[PATCH] myri10ge annotations
[linux-2.6/kvm.git] / drivers / net / myri10ge / myri10ge.c
blobd320bbeeeff6f2019cf3e259c80d4127d82129e7
1 /*************************************************************************
2 * myri10ge.c: Myricom Myri-10G Ethernet driver.
4 * Copyright (C) 2005, 2006 Myricom, Inc.
5 * All rights reserved.
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.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
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
36 * Contact Information:
37 * <help@myri.com>
38 * Myricom, Inc., 325N Santa Anita Avenue, Arcadia, CA 91006
39 *************************************************************************/
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/ip.h>
52 #include <linux/inet.h>
53 #include <linux/in.h>
54 #include <linux/ethtool.h>
55 #include <linux/firmware.h>
56 #include <linux/delay.h>
57 #include <linux/version.h>
58 #include <linux/timer.h>
59 #include <linux/vmalloc.h>
60 #include <linux/crc32.h>
61 #include <linux/moduleparam.h>
62 #include <linux/io.h>
63 #include <net/checksum.h>
64 #include <asm/byteorder.h>
65 #include <asm/io.h>
66 #include <asm/processor.h>
67 #ifdef CONFIG_MTRR
68 #include <asm/mtrr.h>
69 #endif
71 #include "myri10ge_mcp.h"
72 #include "myri10ge_mcp_gen_header.h"
74 #define MYRI10GE_VERSION_STR "1.0.0"
76 MODULE_DESCRIPTION("Myricom 10G driver (10GbE)");
77 MODULE_AUTHOR("Maintainer: help@myri.com");
78 MODULE_VERSION(MYRI10GE_VERSION_STR);
79 MODULE_LICENSE("Dual BSD/GPL");
81 #define MYRI10GE_MAX_ETHER_MTU 9014
83 #define MYRI10GE_ETH_STOPPED 0
84 #define MYRI10GE_ETH_STOPPING 1
85 #define MYRI10GE_ETH_STARTING 2
86 #define MYRI10GE_ETH_RUNNING 3
87 #define MYRI10GE_ETH_OPEN_FAILED 4
89 #define MYRI10GE_EEPROM_STRINGS_SIZE 256
90 #define MYRI10GE_MAX_SEND_DESC_TSO ((65536 / 2048) * 2)
92 #define MYRI10GE_NO_CONFIRM_DATA htonl(0xffffffff)
93 #define MYRI10GE_NO_RESPONSE_RESULT 0xffffffff
95 struct myri10ge_rx_buffer_state {
96 struct sk_buff *skb;
97 DECLARE_PCI_UNMAP_ADDR(bus)
98 DECLARE_PCI_UNMAP_LEN(len)
101 struct myri10ge_tx_buffer_state {
102 struct sk_buff *skb;
103 int last;
104 DECLARE_PCI_UNMAP_ADDR(bus)
105 DECLARE_PCI_UNMAP_LEN(len)
108 struct myri10ge_cmd {
109 u32 data0;
110 u32 data1;
111 u32 data2;
114 struct myri10ge_rx_buf {
115 struct mcp_kreq_ether_recv __iomem *lanai; /* lanai ptr for recv ring */
116 u8 __iomem *wc_fifo; /* w/c rx dma addr fifo address */
117 struct mcp_kreq_ether_recv *shadow; /* host shadow of recv ring */
118 struct myri10ge_rx_buffer_state *info;
119 int cnt;
120 int alloc_fail;
121 int mask; /* number of rx slots -1 */
124 struct myri10ge_tx_buf {
125 struct mcp_kreq_ether_send __iomem *lanai; /* lanai ptr for sendq */
126 u8 __iomem *wc_fifo; /* w/c send fifo address */
127 struct mcp_kreq_ether_send *req_list; /* host shadow of sendq */
128 char *req_bytes;
129 struct myri10ge_tx_buffer_state *info;
130 int mask; /* number of transmit slots -1 */
131 int boundary; /* boundary transmits cannot cross */
132 int req ____cacheline_aligned; /* transmit slots submitted */
133 int pkt_start; /* packets started */
134 int done ____cacheline_aligned; /* transmit slots completed */
135 int pkt_done; /* packets completed */
138 struct myri10ge_rx_done {
139 struct mcp_slot *entry;
140 dma_addr_t bus;
141 int cnt;
142 int idx;
145 struct myri10ge_priv {
146 int running; /* running? */
147 int csum_flag; /* rx_csums? */
148 struct myri10ge_tx_buf tx; /* transmit ring */
149 struct myri10ge_rx_buf rx_small;
150 struct myri10ge_rx_buf rx_big;
151 struct myri10ge_rx_done rx_done;
152 int small_bytes;
153 struct net_device *dev;
154 struct net_device_stats stats;
155 u8 __iomem *sram;
156 int sram_size;
157 unsigned long board_span;
158 unsigned long iomem_base;
159 __be32 __iomem *irq_claim;
160 __be32 __iomem *irq_deassert;
161 char *mac_addr_string;
162 struct mcp_cmd_response *cmd;
163 dma_addr_t cmd_bus;
164 struct mcp_irq_data *fw_stats;
165 dma_addr_t fw_stats_bus;
166 struct pci_dev *pdev;
167 int msi_enabled;
168 __be32 link_state;
169 unsigned int rdma_tags_available;
170 int intr_coal_delay;
171 __be32 __iomem *intr_coal_delay_ptr;
172 int mtrr;
173 int wake_queue;
174 int stop_queue;
175 int down_cnt;
176 wait_queue_head_t down_wq;
177 struct work_struct watchdog_work;
178 struct timer_list watchdog_timer;
179 int watchdog_tx_done;
180 int watchdog_tx_req;
181 int watchdog_resets;
182 int tx_linearized;
183 int pause;
184 char *fw_name;
185 char eeprom_strings[MYRI10GE_EEPROM_STRINGS_SIZE];
186 char fw_version[128];
187 u8 mac_addr[6]; /* eeprom mac address */
188 unsigned long serial_number;
189 int vendor_specific_offset;
190 int fw_multicast_support;
191 u32 devctl;
192 u16 msi_flags;
193 u32 read_dma;
194 u32 write_dma;
195 u32 read_write_dma;
196 u32 link_changes;
197 u32 msg_enable;
200 static char *myri10ge_fw_unaligned = "myri10ge_ethp_z8e.dat";
201 static char *myri10ge_fw_aligned = "myri10ge_eth_z8e.dat";
203 static char *myri10ge_fw_name = NULL;
204 module_param(myri10ge_fw_name, charp, S_IRUGO | S_IWUSR);
205 MODULE_PARM_DESC(myri10ge_fw_name, "Firmware image name\n");
207 static int myri10ge_ecrc_enable = 1;
208 module_param(myri10ge_ecrc_enable, int, S_IRUGO);
209 MODULE_PARM_DESC(myri10ge_ecrc_enable, "Enable Extended CRC on PCI-E\n");
211 static int myri10ge_max_intr_slots = 1024;
212 module_param(myri10ge_max_intr_slots, int, S_IRUGO);
213 MODULE_PARM_DESC(myri10ge_max_intr_slots, "Interrupt queue slots\n");
215 static int myri10ge_small_bytes = -1; /* -1 == auto */
216 module_param(myri10ge_small_bytes, int, S_IRUGO | S_IWUSR);
217 MODULE_PARM_DESC(myri10ge_small_bytes, "Threshold of small packets\n");
219 static int myri10ge_msi = 1; /* enable msi by default */
220 module_param(myri10ge_msi, int, S_IRUGO);
221 MODULE_PARM_DESC(myri10ge_msi, "Enable Message Signalled Interrupts\n");
223 static int myri10ge_intr_coal_delay = 25;
224 module_param(myri10ge_intr_coal_delay, int, S_IRUGO);
225 MODULE_PARM_DESC(myri10ge_intr_coal_delay, "Interrupt coalescing delay\n");
227 static int myri10ge_flow_control = 1;
228 module_param(myri10ge_flow_control, int, S_IRUGO);
229 MODULE_PARM_DESC(myri10ge_flow_control, "Pause parameter\n");
231 static int myri10ge_deassert_wait = 1;
232 module_param(myri10ge_deassert_wait, int, S_IRUGO | S_IWUSR);
233 MODULE_PARM_DESC(myri10ge_deassert_wait,
234 "Wait when deasserting legacy interrupts\n");
236 static int myri10ge_force_firmware = 0;
237 module_param(myri10ge_force_firmware, int, S_IRUGO);
238 MODULE_PARM_DESC(myri10ge_force_firmware,
239 "Force firmware to assume aligned completions\n");
241 static int myri10ge_skb_cross_4k = 0;
242 module_param(myri10ge_skb_cross_4k, int, S_IRUGO | S_IWUSR);
243 MODULE_PARM_DESC(myri10ge_skb_cross_4k,
244 "Can a small skb cross a 4KB boundary?\n");
246 static int myri10ge_initial_mtu = MYRI10GE_MAX_ETHER_MTU - ETH_HLEN;
247 module_param(myri10ge_initial_mtu, int, S_IRUGO);
248 MODULE_PARM_DESC(myri10ge_initial_mtu, "Initial MTU\n");
250 static int myri10ge_napi_weight = 64;
251 module_param(myri10ge_napi_weight, int, S_IRUGO);
252 MODULE_PARM_DESC(myri10ge_napi_weight, "Set NAPI weight\n");
254 static int myri10ge_watchdog_timeout = 1;
255 module_param(myri10ge_watchdog_timeout, int, S_IRUGO);
256 MODULE_PARM_DESC(myri10ge_watchdog_timeout, "Set watchdog timeout\n");
258 static int myri10ge_max_irq_loops = 1048576;
259 module_param(myri10ge_max_irq_loops, int, S_IRUGO);
260 MODULE_PARM_DESC(myri10ge_max_irq_loops,
261 "Set stuck legacy IRQ detection threshold\n");
263 #define MYRI10GE_MSG_DEFAULT NETIF_MSG_LINK
265 static int myri10ge_debug = -1; /* defaults above */
266 module_param(myri10ge_debug, int, 0);
267 MODULE_PARM_DESC(myri10ge_debug, "Debug level (0=none,...,16=all)");
269 #define MYRI10GE_FW_OFFSET 1024*1024
270 #define MYRI10GE_HIGHPART_TO_U32(X) \
271 (sizeof (X) == 8) ? ((u32)((u64)(X) >> 32)) : (0)
272 #define MYRI10GE_LOWPART_TO_U32(X) ((u32)(X))
274 #define myri10ge_pio_copy(to,from,size) __iowrite64_copy(to,from,size/8)
276 static inline void put_be32(__be32 val, __be32 __iomem *p)
278 __raw_writel((__force __u32)val, (__force void __iomem *)p);
281 static int
282 myri10ge_send_cmd(struct myri10ge_priv *mgp, u32 cmd,
283 struct myri10ge_cmd *data, int atomic)
285 struct mcp_cmd *buf;
286 char buf_bytes[sizeof(*buf) + 8];
287 struct mcp_cmd_response *response = mgp->cmd;
288 char __iomem *cmd_addr = mgp->sram + MXGEFW_ETH_CMD;
