ||
699 | if (status != 0) { | ||
700 | dev_err(&mgp->pdev->dev, "failed set interrupt parameters\n"); | ||
701 | return status; | ||
702 | } | ||
703 | __raw_writel(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr); | ||
704 | |||
705 | /* Run a small DMA test. | ||
706 | * The magic multipliers to the length tell the firmware | ||
707 | * to do DMA read, write, or read+write tests. The | ||
708 | * results are returned in cmd.data0. The upper 16 | ||
709 | * bits or the return is the number of transfers completed. | ||
710 | * The lower 16 bits is the time in 0.5us ticks that the | ||
711 | * transfers took to complete. | ||
712 | */ | ||
713 | |||
714 | len = mgp->tx.boundary; | ||
715 | |||
716 | cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus); | ||
717 | cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus); | ||
718 | cmd.data2 = len * 0x10000; | ||
719 | status = myri10ge_send_cmd(mgp, MXGEFW_DMA_TEST, &cmd, 0); | ||
720 | if (status == 0) | ||
721 | mgp->read_dma = ((cmd.data0 >> 16) * len * 2) / | ||
722 | (cmd.data0 & 0xffff); | ||
723 | else | ||
724 | dev_warn(&mgp->pdev->dev, "DMA read benchmark failed: %d\n", | ||
725 | status); | ||
726 | cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus); | ||
727 | cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus); | ||
728 | cmd.data2 = len * 0x1; | ||
729 | status = myri10ge_send_cmd(mgp, MXGEFW_DMA_TEST, &cmd, 0); | ||
730 | if (status == 0) | ||
731 | mgp->write_dma = ((cmd.data0 >> 16) * len * 2) / | ||
732 | (cmd.data0 & 0xffff); | ||
733 | else | ||
734 | dev_warn(&mgp->pdev->dev, "DMA write benchmark failed: %d\n", | ||
735 | status); | ||
736 | |||
737 | cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus); | ||
738 | cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus); | ||
739 | cmd.data2 = len * 0x10001; | ||
740 | status = myri10ge_send_cmd(mgp, MXGEFW_DMA_TEST, &cmd, 0); | ||
741 | if (status == 0) | ||
742 | mgp->read_write_dma = ((cmd.data0 >> 16) * len * 2 * 2) / | ||
743 | (cmd.data0 & 0xffff); | ||
744 | else | ||
745 | dev_warn(&mgp->pdev->dev, | ||
746 | "DMA read/write benchmark failed: %d\n", status); | ||
747 | |||
748 | memset(mgp->rx_done.entry, 0, bytes); | ||
749 | |||
750 | /* reset mcp/driver shared state back to 0 */ | ||
751 | mgp->tx.req = 0; | ||
752 | mgp->tx.done = 0; | ||
753 | mgp->tx.pkt_start = 0; | ||
754 | mgp->tx.pkt_done = 0; | ||
755 | mgp->rx_big.cnt = 0; | ||
756 | mgp->rx_small.cnt = 0; | ||
757 | mgp->rx_done.idx = 0; | ||
758 | mgp->rx_done.cnt = 0; | ||
759 | status = myri10ge_update_mac_address(mgp, mgp->dev->dev_addr); | ||
760 | myri10ge_change_promisc(mgp, 0, 0); | ||
761 | myri10ge_change_pause(mgp, mgp->pause); | ||
762 | return status; | ||
763 | } | ||
764 | |||
765 | static inline void | ||
766 | myri10ge_submit_8rx(struct mcp_kreq_ether_recv __iomem * dst, | ||
767 | struct mcp_kreq_ether_recv *src) | ||
768 | { | ||
769 | u32 low; | ||
770 | |||
771 | low = src->addr_low; | ||
772 | src->addr_low = DMA_32BIT_MASK; | ||
773 | myri10ge_pio_copy(dst, src, 8 * sizeof(*src)); | ||
774 | mb(); | ||
775 | src->addr_low = low; | ||
776 | __raw_writel(low, &dst->addr_low); | ||
777 | mb(); | ||
778 | } | ||
779 | |||
780 | /* | ||
781 | * Set of routunes to get a new receive buffer. Any buffer which | ||
782 | * crosses a 4KB boundary must start on a 4KB boundary due to PCIe | ||
783 | * wdma restrictions. We also try to align any smaller allocation to | ||
784 | * at least a 16 byte boundary for efficiency. We assume the linux | ||
785 | * memory allocator works by powers of 2, and will not return memory | ||
786 | * smaller than 2KB which crosses a 4KB boundary. If it does, we fall | ||
787 | * back to allocating 2x as much space as required. | ||
788 | * | ||
789 | * We intend to replace large (>4KB) skb allocations by using | ||
790 | * pages directly and building a fraglist in the near future. | ||
791 | */ | ||
792 | |||
793 | static inline struct sk_buff *myri10ge_alloc_big(int bytes) | ||
794 | { | ||
795 | struct sk_buff *skb; | ||
796 | unsigned long data, roundup; | ||
797 | |||
798 | skb = dev_alloc_skb(bytes + 4096 + MXGEFW_PAD); | ||
799 | if (skb == NULL) | ||
800 | return NULL; | ||
801 | |||
802 | /* Correct skb->truesize so that socket buffer | ||
803 | * accounting is not confused the rounding we must | ||
804 | * do to satisfy alignment constraints. | ||
805 | */ | ||
806 | skb->truesize -= 4096; | ||
807 | |||
808 | data = (unsigned long)(skb->data); | ||
809 | roundup = (-data) & (4095); | ||
810 | skb_reserve(skb, roundup); | ||
811 | return skb; | ||
812 | } | ||
813 | |||
814 | /* Allocate 2x as much space as required and use whichever portion | ||
815 | * does not cross a 4KB boundary */ | ||
816 | static inline struct sk_buff *myri10ge_alloc_small_safe(unsigned int bytes) | ||
817 | { | ||
818 | struct sk_buff *skb; | ||
819 | unsigned long data, boundary; | ||
820 | |||
821 | skb = dev_alloc_skb(2 * (bytes + MXGEFW_PAD) - 1); | ||
822 | if (unlikely(skb == NULL)) | ||
823 | return NULL; | ||
824 | |||
825 | /* Correct skb->truesize so that socket buffer | ||
826 | * accounting is not confused the rounding we must | ||
827 | * do to satisfy alignment constraints. | ||
828 | */ | ||
829 | skb->truesize -= bytes + MXGEFW_PAD; | ||
830 | |||
831 | data = (unsigned long)(skb->data); | ||
832 | boundary = (data + 4095UL) & ~4095UL; | ||
833 | if ((boundary - data) >= (bytes + MXGEFW_PAD)) | ||
834 | return skb; | ||
835 | |||
836 | skb_reserve(skb, boundary - data); | ||
837 | return skb; | ||
838 | } | ||
839 | |||
840 | /* Allocate just enough space, and verify that the allocated | ||
841 | * space does not cross a 4KB boundary */ | ||
842 | static inline struct sk_buff *myri10ge_alloc_small(int bytes) | ||
843 | { | ||
844 | struct sk_buff *skb; | ||
845 | unsigned long roundup, data, end; | ||
846 | |||
847 | skb = dev_alloc_skb(bytes + 16 + MXGEFW_PAD); | ||
848 | if (unlikely(skb == NULL)) | ||
849 | return NULL; | ||
850 | |||
851 | /* Round allocated buffer to 16 byte boundary */ | ||
852 | data = (unsigned long)(skb->data); | ||
853 | roundup = (-data) & 15UL; | ||
854 | skb_reserve(skb, roundup); | ||
855 | /* Verify that the data buffer does not cross a page boundary */ | ||
856 | data = (unsigned long)(skb->data); | ||
857 | end = data + bytes + MXGEFW_PAD - 1; | ||
858 | if (unlikely(((end >> 12) != (data >> 12)) && (data & 4095UL))) { | ||
859 | printk(KERN_NOTICE | ||
860 | "myri10ge_alloc_small: small skb crossed 4KB boundary\n"); | ||
861 | myri10ge_skb_cross_4k = 1; | ||
862 | dev_kfree_skb_any(skb); | ||
863 | skb = myri10ge_alloc_small_safe(bytes); | ||
864 | } | ||
865 | return skb; | ||
866 | } | ||
867 | |||
868 | static inline int | ||
869 | myri10ge_getbuf(struct myri10ge_rx_buf *rx, struct pci_dev *pdev, int bytes, | ||
870 | int idx) | ||
871 | { | ||
872 | struct sk_buff *skb; | ||
873 | dma_addr_t bus; | ||
874 | int len, retval = 0; | ||
875 | |||
876 | bytes += VLAN_HLEN; /* account for 802.1q vlan tag */ | ||
877 | |||
878 | if ((bytes + MXGEFW_PAD) > (4096 - 16) /* linux overhead */ ) | ||
879 | skb = myri10ge_alloc_big(bytes); | ||
880 | else if (myri10ge_skb_cross_4k) | ||
881 | skb = myri10ge_alloc_small_safe(bytes); | ||
882 | else | ||
883 | skb = myri10ge_alloc_small(bytes); | ||
884 | |||
885 | if (unlikely(skb == NULL)) { | ||
886 | rx->alloc_fail++; | ||
887 | retval = -ENOBUFS; | ||
888 | goto done; | ||
889 | } | ||
890 | |||
891 | /* set len so that it only covers the area we | ||
892 | * need mapped for DMA */ | ||
893 | len = bytes + MXGEFW_PAD; | ||
894 | |||
895 | bus = pci_map_single(pdev, skb->data, len, PCI_DMA_FROMDEVICE); | ||
896 | rx->info[idx].skb = skb; | ||
897 | pci_unmap_addr_set(&rx->info[idx], bus, bus); | ||
898 | pci_unmap_len_set(&rx->info[idx], len, len); | ||
899 | rx->shadow[idx].addr_low = htonl(MYRI10GE_LOWPART_TO_U32(bus)); | ||
900 | rx->shadow[idx].addr_high = htonl(MYRI10GE_HIGHPART_TO_U32(bus)); | ||
901 | |||
902 | done: | ||
903 | /* copy 8 descriptors (64-bytes) to the mcp at a time */ | ||
904 | if ((idx & 7) == 7) { | ||
905 | if (rx->wc_fifo == NULL) | ||
906 | myri10ge_submit_8rx(&rx->lanai[idx - 7], | ||
907 | &rx->shadow[idx - 7]); | ||
908 | else { | ||
909 | mb(); | ||
910 | myri10ge_pio_copy(rx->wc_fifo, | ||
911 | &rx->shadow[idx - 7], 64); | ||
912 | } | ||
913 | } | ||
914 | return retval; | ||
915 | } | ||
916 | |||
917 | static inline void myri10ge_vlan_ip_csum(struct sk_buff *skb, u16 hw_csum) | ||
918 | { | ||
919 | struct vlan_hdr *vh = (struct vlan_hdr *)(skb->data); | ||
920 | |||
921 | if ((skb->protocol == ntohs(ETH_P_8021Q)) && | ||
922 | (vh->h_vlan_encapsulated_proto == htons(ETH_P_IP) || | ||
923 | vh->h_vlan_encapsulated_proto == htons(ETH_P_IPV6))) { | ||
924 | skb->csum = hw_csum; | ||
925 | skb->ip_summed = CHECKSUM_HW; | ||
926 | } | ||
927 | } | ||
928 | |||
929 | static inline unsigned long | ||
930 | myri10ge_rx_done(struct myri10ge_priv *mgp, struct myri10ge_rx_buf *rx, | ||
931 | int bytes, int len, int csum) | ||
932 | { | ||
933 | dma_addr_t bus; | ||
934 | struct sk_buff *skb; | ||
935 | int idx, unmap_len; | ||
936 | |||
937 | idx = rx->cnt & rx->mask; | ||
938 | rx->cnt++; | ||
939 | |||
940 | /* save a pointer to the received skb */ | ||
941 | skb = rx->info[idx].skb; | ||
942 | bus = pci_unmap_addr(&rx->info[idx], bus); | ||
943 | unmap_len = pci_unmap_len(&rx->info[idx], len); | ||
944 | |||
945 | /* try to replace the received skb */ | ||
946 | if (myri10ge_getbuf(rx, mgp->pdev, bytes, idx)) { | ||
947 | /* drop the frame -- the old skbuf is re-cycled */ | ||
948 | mgp->stats.rx_dropped += 1; | ||
949 | return 0; | ||
950 | } | ||
951 | |||
952 | /* unmap the recvd skb */ | ||
953 | pci_unmap_single(mgp->pdev, bus, unmap_len, PCI_DMA_FROMDEVICE); | ||
954 | |||
955 | /* mcp implicitly skips 1st bytes so that packet is properly | ||
956 | * aligned */ | ||
957 | skb_reserve(skb, MXGEFW_PAD); | ||
958 | |||
959 | /* set the length of the frame */ | ||
960 | skb_put(skb, len); | ||
961 | |||
962 | skb->protocol = eth_type_trans(skb, mgp->dev); | ||
963 | skb->dev = mgp->dev; | ||
964 | if (mgp->csum_flag) { | ||
965 | if ((skb->protocol == ntohs(ETH_P_IP)) || | ||
966 | (skb->protocol == ntohs(ETH_P_IPV6))) { | ||
967 | skb->csum = ntohs((u16) csum); | ||
968 | skb->ip_summed = CHECKSUM_HW; | ||
969 | } else | ||
970 | myri10ge_vlan_ip_csum(skb, ntohs((u16) csum)); | ||
971 | } | ||
972 | |||
973 | netif_receive_skb(skb); | ||
974 | mgp->dev->last_rx = jiffies; | ||
975 | return 1; | ||
976 | } | ||
977 | |||
978 | static inline void myri10ge_tx_done(struct myri10ge_priv *mgp, int mcp_index) | ||
979 | { | ||
980 | struct pci_dev *pdev = mgp->pdev; | ||
981 | struct myri10ge_tx_buf *tx = &mgp->tx; | ||
982 | struct sk_buff *skb; | ||
983 | int idx, len; | ||
984 | int limit = 0; | ||
985 | |||
986 | while (tx->pkt_done != mcp_index) { | ||
987 | idx = tx->done & tx->mask; | ||
988 | skb = tx->info[idx].skb; | ||
989 | |||
990 | /* Mark as free */ | ||
991 | tx->info[idx].skb = NULL; | ||
992 | if (tx->info[idx].last) { | ||
993 | tx->pkt_done++; | ||
994 | tx->info[idx].last = 0; | ||
995 | } | ||
996 | tx->done++; | ||
997 | len = pci_unmap_len(&tx->info[idx], len); | ||
998 | pci_unmap_len_set(&tx->info[idx], len, 0); | ||
999 | if (skb) { | ||
1000 | mgp->stats.tx_bytes += skb->len; | ||
1001 | mgp->stats.tx_packets++; | ||
1002 | dev_kfree_skb_irq(skb); | ||
1003 | if (len) | ||
