?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
added 2.6.29.6 aldebaran kernel
[nao-ulib.git] / kernel / 2.6.29.6-aldebaran-rt / drivers / net / myri10ge / myri10ge.c
blob53ff2385da08b96af460d46115a8428c5acffddc
1 /*************************************************************************
2 * myri10ge.c: Myricom Myri-10G Ethernet driver.
3 *
4 * Copyright (C) 2005 - 2009 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 COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20 * AND 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 COPYRIGHT OWNER OR CONTRIBUTORS BE
23 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF 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/dma-mapping.h>
48 #include <linux/etherdevice.h>
49 #include <linux/if_ether.h>
50 #include <linux/if_vlan.h>
51 #include <linux/inet_lro.h>
52 #include <linux/dca.h>
53 #include <linux/ip.h>
54 #include <linux/inet.h>
55 #include <linux/in.h>
56 #include <linux/ethtool.h>
57 #include <linux/firmware.h>
58 #include <linux/delay.h>
59 #include <linux/timer.h>
60 #include <linux/vmalloc.h>
61 #include <linux/crc32.h>
62 #include <linux/moduleparam.h>
63 #include <linux/io.h>
64 #include <linux/log2.h>
65 #include <net/checksum.h>
66 #include <net/ip.h>
67 #include <net/tcp.h>
68 #include <asm/byteorder.h>
69 #include <asm/io.h>
70 #include <asm/processor.h>
71 #ifdef CONFIG_MTRR
72 #include <asm/mtrr.h>
73 #endif
74
75 #include "myri10ge_mcp.h"
76 #include "myri10ge_mcp_gen_header.h"
77
78 #define MYRI10GE_VERSION_STR "1.4.4-1.401"
79
80 MODULE_DESCRIPTION("Myricom 10G driver (10GbE)");
81 MODULE_AUTHOR("Maintainer: help@myri.com");
82 MODULE_VERSION(MYRI10GE_VERSION_STR);
83 MODULE_LICENSE("Dual BSD/GPL");
84
85 #define MYRI10GE_MAX_ETHER_MTU 9014
86
87 #define MYRI10GE_ETH_STOPPED 0
88 #define MYRI10GE_ETH_STOPPING 1
89 #define MYRI10GE_ETH_STARTING 2
90 #define MYRI10GE_ETH_RUNNING 3
91 #define MYRI10GE_ETH_OPEN_FAILED 4
92
93 #define MYRI10GE_EEPROM_STRINGS_SIZE 256
94 #define MYRI10GE_MAX_SEND_DESC_TSO ((65536 / 2048) * 2)
95 #define MYRI10GE_MAX_LRO_DESCRIPTORS 8
96 #define MYRI10GE_LRO_MAX_PKTS 64
97
98 #define MYRI10GE_NO_CONFIRM_DATA htonl(0xffffffff)
99 #define MYRI10GE_NO_RESPONSE_RESULT 0xffffffff
100
101 #define MYRI10GE_ALLOC_ORDER 0
102 #define MYRI10GE_ALLOC_SIZE ((1 << MYRI10GE_ALLOC_ORDER) * PAGE_SIZE)
103 #define MYRI10GE_MAX_FRAGS_PER_FRAME (MYRI10GE_MAX_ETHER_MTU/MYRI10GE_ALLOC_SIZE + 1)
104
105 #define MYRI10GE_MAX_SLICES 32
106
107 struct myri10ge_rx_buffer_state {
108 struct page *page;
109 int page_offset;
110 DECLARE_PCI_UNMAP_ADDR(bus)
111 DECLARE_PCI_UNMAP_LEN(len)
112 };
113
114 struct myri10ge_tx_buffer_state {
115 struct sk_buff *skb;
116 int last;
117 DECLARE_PCI_UNMAP_ADDR(bus)
118 DECLARE_PCI_UNMAP_LEN(len)
119 };
120
121 struct myri10ge_cmd {
122 u32 data0;
123 u32 data1;
124 u32 data2;
125 };
126
127 struct myri10ge_rx_buf {
128 struct mcp_kreq_ether_recv __iomem *lanai; /* lanai ptr for recv ring */
129 struct mcp_kreq_ether_recv *shadow; /* host shadow of recv ring */
130 struct myri10ge_rx_buffer_state *info;
131 struct page *page;
132 dma_addr_t bus;
133 int page_offset;
134 int cnt;
135 int fill_cnt;
136 int alloc_fail;
137 int mask; /* number of rx slots -1 */
138 int watchdog_needed;
139 };
140
141 struct myri10ge_tx_buf {
142 struct mcp_kreq_ether_send __iomem *lanai; /* lanai ptr for sendq */
143 __be32 __iomem *send_go; /* "go" doorbell ptr */
144 __be32 __iomem *send_stop; /* "stop" doorbell ptr */
145 struct mcp_kreq_ether_send *req_list; /* host shadow of sendq */
146 char *req_bytes;
147 struct myri10ge_tx_buffer_state *info;
148 int mask; /* number of transmit slots -1 */
149 int req ____cacheline_aligned; /* transmit slots submitted */
150 int pkt_start; /* packets started */
151 int stop_queue;
152 int linearized;
153 int done ____cacheline_aligned; /* transmit slots completed */
154 int pkt_done; /* packets completed */
155 int wake_queue;
156 int queue_active;
157 };
158
159 struct myri10ge_rx_done {
160 struct mcp_slot *entry;
161 dma_addr_t bus;
162 int cnt;
163 int idx;
164 struct net_lro_mgr lro_mgr;
165 struct net_lro_desc lro_desc[MYRI10GE_MAX_LRO_DESCRIPTORS];
166 };
167
168 struct myri10ge_slice_netstats {
169 unsigned long rx_packets;
170 unsigned long tx_packets;
171 unsigned long rx_bytes;
172 unsigned long tx_bytes;
173 unsigned long rx_dropped;
174 unsigned long tx_dropped;
175 };
176
177 struct myri10ge_slice_state {
178 struct myri10ge_tx_buf tx; /* transmit ring */
179 struct myri10ge_rx_buf rx_small;
180 struct myri10ge_rx_buf rx_big;
181 struct myri10ge_rx_done rx_done;
182 struct net_device *dev;
183 struct napi_struct napi;
184 struct myri10ge_priv *mgp;
185 struct myri10ge_slice_netstats stats;
186 __be32 __iomem *irq_claim;
187 struct mcp_irq_data *fw_stats;
188 dma_addr_t fw_stats_bus;
189 int watchdog_tx_done;
190 int watchdog_tx_req;
191 #ifdef CONFIG_MYRI10GE_DCA
192 int cached_dca_tag;
193 int cpu;
194 __be32 __iomem *dca_tag;
195 #endif
196 char irq_desc[32];
197 };
198
199 struct myri10ge_priv {
200 struct myri10ge_slice_state *ss;
201 int tx_boundary; /* boundary transmits cannot cross */
202 int num_slices;
203 int running; /* running? */
204 int csum_flag; /* rx_csums? */
205 int small_bytes;
206 int big_bytes;
207 int max_intr_slots;
208 struct net_device *dev;
209 struct net_device_stats stats;
210 spinlock_t stats_lock;
211 u8 __iomem *sram;
212 int sram_size;
213 unsigned long board_span;
214 unsigned long iomem_base;
215 __be32 __iomem *irq_deassert;
216 char *mac_addr_string;
217 struct mcp_cmd_response *cmd;
218 dma_addr_t cmd_bus;
219 struct pci_dev *pdev;
220 int msi_enabled;
221 int msix_enabled;
222 struct msix_entry *msix_vectors;
223 #ifdef CONFIG_MYRI10GE_DCA
224 int dca_enabled;
225 #endif
226 u32 link_state;
227 unsigned int rdma_tags_available;
228 int intr_coal_delay;
229 __be32 __iomem *intr_coal_delay_ptr;
230 int mtrr;
231 int wc_enabled;
232 int down_cnt;
233 wait_queue_head_t down_wq;
234 struct work_struct watchdog_work;
235 struct timer_list watchdog_timer;
236 int watchdog_resets;
237 int watchdog_pause;
238 int pause;
239 char *fw_name;
240 char eeprom_strings[MYRI10GE_EEPROM_STRINGS_SIZE];
241 char *product_code_string;
242 char fw_version[128];
243 int fw_ver_major;
244 int fw_ver_minor;
245 int fw_ver_tiny;
246 int adopted_rx_filter_bug;
247 u8 mac_addr[6]; /* eeprom mac address */
248 unsigned long serial_number;
249 int vendor_specific_offset;
250 int fw_multicast_support;
251 unsigned long features;
252 u32 max_tso6;
253 u32 read_dma;
254 u32 write_dma;
255 u32 read_write_dma;
256 u32 link_changes;
257 u32 msg_enable;
258 };
259
260 static char *myri10ge_fw_unaligned = "myri10ge_ethp_z8e.dat";
261 static char *myri10ge_fw_aligned = "myri10ge_eth_z8e.dat";
262 static char *myri10ge_fw_rss_unaligned = "myri10ge_rss_ethp_z8e.dat";
263 static char *myri10ge_fw_rss_aligned = "myri10ge_rss_eth_z8e.dat";
264
265 static char *myri10ge_fw_name = NULL;
266 module_param(myri10ge_fw_name, charp, S_IRUGO | S_IWUSR);
267 MODULE_PARM_DESC(myri10ge_fw_name, "Firmware image name");
268
269 static int myri10ge_ecrc_enable = 1;
270 module_param(myri10ge_ecrc_enable, int, S_IRUGO);
271 MODULE_PARM_DESC(myri10ge_ecrc_enable, "Enable Extended CRC on PCI-E");
272
273 static int myri10ge_small_bytes = -1; /* -1 == auto */
274 module_param(myri10ge_small_bytes, int, S_IRUGO | S_IWUSR);
275 MODULE_PARM_DESC(myri10ge_small_bytes, "Threshold of small packets");
276
277 static int myri10ge_msi = 1; /* enable msi by default */
278 module_param(myri10ge_msi, int, S_IRUGO | S_IWUSR);
279 MODULE_PARM_DESC(myri10ge_msi, "Enable Message Signalled Interrupts");
280
281 static int myri10ge_intr_coal_delay = 75;
282 module_param(myri10ge_intr_coal_delay, int, S_IRUGO);
283 MODULE_PARM_DESC(myri10ge_intr_coal_delay, "Interrupt coalescing delay");
284
285 static int myri10ge_flow_control = 1;
286 module_param(myri10ge_flow_control, int, S_IRUGO);
287 MODULE_PARM_DESC(myri10ge_flow_control, "Pause parameter");
288
289 static int myri10ge_deassert_wait = 1;
290 module_param(myri10ge_deassert_wait, int, S_IRUGO | S_IWUSR);
291 MODULE_PARM_DESC(myri10ge_deassert_wait,
292 "Wait when deasserting legacy interrupts");
293
294 static int myri10ge_force_firmware = 0;
295 module_param(myri10ge_force_firmware, int, S_IRUGO);
296 MODULE_PARM_DESC(myri10ge_force_firmware,
297 "Force firmware to assume aligned completions");
298
299 static int myri10ge_initial_mtu = MYRI10GE_MAX_ETHER_MTU - ETH_HLEN;
300 module_param(myri10ge_initial_mtu, int, S_IRUGO);
301 MODULE_PARM_DESC(myri10ge_initial_mtu, "Initial MTU");
302
303 static int myri10ge_napi_weight = 64;
304 module_param(myri10ge_napi_weight, int, S_IRUGO);
305 MODULE_PARM_DESC(myri10ge_napi_weight, "Set NAPI weight");
306
307 static int myri10ge_watchdog_timeout = 1;
308 module_param(myri10ge_watchdog_timeout, int, S_IRUGO);
309 MODULE_PARM_DESC(myri10ge_watchdog_timeout, "Set watchdog timeout");
310
311 static int myri10ge_max_irq_loops = 1048576;
312 module_param(myri10ge_max_irq_loops, int, S_IRUGO);
313 MODULE_PARM_DESC(myri10ge_max_irq_loops,
314 "Set stuck legacy IRQ detection threshold");
315
316 #define MYRI10GE_MSG_DEFAULT NETIF_MSG_LINK
317
318 static int myri10ge_debug = -1; /* defaults above */
319 module_param(myri10ge_debug, int, 0);
320 MODULE_PARM_DESC(myri10ge_debug, "Debug level (0=none,...,16=all)");
321
322 static int myri10ge_lro = 1;
323 module_param(myri10ge_lro, int, S_IRUGO);
324 MODULE_PARM_DESC(myri10ge_lro, "Enable large receive offload");
325
326 static int myri10ge_lro_max_pkts = MYRI10GE_LRO_MAX_PKTS;
327 module_param(myri10ge_lro_max_pkts, int, S_IRUGO);
328 MODULE_PARM_DESC(myri10ge_lro_max_pkts,
329 "Number of LRO packets to be aggregated");
330
331 static int myri10ge_fill_thresh = 256;
332 module_param(myri10ge_fill_thresh, int, S_IRUGO | S_IWUSR);
333 MODULE_PARM_DESC(myri10ge_fill_thresh, "Number of empty rx slots allowed");
334
335 static int myri10ge_reset_recover = 1;
336
337 static int myri10ge_max_slices = 1;
338 module_param(myri10ge_max_slices, int, S_IRUGO);
339 MODULE_PARM_DESC(myri10ge_max_slices, "Max tx/rx queues");
340
341 static int myri10ge_rss_hash = MXGEFW_RSS_HASH_TYPE_SRC_PORT;
342 module_param(myri10ge_rss_hash, int, S_IRUGO);
343 MODULE_PARM_DESC(myri10ge_rss_hash, "Type of RSS hashing to do");
344
345 static int myri10ge_dca = 1;
346 module_param(myri10ge_dca, int, S_IRUGO);
347 MODULE_PARM_DESC(myri10ge_dca, "Enable DCA if possible");
348
349 #define MYRI10GE_FW_OFFSET 1024*1024
350 #define MYRI10GE_HIGHPART_TO_U32(X) \
351 (sizeof (X) == 8) ? ((u32)((u64)(X) >> 32)) : (0)
352 #define MYRI10GE_LOWPART_TO_U32(X) ((u32)(X))
353
354 #define myri10ge_pio_copy(to,from,size) __iowrite64_copy(to,from,size/8)
355
356 static void myri10ge_set_multicast_list(struct net_device *dev);
357 static int myri10ge_sw_tso(struct sk_buff *skb, struct net_device *dev);
358
359 static inline void put_be32(__be32 val, __be32 __iomem * p)
360 {
361 __raw_writel((__force __u32) val, (__force void __iomem *)p);
362 }
363
364 static int
365 myri10ge_send_cmd(struct myri10ge_priv *mgp, u32 cmd,
366 struct myri10ge_cmd *data, int atomic)
367 {
368 struct mcp_cmd *buf;
369 char buf_bytes[sizeof(*buf) + 8];
370 struct mcp_cmd_response *response = mgp->cmd;
371 char __iomem *cmd_addr = mgp->sram + MXGEFW_ETH_CMD;
372 u32 dma_low, dma_high, result, value;
373 int sleep_total = 0;
374
375 /* ensure buf is aligned to 8 bytes */
376 buf = (struct mcp_cmd *)ALIGN((unsigned long)buf_bytes, 8);
377
378 buf->data0 = htonl(data->data0);
379 buf->data1 = htonl(data->data1);
380 buf->data2 = htonl(data->data2);
381 buf->cmd = htonl(cmd);
382 dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
383 dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
384
385 buf->response_addr.low = htonl(dma_low);
386 buf->response_addr.high = htonl(dma_high);
387 response->result = htonl(MYRI10GE_NO_RESPONSE_RESULT);
388 mb();
389 myri10ge_pio_copy(cmd_addr, buf, sizeof(*buf));
390
391 /* wait up to 15ms. Longest command is the DMA benchmark,
392 * which is capped at 5ms, but runs from a timeout handler
393 * that runs every 7.8ms. So a 15ms timeout leaves us with
394 * a 2.2ms margin
395 */
396 if (atomic) {
397 /* if atomic is set, do not sleep,
398 * and try to get the completion quickly
399 * (1ms will be enough for those commands) */
400 for (sleep_total = 0;
401 sleep_total < 1000
402 && response->result == htonl(MYRI10GE_NO_RESPONSE_RESULT);
403 sleep_total += 10) {
404 udelay(10);
405 mb();
406 }
407 } else {
408 /* use msleep for most command */
409 for (sleep_total = 0;
410 sleep_total < 15
411 && response->result == htonl(MYRI10GE_NO_RESPONSE_RESULT);
412 sleep_total++)
413 msleep(1);
414 }
415
416 result = ntohl(response->result);
417 value = ntohl(response->data);
418 if (result != MYRI10GE_NO_RESPONSE_RESULT) {
419 if (result == 0) {
420 data->data0 = value;
421 return 0;
422 } else if (result == MXGEFW_CMD_UNKNOWN) {
423 return -ENOSYS;
424 } else if (result == MXGEFW_CMD_ERROR_UNALIGNED) {
425 return -E2BIG;
426 } else if (result == MXGEFW_CMD_ERROR_RANGE &&
427 cmd == MXGEFW_CMD_ENABLE_RSS_QUEUES &&
428 (data->
429 data1 & MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES) !=
430 0) {
431 return -ERANGE;
432 } else {
433 dev_err(&mgp->pdev->dev,
434 "command %d failed, result = %d\n",
435 cmd, result);
436 return -ENXIO;
437 }
438 }
439
440 dev_err(&mgp->pdev->dev, "command %d timed out, result = %d\n",
441 cmd, result);
442 return -EAGAIN;
443 }
444
445 /*
446 * The eeprom strings on the lanaiX have the format
447 * SN=x\0
448 * MAC=x:x:x:x:x:x\0
449 * PT:ddd mmm xx xx:xx:xx xx\0
450 * PV:ddd mmm xx xx:xx:xx xx\0
451 */
452 static int myri10ge_read_mac_addr(struct myri10ge_priv *mgp)
453 {
454 char *ptr, *limit;
455 int i;
456
457 ptr = mgp->eeprom_strings;
458 limit = mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE;
459
460 while (*ptr != '\0' && ptr < limit) {
461 if (memcmp(ptr, "MAC=", 4) == 0) {
462 ptr += 4;
463 mgp->mac_addr_string = ptr;
464 for (i = 0; i < 6; i++) {
465 if ((ptr + 2) > limit)
466 goto abort;
467 mgp->mac_addr[i] =
468 simple_strtoul(ptr, &ptr, 16);
469 ptr += 1;
470 }
471 }
472 if (memcmp(ptr, "PC=", 3) == 0) {
473 ptr += 3;
474 mgp->product_code_string = ptr;
475 }
476 if (memcmp((const void *)ptr, "SN=", 3) == 0) {
477 ptr += 3;
478 mgp->serial_number = simple_strtoul(ptr, &ptr, 10);
479 }