289 u32 dma_low, dma_high, result, value;
290 int sleep_total = 0;
292 /* ensure buf is aligned to 8 bytes */
293 buf = (struct mcp_cmd *)ALIGN((unsigned long)buf_bytes, 8);
295 buf->data0 = htonl(data->data0);
296 buf->data1 = htonl(data->data1);
297 buf->data2 = htonl(data->data2);
298 buf->cmd = htonl(cmd);
299 dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
300 dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
302 buf->response_addr.low = htonl(dma_low);
303 buf->response_addr.high = htonl(dma_high);
304 response->result = htonl(MYRI10GE_NO_RESPONSE_RESULT);
305 mb();
306 myri10ge_pio_copy(cmd_addr, buf, sizeof(*buf));
308 /* wait up to 15ms. Longest command is the DMA benchmark,
309 * which is capped at 5ms, but runs from a timeout handler
310 * that runs every 7.8ms. So a 15ms timeout leaves us with
311 * a 2.2ms margin
313 if (atomic) {
314 /* if atomic is set, do not sleep,
315 * and try to get the completion quickly
316 * (1ms will be enough for those commands) */
317 for (sleep_total = 0;
318 sleep_total < 1000
319 && response->result == htonl(MYRI10GE_NO_RESPONSE_RESULT);
320 sleep_total += 10)
321 udelay(10);
322 } else {
323 /* use msleep for most command */
324 for (sleep_total = 0;
325 sleep_total < 15
326 && response->result == htonl(MYRI10GE_NO_RESPONSE_RESULT);
327 sleep_total++)
328 msleep(1);
331 result = ntohl(response->result);
332 value = ntohl(response->data);
333 if (result != MYRI10GE_NO_RESPONSE_RESULT) {
334 if (result == 0) {
335 data->data0 = value;
336 return 0;
337 } else if (result == MXGEFW_CMD_UNKNOWN) {
338 return -ENOSYS;
339 } else {
340 dev_err(&mgp->pdev->dev,
341 "command %d failed, result = %d\n",
342 cmd, result);
343 return -ENXIO;
347 dev_err(&mgp->pdev->dev, "command %d timed out, result = %d\n",
348 cmd, result);
349 return -EAGAIN;
353 * The eeprom strings on the lanaiX have the format
354 * SN=x\0
355 * MAC=x:x:x:x:x:x\0
356 * PT:ddd mmm xx xx:xx:xx xx\0
357 * PV:ddd mmm xx xx:xx:xx xx\0
359 static int myri10ge_read_mac_addr(struct myri10ge_priv *mgp)
361 char *ptr, *limit;
362 int i;
364 ptr = mgp->eeprom_strings;
365 limit = mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE;
367 while (*ptr != '\0' && ptr < limit) {
368 if (memcmp(ptr, "MAC=", 4) == 0) {
369 ptr += 4;
370 mgp->mac_addr_string = ptr;
371 for (i = 0; i < 6; i++) {
372 if ((ptr + 2) > limit)
373 goto abort;
374 mgp->mac_addr[i] =
375 simple_strtoul(ptr, &ptr, 16);
376 ptr += 1;
379 if (memcmp((const void *)ptr, "SN=", 3) == 0) {
380 ptr += 3;
381 mgp->serial_number = simple_strtoul(ptr, &ptr, 10);
383 while (ptr < limit && *ptr++) ;
386 return 0;
388 abort:
389 dev_err(&mgp->pdev->dev, "failed to parse eeprom_strings\n");
390 return -ENXIO;
394 * Enable or disable periodic RDMAs from the host to make certain
395 * chipsets resend dropped PCIe messages
398 static void myri10ge_dummy_rdma(struct myri10ge_priv *mgp, int enable)
400 char __iomem *submit;
401 __be32 buf[16];
402 u32 dma_low, dma_high;
403 int i;
405 /* clear confirmation addr */
406 mgp->cmd->data = 0;
407 mb();
409 /* send a rdma command to the PCIe engine, and wait for the
410 * response in the confirmation address. The firmware should
411 * write a -1 there to indicate it is alive and well
413 dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
414 dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
416 buf[0] = htonl(dma_high); /* confirm addr MSW */
417 buf[1] = htonl(dma_low); /* confirm addr LSW */
418 buf[2] = MYRI10GE_NO_CONFIRM_DATA; /* confirm data */
419 buf[3] = htonl(dma_high); /* dummy addr MSW */
420 buf[4] = htonl(dma_low); /* dummy addr LSW */
421 buf[5] = htonl(enable); /* enable? */
423 submit = mgp->sram + MXGEFW_BOOT_DUMMY_RDMA;
425 myri10ge_pio_copy(submit, &buf, sizeof(buf));
426 for (i = 0; mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 20; i++)
427 msleep(1);
428 if (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA)
429 dev_err(&mgp->pdev->dev, "dummy rdma %s failed\n",
430 (enable ? "enable" : "disable"));
433 static int
434 myri10ge_validate_firmware(struct myri10ge_priv *mgp,
435 struct mcp_gen_header *hdr)
437 struct device *dev = &mgp->pdev->dev;
438 int major, minor;
440 /* check firmware type */
441 if (ntohl(hdr->mcp_type) != MCP_TYPE_ETH) {
442 dev_err(dev, "Bad firmware type: 0x%x\n", ntohl(hdr->mcp_type));
443 return -EINVAL;
446 /* save firmware version for ethtool */
447 strncpy(mgp->fw_version, hdr->version, sizeof(mgp->fw_version));
449 sscanf(mgp->fw_version, "%d.%d", &major, &minor);
451 if (!(major == MXGEFW_VERSION_MAJOR && minor == MXGEFW_VERSION_MINOR)) {
452 dev_err(dev, "Found firmware version %s\n", mgp->fw_version);
453 dev_err(dev, "Driver needs %d.%d\n", MXGEFW_VERSION_MAJOR,
454 MXGEFW_VERSION_MINOR);
455 return -EINVAL;
457 return 0;
460 static int myri10ge_load_hotplug_firmware(struct myri10ge_priv *mgp, u32 * size)
462 unsigned crc, reread_crc;
463 const struct firmware *fw;
464 struct device *dev = &mgp->pdev->dev;
465 struct mcp_gen_header *hdr;
466 size_t hdr_offset;
467 int status;
468 unsigned i;
470 if ((status = request_firmware(&fw, mgp->fw_name, dev)) < 0) {
471 dev_err(dev, "Unable to load %s firmware image via hotplug\n",
472 mgp->fw_name);
473 status = -EINVAL;
474 goto abort_with_nothing;
477 /* check size */
479 if (fw->size >= mgp->sram_size - MYRI10GE_FW_OFFSET ||
480 fw->size < MCP_HEADER_PTR_OFFSET + 4) {
481 dev_err(dev, "Firmware size invalid:%d\n", (int)fw->size);
482 status = -EINVAL;
483 goto abort_with_fw;
486 /* check id */
487 hdr_offset = ntohl(*(__be32 *) (fw->data + MCP_HEADER_PTR_OFFSET));
488 if ((hdr_offset & 3) || hdr_offset + sizeof(*hdr) > fw->size) {
489 dev_err(dev, "Bad firmware file\n");
490 status = -EINVAL;
491 goto abort_with_fw;
493 hdr = (void *)(fw->data + hdr_offset);
495 status = myri10ge_validate_firmware(mgp, hdr);
496 if (status != 0)
497 goto abort_with_fw;
499 crc = crc32(~0, fw->data, fw->size);
500 for (i = 0; i < fw->size; i += 256) {
501 myri10ge_pio_copy(mgp->sram + MYRI10GE_FW_OFFSET + i,
502 fw->data + i,
503 min(256U, (unsigned)(fw->size - i)));
504 mb();
505 readb(mgp->sram);
507 /* corruption checking is good for parity recovery and buggy chipset */
508 memcpy_fromio(fw->data, mgp->sram + MYRI10GE_FW_OFFSET, fw->size);
509 reread_crc = crc32(~0, fw->data, fw->size);
510 if (crc != reread_crc) {
511 dev_err(dev, "CRC failed(fw-len=%u), got 0x%x (expect 0x%x)\n",
512 (unsigned)fw->size, reread_crc, crc);
513 status = -EIO;
514 goto abort_with_fw;
516 *size = (u32) fw->size;
518 abort_with_fw:
519 release_firmware(fw);
521 abort_with_nothing:
522 return status;
525 static int myri10ge_adopt_running_firmware(struct myri10ge_priv *mgp)
527 struct mcp_gen_header *hdr;
528 struct device *dev = &mgp->pdev->dev;
529 const size_t bytes = sizeof(struct mcp_gen_header);
530 size_t hdr_offset;
531 int status;
533 /* find running firmware header */
534 hdr_offset = ntohl(__raw_readl(mgp->sram + MCP_HEADER_PTR_OFFSET));
536 if ((hdr_offset & 3) || hdr_offset + sizeof(*hdr) > mgp->sram_size) {
537 dev_err(dev, "Running firmware has bad header offset (%d)\n",
538 (int)hdr_offset);
539 return -EIO;
542 /* copy header of running firmware from SRAM to host memory to
543 * validate firmware */
544 hdr = kmalloc(bytes, GFP_KERNEL);
545 if (hdr == NULL) {
546 dev_err(dev, "could not malloc firmware hdr\n");
547 return -ENOMEM;
549 memcpy_fromio(hdr, mgp->sram + hdr_offset, bytes);
550 status = myri10ge_validate_firmware(mgp, hdr);
551 kfree(hdr);
552 return status;
555 static int myri10ge_load_firmware(struct myri10ge_priv *mgp)
557 char __iomem *submit;
558 __be32 buf[16];
559 u32 dma_low, dma_high, size;
560 int status, i;
562 size = 0;
563 status = myri10ge_load_hotplug_firmware(mgp, &size);
564 if (status) {
565 dev_warn(&mgp->pdev->dev, "hotplug firmware loading failed\n");
567 /* Do not attempt to adopt firmware if there
568 * was a bad crc */
569 if (status == -EIO)
570 return status;
572 status = myri10ge_adopt_running_firmware(mgp);
573 if (status != 0) {
574 dev_err(&mgp->pdev->dev,
575 "failed to adopt running firmware\n");
576 return status;
578 dev_info(&mgp->pdev->dev,
579 "Successfully adopted running firmware\n");
580 if (mgp->tx.boundary == 4096) {
581 dev_warn(&mgp->pdev->dev,
582 "Using firmware currently running on NIC"
583 ". For optimal\n");
584 dev_warn(&mgp->pdev->dev,
585 "performance consider loading optimized "