1004 | pci_unmap_single(pdev, | ||
1005 | pci_unmap_addr(&tx->info[idx], | ||
1006 | bus), len, | ||
1007 | PCI_DMA_TODEVICE); | ||
1008 | } else { | ||
1009 | if (len) | ||
1010 | pci_unmap_page(pdev, | ||
1011 | pci_unmap_addr(&tx->info[idx], | ||
1012 | bus), len, | ||
1013 | PCI_DMA_TODEVICE); | ||
1014 | } | ||
1015 | |||
1016 | /* limit potential for livelock by only handling | ||
1017 | * 2 full tx rings per call */ | ||
1018 | if (unlikely(++limit > 2 * tx->mask)) | ||
1019 | break; | ||
1020 | } | ||
1021 | /* start the queue if we've stopped it */ | ||
1022 | if (netif_queue_stopped(mgp->dev) | ||
1023 | && tx->req - tx->done < (tx->mask >> 1)) { | ||
1024 | mgp->wake_queue++; | ||
1025 | netif_wake_queue(mgp->dev); | ||
1026 | } | ||
1027 | } | ||
1028 | |||
1029 | static inline void myri10ge_clean_rx_done(struct myri10ge_priv *mgp, int *limit) | ||
1030 | { | ||
1031 | struct myri10ge_rx_done *rx_done = &mgp->rx_done; | ||
1032 | unsigned long rx_bytes = 0; | ||
1033 | unsigned long rx_packets = 0; | ||
1034 | unsigned long rx_ok; | ||
1035 | |||
1036 | int idx = rx_done->idx; | ||
1037 | int cnt = rx_done->cnt; | ||
1038 | u16 length; | ||
1039 | u16 checksum; | ||
1040 | |||
1041 | while (rx_done->entry[idx].length != 0 && *limit != 0) { | ||
1042 | length = ntohs(rx_done->entry[idx].length); | ||
1043 | rx_done->entry[idx].length = 0; | ||
1044 | checksum = ntohs(rx_done->entry[idx].checksum); | ||
1045 | if (length <= mgp->small_bytes) | ||
1046 | rx_ok = myri10ge_rx_done(mgp, &mgp->rx_small, | ||
1047 | mgp->small_bytes, | ||
1048 | length, checksum); | ||
1049 | else | ||
1050 | rx_ok = myri10ge_rx_done(mgp, &mgp->rx_big, | ||
1051 | mgp->dev->mtu + ETH_HLEN, | ||
1052 | length, checksum); | ||
1053 | rx_packets += rx_ok; | ||
1054 | rx_bytes += rx_ok * (unsigned long)length; | ||
1055 | cnt++; | ||
1056 | idx = cnt & (myri10ge_max_intr_slots - 1); | ||
1057 | |||
1058 | /* limit potential for livelock by only handling a | ||
1059 | * limited number of frames. */ | ||
1060 | (*limit)--; | ||
1061 | } | ||
1062 | rx_done->idx = idx; | ||
1063 | rx_done->cnt = cnt; | ||
1064 | mgp->stats.rx_packets += rx_packets; | ||
1065 | mgp->stats.rx_bytes += rx_bytes; | ||
1066 | } | ||
1067 | |||
1068 | static inline void myri10ge_check_statblock(struct myri10ge_priv *mgp) | ||
1069 | { | ||
1070 | struct mcp_irq_data *stats = mgp->fw_stats; | ||
1071 | |||
1072 | if (unlikely(stats->stats_updated)) { | ||
1073 | if (mgp->link_state != stats->link_up) { | ||
1074 | mgp->link_state = stats->link_up; | ||
1075 | if (mgp->link_state) { | ||
1076 | printk(KERN_INFO "myri10ge: %s: link up\n", | ||
1077 | mgp->dev->name); | ||
1078 | netif_carrier_on(mgp->dev); | ||
1079 | } else { | ||
1080 | printk(KERN_INFO "myri10ge: %s: link down\n", | ||
1081 | mgp->dev->name); | ||
1082 | netif_carrier_off(mgp->dev); | ||
1083 | } | ||
1084 | } | ||
1085 | if (mgp->rdma_tags_available != | ||
1086 | ntohl(mgp->fw_stats->rdma_tags_available)) { | ||
1087 | mgp->rdma_tags_available = | ||
1088 | ntohl(mgp->fw_stats->rdma_tags_available); | ||
1089 | printk(KERN_WARNING "myri10ge: %s: RDMA timed out! " | ||
1090 | "%d tags left\n", mgp->dev->name, | ||
1091 | mgp->rdma_tags_available); | ||
1092 | } | ||
1093 | mgp->down_cnt += stats->link_down; | ||
1094 | if (stats->link_down) | ||
1095 | wake_up(&mgp->down_wq); | ||
1096 | } | ||
1097 | } | ||
1098 | |||
1099 | static int myri10ge_poll(struct net_device *netdev, int *budget) | ||
1100 | { | ||
1101 | struct myri10ge_priv *mgp = netdev_priv(netdev); | ||
1102 | struct myri10ge_rx_done *rx_done = &mgp->rx_done; | ||
1103 | int limit, orig_limit, work_done; | ||
1104 | |||
1105 | /* process as many rx events as NAPI will allow */ | ||
1106 | limit = min(*budget, netdev->quota); | ||
1107 | orig_limit = limit; | ||
1108 | myri10ge_clean_rx_done(mgp, &limit); | ||
1109 | work_done = orig_limit - limit; | ||
1110 | *budget -= work_done; | ||
1111 | netdev->quota -= work_done; | ||
1112 | |||
1113 | if (rx_done->entry[rx_done->idx].length == 0 || !netif_running(netdev)) { | ||
1114 | netif_rx_complete(netdev); | ||
1115 | __raw_writel(htonl(3), mgp->irq_claim); | ||
1116 | return 0; | ||
1117 | } | ||
1118 | return 1; | ||
1119 | } | ||
1120 | |||
1121 | static irqreturn_t myri10ge_intr(int irq, void *arg, struct pt_regs *regs) | ||
1122 | { | ||
1123 | struct myri10ge_priv *mgp = arg; | ||
1124 | struct mcp_irq_data *stats = mgp->fw_stats; | ||
1125 | struct myri10ge_tx_buf *tx = &mgp->tx; | ||
1126 | u32 send_done_count; | ||
1127 | int i; | ||
1128 | |||
1129 | /* make sure it is our IRQ, and that the DMA has finished */ | ||
1130 | if (unlikely(!stats->valid)) | ||
1131 | return (IRQ_NONE); | ||
1132 | |||
1133 | /* low bit indicates receives are present, so schedule | ||
1134 | * napi poll handler */ | ||
1135 | if (stats->valid & 1) | ||
1136 | netif_rx_schedule(mgp->dev); | ||
1137 | |||
1138 | if (!mgp->msi_enabled) { | ||
1139 | __raw_writel(0, mgp->irq_deassert); | ||
1140 | if (!myri10ge_deassert_wait) | ||
1141 | stats->valid = 0; | ||
1142 | mb(); | ||
1143 | } else | ||
1144 | stats->valid = 0; | ||
1145 | |||
1146 | /* Wait for IRQ line to go low, if using INTx */ | ||
1147 | i = 0; | ||
1148 | while (1) { | ||
1149 | i++; | ||
1150 | /* check for transmit completes and receives */ | ||
1151 | send_done_count = ntohl(stats->send_done_count); | ||
1152 | if (send_done_count != tx->pkt_done) | ||
1153 | myri10ge_tx_done(mgp, (int)send_done_count); | ||
1154 | if (unlikely(i > myri10ge_max_irq_loops)) { | ||
1155 | printk(KERN_WARNING "myri10ge: %s: irq stuck?\n", | ||
1156 | mgp->dev->name); | ||
1157 | stats->valid = 0; | ||
1158 | schedule_work(&mgp->watchdog_work); | ||
1159 | } | ||
1160 | if (likely(stats->valid == 0)) | ||
1161 | break; | ||
1162 | cpu_relax(); | ||
1163 | barrier(); | ||
1164 | } | ||
1165 | |||
1166 | myri10ge_check_statblock(mgp); | ||
1167 | |||
1168 | __raw_writel(htonl(3), mgp->irq_claim + 1); | ||
1169 | return (IRQ_HANDLED); | ||
1170 | } | ||
1171 | |||
1172 | static int | ||
1173 | myri10ge_get_settings(struct net_device *netdev, struct ethtool_cmd *cmd) | ||
1174 | { | ||
1175 | cmd->autoneg = AUTONEG_DISABLE; | ||
1176 | cmd->speed = SPEED_10000; | ||
1177 | cmd->duplex = DUPLEX_FULL; | ||
1178 | return 0; | ||
1179 | } | ||
1180 | |||
1181 | static void | ||
1182 | myri10ge_get_drvinfo(struct net_device *netdev, struct ethtool_drvinfo *info) | ||
1183 | { | ||
1184 | struct myri10ge_priv *mgp = netdev_priv(netdev); | ||
1185 | |||
1186 | strlcpy(info->driver, "myri10ge", sizeof(info->driver)); | ||
1187 | strlcpy(info->version, MYRI10GE_VERSION_STR, sizeof(info->version)); | ||
1188 | strlcpy(info->fw_version, mgp->fw_version, sizeof(info->fw_version)); | ||
1189 | strlcpy(info->bus_info, pci_name(mgp->pdev), sizeof(info->bus_info)); | ||
1190 | } | ||
1191 | |||
1192 | static int | ||
1193 | myri10ge_get_coalesce(struct net_device *netdev, struct ethtool_coalesce *coal) | ||
1194 | { | ||
1195 | struct myri10ge_priv *mgp = netdev_priv(netdev); | ||
1196 | coal->rx_coalesce_usecs = mgp->intr_coal_delay; | ||
1197 | return 0; | ||
1198 | } | ||
1199 | |||
1200 | static int | ||
1201 | myri10ge_set_coalesce(struct net_device *netdev, struct ethtool_coalesce *coal) | ||
1202 | { | ||
1203 | struct myri10ge_priv *mgp = netdev_priv(netdev); | ||
1204 | |||
1205 | mgp->intr_coal_delay = coal->rx_coalesce_usecs; | ||
1206 | __raw_writel(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr); | ||
1207 | return 0; | ||
1208 | } | ||
1209 | |||
1210 | static void | ||
1211 | myri10ge_get_pauseparam(struct net_device *netdev, | ||
1212 | struct ethtool_pauseparam *pause) | ||
1213 | { | ||
1214 | struct myri10ge_priv *mgp = netdev_priv(netdev); | ||
1215 | |||
1216 | pause->autoneg = 0; | ||
1217 | pause->rx_pause = mgp->pause; | ||
1218 | pause->tx_pause = mgp->pause; | ||
1219 | } | ||
1220 | |||
1221 | static int | ||
1222 | myri10ge_set_pauseparam(struct net_device *netdev, | ||
1223 | struct ethtool_pauseparam *pause) | ||
1224 | { | ||
1225 | struct myri10ge_priv *mgp = netdev_priv(netdev); | ||
1226 | |||
1227 | if (pause->tx_pause != mgp->pause) | ||
1228 | return myri10ge_change_pause(mgp, pause->tx_pause); | ||
1229 | if (pause->rx_pause != mgp->pause) | ||
1230 | return myri10ge_change_pause(mgp, pause->tx_pause); | ||
1231 | if (pause->autoneg != 0) | ||
1232 | return -EINVAL; | ||
1233 | return 0; | ||
1234 | } | ||
1235 | |||
1236 | static void | ||
1237 | myri10ge_get_ringparam(struct net_device *netdev, | ||
1238 | struct ethtool_ringparam *ring) | ||
1239 | { | ||
1240 | struct myri10ge_priv *mgp = netdev_priv(netdev); | ||
1241 | |||
1242 | ring->rx_mini_max_pending = mgp->rx_small.mask + 1; | ||
1243 | ring->rx_max_pending = mgp->rx_big.mask + 1; | ||
1244 | ring->rx_jumbo_max_pending = 0; | ||
1245 | ring->tx_max_pending = mgp->rx_small.mask + 1; | ||
1246 | ring->rx_mini_pending = ring->rx_mini_max_pending; | ||
1247 | ring->rx_pending = ring->rx_max_pending; | ||
1248 | ring->rx_jumbo_pending = ring->rx_jumbo_max_pending; | ||
1249 | ring->tx_pending = ring->tx_max_pending; | ||
1250 | } | ||
1251 | |||
1252 | static u32 myri10ge_get_rx_csum(struct net_device *netdev) | ||
1253 | { | ||
1254 | struct myri10ge_priv *mgp = netdev_priv(netdev); | ||
1255 | if (mgp->csum_flag) | ||
1256 | return 1; | ||
1257 | else | ||
1258 | return 0; | ||
1259 | } | ||
1260 | |||
1261 | static int myri10ge_set_rx_csum(struct net_device *netdev, u32 csum_enabled) | ||
1262 | { | ||
1263 | struct myri10ge_priv *mgp = netdev_priv(netdev); | ||
1264 | if (csum_enabled) | ||
1265 | mgp->csum_flag = MXGEFW_FLAGS_CKSUM; | ||
1266 | else | ||
1267 | mgp->csum_flag = 0; | ||
1268 | return 0; | ||
1269 | } | ||
1270 | |||
1271 | static const char myri10ge_gstrings_stats[][ETH_GSTRING_LEN] = { | ||
1272 | "rx_packets", "tx_packets", "rx_bytes", "tx_bytes", "rx_errors", | ||
1273 | "tx_errors", "rx_dropped", "tx_dropped", "multicast", "collisions", | ||
1274 | "rx_length_errors", "rx_over_errors", "rx_crc_errors", | ||
1275 | "rx_frame_errors", "rx_fifo_errors", "rx_missed_errors", | ||
1276 | "tx_aborted_errors", "tx_carrier_errors", "tx_fifo_errors", | ||
1277 | "tx_heartbeat_errors", "tx_window_errors", | ||
1278 | /* device-specific stats */ | ||
1279 | "read_dma_bw_MBs", "write_dma_bw_MBs", "read_write_dma_bw_MBs", | ||
1280 | "serial_number", "tx_pkt_start", "tx_pkt_done", | ||
1281 | "tx_req", "tx_done", "rx_small_cnt", "rx_big_cnt", | ||
1282 | "wake_queue", "stop_queue", "watchdog_resets", "tx_linearized", | ||
1283 | "link_up", "dropped_link_overflow", "dropped_link_error_or_filtered", | ||
1284 | "dropped_runt", "dropped_overrun", "dropped_no_small_buffer", | ||
1285 | "dropped_no_big_buffer" | ||
1286 | }; | ||
1287 | |||
1288 | #define MYRI10GE_NET_STATS_LEN 21 | ||
1289 | #define MYRI10GE_STATS_LEN sizeof(myri10ge_gstrings_stats) / ETH_GSTRING_LEN | ||
1290 | |||
1291 | static void | ||
1292 | myri10ge_get_strings(struct net_device *netdev, u32 stringset, u8 * data) | ||
1293 | { | ||
1294 | switch (stringset) { | ||
1295 | case ETH_SS_STATS: | ||
1296 | memcpy(data, *myri10ge_gstrings_stats, | ||
1297 | sizeof(myri10ge_gstrings_stats)); | ||
1298 | break; | ||
1299 | } | ||
1300 | } | ||
1301 | |||
1302 | static int myri10ge_get_stats_count(struct net_device *netdev) | ||
1303 | { | ||
1304 | return MYRI10GE_STATS_LEN; | ||
1305 | } | ||
1306 | |||
1307 | static void | ||