480 while (ptr < limit && *ptr++) ;
481 }
482
483 return 0;
484
485 abort:
486 dev_err(&mgp->pdev->dev, "failed to parse eeprom_strings\n");
487 return -ENXIO;
488 }
489
490 /*
491 * Enable or disable periodic RDMAs from the host to make certain
492 * chipsets resend dropped PCIe messages
493 */
494
495 static void myri10ge_dummy_rdma(struct myri10ge_priv *mgp, int enable)
496 {
497 char __iomem *submit;
498 __be32 buf[16] __attribute__ ((__aligned__(8)));
499 u32 dma_low, dma_high;
500 int i;
501
502 /* clear confirmation addr */
503 mgp->cmd->data = 0;
504 mb();
505
506 /* send a rdma command to the PCIe engine, and wait for the
507 * response in the confirmation address. The firmware should
508 * write a -1 there to indicate it is alive and well
509 */
510 dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
511 dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
512
513 buf[0] = htonl(dma_high); /* confirm addr MSW */
514 buf[1] = htonl(dma_low); /* confirm addr LSW */
515 buf[2] = MYRI10GE_NO_CONFIRM_DATA; /* confirm data */
516 buf[3] = htonl(dma_high); /* dummy addr MSW */
517 buf[4] = htonl(dma_low); /* dummy addr LSW */
518 buf[5] = htonl(enable); /* enable? */
519
520 submit = mgp->sram + MXGEFW_BOOT_DUMMY_RDMA;
521
522 myri10ge_pio_copy(submit, &buf, sizeof(buf));
523 for (i = 0; mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 20; i++)
524 msleep(1);
525 if (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA)
526 dev_err(&mgp->pdev->dev, "dummy rdma %s failed\n",
527 (enable ? "enable" : "disable"));
528 }
529
530 static int
531 myri10ge_validate_firmware(struct myri10ge_priv *mgp,
532 struct mcp_gen_header *hdr)
533 {
534 struct device *dev = &mgp->pdev->dev;
535
536 /* check firmware type */
537 if (ntohl(hdr->mcp_type) != MCP_TYPE_ETH) {
538 dev_err(dev, "Bad firmware type: 0x%x\n", ntohl(hdr->mcp_type));
539 return -EINVAL;
540 }
541
542 /* save firmware version for ethtool */
543 strncpy(mgp->fw_version, hdr->version, sizeof(mgp->fw_version));
544
545 sscanf(mgp->fw_version, "%d.%d.%d", &mgp->fw_ver_major,
546 &mgp->fw_ver_minor, &mgp->fw_ver_tiny);
547
548 if (!(mgp->fw_ver_major == MXGEFW_VERSION_MAJOR
549 && mgp->fw_ver_minor == MXGEFW_VERSION_MINOR)) {
550 dev_err(dev, "Found firmware version %s\n", mgp->fw_version);
551 dev_err(dev, "Driver needs %d.%d\n", MXGEFW_VERSION_MAJOR,
552 MXGEFW_VERSION_MINOR);
553 return -EINVAL;
554 }
555 return 0;
556 }
557
558 static int myri10ge_load_hotplug_firmware(struct myri10ge_priv *mgp, u32 * size)
559 {
560 unsigned crc, reread_crc;
561 const struct firmware *fw;
562 struct device *dev = &mgp->pdev->dev;
563 unsigned char *fw_readback;
564 struct mcp_gen_header *hdr;
565 size_t hdr_offset;
566 int status;
567 unsigned i;
568
569 if ((status = request_firmware(&fw, mgp->fw_name, dev)) < 0) {
570 dev_err(dev, "Unable to load %s firmware image via hotplug\n",
571 mgp->fw_name);
572 status = -EINVAL;
573 goto abort_with_nothing;
574 }
575
576 /* check size */
577
578 if (fw->size >= mgp->sram_size - MYRI10GE_FW_OFFSET ||
579 fw->size < MCP_HEADER_PTR_OFFSET + 4) {
580 dev_err(dev, "Firmware size invalid:%d\n", (int)fw->size);
581 status = -EINVAL;
582 goto abort_with_fw;
583 }
584
585 /* check id */
586 hdr_offset = ntohl(*(__be32 *) (fw->data + MCP_HEADER_PTR_OFFSET));
587 if ((hdr_offset & 3) || hdr_offset + sizeof(*hdr) > fw->size) {
588 dev_err(dev, "Bad firmware file\n");
589 status = -EINVAL;
590 goto abort_with_fw;
591 }
592 hdr = (void *)(fw->data + hdr_offset);
593
594 status = myri10ge_validate_firmware(mgp, hdr);
595 if (status != 0)
596 goto abort_with_fw;
597
598 crc = crc32(~0, fw->data, fw->size);
599 for (i = 0; i < fw->size; i += 256) {
600 myri10ge_pio_copy(mgp->sram + MYRI10GE_FW_OFFSET + i,
601 fw->data + i,
602 min(256U, (unsigned)(fw->size - i)));
603 mb();
604 readb(mgp->sram);
605 }
606 fw_readback = vmalloc(fw->size);
607 if (!fw_readback) {
608 status = -ENOMEM;
609 goto abort_with_fw;
610 }
611 /* corruption checking is good for parity recovery and buggy chipset */
612 memcpy_fromio(fw_readback, mgp->sram + MYRI10GE_FW_OFFSET, fw->size);
613 reread_crc = crc32(~0, fw_readback, fw->size);
614 vfree(fw_readback);
615 if (crc != reread_crc) {
616 dev_err(dev, "CRC failed(fw-len=%u), got 0x%x (expect 0x%x)\n",
617 (unsigned)fw->size, reread_crc, crc);
618 status = -EIO;
619 goto abort_with_fw;
620 }
621 *size = (u32) fw->size;
622
623 abort_with_fw:
624 release_firmware(fw);
625
626 abort_with_nothing:
627 return status;
628 }
629
630 static int myri10ge_adopt_running_firmware(struct myri10ge_priv *mgp)
631 {
632 struct mcp_gen_header *hdr;
633 struct device *dev = &mgp->pdev->dev;
634 const size_t bytes = sizeof(struct mcp_gen_header);
635 size_t hdr_offset;
636 int status;
637
638 /* find running firmware header */
639 hdr_offset = swab32(readl(mgp->sram + MCP_HEADER_PTR_OFFSET));
640
641 if ((hdr_offset & 3) || hdr_offset + sizeof(*hdr) > mgp->sram_size) {
642 dev_err(dev, "Running firmware has bad header offset (%d)\n",
643 (int)hdr_offset);
644 return -EIO;
645 }
646
647 /* copy header of running firmware from SRAM to host memory to
648 * validate firmware */
649 hdr = kmalloc(bytes, GFP_KERNEL);
650 if (hdr == NULL) {
651 dev_err(dev, "could not malloc firmware hdr\n");
652 return -ENOMEM;
653 }
654 memcpy_fromio(hdr, mgp->sram + hdr_offset, bytes);
655 status = myri10ge_validate_firmware(mgp, hdr);
656 kfree(hdr);
657
658 /* check to see if adopted firmware has bug where adopting
659 * it will cause broadcasts to be filtered unless the NIC
660 * is kept in ALLMULTI mode */
661 if (mgp->fw_ver_major == 1 && mgp->fw_ver_minor == 4 &&
662 mgp->fw_ver_tiny >= 4 && mgp->fw_ver_tiny <= 11) {
663 mgp->adopted_rx_filter_bug = 1;
664 dev_warn(dev, "Adopting fw %d.%d.%d: "
665 "working around rx filter bug\n",
666 mgp->fw_ver_major, mgp->fw_ver_minor,
667 mgp->fw_ver_tiny);
668 }
669 return status;
670 }
671
672 static int myri10ge_get_firmware_capabilities(struct myri10ge_priv *mgp)
673 {
674 struct myri10ge_cmd cmd;
675 int status;
676
677 /* probe for IPv6 TSO support */
678 mgp->features = NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_TSO;
679 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_MAX_TSO6_HDR_SIZE,
680 &cmd, 0);
681 if (status == 0) {
682 mgp->max_tso6 = cmd.data0;
683 mgp->features |= NETIF_F_TSO6;
684 }
685
686 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_RX_RING_SIZE, &cmd, 0);
687 if (status != 0) {
688 dev_err(&mgp->pdev->dev,
689 "failed MXGEFW_CMD_GET_RX_RING_SIZE\n");
690 return -ENXIO;
691 }
692
693 mgp->max_intr_slots = 2 * (cmd.data0 / sizeof(struct mcp_dma_addr));
694
695 return 0;
696 }
697
698 static int myri10ge_load_firmware(struct myri10ge_priv *mgp, int adopt)
699 {
700 char __iomem *submit;
701 __be32 buf[16] __attribute__ ((__aligned__(8)));
702 u32 dma_low, dma_high, size;
703 int status, i;
704
705 size = 0;
706 status = myri10ge_load_hotplug_firmware(mgp, &size);
707 if (status) {
708 if (!adopt)
709 return status;
710 dev_warn(&mgp->pdev->dev, "hotplug firmware loading failed\n");
711
712 /* Do not attempt to adopt firmware if there
713 * was a bad crc */
714 if (status == -EIO)
715 return status;
716
717 status = myri10ge_adopt_running_firmware(mgp);
718 if (status != 0) {
719 dev_err(&mgp->pdev->dev,
720 "failed to adopt running firmware\n");
721 return status;
722 }
723 dev_info(&mgp->pdev->dev,
724 "Successfully adopted running firmware\n");
725 if (mgp->tx_boundary == 4096) {
726 dev_warn(&mgp->pdev->dev,
727 "Using firmware currently running on NIC"
728 ". For optimal\n");
729 dev_warn(&mgp->pdev->dev,
730 "performance consider loading optimized "
731 "firmware\n");
732 dev_warn(&mgp->pdev->dev, "via hotplug\n");
733 }
734
735 mgp->fw_name = "adopted";
736 mgp->tx_boundary = 2048;
737 myri10ge_dummy_rdma(mgp, 1);
738 status = myri10ge_get_firmware_capabilities(mgp);
739 return status;
740 }
741
742 /* clear confirmation addr */
743 mgp->cmd->data = 0;
744 mb();
745
746 /* send a reload command to the bootstrap MCP, and wait for the
747 * response in the confirmation address. The firmware should
748 * write a -1 there to indicate it is alive and well
749 */
750 dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
751 dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
752
753 buf[0] = htonl(dma_high); /* confirm addr MSW */
754 buf[1] = htonl(dma_low); /* confirm addr LSW */
755 buf[2] = MYRI10GE_NO_CONFIRM_DATA; /* confirm data */
756
757 /* FIX: All newest firmware should un-protect the bottom of
758 * the sram before handoff. However, the very first interfaces
759 * do not. Therefore the handoff copy must skip the first 8 bytes
760 */
761 buf[3] = htonl(MYRI10GE_FW_OFFSET + 8); /* where the code starts */
762 buf[4] = htonl(size - 8); /* length of code */
763 buf[5] = htonl(8); /* where to copy to */
764 buf[6] = htonl(0); /* where to jump to */
765
766 submit = mgp->sram + MXGEFW_BOOT_HANDOFF;
767
768 myri10ge_pio_copy(submit, &buf, sizeof(buf));
769 mb();
770 msleep(1);
771 mb();
772 i = 0;
773 while (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 9) {
774 msleep(1 << i);
775 i++;
776 }
777 if (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA) {
778 dev_err(&mgp->pdev->dev, "handoff failed\n");
779 return -ENXIO;
780 }
781 myri10ge_dummy_rdma(mgp, 1);
782 status = myri10ge_get_firmware_capabilities(mgp);
783
784 return status;
785 }
786
787 static int myri10ge_update_mac_address(struct myri10ge_priv *mgp, u8 * addr)
788 {
789 struct myri10ge_cmd cmd;
790 int status;
791
792 cmd.data0 = ((addr[0] << 24) | (addr[1] << 16)
793 | (addr[2] << 8) | addr[3]);
794
795 cmd.data1 = ((addr[4] << 8) | (addr[5]));
796
797 status = myri10ge_send_cmd(mgp, MXGEFW_SET_MAC_ADDRESS, &cmd, 0);
798 return status;
799 }
800
801 static int myri10ge_change_pause(struct myri10ge_priv *mgp, int pause)
802 {
803 struct myri10ge_cmd cmd;
804 int status, ctl;
805
806 ctl = pause ? MXGEFW_ENABLE_FLOW_CONTROL : MXGEFW_DISABLE_FLOW_CONTROL;
807 status = myri10ge_send_cmd(mgp, ctl, &cmd, 0);
808
809 if (status) {
810 printk(KERN_ERR
811 "myri10ge: %s: Failed to set flow control mode\n",
812 mgp->dev->name);
813 return status;
814 }
815 mgp->pause = pause;
816 return 0;
817 }
818
819 static void
820 myri10ge_change_promisc(struct myri10ge_priv *mgp, int promisc, int atomic)
821 {
822 struct myri10ge_cmd cmd;
823 int status, ctl;
824
825 ctl = promisc ? MXGEFW_ENABLE_PROMISC : MXGEFW_DISABLE_PROMISC;
826 status = myri10ge_send_cmd(mgp, ctl, &cmd, atomic);
827 if (status)
828 printk(KERN_ERR "myri10ge: %s: Failed to set promisc mode\n",
829 mgp->dev->name);
830 }
831
832 static int myri10ge_dma_test(struct myri10ge_priv *mgp, int test_type)
833 {
834 struct myri10ge_cmd cmd;
835 int status;
836 u32 len;
837 struct page *dmatest_page;
838 dma_addr_t dmatest_bus;
839 char *test = " ";
840
841 dmatest_page = alloc_page(GFP_KERNEL);
842 if (!dmatest_page)
843 return -ENOMEM;
844 dmatest_bus = pci_map_page(mgp->pdev, dmatest_page, 0, PAGE_SIZE,
845 DMA_BIDIRECTIONAL);
846
847 /* Run a small DMA test.
848 * The magic multipliers to the length tell the firmware
849 * to do DMA read, write, or read+write tests. The
850 * results are returned in cmd.data0. The upper 16
851 * bits or the return is the number of transfers completed.
852 * The lower 16 bits is the time in 0.5us ticks that the
853 * transfers took to complete.
854 */
855
856 len = mgp->tx_boundary;
857
858 cmd.data0 = MYRI10GE_LOWPART_TO_U32(dmatest_bus);
859 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(dmatest_bus);
860 cmd.data2 = len * 0x10000;
861 status = myri10ge_send_cmd(mgp, test_type, &cmd, 0);
862 if (status != 0) {
863 test = "read";
864 goto abort;
865 }
866 mgp->read_dma = ((cmd.data0 >> 16) * len * 2) / (cmd.data0 & 0xffff);
867 cmd.data0 = MYRI10GE_LOWPART_TO_U32(dmatest_bus);
868 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(dmatest_bus);
869 cmd.data2 = len * 0x1;
870 status = myri10ge_send_cmd(mgp, test_type, &cmd, 0);
871 if (status != 0) {
872 test = "write";
873 goto abort;
874 }
875 mgp->write_dma = ((cmd.data0 >> 16) * len * 2) / (cmd.data0 & 0xffff);
876
877 cmd.data0 = MYRI10GE_LOWPART_TO_U32(dmatest_bus);
878 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(dmatest_bus);
879 cmd.data2 = len * 0x10001;
880 status = myri10ge_send_cmd(mgp, test_type, &cmd, 0);
881 if (status != 0) {
882 test = "read/write";
883 goto abort;
884 }
885 mgp->read_write_dma = ((cmd.data0 >> 16) * len * 2 * 2) /
886 (cmd.data0 & 0xffff);
887
888 abort:
889 pci_unmap_page(mgp->pdev, dmatest_bus, PAGE_SIZE, DMA_BIDIRECTIONAL);
890 put_page(dmatest_page);
891
892 if (status != 0 && test_type != MXGEFW_CMD_UNALIGNED_TEST)
893 dev_warn(&mgp->pdev->dev, "DMA %s benchmark failed: %d\n",
894 test, status);
895
896 return status;
897 }
898
899 static int myri10ge_reset(struct myri10ge_priv *mgp)
900 {
901 struct myri10ge_cmd cmd;
902 struct myri10ge_slice_state *ss;
903 int i, status;
904 size_t bytes;
905 #ifdef CONFIG_MYRI10GE_DCA
906 unsigned long dca_tag_off;
907 #endif
908
909 /* try to send a reset command to the card to see if it
910 * is alive */
911 memset(&cmd, 0, sizeof(cmd));
912 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_RESET, &cmd, 0);
913 if (status != 0) {
914 dev_err(&mgp->pdev->dev, "failed reset\n");
915 return -ENXIO;
916 }
917
918 (void)myri10ge_dma_test(mgp, MXGEFW_DMA_TEST);
919 /*
920 * Use non-ndis mcp_slot (eg, 4 bytes total,
921 * no toeplitz hash value returned. Older firmware will
922 * not understand this command, but will use the correct
923 * sized mcp_slot, so we ignore error returns
924 */
925 cmd.data0 = MXGEFW_RSS_MCP_SLOT_TYPE_MIN;
926 (void)myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_RSS_MCP_SLOT_TYPE, &cmd, 0);
927
928 /* Now exchange information about interrupts */
929
930 bytes = mgp->max_intr_slots * sizeof(*mgp->ss[0].rx_done.entry);
931 cmd.data0 = (u32) bytes;
932 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_SIZE, &cmd, 0);
933
934 /*
935 * Even though we already know how many slices are supported
936 * via myri10ge_probe_slices() MXGEFW_CMD_GET_MAX_RSS_QUEUES
937 * has magic side effects, and must be called after a reset.
938 * It must be called prior to calling any RSS related cmds,
939 * including assigning an interrupt queue for anything but
940 * slice 0. It must also be called *after*
941 * MXGEFW_CMD_SET_INTRQ_SIZE, since the intrq size is used by
942 * the firmware to compute offsets.