586 "firmware\n");
587 dev_warn(&mgp->pdev->dev, "via hotplug\n");
590 mgp->fw_name = "adopted";
591 mgp->tx.boundary = 2048;
592 return status;
595 /* clear confirmation addr */
596 mgp->cmd->data = 0;
597 mb();
599 /* send a reload command to the bootstrap MCP, and wait for the
600 * response in the confirmation address. The firmware should
601 * write a -1 there to indicate it is alive and well
603 dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
604 dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
606 buf[0] = htonl(dma_high); /* confirm addr MSW */
607 buf[1] = htonl(dma_low); /* confirm addr LSW */
608 buf[2] = MYRI10GE_NO_CONFIRM_DATA; /* confirm data */
610 /* FIX: All newest firmware should un-protect the bottom of
611 * the sram before handoff. However, the very first interfaces
612 * do not. Therefore the handoff copy must skip the first 8 bytes
614 buf[3] = htonl(MYRI10GE_FW_OFFSET + 8); /* where the code starts */
615 buf[4] = htonl(size - 8); /* length of code */
616 buf[5] = htonl(8); /* where to copy to */
617 buf[6] = htonl(0); /* where to jump to */
619 submit = mgp->sram + MXGEFW_BOOT_HANDOFF;
621 myri10ge_pio_copy(submit, &buf, sizeof(buf));
622 mb();
623 msleep(1);
624 mb();
625 i = 0;
626 while (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 20) {
627 msleep(1);
628 i++;
630 if (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA) {
631 dev_err(&mgp->pdev->dev, "handoff failed\n");
632 return -ENXIO;
634 dev_info(&mgp->pdev->dev, "handoff confirmed\n");
635 myri10ge_dummy_rdma(mgp, 1);
637 return 0;
640 static int myri10ge_update_mac_address(struct myri10ge_priv *mgp, u8 * addr)
642 struct myri10ge_cmd cmd;
643 int status;
645 cmd.data0 = ((addr[0] << 24) | (addr[1] << 16)
646 | (addr[2] << 8) | addr[3]);
648 cmd.data1 = ((addr[4] << 8) | (addr[5]));
650 status = myri10ge_send_cmd(mgp, MXGEFW_SET_MAC_ADDRESS, &cmd, 0);
651 return status;
654 static int myri10ge_change_pause(struct myri10ge_priv *mgp, int pause)
656 struct myri10ge_cmd cmd;
657 int status, ctl;
659 ctl = pause ? MXGEFW_ENABLE_FLOW_CONTROL : MXGEFW_DISABLE_FLOW_CONTROL;
660 status = myri10ge_send_cmd(mgp, ctl, &cmd, 0);
662 if (status) {
663 printk(KERN_ERR
664 "myri10ge: %s: Failed to set flow control mode\n",
665 mgp->dev->name);
666 return status;
668 mgp->pause = pause;
669 return 0;
672 static void
673 myri10ge_change_promisc(struct myri10ge_priv *mgp, int promisc, int atomic)
675 struct myri10ge_cmd cmd;
676 int status, ctl;
678 ctl = promisc ? MXGEFW_ENABLE_PROMISC : MXGEFW_DISABLE_PROMISC;
679 status = myri10ge_send_cmd(mgp, ctl, &cmd, atomic);
680 if (status)
681 printk(KERN_ERR "myri10ge: %s: Failed to set promisc mode\n",
682 mgp->dev->name);
685 static int myri10ge_reset(struct myri10ge_priv *mgp)
687 struct myri10ge_cmd cmd;
688 int status;
689 size_t bytes;
690 u32 len;
692 /* try to send a reset command to the card to see if it
693 * is alive */
694 memset(&cmd, 0, sizeof(cmd));
695 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_RESET, &cmd, 0);
696 if (status != 0) {
697 dev_err(&mgp->pdev->dev, "failed reset\n");
698 return -ENXIO;
701 /* Now exchange information about interrupts */
703 bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
704 memset(mgp->rx_done.entry, 0, bytes);
705 cmd.data0 = (u32) bytes;
706 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_SIZE, &cmd, 0);
707 cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus);
708 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus);
709 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_DMA, &cmd, 0);
711 status |=
712 myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_IRQ_ACK_OFFSET, &cmd, 0);
713 mgp->irq_claim = (__iomem __be32 *) (mgp->sram + cmd.data0);
714 if (!mgp->msi_enabled) {
715 status |= myri10ge_send_cmd
716 (mgp, MXGEFW_CMD_GET_IRQ_DEASSERT_OFFSET, &cmd, 0);
717 mgp->irq_deassert = (__iomem __be32 *) (mgp->sram + cmd.data0);
720 status |= myri10ge_send_cmd
721 (mgp, MXGEFW_CMD_GET_INTR_COAL_DELAY_OFFSET, &cmd, 0);
722 mgp->intr_coal_delay_ptr = (__iomem __be32 *) (mgp->sram + cmd.data0);
723 if (status != 0) {
724 dev_err(&mgp->pdev->dev, "failed set interrupt parameters\n");
725 return status;
727 put_be32(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
729 /* Run a small DMA test.
730 * The magic multipliers to the length tell the firmware
731 * to do DMA read, write, or read+write tests. The
732 * results are returned in cmd.data0. The upper 16
733 * bits or the return is the number of transfers completed.
734 * The lower 16 bits is the time in 0.5us ticks that the
735 * transfers took to complete.
738 len = mgp->tx.boundary;
740 cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus);
741 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus);
742 cmd.data2 = len * 0x10000;
743 status = myri10ge_send_cmd(mgp, MXGEFW_DMA_TEST, &cmd, 0);
744 if (status == 0)
745 mgp->read_dma = ((cmd.data0 >> 16) * len * 2) /
746 (cmd.data0 & 0xffff);
747 else
748 dev_warn(&mgp->pdev->dev, "DMA read benchmark failed: %d\n",
749 status);
750 cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus);
751 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus);
752 cmd.data2 = len * 0x1;
753 status = myri10ge_send_cmd(mgp, MXGEFW_DMA_TEST, &cmd, 0);
754 if (status == 0)
755 mgp->write_dma = ((cmd.data0 >> 16) * len * 2) /
756 (cmd.data0 & 0xffff);
757 else
758 dev_warn(&mgp->pdev->dev, "DMA write benchmark failed: %d\n",
759 status);
761 cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->rx_done.bus);
762 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->rx_done.bus);
763 cmd.data2 = len * 0x10001;
764 status = myri10ge_send_cmd(mgp, MXGEFW_DMA_TEST, &cmd, 0);
765 if (status == 0)
766 mgp->read_write_dma = ((cmd.data0 >> 16) * len * 2 * 2) /
767 (cmd.data0 & 0xffff);
768 else
769 dev_warn(&mgp->pdev->dev,
770 "DMA read/write benchmark failed: %d\n", status);
772 memset(mgp->rx_done.entry, 0, bytes);
774 /* reset mcp/driver shared state back to 0 */
775 mgp->tx.req = 0;
776 mgp->tx.done = 0;
777 mgp->tx.pkt_start = 0;
778 mgp->tx.pkt_done = 0;
779 mgp->rx_big.cnt = 0;
780 mgp->rx_small.cnt = 0;
781 mgp->rx_done.idx = 0;
782 mgp->rx_done.cnt = 0;
783 mgp->link_changes = 0;
784 status = myri10ge_update_mac_address(mgp, mgp->dev->dev_addr);
785 myri10ge_change_promisc(mgp, 0, 0);
786 myri10ge_change_pause(mgp, mgp->pause);
787 return status;
790 static inline void
791 myri10ge_submit_8rx(struct mcp_kreq_ether_recv __iomem * dst,
792 struct mcp_kreq_ether_recv *src)
794 __be32 low;
796 low = src->addr_low;
797 src->addr_low = htonl(DMA_32BIT_MASK);
798 myri10ge_pio_copy(dst, src, 8 * sizeof(*src));
799 mb();
800 src->addr_low = low;
801 put_be32(low, &dst->addr_low);
802 mb();
806 * Set of routines to get a new receive buffer. Any buffer which
807 * crosses a 4KB boundary must start on a 4KB boundary due to PCIe
808 * wdma restrictions. We also try to align any smaller allocation to
809 * at least a 16 byte boundary for efficiency. We assume the linux
810 * memory allocator works by powers of 2, and will not return memory
811 * smaller than 2KB which crosses a 4KB boundary. If it does, we fall
812 * back to allocating 2x as much space as required.
814 * We intend to replace large (>4KB) skb allocations by using
815 * pages directly and building a fraglist in the near future.
818 static inline struct sk_buff *myri10ge_alloc_big(struct net_device *dev,
819 int bytes)
821 struct sk_buff *skb;
822 unsigned long data, roundup;
824 skb = netdev_alloc_skb(dev, bytes + 4096 + MXGEFW_PAD);
825 if (skb == NULL)
826 return NULL;
828 /* Correct skb->truesize so that socket buffer
829 * accounting is not confused the rounding we must
830 * do to satisfy alignment constraints.
832 skb->truesize -= 4096;
834 data = (unsigned long)(skb->data);
835 roundup = (-data) & (4095);
836 skb_reserve(skb, roundup);
837 return skb;
840 /* Allocate 2x as much space as required and use whichever portion
841 * does not cross a 4KB boundary */
842 static inline struct sk_buff *myri10ge_alloc_small_safe(struct net_device *dev,
843 unsigned int bytes)
845 struct sk_buff *skb;
846 unsigned long data, boundary;
848 skb = netdev_alloc_skb(dev, 2 * (bytes + MXGEFW_PAD) - 1);
849 if (unlikely(skb == NULL))
850 return NULL;
852 /* Correct skb->truesize so that socket buffer
853 * accounting is not confused the rounding we must
854 * do to satisfy alignment constraints.