1308 | myri10ge_get_ethtool_stats(struct net_device *netdev, | ||
1309 | struct ethtool_stats *stats, u64 * data) | ||
1310 | { | ||
1311 | struct myri10ge_priv *mgp = netdev_priv(netdev); | ||
1312 | int i; | ||
1313 | |||
1314 | for (i = 0; i < MYRI10GE_NET_STATS_LEN; i++) | ||
1315 | data[i] = ((unsigned long *)&mgp->stats)[i]; | ||
1316 | |||
1317 | data[i++] = (unsigned int)mgp->read_dma; | ||
1318 | data[i++] = (unsigned int)mgp->write_dma; | ||
1319 | data[i++] = (unsigned int)mgp->read_write_dma; | ||
1320 | data[i++] = (unsigned int)mgp->serial_number; | ||
1321 | data[i++] = (unsigned int)mgp->tx.pkt_start; | ||
1322 | data[i++] = (unsigned int)mgp->tx.pkt_done; | ||
1323 | data[i++] = (unsigned int)mgp->tx.req; | ||
1324 | data[i++] = (unsigned int)mgp->tx.done; | ||
1325 | data[i++] = (unsigned int)mgp->rx_small.cnt; | ||
1326 | data[i++] = (unsigned int)mgp->rx_big.cnt; | ||
1327 | data[i++] = (unsigned int)mgp->wake_queue; | ||
1328 | data[i++] = (unsigned int)mgp->stop_queue; | ||
1329 | data[i++] = (unsigned int)mgp->watchdog_resets; | ||
1330 | data[i++] = (unsigned int)mgp->tx_linearized; | ||
1331 | data[i++] = (unsigned int)ntohl(mgp->fw_stats->link_up); | ||
1332 | data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_link_overflow); | ||
1333 | data[i++] = | ||
1334 | (unsigned int)ntohl(mgp->fw_stats->dropped_link_error_or_filtered); | ||
1335 | data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_runt); | ||
1336 | data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_overrun); | ||
1337 | data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_no_small_buffer); | ||
1338 | data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_no_big_buffer); | ||
1339 | } | ||
1340 | |||
1341 | static struct ethtool_ops myri10ge_ethtool_ops = { | ||
1342 | .get_settings = myri10ge_get_settings, | ||
1343 | .get_drvinfo = myri10ge_get_drvinfo, | ||
1344 | .get_coalesce = myri10ge_get_coalesce, | ||
1345 | .set_coalesce = myri10ge_set_coalesce, | ||
1346 | .get_pauseparam = myri10ge_get_pauseparam, | ||
1347 | .set_pauseparam = myri10ge_set_pauseparam, | ||
1348 | .get_ringparam = myri10ge_get_ringparam, | ||
1349 | .get_rx_csum = myri10ge_get_rx_csum, | ||
1350 | .set_rx_csum = myri10ge_set_rx_csum, | ||
1351 | .get_tx_csum = ethtool_op_get_tx_csum, | ||
1352 | .set_tx_csum = ethtool_op_set_tx_csum, | ||
1353 | .get_sg = ethtool_op_get_sg, | ||
1354 | .set_sg = ethtool_op_set_sg, | ||
1355 | #ifdef NETIF_F_TSO | ||
1356 | .get_tso = ethtool_op_get_tso, | ||
1357 | .set_tso = ethtool_op_set_tso, | ||
1358 | #endif | ||
1359 | .get_strings = myri10ge_get_strings, | ||
1360 | .get_stats_count = myri10ge_get_stats_count, | ||
1361 | .get_ethtool_stats = myri10ge_get_ethtool_stats | ||
1362 | }; | ||
1363 | |||
1364 | static int myri10ge_allocate_rings(struct net_device *dev) | ||
1365 | { | ||
1366 | struct myri10ge_priv *mgp; | ||
1367 | struct myri10ge_cmd cmd; | ||
1368 | int tx_ring_size, rx_ring_size; | ||
1369 | int tx_ring_entries, rx_ring_entries; | ||
1370 | int i, status; | ||
1371 | size_t bytes; | ||
1372 | |||
1373 | mgp = netdev_priv(dev); | ||
1374 | |||
1375 | /* get ring sizes */ | ||
1376 | |||
1377 | status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_RING_SIZE, &cmd, 0); | ||
1378 | tx_ring_size = cmd.data0; | ||
1379 | status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_RX_RING_SIZE, &cmd, 0); | ||
1380 | rx_ring_size = cmd.data0; | ||
1381 | |||
1382 | tx_ring_entries = tx_ring_size / sizeof(struct mcp_kreq_ether_send); | ||
1383 | rx_ring_entries = rx_ring_size / sizeof(struct mcp_dma_addr); | ||
1384 | mgp->tx.mask = tx_ring_entries - 1; | ||
1385 | mgp->rx_small.mask = mgp->rx_big.mask = rx_ring_entries - 1; | ||
1386 | |||
1387 | /* allocate the host shadow rings */ | ||
1388 | |||
1389 | bytes = 8 + (MYRI10GE_MAX_SEND_DESC_TSO + 4) | ||
1390 | * sizeof(*mgp->tx.req_list); | ||
1391 | mgp->tx.req_bytes = kzalloc(bytes, GFP_KERNEL); | ||
1392 | if (mgp->tx.req_bytes == NULL) | ||
1393 | goto abort_with_nothing; | ||
1394 | |||
1395 | /* ensure req_list entries are aligned to 8 bytes */ | ||
1396 | mgp->tx.req_list = (struct mcp_kreq_ether_send *) | ||
1397 | ALIGN((unsigned long)mgp->tx.req_bytes, 8); | ||
1398 | |||
1399 | bytes = rx_ring_entries * sizeof(*mgp->rx_small.shadow); | ||
1400 | mgp->rx_small.shadow = kzalloc(bytes, GFP_KERNEL); | ||
1401 | if (mgp->rx_small.shadow == NULL) | ||
1402 | goto abort_with_tx_req_bytes; | ||
1403 | |||
1404 | bytes = rx_ring_entries * sizeof(*mgp->rx_big.shadow); | ||
1405 | mgp->rx_big.shadow = kzalloc(bytes, GFP_KERNEL); | ||
1406 | if (mgp->rx_big.shadow == NULL) | ||
1407 | goto abort_with_rx_small_shadow; | ||
1408 | |||
1409 | /* allocate the host info rings */ | ||
1410 | |||
1411 | bytes = tx_ring_entries * sizeof(*mgp->tx.info); | ||
1412 | mgp->tx.info = kzalloc(bytes, GFP_KERNEL); | ||
1413 | if (mgp->tx.info == NULL) | ||
1414 | goto abort_with_rx_big_shadow; | ||
1415 | |||
1416 | bytes = rx_ring_entries * sizeof(*mgp->rx_small.info); | ||
1417 | mgp->rx_small.info = kzalloc(bytes, GFP_KERNEL); | ||
1418 | if (mgp->rx_small.info == NULL) | ||
1419 | goto abort_with_tx_info; | ||
1420 | |||
1421 | bytes = rx_ring_entries * sizeof(*mgp->rx_big.info); | ||
1422 | mgp->rx_big.info = kzalloc(bytes, GFP_KERNEL); | ||
1423 | if (mgp->rx_big.info == NULL) | ||
1424 | goto abort_with_rx_small_info; | ||
1425 | |||
1426 | /* Fill the receive rings */ | ||
1427 | |||
1428 | for (i = 0; i <= mgp->rx_small.mask; i++) { | ||
1429 | status = myri10ge_getbuf(&mgp->rx_small, mgp->pdev, | ||
1430 | mgp->small_bytes, i); | ||
1431 | if (status) { | ||
1432 | printk(KERN_ERR | ||
1433 | "myri10ge: %s: alloced only %d small bufs\n", | ||
1434 | dev->name, i); | ||
1435 | goto abort_with_rx_small_ring; | ||
1436 | } | ||
1437 | } | ||
1438 | |||
1439 | for (i = 0; i <= mgp->rx_big.mask; i++) { | ||
1440 | status = | ||
1441 | myri10ge_getbuf(&mgp->rx_big, mgp->pdev, | ||
1442 | dev->mtu + ETH_HLEN, i); | ||
1443 | if (status) { | ||
1444 | printk(KERN_ERR | ||
1445 | "myri10ge: %s: alloced only %d big bufs\n", | ||
1446 | dev->name, i); | ||
1447 | goto abort_with_rx_big_ring; | ||
1448 | } | ||
1449 | } | ||
1450 | |||
1451 | return 0; | ||
1452 | |||
1453 | abort_with_rx_big_ring: | ||
1454 | for (i = 0; i <= mgp->rx_big.mask; i++) { | ||
1455 | if (mgp->rx_big.info[i].skb != NULL) | ||
1456 | dev_kfree_skb_any(mgp->rx_big.info[i].skb); | ||
1457 | if (pci_unmap_len(&mgp->rx_big.info[i], len)) | ||
1458 | pci_unmap_single(mgp->pdev, | ||
1459 | pci_unmap_addr(&mgp->rx_big.info[i], | ||
1460 | bus), | ||
1461 | pci_unmap_len(&mgp->rx_big.info[i], | ||
1462 | len), | ||
1463 | PCI_DMA_FROMDEVICE); | ||
1464 | } | ||
1465 | |||
1466 | abort_with_rx_small_ring: | ||
1467 | for (i = 0; i <= mgp->rx_small.mask; i++) { | ||
1468 | if (mgp->rx_small.info[i].skb != NULL) | ||
1469 | dev_kfree_skb_any(mgp->rx_small.info[i].skb); | ||
1470 | if (pci_unmap_len(&mgp->rx_small.info[i], len)) | ||
1471 | pci_unmap_single(mgp->pdev, | ||
1472 | pci_unmap_addr(&mgp->rx_small.info[i], | ||
1473 | bus), | ||
1474 | pci_unmap_len(&mgp->rx_small.info[i], | ||
1475 | len), | ||
1476 | PCI_DMA_FROMDEVICE); | ||
1477 | } | ||
1478 | kfree(mgp->rx_big.info); | ||
1479 | |||
1480 | abort_with_rx_small_info: | ||
1481 | kfree(mgp->rx_small.info); | ||
1482 | |||
1483 | abort_with_tx_info: | ||
1484 | kfree(mgp->tx.info); | ||
1485 | |||
1486 | abort_with_rx_big_shadow: | ||
1487 | kfree(mgp->rx_big.shadow); | ||
1488 | |||
1489 | abort_with_rx_small_shadow: | ||
1490 | kfree(mgp->rx_small.shadow); | ||
1491 | |||
1492 | abort_with_tx_req_bytes: | ||
1493 | kfree(mgp->tx.req_bytes); | ||
1494 | mgp->tx.req_bytes = NULL; | ||
1495 | mgp->tx.req_list = NULL; | ||
1496 | |||
1497 | abort_with_nothing: | ||
1498 | return status; | ||
1499 | } | ||
1500 | |||
1501 | static void myri10ge_free_rings(struct net_device *dev) | ||
1502 | { | ||
1503 | struct myri10ge_priv *mgp; | ||
1504 | struct sk_buff *skb; | ||
1505 | struct myri10ge_tx_buf *tx; | ||
1506 | int i, len, idx; | ||
1507 | |||
1508 | mgp = netdev_priv(dev); | ||
1509 | |||
1510 | for (i = 0; i <= mgp->rx_big.mask; i++) { | ||
1511 | if (mgp->rx_big.info[i].skb != NULL) | ||
1512 | dev_kfree_skb_any(mgp->rx_big.info[i].skb); | ||
1513 | if (pci_unmap_len(&mgp->rx_big.info[i], len)) | ||
1514 | pci_unmap_single(mgp->pdev, | ||
1515 | pci_unmap_addr(&mgp->rx_big.info[i], | ||
1516 | bus), | ||
1517 | pci_unmap_len(&mgp->rx_big.info[i], | ||
1518 | len), | ||
1519 | PCI_DMA_FROMDEVICE); | ||
1520 | } | ||
1521 | |||
1522 | for (i = 0; i <= mgp->rx_small.mask; i++) { | ||
1523 | if (mgp->rx_small.info[i].skb != NULL) | ||
1524 | dev_kfree_skb_any(mgp->rx_small.info[i].skb); | ||
1525 | if (pci_unmap_len(&mgp->rx_small.info[i], len)) | ||
1526 | pci_unmap_single(mgp->pdev, | ||
1527 | pci_unmap_addr(&mgp->rx_small.info[i], | ||
1528 | bus), | ||
1529 | pci_unmap_len(&mgp->rx_small.info[i], | ||
1530 | len), | ||
1531 | PCI_DMA_FROMDEVICE); | ||
1532 | } | ||
1533 | |||
1534 | tx = &mgp->tx; | ||
1535 | while (tx->done != tx->req) { | ||
1536 | idx = tx->done & tx->mask; | ||
1537 | skb = tx->info[idx].skb; | ||
1538 | |||
1539 | /* Mark as free */ | ||
1540 | tx->info[idx].skb = NULL; | ||
1541 | tx->done++; | ||
1542 | len = pci_unmap_len(&tx->info[idx], len); | ||
1543 | pci_unmap_len_set(&tx->info[idx], len, 0); | ||
1544 | if (skb) { | ||
1545 | mgp->stats.tx_dropped++; | ||
1546 | dev_kfree_skb_any(skb); | ||
1547 | if (len) | ||
1548 | pci_unmap_single(mgp->pdev, | ||
1549 | pci_unmap_addr(&tx->info[idx], | ||
1550 | bus), len, | ||
1551 | PCI_DMA_TODEVICE); | ||
1552 | } else { | ||
1553 | if (len) | ||
1554 | pci_unmap_page(mgp->pdev, | ||
1555 | pci_unmap_addr(&tx->info[idx], | ||
1556 | bus), len, | ||
1557 | PCI_DMA_TODEVICE); | ||
1558 | } | ||
1559 | } | ||
1560 | kfree(mgp->rx_big.info); | ||
1561 | |||
1562 | kfree(mgp->rx_small.info); | ||
1563 | |||
1564 | kfree(mgp->tx.info); | ||
1565 | |||
1566 | kfree(mgp->rx_big.shadow); | ||
1567 | |||
1568 | kfree(mgp->rx_small.shadow); | ||
1569 | |||
1570 | kfree(mgp->tx.req_bytes); | ||
1571 | mgp->tx.req_bytes = NULL; | ||
1572 | mgp->tx.req_list = NULL; | ||
1573 | } | ||
1574 | |||
1575 | static int myri10ge_open(struct net_device *dev) | ||
1576 | { | ||
1577 | struct myri10ge_priv *mgp; | ||
1578 | struct myri10ge_cmd cmd; | ||
1579 | int status, big_pow2; | ||
1580 | |||
1581 | mgp = netdev_priv(dev); | ||
1582 | |||
1583 | if (mgp->running != MYRI10GE_ETH_STOPPED) | ||
1584 | return -EBUSY; | ||
1585 | |||
1586 | mgp->running = MYRI10GE_ETH_STARTING; | ||
1587 | status = myri10ge_reset(mgp); | ||
1588 | if (status != 0) { | ||
1589 | printk(KERN_ERR "myri10ge: %s: failed reset\n", dev->name); | ||
1590 | mgp->running = MYRI10GE_ETH_STOPPED; | ||
1591 | return -ENXIO; | ||
1592 | } | ||
1593 | |||
1594 | /* decide what small buffer size to use. For good TCP rx | ||
1595 | * performance, it is important to not receive 1514 byte | ||
1596 | * frames into jumbo buffers, as it confuses the socket buffer | ||
1597 | * accounting code, leading to drops and erratic performance. | ||
1598 | */ | ||
1599 | |||
1600 | if (dev->mtu <= ETH_DATA_LEN) | ||
1601 | mgp->small_bytes = 128; /* enough for a TCP header */ | ||
1602 | else | ||
1603 | mgp->small_bytes = ETH_FRAME_LEN; /* enough for an ETH_DATA_LEN frame */ | ||
1604 | |||
1605 | /* Override the small buffer size? */ | ||