943 */
944
945 if (mgp->num_slices > 1) {
946
947 /* ask the maximum number of slices it supports */
948 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_MAX_RSS_QUEUES,
949 &cmd, 0);
950 if (status != 0) {
951 dev_err(&mgp->pdev->dev,
952 "failed to get number of slices\n");
953 }
954
955 /*
956 * MXGEFW_CMD_ENABLE_RSS_QUEUES must be called prior
957 * to setting up the interrupt queue DMA
958 */
959
960 cmd.data0 = mgp->num_slices;
961 cmd.data1 = MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE;
962 if (mgp->dev->real_num_tx_queues > 1)
963 cmd.data1 |= MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES;
964 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ENABLE_RSS_QUEUES,
965 &cmd, 0);
966
967 /* Firmware older than 1.4.32 only supports multiple
968 * RX queues, so if we get an error, first retry using a
969 * single TX queue before giving up */
970 if (status != 0 && mgp->dev->real_num_tx_queues > 1) {
971 mgp->dev->real_num_tx_queues = 1;
972 cmd.data0 = mgp->num_slices;
973 cmd.data1 = MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE;
974 status = myri10ge_send_cmd(mgp,
975 MXGEFW_CMD_ENABLE_RSS_QUEUES,
976 &cmd, 0);
977 }
978
979 if (status != 0) {
980 dev_err(&mgp->pdev->dev,
981 "failed to set number of slices\n");
982
983 return status;
984 }
985 }
986 for (i = 0; i < mgp->num_slices; i++) {
987 ss = &mgp->ss[i];
988 cmd.data0 = MYRI10GE_LOWPART_TO_U32(ss->rx_done.bus);
989 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(ss->rx_done.bus);
990 cmd.data2 = i;
991 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_DMA,
992 &cmd, 0);
993 };
994
995 status |=
996 myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_IRQ_ACK_OFFSET, &cmd, 0);
997 for (i = 0; i < mgp->num_slices; i++) {
998 ss = &mgp->ss[i];
999 ss->irq_claim =
1000 (__iomem __be32 *) (mgp->sram + cmd.data0 + 8 * i);
1001 }
1002 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_IRQ_DEASSERT_OFFSET,
1003 &cmd, 0);
1004 mgp->irq_deassert = (__iomem __be32 *) (mgp->sram + cmd.data0);
1005
1006 status |= myri10ge_send_cmd
1007 (mgp, MXGEFW_CMD_GET_INTR_COAL_DELAY_OFFSET, &cmd, 0);
1008 mgp->intr_coal_delay_ptr = (__iomem __be32 *) (mgp->sram + cmd.data0);
1009 if (status != 0) {
1010 dev_err(&mgp->pdev->dev, "failed set interrupt parameters\n");
1011 return status;
1012 }
1013 put_be32(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
1014
1015 #ifdef CONFIG_MYRI10GE_DCA
1016 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_DCA_OFFSET, &cmd, 0);
1017 dca_tag_off = cmd.data0;
1018 for (i = 0; i < mgp->num_slices; i++) {
1019 ss = &mgp->ss[i];
1020 if (status == 0) {
1021 ss->dca_tag = (__iomem __be32 *)
1022 (mgp->sram + dca_tag_off + 4 * i);
1023 } else {
1024 ss->dca_tag = NULL;
1025 }
1026 }
1027 #endif /* CONFIG_MYRI10GE_DCA */
1028
1029 /* reset mcp/driver shared state back to 0 */
1030
1031 mgp->link_changes = 0;
1032 for (i = 0; i < mgp->num_slices; i++) {
1033 ss = &mgp->ss[i];
1034
1035 memset(ss->rx_done.entry, 0, bytes);
1036 ss->tx.req = 0;
1037 ss->tx.done = 0;
1038 ss->tx.pkt_start = 0;
1039 ss->tx.pkt_done = 0;
1040 ss->rx_big.cnt = 0;
1041 ss->rx_small.cnt = 0;
1042 ss->rx_done.idx = 0;
1043 ss->rx_done.cnt = 0;
1044 ss->tx.wake_queue = 0;
1045 ss->tx.stop_queue = 0;
1046 }
1047
1048 status = myri10ge_update_mac_address(mgp, mgp->dev->dev_addr);
1049 myri10ge_change_pause(mgp, mgp->pause);
1050 myri10ge_set_multicast_list(mgp->dev);
1051 return status;
1052 }
1053
1054 #ifdef CONFIG_MYRI10GE_DCA
1055 static void
1056 myri10ge_write_dca(struct myri10ge_slice_state *ss, int cpu, int tag)
1057 {
1058 ss->cpu = cpu;
1059 ss->cached_dca_tag = tag;
1060 put_be32(htonl(tag), ss->dca_tag);
1061 }
1062
1063 static inline void myri10ge_update_dca(struct myri10ge_slice_state *ss)
1064 {
1065 int cpu = get_cpu();
1066 int tag;
1067
1068 if (cpu != ss->cpu) {
1069 tag = dca_get_tag(cpu);
1070 if (ss->cached_dca_tag != tag)
1071 myri10ge_write_dca(ss, cpu, tag);
1072 }
1073 put_cpu();
1074 }
1075
1076 static void myri10ge_setup_dca(struct myri10ge_priv *mgp)
1077 {
1078 int err, i;
1079 struct pci_dev *pdev = mgp->pdev;
1080
1081 if (mgp->ss[0].dca_tag == NULL || mgp->dca_enabled)
1082 return;
1083 if (!myri10ge_dca) {
1084 dev_err(&pdev->dev, "dca disabled by administrator\n");
1085 return;
1086 }
1087 err = dca_add_requester(&pdev->dev);
1088 if (err) {
1089 if (err != -ENODEV)
1090 dev_err(&pdev->dev,
1091 "dca_add_requester() failed, err=%d\n", err);
1092 return;
1093 }
1094 mgp->dca_enabled = 1;
1095 for (i = 0; i < mgp->num_slices; i++)
1096 myri10ge_write_dca(&mgp->ss[i], -1, 0);
1097 }
1098
1099 static void myri10ge_teardown_dca(struct myri10ge_priv *mgp)
1100 {
1101 struct pci_dev *pdev = mgp->pdev;
1102 int err;
1103
1104 if (!mgp->dca_enabled)
1105 return;
1106 mgp->dca_enabled = 0;
1107 err = dca_remove_requester(&pdev->dev);
1108 }
1109
1110 static int myri10ge_notify_dca_device(struct device *dev, void *data)
1111 {
1112 struct myri10ge_priv *mgp;
1113 unsigned long event;
1114
1115 mgp = dev_get_drvdata(dev);
1116 event = *(unsigned long *)data;
1117
1118 if (event == DCA_PROVIDER_ADD)
1119 myri10ge_setup_dca(mgp);
1120 else if (event == DCA_PROVIDER_REMOVE)
1121 myri10ge_teardown_dca(mgp);
1122 return 0;
1123 }
1124 #endif /* CONFIG_MYRI10GE_DCA */
1125
1126 static inline void
1127 myri10ge_submit_8rx(struct mcp_kreq_ether_recv __iomem * dst,
1128 struct mcp_kreq_ether_recv *src)
1129 {
1130 __be32 low;
1131
1132 low = src->addr_low;
1133 src->addr_low = htonl(DMA_32BIT_MASK);
1134 myri10ge_pio_copy(dst, src, 4 * sizeof(*src));
1135 mb();
1136 myri10ge_pio_copy(dst + 4, src + 4, 4 * sizeof(*src));
1137 mb();
1138 src->addr_low = low;
1139 put_be32(low, &dst->addr_low);
1140 mb();
1141 }
1142
1143 static inline void myri10ge_vlan_ip_csum(struct sk_buff *skb, __wsum hw_csum)
1144 {
1145 struct vlan_hdr *vh = (struct vlan_hdr *)(skb->data);
1146
1147 if ((skb->protocol == htons(ETH_P_8021Q)) &&
1148 (vh->h_vlan_encapsulated_proto == htons(ETH_P_IP) ||
1149 vh->h_vlan_encapsulated_proto == htons(ETH_P_IPV6))) {
1150 skb->csum = hw_csum;
1151 skb->ip_summed = CHECKSUM_COMPLETE;
1152 }
1153 }
1154
1155 static inline void
1156 myri10ge_rx_skb_build(struct sk_buff *skb, u8 * va,
1157 struct skb_frag_struct *rx_frags, int len, int hlen)
1158 {
1159 struct skb_frag_struct *skb_frags;
1160
1161 skb->len = skb->data_len = len;
1162 skb->truesize = len + sizeof(struct sk_buff);
1163 /* attach the page(s) */
1164
1165 skb_frags = skb_shinfo(skb)->frags;
1166 while (len > 0) {
1167 memcpy(skb_frags, rx_frags, sizeof(*skb_frags));
1168 len -= rx_frags->size;
1169 skb_frags++;
1170 rx_frags++;
1171 skb_shinfo(skb)->nr_frags++;
1172 }
1173
1174 /* pskb_may_pull is not available in irq context, but
1175 * skb_pull() (for ether_pad and eth_type_trans()) requires
1176 * the beginning of the packet in skb_headlen(), move it
1177 * manually */
1178 skb_copy_to_linear_data(skb, va, hlen);
1179 skb_shinfo(skb)->frags[0].page_offset += hlen;
1180 skb_shinfo(skb)->frags[0].size -= hlen;
1181 skb->data_len -= hlen;
1182 skb->tail += hlen;
1183 skb_pull(skb, MXGEFW_PAD);
1184 }
1185
1186 static void
1187 myri10ge_alloc_rx_pages(struct myri10ge_priv *mgp, struct myri10ge_rx_buf *rx,
1188 int bytes, int watchdog)
1189 {
1190 struct page *page;
1191 int idx;
1192
1193 if (unlikely(rx->watchdog_needed && !watchdog))
1194 return;
1195
1196 /* try to refill entire ring */
1197 while (rx->fill_cnt != (rx->cnt + rx->mask + 1)) {
1198 idx = rx->fill_cnt & rx->mask;
1199 if (rx->page_offset + bytes <= MYRI10GE_ALLOC_SIZE) {
1200 /* we can use part of previous page */
1201 get_page(rx->page);
1202 } else {
1203 /* we need a new page */
1204 page =
1205 alloc_pages(GFP_ATOMIC | __GFP_COMP,
1206 MYRI10GE_ALLOC_ORDER);
1207 if (unlikely(page == NULL)) {
1208 if (rx->fill_cnt - rx->cnt < 16)
1209 rx->watchdog_needed = 1;
1210 return;
1211 }
1212 rx->page = page;
1213 rx->page_offset = 0;
1214 rx->bus = pci_map_page(mgp->pdev, page, 0,
1215 MYRI10GE_ALLOC_SIZE,
1216 PCI_DMA_FROMDEVICE);
1217 }
1218 rx->info[idx].page = rx->page;
1219 rx->info[idx].page_offset = rx->page_offset;
1220 /* note that this is the address of the start of the
1221 * page */
1222 pci_unmap_addr_set(&rx->info[idx], bus, rx->bus);
1223 rx->shadow[idx].addr_low =
1224 htonl(MYRI10GE_LOWPART_TO_U32(rx->bus) + rx->page_offset);
1225 rx->shadow[idx].addr_high =
1226 htonl(MYRI10GE_HIGHPART_TO_U32(rx->bus));
1227
1228 /* start next packet on a cacheline boundary */
1229 rx->page_offset += SKB_DATA_ALIGN(bytes);
1230
1231 #if MYRI10GE_ALLOC_SIZE > 4096
1232 /* don't cross a 4KB boundary */
1233 if ((rx->page_offset >> 12) !=
1234 ((rx->page_offset + bytes - 1) >> 12))
1235 rx->page_offset = (rx->page_offset + 4096) & ~4095;
1236 #endif
1237 rx->fill_cnt++;
1238
1239 /* copy 8 descriptors to the firmware at a time */
1240 if ((idx & 7) == 7) {
1241 myri10ge_submit_8rx(&rx->lanai[idx - 7],
1242 &rx->shadow[idx - 7]);
1243 }
1244 }
1245 }
1246
1247 static inline void
1248 myri10ge_unmap_rx_page(struct pci_dev *pdev,
1249 struct myri10ge_rx_buffer_state *info, int bytes)
1250 {
1251 /* unmap the recvd page if we're the only or last user of it */
1252 if (bytes >= MYRI10GE_ALLOC_SIZE / 2 ||
1253 (info->page_offset + 2 * bytes) > MYRI10GE_ALLOC_SIZE) {
1254 pci_unmap_page(pdev, (pci_unmap_addr(info, bus)
1255 & ~(MYRI10GE_ALLOC_SIZE - 1)),
1256 MYRI10GE_ALLOC_SIZE, PCI_DMA_FROMDEVICE);
1257 }
1258 }
1259
1260 #define MYRI10GE_HLEN 64 /* The number of bytes to copy from a
1261 * page into an skb */
1262
1263 static inline int
1264 myri10ge_rx_done(struct myri10ge_slice_state *ss, struct myri10ge_rx_buf *rx,
1265 int bytes, int len, __wsum csum)
1266 {
1267 struct myri10ge_priv *mgp = ss->mgp;
1268 struct sk_buff *skb;
1269 struct skb_frag_struct rx_frags[MYRI10GE_MAX_FRAGS_PER_FRAME];
1270 int i, idx, hlen, remainder;
1271 struct pci_dev *pdev = mgp->pdev;
1272 struct net_device *dev = mgp->dev;
1273 u8 *va;
1274
1275 len += MXGEFW_PAD;
1276 idx = rx->cnt & rx->mask;
1277 va = page_address(rx->info[idx].page) + rx->info[idx].page_offset;
1278 prefetch(va);
1279 /* Fill skb_frag_struct(s) with data from our receive */
1280 for (i = 0, remainder = len; remainder > 0; i++) {
1281 myri10ge_unmap_rx_page(pdev, &rx->info[idx], bytes);
1282 rx_frags[i].page = rx->info[idx].page;
1283 rx_frags[i].page_offset = rx->info[idx].page_offset;
1284 if (remainder < MYRI10GE_ALLOC_SIZE)
1285 rx_frags[i].size = remainder;
1286 else
1287 rx_frags[i].size = MYRI10GE_ALLOC_SIZE;
1288 rx->cnt++;
1289 idx = rx->cnt & rx->mask;
1290 remainder -= MYRI10GE_ALLOC_SIZE;
1291 }
1292
1293 if (mgp->csum_flag && myri10ge_lro) {
1294 rx_frags[0].page_offset += MXGEFW_PAD;
1295 rx_frags[0].size -= MXGEFW_PAD;
1296 len -= MXGEFW_PAD;
1297 lro_receive_frags(&ss->rx_done.lro_mgr, rx_frags,
1298 /* opaque, will come back in get_frag_header */
1299 len, len,
1300 (void *)(__force unsigned long)csum, csum);
1301
1302 return 1;
1303 }
1304
1305 hlen = MYRI10GE_HLEN > len ? len : MYRI10GE_HLEN;
1306
1307 /* allocate an skb to attach the page(s) to. This is done
1308 * after trying LRO, so as to avoid skb allocation overheads */
1309
1310 skb = netdev_alloc_skb(dev, MYRI10GE_HLEN + 16);
1311 if (unlikely(skb == NULL)) {
1312 ss->stats.rx_dropped++;
1313 do {
1314 i--;
1315 put_page(rx_frags[i].page);
1316 } while (i != 0);
1317 return 0;
1318 }
1319
1320 /* Attach the pages to the skb, and trim off any padding */
1321 myri10ge_rx_skb_build(skb, va, rx_frags, len, hlen);
1322 if (skb_shinfo(skb)->frags[0].size <= 0) {
1323 put_page(skb_shinfo(skb)->frags[0].page);
1324 skb_shinfo(skb)->nr_frags = 0;
1325 }
1326 skb->protocol = eth_type_trans(skb, dev);
1327
1328 if (mgp->csum_flag) {
1329 if ((skb->protocol == htons(ETH_P_IP)) ||
1330 (skb->protocol == htons(ETH_P_IPV6))) {
1331 skb->csum = csum;
1332 skb->ip_summed = CHECKSUM_COMPLETE;
1333 } else
1334 myri10ge_vlan_ip_csum(skb, csum);
1335 }
1336 netif_receive_skb(skb);
1337 return 1;
1338 }
1339
1340 static inline void
1341 myri10ge_tx_done(struct myri10ge_slice_state *ss, int mcp_index)
1342 {
1343 struct pci_dev *pdev = ss->mgp->pdev;
1344 struct myri10ge_tx_buf *tx = &ss->tx;
1345 struct netdev_queue *dev_queue;
1346 struct sk_buff *skb;
1347 int idx, len;
1348
1349 while (tx->pkt_done != mcp_index) {
1350 idx = tx->done & tx->mask;
1351 skb = tx->info[idx].skb;
1352
1353 /* Mark as free */
1354 tx->info[idx].skb = NULL;
1355 if (tx->info[idx].last) {
1356 tx->pkt_done++;
1357 tx->info[idx].last = 0;
1358 }
1359 tx->done++;
1360 len = pci_unmap_len(&tx->info[idx], len);
1361 pci_unmap_len_set(&tx->info[idx], len, 0);
1362 if (skb) {
1363 ss->stats.tx_bytes += skb->len;
1364 ss->stats.tx_packets++;
1365 dev_kfree_skb_irq(skb);
1366 if (len)
1367 pci_unmap_single(pdev,
1368 pci_unmap_addr(&tx->info[idx],
1369 bus), len,
1370 PCI_DMA_TODEVICE);
1371 } else {
1372 if (len)
1373 pci_unmap_page(pdev,
1374 pci_unmap_addr(&tx->info[idx],
1375 bus), len,
1376 PCI_DMA_TODEVICE);
1377 }
1378 }
1379
1380 dev_queue = netdev_get_tx_queue(ss->dev, ss - ss->mgp->ss);
1381 /*
1382 * Make a minimal effort to prevent the NIC from polling an
1383 * idle tx queue. If we can't get the lock we leave the queue
1384 * active. In this case, either a thread was about to start
1385 * using the queue anyway, or we lost a race and the NIC will
1386 * waste some of its resources polling an inactive queue for a
1387 * while.
1388 */
1389
1390 if ((ss->mgp->dev->real_num_tx_queues > 1) &&
1391 __netif_tx_trylock(dev_queue)) {
1392 if (tx->req == tx->done) {
1393 tx->queue_active = 0;
1394 put_be32(htonl(1), tx->send_stop);
1395 mb();
1396 mmiowb();
1397 }
1398 __netif_tx_unlock(dev_queue);
1399 }
1400
1401 /* start the queue if we've stopped it */
1402 if (netif_tx_queue_stopped(dev_queue)
1403 && tx->req - tx->done < (tx->mask >> 1)) {
1404 tx->wake_queue++;
1405 netif_tx_wake_queue(dev_queue);
1406 }
1407 }
1408
1409 static inline int
1410 myri10ge_clean_rx_done(struct myri10ge_slice_state *ss, int budget)
1411 {
1412 struct myri10ge_rx_done *rx_done = &ss->rx_done;
1413 struct myri10ge_priv *mgp = ss->mgp;
1414 unsigned long rx_bytes = 0;
1415 unsigned long rx_packets = 0;
1416 unsigned long rx_ok;
1417
1418 int idx = rx_done->idx;
1419 int cnt = rx_done->cnt;
1420 int work_done = 0;
1421 u16 length;
1422 __wsum checksum;
1423
1424 while (rx_done->entry[idx].length != 0 && work_done < budget) {
1425 length = ntohs(rx_done->entry[idx].length);
1426 rx_done->entry[idx].length = 0;
1427 checksum = csum_unfold(rx_done->entry[idx].checksum);
1428 if (length <= mgp->small_bytes)
1429 rx_ok = myri10ge_rx_done(ss, &ss->rx_small,
1430 mgp->small_bytes,
1431 length, checksum);
1432 else
1433 rx_ok = myri10ge_rx_done(ss, &ss->rx_big,
1434 mgp->big_bytes,
1435 length, checksum);
1436 rx_packets += rx_ok;
1437 rx_bytes += rx_ok * (unsigned long)length;
1438 cnt++;
1439 idx = cnt & (mgp->max_intr_slots - 1);
1440 work_done++;
1441 }
1442 rx_done->idx = idx;
1443 rx_done->cnt = cnt;
1444 ss->stats.rx_packets += rx_packets;
1445 ss->stats.rx_bytes += rx_bytes;
1446
1447 if (myri10ge_lro)
1448 lro_flush_all(&rx_done->lro_mgr);
1449
1450 /* restock receive rings if needed */
1451 if (ss->rx_small.fill_cnt - ss->rx_small.cnt < myri10ge_fill_thresh)
1452 myri10ge_alloc_rx_pages(mgp, &ss->rx_small,
1453 mgp->small_bytes + MXGEFW_PAD, 0);
1454 if (ss->rx_big.fill_cnt - ss->rx_big.cnt < myri10ge_fill_thresh)
1455 myri10ge_alloc_rx_pages(mgp, &ss->rx_big, mgp->big_bytes, 0);
1456
1457 return work_done;
1458 }
1459
1460 static inline void myri10ge_check_statblock(struct myri10ge_priv *mgp)
1461 {
1462 struct mcp_irq_data *stats = mgp->ss[0].fw_stats;
1463
1464 if (unlikely(stats->stats_updated)) {
1465 unsigned link_up = ntohl(stats->link_up);
1466 if (mgp->link_state != link_up) {
1467 mgp->link_state = link_up;
1468
1469 if (mgp->link_state == MXGEFW_LINK_UP) {
1470 if (netif_msg_link(mgp))
1471 printk(KERN_INFO
1472 "myri10ge: %s: link up\n",
1473 mgp->dev->name);
1474 netif_carrier_on(mgp->dev);
1475 mgp->link_changes++;
1476 } else {
1477 if (netif_msg_link(mgp))
1478 printk(KERN_INFO
1479 "myri10ge: %s: link %s\n",
1480 mgp->dev->name,
1481 (link_up == MXGEFW_LINK_MYRINET ?
1482 "mismatch (Myrinet detected)" :
1483 "down"));
1484 netif_carrier_off(mgp->dev);
1485 mgp->link_changes++;
1486 }
1487 }
1488 if (mgp->rdma_tags_available !=
1489 ntohl(stats->rdma_tags_available)) {
1490 mgp->rdma_tags_available =
1491 ntohl(stats->rdma_tags_available);
1492 printk(KERN_WARNING "myri10ge: %s: RDMA timed out! "
1493 "%d tags left\n", mgp->dev->name,
1494 mgp->rdma_tags_available);
1495 }
1496 mgp->down_cnt += stats->link_down;
1497 if (stats->link_down)
1498 wake_up(&mgp->down_wq);
1499 }
1500 }
1501
1502 static int myri10ge_poll(struct napi_struct *napi, int budget)
1503 {
1504 struct myri10ge_slice_state *ss =
1505 container_of(napi, struct myri10ge_slice_state, napi);
1506 int work_done;
1507
1508 #ifdef CONFIG_MYRI10GE_DCA
1509 if (ss->mgp->dca_enabled)
1510 myri10ge_update_dca(ss);
1511 #endif
1512
1513 /* process as many rx events as NAPI will allow */
1514 work_done = myri10ge_clean_rx_done(ss, budget);
1515
1516 if (work_done < budget) {
1517 netif_rx_complete(napi);
1518 put_be32(htonl(3), ss->irq_claim);
1519 }
1520 return work_done;
1521 }
1522
1523 static irqreturn_t myri10ge_intr(int irq, void *arg)
1524 {
1525 struct myri10ge_slice_state *ss = arg;
1526 struct myri10ge_priv *mgp = ss->mgp;
1527 struct mcp_irq_data *stats = ss->fw_stats;
1528 struct myri10ge_tx_buf *tx = &ss->tx;
1529 u32 send_done_count;
1530 int i;
1531
1532 /* an interrupt on a non-zero receive-only slice is implicitly
1533 * valid since MSI-X irqs are not shared */
1534 if ((mgp->dev->real_num_tx_queues == 1) && (ss != mgp->ss)) {
1535 netif_rx_schedule(&ss->napi);
1536 return (IRQ_HANDLED);
1537 }
1538
1539 /* make sure it is our IRQ, and that the DMA has finished */
1540 if (unlikely(!stats->valid))
1541 return (IRQ_NONE);
1542
1543 /* low bit indicates receives are present, so schedule
1544 * napi poll handler */
1545 if (stats->valid & 1)
1546 netif_rx_schedule(&ss->napi);
1547
1548 if (!mgp->msi_enabled && !mgp->msix_enabled) {
1549 put_be32(0, mgp->irq_deassert);
1550 if (!myri10ge_deassert_wait)
1551 stats->valid = 0;
1552 mb();
1553 } else
1554 stats->valid = 0;
1555
1556 /* Wait for IRQ line to go low, if using INTx */
1557 i = 0;
1558 while (1) {
1559 i++;
1560 /* check for transmit completes and receives */
1561 send_done_count = ntohl(stats->send_done_count);
1562 if (send_done_count != tx->pkt_done)
1563 myri10ge_tx_done(ss, (int)send_done_count);
1564 if (unlikely(i > myri10ge_max_irq_loops)) {
1565 printk(KERN_WARNING "myri10ge: %s: irq stuck?\n",
1566 mgp->dev->name);
1567 stats->valid = 0;
1568 schedule_work(&mgp->watchdog_work);
1569 }
1570 if (likely(stats->valid == 0))
1571 break;
1572 cpu_relax();
1573 barrier();
1574 }
1575
1576 /* Only slice 0 updates stats */
1577 if (ss == mgp->ss)
1578 myri10ge_check_statblock(mgp);
1579
1580 put_be32(htonl(3), ss->irq_claim + 1);
1581 return (IRQ_HANDLED);
1582 }
1583
1584 static int
1585 myri10ge_get_settings(struct net_device *netdev, struct ethtool_cmd *cmd)
1586 {
1587 struct myri10ge_priv *mgp = netdev_priv(netdev);
1588 char *ptr;
1589 int i;
1590
1591 cmd->autoneg = AUTONEG_DISABLE;
1592 cmd->speed = SPEED_10000;
1593 cmd->duplex = DUPLEX_FULL;
1594
1595 /*
1596 * parse the product code to deterimine the interface type
1597 * (CX4, XFP, Quad Ribbon Fiber) by looking at the character
1598 * after the 3rd dash in the driver's cached copy of the
1599 * EEPROM's product code string.