856 skb->truesize -= bytes + MXGEFW_PAD;
858 data = (unsigned long)(skb->data);
859 boundary = (data + 4095UL) & ~4095UL;
860 if ((boundary - data) >= (bytes + MXGEFW_PAD))
861 return skb;
863 skb_reserve(skb, boundary - data);
864 return skb;
867 /* Allocate just enough space, and verify that the allocated
868 * space does not cross a 4KB boundary */
869 static inline struct sk_buff *myri10ge_alloc_small(struct net_device *dev,
870 int bytes)
872 struct sk_buff *skb;
873 unsigned long roundup, data, end;
875 skb = netdev_alloc_skb(dev, bytes + 16 + MXGEFW_PAD);
876 if (unlikely(skb == NULL))
877 return NULL;
879 /* Round allocated buffer to 16 byte boundary */
880 data = (unsigned long)(skb->data);
881 roundup = (-data) & 15UL;
882 skb_reserve(skb, roundup);
883 /* Verify that the data buffer does not cross a page boundary */
884 data = (unsigned long)(skb->data);
885 end = data + bytes + MXGEFW_PAD - 1;
886 if (unlikely(((end >> 12) != (data >> 12)) && (data & 4095UL))) {
887 printk(KERN_NOTICE
888 "myri10ge_alloc_small: small skb crossed 4KB boundary\n");
889 myri10ge_skb_cross_4k = 1;
890 dev_kfree_skb_any(skb);
891 skb = myri10ge_alloc_small_safe(dev, bytes);
893 return skb;
896 static inline int
897 myri10ge_getbuf(struct myri10ge_rx_buf *rx, struct myri10ge_priv *mgp,
898 int bytes, int idx)
900 struct net_device *dev = mgp->dev;
901 struct pci_dev *pdev = mgp->pdev;
902 struct sk_buff *skb;
903 dma_addr_t bus;
904 int len, retval = 0;
906 bytes += VLAN_HLEN; /* account for 802.1q vlan tag */
908 if ((bytes + MXGEFW_PAD) > (4096 - 16) /* linux overhead */ )
909 skb = myri10ge_alloc_big(dev, bytes);
910 else if (myri10ge_skb_cross_4k)
911 skb = myri10ge_alloc_small_safe(dev, bytes);
912 else
913 skb = myri10ge_alloc_small(dev, bytes);
915 if (unlikely(skb == NULL)) {
916 rx->alloc_fail++;
917 retval = -ENOBUFS;
918 goto done;
921 /* set len so that it only covers the area we
922 * need mapped for DMA */
923 len = bytes + MXGEFW_PAD;
925 bus = pci_map_single(pdev, skb->data, len, PCI_DMA_FROMDEVICE);
926 rx->info[idx].skb = skb;
927 pci_unmap_addr_set(&rx->info[idx], bus, bus);
928 pci_unmap_len_set(&rx->info[idx], len, len);
929 rx->shadow[idx].addr_low = htonl(MYRI10GE_LOWPART_TO_U32(bus));
930 rx->shadow[idx].addr_high = htonl(MYRI10GE_HIGHPART_TO_U32(bus));
932 done:
933 /* copy 8 descriptors (64-bytes) to the mcp at a time */
934 if ((idx & 7) == 7) {
935 if (rx->wc_fifo == NULL)
936 myri10ge_submit_8rx(&rx->lanai[idx - 7],
937 &rx->shadow[idx - 7]);
938 else {
939 mb();
940 myri10ge_pio_copy(rx->wc_fifo,
941 &rx->shadow[idx - 7], 64);
944 return retval;
947 static inline void myri10ge_vlan_ip_csum(struct sk_buff *skb, __wsum hw_csum)
949 struct vlan_hdr *vh = (struct vlan_hdr *)(skb->data);
951 if ((skb->protocol == htons(ETH_P_8021Q)) &&
952 (vh->h_vlan_encapsulated_proto == htons(ETH_P_IP) ||
953 vh->h_vlan_encapsulated_proto == htons(ETH_P_IPV6))) {
954 skb->csum = hw_csum;
955 skb->ip_summed = CHECKSUM_COMPLETE;
959 static inline unsigned long
960 myri10ge_rx_done(struct myri10ge_priv *mgp, struct myri10ge_rx_buf *rx,
961 int bytes, int len, __wsum csum)
963 dma_addr_t bus;
964 struct sk_buff *skb;
965 int idx, unmap_len;
967 idx = rx->cnt & rx->mask;
968 rx->cnt++;
970 /* save a pointer to the received skb */
971 skb = rx->info[idx].skb;
972 bus = pci_unmap_addr(&rx->info[idx], bus);
973 unmap_len = pci_unmap_len(&rx->info[idx], len);
975 /* try to replace the received skb */
976 if (myri10ge_getbuf(rx, mgp, bytes, idx)) {
977 /* drop the frame -- the old skbuf is re-cycled */
978 mgp->stats.rx_dropped += 1;
979 return 0;
982 /* unmap the recvd skb */
983 pci_unmap_single(mgp->pdev, bus, unmap_len, PCI_DMA_FROMDEVICE);
985 /* mcp implicitly skips 1st bytes so that packet is properly
986 * aligned */
987 skb_reserve(skb, MXGEFW_PAD);
989 /* set the length of the frame */
990 skb_put(skb, len);
992 skb->protocol = eth_type_trans(skb, mgp->dev);
993 if (mgp->csum_flag) {
994 if ((skb->protocol == htons(ETH_P_IP)) ||
995 (skb->protocol == htons(ETH_P_IPV6))) {
996 skb->csum = csum;
997 skb->ip_summed = CHECKSUM_COMPLETE;
998 } else
999 myri10ge_vlan_ip_csum(skb, csum);
1002 netif_receive_skb(skb);
1003 mgp->dev->last_rx = jiffies;
1004 return 1;
1007 static inline void myri10ge_tx_done(struct myri10ge_priv *mgp, int mcp_index)
1009 struct pci_dev *pdev = mgp->pdev;
1010 struct myri10ge_tx_buf *tx = &mgp->tx;
1011 struct sk_buff *skb;
1012 int idx, len;
1013 int limit = 0;
1015 while (tx->pkt_done != mcp_index) {
1016 idx = tx->done & tx->mask;
1017 skb = tx->info[idx].skb;
1019 /* Mark as free */
1020 tx->info[idx].skb = NULL;
1021 if (tx->info[idx].last) {
1022 tx->pkt_done++;
1023 tx->info[idx].last = 0;
1025 tx->done++;
1026 len = pci_unmap_len(&tx->info[idx], len);
1027 pci_unmap_len_set(&tx->info[idx], len, 0);
1028 if (skb) {
1029 mgp->stats.tx_bytes += skb->len;
1030 mgp->stats.tx_packets++;
1031 dev_kfree_skb_irq(skb);
1032 if (len)
1033 pci_unmap_single(pdev,
1034 pci_unmap_addr(&tx->info[idx],
1035 bus), len,
1036 PCI_DMA_TODEVICE);
1037 } else {
1038 if (len)
1039 pci_unmap_page(pdev,
1040 pci_unmap_addr(&tx->info[idx],
1041 bus), len,
1042 PCI_DMA_TODEVICE);
1045 /* limit potential for livelock by only handling
1046 * 2 full tx rings per call */
1047 if (unlikely(++limit > 2 * tx->mask))
1048 break;
1050 /* start the queue if we've stopped it */
1051 if (netif_queue_stopped(mgp->dev)
1052 && tx->req - tx->done < (tx->mask >> 1)) {
1053 mgp->wake_queue++;
1054 netif_wake_queue(mgp->dev);
1058 static inline void myri10ge_clean_rx_done(struct myri10ge_priv *mgp, int *limit)
1060 struct myri10ge_rx_done *rx_done = &mgp->rx_done;
1061 unsigned long rx_bytes = 0;
1062 unsigned long rx_packets = 0;
1063 unsigned long rx_ok;
1065 int idx = rx_done->idx;
1066 int cnt = rx_done->cnt;
1067 u16 length;
1068 __wsum checksum;
1070 while (rx_done->entry[idx].length != 0 && *limit != 0) {
1071 length = ntohs(rx_done->entry[idx].length);
1072 rx_done->entry[idx].length = 0;
1073 checksum = csum_unfold(rx_done->entry[idx].checksum);
1074 if (length <= mgp->small_bytes)
1075 rx_ok = myri10ge_rx_done(mgp, &mgp->rx_small,
1076 mgp->small_bytes,
1077 length, checksum);
1078 else
1079 rx_ok = myri10ge_rx_done(mgp, &mgp->rx_big,
1080 mgp->dev->mtu + ETH_HLEN,
1081 length, checksum);
1082 rx_packets += rx_ok;
1083 rx_bytes += rx_ok * (unsigned long)length;
1084 cnt++;
1085 idx = cnt & (myri10ge_max_intr_slots - 1);
1087 /* limit potential for livelock by only handling a
1088 * limited number of frames. */
1089 (*limit)--;
1091 rx_done->idx = idx;
1092 rx_done->cnt = cnt;
1093 mgp->stats.rx_packets += rx_packets;
1094 mgp->stats.rx_bytes += rx_bytes;
1097 static inline void myri10ge_check_statblock(struct myri10ge_priv *mgp)
1099 struct mcp_irq_data *stats = mgp->fw_stats;
1101 if (unlikely(stats->stats_updated)) {
1102 if (mgp->link_state != stats->link_up) {
1103 mgp->link_state = stats->link_up;
1104 if (mgp->link_state) {
1105 if (netif_msg_link(mgp))
1106 printk(KERN_INFO
1107 "myri10ge: %s: link up\n",
1108 mgp->dev->name);
1109 netif_carrier_on(mgp->dev);
1110 mgp->link_changes++;
1111 } else {
1112 if (netif_msg_link(mgp))
1113 printk(KERN_INFO
1114 "myri10ge: %s: link down\n",
1115 mgp->dev->name);
1116 netif_carrier_off(mgp->dev);
1117 mgp->link_changes++;
1120 if (mgp->rdma_tags_available !=
1121 ntohl(mgp->fw_stats->rdma_tags_available)) {
1122 mgp->rdma_tags_available =
1123 ntohl(mgp->fw_stats->rdma_tags_available);
1124 printk(KERN_WARNING "myri10ge: %s: RDMA timed out! "
1125 "%d tags left\n", mgp->dev->name,
1126 mgp->rdma_tags_available);
1128 mgp->down_cnt += stats->link_down;
1129 if (stats->link_down)
1130 wake_up(&mgp->down_wq);
1134 static int myri10ge_poll(struct net_device *netdev, int *budget)
1136 struct myri10ge_priv *mgp = netdev_priv(netdev);
1137 struct myri10ge_rx_done *rx_done = &mgp->rx_done;