1606 | if (myri10ge_small_bytes > 0) | ||
1607 | mgp->small_bytes = myri10ge_small_bytes; | ||
1608 | |||
1609 | /* If the user sets an obscenely small MTU, adjust the small | ||
1610 | * bytes down to nearly nothing */ | ||
1611 | if (mgp->small_bytes >= (dev->mtu + ETH_HLEN)) | ||
1612 | mgp->small_bytes = 64; | ||
1613 | |||
1614 | /* get the lanai pointers to the send and receive rings */ | ||
1615 | |||
1616 | status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_OFFSET, &cmd, 0); | ||
1617 | mgp->tx.lanai = | ||
1618 | (struct mcp_kreq_ether_send __iomem *)(mgp->sram + cmd.data0); | ||
1619 | |||
1620 | status |= | ||
1621 | myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SMALL_RX_OFFSET, &cmd, 0); | ||
1622 | mgp->rx_small.lanai = | ||
1623 | (struct mcp_kreq_ether_recv __iomem *)(mgp->sram + cmd.data0); | ||
1624 | |||
1625 | status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_BIG_RX_OFFSET, &cmd, 0); | ||
1626 | mgp->rx_big.lanai = | ||
1627 | (struct mcp_kreq_ether_recv __iomem *)(mgp->sram + cmd.data0); | ||
1628 | |||
1629 | if (status != 0) { | ||
1630 | printk(KERN_ERR | ||
1631 | "myri10ge: %s: failed to get ring sizes or locations\n", | ||
1632 | dev->name); | ||
1633 | mgp->running = MYRI10GE_ETH_STOPPED; | ||
1634 | return -ENXIO; | ||
1635 | } | ||
1636 | |||
1637 | if (mgp->mtrr >= 0) { | ||
1638 | mgp->tx.wc_fifo = (u8 __iomem *) mgp->sram + 0x200000; | ||
1639 | mgp->rx_small.wc_fifo = (u8 __iomem *) mgp->sram + 0x300000; | ||
1640 | mgp->rx_big.wc_fifo = (u8 __iomem *) mgp->sram + 0x340000; | ||
1641 | } else { | ||
1642 | mgp->tx.wc_fifo = NULL; | ||
1643 | mgp->rx_small.wc_fifo = NULL; | ||
1644 | mgp->rx_big.wc_fifo = NULL; | ||
1645 | } | ||
1646 | |||
1647 | status = myri10ge_allocate_rings(dev); | ||
1648 | if (status != 0) | ||
1649 | goto abort_with_nothing; | ||
1650 | |||
1651 | /* Firmware needs the big buff size as a power of 2. Lie and | ||
1652 | * tell him the buffer is larger, because we only use 1 | ||
1653 | * buffer/pkt, and the mtu will prevent overruns. | ||
1654 | */ | ||
1655 | big_pow2 = dev->mtu + ETH_HLEN + MXGEFW_PAD; | ||
1656 | while ((big_pow2 & (big_pow2 - 1)) != 0) | ||
1657 | big_pow2++; | ||
1658 | |||
1659 | /* now give firmware buffers sizes, and MTU */ | ||
1660 | cmd.data0 = dev->mtu + ETH_HLEN + VLAN_HLEN; | ||
1661 | status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_MTU, &cmd, 0); | ||
1662 | cmd.data0 = mgp->small_bytes; | ||
1663 | status |= | ||
1664 | myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_SMALL_BUFFER_SIZE, &cmd, 0); | ||
1665 | cmd.data0 = big_pow2; | ||
1666 | status |= | ||
1667 | myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_BIG_BUFFER_SIZE, &cmd, 0); | ||
1668 | if (status) { | ||
1669 | printk(KERN_ERR "myri10ge: %s: Couldn't set buffer sizes\n", | ||
1670 | dev->name); | ||
1671 | goto abort_with_rings; | ||
1672 | } | ||
1673 | |||
1674 | cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->fw_stats_bus); | ||
1675 | cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->fw_stats_bus); | ||
1676 | status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_STATS_DMA, &cmd, 0); | ||
1677 | if (status) { | ||
1678 | printk(KERN_ERR "myri10ge: %s: Couldn't set stats DMA\n", | ||
1679 | dev->name); | ||
1680 | goto abort_with_rings; | ||
1681 | } | ||
1682 | |||
1683 | mgp->link_state = -1; | ||
1684 | mgp->rdma_tags_available = 15; | ||
1685 | |||
1686 | netif_poll_enable(mgp->dev); /* must happen prior to any irq */ | ||
1687 | |||
1688 | status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_UP, &cmd, 0); | ||
1689 | if (status) { | ||
1690 | printk(KERN_ERR "myri10ge: %s: Couldn't bring up link\n", | ||
1691 | dev->name); | ||
1692 | goto abort_with_rings; | ||
1693 | } | ||
1694 | |||
1695 | mgp->wake_queue = 0; | ||
1696 | mgp->stop_queue = 0; | ||
1697 | mgp->running = MYRI10GE_ETH_RUNNING; | ||
1698 | mgp->watchdog_timer.expires = jiffies + myri10ge_watchdog_timeout * HZ; | ||
1699 | add_timer(&mgp->watchdog_timer); | ||
1700 | netif_wake_queue(dev); | ||
1701 | return 0; | ||
1702 | |||
1703 | abort_with_rings: | ||
1704 | myri10ge_free_rings(dev); | ||
1705 | |||
1706 | abort_with_nothing: | ||
1707 | mgp->running = MYRI10GE_ETH_STOPPED; | ||
1708 | return -ENOMEM; | ||
1709 | } | ||
1710 | |||
1711 | static int myri10ge_close(struct net_device *dev) | ||
1712 | { | ||
1713 | struct myri10ge_priv *mgp; | ||
1714 | struct myri10ge_cmd cmd; | ||
1715 | int status, old_down_cnt; | ||
1716 | |||
1717 | mgp = netdev_priv(dev); | ||
1718 | |||
1719 | if (mgp->running != MYRI10GE_ETH_RUNNING) | ||
1720 | return 0; | ||
1721 | |||
1722 | if (mgp->tx.req_bytes == NULL) | ||
1723 | return 0; | ||
1724 | |||
1725 | del_timer_sync(&mgp->watchdog_timer); | ||
1726 | mgp->running = MYRI10GE_ETH_STOPPING; | ||
1727 | netif_poll_disable(mgp->dev); | ||
1728 | netif_carrier_off(dev); | ||
1729 | netif_stop_queue(dev); | ||
1730 | old_down_cnt = mgp->down_cnt; | ||
1731 | mb(); | ||
1732 | status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_DOWN, &cmd, 0); | ||
1733 | if (status) | ||
1734 | printk(KERN_ERR "myri10ge: %s: Couldn't bring down link\n", | ||
1735 | dev->name); | ||
1736 | |||
1737 | wait_event_timeout(mgp->down_wq, old_down_cnt != mgp->down_cnt, HZ); | ||
1738 | if (old_down_cnt == mgp->down_cnt) | ||
1739 | printk(KERN_ERR "myri10ge: %s never got down irq\n", dev->name); | ||
1740 | |||
1741 | netif_tx_disable(dev); | ||
1742 | |||
1743 | myri10ge_free_rings(dev); | ||
1744 | |||
1745 | mgp->running = MYRI10GE_ETH_STOPPED; | ||
1746 | return 0; | ||
1747 | } | ||
1748 | |||
1749 | /* copy an array of struct mcp_kreq_ether_send's to the mcp. Copy | ||
1750 | * backwards one at a time and handle ring wraps */ | ||
1751 | |||
1752 | static inline void | ||
1753 | myri10ge_submit_req_backwards(struct myri10ge_tx_buf *tx, | ||
1754 | struct mcp_kreq_ether_send *src, int cnt) | ||
1755 | { | ||
1756 | int idx, starting_slot; | ||
1757 | starting_slot = tx->req; | ||
1758 | while (cnt > 1) { | ||
1759 | cnt--; | ||
1760 | idx = (starting_slot + cnt) & tx->mask; | ||
1761 | myri10ge_pio_copy(&tx->lanai[idx], &src[cnt], sizeof(*src)); | ||
1762 | mb(); | ||
1763 | } | ||
1764 | } | ||
1765 | |||
1766 | /* | ||
1767 | * copy an array of struct mcp_kreq_ether_send's to the mcp. Copy | ||
1768 | * at most 32 bytes at a time, so as to avoid involving the software | ||
1769 | * pio handler in the nic. We re-write the first segment's flags | ||
1770 | * to mark them valid only after writing the entire chain. | ||
1771 | */ | ||
1772 | |||
1773 | static inline void | ||
1774 | myri10ge_submit_req(struct myri10ge_tx_buf *tx, struct mcp_kreq_ether_send *src, | ||
1775 | int cnt) | ||
1776 | { | ||
1777 | int idx, i; | ||
1778 | struct mcp_kreq_ether_send __iomem *dstp, *dst; | ||
1779 | struct mcp_kreq_ether_send *srcp; | ||
1780 | u8 last_flags; | ||
1781 | |||
1782 | idx = tx->req & tx->mask; | ||
1783 | |||
1784 | last_flags = src->flags; | ||
1785 | src->flags = 0; | ||
1786 | mb(); | ||
1787 | dst = dstp = &tx->lanai[idx]; | ||
1788 | srcp = src; | ||
1789 | |||
1790 | if ((idx + cnt) < tx->mask) { | ||
1791 | for (i = 0; i < (cnt - 1); i += 2) { | ||
1792 | myri10ge_pio_copy(dstp, srcp, 2 * sizeof(*src)); | ||
1793 | mb(); /* force write every 32 bytes */ | ||
1794 | srcp += 2; | ||
1795 | dstp += 2; | ||
1796 | } | ||
1797 | } else { | ||
1798 | /* submit all but the first request, and ensure | ||
1799 | * that it is submitted below */ | ||
1800 | myri10ge_submit_req_backwards(tx, src, cnt); | ||
1801 | i = 0; | ||
1802 | } | ||
1803 | if (i < cnt) { | ||
1804 | /* submit the first request */ | ||
1805 | myri10ge_pio_copy(dstp, srcp, sizeof(*src)); | ||
1806 | mb(); /* barrier before setting valid flag */ | ||
1807 | } | ||
1808 | |||
1809 | /* re-write the last 32-bits with the valid flags */ | ||
1810 | src->flags = last_flags; | ||
1811 | __raw_writel(*((u32 *) src + 3), (u32 __iomem *) dst + 3); | ||
1812 | tx->req += cnt; | ||
1813 | mb(); | ||
1814 | } | ||
1815 | |||
1816 | static inline void | ||
1817 | myri10ge_submit_req_wc(struct myri10ge_tx_buf *tx, | ||
1818 | struct mcp_kreq_ether_send *src, int cnt) | ||
1819 | { | ||
1820 | tx->req += cnt; | ||
1821 | mb(); | ||
1822 | while (cnt >= 4) { | ||
1823 | myri10ge_pio_copy(tx->wc_fifo, src, 64); | ||
1824 | mb(); | ||
1825 | src += 4; | ||
1826 | cnt -= 4; | ||
1827 | } | ||
1828 | if (cnt > 0) { | ||
1829 | /* pad it to 64 bytes. The src is 64 bytes bigger than it | ||
1830 | * needs to be so that we don't overrun it */ | ||
1831 | myri10ge_pio_copy(tx->wc_fifo + (cnt << 18), src, 64); | ||
1832 | mb(); | ||
1833 | } | ||
1834 | } | ||
1835 | |||
1836 | /* | ||
1837 | * Transmit a packet. We need to split the packet so that a single | ||
1838 | * segment does not cross myri10ge->tx.boundary, so this makes segment | ||
1839 | * counting tricky. So rather than try to count segments up front, we | ||
1840 | * just give up if there are too few segments to hold a reasonably | ||
1841 | * fragmented packet currently available. If we run | ||
1842 | * out of segments while preparing a packet for DMA, we just linearize | ||
1843 | * it and try again. | ||
1844 | */ | ||
1845 | |||
1846 | static int myri10ge_xmit(struct sk_buff *skb, struct net_device *dev) | ||
1847 | { | ||
1848 | struct myri10ge_priv *mgp = netdev_priv(dev); | ||
1849 | struct mcp_kreq_ether_send *req; | ||
1850 | struct myri10ge_tx_buf *tx = &mgp->tx; | ||
1851 | struct skb_frag_struct *frag; | ||
1852 | dma_addr_t bus; | ||
1853 | u32 low, high_swapped; | ||
1854 | unsigned int len; | ||
1855 | int idx, last_idx, avail, frag_cnt, frag_idx, count, mss, max_segments; | ||
1856 | u16 pseudo_hdr_offset, cksum_offset; | ||
1857 | int cum_len, seglen, boundary, rdma_count; | ||
1858 | u8 flags, odd_flag; | ||
1859 | |||
1860 | again: | ||
1861 | req = tx->req_list; | ||
1862 | avail = tx->mask - 1 - (tx->req - tx->done); | ||
1863 | |||
1864 | mss = 0; | ||
1865 | max_segments = MXGEFW_MAX_SEND_DESC; | ||
1866 | |||
1867 | #ifdef NETIF_F_TSO | ||
1868 | if (skb->len > (dev->mtu + ETH_HLEN)) { | ||
1869 | mss = skb_shinfo(skb)->tso_size; | ||
1870 | if (mss != 0) | ||
1871 | max_segments = MYRI10GE_MAX_SEND_DESC_TSO; | ||
1872 | } | ||
1873 | #endif /*NETIF_F_TSO */ | ||
1874 | |||
1875 | if ((unlikely(avail < max_segments))) { | ||
1876 | /* we are out of transmit resources */ | ||
1877 | mgp->stop_queue++; | ||
1878 | netif_stop_queue(dev); | ||
1879 | return 1; | ||
1880 | } | ||
1881 | |||
1882 | /* Setup checksum offloading, if needed */ | ||
1883 | cksum_offset = 0; | ||
1884 | pseudo_hdr_offset = 0; | ||
1885 | odd_flag = 0; | ||
1886 | flags = (MXGEFW_FLAGS_NO_TSO | MXGEFW_FLAGS_FIRST); | ||
1887 | if (likely(skb->ip_summed == CHECKSUM_HW)) { | ||
1888 | cksum_offset = (skb->h.raw - skb->data); | ||
1889 | pseudo_hdr_offset = (skb->h.raw + skb->csum) - skb->data; | ||
1890 | /* If the headers are excessively large, then we must | ||
1891 | * fall back to a software checksum */ | ||
1892 | if (unlikely(cksum_offset > 255 || pseudo_hdr_offset > 127)) { | ||
1893 | if (skb_checksum_help(skb, 0)) | ||
1894 | goto drop; | ||
1895 | cksum_offset = 0; | ||
1896 | pseudo_hdr_offset = 0; | ||
1897 | } else { | ||
1898 | pseudo_hdr_offset = htons(pseudo_hdr_offset); | ||
1899 | odd_flag = MXGEFW_FLAGS_ALIGN_ODD; | ||
1900 | flags |= MXGEFW_FLAGS_CKSUM; | ||
1901 | } | ||
1902 | } | ||
1903 | |||
1904 | cum_len = 0; | ||
1905 | |||