1600 */
1601 ptr = mgp->product_code_string;
1602 if (ptr == NULL) {
1603 printk(KERN_ERR "myri10ge: %s: Missing product code\n",
1604 netdev->name);
1605 return 0;
1606 }
1607 for (i = 0; i < 3; i++, ptr++) {
1608 ptr = strchr(ptr, '-');
1609 if (ptr == NULL) {
1610 printk(KERN_ERR "myri10ge: %s: Invalid product "
1611 "code %s\n", netdev->name,
1612 mgp->product_code_string);
1613 return 0;
1614 }
1615 }
1616 if (*ptr == 'R' || *ptr == 'Q') {
1617 /* We've found either an XFP or quad ribbon fiber */
1618 cmd->port = PORT_FIBRE;
1619 }
1620 return 0;
1621 }
1622
1623 static void
1624 myri10ge_get_drvinfo(struct net_device *netdev, struct ethtool_drvinfo *info)
1625 {
1626 struct myri10ge_priv *mgp = netdev_priv(netdev);
1627
1628 strlcpy(info->driver, "myri10ge", sizeof(info->driver));
1629 strlcpy(info->version, MYRI10GE_VERSION_STR, sizeof(info->version));
1630 strlcpy(info->fw_version, mgp->fw_version, sizeof(info->fw_version));
1631 strlcpy(info->bus_info, pci_name(mgp->pdev), sizeof(info->bus_info));
1632 }
1633
1634 static int
1635 myri10ge_get_coalesce(struct net_device *netdev, struct ethtool_coalesce *coal)
1636 {
1637 struct myri10ge_priv *mgp = netdev_priv(netdev);
1638
1639 coal->rx_coalesce_usecs = mgp->intr_coal_delay;
1640 return 0;
1641 }
1642
1643 static int
1644 myri10ge_set_coalesce(struct net_device *netdev, struct ethtool_coalesce *coal)
1645 {
1646 struct myri10ge_priv *mgp = netdev_priv(netdev);
1647
1648 mgp->intr_coal_delay = coal->rx_coalesce_usecs;
1649 put_be32(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
1650 return 0;
1651 }
1652
1653 static void
1654 myri10ge_get_pauseparam(struct net_device *netdev,
1655 struct ethtool_pauseparam *pause)
1656 {
1657 struct myri10ge_priv *mgp = netdev_priv(netdev);
1658
1659 pause->autoneg = 0;
1660 pause->rx_pause = mgp->pause;
1661 pause->tx_pause = mgp->pause;
1662 }
1663
1664 static int
1665 myri10ge_set_pauseparam(struct net_device *netdev,
1666 struct ethtool_pauseparam *pause)
1667 {
1668 struct myri10ge_priv *mgp = netdev_priv(netdev);
1669
1670 if (pause->tx_pause != mgp->pause)
1671 return myri10ge_change_pause(mgp, pause->tx_pause);
1672 if (pause->rx_pause != mgp->pause)
1673 return myri10ge_change_pause(mgp, pause->tx_pause);
1674 if (pause->autoneg != 0)
1675 return -EINVAL;
1676 return 0;
1677 }
1678
1679 static void
1680 myri10ge_get_ringparam(struct net_device *netdev,
1681 struct ethtool_ringparam *ring)
1682 {
1683 struct myri10ge_priv *mgp = netdev_priv(netdev);
1684
1685 ring->rx_mini_max_pending = mgp->ss[0].rx_small.mask + 1;
1686 ring->rx_max_pending = mgp->ss[0].rx_big.mask + 1;
1687 ring->rx_jumbo_max_pending = 0;
1688 ring->tx_max_pending = mgp->ss[0].rx_small.mask + 1;
1689 ring->rx_mini_pending = ring->rx_mini_max_pending;
1690 ring->rx_pending = ring->rx_max_pending;
1691 ring->rx_jumbo_pending = ring->rx_jumbo_max_pending;
1692 ring->tx_pending = ring->tx_max_pending;
1693 }
1694
1695 static u32 myri10ge_get_rx_csum(struct net_device *netdev)
1696 {
1697 struct myri10ge_priv *mgp = netdev_priv(netdev);
1698
1699 if (mgp->csum_flag)
1700 return 1;
1701 else
1702 return 0;
1703 }
1704
1705 static int myri10ge_set_rx_csum(struct net_device *netdev, u32 csum_enabled)
1706 {
1707 struct myri10ge_priv *mgp = netdev_priv(netdev);
1708
1709 if (csum_enabled)
1710 mgp->csum_flag = MXGEFW_FLAGS_CKSUM;
1711 else
1712 mgp->csum_flag = 0;
1713 return 0;
1714 }
1715
1716 static int myri10ge_set_tso(struct net_device *netdev, u32 tso_enabled)
1717 {
1718 struct myri10ge_priv *mgp = netdev_priv(netdev);
1719 unsigned long flags = mgp->features & (NETIF_F_TSO6 | NETIF_F_TSO);
1720
1721 if (tso_enabled)
1722 netdev->features |= flags;
1723 else
1724 netdev->features &= ~flags;
1725 return 0;
1726 }
1727
1728 static const char myri10ge_gstrings_main_stats[][ETH_GSTRING_LEN] = {
1729 "rx_packets", "tx_packets", "rx_bytes", "tx_bytes", "rx_errors",
1730 "tx_errors", "rx_dropped", "tx_dropped", "multicast", "collisions",
1731 "rx_length_errors", "rx_over_errors", "rx_crc_errors",
1732 "rx_frame_errors", "rx_fifo_errors", "rx_missed_errors",
1733 "tx_aborted_errors", "tx_carrier_errors", "tx_fifo_errors",
1734 "tx_heartbeat_errors", "tx_window_errors",
1735 /* device-specific stats */
1736 "tx_boundary", "WC", "irq", "MSI", "MSIX",
1737 "read_dma_bw_MBs", "write_dma_bw_MBs", "read_write_dma_bw_MBs",
1738 "serial_number", "watchdog_resets",
1739 #ifdef CONFIG_MYRI10GE_DCA
1740 "dca_capable_firmware", "dca_device_present",
1741 #endif
1742 "link_changes", "link_up", "dropped_link_overflow",
1743 "dropped_link_error_or_filtered",
1744 "dropped_pause", "dropped_bad_phy", "dropped_bad_crc32",
1745 "dropped_unicast_filtered", "dropped_multicast_filtered",
1746 "dropped_runt", "dropped_overrun", "dropped_no_small_buffer",
1747 "dropped_no_big_buffer"
1748 };
1749
1750 static const char myri10ge_gstrings_slice_stats[][ETH_GSTRING_LEN] = {
1751 "----------- slice ---------",
1752 "tx_pkt_start", "tx_pkt_done", "tx_req", "tx_done",
1753 "rx_small_cnt", "rx_big_cnt",
1754 "wake_queue", "stop_queue", "tx_linearized", "LRO aggregated",
1755 "LRO flushed",
1756 "LRO avg aggr", "LRO no_desc"
1757 };
1758
1759 #define MYRI10GE_NET_STATS_LEN 21
1760 #define MYRI10GE_MAIN_STATS_LEN ARRAY_SIZE(myri10ge_gstrings_main_stats)
1761 #define MYRI10GE_SLICE_STATS_LEN ARRAY_SIZE(myri10ge_gstrings_slice_stats)
1762
1763 static void
1764 myri10ge_get_strings(struct net_device *netdev, u32 stringset, u8 * data)
1765 {
1766 struct myri10ge_priv *mgp = netdev_priv(netdev);
1767 int i;
1768
1769 switch (stringset) {
1770 case ETH_SS_STATS:
1771 memcpy(data, *myri10ge_gstrings_main_stats,
1772 sizeof(myri10ge_gstrings_main_stats));
1773 data += sizeof(myri10ge_gstrings_main_stats);
1774 for (i = 0; i < mgp->num_slices; i++) {
1775 memcpy(data, *myri10ge_gstrings_slice_stats,
1776 sizeof(myri10ge_gstrings_slice_stats));
1777 data += sizeof(myri10ge_gstrings_slice_stats);
1778 }
1779 break;
1780 }
1781 }
1782
1783 static int myri10ge_get_sset_count(struct net_device *netdev, int sset)
1784 {
1785 struct myri10ge_priv *mgp = netdev_priv(netdev);
1786
1787 switch (sset) {
1788 case ETH_SS_STATS:
1789 return MYRI10GE_MAIN_STATS_LEN +
1790 mgp->num_slices * MYRI10GE_SLICE_STATS_LEN;
1791 default:
1792 return -EOPNOTSUPP;
1793 }
1794 }
1795
1796 static void
1797 myri10ge_get_ethtool_stats(struct net_device *netdev,
1798 struct ethtool_stats *stats, u64 * data)
1799 {
1800 struct myri10ge_priv *mgp = netdev_priv(netdev);
1801 struct myri10ge_slice_state *ss;
1802 int slice;
1803 int i;
1804
1805 for (i = 0; i < MYRI10GE_NET_STATS_LEN; i++)
1806 data[i] = ((unsigned long *)&mgp->stats)[i];
1807
1808 data[i++] = (unsigned int)mgp->tx_boundary;
1809 data[i++] = (unsigned int)mgp->wc_enabled;
1810 data[i++] = (unsigned int)mgp->pdev->irq;
1811 data[i++] = (unsigned int)mgp->msi_enabled;
1812 data[i++] = (unsigned int)mgp->msix_enabled;
1813 data[i++] = (unsigned int)mgp->read_dma;
1814 data[i++] = (unsigned int)mgp->write_dma;
1815 data[i++] = (unsigned int)mgp->read_write_dma;
1816 data[i++] = (unsigned int)mgp->serial_number;
1817 data[i++] = (unsigned int)mgp->watchdog_resets;
1818 #ifdef CONFIG_MYRI10GE_DCA
1819 data[i++] = (unsigned int)(mgp->ss[0].dca_tag != NULL);
1820 data[i++] = (unsigned int)(mgp->dca_enabled);
1821 #endif
1822 data[i++] = (unsigned int)mgp->link_changes;
1823
1824 /* firmware stats are useful only in the first slice */
1825 ss = &mgp->ss[0];
1826 data[i++] = (unsigned int)ntohl(ss->fw_stats->link_up);
1827 data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_link_overflow);
1828 data[i++] =
1829 (unsigned int)ntohl(ss->fw_stats->dropped_link_error_or_filtered);
1830 data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_pause);
1831 data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_bad_phy);
1832 data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_bad_crc32);
1833 data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_unicast_filtered);
1834 data[i++] =
1835 (unsigned int)ntohl(ss->fw_stats->dropped_multicast_filtered);
1836 data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_runt);
1837 data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_overrun);
1838 data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_no_small_buffer);
1839 data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_no_big_buffer);
1840
1841 for (slice = 0; slice < mgp->num_slices; slice++) {
1842 ss = &mgp->ss[slice];
1843 data[i++] = slice;
1844 data[i++] = (unsigned int)ss->tx.pkt_start;
1845 data[i++] = (unsigned int)ss->tx.pkt_done;
1846 data[i++] = (unsigned int)ss->tx.req;
1847 data[i++] = (unsigned int)ss->tx.done;
1848 data[i++] = (unsigned int)ss->rx_small.cnt;
1849 data[i++] = (unsigned int)ss->rx_big.cnt;
1850 data[i++] = (unsigned int)ss->tx.wake_queue;
1851 data[i++] = (unsigned int)ss->tx.stop_queue;
1852 data[i++] = (unsigned int)ss->tx.linearized;
1853 data[i++] = ss->rx_done.lro_mgr.stats.aggregated;
1854 data[i++] = ss->rx_done.lro_mgr.stats.flushed;
1855 if (ss->rx_done.lro_mgr.stats.flushed)
1856 data[i++] = ss->rx_done.lro_mgr.stats.aggregated /
1857 ss->rx_done.lro_mgr.stats.flushed;
1858 else
1859 data[i++] = 0;
1860 data[i++] = ss->rx_done.lro_mgr.stats.no_desc;
1861 }
1862 }
1863
1864 static void myri10ge_set_msglevel(struct net_device *netdev, u32 value)
1865 {
1866 struct myri10ge_priv *mgp = netdev_priv(netdev);
1867 mgp->msg_enable = value;
1868 }
1869
1870 static u32 myri10ge_get_msglevel(struct net_device *netdev)
1871 {
1872 struct myri10ge_priv *mgp = netdev_priv(netdev);
1873 return mgp->msg_enable;
1874 }
1875
1876 static const struct ethtool_ops myri10ge_ethtool_ops = {
1877 .get_settings = myri10ge_get_settings,
1878 .get_drvinfo = myri10ge_get_drvinfo,
1879 .get_coalesce = myri10ge_get_coalesce,
1880 .set_coalesce = myri10ge_set_coalesce,
1881 .get_pauseparam = myri10ge_get_pauseparam,
1882 .set_pauseparam = myri10ge_set_pauseparam,
1883 .get_ringparam = myri10ge_get_ringparam,
1884 .get_rx_csum = myri10ge_get_rx_csum,
1885 .set_rx_csum = myri10ge_set_rx_csum,
1886 .set_tx_csum = ethtool_op_set_tx_hw_csum,
1887 .set_sg = ethtool_op_set_sg,
1888 .set_tso = myri10ge_set_tso,
1889 .get_link = ethtool_op_get_link,
1890 .get_strings = myri10ge_get_strings,
1891 .get_sset_count = myri10ge_get_sset_count,
1892 .get_ethtool_stats = myri10ge_get_ethtool_stats,
1893 .set_msglevel = myri10ge_set_msglevel,
1894 .get_msglevel = myri10ge_get_msglevel
1895 };
1896
1897 static int myri10ge_allocate_rings(struct myri10ge_slice_state *ss)
1898 {
1899 struct myri10ge_priv *mgp = ss->mgp;
1900 struct myri10ge_cmd cmd;
1901 struct net_device *dev = mgp->dev;
1902 int tx_ring_size, rx_ring_size;
1903 int tx_ring_entries, rx_ring_entries;
1904 int i, slice, status;
1905 size_t bytes;
1906
1907 /* get ring sizes */
1908 slice = ss - mgp->ss;
1909 cmd.data0 = slice;
1910 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_RING_SIZE, &cmd, 0);
1911 tx_ring_size = cmd.data0;
1912 cmd.data0 = slice;
1913 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_RX_RING_SIZE, &cmd, 0);
1914 if (status != 0)
1915 return status;
1916 rx_ring_size = cmd.data0;
1917
1918 tx_ring_entries = tx_ring_size / sizeof(struct mcp_kreq_ether_send);
1919 rx_ring_entries = rx_ring_size / sizeof(struct mcp_dma_addr);
1920 ss->tx.mask = tx_ring_entries - 1;
1921 ss->rx_small.mask = ss->rx_big.mask = rx_ring_entries - 1;
1922
1923 status = -ENOMEM;
1924
1925 /* allocate the host shadow rings */
1926
1927 bytes = 8 + (MYRI10GE_MAX_SEND_DESC_TSO + 4)
1928 * sizeof(*ss->tx.req_list);
1929 ss->tx.req_bytes = kzalloc(bytes, GFP_KERNEL);
1930 if (ss->tx.req_bytes == NULL)
1931 goto abort_with_nothing;
1932
1933 /* ensure req_list entries are aligned to 8 bytes */
1934 ss->tx.req_list = (struct mcp_kreq_ether_send *)
1935 ALIGN((unsigned long)ss->tx.req_bytes, 8);
1936 ss->tx.queue_active = 0;
1937
1938 bytes = rx_ring_entries * sizeof(*ss->rx_small.shadow);
1939 ss->rx_small.shadow = kzalloc(bytes, GFP_KERNEL);
1940 if (ss->rx_small.shadow == NULL)
1941 goto abort_with_tx_req_bytes;
1942
1943 bytes = rx_ring_entries * sizeof(*ss->rx_big.shadow);
1944 ss->rx_big.shadow = kzalloc(bytes, GFP_KERNEL);
1945 if (ss->rx_big.shadow == NULL)
1946 goto abort_with_rx_small_shadow;
1947
1948 /* allocate the host info rings */
1949
1950 bytes = tx_ring_entries * sizeof(*ss->tx.info);
1951 ss->tx.info = kzalloc(bytes, GFP_KERNEL);
1952 if (ss->tx.info == NULL)
1953 goto abort_with_rx_big_shadow;
1954
1955 bytes = rx_ring_entries * sizeof(*ss->rx_small.info);
1956 ss->rx_small.info = kzalloc(bytes, GFP_KERNEL);
1957 if (ss->rx_small.info == NULL)
1958 goto abort_with_tx_info;
1959
1960 bytes = rx_ring_entries * sizeof(*ss->rx_big.info);
1961 ss->rx_big.info = kzalloc(bytes, GFP_KERNEL);
1962 if (ss->rx_big.info == NULL)
1963 goto abort_with_rx_small_info;
1964
1965 /* Fill the receive rings */
1966 ss->rx_big.cnt = 0;
1967 ss->rx_small.cnt = 0;
1968 ss->rx_big.fill_cnt = 0;
1969 ss->rx_small.fill_cnt = 0;
1970 ss->rx_small.page_offset = MYRI10GE_ALLOC_SIZE;
1971 ss->rx_big.page_offset = MYRI10GE_ALLOC_SIZE;
1972 ss->rx_small.watchdog_needed = 0;
1973 ss->rx_big.watchdog_needed = 0;
1974 myri10ge_alloc_rx_pages(mgp, &ss->rx_small,
1975 mgp->small_bytes + MXGEFW_PAD, 0);
1976
1977 if (ss->rx_small.fill_cnt < ss->rx_small.mask + 1) {
1978 printk(KERN_ERR
1979 "myri10ge: %s:slice-%d: alloced only %d small bufs\n",
1980 dev->name, slice, ss->rx_small.fill_cnt);
1981 goto abort_with_rx_small_ring;
1982 }
1983
1984 myri10ge_alloc_rx_pages(mgp, &ss->rx_big, mgp->big_bytes, 0);
1985 if (ss->rx_big.fill_cnt < ss->rx_big.mask + 1) {
1986 printk(KERN_ERR
1987 "myri10ge: %s:slice-%d: alloced only %d big bufs\n",
1988 dev->name, slice, ss->rx_big.fill_cnt);
1989 goto abort_with_rx_big_ring;
1990 }
1991
1992 return 0;
1993
1994 abort_with_rx_big_ring:
1995 for (i = ss->rx_big.cnt; i < ss->rx_big.fill_cnt; i++) {
1996 int idx = i & ss->rx_big.mask;
1997 myri10ge_unmap_rx_page(mgp->pdev, &ss->rx_big.info[idx],
1998 mgp->big_bytes);
1999 put_page(ss->rx_big.info[idx].page);
2000 }
2001
2002 abort_with_rx_small_ring:
2003 for (i = ss->rx_small.cnt; i < ss->rx_small.fill_cnt; i++) {
2004 int idx = i & ss->rx_small.mask;
2005 myri10ge_unmap_rx_page(mgp->pdev, &ss->rx_small.info[idx],
2006 mgp->small_bytes + MXGEFW_PAD);
2007 put_page(ss->rx_small.info[idx].page);
2008 }
2009
2010 kfree(ss->rx_big.info);
2011
2012 abort_with_rx_small_info:
2013 kfree(ss->rx_small.info);
2014
2015 abort_with_tx_info:
2016 kfree(ss->tx.info);
2017
2018 abort_with_rx_big_shadow:
2019 kfree(ss->rx_big.shadow);
2020
2021 abort_with_rx_small_shadow:
2022 kfree(ss->rx_small.shadow);
2023
2024 abort_with_tx_req_bytes:
2025 kfree(ss->tx.req_bytes);
2026 ss->tx.req_bytes = NULL;
2027 ss->tx.req_list = NULL;
2028
2029 abort_with_nothing:
2030 return status;
2031 }
2032
2033 static void myri10ge_free_rings(struct myri10ge_slice_state *ss)
2034 {
2035 struct myri10ge_priv *mgp = ss->mgp;
2036 struct sk_buff *skb;
2037 struct myri10ge_tx_buf *tx;
2038 int i, len, idx;
2039
2040 /* If not allocated, skip it */
2041 if (ss->tx.req_list == NULL)
2042 return;
2043
2044 for (i = ss->rx_big.cnt; i < ss->rx_big.fill_cnt; i++) {
2045 idx = i & ss->rx_big.mask;
2046 if (i == ss->rx_big.fill_cnt - 1)
2047 ss->rx_big.info[idx].page_offset = MYRI10GE_ALLOC_SIZE;
2048 myri10ge_unmap_rx_page(mgp->pdev, &ss->rx_big.info[idx],
2049 mgp->big_bytes);
2050 put_page(ss->rx_big.info[idx].page);
2051 }
2052
2053 for (i = ss->rx_small.cnt; i < ss->rx_small.fill_cnt; i++) {
2054 idx = i & ss->rx_small.mask;
2055 if (i == ss->rx_small.fill_cnt - 1)