1138 int limit, orig_limit, work_done;
1140 /* process as many rx events as NAPI will allow */
1141 limit = min(*budget, netdev->quota);
1142 orig_limit = limit;
1143 myri10ge_clean_rx_done(mgp, &limit);
1144 work_done = orig_limit - limit;
1145 *budget -= work_done;
1146 netdev->quota -= work_done;
1148 if (rx_done->entry[rx_done->idx].length == 0 || !netif_running(netdev)) {
1149 netif_rx_complete(netdev);
1150 put_be32(htonl(3), mgp->irq_claim);
1151 return 0;
1153 return 1;
1156 static irqreturn_t myri10ge_intr(int irq, void *arg)
1158 struct myri10ge_priv *mgp = arg;
1159 struct mcp_irq_data *stats = mgp->fw_stats;
1160 struct myri10ge_tx_buf *tx = &mgp->tx;
1161 u32 send_done_count;
1162 int i;
1164 /* make sure it is our IRQ, and that the DMA has finished */
1165 if (unlikely(!stats->valid))
1166 return (IRQ_NONE);
1168 /* low bit indicates receives are present, so schedule
1169 * napi poll handler */
1170 if (stats->valid & 1)
1171 netif_rx_schedule(mgp->dev);
1173 if (!mgp->msi_enabled) {
1174 put_be32(0, mgp->irq_deassert);
1175 if (!myri10ge_deassert_wait)
1176 stats->valid = 0;
1177 mb();
1178 } else
1179 stats->valid = 0;
1181 /* Wait for IRQ line to go low, if using INTx */
1182 i = 0;
1183 while (1) {
1184 i++;
1185 /* check for transmit completes and receives */
1186 send_done_count = ntohl(stats->send_done_count);
1187 if (send_done_count != tx->pkt_done)
1188 myri10ge_tx_done(mgp, (int)send_done_count);
1189 if (unlikely(i > myri10ge_max_irq_loops)) {
1190 printk(KERN_WARNING "myri10ge: %s: irq stuck?\n",
1191 mgp->dev->name);
1192 stats->valid = 0;
1193 schedule_work(&mgp->watchdog_work);
1195 if (likely(stats->valid == 0))
1196 break;
1197 cpu_relax();
1198 barrier();
1201 myri10ge_check_statblock(mgp);
1203 put_be32(htonl(3), mgp->irq_claim + 1);
1204 return (IRQ_HANDLED);
1207 static int
1208 myri10ge_get_settings(struct net_device *netdev, struct ethtool_cmd *cmd)
1210 cmd->autoneg = AUTONEG_DISABLE;
1211 cmd->speed = SPEED_10000;
1212 cmd->duplex = DUPLEX_FULL;
1213 return 0;
1216 static void
1217 myri10ge_get_drvinfo(struct net_device *netdev, struct ethtool_drvinfo *info)
1219 struct myri10ge_priv *mgp = netdev_priv(netdev);
1221 strlcpy(info->driver, "myri10ge", sizeof(info->driver));
1222 strlcpy(info->version, MYRI10GE_VERSION_STR, sizeof(info->version));
1223 strlcpy(info->fw_version, mgp->fw_version, sizeof(info->fw_version));
1224 strlcpy(info->bus_info, pci_name(mgp->pdev), sizeof(info->bus_info));
1227 static int
1228 myri10ge_get_coalesce(struct net_device *netdev, struct ethtool_coalesce *coal)
1230 struct myri10ge_priv *mgp = netdev_priv(netdev);
1231 coal->rx_coalesce_usecs = mgp->intr_coal_delay;
1232 return 0;
1235 static int
1236 myri10ge_set_coalesce(struct net_device *netdev, struct ethtool_coalesce *coal)
1238 struct myri10ge_priv *mgp = netdev_priv(netdev);
1240 mgp->intr_coal_delay = coal->rx_coalesce_usecs;
1241 put_be32(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
1242 return 0;
1245 static void
1246 myri10ge_get_pauseparam(struct net_device *netdev,
1247 struct ethtool_pauseparam *pause)
1249 struct myri10ge_priv *mgp = netdev_priv(netdev);
1251 pause->autoneg = 0;
1252 pause->rx_pause = mgp->pause;
1253 pause->tx_pause = mgp->pause;
1256 static int
1257 myri10ge_set_pauseparam(struct net_device *netdev,
1258 struct ethtool_pauseparam *pause)
1260 struct myri10ge_priv *mgp = netdev_priv(netdev);
1262 if (pause->tx_pause != mgp->pause)
1263 return myri10ge_change_pause(mgp, pause->tx_pause);
1264 if (pause->rx_pause != mgp->pause)
1265 return myri10ge_change_pause(mgp, pause->tx_pause);
1266 if (pause->autoneg != 0)
1267 return -EINVAL;
1268 return 0;
1271 static void
1272 myri10ge_get_ringparam(struct net_device *netdev,
1273 struct ethtool_ringparam *ring)
1275 struct myri10ge_priv *mgp = netdev_priv(netdev);
1277 ring->rx_mini_max_pending = mgp->rx_small.mask + 1;
1278 ring->rx_max_pending = mgp->rx_big.mask + 1;
1279 ring->rx_jumbo_max_pending = 0;
1280 ring->tx_max_pending = mgp->rx_small.mask + 1;
1281 ring->rx_mini_pending = ring->rx_mini_max_pending;
1282 ring->rx_pending = ring->rx_max_pending;
1283 ring->rx_jumbo_pending = ring->rx_jumbo_max_pending;
1284 ring->tx_pending = ring->tx_max_pending;
1287 static u32 myri10ge_get_rx_csum(struct net_device *netdev)
1289 struct myri10ge_priv *mgp = netdev_priv(netdev);
1290 if (mgp->csum_flag)
1291 return 1;
1292 else
1293 return 0;
1296 static int myri10ge_set_rx_csum(struct net_device *netdev, u32 csum_enabled)
1298 struct myri10ge_priv *mgp = netdev_priv(netdev);
1299 if (csum_enabled)
1300 mgp->csum_flag = MXGEFW_FLAGS_CKSUM;
1301 else
1302 mgp->csum_flag = 0;
1303 return 0;
1306 static const char myri10ge_gstrings_stats[][ETH_GSTRING_LEN] = {
1307 "rx_packets", "tx_packets", "rx_bytes", "tx_bytes", "rx_errors",
1308 "tx_errors", "rx_dropped", "tx_dropped", "multicast", "collisions",
1309 "rx_length_errors", "rx_over_errors", "rx_crc_errors",
1310 "rx_frame_errors", "rx_fifo_errors", "rx_missed_errors",
1311 "tx_aborted_errors", "tx_carrier_errors", "tx_fifo_errors",
1312 "tx_heartbeat_errors", "tx_window_errors",
1313 /* device-specific stats */
1314 "tx_boundary", "WC", "irq", "MSI",
1315 "read_dma_bw_MBs", "write_dma_bw_MBs", "read_write_dma_bw_MBs",
1316 "serial_number", "tx_pkt_start", "tx_pkt_done",
1317 "tx_req", "tx_done", "rx_small_cnt", "rx_big_cnt",
1318 "wake_queue", "stop_queue", "watchdog_resets", "tx_linearized",
1319 "link_changes", "link_up", "dropped_link_overflow",
1320 "dropped_link_error_or_filtered", "dropped_multicast_filtered",
1321 "dropped_runt", "dropped_overrun", "dropped_no_small_buffer",
1322 "dropped_no_big_buffer"
1325 #define MYRI10GE_NET_STATS_LEN 21
1326 #define MYRI10GE_STATS_LEN sizeof(myri10ge_gstrings_stats) / ETH_GSTRING_LEN
1328 static void
1329 myri10ge_get_strings(struct net_device *netdev, u32 stringset, u8 * data)
1331 switch (stringset) {
1332 case ETH_SS_STATS:
1333 memcpy(data, *myri10ge_gstrings_stats,
1334 sizeof(myri10ge_gstrings_stats));
1335 break;
1339 static int myri10ge_get_stats_count(struct net_device *netdev)
1341 return MYRI10GE_STATS_LEN;
1344 static void
1345 myri10ge_get_ethtool_stats(struct net_device *netdev,
1346 struct ethtool_stats *stats, u64 * data)
1348 struct myri10ge_priv *mgp = netdev_priv(netdev);
1349 int i;
1351 for (i = 0; i < MYRI10GE_NET_STATS_LEN; i++)
1352 data[i] = ((unsigned long *)&mgp->stats)[i];
1354 data[i++] = (unsigned int)mgp->tx.boundary;
1355 data[i++] = (unsigned int)(mgp->mtrr >= 0);
1356 data[i++] = (unsigned int)mgp->pdev->irq;
1357 data[i++] = (unsigned int)mgp->msi_enabled;
1358 data[i++] = (unsigned int)mgp->read_dma;
1359 data[i++] = (unsigned int)mgp->write_dma;
1360 data[i++] = (unsigned int)mgp->read_write_dma;
1361 data[i++] = (unsigned int)mgp->serial_number;
1362 data[i++] = (unsigned int)mgp->tx.pkt_start;
1363 data[i++] = (unsigned int)mgp->tx.pkt_done;
1364 data[i++] = (unsigned int)mgp->tx.req;
1365 data[i++] = (unsigned int)mgp->tx.done;
1366 data[i++] = (unsigned int)mgp->rx_small.cnt;
1367 data[i++] = (unsigned int)mgp->rx_big.cnt;
1368 data[i++] = (unsigned int)mgp->wake_queue;
1369 data[i++] = (unsigned int)mgp->stop_queue;
1370 data[i++] = (unsigned int)mgp->watchdog_resets;
1371 data[i++] = (unsigned int)mgp->tx_linearized;
1372 data[i++] = (unsigned int)mgp->link_changes;
1373 data[i++] = (unsigned int)ntohl(mgp->fw_stats->link_up);
1374 data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_link_overflow);
1375 data[i++] =
1376 (unsigned int)ntohl(mgp->fw_stats->dropped_link_error_or_filtered);
1377 data[i++] =
1378 (unsigned int)ntohl(mgp->fw_stats->dropped_multicast_filtered);
1379 data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_runt);
1380 data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_overrun);