1906 | #ifdef NETIF_F_TSO | ||
1907 | if (mss) { /* TSO */ | ||
1908 | /* this removes any CKSUM flag from before */ | ||
1909 | flags = (MXGEFW_FLAGS_TSO_HDR | MXGEFW_FLAGS_FIRST); | ||
1910 | |||
1911 | /* negative cum_len signifies to the | ||
1912 | * send loop that we are still in the | ||
1913 | * header portion of the TSO packet. | ||
1914 | * TSO header must be at most 134 bytes long */ | ||
1915 | cum_len = -((skb->h.raw - skb->data) + (skb->h.th->doff << 2)); | ||
1916 | |||
1917 | /* for TSO, pseudo_hdr_offset holds mss. | ||
1918 | * The firmware figures out where to put | ||
1919 | * the checksum by parsing the header. */ | ||
1920 | pseudo_hdr_offset = htons(mss); | ||
1921 | } else | ||
1922 | #endif /*NETIF_F_TSO */ | ||
1923 | /* Mark small packets, and pad out tiny packets */ | ||
1924 | if (skb->len <= MXGEFW_SEND_SMALL_SIZE) { | ||
1925 | flags |= MXGEFW_FLAGS_SMALL; | ||
1926 | |||
1927 | /* pad frames to at least ETH_ZLEN bytes */ | ||
1928 | if (unlikely(skb->len < ETH_ZLEN)) { | ||
1929 | skb = skb_padto(skb, ETH_ZLEN); | ||
1930 | if (skb == NULL) { | ||
1931 | /* The packet is gone, so we must | ||
1932 | * return 0 */ | ||
1933 | mgp->stats.tx_dropped += 1; | ||
1934 | return 0; | ||
1935 | } | ||
1936 | /* adjust the len to account for the zero pad | ||
1937 | * so that the nic can know how long it is */ | ||
1938 | skb->len = ETH_ZLEN; | ||
1939 | } | ||
1940 | } | ||
1941 | |||
1942 | /* map the skb for DMA */ | ||
1943 | len = skb->len - skb->data_len; | ||
1944 | idx = tx->req & tx->mask; | ||
1945 | tx->info[idx].skb = skb; | ||
1946 | bus = pci_map_single(mgp->pdev, skb->data, len, PCI_DMA_TODEVICE); | ||
1947 | pci_unmap_addr_set(&tx->info[idx], bus, bus); | ||
1948 | pci_unmap_len_set(&tx->info[idx], len, len); | ||
1949 | |||
1950 | frag_cnt = skb_shinfo(skb)->nr_frags; | ||
1951 | frag_idx = 0; | ||
1952 | count = 0; | ||
1953 | rdma_count = 0; | ||
1954 | |||
1955 | /* "rdma_count" is the number of RDMAs belonging to the | ||
1956 | * current packet BEFORE the current send request. For | ||
1957 | * non-TSO packets, this is equal to "count". | ||
1958 | * For TSO packets, rdma_count needs to be reset | ||
1959 | * to 0 after a segment cut. | ||
1960 | * | ||
1961 | * The rdma_count field of the send request is | ||
1962 | * the number of RDMAs of the packet starting at | ||
1963 | * that request. For TSO send requests with one ore more cuts | ||
1964 | * in the middle, this is the number of RDMAs starting | ||
1965 | * after the last cut in the request. All previous | ||
1966 | * segments before the last cut implicitly have 1 RDMA. | ||
1967 | * | ||
1968 | * Since the number of RDMAs is not known beforehand, | ||
1969 | * it must be filled-in retroactively - after each | ||
1970 | * segmentation cut or at the end of the entire packet. | ||
1971 | */ | ||
1972 | |||
1973 | while (1) { | ||
1974 | /* Break the SKB or Fragment up into pieces which | ||
1975 | * do not cross mgp->tx.boundary */ | ||
1976 | low = MYRI10GE_LOWPART_TO_U32(bus); | ||
1977 | high_swapped = htonl(MYRI10GE_HIGHPART_TO_U32(bus)); | ||
1978 | while (len) { | ||
1979 | u8 flags_next; | ||
1980 | int cum_len_next; | ||
1981 | |||
1982 | if (unlikely(count == max_segments)) | ||
1983 | goto abort_linearize; | ||
1984 | |||
1985 | boundary = (low + tx->boundary) & ~(tx->boundary - 1); | ||
1986 | seglen = boundary - low; | ||
1987 | if (seglen > len) | ||
1988 | seglen = len; | ||
1989 | flags_next = flags & ~MXGEFW_FLAGS_FIRST; | ||
1990 | cum_len_next = cum_len + seglen; | ||
1991 | #ifdef NETIF_F_TSO | ||
1992 | if (mss) { /* TSO */ | ||
1993 | (req - rdma_count)->rdma_count = rdma_count + 1; | ||
1994 | |||
1995 | if (likely(cum_len >= 0)) { /* payload */ | ||
1996 | int next_is_first, chop; | ||
1997 | |||
1998 | chop = (cum_len_next > mss); | ||
1999 | cum_len_next = cum_len_next % mss; | ||
2000 | next_is_first = (cum_len_next == 0); | ||
2001 | flags |= chop * MXGEFW_FLAGS_TSO_CHOP; | ||
2002 | flags_next |= next_is_first * | ||
2003 | MXGEFW_FLAGS_FIRST; | ||
2004 | rdma_count |= -(chop | next_is_first); | ||
2005 | rdma_count += chop & !next_is_first; | ||
2006 | } else if (likely(cum_len_next >= 0)) { /* header ends */ | ||
2007 | int small; | ||
2008 | |||
2009 | rdma_count = -1; | ||
2010 | cum_len_next = 0; | ||
2011 | seglen = -cum_len; | ||
2012 | small = (mss <= MXGEFW_SEND_SMALL_SIZE); | ||
2013 | flags_next = MXGEFW_FLAGS_TSO_PLD | | ||
2014 | MXGEFW_FLAGS_FIRST | | ||
2015 | (small * MXGEFW_FLAGS_SMALL); | ||
2016 | } | ||
2017 | } | ||
2018 | #endif /* NETIF_F_TSO */ | ||
2019 | req->addr_high = high_swapped; | ||
2020 | req->addr_low = htonl(low); | ||
2021 | req->pseudo_hdr_offset = pseudo_hdr_offset; | ||
2022 | req->pad = 0; /* complete solid 16-byte block; does this matter? */ | ||
2023 | req->rdma_count = 1; | ||
2024 | req->length = htons(seglen); | ||
2025 | req->cksum_offset = cksum_offset; | ||
2026 | req->flags = flags | ((cum_len & 1) * odd_flag); | ||
2027 | |||
2028 | low += seglen; | ||
2029 | len -= seglen; | ||
2030 | cum_len = cum_len_next; | ||
2031 | flags = flags_next; | ||
2032 | req++; | ||
2033 | count++; | ||
2034 | rdma_count++; | ||
2035 | if (unlikely(cksum_offset > seglen)) | ||
2036 | cksum_offset -= seglen; | ||
2037 | else | ||
2038 | cksum_offset = 0; | ||
2039 | } | ||
2040 | if (frag_idx == frag_cnt) | ||
2041 | break; | ||
2042 | |||
2043 | /* map next fragment for DMA */ | ||
2044 | idx = (count + tx->req) & tx->mask; | ||
2045 | frag = &skb_shinfo(skb)->frags[frag_idx]; | ||
2046 | frag_idx++; | ||
2047 | len = frag->size; | ||
2048 | bus = pci_map_page(mgp->pdev, frag->page, frag->page_offset, | ||
2049 | len, PCI_DMA_TODEVICE); | ||
2050 | pci_unmap_addr_set(&tx->info[idx], bus, bus); | ||
2051 | pci_unmap_len_set(&tx->info[idx], len, len); | ||
2052 | } | ||
2053 | |||
2054 | (req - rdma_count)->rdma_count = rdma_count; | ||
2055 | #ifdef NETIF_F_TSO | ||
2056 | if (mss) | ||
2057 | do { | ||
2058 | req--; | ||
2059 | req->flags |= MXGEFW_FLAGS_TSO_LAST; | ||
2060 | } while (!(req->flags & (MXGEFW_FLAGS_TSO_CHOP | | ||
2061 | MXGEFW_FLAGS_FIRST))); | ||
2062 | #endif | ||
2063 | idx = ((count - 1) + tx->req) & tx->mask; | ||
2064 | tx->info[idx].last = 1; | ||
2065 | if (tx->wc_fifo == NULL) | ||
2066 | myri10ge_submit_req(tx, tx->req_list, count); | ||
2067 | else | ||
2068 | myri10ge_submit_req_wc(tx, tx->req_list, count); | ||
2069 | tx->pkt_start++; | ||
2070 | if ((avail - count) < MXGEFW_MAX_SEND_DESC) { | ||
2071 | mgp->stop_queue++; | ||
2072 | netif_stop_queue(dev); | ||
2073 | } | ||
2074 | dev->trans_start = jiffies; | ||
2075 | return 0; | ||
2076 | |||
2077 | abort_linearize: | ||
2078 | /* Free any DMA resources we've alloced and clear out the skb | ||
2079 | * slot so as to not trip up assertions, and to avoid a | ||
2080 | * double-free if linearizing fails */ | ||
2081 | |||
2082 | last_idx = (idx + 1) & tx->mask; | ||
2083 | idx = tx->req & tx->mask; | ||
2084 | tx->info[idx].skb = NULL; | ||
2085 | do { | ||
2086 | len = pci_unmap_len(&tx->info[idx], len); | ||
2087 | if (len) { | ||
2088 | if (tx->info[idx].skb != NULL) | ||
2089 | pci_unmap_single(mgp->pdev, | ||
2090 | pci_unmap_addr(&tx->info[idx], | ||
2091 | bus), len, | ||
2092 | PCI_DMA_TODEVICE); | ||
2093 | else | ||
2094 | pci_unmap_page(mgp->pdev, | ||
2095 | pci_unmap_addr(&tx->info[idx], | ||
2096 | bus), len, | ||
2097 | PCI_DMA_TODEVICE); | ||
2098 | pci_unmap_len_set(&tx->info[idx], len, 0); | ||
2099 | tx->info[idx].skb = NULL; | ||
2100 | } | ||
2101 | idx = (idx + 1) & tx->mask; | ||
2102 | } while (idx != last_idx); | ||
2103 | if (skb_shinfo(skb)->tso_size) { | ||
2104 | printk(KERN_ERR | ||
2105 | "myri10ge: %s: TSO but wanted to linearize?!?!?\n", | ||
2106 | mgp->dev->name); | ||
2107 | goto drop; | ||
2108 | } | ||
2109 | |||
2110 | if (skb_linearize(skb, GFP_ATOMIC)) | ||
2111 | goto drop; | ||
2112 | |||
2113 | mgp->tx_linearized++; | ||
2114 | goto again; | ||
2115 | |||
2116 | drop: | ||
2117 | dev_kfree_skb_any(skb); | ||
2118 | mgp->stats.tx_dropped += 1; | ||
2119 | return 0; | ||
2120 | |||
2121 | } | ||
2122 | |||
2123 | static struct net_device_stats *myri10ge_get_stats(struct net_device *dev) | ||
2124 | { | ||
2125 | struct myri10ge_priv *mgp = netdev_priv(dev); | ||
2126 | return &mgp->stats; | ||
2127 | } | ||
2128 | |||
2129 | static void myri10ge_set_multicast_list(struct net_device *dev) | ||
2130 | { | ||
2131 | /* can be called from atomic contexts, | ||
2132 | * pass 1 to force atomicity in myri10ge_send_cmd() */ | ||
2133 | myri10ge_change_promisc(netdev_priv(dev), dev->flags & IFF_PROMISC, 1); | ||
2134 | } | ||
2135 | |||
2136 | static int myri10ge_set_mac_address(struct net_device *dev, void *addr) | ||
2137 | { | ||
2138 | struct sockaddr *sa = addr; | ||
2139 | struct myri10ge_priv *mgp = netdev_priv(dev); | ||
2140 | int status; | ||
2141 | |||
2142 | if (!is_valid_ether_addr(sa->sa_data)) | ||
2143 | return -EADDRNOTAVAIL; | ||
2144 | |||
2145 | status = myri10ge_update_mac_address(mgp, sa->sa_data); | ||
2146 | if (status != 0) { | ||
2147 | printk(KERN_ERR | ||
2148 | "myri10ge: %s: changing mac address failed with %d\n", | ||
2149 | dev->name, status); | ||
2150 | return status; | ||
2151 | } | ||
2152 | |||
2153 | /* change the dev structure */ | ||
2154 | memcpy(dev->dev_addr, sa->sa_data, 6); | ||
2155 | return 0; | ||
2156 | } | ||
2157 | |||
2158 | static int myri10ge_change_mtu(struct net_device *dev, int new_mtu) | ||
2159 | { | ||
2160 | struct myri10ge_priv *mgp = netdev_priv(dev); | ||
2161 | int error = 0; | ||
2162 | |||
2163 | if ((new_mtu < 68) || (ETH_HLEN + new_mtu > MYRI10GE_MAX_ETHER_MTU)) { | ||
2164 | printk(KERN_ERR "myri10ge: %s: new mtu (%d) is not valid\n", | ||
2165 | dev->name, new_mtu); | ||
2166 | return -EINVAL; | ||
2167 | } | ||
2168 | printk(KERN_INFO "%s: changing mtu from %d to %d\n", | ||
2169 | dev->name, dev->mtu, new_mtu); | ||
2170 | if (mgp->running) { | ||
2171 | /* if we change the mtu on an active device, we must | ||
2172 | * reset the device so the firmware sees the change */ | ||
2173 | myri10ge_close(dev); | ||
2174 | dev->mtu = new_mtu; | ||
2175 | myri10ge_open(dev); | ||
2176 | } else | ||
2177 | dev->mtu = new_mtu; | ||
2178 | |||
2179 | return error; | ||
2180 | } | ||
2181 | |||
2182 | /* | ||
2183 | * Enable ECRC to align PCI-E Completion packets on an 8-byte boundary. | ||
2184 | * Only do it if the bridge is a root port since we don't want to disturb | ||
2185 | * any other device, except if forced with myri10ge_ecrc_enable > 1. | ||
2186 | */ | ||
2187 | |||
2188 | #define PCI_DEVICE_ID_NVIDIA_NFORCE_CK804_PCIE 0x005d | ||
2189 | |||
2190 | static void myri10ge_enable_ecrc(struct myri10ge_priv *mgp) | ||
2191 | { | ||
2192 | struct pci_dev *bridge = mgp->pdev->bus->self; | ||
2193 | struct device *dev = &mgp->pdev->dev; | ||
2194 | unsigned cap; | ||
2195 | unsigned err_cap; | ||
2196 | u16 val; | ||
2197 | u8 ext_type; | ||
2198 | int ret; | ||
2199 | |||
2200 | if (!myri10ge_ecrc_enable || !bridge) | ||
2201 | return; | ||
2202 | |||
2203 | /* check that the bridge is a root port */ | ||
2204 | cap = pci_find_capability(bridge, PCI_CAP_ID_EXP); | ||
2205 | pci_read_config_word(bridge, cap + PCI_CAP_FLAGS, &val); | ||
2206 | ext_type = (val & PCI_EXP_FLAGS_TYPE) >> 4; | ||
2207 | if (ext_type != PCI_EXP_TYPE_ROOT_PORT) { | ||
2208 | if (myri10ge_ecrc_enable > 1) { | ||
2209 | struct pci_dev *old_bridge = bridge; | ||