2056 ss->rx_small.info[idx].page_offset =
2057 MYRI10GE_ALLOC_SIZE;
2058 myri10ge_unmap_rx_page(mgp->pdev, &ss->rx_small.info[idx],
2059 mgp->small_bytes + MXGEFW_PAD);
2060 put_page(ss->rx_small.info[idx].page);
2061 }
2062 tx = &ss->tx;
2063 while (tx->done != tx->req) {
2064 idx = tx->done & tx->mask;
2065 skb = tx->info[idx].skb;
2066
2067 /* Mark as free */
2068 tx->info[idx].skb = NULL;
2069 tx->done++;
2070 len = pci_unmap_len(&tx->info[idx], len);
2071 pci_unmap_len_set(&tx->info[idx], len, 0);
2072 if (skb) {
2073 ss->stats.tx_dropped++;
2074 dev_kfree_skb_any(skb);
2075 if (len)
2076 pci_unmap_single(mgp->pdev,
2077 pci_unmap_addr(&tx->info[idx],
2078 bus), len,
2079 PCI_DMA_TODEVICE);
2080 } else {
2081 if (len)
2082 pci_unmap_page(mgp->pdev,
2083 pci_unmap_addr(&tx->info[idx],
2084 bus), len,
2085 PCI_DMA_TODEVICE);
2086 }
2087 }
2088 kfree(ss->rx_big.info);
2089
2090 kfree(ss->rx_small.info);
2091
2092 kfree(ss->tx.info);
2093
2094 kfree(ss->rx_big.shadow);
2095
2096 kfree(ss->rx_small.shadow);
2097
2098 kfree(ss->tx.req_bytes);
2099 ss->tx.req_bytes = NULL;
2100 ss->tx.req_list = NULL;
2101 }
2102
2103 static int myri10ge_request_irq(struct myri10ge_priv *mgp)
2104 {
2105 struct pci_dev *pdev = mgp->pdev;
2106 struct myri10ge_slice_state *ss;
2107 struct net_device *netdev = mgp->dev;
2108 int i;
2109 int status;
2110
2111 mgp->msi_enabled = 0;
2112 mgp->msix_enabled = 0;
2113 status = 0;
2114 if (myri10ge_msi) {
2115 if (mgp->num_slices > 1) {
2116 status =
2117 pci_enable_msix(pdev, mgp->msix_vectors,
2118 mgp->num_slices);
2119 if (status == 0) {
2120 mgp->msix_enabled = 1;
2121 } else {
2122 dev_err(&pdev->dev,
2123 "Error %d setting up MSI-X\n", status);
2124 return status;
2125 }
2126 }
2127 if (mgp->msix_enabled == 0) {
2128 status = pci_enable_msi(pdev);
2129 if (status != 0) {
2130 dev_err(&pdev->dev,
2131 "Error %d setting up MSI; falling back to xPIC\n",
2132 status);
2133 } else {
2134 mgp->msi_enabled = 1;
2135 }
2136 }
2137 }
2138 if (mgp->msix_enabled) {
2139 for (i = 0; i < mgp->num_slices; i++) {
2140 ss = &mgp->ss[i];
2141 snprintf(ss->irq_desc, sizeof(ss->irq_desc),
2142 "%s:slice-%d", netdev->name, i);
2143 status = request_irq(mgp->msix_vectors[i].vector,
2144 myri10ge_intr, 0, ss->irq_desc,
2145 ss);
2146 if (status != 0) {
2147 dev_err(&pdev->dev,
2148 "slice %d failed to allocate IRQ\n", i);
2149 i--;
2150 while (i >= 0) {
2151 free_irq(mgp->msix_vectors[i].vector,
2152 &mgp->ss[i]);
2153 i--;
2154 }
2155 pci_disable_msix(pdev);
2156 return status;
2157 }
2158 }
2159 } else {
2160 status = request_irq(pdev->irq, myri10ge_intr, IRQF_SHARED,
2161 mgp->dev->name, &mgp->ss[0]);
2162 if (status != 0) {
2163 dev_err(&pdev->dev, "failed to allocate IRQ\n");
2164 if (mgp->msi_enabled)
2165 pci_disable_msi(pdev);
2166 }
2167 }
2168 return status;
2169 }
2170
2171 static void myri10ge_free_irq(struct myri10ge_priv *mgp)
2172 {
2173 struct pci_dev *pdev = mgp->pdev;
2174 int i;
2175
2176 if (mgp->msix_enabled) {
2177 for (i = 0; i < mgp->num_slices; i++)
2178 free_irq(mgp->msix_vectors[i].vector, &mgp->ss[i]);
2179 } else {
2180 free_irq(pdev->irq, &mgp->ss[0]);
2181 }
2182 if (mgp->msi_enabled)
2183 pci_disable_msi(pdev);
2184 if (mgp->msix_enabled)
2185 pci_disable_msix(pdev);
2186 }
2187
2188 static int
2189 myri10ge_get_frag_header(struct skb_frag_struct *frag, void **mac_hdr,
2190 void **ip_hdr, void **tcpudp_hdr,
2191 u64 * hdr_flags, void *priv)
2192 {
2193 struct ethhdr *eh;
2194 struct vlan_ethhdr *veh;
2195 struct iphdr *iph;
2196 u8 *va = page_address(frag->page) + frag->page_offset;
2197 unsigned long ll_hlen;
2198 /* passed opaque through lro_receive_frags() */
2199 __wsum csum = (__force __wsum) (unsigned long)priv;
2200
2201 /* find the mac header, aborting if not IPv4 */
2202
2203 eh = (struct ethhdr *)va;
2204 *mac_hdr = eh;
2205 ll_hlen = ETH_HLEN;
2206 if (eh->h_proto != htons(ETH_P_IP)) {
2207 if (eh->h_proto == htons(ETH_P_8021Q)) {
2208 veh = (struct vlan_ethhdr *)va;
2209 if (veh->h_vlan_encapsulated_proto != htons(ETH_P_IP))
2210 return -1;
2211
2212 ll_hlen += VLAN_HLEN;
2213
2214 /*
2215 * HW checksum starts ETH_HLEN bytes into
2216 * frame, so we must subtract off the VLAN
2217 * header's checksum before csum can be used
2218 */
2219 csum = csum_sub(csum, csum_partial(va + ETH_HLEN,
2220 VLAN_HLEN, 0));
2221 } else {
2222 return -1;
2223 }
2224 }
2225 *hdr_flags = LRO_IPV4;
2226
2227 iph = (struct iphdr *)(va + ll_hlen);
2228 *ip_hdr = iph;
2229 if (iph->protocol != IPPROTO_TCP)
2230 return -1;
2231 if (iph->frag_off & htons(IP_MF | IP_OFFSET))
2232 return -1;
2233 *hdr_flags |= LRO_TCP;
2234 *tcpudp_hdr = (u8 *) (*ip_hdr) + (iph->ihl << 2);
2235
2236 /* verify the IP checksum */
2237 if (unlikely(ip_fast_csum((u8 *) iph, iph->ihl)))
2238 return -1;
2239
2240 /* verify the checksum */
2241 if (unlikely(csum_tcpudp_magic(iph->saddr, iph->daddr,
2242 ntohs(iph->tot_len) - (iph->ihl << 2),
2243 IPPROTO_TCP, csum)))
2244 return -1;
2245
2246 return 0;
2247 }
2248
2249 static int myri10ge_get_txrx(struct myri10ge_priv *mgp, int slice)
2250 {
2251 struct myri10ge_cmd cmd;
2252 struct myri10ge_slice_state *ss;
2253 int status;
2254
2255 ss = &mgp->ss[slice];
2256 status = 0;
2257 if (slice == 0 || (mgp->dev->real_num_tx_queues > 1)) {
2258 cmd.data0 = slice;
2259 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_OFFSET,
2260 &cmd, 0);
2261 ss->tx.lanai = (struct mcp_kreq_ether_send __iomem *)
2262 (mgp->sram + cmd.data0);
2263 }
2264 cmd.data0 = slice;
2265 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SMALL_RX_OFFSET,
2266 &cmd, 0);
2267 ss->rx_small.lanai = (struct mcp_kreq_ether_recv __iomem *)
2268 (mgp->sram + cmd.data0);
2269
2270 cmd.data0 = slice;
2271 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_BIG_RX_OFFSET, &cmd, 0);
2272 ss->rx_big.lanai = (struct mcp_kreq_ether_recv __iomem *)
2273 (mgp->sram + cmd.data0);
2274
2275 ss->tx.send_go = (__iomem __be32 *)
2276 (mgp->sram + MXGEFW_ETH_SEND_GO + 64 * slice);
2277 ss->tx.send_stop = (__iomem __be32 *)
2278 (mgp->sram + MXGEFW_ETH_SEND_STOP + 64 * slice);
2279 return status;
2280
2281 }
2282
2283 static int myri10ge_set_stats(struct myri10ge_priv *mgp, int slice)
2284 {
2285 struct myri10ge_cmd cmd;
2286 struct myri10ge_slice_state *ss;
2287 int status;
2288
2289 ss = &mgp->ss[slice];
2290 cmd.data0 = MYRI10GE_LOWPART_TO_U32(ss->fw_stats_bus);
2291 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(ss->fw_stats_bus);
2292 cmd.data2 = sizeof(struct mcp_irq_data) | (slice << 16);
2293 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_STATS_DMA_V2, &cmd, 0);
2294 if (status == -ENOSYS) {
2295 dma_addr_t bus = ss->fw_stats_bus;
2296 if (slice != 0)
2297 return -EINVAL;
2298 bus += offsetof(struct mcp_irq_data, send_done_count);
2299 cmd.data0 = MYRI10GE_LOWPART_TO_U32(bus);
2300 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(bus);
2301 status = myri10ge_send_cmd(mgp,
2302 MXGEFW_CMD_SET_STATS_DMA_OBSOLETE,
2303 &cmd, 0);
2304 /* Firmware cannot support multicast without STATS_DMA_V2 */
2305 mgp->fw_multicast_support = 0;
2306 } else {
2307 mgp->fw_multicast_support = 1;
2308 }
2309 return 0;
2310 }
2311
2312 static int myri10ge_open(struct net_device *dev)
2313 {
2314 struct myri10ge_slice_state *ss;
2315 struct myri10ge_priv *mgp = netdev_priv(dev);
2316 struct myri10ge_cmd cmd;
2317 int i, status, big_pow2, slice;
2318 u8 *itable;
2319 struct net_lro_mgr *lro_mgr;
2320
2321 if (mgp->running != MYRI10GE_ETH_STOPPED)
2322 return -EBUSY;
2323
2324 mgp->running = MYRI10GE_ETH_STARTING;
2325 status = myri10ge_reset(mgp);
2326 if (status != 0) {
2327 printk(KERN_ERR "myri10ge: %s: failed reset\n", dev->name);
2328 goto abort_with_nothing;
2329 }
2330
2331 if (mgp->num_slices > 1) {
2332 cmd.data0 = mgp->num_slices;
2333 cmd.data1 = MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE;
2334 if (mgp->dev->real_num_tx_queues > 1)
2335 cmd.data1 |= MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES;
2336 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ENABLE_RSS_QUEUES,
2337 &cmd, 0);
2338 if (status != 0) {
2339 printk(KERN_ERR
2340 "myri10ge: %s: failed to set number of slices\n",
2341 dev->name);
2342 goto abort_with_nothing;
2343 }
2344 /* setup the indirection table */
2345 cmd.data0 = mgp->num_slices;
2346 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_RSS_TABLE_SIZE,
2347 &cmd, 0);
2348
2349 status |= myri10ge_send_cmd(mgp,
2350 MXGEFW_CMD_GET_RSS_TABLE_OFFSET,
2351 &cmd, 0);
2352 if (status != 0) {
2353 printk(KERN_ERR
2354 "myri10ge: %s: failed to setup rss tables\n",
2355 dev->name);
2356 goto abort_with_nothing;
2357 }
2358
2359 /* just enable an identity mapping */
2360 itable = mgp->sram + cmd.data0;
2361 for (i = 0; i < mgp->num_slices; i++)
2362 __raw_writeb(i, &itable[i]);
2363
2364 cmd.data0 = 1;
2365 cmd.data1 = myri10ge_rss_hash;
2366 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_RSS_ENABLE,
2367 &cmd, 0);
2368 if (status != 0) {
2369 printk(KERN_ERR
2370 "myri10ge: %s: failed to enable slices\n",
2371 dev->name);
2372 goto abort_with_nothing;
2373 }
2374 }
2375
2376 status = myri10ge_request_irq(mgp);
2377 if (status != 0)
2378 goto abort_with_nothing;
2379
2380 /* decide what small buffer size to use. For good TCP rx
2381 * performance, it is important to not receive 1514 byte
2382 * frames into jumbo buffers, as it confuses the socket buffer
2383 * accounting code, leading to drops and erratic performance.
2384 */
2385
2386 if (dev->mtu <= ETH_DATA_LEN)
2387 /* enough for a TCP header */
2388 mgp->small_bytes = (128 > SMP_CACHE_BYTES)
2389 ? (128 - MXGEFW_PAD)
2390 : (SMP_CACHE_BYTES - MXGEFW_PAD);
2391 else
2392 /* enough for a vlan encapsulated ETH_DATA_LEN frame */
2393 mgp->small_bytes = VLAN_ETH_FRAME_LEN;
2394
2395 /* Override the small buffer size? */
2396 if (myri10ge_small_bytes > 0)
2397 mgp->small_bytes = myri10ge_small_bytes;
2398
2399 /* Firmware needs the big buff size as a power of 2. Lie and
2400 * tell him the buffer is larger, because we only use 1
2401 * buffer/pkt, and the mtu will prevent overruns.
2402 */
2403 big_pow2 = dev->mtu + ETH_HLEN + VLAN_HLEN + MXGEFW_PAD;
2404 if (big_pow2 < MYRI10GE_ALLOC_SIZE / 2) {
2405 while (!is_power_of_2(big_pow2))
2406 big_pow2++;
2407 mgp->big_bytes = dev->mtu + ETH_HLEN + VLAN_HLEN + MXGEFW_PAD;
2408 } else {
2409 big_pow2 = MYRI10GE_ALLOC_SIZE;
2410 mgp->big_bytes = big_pow2;
2411 }
2412
2413 /* setup the per-slice data structures */
2414 for (slice = 0; slice < mgp->num_slices; slice++) {
2415 ss = &mgp->ss[slice];
2416
2417 status = myri10ge_get_txrx(mgp, slice);
2418 if (status != 0) {
2419 printk(KERN_ERR
2420 "myri10ge: %s: failed to get ring sizes or locations\n",
2421 dev->name);
2422 goto abort_with_rings;
2423 }
2424 status = myri10ge_allocate_rings(ss);
2425 if (status != 0)
2426 goto abort_with_rings;
2427
2428 /* only firmware which supports multiple TX queues
2429 * supports setting up the tx stats on non-zero
2430 * slices */
2431 if (slice == 0 || mgp->dev->real_num_tx_queues > 1)
2432 status = myri10ge_set_stats(mgp, slice);
2433 if (status) {
2434 printk(KERN_ERR
2435 "myri10ge: %s: Couldn't set stats DMA\n",
2436 dev->name);
2437 goto abort_with_rings;
2438 }
2439
2440 lro_mgr = &ss->rx_done.lro_mgr;
2441 lro_mgr->dev = dev;
2442 lro_mgr->features = LRO_F_NAPI;
2443 lro_mgr->ip_summed = CHECKSUM_COMPLETE;
2444 lro_mgr->ip_summed_aggr = CHECKSUM_UNNECESSARY;
2445 lro_mgr->max_desc = MYRI10GE_MAX_LRO_DESCRIPTORS;
2446 lro_mgr->lro_arr = ss->rx_done.lro_desc;
2447 lro_mgr->get_frag_header = myri10ge_get_frag_header;
2448 lro_mgr->max_aggr = myri10ge_lro_max_pkts;
2449 lro_mgr->frag_align_pad = 2;
2450 if (lro_mgr->max_aggr > MAX_SKB_FRAGS)
2451 lro_mgr->max_aggr = MAX_SKB_FRAGS;
2452
2453 /* must happen prior to any irq */
2454 napi_enable(&(ss)->napi);
2455 }
2456
2457 /* now give firmware buffers sizes, and MTU */
2458 cmd.data0 = dev->mtu + ETH_HLEN + VLAN_HLEN;
2459 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_MTU, &cmd, 0);
2460 cmd.data0 = mgp->small_bytes;
2461 status |=
2462 myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_SMALL_BUFFER_SIZE, &cmd, 0);
2463 cmd.data0 = big_pow2;
2464 status |=
2465 myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_BIG_BUFFER_SIZE, &cmd, 0);
2466 if (status) {
2467 printk(KERN_ERR "myri10ge: %s: Couldn't set buffer sizes\n",
2468 dev->name);
2469 goto abort_with_rings;
2470 }
2471
2472 /*
2473 * Set Linux style TSO mode; this is needed only on newer
2474 * firmware versions. Older versions default to Linux
2475 * style TSO
2476 */
2477 cmd.data0 = 0;
2478 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_TSO_MODE, &cmd, 0);
2479 if (status && status != -ENOSYS) {
2480 printk(KERN_ERR "myri10ge: %s: Couldn't set TSO mode\n",
2481 dev->name);
2482 goto abort_with_rings;
2483 }
2484
2485 mgp->link_state = ~0U;
2486 mgp->rdma_tags_available = 15;
2487
2488 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_UP, &cmd, 0);
2489 if (status) {
2490 printk(KERN_ERR "myri10ge: %s: Couldn't bring up link\n",
2491 dev->name);
2492 goto abort_with_rings;
2493 }
2494
2495 mgp->running = MYRI10GE_ETH_RUNNING;
2496 mgp->watchdog_timer.expires = jiffies + myri10ge_watchdog_timeout * HZ;
2497 add_timer(&mgp->watchdog_timer);
2498 netif_tx_wake_all_queues(dev);
2499
2500 return 0;
2501
2502 abort_with_rings:
2503 while (slice) {
2504 slice--;
2505 napi_disable(&mgp->ss[slice].napi);
2506 }
2507 for (i = 0; i < mgp->num_slices; i++)
2508 myri10ge_free_rings(&mgp->ss[i]);
2509
2510 myri10ge_free_irq(mgp);
2511
2512 abort_with_nothing:
2513 mgp->running = MYRI10GE_ETH_STOPPED;
2514 return -ENOMEM;
2515 }
2516
2517 static int myri10ge_close(struct net_device *dev)
2518 {
2519 struct myri10ge_priv *mgp = netdev_priv(dev);
2520 struct myri10ge_cmd cmd;
2521 int status, old_down_cnt;
2522 int i;
2523
2524 if (mgp->running != MYRI10GE_ETH_RUNNING)
2525 return 0;
2526
2527 if (mgp->ss[0].tx.req_bytes == NULL)
2528 return 0;
2529
2530 del_timer_sync(&mgp->watchdog_timer);
2531 mgp->running = MYRI10GE_ETH_STOPPING;
2532 for (i = 0; i < mgp->num_slices; i++) {
2533 napi_disable(&mgp->ss[i].napi);
2534 }
2535 netif_carrier_off(dev);
2536
2537 netif_tx_stop_all_queues(dev);
2538 old_down_cnt = mgp->down_cnt;
2539 mb();
2540 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_DOWN, &cmd, 0);
2541 if (status)
2542 printk(KERN_ERR "myri10ge: %s: Couldn't bring down link\n",
2543 dev->name);
2544
2545 wait_event_timeout(mgp->down_wq, old_down_cnt != mgp->down_cnt, HZ);
2546 if (old_down_cnt == mgp->down_cnt)
2547 printk(KERN_ERR "myri10ge: %s never got down irq\n", dev->name);
2548
2549 netif_tx_disable(dev);
2550 myri10ge_free_irq(mgp);
2551 for (i = 0; i < mgp->num_slices; i++)
2552 myri10ge_free_rings(&mgp->ss[i]);
2553
2554 mgp->running = MYRI10GE_ETH_STOPPED;
2555 return 0;
2556 }
2557
2558 /* copy an array of struct mcp_kreq_ether_send's to the mcp. Copy
2559 * backwards one at a time and handle ring wraps */
2560
2561 static inline void
2562 myri10ge_submit_req_backwards(struct myri10ge_tx_buf *tx,
2563 struct mcp_kreq_ether_send *src, int cnt)
2564 {
2565 int idx, starting_slot;
2566 starting_slot = tx->req;
2567 while (cnt > 1) {
2568 cnt--;
2569 idx = (starting_slot + cnt) & tx->mask;
2570 myri10ge_pio_copy(&tx->lanai[idx], &src[cnt], sizeof(*src));
2571 mb();
2572 }
2573 }
2574
2575 /*
2576 * copy an array of struct mcp_kreq_ether_send's to the mcp. Copy
2577 * at most 32 bytes at a time, so as to avoid involving the software
2578 * pio handler in the nic. We re-write the first segment's flags
2579 * to mark them valid only after writing the entire chain.