1381 data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_no_small_buffer);
1382 data[i++] = (unsigned int)ntohl(mgp->fw_stats->dropped_no_big_buffer);
1385 static void myri10ge_set_msglevel(struct net_device *netdev, u32 value)
1387 struct myri10ge_priv *mgp = netdev_priv(netdev);
1388 mgp->msg_enable = value;
1391 static u32 myri10ge_get_msglevel(struct net_device *netdev)
1393 struct myri10ge_priv *mgp = netdev_priv(netdev);
1394 return mgp->msg_enable;
1397 static const struct ethtool_ops myri10ge_ethtool_ops = {
1398 .get_settings = myri10ge_get_settings,
1399 .get_drvinfo = myri10ge_get_drvinfo,
1400 .get_coalesce = myri10ge_get_coalesce,
1401 .set_coalesce = myri10ge_set_coalesce,
1402 .get_pauseparam = myri10ge_get_pauseparam,
1403 .set_pauseparam = myri10ge_set_pauseparam,
1404 .get_ringparam = myri10ge_get_ringparam,
1405 .get_rx_csum = myri10ge_get_rx_csum,
1406 .set_rx_csum = myri10ge_set_rx_csum,
1407 .get_tx_csum = ethtool_op_get_tx_csum,
1408 .set_tx_csum = ethtool_op_set_tx_hw_csum,
1409 .get_sg = ethtool_op_get_sg,
1410 .set_sg = ethtool_op_set_sg,
1411 #ifdef NETIF_F_TSO
1412 .get_tso = ethtool_op_get_tso,
1413 .set_tso = ethtool_op_set_tso,
1414 #endif
1415 .get_strings = myri10ge_get_strings,
1416 .get_stats_count = myri10ge_get_stats_count,
1417 .get_ethtool_stats = myri10ge_get_ethtool_stats,
1418 .set_msglevel = myri10ge_set_msglevel,
1419 .get_msglevel = myri10ge_get_msglevel
1422 static int myri10ge_allocate_rings(struct net_device *dev)
1424 struct myri10ge_priv *mgp;
1425 struct myri10ge_cmd cmd;
1426 int tx_ring_size, rx_ring_size;
1427 int tx_ring_entries, rx_ring_entries;
1428 int i, status;
1429 size_t bytes;
1431 mgp = netdev_priv(dev);
1433 /* get ring sizes */
1435 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_RING_SIZE, &cmd, 0);
1436 tx_ring_size = cmd.data0;
1437 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_RX_RING_SIZE, &cmd, 0);
1438 rx_ring_size = cmd.data0;
1440 tx_ring_entries = tx_ring_size / sizeof(struct mcp_kreq_ether_send);
1441 rx_ring_entries = rx_ring_size / sizeof(struct mcp_dma_addr);
1442 mgp->tx.mask = tx_ring_entries - 1;
1443 mgp->rx_small.mask = mgp->rx_big.mask = rx_ring_entries - 1;
1445 /* allocate the host shadow rings */
1447 bytes = 8 + (MYRI10GE_MAX_SEND_DESC_TSO + 4)
1448 * sizeof(*mgp->tx.req_list);
1449 mgp->tx.req_bytes = kzalloc(bytes, GFP_KERNEL);
1450 if (mgp->tx.req_bytes == NULL)
1451 goto abort_with_nothing;
1453 /* ensure req_list entries are aligned to 8 bytes */
1454 mgp->tx.req_list = (struct mcp_kreq_ether_send *)
1455 ALIGN((unsigned long)mgp->tx.req_bytes, 8);
1457 bytes = rx_ring_entries * sizeof(*mgp->rx_small.shadow);
1458 mgp->rx_small.shadow = kzalloc(bytes, GFP_KERNEL);
1459 if (mgp->rx_small.shadow == NULL)
1460 goto abort_with_tx_req_bytes;
1462 bytes = rx_ring_entries * sizeof(*mgp->rx_big.shadow);
1463 mgp->rx_big.shadow = kzalloc(bytes, GFP_KERNEL);
1464 if (mgp->rx_big.shadow == NULL)
1465 goto abort_with_rx_small_shadow;
1467 /* allocate the host info rings */
1469 bytes = tx_ring_entries * sizeof(*mgp->tx.info);
1470 mgp->tx.info = kzalloc(bytes, GFP_KERNEL);
1471 if (mgp->tx.info == NULL)
1472 goto abort_with_rx_big_shadow;
1474 bytes = rx_ring_entries * sizeof(*mgp->rx_small.info);
1475 mgp->rx_small.info = kzalloc(bytes, GFP_KERNEL);
1476 if (mgp->rx_small.info == NULL)
1477 goto abort_with_tx_info;
1479 bytes = rx_ring_entries * sizeof(*mgp->rx_big.info);
1480 mgp->rx_big.info = kzalloc(bytes, GFP_KERNEL);
1481 if (mgp->rx_big.info == NULL)
1482 goto abort_with_rx_small_info;
1484 /* Fill the receive rings */
1486 for (i = 0; i <= mgp->rx_small.mask; i++) {
1487 status = myri10ge_getbuf(&mgp->rx_small, mgp,
1488 mgp->small_bytes, i);
1489 if (status) {
1490 printk(KERN_ERR
1491 "myri10ge: %s: alloced only %d small bufs\n",
1492 dev->name, i);
1493 goto abort_with_rx_small_ring;
1497 for (i = 0; i <= mgp->rx_big.mask; i++) {
1498 status =
1499 myri10ge_getbuf(&mgp->rx_big, mgp, dev->mtu + ETH_HLEN, i);
1500 if (status) {
1501 printk(KERN_ERR
1502 "myri10ge: %s: alloced only %d big bufs\n",
1503 dev->name, i);
1504 goto abort_with_rx_big_ring;
1508 return 0;
1510 abort_with_rx_big_ring:
1511 for (i = 0; i <= mgp->rx_big.mask; i++) {
1512 if (mgp->rx_big.info[i].skb != NULL)
1513 dev_kfree_skb_any(mgp->rx_big.info[i].skb);
1514 if (pci_unmap_len(&mgp->rx_big.info[i], len))
1515 pci_unmap_single(mgp->pdev,
1516 pci_unmap_addr(&mgp->rx_big.info[i],
1517 bus),
1518 pci_unmap_len(&mgp->rx_big.info[i],
1519 len),
1520 PCI_DMA_FROMDEVICE);
1523 abort_with_rx_small_ring:
1524 for (i = 0; i <= mgp->rx_small.mask; i++) {
1525 if (mgp->rx_small.info[i].skb != NULL)
1526 dev_kfree_skb_any(mgp->rx_small.info[i].skb);
1527 if (pci_unmap_len(&mgp->rx_small.info[i], len))
1528 pci_unmap_single(mgp->pdev,
1529 pci_unmap_addr(&mgp->rx_small.info[i],
1530 bus),
1531 pci_unmap_len(&mgp->rx_small.info[i],
1532 len),
1533 PCI_DMA_FROMDEVICE);
1535 kfree(mgp->rx_big.info);
1537 abort_with_rx_small_info:
1538 kfree(mgp->rx_small.info);
1540 abort_with_tx_info:
1541 kfree(mgp->tx.info);
1543 abort_with_rx_big_shadow:
1544 kfree(mgp->rx_big.shadow);
1546 abort_with_rx_small_shadow:
1547 kfree(mgp->rx_small.shadow);
1549 abort_with_tx_req_bytes:
1550 kfree(mgp->tx.req_bytes);
1551 mgp->tx.req_bytes = NULL;
1552 mgp->tx.req_list = NULL;
1554 abort_with_nothing:
1555 return status;
1558 static void myri10ge_free_rings(struct net_device *dev)
1560 struct myri10ge_priv *mgp;
1561 struct sk_buff *skb;
1562 struct myri10ge_tx_buf *tx;
1563 int i, len, idx;
1565 mgp = netdev_priv(dev);
1567 for (i = 0; i <= mgp->rx_big.mask; i++) {
1568 if (mgp->rx_big.info[i].skb != NULL)
1569 dev_kfree_skb_any(mgp->rx_big.info[i].skb);
1570 if (pci_unmap_len(&mgp->rx_big.info[i], len))
1571 pci_unmap_single(mgp->pdev,
1572 pci_unmap_addr(&mgp->rx_big.info[i],
1573 bus),
1574 pci_unmap_len(&mgp->rx_big.info[i],
1575 len),
1576 PCI_DMA_FROMDEVICE);
1579 for (i = 0; i <= mgp->rx_small.mask; i++) {
1580 if (mgp->rx_small.info[i].skb != NULL)
1581 dev_kfree_skb_any(mgp->rx_small.info[i].skb);
1582 if (pci_unmap_len(&mgp->rx_small.info[i], len))
1583 pci_unmap_single(mgp->pdev,
1584 pci_unmap_addr(&mgp->rx_small.info[i],
1585 bus),
1586 pci_unmap_len(&mgp->rx_small.info[i],
1587 len),
1588 PCI_DMA_FROMDEVICE);
1591 tx = &mgp->tx;
1592 while (tx->done != tx->req) {
1593 idx = tx->done & tx->mask;
1594 skb = tx->info[idx].skb;
1596 /* Mark as free */
1597 tx->info[idx].skb = NULL;
1598 tx->done++;
1599 len = pci_unmap_len(&tx->info[idx], len);
1600 pci_unmap_len_set(&tx->info[idx], len, 0);
1601 if (skb) {
1602 mgp->stats.tx_dropped++;
1603 dev_kfree_skb_any(skb);
1604 if (len)
1605 pci_unmap_single(mgp->pdev,
1606 pci_unmap_addr(&tx->info[idx],
1607 bus), len,
1608 PCI_DMA_TODEVICE);
1609 } else {
1610 if (len)
1611 pci_unmap_page(mgp->pdev,
1612 pci_unmap_addr(&tx->info[idx],
1613 bus), len,
1614 PCI_DMA_TODEVICE);
1617 kfree(mgp->rx_big.info);
1619 kfree(mgp->rx_small.info);
1621 kfree(mgp->tx.info);
1623 kfree(mgp->rx_big.shadow);
1625 kfree(mgp->rx_small.shadow);
1627 kfree(mgp->tx.req_bytes);
1628 mgp->tx.req_bytes = NULL;
1629 mgp->tx.req_list = NULL;
1632 static int myri10ge_open(struct net_device *dev)
1634 struct myri10ge_priv *mgp;
1635 struct myri10ge_cmd cmd;
1636 int status, big_pow2;
1638 mgp = netdev_priv(dev);
1640 if (mgp->running != MYRI10GE_ETH_STOPPED)
1641 return -EBUSY;
1643 mgp->running = MYRI10GE_ETH_STARTING;
1644 status = myri10ge_reset(mgp);
1645 if (status != 0) {
1646 printk(KERN_ERR "myri10ge: %s: failed reset\n", dev->name);
1647 mgp->running = MYRI10GE_ETH_STOPPED;
1648 return -ENXIO;
1651 /* decide what small buffer size to use. For good TCP rx
1652 * performance, it is important to not receive 1514 byte
1653 * frames into jumbo buffers, as it confuses the socket buffer
1654 * accounting code, leading to drops and erratic performance.