2210 | |||
2211 | /* Walk the hierarchy up to the root port | ||
2212 | * where ECRC has to be enabled */ | ||
2213 | do { | ||
2214 | bridge = bridge->bus->self; | ||
2215 | if (!bridge) { | ||
2216 | dev_err(dev, | ||
2217 | "Failed to find root port" | ||
2218 | " to force ECRC\n"); | ||
2219 | return; | ||
2220 | } | ||
2221 | cap = | ||
2222 | pci_find_capability(bridge, PCI_CAP_ID_EXP); | ||
2223 | pci_read_config_word(bridge, | ||
2224 | cap + PCI_CAP_FLAGS, &val); | ||
2225 | ext_type = (val & PCI_EXP_FLAGS_TYPE) >> 4; | ||
2226 | } while (ext_type != PCI_EXP_TYPE_ROOT_PORT); | ||
2227 | |||
2228 | dev_info(dev, | ||
2229 | "Forcing ECRC on non-root port %s" | ||
2230 | " (enabling on root port %s)\n", | ||
2231 | pci_name(old_bridge), pci_name(bridge)); | ||
2232 | } else { | ||
2233 | dev_err(dev, | ||
2234 | "Not enabling ECRC on non-root port %s\n", | ||
2235 | pci_name(bridge)); | ||
2236 | return; | ||
2237 | } | ||
2238 | } | ||
2239 | |||
2240 | cap = pci_find_ext_capability(bridge, PCI_EXT_CAP_ID_ERR); | ||
2241 | /* nvidia ext cap is not always linked in ext cap chain */ | ||
2242 | if (!cap | ||
2243 | && bridge->vendor == PCI_VENDOR_ID_NVIDIA | ||
2244 | && bridge->device == PCI_DEVICE_ID_NVIDIA_NFORCE_CK804_PCIE) | ||
2245 | cap = 0x160; | ||
2246 | |||
2247 | if (!cap) | ||
2248 | return; | ||
2249 | |||
2250 | ret = pci_read_config_dword(bridge, cap + PCI_ERR_CAP, &err_cap); | ||
2251 | if (ret) { | ||
2252 | dev_err(dev, "failed reading ext-conf-space of %s\n", | ||
2253 | pci_name(bridge)); | ||
2254 | dev_err(dev, "\t pci=nommconf in use? " | ||
2255 | "or buggy/incomplete/absent ACPI MCFG attr?\n"); | ||
2256 | return; | ||
2257 | } | ||
2258 | if (!(err_cap & PCI_ERR_CAP_ECRC_GENC)) | ||
2259 | return; | ||
2260 | |||
2261 | err_cap |= PCI_ERR_CAP_ECRC_GENE; | ||
2262 | pci_write_config_dword(bridge, cap + PCI_ERR_CAP, err_cap); | ||
2263 | dev_info(dev, "Enabled ECRC on upstream bridge %s\n", pci_name(bridge)); | ||
2264 | mgp->tx.boundary = 4096; | ||
2265 | mgp->fw_name = myri10ge_fw_aligned; | ||
2266 | } | ||
2267 | |||
2268 | /* | ||
2269 | * The Lanai Z8E PCI-E interface achieves higher Read-DMA throughput | ||
2270 | * when the PCI-E Completion packets are aligned on an 8-byte | ||
2271 | * boundary. Some PCI-E chip sets always align Completion packets; on | ||
2272 | * the ones that do not, the alignment can be enforced by enabling | ||
2273 | * ECRC generation (if supported). | ||
2274 | * | ||
2275 | * When PCI-E Completion packets are not aligned, it is actually more | ||
2276 | * efficient to limit Read-DMA transactions to 2KB, rather than 4KB. | ||
2277 | * | ||
2278 | * If the driver can neither enable ECRC nor verify that it has | ||
2279 | * already been enabled, then it must use a firmware image which works | ||
2280 | * around unaligned completion packets (myri10ge_ethp_z8e.dat), and it | ||
2281 | * should also ensure that it never gives the device a Read-DMA which is | ||
2282 | * larger than 2KB by setting the tx.boundary to 2KB. If ECRC is | ||
2283 | * enabled, then the driver should use the aligned (myri10ge_eth_z8e.dat) | ||
2284 | * firmware image, and set tx.boundary to 4KB. | ||
2285 | */ | ||
2286 | |||
2287 | #define PCI_DEVICE_ID_SERVERWORKS_HT2000_PCIE 0x0132 | ||
2288 | |||
2289 | static void myri10ge_select_firmware(struct myri10ge_priv *mgp) | ||
2290 | { | ||
2291 | struct pci_dev *bridge = mgp->pdev->bus->self; | ||
2292 | |||
2293 | mgp->tx.boundary = 2048; | ||
2294 | mgp->fw_name = myri10ge_fw_unaligned; | ||
2295 | |||
2296 | if (myri10ge_force_firmware == 0) { | ||
2297 | myri10ge_enable_ecrc(mgp); | ||
2298 | |||
2299 | /* Check to see if the upstream bridge is known to | ||
2300 | * provide aligned completions */ | ||
2301 | if (bridge | ||
2302 | /* ServerWorks HT2000/HT1000 */ | ||
2303 | && bridge->vendor == PCI_VENDOR_ID_SERVERWORKS | ||
2304 | && bridge->device == | ||
2305 | PCI_DEVICE_ID_SERVERWORKS_HT2000_PCIE) { | ||
2306 | dev_info(&mgp->pdev->dev, | ||
2307 | "Assuming aligned completions (0x%x:0x%x)\n", | ||
2308 | bridge->vendor, bridge->device); | ||
2309 | mgp->tx.boundary = 4096; | ||
2310 | mgp->fw_name = myri10ge_fw_aligned; | ||
2311 | } | ||
2312 | } else { | ||
2313 | if (myri10ge_force_firmware == 1) { | ||
2314 | dev_info(&mgp->pdev->dev, | ||
2315 | "Assuming aligned completions (forced)\n"); | ||
2316 | mgp->tx.boundary = 4096; | ||
2317 | mgp->fw_name = myri10ge_fw_aligned; | ||
2318 | } else { | ||
2319 | dev_info(&mgp->pdev->dev, | ||
2320 | "Assuming unaligned completions (forced)\n"); | ||
2321 | mgp->tx.boundary = 2048; | ||
2322 | mgp->fw_name = myri10ge_fw_unaligned; | ||
2323 | } | ||
2324 | } | ||
2325 | if (myri10ge_fw_name != NULL) { | ||
2326 | dev_info(&mgp->pdev->dev, "overriding firmware to %s\n", | ||
2327 | myri10ge_fw_name); | ||
2328 | mgp->fw_name = myri10ge_fw_name; | ||
2329 | } | ||
2330 | } | ||
2331 | |||
2332 | static void myri10ge_save_state(struct myri10ge_priv *mgp) | ||
2333 | { | ||
2334 | struct pci_dev *pdev = mgp->pdev; | ||
2335 | int cap; | ||
2336 | |||
2337 | pci_save_state(pdev); | ||
2338 | /* now save PCIe and MSI state that Linux will not | ||
2339 | * save for us */ | ||
2340 | cap = pci_find_capability(pdev, PCI_CAP_ID_EXP); | ||
2341 | pci_read_config_dword(pdev, cap + PCI_EXP_DEVCTL, &mgp->devctl); | ||
2342 | cap = pci_find_capability(pdev, PCI_CAP_ID_MSI); | ||
2343 | pci_read_config_word(pdev, cap + PCI_MSI_FLAGS, &mgp->msi_flags); | ||
2344 | } | ||
2345 | |||
2346 | static void myri10ge_restore_state(struct myri10ge_priv *mgp) | ||
2347 | { | ||
2348 | struct pci_dev *pdev = mgp->pdev; | ||
2349 | int cap; | ||
2350 | |||
2351 | /* restore PCIe and MSI state that linux will not */ | ||
2352 | cap = pci_find_capability(pdev, PCI_CAP_ID_EXP); | ||
2353 | pci_write_config_dword(pdev, cap + PCI_CAP_ID_EXP, mgp->devctl); | ||
2354 | cap = pci_find_capability(pdev, PCI_CAP_ID_MSI); | ||
2355 | pci_write_config_word(pdev, cap + PCI_MSI_FLAGS, mgp->msi_flags); | ||
2356 | |||
2357 | pci_restore_state(pdev); | ||
2358 | } | ||
2359 | |||
2360 | #ifdef CONFIG_PM | ||
2361 | |||
2362 | static int myri10ge_suspend(struct pci_dev *pdev, pm_message_t state) | ||
2363 | { | ||
2364 | struct myri10ge_priv *mgp; | ||
2365 | struct net_device *netdev; | ||
2366 | |||
2367 | mgp = pci_get_drvdata(pdev); | ||
2368 | if (mgp == NULL) | ||
2369 | return -EINVAL; | ||
2370 | netdev = mgp->dev; | ||
2371 | |||
2372 | netif_device_detach(netdev); | ||
2373 | if (netif_running(netdev)) { | ||
2374 | printk(KERN_INFO "myri10ge: closing %s\n", netdev->name); | ||
2375 | rtnl_lock(); | ||
2376 | myri10ge_close(netdev); | ||
2377 | rtnl_unlock(); | ||
2378 | } | ||
2379 | myri10ge_dummy_rdma(mgp, 0); | ||
2380 | free_irq(pdev->irq, mgp); | ||
2381 | myri10ge_save_state(mgp); | ||
2382 | pci_disable_device(pdev); | ||
2383 | pci_set_power_state(pdev, pci_choose_state(pdev, state)); | ||
2384 | return 0; | ||
2385 | } | ||
2386 | |||
2387 | static int myri10ge_resume(struct pci_dev *pdev) | ||
2388 | { | ||
2389 | struct myri10ge_priv *mgp; | ||
2390 | struct net_device *netdev; | ||
2391 | int status; | ||
2392 | u16 vendor; | ||
2393 | |||
2394 | mgp = pci_get_drvdata(pdev); | ||
2395 | if (mgp == NULL) | ||
2396 | return -EINVAL; | ||
2397 | netdev = mgp->dev; | ||
2398 | pci_set_power_state(pdev, 0); /* zeros conf space as a side effect */ | ||
2399 | msleep(5); /* give card time to respond */ | ||
2400 | pci_read_config_word(mgp->pdev, PCI_VENDOR_ID, &vendor); | ||
2401 | if (vendor == 0xffff) { | ||
2402 | printk(KERN_ERR "myri10ge: %s: device disappeared!\n", | ||
2403 | mgp->dev->name); | ||
2404 | return -EIO; | ||
2405 | } | ||
2406 | myri10ge_restore_state(mgp); | ||
2407 | pci_enable_device(pdev); | ||
2408 | pci_set_master(pdev); | ||
2409 | |||
2410 | status = request_irq(pdev->irq, myri10ge_intr, SA_SHIRQ, | ||
2411 | netdev->name, mgp); | ||
2412 | if (status != 0) { | ||
2413 | dev_err(&pdev->dev, "failed to allocate IRQ\n"); | ||
2414 | goto abort_with_msi; | ||
2415 | } | ||
2416 | |||
2417 | myri10ge_reset(mgp); | ||
2418 | myri10ge_dummy_rdma(mgp, mgp->tx.boundary != 4096); | ||
2419 | |||
2420 | /* Save configuration space to be restored if the | ||
2421 | * nic resets due to a parity error */ | ||
2422 | myri10ge_save_state(mgp); | ||
2423 | |||
2424 | if (netif_running(netdev)) { | ||
2425 | rtnl_lock(); | ||
2426 | myri10ge_open(netdev); | ||
2427 | rtnl_unlock(); | ||
2428 | } | ||
2429 | netif_device_attach(netdev); | ||
2430 | |||
2431 | return 0; | ||
2432 | |||
2433 | abort_with_msi: | ||
2434 | return -EIO; | ||
2435 | |||
2436 | } | ||
2437 | |||
2438 | #endif /* CONFIG_PM */ | ||
2439 | |||
2440 | static u32 myri10ge_read_reboot(struct myri10ge_priv *mgp) | ||
2441 | { | ||
2442 | struct pci_dev *pdev = mgp->pdev; | ||
2443 | int vs = mgp->vendor_specific_offset; | ||
2444 | u32 reboot; | ||
2445 | |||
2446 | /*enter read32 mode */ | ||
2447 | pci_write_config_byte(pdev, vs + 0x10, 0x3); | ||
2448 | |||
2449 | /*read REBOOT_STATUS (0xfffffff0) */ | ||
2450 | pci_write_config_dword(pdev, vs + 0x18, 0xfffffff0); | ||
2451 | pci_read_config_dword(pdev, vs + 0x14, &reboot); | ||
2452 | return reboot; | ||
2453 | } | ||
2454 | |||
2455 | /* | ||
2456 | * This watchdog is used to check whether the board has suffered | ||
2457 | * from a parity error and needs to be recovered. | ||
2458 | */ | ||
2459 | static void myri10ge_watchdog(void *arg) | ||
2460 | { | ||
2461 | struct myri10ge_priv *mgp = arg; | ||
2462 | u32 reboot; | ||
2463 | int status; | ||
2464 | u16 cmd, vendor; | ||
2465 | |||
2466 | mgp->watchdog_resets++; | ||
2467 | pci_read_config_word(mgp->pdev, PCI_COMMAND, &cmd); | ||
2468 | if ((cmd & PCI_COMMAND_MASTER) == 0) { | ||
2469 | /* Bus master DMA disabled? Check to see | ||
2470 | * if the card rebooted due to a parity error | ||
2471 | * For now, just report it */ | ||
2472 | reboot = myri10ge_read_reboot(mgp); | ||
2473 | printk(KERN_ERR | ||
2474 | "myri10ge: %s: NIC rebooted (0x%x), resetting\n", | ||
2475 | mgp->dev->name, reboot); | ||
2476 | /* | ||
2477 | * A rebooted nic will come back with config space as | ||
2478 | * it was after power was applied to PCIe bus. | ||
2479 | * Attempt to restore config space which was saved | ||
2480 | * when the driver was loaded, or the last time the | ||
2481 | * nic was resumed from power saving mode. | ||
2482 | */ | ||
2483 | myri10ge_restore_state(mgp); | ||
2484 | } else { | ||
2485 | /* if we get back -1's from our slot, perhaps somebody | ||
2486 | * powered off our card. Don't try to reset it in | ||
2487 | * this case */ | ||
2488 | if (cmd == 0xffff) { | ||
2489 | pci_read_config_word(mgp->pdev, PCI_VENDOR_ID, &vendor); | ||
2490 | if (vendor == 0xffff) { | ||
2491 | printk(KERN_ERR | ||
2492 | "myri10ge: %s: device disappeared!\n", | ||
2493 | mgp->dev->name); | ||
2494 | return; | ||
2495 | } | ||
2496 | } | ||
2497 | /* Perhaps it is a software error. Try to reset */ | ||
2498 | |||
2499 | printk(KERN_ERR "myri10ge: %s: device timeout, resetting\n", | ||
2500 | mgp->dev->name); | ||
2501 | printk(KERN_INFO "myri10ge: %s: %d %d %d %d %d\n", | ||
2502 | mgp->dev->name, mgp->tx.req, mgp->tx.done, | ||
2503 | mgp->tx.pkt_start, mgp->tx.pkt_done, | ||
2504 | (int)ntohl(mgp->fw_stats->send_done_count)); | ||
2505 | msleep(2000); | ||
2506 | printk(KERN_INFO "myri10ge: %s: %d %d %d %d %d\n", | ||
2507 | mgp->dev->name, mgp->tx.req, mgp->tx.done, | ||