2580 */
2581
2582 static inline void
2583 myri10ge_submit_req(struct myri10ge_tx_buf *tx, struct mcp_kreq_ether_send *src,
2584 int cnt)
2585 {
2586 int idx, i;
2587 struct mcp_kreq_ether_send __iomem *dstp, *dst;
2588 struct mcp_kreq_ether_send *srcp;
2589 u8 last_flags;
2590
2591 idx = tx->req & tx->mask;
2592
2593 last_flags = src->flags;
2594 src->flags = 0;
2595 mb();
2596 dst = dstp = &tx->lanai[idx];
2597 srcp = src;
2598
2599 if ((idx + cnt) < tx->mask) {
2600 for (i = 0; i < (cnt - 1); i += 2) {
2601 myri10ge_pio_copy(dstp, srcp, 2 * sizeof(*src));
2602 mb(); /* force write every 32 bytes */
2603 srcp += 2;
2604 dstp += 2;
2605 }
2606 } else {
2607 /* submit all but the first request, and ensure
2608 * that it is submitted below */
2609 myri10ge_submit_req_backwards(tx, src, cnt);
2610 i = 0;
2611 }
2612 if (i < cnt) {
2613 /* submit the first request */
2614 myri10ge_pio_copy(dstp, srcp, sizeof(*src));
2615 mb(); /* barrier before setting valid flag */
2616 }
2617
2618 /* re-write the last 32-bits with the valid flags */
2619 src->flags = last_flags;
2620 put_be32(*((__be32 *) src + 3), (__be32 __iomem *) dst + 3);
2621 tx->req += cnt;
2622 mb();
2623 }
2624
2625 /*
2626 * Transmit a packet. We need to split the packet so that a single
2627 * segment does not cross myri10ge->tx_boundary, so this makes segment
2628 * counting tricky. So rather than try to count segments up front, we
2629 * just give up if there are too few segments to hold a reasonably
2630 * fragmented packet currently available. If we run
2631 * out of segments while preparing a packet for DMA, we just linearize
2632 * it and try again.
2633 */
2634
2635 static int myri10ge_xmit(struct sk_buff *skb, struct net_device *dev)
2636 {
2637 struct myri10ge_priv *mgp = netdev_priv(dev);
2638 struct myri10ge_slice_state *ss;
2639 struct mcp_kreq_ether_send *req;
2640 struct myri10ge_tx_buf *tx;
2641 struct skb_frag_struct *frag;
2642 struct netdev_queue *netdev_queue;
2643 dma_addr_t bus;
2644 u32 low;
2645 __be32 high_swapped;
2646 unsigned int len;
2647 int idx, last_idx, avail, frag_cnt, frag_idx, count, mss, max_segments;
2648 u16 pseudo_hdr_offset, cksum_offset, queue;
2649 int cum_len, seglen, boundary, rdma_count;
2650 u8 flags, odd_flag;
2651
2652 queue = skb_get_queue_mapping(skb);
2653 ss = &mgp->ss[queue];
2654 netdev_queue = netdev_get_tx_queue(mgp->dev, queue);
2655 tx = &ss->tx;
2656
2657 again:
2658 req = tx->req_list;
2659 avail = tx->mask - 1 - (tx->req - tx->done);
2660
2661 mss = 0;
2662 max_segments = MXGEFW_MAX_SEND_DESC;
2663
2664 if (skb_is_gso(skb)) {
2665 mss = skb_shinfo(skb)->gso_size;
2666 max_segments = MYRI10GE_MAX_SEND_DESC_TSO;
2667 }
2668
2669 if ((unlikely(avail < max_segments))) {
2670 /* we are out of transmit resources */
2671 tx->stop_queue++;
2672 netif_tx_stop_queue(netdev_queue);
2673 return 1;
2674 }
2675
2676 /* Setup checksum offloading, if needed */
2677 cksum_offset = 0;
2678 pseudo_hdr_offset = 0;
2679 odd_flag = 0;
2680 flags = (MXGEFW_FLAGS_NO_TSO | MXGEFW_FLAGS_FIRST);
2681 if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
2682 cksum_offset = skb_transport_offset(skb);
2683 pseudo_hdr_offset = cksum_offset + skb->csum_offset;
2684 /* If the headers are excessively large, then we must
2685 * fall back to a software checksum */
2686 if (unlikely(!mss && (cksum_offset > 255 ||
2687 pseudo_hdr_offset > 127))) {
2688 if (skb_checksum_help(skb))
2689 goto drop;
2690 cksum_offset = 0;
2691 pseudo_hdr_offset = 0;
2692 } else {
2693 odd_flag = MXGEFW_FLAGS_ALIGN_ODD;
2694 flags |= MXGEFW_FLAGS_CKSUM;
2695 }
2696 }
2697
2698 cum_len = 0;
2699
2700 if (mss) { /* TSO */
2701 /* this removes any CKSUM flag from before */
2702 flags = (MXGEFW_FLAGS_TSO_HDR | MXGEFW_FLAGS_FIRST);
2703
2704 /* negative cum_len signifies to the
2705 * send loop that we are still in the
2706 * header portion of the TSO packet.
2707 * TSO header can be at most 1KB long */
2708 cum_len = -(skb_transport_offset(skb) + tcp_hdrlen(skb));
2709
2710 /* for IPv6 TSO, the checksum offset stores the
2711 * TCP header length, to save the firmware from
2712 * the need to parse the headers */
2713 if (skb_is_gso_v6(skb)) {
2714 cksum_offset = tcp_hdrlen(skb);
2715 /* Can only handle headers <= max_tso6 long */
2716 if (unlikely(-cum_len > mgp->max_tso6))
2717 return myri10ge_sw_tso(skb, dev);
2718 }
2719 /* for TSO, pseudo_hdr_offset holds mss.
2720 * The firmware figures out where to put
2721 * the checksum by parsing the header. */
2722 pseudo_hdr_offset = mss;
2723 } else
2724 /* Mark small packets, and pad out tiny packets */
2725 if (skb->len <= MXGEFW_SEND_SMALL_SIZE) {
2726 flags |= MXGEFW_FLAGS_SMALL;
2727
2728 /* pad frames to at least ETH_ZLEN bytes */
2729 if (unlikely(skb->len < ETH_ZLEN)) {
2730 if (skb_padto(skb, ETH_ZLEN)) {
2731 /* The packet is gone, so we must
2732 * return 0 */
2733 ss->stats.tx_dropped += 1;
2734 return 0;
2735 }
2736 /* adjust the len to account for the zero pad
2737 * so that the nic can know how long it is */
2738 skb->len = ETH_ZLEN;
2739 }
2740 }
2741
2742 /* map the skb for DMA */
2743 len = skb->len - skb->data_len;
2744 idx = tx->req & tx->mask;
2745 tx->info[idx].skb = skb;
2746 bus = pci_map_single(mgp->pdev, skb->data, len, PCI_DMA_TODEVICE);
2747 pci_unmap_addr_set(&tx->info[idx], bus, bus);
2748 pci_unmap_len_set(&tx->info[idx], len, len);
2749
2750 frag_cnt = skb_shinfo(skb)->nr_frags;
2751 frag_idx = 0;
2752 count = 0;
2753 rdma_count = 0;
2754
2755 /* "rdma_count" is the number of RDMAs belonging to the
2756 * current packet BEFORE the current send request. For
2757 * non-TSO packets, this is equal to "count".
2758 * For TSO packets, rdma_count needs to be reset
2759 * to 0 after a segment cut.
2760 *
2761 * The rdma_count field of the send request is
2762 * the number of RDMAs of the packet starting at
2763 * that request. For TSO send requests with one ore more cuts
2764 * in the middle, this is the number of RDMAs starting
2765 * after the last cut in the request. All previous
2766 * segments before the last cut implicitly have 1 RDMA.
2767 *
2768 * Since the number of RDMAs is not known beforehand,
2769 * it must be filled-in retroactively - after each
2770 * segmentation cut or at the end of the entire packet.
2771 */
2772
2773 while (1) {
2774 /* Break the SKB or Fragment up into pieces which
2775 * do not cross mgp->tx_boundary */
2776 low = MYRI10GE_LOWPART_TO_U32(bus);
2777 high_swapped = htonl(MYRI10GE_HIGHPART_TO_U32(bus));
2778 while (len) {
2779 u8 flags_next;
2780 int cum_len_next;
2781
2782 if (unlikely(count == max_segments))
2783 goto abort_linearize;
2784
2785 boundary =
2786 (low + mgp->tx_boundary) & ~(mgp->tx_boundary - 1);
2787 seglen = boundary - low;
2788 if (seglen > len)
2789 seglen = len;
2790 flags_next = flags & ~MXGEFW_FLAGS_FIRST;
2791 cum_len_next = cum_len + seglen;
2792 if (mss) { /* TSO */
2793 (req - rdma_count)->rdma_count = rdma_count + 1;
2794
2795 if (likely(cum_len >= 0)) { /* payload */
2796 int next_is_first, chop;
2797
2798 chop = (cum_len_next > mss);
2799 cum_len_next = cum_len_next % mss;
2800 next_is_first = (cum_len_next == 0);
2801 flags |= chop * MXGEFW_FLAGS_TSO_CHOP;
2802 flags_next |= next_is_first *
2803 MXGEFW_FLAGS_FIRST;
2804 rdma_count |= -(chop | next_is_first);
2805 rdma_count += chop & !next_is_first;
2806 } else if (likely(cum_len_next >= 0)) { /* header ends */
2807 int small;
2808
2809 rdma_count = -1;
2810 cum_len_next = 0;
2811 seglen = -cum_len;
2812 small = (mss <= MXGEFW_SEND_SMALL_SIZE);
2813 flags_next = MXGEFW_FLAGS_TSO_PLD |
2814 MXGEFW_FLAGS_FIRST |
2815 (small * MXGEFW_FLAGS_SMALL);
2816 }
2817 }
2818 req->addr_high = high_swapped;
2819 req->addr_low = htonl(low);
2820 req->pseudo_hdr_offset = htons(pseudo_hdr_offset);
2821 req->pad = 0; /* complete solid 16-byte block; does this matter? */
2822 req->rdma_count = 1;
2823 req->length = htons(seglen);
2824 req->cksum_offset = cksum_offset;
2825 req->flags = flags | ((cum_len & 1) * odd_flag);
2826
2827 low += seglen;
2828 len -= seglen;
2829 cum_len = cum_len_next;
2830 flags = flags_next;
2831 req++;
2832 count++;
2833 rdma_count++;
2834 if (cksum_offset != 0 && !(mss && skb_is_gso_v6(skb))) {
2835 if (unlikely(cksum_offset > seglen))
2836 cksum_offset -= seglen;
2837 else
2838 cksum_offset = 0;
2839 }
2840 }
2841 if (frag_idx == frag_cnt)
2842 break;
2843
2844 /* map next fragment for DMA */
2845 idx = (count + tx->req) & tx->mask;
2846 frag = &skb_shinfo(skb)->frags[frag_idx];
2847 frag_idx++;
2848 len = frag->size;
2849 bus = pci_map_page(mgp->pdev, frag->page, frag->page_offset,
2850 len, PCI_DMA_TODEVICE);
2851 pci_unmap_addr_set(&tx->info[idx], bus, bus);
2852 pci_unmap_len_set(&tx->info[idx], len, len);
2853 }
2854
2855 (req - rdma_count)->rdma_count = rdma_count;
2856 if (mss)
2857 do {
2858 req--;
2859 req->flags |= MXGEFW_FLAGS_TSO_LAST;
2860 } while (!(req->flags & (MXGEFW_FLAGS_TSO_CHOP |
2861 MXGEFW_FLAGS_FIRST)));
2862 idx = ((count - 1) + tx->req) & tx->mask;
2863 tx->info[idx].last = 1;
2864 myri10ge_submit_req(tx, tx->req_list, count);
2865 /* if using multiple tx queues, make sure NIC polls the
2866 * current slice */
2867 if ((mgp->dev->real_num_tx_queues > 1) && tx->queue_active == 0) {
2868 tx->queue_active = 1;
2869 put_be32(htonl(1), tx->send_go);
2870 mb();
2871 mmiowb();
2872 }
2873 tx->pkt_start++;
2874 if ((avail - count) < MXGEFW_MAX_SEND_DESC) {
2875 tx->stop_queue++;
2876 netif_tx_stop_queue(netdev_queue);
2877 }
2878 dev->trans_start = jiffies;
2879 return 0;
2880
2881 abort_linearize:
2882 /* Free any DMA resources we've alloced and clear out the skb
2883 * slot so as to not trip up assertions, and to avoid a
2884 * double-free if linearizing fails */
2885
2886 last_idx = (idx + 1) & tx->mask;
2887 idx = tx->req & tx->mask;
2888 tx->info[idx].skb = NULL;
2889 do {
2890 len = pci_unmap_len(&tx->info[idx], len);
2891 if (len) {
2892 if (tx->info[idx].skb != NULL)
2893 pci_unmap_single(mgp->pdev,
2894 pci_unmap_addr(&tx->info[idx],
2895 bus), len,
2896 PCI_DMA_TODEVICE);
2897 else
2898 pci_unmap_page(mgp->pdev,
2899 pci_unmap_addr(&tx->info[idx],
2900 bus), len,
2901 PCI_DMA_TODEVICE);
2902 pci_unmap_len_set(&tx->info[idx], len, 0);
2903 tx->info[idx].skb = NULL;
2904 }
2905 idx = (idx + 1) & tx->mask;
2906 } while (idx != last_idx);
2907 if (skb_is_gso(skb)) {
2908 printk(KERN_ERR
2909 "myri10ge: %s: TSO but wanted to linearize?!?!?\n",
2910 mgp->dev->name);
2911 goto drop;
2912 }
2913
2914 if (skb_linearize(skb))
2915 goto drop;
2916
2917 tx->linearized++;
2918 goto again;
2919
2920 drop:
2921 dev_kfree_skb_any(skb);
2922 ss->stats.tx_dropped += 1;
2923 return 0;
2924
2925 }
2926
2927 static int myri10ge_sw_tso(struct sk_buff *skb, struct net_device *dev)
2928 {
2929 struct sk_buff *segs, *curr;
2930 struct myri10ge_priv *mgp = netdev_priv(dev);
2931 struct myri10ge_slice_state *ss;
2932 int status;
2933
2934 segs = skb_gso_segment(skb, dev->features & ~NETIF_F_TSO6);
2935 if (IS_ERR(segs))
2936 goto drop;
2937
2938 while (segs) {
2939 curr = segs;
2940 segs = segs->next;
2941 curr->next = NULL;
2942 status = myri10ge_xmit(curr, dev);
2943 if (status != 0) {
2944 dev_kfree_skb_any(curr);
2945 if (segs != NULL) {
2946 curr = segs;
2947 segs = segs->next;
2948 curr->next = NULL;
2949 dev_kfree_skb_any(segs);
2950 }
2951 goto drop;
2952 }
2953 }
2954 dev_kfree_skb_any(skb);
2955 return 0;
2956
2957 drop:
2958 ss = &mgp->ss[skb_get_queue_mapping(skb)];
2959 dev_kfree_skb_any(skb);
2960 ss->stats.tx_dropped += 1;
2961 return 0;
2962 }
2963
2964 static struct net_device_stats *myri10ge_get_stats(struct net_device *dev)
2965 {
2966 struct myri10ge_priv *mgp = netdev_priv(dev);
2967 struct myri10ge_slice_netstats *slice_stats;
2968 struct net_device_stats *stats = &mgp->stats;
2969 int i;
2970
2971 memset(stats, 0, sizeof(*stats));
2972 for (i = 0; i < mgp->num_slices; i++) {
2973 slice_stats = &mgp->ss[i].stats;
2974 stats->rx_packets += slice_stats->rx_packets;
2975 stats->tx_packets += slice_stats->tx_packets;
2976 stats->rx_bytes += slice_stats->rx_bytes;
2977 stats->tx_bytes += slice_stats->tx_bytes;
2978 stats->rx_dropped += slice_stats->rx_dropped;
2979 stats->tx_dropped += slice_stats->tx_dropped;
2980 }
2981 return stats;
2982 }
2983
2984 static void myri10ge_set_multicast_list(struct net_device *dev)
2985 {
2986 struct myri10ge_priv *mgp = netdev_priv(dev);
2987 struct myri10ge_cmd cmd;
2988 struct dev_mc_list *mc_list;
2989 __be32 data[2] = { 0, 0 };
2990 int err;
2991
2992 /* can be called from atomic contexts,
2993 * pass 1 to force atomicity in myri10ge_send_cmd() */
2994 myri10ge_change_promisc(mgp, dev->flags & IFF_PROMISC, 1);
2995
2996 /* This firmware is known to not support multicast */
2997 if (!mgp->fw_multicast_support)
2998 return;
2999
3000 /* Disable multicast filtering */
3001
3002 err = myri10ge_send_cmd(mgp, MXGEFW_ENABLE_ALLMULTI, &cmd, 1);
3003 if (err != 0) {
3004 printk(KERN_ERR "myri10ge: %s: Failed MXGEFW_ENABLE_ALLMULTI,"
3005 " error status: %d\n", dev->name, err);
3006 goto abort;
3007 }
3008
3009 if ((dev->flags & IFF_ALLMULTI) || mgp->adopted_rx_filter_bug) {
3010 /* request to disable multicast filtering, so quit here */
3011 return;
3012 }
3013
3014 /* Flush the filters */
3015
3016 err = myri10ge_send_cmd(mgp, MXGEFW_LEAVE_ALL_MULTICAST_GROUPS,
3017 &cmd, 1);
3018 if (err != 0) {
3019 printk(KERN_ERR
3020 "myri10ge: %s: Failed MXGEFW_LEAVE_ALL_MULTICAST_GROUPS"
3021 ", error status: %d\n", dev->name, err);
3022 goto abort;
3023 }
3024
3025 /* Walk the multicast list, and add each address */
3026 for (mc_list = dev->mc_list; mc_list != NULL; mc_list = mc_list->next) {
3027 memcpy(data, &mc_list->dmi_addr, 6);
3028 cmd.data0 = ntohl(data[0]);
3029 cmd.data1 = ntohl(data[1]);
3030 err = myri10ge_send_cmd(mgp, MXGEFW_JOIN_MULTICAST_GROUP,
3031 &cmd, 1);
3032
3033 if (err != 0) {
3034 printk(KERN_ERR "myri10ge: %s: Failed "
3035 "MXGEFW_JOIN_MULTICAST_GROUP, error status:"
3036 "%d\t", dev->name, err);
3037 printk(KERN_ERR "MAC %pM\n", mc_list->dmi_addr);
3038 goto abort;
3039 }
3040 }
3041 /* Enable multicast filtering */
3042 err = myri10ge_send_cmd(mgp, MXGEFW_DISABLE_ALLMULTI, &cmd, 1);
3043 if (err != 0) {
3044 printk(KERN_ERR "myri10ge: %s: Failed MXGEFW_DISABLE_ALLMULTI,"
3045 "error status: %d\n", dev->name, err);
3046 goto abort;
3047 }
3048
3049 return;
3050
3051 abort:
3052 return;
3053 }
3054
3055 static int myri10ge_set_mac_address(struct net_device *dev, void *addr)
3056 {
3057 struct sockaddr *sa = addr;
3058 struct myri10ge_priv *mgp = netdev_priv(dev);
3059 int status;
3060
3061 if (!is_valid_ether_addr(sa->sa_data))
3062 return -EADDRNOTAVAIL;
3063
3064 status = myri10ge_update_mac_address(mgp, sa->sa_data);
3065 if (status != 0) {
3066 printk(KERN_ERR
3067 "myri10ge: %s: changing mac address failed with %d\n",
3068 dev->name, status);
3069 return status;
3070 }
3071
3072 /* change the dev structure */
3073 memcpy(dev->dev_addr, sa->sa_data, 6);
3074 return 0;
3075 }
3076
3077 static int myri10ge_change_mtu(struct net_device *dev, int new_mtu)
3078 {
3079 struct myri10ge_priv *mgp = netdev_priv(dev);
3080 int error = 0;
3081
3082 if ((new_mtu < 68) || (ETH_HLEN + new_mtu > MYRI10GE_MAX_ETHER_MTU)) {
3083 printk(KERN_ERR "myri10ge: %s: new mtu (%d) is not valid\n",
3084 dev->name, new_mtu);
3085 return -EINVAL;
3086 }
3087 printk(KERN_INFO "%s: changing mtu from %d to %d\n",
3088 dev->name, dev->mtu, new_mtu);
3089 if (mgp->running) {
3090 /* if we change the mtu on an active device, we must
3091 * reset the device so the firmware sees the change */
3092 myri10ge_close(dev);
3093 dev->mtu = new_mtu;
3094 myri10ge_open(dev);
3095 } else
3096 dev->mtu = new_mtu;
3097
3098 return error;
3099 }
3100
3101 /*
3102 * Enable ECRC to align PCI-E Completion packets on an 8-byte boundary.