1657 if (dev->mtu <= ETH_DATA_LEN)
1658 mgp->small_bytes = 128; /* enough for a TCP header */
1659 else
1660 mgp->small_bytes = ETH_FRAME_LEN; /* enough for an ETH_DATA_LEN frame */
1662 /* Override the small buffer size? */
1663 if (myri10ge_small_bytes > 0)
1664 mgp->small_bytes = myri10ge_small_bytes;
1666 /* If the user sets an obscenely small MTU, adjust the small
1667 * bytes down to nearly nothing */
1668 if (mgp->small_bytes >= (dev->mtu + ETH_HLEN))
1669 mgp->small_bytes = 64;
1671 /* get the lanai pointers to the send and receive rings */
1673 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_OFFSET, &cmd, 0);
1674 mgp->tx.lanai =
1675 (struct mcp_kreq_ether_send __iomem *)(mgp->sram + cmd.data0);
1677 status |=
1678 myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SMALL_RX_OFFSET, &cmd, 0);
1679 mgp->rx_small.lanai =
1680 (struct mcp_kreq_ether_recv __iomem *)(mgp->sram + cmd.data0);
1682 status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_BIG_RX_OFFSET, &cmd, 0);
1683 mgp->rx_big.lanai =
1684 (struct mcp_kreq_ether_recv __iomem *)(mgp->sram + cmd.data0);
1686 if (status != 0) {
1687 printk(KERN_ERR
1688 "myri10ge: %s: failed to get ring sizes or locations\n",
1689 dev->name);
1690 mgp->running = MYRI10GE_ETH_STOPPED;
1691 return -ENXIO;
1694 if (mgp->mtrr >= 0) {
1695 mgp->tx.wc_fifo = (u8 __iomem *) mgp->sram + MXGEFW_ETH_SEND_4;
1696 mgp->rx_small.wc_fifo =
1697 (u8 __iomem *) mgp->sram + MXGEFW_ETH_RECV_SMALL;
1698 mgp->rx_big.wc_fifo =
1699 (u8 __iomem *) mgp->sram + MXGEFW_ETH_RECV_BIG;
1700 } else {
1701 mgp->tx.wc_fifo = NULL;
1702 mgp->rx_small.wc_fifo = NULL;
1703 mgp->rx_big.wc_fifo = NULL;
1706 status = myri10ge_allocate_rings(dev);
1707 if (status != 0)
1708 goto abort_with_nothing;
1710 /* Firmware needs the big buff size as a power of 2. Lie and
1711 * tell him the buffer is larger, because we only use 1
1712 * buffer/pkt, and the mtu will prevent overruns.
1714 big_pow2 = dev->mtu + ETH_HLEN + MXGEFW_PAD;
1715 while ((big_pow2 & (big_pow2 - 1)) != 0)
1716 big_pow2++;
1718 /* now give firmware buffers sizes, and MTU */
1719 cmd.data0 = dev->mtu + ETH_HLEN + VLAN_HLEN;
1720 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_MTU, &cmd, 0);
1721 cmd.data0 = mgp->small_bytes;
1722 status |=
1723 myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_SMALL_BUFFER_SIZE, &cmd, 0);
1724 cmd.data0 = big_pow2;
1725 status |=
1726 myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_BIG_BUFFER_SIZE, &cmd, 0);
1727 if (status) {
1728 printk(KERN_ERR "myri10ge: %s: Couldn't set buffer sizes\n",
1729 dev->name);
1730 goto abort_with_rings;
1733 cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->fw_stats_bus);
1734 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->fw_stats_bus);
1735 cmd.data2 = sizeof(struct mcp_irq_data);
1736 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_STATS_DMA_V2, &cmd, 0);
1737 if (status == -ENOSYS) {
1738 dma_addr_t bus = mgp->fw_stats_bus;
1739 bus += offsetof(struct mcp_irq_data, send_done_count);
1740 cmd.data0 = MYRI10GE_LOWPART_TO_U32(bus);
1741 cmd.data1 = MYRI10GE_HIGHPART_TO_U32(bus);
1742 status = myri10ge_send_cmd(mgp,
1743 MXGEFW_CMD_SET_STATS_DMA_OBSOLETE,
1744 &cmd, 0);
1745 /* Firmware cannot support multicast without STATS_DMA_V2 */
1746 mgp->fw_multicast_support = 0;
1747 } else {
1748 mgp->fw_multicast_support = 1;
1750 if (status) {
1751 printk(KERN_ERR "myri10ge: %s: Couldn't set stats DMA\n",
1752 dev->name);
1753 goto abort_with_rings;
1756 mgp->link_state = htonl(~0U);
1757 mgp->rdma_tags_available = 15;
1759 netif_poll_enable(mgp->dev); /* must happen prior to any irq */
1761 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_UP, &cmd, 0);
1762 if (status) {
1763 printk(KERN_ERR "myri10ge: %s: Couldn't bring up link\n",
1764 dev->name);
1765 goto abort_with_rings;
1768 mgp->wake_queue = 0;
1769 mgp->stop_queue = 0;
1770 mgp->running = MYRI10GE_ETH_RUNNING;
1771 mgp->watchdog_timer.expires = jiffies + myri10ge_watchdog_timeout * HZ;
1772 add_timer(&mgp->watchdog_timer);
1773 netif_wake_queue(dev);
1774 return 0;
1776 abort_with_rings:
1777 myri10ge_free_rings(dev);
1779 abort_with_nothing:
1780 mgp->running = MYRI10GE_ETH_STOPPED;
1781 return -ENOMEM;
1784 static int myri10ge_close(struct net_device *dev)
1786 struct myri10ge_priv *mgp;
1787 struct myri10ge_cmd cmd;
1788 int status, old_down_cnt;
1790 mgp = netdev_priv(dev);
1792 if (mgp->running != MYRI10GE_ETH_RUNNING)
1793 return 0;
1795 if (mgp->tx.req_bytes == NULL)
1796 return 0;
1798 del_timer_sync(&mgp->watchdog_timer);
1799 mgp->running = MYRI10GE_ETH_STOPPING;
1800 netif_poll_disable(mgp->dev);
1801 netif_carrier_off(dev);
1802 netif_stop_queue(dev);
1803 old_down_cnt = mgp->down_cnt;
1804 mb();
1805 status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_DOWN, &cmd, 0);
1806 if (status)
1807 printk(KERN_ERR "myri10ge: %s: Couldn't bring down link\n",
1808 dev->name);
1810 wait_event_timeout(mgp->down_wq, old_down_cnt != mgp->down_cnt, HZ);
1811 if (old_down_cnt == mgp->down_cnt)
1812 printk(KERN_ERR "myri10ge: %s never got down irq\n", dev->name);
1814 netif_tx_disable(dev);
1816 myri10ge_free_rings(dev);
1818 mgp->running = MYRI10GE_ETH_STOPPED;
1819 return 0;
1822 /* copy an array of struct mcp_kreq_ether_send's to the mcp. Copy
1823 * backwards one at a time and handle ring wraps */
1825 static inline void
1826 myri10ge_submit_req_backwards(struct myri10ge_tx_buf *tx,
1827 struct mcp_kreq_ether_send *src, int cnt)
1829 int idx, starting_slot;
1830 starting_slot = tx->req;
1831 while (cnt > 1) {
1832 cnt--;
1833 idx = (starting_slot + cnt) & tx->mask;
1834 myri10ge_pio_copy(&tx->lanai[idx], &src[cnt], sizeof(*src));
1835 mb();
1840 * copy an array of struct mcp_kreq_ether_send's to the mcp. Copy
1841 * at most 32 bytes at a time, so as to avoid involving the software
1842 * pio handler in the nic. We re-write the first segment's flags
1843 * to mark them valid only after writing the entire chain.
1846 static inline void
1847 myri10ge_submit_req(struct myri10ge_tx_buf *tx, struct mcp_kreq_ether_send *src,
1848 int cnt)
1850 int idx, i;
1851 struct mcp_kreq_ether_send __iomem *dstp, *dst;
1852 struct mcp_kreq_ether_send *srcp;
1853 u8 last_flags;
1855 idx = tx->req & tx->mask;
1857 last_flags = src->flags;
1858 src->flags = 0;
1859 mb();
1860 dst = dstp = &tx->lanai[idx];
1861 srcp = src;
1863 if ((idx + cnt) < tx->mask) {
1864 for (i = 0; i < (cnt - 1); i += 2) {
1865 myri10ge_pio_copy(dstp, srcp, 2 * sizeof(*src));
1866 mb(); /* force write every 32 bytes */
1867 srcp += 2;
1868 dstp += 2;
1870 } else {
1871 /* submit all but the first request, and ensure
1872 * that it is submitted below */
1873 myri10ge_submit_req_backwards(tx, src, cnt);
1874 i = 0;
1876 if (i < cnt) {
1877 /* submit the first request */
1878 myri10ge_pio_copy(dstp, srcp, sizeof(*src));
1879 mb(); /* barrier before setting valid flag */
1882 /* re-write the last 32-bits with the valid flags */
1883 src->flags = last_flags;
1884 put_be32(*((__be32 *) src + 3), (__be32 __iomem *) dst + 3);
1885 tx->req += cnt;
1886 mb();
1889 static inline void
1890 myri10ge_submit_req_wc(struct myri10ge_tx_buf *tx,
1891 struct mcp_kreq_ether_send *src, int cnt)
1893 tx->req += cnt;
1894 mb();
1895 while (cnt >= 4) {
1896 myri10ge_pio_copy(tx->wc_fifo, src, 64);
1897 mb();
1898 src += 4;
1899 cnt -= 4;
1901 if (cnt > 0) {
1902 /* pad it to 64 bytes. The src is 64 bytes bigger than it
1903 * needs to be so that we don't overrun it */
1904 myri10ge_pio_copy(tx->wc_fifo + MXGEFW_ETH_SEND_OFFSET(cnt),
1905 src, 64);
1906 mb();
1911 * Transmit a packet. We need to split the packet so that a single
1912 * segment does not cross myri10ge->tx.boundary, so this makes segment
1913 * counting tricky. So rather than try to count segments up front, we
1914 * just give up if there are too few segments to hold a reasonably
1915 * fragmented packet currently available. If we run
1916 * out of segments while preparing a packet for DMA, we just linearize
1917 * it and try again.