2508 | mgp->tx.pkt_start, mgp->tx.pkt_done, | ||
2509 | (int)ntohl(mgp->fw_stats->send_done_count)); | ||
2510 | } | ||
2511 | rtnl_lock(); | ||
2512 | myri10ge_close(mgp->dev); | ||
2513 | status = myri10ge_load_firmware(mgp); | ||
2514 | if (status != 0) | ||
2515 | printk(KERN_ERR "myri10ge: %s: failed to load firmware\n", | ||
2516 | mgp->dev->name); | ||
2517 | else | ||
2518 | myri10ge_open(mgp->dev); | ||
2519 | rtnl_unlock(); | ||
2520 | } | ||
2521 | |||
2522 | /* | ||
2523 | * We use our own timer routine rather than relying upon | ||
2524 | * netdev->tx_timeout because we have a very large hardware transmit | ||
2525 | * queue. Due to the large queue, the netdev->tx_timeout function | ||
2526 | * cannot detect a NIC with a parity error in a timely fashion if the | ||
2527 | * NIC is lightly loaded. | ||
2528 | */ | ||
2529 | static void myri10ge_watchdog_timer(unsigned long arg) | ||
2530 | { | ||
2531 | struct myri10ge_priv *mgp; | ||
2532 | |||
2533 | mgp = (struct myri10ge_priv *)arg; | ||
2534 | if (mgp->tx.req != mgp->tx.done && | ||
2535 | mgp->tx.done == mgp->watchdog_tx_done) | ||
2536 | /* nic seems like it might be stuck.. */ | ||
2537 | schedule_work(&mgp->watchdog_work); | ||
2538 | else | ||
2539 | /* rearm timer */ | ||
2540 | mod_timer(&mgp->watchdog_timer, | ||
2541 | jiffies + myri10ge_watchdog_timeout * HZ); | ||
2542 | |||
2543 | mgp->watchdog_tx_done = mgp->tx.done; | ||
2544 | } | ||
2545 | |||
2546 | static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent) | ||
2547 | { | ||
2548 | struct net_device *netdev; | ||
2549 | struct myri10ge_priv *mgp; | ||
2550 | struct device *dev = &pdev->dev; | ||
2551 | size_t bytes; | ||
2552 | int i; | ||
2553 | int status = -ENXIO; | ||
2554 | int cap; | ||
2555 | int dac_enabled; | ||
2556 | u16 val; | ||
2557 | |||
2558 | netdev = alloc_etherdev(sizeof(*mgp)); | ||
2559 | if (netdev == NULL) { | ||
2560 | dev_err(dev, "Could not allocate ethernet device\n"); | ||
2561 | return -ENOMEM; | ||
2562 | } | ||
2563 | |||
2564 | mgp = netdev_priv(netdev); | ||
2565 | memset(mgp, 0, sizeof(*mgp)); | ||
2566 | mgp->dev = netdev; | ||
2567 | mgp->pdev = pdev; | ||
2568 | mgp->csum_flag = MXGEFW_FLAGS_CKSUM; | ||
2569 | mgp->pause = myri10ge_flow_control; | ||
2570 | mgp->intr_coal_delay = myri10ge_intr_coal_delay; | ||
2571 | init_waitqueue_head(&mgp->down_wq); | ||
2572 | |||
2573 | if (pci_enable_device(pdev)) { | ||
2574 | dev_err(&pdev->dev, "pci_enable_device call failed\n"); | ||
2575 | status = -ENODEV; | ||
2576 | goto abort_with_netdev; | ||
2577 | } | ||
2578 | myri10ge_select_firmware(mgp); | ||
2579 | |||
2580 | /* Find the vendor-specific cap so we can check | ||
2581 | * the reboot register later on */ | ||
2582 | mgp->vendor_specific_offset | ||
2583 | = pci_find_capability(pdev, PCI_CAP_ID_VNDR); | ||
2584 | |||
2585 | /* Set our max read request to 4KB */ | ||
2586 | cap = pci_find_capability(pdev, PCI_CAP_ID_EXP); | ||
2587 | if (cap < 64) { | ||
2588 | dev_err(&pdev->dev, "Bad PCI_CAP_ID_EXP location %d\n", cap); | ||
2589 | goto abort_with_netdev; | ||
2590 | } | ||
2591 | status = pci_read_config_word(pdev, cap + PCI_EXP_DEVCTL, &val); | ||
2592 | if (status != 0) { | ||
2593 | dev_err(&pdev->dev, "Error %d reading PCI_EXP_DEVCTL\n", | ||
2594 | status); | ||
2595 | goto abort_with_netdev; | ||
2596 | } | ||
2597 | val = (val & ~PCI_EXP_DEVCTL_READRQ) | (5 << 12); | ||
2598 | status = pci_write_config_word(pdev, cap + PCI_EXP_DEVCTL, val); | ||
2599 | if (status != 0) { | ||
2600 | dev_err(&pdev->dev, "Error %d writing PCI_EXP_DEVCTL\n", | ||
2601 | status); | ||
2602 | goto abort_with_netdev; | ||
2603 | } | ||
2604 | |||
2605 | pci_set_master(pdev); | ||
2606 | dac_enabled = 1; | ||
2607 | status = pci_set_dma_mask(pdev, DMA_64BIT_MASK); | ||
2608 | if (status != 0) { | ||
2609 | dac_enabled = 0; | ||
2610 | dev_err(&pdev->dev, | ||
2611 | "64-bit pci address mask was refused, trying 32-bit"); | ||
2612 | status = pci_set_dma_mask(pdev, DMA_32BIT_MASK); | ||
2613 | } | ||
2614 | if (status != 0) { | ||
2615 | dev_err(&pdev->dev, "Error %d setting DMA mask\n", status); | ||
2616 | goto abort_with_netdev; | ||
2617 | } | ||
2618 | mgp->cmd = pci_alloc_consistent(pdev, sizeof(*mgp->cmd), &mgp->cmd_bus); | ||
2619 | if (mgp->cmd == NULL) | ||
2620 | goto abort_with_netdev; | ||
2621 | |||
2622 | mgp->fw_stats = pci_alloc_consistent(pdev, sizeof(*mgp->fw_stats), | ||
2623 | &mgp->fw_stats_bus); | ||
2624 | if (mgp->fw_stats == NULL) | ||
2625 | goto abort_with_cmd; | ||
2626 | |||
2627 | mgp->board_span = pci_resource_len(pdev, 0); | ||
2628 | mgp->iomem_base = pci_resource_start(pdev, 0); | ||
2629 | mgp->mtrr = -1; | ||
2630 | #ifdef CONFIG_MTRR | ||
2631 | mgp->mtrr = mtrr_add(mgp->iomem_base, mgp->board_span, | ||
2632 | MTRR_TYPE_WRCOMB, 1); | ||
2633 | #endif | ||
2634 | /* Hack. need to get rid of these magic numbers */ | ||
2635 | mgp->sram_size = | ||
2636 | 2 * 1024 * 1024 - (2 * (48 * 1024) + (32 * 1024)) - 0x100; | ||
2637 | if (mgp->sram_size > mgp->board_span) { | ||
2638 | dev_err(&pdev->dev, "board span %ld bytes too small\n", | ||
2639 | mgp->board_span); | ||
2640 | goto abort_with_wc; | ||
2641 | } | ||
2642 | mgp->sram = ioremap(mgp->iomem_base, mgp->board_span); | ||
2643 | if (mgp->sram == NULL) { | ||
2644 | dev_err(&pdev->dev, "ioremap failed for %ld bytes at 0x%lx\n", | ||
2645 | mgp->board_span, mgp->iomem_base); | ||
2646 | status = -ENXIO; | ||
2647 | goto abort_with_wc; | ||
2648 | } | ||
2649 | memcpy_fromio(mgp->eeprom_strings, | ||
2650 | mgp->sram + mgp->sram_size - MYRI10GE_EEPROM_STRINGS_SIZE, | ||
2651 | MYRI10GE_EEPROM_STRINGS_SIZE); | ||
2652 | memset(mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE - 2, 0, 2); | ||
2653 | status = myri10ge_read_mac_addr(mgp); | ||
2654 | if (status) | ||
2655 | goto abort_with_ioremap; | ||
2656 | |||
2657 | for (i = 0; i < ETH_ALEN; i++) | ||
2658 | netdev->dev_addr[i] = mgp->mac_addr[i]; | ||
2659 | |||
2660 | /* allocate rx done ring */ | ||
2661 | bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry); | ||
2662 | mgp->rx_done.entry = | ||
2663 | pci_alloc_consistent(pdev, bytes, &mgp->rx_done.bus); | ||
2664 | if (mgp->rx_done.entry == NULL) | ||
2665 | goto abort_with_ioremap; | ||
2666 | memset(mgp->rx_done.entry, 0, bytes); | ||
2667 | |||
2668 | status = myri10ge_load_firmware(mgp); | ||
2669 | if (status != 0) { | ||
2670 | dev_err(&pdev->dev, "failed to load firmware\n"); | ||
2671 | goto abort_with_rx_done; | ||
2672 | } | ||
2673 | |||
2674 | status = myri10ge_reset(mgp); | ||
2675 | if (status != 0) { | ||
2676 | dev_err(&pdev->dev, "failed reset\n"); | ||
2677 | goto abort_with_firmware; | ||
2678 | } | ||
2679 | |||
2680 | if (myri10ge_msi) { | ||
2681 | status = pci_enable_msi(pdev); | ||
2682 | if (status != 0) | ||
2683 | dev_err(&pdev->dev, | ||
2684 | "Error %d setting up MSI; falling back to xPIC\n", | ||
2685 | status); | ||
2686 | else | ||
2687 | mgp->msi_enabled = 1; | ||
2688 | } | ||
2689 | |||
2690 | status = request_irq(pdev->irq, myri10ge_intr, SA_SHIRQ, | ||
2691 | netdev->name, mgp); | ||
2692 | if (status != 0) { | ||
2693 | dev_err(&pdev->dev, "failed to allocate IRQ\n"); | ||
2694 | goto abort_with_firmware; | ||
2695 | } | ||
2696 | |||
2697 | pci_set_drvdata(pdev, mgp); | ||
2698 | if ((myri10ge_initial_mtu + ETH_HLEN) > MYRI10GE_MAX_ETHER_MTU) | ||
2699 | myri10ge_initial_mtu = MYRI10GE_MAX_ETHER_MTU - ETH_HLEN; | ||
2700 | if ((myri10ge_initial_mtu + ETH_HLEN) < 68) | ||
2701 | myri10ge_initial_mtu = 68; | ||
2702 | netdev->mtu = myri10ge_initial_mtu; | ||
2703 | netdev->open = myri10ge_open; | ||
2704 | netdev->stop = myri10ge_close; | ||
2705 | netdev->hard_start_xmit = myri10ge_xmit; | ||
2706 | netdev->get_stats = myri10ge_get_stats; | ||
2707 | netdev->base_addr = mgp->iomem_base; | ||
2708 | netdev->irq = pdev->irq; | ||
2709 | netdev->change_mtu = myri10ge_change_mtu; | ||
2710 | netdev->set_multicast_list = myri10ge_set_multicast_list; | ||
2711 | netdev->set_mac_address = myri10ge_set_mac_address; | ||
2712 | netdev->features = NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_TSO; | ||
2713 | if (dac_enabled) | ||
2714 | netdev->features |= NETIF_F_HIGHDMA; | ||
2715 | netdev->poll = myri10ge_poll; | ||
2716 | netdev->weight = myri10ge_napi_weight; | ||
2717 | |||
2718 | /* Save configuration space to be restored if the | ||
2719 | * nic resets due to a parity error */ | ||
2720 | myri10ge_save_state(mgp); | ||
2721 | /* Restore state immediately since pci_save_msi_state disables MSI */ | ||
2722 | myri10ge_restore_state(mgp); | ||
2723 | |||
2724 | /* Setup the watchdog timer */ | ||
2725 | setup_timer(&mgp->watchdog_timer, myri10ge_watchdog_timer, | ||
2726 | (unsigned long)mgp); | ||
2727 | |||
2728 | SET_ETHTOOL_OPS(netdev, &myri10ge_ethtool_ops); | ||
2729 | INIT_WORK(&mgp->watchdog_work, myri10ge_watchdog, mgp); | ||
2730 | status = register_netdev(netdev); | ||
2731 | if (status != 0) { | ||
2732 | dev_err(&pdev->dev, "register_netdev failed: %d\n", status); | ||
2733 | goto abort_with_irq; | ||
2734 | } | ||
2735 | |||
2736 | printk(KERN_INFO "myri10ge: %s: %s IRQ %d, tx bndry %d, fw %s, WC %s\n", | ||
2737 | netdev->name, (mgp->msi_enabled ? "MSI" : "xPIC"), | ||
2738 | pdev->irq, mgp->tx.boundary, mgp->fw_name, | ||
2739 | (mgp->mtrr >= 0 ? "Enabled" : "Disabled")); | ||
2740 | |||
2741 | return 0; | ||
2742 | |||
2743 | abort_with_irq: | ||
2744 | free_irq(pdev->irq, mgp); | ||
2745 | if (mgp->msi_enabled) | ||
2746 | pci_disable_msi(pdev); | ||
2747 | |||
2748 | abort_with_firmware: | ||
2749 | myri10ge_dummy_rdma(mgp, 0); | ||
2750 | |||
2751 | abort_with_rx_done: | ||
2752 | bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry); | ||
2753 | pci_free_consistent(pdev, bytes, mgp->rx_done.entry, mgp->rx_done.bus); | ||
2754 | |||
2755 | abort_with_ioremap: | ||
2756 | iounmap(mgp->sram); | ||
2757 | |||
2758 | abort_with_wc: | ||
2759 | #ifdef CONFIG_MTRR | ||
2760 | if (mgp->mtrr >= 0) | ||
2761 | mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span); | ||
2762 | #endif | ||
2763 | pci_free_consistent(pdev, sizeof(*mgp->fw_stats), | ||
2764 | mgp->fw_stats, mgp->fw_stats_bus); | ||
2765 | |||
2766 | abort_with_cmd: | ||
2767 | pci_free_consistent(pdev, sizeof(*mgp->cmd), mgp->cmd, mgp->cmd_bus); | ||
2768 | |||
2769 | abort_with_netdev: | ||
2770 | |||
2771 | free_netdev(netdev); | ||
2772 | return status; | ||
2773 | } | ||
2774 | |||
2775 | /* | ||
2776 | * myri10ge_remove | ||
2777 | * | ||
2778 | * Does what is necessary to shutdown one Myrinet device. Called | ||
2779 | * once for each Myrinet card by the kernel when a module is | ||
2780 | * unloaded. | ||
2781 | */ | ||
2782 | static void myri10ge_remove(struct pci_dev *pdev) | ||
2783 | { | ||
2784 | struct myri10ge_priv *mgp; | ||
2785 | struct net_device *netdev; | ||
2786 | size_t bytes; | ||
2787 | |||
2788 | mgp = pci_get_drvdata(pdev); | ||
2789 | if (mgp == NULL) | ||
2790 | return; | ||
2791 | |||
2792 | flush_scheduled_work(); | ||
2793 | netdev = mgp->dev; | ||
2794 | unregister_netdev(netdev); | ||
2795 | free_irq(pdev->irq, mgp); | ||
2796 | if (mgp->msi_enabled) | ||
2797 | pci_disable_msi(pdev); | ||
2798 | |||
2799 | myri10ge_dummy_rdma(mgp, 0); | ||
2800 | |||
2801 | bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry); | ||
2802 | pci_free_consistent(pdev, bytes, mgp->rx_done.entry, mgp->rx_done.bus); | ||
2803 | |||
2804 | iounmap(mgp->sram); | ||