3103 * Only do it if the bridge is a root port since we don't want to disturb
3104 * any other device, except if forced with myri10ge_ecrc_enable > 1.
3105 */
3106
3107 static void myri10ge_enable_ecrc(struct myri10ge_priv *mgp)
3108 {
3109 struct pci_dev *bridge = mgp->pdev->bus->self;
3110 struct device *dev = &mgp->pdev->dev;
3111 unsigned cap;
3112 unsigned err_cap;
3113 u16 val;
3114 u8 ext_type;
3115 int ret;
3116
3117 if (!myri10ge_ecrc_enable || !bridge)
3118 return;
3119
3120 /* check that the bridge is a root port */
3121 cap = pci_find_capability(bridge, PCI_CAP_ID_EXP);
3122 pci_read_config_word(bridge, cap + PCI_CAP_FLAGS, &val);
3123 ext_type = (val & PCI_EXP_FLAGS_TYPE) >> 4;
3124 if (ext_type != PCI_EXP_TYPE_ROOT_PORT) {
3125 if (myri10ge_ecrc_enable > 1) {
3126 struct pci_dev *prev_bridge, *old_bridge = bridge;
3127
3128 /* Walk the hierarchy up to the root port
3129 * where ECRC has to be enabled */
3130 do {
3131 prev_bridge = bridge;
3132 bridge = bridge->bus->self;
3133 if (!bridge || prev_bridge == bridge) {
3134 dev_err(dev,
3135 "Failed to find root port"
3136 " to force ECRC\n");
3137 return;
3138 }
3139 cap =
3140 pci_find_capability(bridge, PCI_CAP_ID_EXP);
3141 pci_read_config_word(bridge,
3142 cap + PCI_CAP_FLAGS, &val);
3143 ext_type = (val & PCI_EXP_FLAGS_TYPE) >> 4;
3144 } while (ext_type != PCI_EXP_TYPE_ROOT_PORT);
3145
3146 dev_info(dev,
3147 "Forcing ECRC on non-root port %s"
3148 " (enabling on root port %s)\n",
3149 pci_name(old_bridge), pci_name(bridge));
3150 } else {
3151 dev_err(dev,
3152 "Not enabling ECRC on non-root port %s\n",
3153 pci_name(bridge));
3154 return;
3155 }
3156 }
3157
3158 cap = pci_find_ext_capability(bridge, PCI_EXT_CAP_ID_ERR);
3159 if (!cap)
3160 return;
3161
3162 ret = pci_read_config_dword(bridge, cap + PCI_ERR_CAP, &err_cap);
3163 if (ret) {
3164 dev_err(dev, "failed reading ext-conf-space of %s\n",
3165 pci_name(bridge));
3166 dev_err(dev, "\t pci=nommconf in use? "
3167 "or buggy/incomplete/absent ACPI MCFG attr?\n");
3168 return;
3169 }
3170 if (!(err_cap & PCI_ERR_CAP_ECRC_GENC))
3171 return;
3172
3173 err_cap |= PCI_ERR_CAP_ECRC_GENE;
3174 pci_write_config_dword(bridge, cap + PCI_ERR_CAP, err_cap);
3175 dev_info(dev, "Enabled ECRC on upstream bridge %s\n", pci_name(bridge));
3176 }
3177
3178 /*
3179 * The Lanai Z8E PCI-E interface achieves higher Read-DMA throughput
3180 * when the PCI-E Completion packets are aligned on an 8-byte
3181 * boundary. Some PCI-E chip sets always align Completion packets; on
3182 * the ones that do not, the alignment can be enforced by enabling
3183 * ECRC generation (if supported).
3184 *
3185 * When PCI-E Completion packets are not aligned, it is actually more
3186 * efficient to limit Read-DMA transactions to 2KB, rather than 4KB.
3187 *
3188 * If the driver can neither enable ECRC nor verify that it has
3189 * already been enabled, then it must use a firmware image which works
3190 * around unaligned completion packets (myri10ge_rss_ethp_z8e.dat), and it
3191 * should also ensure that it never gives the device a Read-DMA which is
3192 * larger than 2KB by setting the tx_boundary to 2KB. If ECRC is
3193 * enabled, then the driver should use the aligned (myri10ge_rss_eth_z8e.dat)
3194 * firmware image, and set tx_boundary to 4KB.
3195 */
3196
3197 static void myri10ge_firmware_probe(struct myri10ge_priv *mgp)
3198 {
3199 struct pci_dev *pdev = mgp->pdev;
3200 struct device *dev = &pdev->dev;
3201 int status;
3202
3203 mgp->tx_boundary = 4096;
3204 /*
3205 * Verify the max read request size was set to 4KB
3206 * before trying the test with 4KB.
3207 */
3208 status = pcie_get_readrq(pdev);
3209 if (status < 0) {
3210 dev_err(dev, "Couldn't read max read req size: %d\n", status);
3211 goto abort;
3212 }
3213 if (status != 4096) {
3214 dev_warn(dev, "Max Read Request size != 4096 (%d)\n", status);
3215 mgp->tx_boundary = 2048;
3216 }
3217 /*
3218 * load the optimized firmware (which assumes aligned PCIe
3219 * completions) in order to see if it works on this host.
3220 */
3221 mgp->fw_name = myri10ge_fw_aligned;
3222 status = myri10ge_load_firmware(mgp, 1);
3223 if (status != 0) {
3224 goto abort;
3225 }
3226
3227 /*
3228 * Enable ECRC if possible
3229 */
3230 myri10ge_enable_ecrc(mgp);
3231
3232 /*
3233 * Run a DMA test which watches for unaligned completions and
3234 * aborts on the first one seen.
3235 */
3236
3237 status = myri10ge_dma_test(mgp, MXGEFW_CMD_UNALIGNED_TEST);
3238 if (status == 0)
3239 return; /* keep the aligned firmware */
3240
3241 if (status != -E2BIG)
3242 dev_warn(dev, "DMA test failed: %d\n", status);
3243 if (status == -ENOSYS)
3244 dev_warn(dev, "Falling back to ethp! "
3245 "Please install up to date fw\n");
3246 abort:
3247 /* fall back to using the unaligned firmware */
3248 mgp->tx_boundary = 2048;
3249 mgp->fw_name = myri10ge_fw_unaligned;
3250
3251 }
3252
3253 static void myri10ge_select_firmware(struct myri10ge_priv *mgp)
3254 {
3255 if (myri10ge_force_firmware == 0) {
3256 int link_width, exp_cap;
3257 u16 lnk;
3258
3259 exp_cap = pci_find_capability(mgp->pdev, PCI_CAP_ID_EXP);
3260 pci_read_config_word(mgp->pdev, exp_cap + PCI_EXP_LNKSTA, &lnk);
3261 link_width = (lnk >> 4) & 0x3f;
3262
3263 /* Check to see if Link is less than 8 or if the
3264 * upstream bridge is known to provide aligned
3265 * completions */
3266 if (link_width < 8) {
3267 dev_info(&mgp->pdev->dev, "PCIE x%d Link\n",
3268 link_width);
3269 mgp->tx_boundary = 4096;
3270 mgp->fw_name = myri10ge_fw_aligned;
3271 } else {
3272 myri10ge_firmware_probe(mgp);
3273 }
3274 } else {
3275 if (myri10ge_force_firmware == 1) {
3276 dev_info(&mgp->pdev->dev,
3277 "Assuming aligned completions (forced)\n");
3278 mgp->tx_boundary = 4096;
3279 mgp->fw_name = myri10ge_fw_aligned;
3280 } else {
3281 dev_info(&mgp->pdev->dev,
3282 "Assuming unaligned completions (forced)\n");
3283 mgp->tx_boundary = 2048;
3284 mgp->fw_name = myri10ge_fw_unaligned;
3285 }
3286 }
3287 if (myri10ge_fw_name != NULL) {
3288 dev_info(&mgp->pdev->dev, "overriding firmware to %s\n",
3289 myri10ge_fw_name);
3290 mgp->fw_name = myri10ge_fw_name;
3291 }
3292 }
3293
3294 #ifdef CONFIG_PM
3295 static int myri10ge_suspend(struct pci_dev *pdev, pm_message_t state)
3296 {
3297 struct myri10ge_priv *mgp;
3298 struct net_device *netdev;
3299
3300 mgp = pci_get_drvdata(pdev);
3301 if (mgp == NULL)
3302 return -EINVAL;
3303 netdev = mgp->dev;
3304
3305 netif_device_detach(netdev);
3306 if (netif_running(netdev)) {
3307 printk(KERN_INFO "myri10ge: closing %s\n", netdev->name);
3308 rtnl_lock();
3309 myri10ge_close(netdev);
3310 rtnl_unlock();
3311 }
3312 myri10ge_dummy_rdma(mgp, 0);
3313 pci_save_state(pdev);
3314 pci_disable_device(pdev);
3315
3316 return pci_set_power_state(pdev, pci_choose_state(pdev, state));
3317 }
3318
3319 static int myri10ge_resume(struct pci_dev *pdev)
3320 {
3321 struct myri10ge_priv *mgp;
3322 struct net_device *netdev;
3323 int status;
3324 u16 vendor;
3325
3326 mgp = pci_get_drvdata(pdev);
3327 if (mgp == NULL)
3328 return -EINVAL;
3329 netdev = mgp->dev;
3330 pci_set_power_state(pdev, 0); /* zeros conf space as a side effect */
3331 msleep(5); /* give card time to respond */
3332 pci_read_config_word(mgp->pdev, PCI_VENDOR_ID, &vendor);
3333 if (vendor == 0xffff) {
3334 printk(KERN_ERR "myri10ge: %s: device disappeared!\n",
3335 mgp->dev->name);
3336 return -EIO;
3337 }
3338
3339 status = pci_restore_state(pdev);
3340 if (status)
3341 return status;
3342
3343 status = pci_enable_device(pdev);
3344 if (status) {
3345 dev_err(&pdev->dev, "failed to enable device\n");
3346 return status;
3347 }
3348
3349 pci_set_master(pdev);
3350
3351 myri10ge_reset(mgp);
3352 myri10ge_dummy_rdma(mgp, 1);
3353
3354 /* Save configuration space to be restored if the
3355 * nic resets due to a parity error */
3356 pci_save_state(pdev);
3357
3358 if (netif_running(netdev)) {
3359 rtnl_lock();
3360 status = myri10ge_open(netdev);
3361 rtnl_unlock();
3362 if (status != 0)
3363 goto abort_with_enabled;
3364
3365 }
3366 netif_device_attach(netdev);
3367
3368 return 0;
3369
3370 abort_with_enabled:
3371 pci_disable_device(pdev);
3372 return -EIO;
3373
3374 }
3375 #endif /* CONFIG_PM */
3376
3377 static u32 myri10ge_read_reboot(struct myri10ge_priv *mgp)
3378 {
3379 struct pci_dev *pdev = mgp->pdev;
3380 int vs = mgp->vendor_specific_offset;
3381 u32 reboot;
3382
3383 /*enter read32 mode */
3384 pci_write_config_byte(pdev, vs + 0x10, 0x3);
3385
3386 /*read REBOOT_STATUS (0xfffffff0) */
3387 pci_write_config_dword(pdev, vs + 0x18, 0xfffffff0);
3388 pci_read_config_dword(pdev, vs + 0x14, &reboot);
3389 return reboot;
3390 }
3391
3392 /*
3393 * This watchdog is used to check whether the board has suffered
3394 * from a parity error and needs to be recovered.
3395 */
3396 static void myri10ge_watchdog(struct work_struct *work)
3397 {
3398 struct myri10ge_priv *mgp =
3399 container_of(work, struct myri10ge_priv, watchdog_work);
3400 struct myri10ge_tx_buf *tx;
3401 u32 reboot;
3402 int status;
3403 int i;
3404 u16 cmd, vendor;
3405
3406 mgp->watchdog_resets++;
3407 pci_read_config_word(mgp->pdev, PCI_COMMAND, &cmd);
3408 if ((cmd & PCI_COMMAND_MASTER) == 0) {
3409 /* Bus master DMA disabled? Check to see
3410 * if the card rebooted due to a parity error
3411 * For now, just report it */
3412 reboot = myri10ge_read_reboot(mgp);
3413 printk(KERN_ERR
3414 "myri10ge: %s: NIC rebooted (0x%x),%s resetting\n",
3415 mgp->dev->name, reboot,
3416 myri10ge_reset_recover ? " " : " not");
3417 if (myri10ge_reset_recover == 0)
3418 return;
3419
3420 myri10ge_reset_recover--;
3421
3422 /*
3423 * A rebooted nic will come back with config space as
3424 * it was after power was applied to PCIe bus.
3425 * Attempt to restore config space which was saved
3426 * when the driver was loaded, or the last time the
3427 * nic was resumed from power saving mode.
3428 */
3429 pci_restore_state(mgp->pdev);
3430
3431 /* save state again for accounting reasons */
3432 pci_save_state(mgp->pdev);
3433
3434 } else {
3435 /* if we get back -1's from our slot, perhaps somebody
3436 * powered off our card. Don't try to reset it in
3437 * this case */
3438 if (cmd == 0xffff) {
3439 pci_read_config_word(mgp->pdev, PCI_VENDOR_ID, &vendor);
3440 if (vendor == 0xffff) {
3441 printk(KERN_ERR
3442 "myri10ge: %s: device disappeared!\n",
3443 mgp->dev->name);
3444 return;
3445 }
3446 }
3447 /* Perhaps it is a software error. Try to reset */
3448
3449 printk(KERN_ERR "myri10ge: %s: device timeout, resetting\n",
3450 mgp->dev->name);
3451 for (i = 0; i < mgp->num_slices; i++) {
3452 tx = &mgp->ss[i].tx;
3453 printk(KERN_INFO
3454 "myri10ge: %s: (%d): %d %d %d %d %d %d\n",
3455 mgp->dev->name, i, tx->queue_active, tx->req,
3456 tx->done, tx->pkt_start, tx->pkt_done,
3457 (int)ntohl(mgp->ss[i].fw_stats->
3458 send_done_count));
3459 msleep(2000);
3460 printk(KERN_INFO
3461 "myri10ge: %s: (%d): %d %d %d %d %d %d\n",
3462 mgp->dev->name, i, tx->queue_active, tx->req,
3463 tx->done, tx->pkt_start, tx->pkt_done,
3464 (int)ntohl(mgp->ss[i].fw_stats->
3465 send_done_count));
3466 }
3467 }
3468
3469 rtnl_lock();
3470 myri10ge_close(mgp->dev);
3471 status = myri10ge_load_firmware(mgp, 1);
3472 if (status != 0)
3473 printk(KERN_ERR "myri10ge: %s: failed to load firmware\n",
3474 mgp->dev->name);
3475 else
3476 myri10ge_open(mgp->dev);
3477 rtnl_unlock();
3478 }
3479
3480 /*
3481 * We use our own timer routine rather than relying upon
3482 * netdev->tx_timeout because we have a very large hardware transmit
3483 * queue. Due to the large queue, the netdev->tx_timeout function
3484 * cannot detect a NIC with a parity error in a timely fashion if the
3485 * NIC is lightly loaded.
3486 */
3487 static void myri10ge_watchdog_timer(unsigned long arg)
3488 {
3489 struct myri10ge_priv *mgp;
3490 struct myri10ge_slice_state *ss;
3491 int i, reset_needed;
3492 u32 rx_pause_cnt;
3493
3494 mgp = (struct myri10ge_priv *)arg;
3495
3496 rx_pause_cnt = ntohl(mgp->ss[0].fw_stats->dropped_pause);
3497 for (i = 0, reset_needed = 0;
3498 i < mgp->num_slices && reset_needed == 0; ++i) {
3499
3500 ss = &mgp->ss[i];
3501 if (ss->rx_small.watchdog_needed) {
3502 myri10ge_alloc_rx_pages(mgp, &ss->rx_small,
3503 mgp->small_bytes + MXGEFW_PAD,
3504 1);
3505 if (ss->rx_small.fill_cnt - ss->rx_small.cnt >=
3506 myri10ge_fill_thresh)
3507 ss->rx_small.watchdog_needed = 0;
3508 }
3509 if (ss->rx_big.watchdog_needed) {
3510 myri10ge_alloc_rx_pages(mgp, &ss->rx_big,
3511 mgp->big_bytes, 1);
3512 if (ss->rx_big.fill_cnt - ss->rx_big.cnt >=
3513 myri10ge_fill_thresh)
3514 ss->rx_big.watchdog_needed = 0;
3515 }
3516
3517 if (ss->tx.req != ss->tx.done &&
3518 ss->tx.done == ss->watchdog_tx_done &&
3519 ss->watchdog_tx_req != ss->watchdog_tx_done) {
3520 /* nic seems like it might be stuck.. */
3521 if (rx_pause_cnt != mgp->watchdog_pause) {
3522 if (net_ratelimit())
3523 printk(KERN_WARNING
3524 "myri10ge %s slice %d:"
3525 "TX paused, check link partner\n",
3526 mgp->dev->name, i);
3527 } else {
3528 printk(KERN_WARNING
3529 "myri10ge %s slice %d stuck:",
3530 mgp->dev->name, i);
3531 reset_needed = 1;
3532 }
3533 }
3534 ss->watchdog_tx_done = ss->tx.done;
3535 ss->watchdog_tx_req = ss->tx.req;
3536 }
3537 mgp->watchdog_pause = rx_pause_cnt;
3538
3539 if (reset_needed) {
3540 schedule_work(&mgp->watchdog_work);
3541 } else {
3542 /* rearm timer */
3543 mod_timer(&mgp->watchdog_timer,
3544 jiffies + myri10ge_watchdog_timeout * HZ);
3545 }
3546 }
3547
3548 static void myri10ge_free_slices(struct myri10ge_priv *mgp)
3549 {
3550 struct myri10ge_slice_state *ss;
3551 struct pci_dev *pdev = mgp->pdev;
3552 size_t bytes;
3553 int i;
3554
3555 if (mgp->ss == NULL)
3556 return;
3557
3558 for (i = 0; i < mgp->num_slices; i++) {
3559 ss = &mgp->ss[i];
3560 if (ss->rx_done.entry != NULL) {
3561 bytes = mgp->max_intr_slots *
3562 sizeof(*ss->rx_done.entry);
3563 dma_free_coherent(&pdev->dev, bytes,
3564 ss->rx_done.entry, ss->rx_done.bus);
3565 ss->rx_done.entry = NULL;
3566 }
3567 if (ss->fw_stats != NULL) {
3568 bytes = sizeof(*ss->fw_stats);
3569 dma_free_coherent(&pdev->dev, bytes,
3570 ss->fw_stats, ss->fw_stats_bus);
3571 ss->fw_stats = NULL;
3572 }
3573 }
3574 kfree(mgp->ss);
3575 mgp->ss = NULL;
3576 }
3577
3578 static int myri10ge_alloc_slices(struct myri10ge_priv *mgp)
3579 {
3580 struct myri10ge_slice_state *ss;
3581 struct pci_dev *pdev = mgp->pdev;
3582 size_t bytes;
3583 int i;
3584
3585 bytes = sizeof(*mgp->ss) * mgp->num_slices;
3586 mgp->ss = kzalloc(bytes, GFP_KERNEL);
3587 if (mgp->ss == NULL) {
3588 return -ENOMEM;
3589 }
3590
3591 for (i = 0; i < mgp->num_slices; i++) {
3592 ss = &mgp->ss[i];
3593 bytes = mgp->max_intr_slots * sizeof(*ss->rx_done.entry);
3594 ss->rx_done.entry = dma_alloc_coherent(&pdev->dev, bytes,
3595 &ss->rx_done.bus,
3596 GFP_KERNEL);
3597 if (ss->rx_done.entry == NULL)
3598 goto abort;
3599 memset(ss->rx_done.entry, 0, bytes);
3600 bytes = sizeof(*ss->fw_stats);
3601 ss->fw_stats = dma_alloc_coherent(&pdev->dev, bytes,
3602 &ss->fw_stats_bus,
3603 GFP_KERNEL);
3604 if (ss->fw_stats == NULL)
3605 goto abort;
3606 ss->mgp = mgp;
3607 ss->dev = mgp->dev;
3608 netif_napi_add(ss->dev, &ss->napi, myri10ge_poll,
3609 myri10ge_napi_weight);
3610 }
3611 return 0;
3612 abort:
3613 myri10ge_free_slices(mgp);
3614 return -ENOMEM;
3615 }
3616
3617 /*
3618 * This function determines the number of slices supported.