2805 | |||
2806 | #ifdef CONFIG_MTRR | ||
2807 | if (mgp->mtrr >= 0) | ||
2808 | mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span); | ||
2809 | #endif | ||
2810 | pci_free_consistent(pdev, sizeof(*mgp->fw_stats), | ||
2811 | mgp->fw_stats, mgp->fw_stats_bus); | ||
2812 | |||
2813 | pci_free_consistent(pdev, sizeof(*mgp->cmd), mgp->cmd, mgp->cmd_bus); | ||
2814 | |||
2815 | free_netdev(netdev); | ||
2816 | pci_set_drvdata(pdev, NULL); | ||
2817 | } | ||
2818 | |||
2819 | #define PCI_DEVICE_ID_MYIRCOM_MYRI10GE_Z8E 0x0008 | ||
2820 | |||
2821 | static struct pci_device_id myri10ge_pci_tbl[] = { | ||
2822 | {PCI_DEVICE(PCI_VENDOR_ID_MYRICOM, PCI_DEVICE_ID_MYIRCOM_MYRI10GE_Z8E)}, | ||
2823 | {0}, | ||
2824 | }; | ||
2825 | |||
2826 | static struct pci_driver myri10ge_driver = { | ||
2827 | .name = "myri10ge", | ||
2828 | .probe = myri10ge_probe, | ||
2829 | .remove = myri10ge_remove, | ||
2830 | .id_table = myri10ge_pci_tbl, | ||
2831 | #ifdef CONFIG_PM | ||
2832 | .suspend = myri10ge_suspend, | ||
2833 | .resume = myri10ge_resume, | ||
2834 | #endif | ||
2835 | }; | ||
2836 | |||
2837 | static __init int myri10ge_init_module(void) | ||
2838 | { | ||
2839 | printk(KERN_INFO "%s: Version %s\n", myri10ge_driver.name, | ||
2840 | MYRI10GE_VERSION_STR); | ||
2841 | return pci_register_driver(&myri10ge_driver); | ||
2842 | } | ||
2843 | |||
2844 | module_init(myri10ge_init_module); | ||
2845 | |||
2846 | static __exit void myri10ge_cleanup_module(void) | ||
2847 | { | ||
2848 | pci_unregister_driver(&myri10ge_driver); | ||
2849 | } | ||
2850 | |||
2851 | module_exit(myri10ge_cleanup_module); | ||
diff --git a/drivers/net/myri10ge/myri10ge_mcp.h b/drivers/net/myri10ge/myri10ge_mcp.h new file mode 100644 index 000000000000..ac23e18db63d --- /dev/null +++ b/drivers/net/myri10ge/myri10ge_mcp.h | |||
@@ -0,0 +1,205 @@ | |||
1 | #ifndef __MYRI10GE_MCP_H__ | ||
2 | #define __MYRI10GE_MCP_H__ | ||
3 | |||
4 | #define MXGEFW_VERSION_MAJOR 1 | ||
5 | #define MXGEFW_VERSION_MINOR 4 | ||
6 | |||
7 | /* 8 Bytes */ | ||
8 | struct mcp_dma_addr { | ||
9 | u32 high; | ||
10 | u32 low; | ||
11 | }; | ||
12 | |||
13 | /* 4 Bytes */ | ||
14 | struct mcp_slot { | ||
15 | u16 checksum; | ||
16 | u16 length; | ||
17 | }; | ||
18 | |||
19 | /* 64 Bytes */ | ||
20 | struct mcp_cmd { | ||
21 | u32 cmd; | ||
22 | u32 data0; /* will be low portion if data > 32 bits */ | ||
23 | /* 8 */ | ||
24 | u32 data1; /* will be high portion if data > 32 bits */ | ||
25 | u32 data2; /* currently unused.. */ | ||
26 | /* 16 */ | ||
27 | struct mcp_dma_addr response_addr; | ||
28 | /* 24 */ | ||
29 | u8 pad[40]; | ||
30 | }; | ||
31 | |||
32 | /* 8 Bytes */ | ||
33 | struct mcp_cmd_response { | ||
34 | u32 data; | ||
35 | u32 result; | ||
36 | }; | ||
37 | |||
38 | /* | ||
39 | * flags used in mcp_kreq_ether_send_t: | ||
40 | * | ||
41 | * The SMALL flag is only needed in the first segment. It is raised | ||
42 | * for packets that are total less or equal 512 bytes. | ||
43 | * | ||
44 | * The CKSUM flag must be set in all segments. | ||
45 | * | ||
46 | * The PADDED flags is set if the packet needs to be padded, and it | ||
47 | * must be set for all segments. | ||
48 | * | ||
49 | * The MXGEFW_FLAGS_ALIGN_ODD must be set if the cumulative | ||
50 | * length of all previous segments was odd. | ||
51 | */ | ||
52 | |||
53 | #define MXGEFW_FLAGS_SMALL 0x1 | ||
54 | #define MXGEFW_FLAGS_TSO_HDR 0x1 | ||
55 | #define MXGEFW_FLAGS_FIRST 0x2 | ||
56 | #define MXGEFW_FLAGS_ALIGN_ODD 0x4 | ||
57 | #define MXGEFW_FLAGS_CKSUM 0x8 | ||
58 | #define MXGEFW_FLAGS_TSO_LAST 0x8 | ||
59 | #define MXGEFW_FLAGS_NO_TSO 0x10 | ||
60 | #define MXGEFW_FLAGS_TSO_CHOP 0x10 | ||
61 | #define MXGEFW_FLAGS_TSO_PLD 0x20 | ||
62 | |||
63 | #define MXGEFW_SEND_SMALL_SIZE 1520 | ||
64 | #define MXGEFW_MAX_MTU 9400 | ||
65 | |||
66 | union mcp_pso_or_cumlen { | ||
67 | u16 pseudo_hdr_offset; | ||
68 | u16 cum_len; | ||
69 | }; | ||
70 | |||
71 | #define MXGEFW_MAX_SEND_DESC 12 | ||
72 | #define MXGEFW_PAD 2 | ||
73 | |||
74 | /* 16 Bytes */ | ||
75 | struct mcp_kreq_ether_send { | ||
76 | u32 addr_high; | ||
77 | u32 addr_low; | ||
78 | u16 pseudo_hdr_offset; | ||
79 | u16 length; | ||
80 | u8 pad; | ||
81 | u8 rdma_count; | ||
82 | u8 cksum_offset; /* where to start computing cksum */ | ||
83 | u8 flags; /* as defined above */ | ||
84 | }; | ||
85 | |||
86 | /* 8 Bytes */ | ||
87 | struct mcp_kreq_ether_recv { | ||
88 | u32 addr_high; | ||
89 | u32 addr_low; | ||
90 | }; | ||
91 | |||
92 | /* Commands */ | ||
93 | |||
94 | #define MXGEFW_CMD_OFFSET 0xf80000 | ||
95 | |||
96 | enum myri10ge_mcp_cmd_type { | ||
97 | MXGEFW_CMD_NONE = 0, | ||
98 | /* Reset the mcp, it is left in a safe state, waiting | ||
99 | * for the driver to set all its parameters */ | ||
100 | MXGEFW_CMD_RESET, | ||
101 | |||
102 | /* get the version number of the current firmware.. | ||
103 | * (may be available in the eeprom strings..? */ | ||
104 | MXGEFW_GET_MCP_VERSION, | ||
105 | |||
106 | /* Parameters which must be set by the driver before it can | ||
107 | * issue MXGEFW_CMD_ETHERNET_UP. They persist until the next | ||
108 | * MXGEFW_CMD_RESET is issued */ | ||
109 | |||
110 | MXGEFW_CMD_SET_INTRQ_DMA, | ||
111 | MXGEFW_CMD_SET_BIG_BUFFER_SIZE, /* in bytes, power of 2 */ | ||
112 | MXGEFW_CMD_SET_SMALL_BUFFER_SIZE, /* in bytes */ | ||
113 | |||
114 | /* Parameters which refer to lanai SRAM addresses where the | ||
115 | * driver must issue PIO writes for various things */ | ||
116 | |||
117 | MXGEFW_CMD_GET_SEND_OFFSET, | ||
118 | MXGEFW_CMD_GET_SMALL_RX_OFFSET, | ||
119 | MXGEFW_CMD_GET_BIG_RX_OFFSET, | ||
120 | MXGEFW_CMD_GET_IRQ_ACK_OFFSET, | ||
121 | MXGEFW_CMD_GET_IRQ_DEASSERT_OFFSET, | ||
122 | |||
123 | /* Parameters which refer to rings stored on the MCP, | ||
124 | * and whose size is controlled by the mcp */ | ||
125 | |||
126 | MXGEFW_CMD_GET_SEND_RING_SIZE, /* in bytes */ | ||
127 | MXGEFW_CMD_GET_RX_RING_SIZE, /* in bytes */ | ||
128 | |||
129 | /* Parameters which refer to rings stored in the host, | ||
130 | * and whose size is controlled by the host. Note that | ||
131 | * all must be physically contiguous and must contain | ||
132 | * a power of 2 number of entries. */ | ||
133 | |||
134 | MXGEFW_CMD_SET_INTRQ_SIZE, /* in bytes */ | ||
135 | |||
136 | /* command to bring ethernet interface up. Above parameters | ||
137 | * (plus mtu & mac address) must have been exchanged prior | ||
138 | * to issuing this command */ | ||
139 | MXGEFW_CMD_ETHERNET_UP, | ||
140 | |||
141 | /* command to bring ethernet interface down. No further sends | ||
142 | * or receives may be processed until an MXGEFW_CMD_ETHERNET_UP | ||
143 | * is issued, and all interrupt queues must be flushed prior | ||
144 | * to ack'ing this command */ | ||
145 | |||
146 | MXGEFW_CMD_ETHERNET_DOWN, | ||
147 | |||
148 | /* commands the driver may issue live, without resetting | ||
149 | * the nic. Note that increasing the mtu "live" should | ||
150 | * only be done if the driver has already supplied buffers | ||
151 | * sufficiently large to handle the new mtu. Decreasing | ||
152 | * the mtu live is safe */ | ||
153 | |||
154 | MXGEFW_CMD_SET_MTU, | ||
155 | MXGEFW_CMD_GET_INTR_COAL_DELAY_OFFSET, /* in microseconds */ | ||
156 | MXGEFW_CMD_SET_STATS_INTERVAL, /* in microseconds */ | ||
157 | MXGEFW_CMD_SET_STATS_DMA, | ||
158 | |||
159 | MXGEFW_ENABLE_PROMISC, | ||
160 | MXGEFW_DISABLE_PROMISC, | ||
161 | MXGEFW_SET_MAC_ADDRESS, | ||
162 | |||
163 | MXGEFW_ENABLE_FLOW_CONTROL, | ||
164 | MXGEFW_DISABLE_FLOW_CONTROL, | ||
165 | |||
166 | /* do a DMA test | ||
167 | * data0,data1 = DMA address | ||
168 | * data2 = RDMA length (MSH), WDMA length (LSH) | ||
169 | * command return data = repetitions (MSH), 0.5-ms ticks (LSH) | ||
170 | */ | ||
171 | MXGEFW_DMA_TEST | ||
172 | }; | ||
173 | |||
174 | enum myri10ge_mcp_cmd_status { | ||
175 | MXGEFW_CMD_OK = 0, | ||
176 | MXGEFW_CMD_UNKNOWN, | ||
177 | MXGEFW_CMD_ERROR_RANGE, | ||
178 | MXGEFW_CMD_ERROR_BUSY, | ||
179 | MXGEFW_CMD_ERROR_EMPTY, | ||
180 | MXGEFW_CMD_ERROR_CLOSED, | ||
181 | MXGEFW_CMD_ERROR_HASH_ERROR, | ||
182 | MXGEFW_CMD_ERROR_BAD_PORT, | ||
183 | MXGEFW_CMD_ERROR_RESOURCES | ||
184 | }; | ||
185 | |||
186 | /* 40 Bytes */ | ||
187 | struct mcp_irq_data { | ||
188 | u32 send_done_count; | ||
189 | |||
190 | u32 link_up; | ||
191 | u32 dropped_link_overflow; | ||
192 | u32 dropped_link_error_or_filtered; | ||
193 | u32 dropped_runt; | ||
194 | u32 dropped_overrun; | ||
195 | u32 dropped_no_small_buffer; | ||
196 | u32 dropped_no_big_buffer; | ||
197 | u32 rdma_tags_available; | ||
198 | |||
199 | u8 tx_stopped; | ||
200 | u8 link_down; | ||
201 | u8 stats_updated; | ||
202 | u8 valid; | ||
203 | }; | ||
204 | |||
205 | #endif /* __MYRI10GE_MCP_H__ */ | ||
diff --git a/drivers/net/myri10ge/myri10ge_mcp_gen_header.h b/drivers/net/myri10ge/myri10ge_mcp_gen_header.h new file mode 100644 index 000000000000..1d3a35f10789 --- /dev/null +++ b/drivers/net/myri10ge/myri10ge_mcp_gen_header.h | |||
@@ -0,0 +1,58 @@ | |||
1 | #ifndef __MYRI10GE_MCP_GEN_HEADER_H__ | ||
2 | #define __MYRI10GE_MCP_GEN_HEADER_H__ | ||
3 | |||
4 | /* this file define a standard header used as a first entry point to | ||
5 | * exchange information between firmware/driver and driver. The | ||
6 | * header structure can be anywhere in the mcp. It will usually be in | ||
7 | * the .data section, because some fields needs to be initialized at | ||
8 | * compile time. | ||
9 | * The 32bit word at offset MX_HEADER_PTR_OFFSET in the mcp must | ||
10 | * contains the location of the header. | ||
11 | * | ||
12 | * Typically a MCP will start with the following: | ||
13 | * .text | ||
14 | * .space 52 ! to help catch MEMORY_INT errors | ||
15 | * bt start ! jump to real code | ||
16 | * nop | ||
17 | * .long _gen_mcp_header | ||
18 | * | ||
19 | * The source will have a definition like: | ||
20 | * | ||
21 | * mcp_gen_header_t gen_mcp_header = { | ||
22 | * .header_length = sizeof(mcp_gen_header_t), | ||
23 | * .mcp_type = MCP_TYPE_XXX, | ||
24 | * .version = "something $Id: mcp_gen_header.h,v 1.2 2006/05/13 10:04:35 bgoglin Exp $", | ||
25 | * .mcp_globals = (unsigned)&Globals | ||
26 | * }; | ||
27 | */ | ||
28 | |||
29 | #define MCP_HEADER_PTR_OFFSET 0x3c | ||
30 | |||
31 | #define MCP_TYPE_MX 0x4d582020 /* "MX " */ | ||
32 | #define MCP_TYPE_PCIE 0x70636965 /* "PCIE" pcie-only MCP */ | ||
33 | #define MCP_TYPE_ETH 0x45544820 /* "ETH " */ | ||
34 | #define MCP_TYPE_MCP0 0x4d435030 /* "MCP0" */ | ||
35 | |||
36 | struct mcp_gen_header { | ||
37 | /* the first 4 fields are filled at compile time */ | ||
38 | unsigned header_length; | ||
39 | unsigned mcp_type; | ||
40 | char version[128]; | ||
41 | unsigned mcp_globals; /* pointer to mcp-type specific structure */ | ||
42 | |||
43 | /* filled by the MCP at run-time */ | ||
44 | unsigned sram_size; | ||
45 | unsigned string_specs; /* either the original STRING_SPECS or a superset */ | ||
46 | unsigned string_specs_len; | ||
47 | |||
48 | /* Fields above this comment are guaranteed to be present. | ||
49 | * | ||
50 | * Fields below this comment are extensions added in later versions | ||
51 | * of this struct, drivers should compare the header_length against | ||
52 | * offsetof(field) to check wether a given MCP implements them. | ||
53 | * | ||
54 | * Never remove any field. Keep everything naturally align. | ||
55 | */ | ||
56 | }; | ||
57 | |||
58 | #endif /* __MYRI10GE_MCP_GEN_HEADER_H__ */ | ||