3619 * The number slices is the minumum of the number of CPUS,
3620 * the number of MSI-X irqs supported, the number of slices
3621 * supported by the firmware
3622 */
3623 static void myri10ge_probe_slices(struct myri10ge_priv *mgp)
3624 {
3625 struct myri10ge_cmd cmd;
3626 struct pci_dev *pdev = mgp->pdev;
3627 char *old_fw;
3628 int i, status, ncpus, msix_cap;
3629
3630 mgp->num_slices = 1;
3631 msix_cap = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
3632 ncpus = num_online_cpus();
3633
3634 if (myri10ge_max_slices == 1 || msix_cap == 0 ||
3635 (myri10ge_max_slices == -1 && ncpus < 2))
3636 return;
3637
3638 /* try to load the slice aware rss firmware */
3639 old_fw = mgp->fw_name;
3640 if (myri10ge_fw_name != NULL) {
3641 dev_info(&mgp->pdev->dev, "overriding rss firmware to %s\n",
3642 myri10ge_fw_name);
3643 mgp->fw_name = myri10ge_fw_name;
3644 } else if (old_fw == myri10ge_fw_aligned)
3645 mgp->fw_name = myri10ge_fw_rss_aligned;
3646 else
3647 mgp->fw_name = myri10ge_fw_rss_unaligned;
3648 status = myri10ge_load_firmware(mgp, 0);
3649 if (status != 0) {
3650 dev_info(&pdev->dev, "Rss firmware not found\n");
3651 return;
3652 }
3653
3654 /* hit the board with a reset to ensure it is alive */
3655 memset(&cmd, 0, sizeof(cmd));
3656 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_RESET, &cmd, 0);
3657 if (status != 0) {
3658 dev_err(&mgp->pdev->dev, "failed reset\n");
3659 goto abort_with_fw;
3660 return;
3661 }
3662
3663 mgp->max_intr_slots = cmd.data0 / sizeof(struct mcp_slot);
3664
3665 /* tell it the size of the interrupt queues */
3666 cmd.data0 = mgp->max_intr_slots * sizeof(struct mcp_slot);
3667 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_SIZE, &cmd, 0);
3668 if (status != 0) {
3669 dev_err(&mgp->pdev->dev, "failed MXGEFW_CMD_SET_INTRQ_SIZE\n");
3670 goto abort_with_fw;
3671 }
3672
3673 /* ask the maximum number of slices it supports */
3674 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_MAX_RSS_QUEUES, &cmd, 0);
3675 if (status != 0)
3676 goto abort_with_fw;
3677 else
3678 mgp->num_slices = cmd.data0;
3679
3680 /* Only allow multiple slices if MSI-X is usable */
3681 if (!myri10ge_msi) {
3682 goto abort_with_fw;
3683 }
3684
3685 /* if the admin did not specify a limit to how many
3686 * slices we should use, cap it automatically to the
3687 * number of CPUs currently online */
3688 if (myri10ge_max_slices == -1)
3689 myri10ge_max_slices = ncpus;
3690
3691 if (mgp->num_slices > myri10ge_max_slices)
3692 mgp->num_slices = myri10ge_max_slices;
3693
3694 /* Now try to allocate as many MSI-X vectors as we have
3695 * slices. We give up on MSI-X if we can only get a single
3696 * vector. */
3697
3698 mgp->msix_vectors = kzalloc(mgp->num_slices *
3699 sizeof(*mgp->msix_vectors), GFP_KERNEL);
3700 if (mgp->msix_vectors == NULL)
3701 goto disable_msix;
3702 for (i = 0; i < mgp->num_slices; i++) {
3703 mgp->msix_vectors[i].entry = i;
3704 }
3705
3706 while (mgp->num_slices > 1) {
3707 /* make sure it is a power of two */
3708 while (!is_power_of_2(mgp->num_slices))
3709 mgp->num_slices--;
3710 if (mgp->num_slices == 1)
3711 goto disable_msix;
3712 status = pci_enable_msix(pdev, mgp->msix_vectors,
3713 mgp->num_slices);
3714 if (status == 0) {
3715 pci_disable_msix(pdev);
3716 return;
3717 }
3718 if (status > 0)
3719 mgp->num_slices = status;
3720 else
3721 goto disable_msix;
3722 }
3723
3724 disable_msix:
3725 if (mgp->msix_vectors != NULL) {
3726 kfree(mgp->msix_vectors);
3727 mgp->msix_vectors = NULL;
3728 }
3729
3730 abort_with_fw:
3731 mgp->num_slices = 1;
3732 mgp->fw_name = old_fw;
3733 myri10ge_load_firmware(mgp, 0);
3734 }
3735
3736 static const struct net_device_ops myri10ge_netdev_ops = {
3737 .ndo_open = myri10ge_open,
3738 .ndo_stop = myri10ge_close,
3739 .ndo_start_xmit = myri10ge_xmit,
3740 .ndo_get_stats = myri10ge_get_stats,
3741 .ndo_validate_addr = eth_validate_addr,
3742 .ndo_change_mtu = myri10ge_change_mtu,
3743 .ndo_set_multicast_list = myri10ge_set_multicast_list,
3744 .ndo_set_mac_address = myri10ge_set_mac_address,
3745 };
3746
3747 static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3748 {
3749 struct net_device *netdev;
3750 struct myri10ge_priv *mgp;
3751 struct device *dev = &pdev->dev;
3752 int i;
3753 int status = -ENXIO;
3754 int dac_enabled;
3755 unsigned hdr_offset, ss_offset;
3756
3757 netdev = alloc_etherdev_mq(sizeof(*mgp), MYRI10GE_MAX_SLICES);
3758 if (netdev == NULL) {
3759 dev_err(dev, "Could not allocate ethernet device\n");
3760 return -ENOMEM;
3761 }
3762
3763 SET_NETDEV_DEV(netdev, &pdev->dev);
3764
3765 mgp = netdev_priv(netdev);
3766 mgp->dev = netdev;
3767 mgp->pdev = pdev;
3768 mgp->csum_flag = MXGEFW_FLAGS_CKSUM;
3769 mgp->pause = myri10ge_flow_control;
3770 mgp->intr_coal_delay = myri10ge_intr_coal_delay;
3771 mgp->msg_enable = netif_msg_init(myri10ge_debug, MYRI10GE_MSG_DEFAULT);
3772 init_waitqueue_head(&mgp->down_wq);
3773
3774 if (pci_enable_device(pdev)) {
3775 dev_err(&pdev->dev, "pci_enable_device call failed\n");
3776 status = -ENODEV;
3777 goto abort_with_netdev;
3778 }
3779
3780 /* Find the vendor-specific cap so we can check
3781 * the reboot register later on */
3782 mgp->vendor_specific_offset
3783 = pci_find_capability(pdev, PCI_CAP_ID_VNDR);
3784
3785 /* Set our max read request to 4KB */
3786 status = pcie_set_readrq(pdev, 4096);
3787 if (status != 0) {
3788 dev_err(&pdev->dev, "Error %d writing PCI_EXP_DEVCTL\n",
3789 status);
3790 goto abort_with_enabled;
3791 }
3792
3793 pci_set_master(pdev);
3794 dac_enabled = 1;
3795 status = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
3796 if (status != 0) {
3797 dac_enabled = 0;
3798 dev_err(&pdev->dev,
3799 "64-bit pci address mask was refused, "
3800 "trying 32-bit\n");
3801 status = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
3802 }
3803 if (status != 0) {
3804 dev_err(&pdev->dev, "Error %d setting DMA mask\n", status);
3805 goto abort_with_enabled;
3806 }
3807 (void)pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
3808 mgp->cmd = dma_alloc_coherent(&pdev->dev, sizeof(*mgp->cmd),
3809 &mgp->cmd_bus, GFP_KERNEL);
3810 if (mgp->cmd == NULL)
3811 goto abort_with_enabled;
3812
3813 mgp->board_span = pci_resource_len(pdev, 0);
3814 mgp->iomem_base = pci_resource_start(pdev, 0);
3815 mgp->mtrr = -1;
3816 mgp->wc_enabled = 0;
3817 #ifdef CONFIG_MTRR
3818 mgp->mtrr = mtrr_add(mgp->iomem_base, mgp->board_span,
3819 MTRR_TYPE_WRCOMB, 1);
3820 if (mgp->mtrr >= 0)
3821 mgp->wc_enabled = 1;
3822 #endif
3823 mgp->sram = ioremap_wc(mgp->iomem_base, mgp->board_span);
3824 if (mgp->sram == NULL) {
3825 dev_err(&pdev->dev, "ioremap failed for %ld bytes at 0x%lx\n",
3826 mgp->board_span, mgp->iomem_base);
3827 status = -ENXIO;
3828 goto abort_with_mtrr;
3829 }
3830 hdr_offset =
3831 ntohl(__raw_readl(mgp->sram + MCP_HEADER_PTR_OFFSET)) & 0xffffc;
3832 ss_offset = hdr_offset + offsetof(struct mcp_gen_header, string_specs);
3833 mgp->sram_size = ntohl(__raw_readl(mgp->sram + ss_offset));
3834 if (mgp->sram_size > mgp->board_span ||
3835 mgp->sram_size <= MYRI10GE_FW_OFFSET) {
3836 dev_err(&pdev->dev,
3837 "invalid sram_size %dB or board span %ldB\n",
3838 mgp->sram_size, mgp->board_span);
3839 goto abort_with_ioremap;
3840 }
3841 memcpy_fromio(mgp->eeprom_strings,
3842 mgp->sram + mgp->sram_size, MYRI10GE_EEPROM_STRINGS_SIZE);
3843 memset(mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE - 2, 0, 2);
3844 status = myri10ge_read_mac_addr(mgp);
3845 if (status)
3846 goto abort_with_ioremap;
3847
3848 for (i = 0; i < ETH_ALEN; i++)
3849 netdev->dev_addr[i] = mgp->mac_addr[i];
3850
3851 myri10ge_select_firmware(mgp);
3852
3853 status = myri10ge_load_firmware(mgp, 1);
3854 if (status != 0) {
3855 dev_err(&pdev->dev, "failed to load firmware\n");
3856 goto abort_with_ioremap;
3857 }
3858 myri10ge_probe_slices(mgp);
3859 status = myri10ge_alloc_slices(mgp);
3860 if (status != 0) {
3861 dev_err(&pdev->dev, "failed to alloc slice state\n");
3862 goto abort_with_firmware;
3863 }
3864 netdev->real_num_tx_queues = mgp->num_slices;
3865 status = myri10ge_reset(mgp);
3866 if (status != 0) {
3867 dev_err(&pdev->dev, "failed reset\n");
3868 goto abort_with_slices;
3869 }
3870 #ifdef CONFIG_MYRI10GE_DCA
3871 myri10ge_setup_dca(mgp);
3872 #endif
3873 pci_set_drvdata(pdev, mgp);
3874 if ((myri10ge_initial_mtu + ETH_HLEN) > MYRI10GE_MAX_ETHER_MTU)
3875 myri10ge_initial_mtu = MYRI10GE_MAX_ETHER_MTU - ETH_HLEN;
3876 if ((myri10ge_initial_mtu + ETH_HLEN) < 68)
3877 myri10ge_initial_mtu = 68;
3878
3879 netdev->netdev_ops = &myri10ge_netdev_ops;
3880 netdev->mtu = myri10ge_initial_mtu;
3881 netdev->base_addr = mgp->iomem_base;
3882 netdev->features = mgp->features;
3883
3884 if (dac_enabled)
3885 netdev->features |= NETIF_F_HIGHDMA;
3886
3887 /* make sure we can get an irq, and that MSI can be
3888 * setup (if available). Also ensure netdev->irq
3889 * is set to correct value if MSI is enabled */
3890 status = myri10ge_request_irq(mgp);
3891 if (status != 0)
3892 goto abort_with_firmware;
3893 netdev->irq = pdev->irq;
3894 myri10ge_free_irq(mgp);
3895
3896 /* Save configuration space to be restored if the
3897 * nic resets due to a parity error */
3898 pci_save_state(pdev);
3899
3900 /* Setup the watchdog timer */
3901 setup_timer(&mgp->watchdog_timer, myri10ge_watchdog_timer,
3902 (unsigned long)mgp);
3903
3904 SET_ETHTOOL_OPS(netdev, &myri10ge_ethtool_ops);
3905 INIT_WORK(&mgp->watchdog_work, myri10ge_watchdog);
3906 status = register_netdev(netdev);
3907 if (status != 0) {
3908 dev_err(&pdev->dev, "register_netdev failed: %d\n", status);
3909 goto abort_with_state;
3910 }
3911 if (mgp->msix_enabled)
3912 dev_info(dev, "%d MSI-X IRQs, tx bndry %d, fw %s, WC %s\n",
3913 mgp->num_slices, mgp->tx_boundary, mgp->fw_name,
3914 (mgp->wc_enabled ? "Enabled" : "Disabled"));
3915 else
3916 dev_info(dev, "%s IRQ %d, tx bndry %d, fw %s, WC %s\n",
3917 mgp->msi_enabled ? "MSI" : "xPIC",
3918 netdev->irq, mgp->tx_boundary, mgp->fw_name,
3919 (mgp->wc_enabled ? "Enabled" : "Disabled"));
3920
3921 return 0;
3922
3923 abort_with_state:
3924 pci_restore_state(pdev);
3925
3926 abort_with_slices:
3927 myri10ge_free_slices(mgp);
3928
3929 abort_with_firmware:
3930 myri10ge_dummy_rdma(mgp, 0);
3931
3932 abort_with_ioremap:
3933 if (mgp->mac_addr_string != NULL)
3934 dev_err(&pdev->dev,
3935 "myri10ge_probe() failed: MAC=%s, SN=%ld\n",
3936 mgp->mac_addr_string, mgp->serial_number);
3937 iounmap(mgp->sram);
3938
3939 abort_with_mtrr:
3940 #ifdef CONFIG_MTRR
3941 if (mgp->mtrr >= 0)
3942 mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span);
3943 #endif
3944 dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
3945 mgp->cmd, mgp->cmd_bus);
3946
3947 abort_with_enabled:
3948 pci_disable_device(pdev);
3949
3950 abort_with_netdev:
3951 free_netdev(netdev);
3952 return status;
3953 }
3954
3955 /*
3956 * myri10ge_remove
3957 *
3958 * Does what is necessary to shutdown one Myrinet device. Called
3959 * once for each Myrinet card by the kernel when a module is
3960 * unloaded.
3961 */
3962 static void myri10ge_remove(struct pci_dev *pdev)
3963 {
3964 struct myri10ge_priv *mgp;
3965 struct net_device *netdev;
3966
3967 mgp = pci_get_drvdata(pdev);
3968 if (mgp == NULL)
3969 return;
3970
3971 flush_scheduled_work();
3972 netdev = mgp->dev;
3973 unregister_netdev(netdev);
3974
3975 #ifdef CONFIG_MYRI10GE_DCA
3976 myri10ge_teardown_dca(mgp);
3977 #endif
3978 myri10ge_dummy_rdma(mgp, 0);
3979
3980 /* avoid a memory leak */
3981 pci_restore_state(pdev);
3982
3983 iounmap(mgp->sram);
3984
3985 #ifdef CONFIG_MTRR
3986 if (mgp->mtrr >= 0)
3987 mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span);
3988 #endif
3989 myri10ge_free_slices(mgp);
3990 if (mgp->msix_vectors != NULL)
3991 kfree(mgp->msix_vectors);
3992 dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
3993 mgp->cmd, mgp->cmd_bus);
3994
3995 free_netdev(netdev);
3996 pci_disable_device(pdev);
3997 pci_set_drvdata(pdev, NULL);
3998 }
3999
4000 #define PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E 0x0008
4001 #define PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E_9 0x0009
4002
4003 static struct pci_device_id myri10ge_pci_tbl[] = {
4004 {PCI_DEVICE(PCI_VENDOR_ID_MYRICOM, PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E)},
4005 {PCI_DEVICE
4006 (PCI_VENDOR_ID_MYRICOM, PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E_9)},
4007 {0},
4008 };
4009
4010 static struct pci_driver myri10ge_driver = {
4011 .name = "myri10ge",
4012 .probe = myri10ge_probe,
4013 .remove = myri10ge_remove,
4014 .id_table = myri10ge_pci_tbl,
4015 #ifdef CONFIG_PM
4016 .suspend = myri10ge_suspend,
4017 .resume = myri10ge_resume,
4018 #endif
4019 };
4020
4021 #ifdef CONFIG_MYRI10GE_DCA
4022 static int
4023 myri10ge_notify_dca(struct notifier_block *nb, unsigned long event, void *p)
4024 {
4025 int err = driver_for_each_device(&myri10ge_driver.driver,
4026 NULL, &event,
4027 myri10ge_notify_dca_device);
4028
4029 if (err)
4030 return NOTIFY_BAD;
4031 return NOTIFY_DONE;
4032 }
4033
4034 static struct notifier_block myri10ge_dca_notifier = {
4035 .notifier_call = myri10ge_notify_dca,
4036 .next = NULL,
4037 .priority = 0,
4038 };
4039 #endif /* CONFIG_MYRI10GE_DCA */
4040
4041 static __init int myri10ge_init_module(void)
4042 {
4043 printk(KERN_INFO "%s: Version %s\n", myri10ge_driver.name,
4044 MYRI10GE_VERSION_STR);
4045
4046 if (myri10ge_rss_hash > MXGEFW_RSS_HASH_TYPE_MAX) {
4047 printk(KERN_ERR
4048 "%s: Illegal rssh hash type %d, defaulting to source port\n",
4049 myri10ge_driver.name, myri10ge_rss_hash);
4050 myri10ge_rss_hash = MXGEFW_RSS_HASH_TYPE_SRC_PORT;
4051 }
4052 #ifdef CONFIG_MYRI10GE_DCA
4053 dca_register_notify(&myri10ge_dca_notifier);
4054 #endif
4055 if (myri10ge_max_slices > MYRI10GE_MAX_SLICES)
4056 myri10ge_max_slices = MYRI10GE_MAX_SLICES;
4057
4058 return pci_register_driver(&myri10ge_driver);
4059 }
4060
4061 module_init(myri10ge_init_module);
4062
4063 static __exit void myri10ge_cleanup_module(void)
4064 {
4065 #ifdef CONFIG_MYRI10GE_DCA
4066 dca_unregister_notify(&myri10ge_dca_notifier);
4067 #endif
4068 pci_unregister_driver(&myri10ge_driver);
4069 }
4070
4071 module_exit(myri10ge_cleanup_module);
Nao low-level kernelspace/userspace library that runs on our robots, Nao-Team HTWK