1920 static int myri10ge_xmit(struct sk_buff *skb, struct net_device *dev)
1922 struct myri10ge_priv *mgp = netdev_priv(dev);
1923 struct mcp_kreq_ether_send *req;
1924 struct myri10ge_tx_buf *tx = &mgp->tx;
1925 struct skb_frag_struct *frag;
1926 dma_addr_t bus;
1927 u32 low;
1928 __be32 high_swapped;
1929 unsigned int len;
1930 int idx, last_idx, avail, frag_cnt, frag_idx, count, mss, max_segments;
1931 u16 pseudo_hdr_offset, cksum_offset;
1932 int cum_len, seglen, boundary, rdma_count;
1933 u8 flags, odd_flag;
1935 again:
1936 req = tx->req_list;
1937 avail = tx->mask - 1 - (tx->req - tx->done);
1939 mss = 0;
1940 max_segments = MXGEFW_MAX_SEND_DESC;
1942 #ifdef NETIF_F_TSO
1943 if (skb->len > (dev->mtu + ETH_HLEN)) {
1944 mss = skb_shinfo(skb)->gso_size;
1945 if (mss != 0)
1946 max_segments = MYRI10GE_MAX_SEND_DESC_TSO;
1948 #endif /*NETIF_F_TSO */
1950 if ((unlikely(avail < max_segments))) {
1951 /* we are out of transmit resources */
1952 mgp->stop_queue++;
1953 netif_stop_queue(dev);
1954 return 1;
1957 /* Setup checksum offloading, if needed */
1958 cksum_offset = 0;
1959 pseudo_hdr_offset = 0;
1960 odd_flag = 0;
1961 flags = (MXGEFW_FLAGS_NO_TSO | MXGEFW_FLAGS_FIRST);
1962 if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1963 cksum_offset = (skb->h.raw - skb->data);
1964 pseudo_hdr_offset = cksum_offset + skb->csum_offset;
1965 /* If the headers are excessively large, then we must
1966 * fall back to a software checksum */
1967 if (unlikely(cksum_offset > 255 || pseudo_hdr_offset > 127)) {
1968 if (skb_checksum_help(skb))
1969 goto drop;
1970 cksum_offset = 0;
1971 pseudo_hdr_offset = 0;
1972 } else {
1973 odd_flag = MXGEFW_FLAGS_ALIGN_ODD;
1974 flags |= MXGEFW_FLAGS_CKSUM;
1978 cum_len = 0;
1980 #ifdef NETIF_F_TSO
1981 if (mss) { /* TSO */
1982 /* this removes any CKSUM flag from before */
1983 flags = (MXGEFW_FLAGS_TSO_HDR | MXGEFW_FLAGS_FIRST);
1985 /* negative cum_len signifies to the
1986 * send loop that we are still in the
1987 * header portion of the TSO packet.
1988 * TSO header must be at most 134 bytes long */
1989 cum_len = -((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
1991 /* for TSO, pseudo_hdr_offset holds mss.
1992 * The firmware figures out where to put
1993 * the checksum by parsing the header. */
1994 pseudo_hdr_offset = mss;
1995 } else
1996 #endif /*NETIF_F_TSO */
1997 /* Mark small packets, and pad out tiny packets */
1998 if (skb->len <= MXGEFW_SEND_SMALL_SIZE) {
1999 flags |= MXGEFW_FLAGS_SMALL;
2001 /* pad frames to at least ETH_ZLEN bytes */
2002 if (unlikely(skb->len < ETH_ZLEN)) {
2003 if (skb_padto(skb, ETH_ZLEN)) {
2004 /* The packet is gone, so we must
2005 * return 0 */
2006 mgp->stats.tx_dropped += 1;
2007 return 0;
2009 /* adjust the len to account for the zero pad
2010 * so that the nic can know how long it is */
2011 skb->len = ETH_ZLEN;
2015 /* map the skb for DMA */
2016 len = skb->len - skb->data_len;
2017 idx = tx->req & tx->mask;
2018 tx->info[idx].skb = skb;
2019 bus = pci_map_single(mgp->pdev, skb->data, len, PCI_DMA_TODEVICE);
2020 pci_unmap_addr_set(&tx->info[idx], bus, bus);
2021 pci_unmap_len_set(&tx->info[idx], len, len);
2023 frag_cnt = skb_shinfo(skb)->nr_frags;
2024 frag_idx = 0;
2025 count = 0;
2026 rdma_count = 0;
2028 /* "rdma_count" is the number of RDMAs belonging to the
2029 * current packet BEFORE the current send request. For
2030 * non-TSO packets, this is equal to "count".
2031 * For TSO packets, rdma_count needs to be reset
2032 * to 0 after a segment cut.
2034 * The rdma_count field of the send request is
2035 * the number of RDMAs of the packet starting at
2036 * that request. For TSO send requests with one ore more cuts
2037 * in the middle, this is the number of RDMAs starting
2038 * after the last cut in the request. All previous
2039 * segments before the last cut implicitly have 1 RDMA.
2041 * Since the number of RDMAs is not known beforehand,
2042 * it must be filled-in retroactively - after each
2043 * segmentation cut or at the end of the entire packet.
2046 while (1) {
2047 /* Break the SKB or Fragment up into pieces which
2048 * do not cross mgp->tx.boundary */
2049 low = MYRI10GE_LOWPART_TO_U32(bus);
2050 high_swapped = htonl(MYRI10GE_HIGHPART_TO_U32(bus));
2051 while (len) {
2052 u8 flags_next;
2053 int cum_len_next;
2055 if (unlikely(count == max_segments))
2056 goto abort_linearize;
2058 boundary = (low + tx->boundary) & ~(tx->boundary - 1);
2059 seglen = boundary - low;
2060 if (seglen > len)
2061 seglen = len;
2062 flags_next = flags & ~MXGEFW_FLAGS_FIRST;
2063 cum_len_next = cum_len + seglen;
2064 #ifdef NETIF_F_TSO
2065 if (mss) { /* TSO */
2066 (req - rdma_count)->rdma_count = rdma_count + 1;
2068 if (likely(cum_len >= 0)) { /* payload */
2069 int next_is_first, chop;
2071 chop = (cum_len_next > mss);
2072 cum_len_next = cum_len_next % mss;
2073 next_is_first = (cum_len_next == 0);
2074 flags |= chop * MXGEFW_FLAGS_TSO_CHOP;
2075 flags_next |= next_is_first *
2076 MXGEFW_FLAGS_FIRST;
2077 rdma_count |= -(chop | next_is_first);
2078 rdma_count += chop & !next_is_first;
2079 } else if (likely(cum_len_next >= 0)) { /* header ends */
2080 int small;
2082 rdma_count = -1;
2083 cum_len_next = 0;
2084 seglen = -cum_len;
2085 small = (mss <= MXGEFW_SEND_SMALL_SIZE);
2086 flags_next = MXGEFW_FLAGS_TSO_PLD |
2087 MXGEFW_FLAGS_FIRST |
2088 (small * MXGEFW_FLAGS_SMALL);
2091 #endif /* NETIF_F_TSO */
2092 req->addr_high = high_swapped;
2093 req->addr_low = htonl(low);
2094 req->pseudo_hdr_offset = htons(pseudo_hdr_offset);
2095 req->pad = 0; /* complete solid 16-byte block; does this matter? */
2096 req->rdma_count = 1;
2097 req->length = htons(seglen);
2098 req->cksum_offset = cksum_offset;
2099 req->flags = flags | ((cum_len & 1) * odd_flag);
2101 low += seglen;
2102 len -= seglen;
2103 cum_len = cum_len_next;
2104 flags = flags_next;
2105 req++;
2106 count++;
2107 rdma_count++;
2108 if (unlikely(cksum_offset > seglen))
2109 cksum_offset -= seglen;
2110 else
2111 cksum_offset = 0;
2113 if (frag_idx == frag_cnt)
2114 break;
2116 /* map next fragment for DMA */
2117 idx = (count + tx->req) & tx->mask;
2118 frag = &skb_shinfo(skb)->frags[frag_idx];
2119 frag_idx++;
2120 len = frag->size;
2121 bus = pci_map_page(mgp->pdev, frag->page, frag->page_offset,
2122 len, PCI_DMA_TODEVICE);
2123 pci_unmap_addr_set(&tx->info[idx], bus, bus);
2124 pci_unmap_len_set(&tx->info[idx], len, len);
2127 (req - rdma_count)->rdma_count = rdma_count;
2128 #ifdef NETIF_F_TSO
2129 if (mss)
2130 do {
2131 req--;
2132 req->flags |= MXGEFW_FLAGS_TSO_LAST;
2133 } while (!(req->flags & (MXGEFW_FLAGS_TSO_CHOP |
2134 MXGEFW_FLAGS_FIRST)));
2135 #endif
2136 idx = ((count - 1) + tx->req) & tx->mask;
2137 tx->info[idx].last = 1;
2138 if (tx->wc_fifo == NULL)
2139 myri10ge_submit_req(tx, tx->req_list, count);
2140 else
2141 myri10ge_submit_req_wc(tx, tx->req_list, count);
2142 tx->pkt_start++;
2143 if ((avail - count) < MXGEFW_MAX_SEND_DESC) {
2144 mgp->stop_queue++;
2145 netif_stop_queue(dev);
2147 dev->trans_start = jiffies;
2148 return 0;
2150 abort_linearize:
2151 /* Free any DMA resources we've alloced and clear out the skb
2152 * slot so as to not trip up assertions, and to avoid a
2153 * double-free if linearizing fails */
2155 last_idx = (idx + 1) & tx->mask;
2156 idx = tx->req & tx->mask;
2157 tx->info[idx].skb = NULL;
2158 do {
2159 len = pci_unmap_len(&tx->info[idx], len);
2160 if (len) {
2161 if (tx->info[idx].skb != NULL)
2162 pci_unmap_single(mgp->pdev,
2163 pci_unmap_addr(&tx->info[idx],
2164 bus), len,
2165 PCI_DMA_TODEVICE);
2166 else
2167 pci_unmap_page(mgp->pdev,
2168 pci_unmap_addr(&tx->info[idx],
2169 bus), len,
2170 PCI_DMA_TODEVICE);
2171 pci_unmap_len_set(&tx->info[idx], len, 0);
2172 tx->info[idx].skb = NULL;
2174 idx = (idx + 1) & tx->mask;
2175 } while (idx != last_idx);
2176 if (skb_is_gso(skb)) {
2177 printk(KERN_ERR
2178 "myri10ge: %s: TSO but wanted to linearize?!?!?\n",
2179 mgp->dev->name);
2180 goto drop;
2183 if (skb_linearize(skb))
2184 goto drop;
2186 mgp->tx_linearized++;
2187 goto again;
2189 drop:
2190 dev_kfree_skb_any(skb);
2191 mgp->stats.tx_dropped += 1;
2192 return 0;
2196 static struct net_device_stats *myri10ge_get_stats(struct net_device *dev)
2198 struct myri10ge_priv *mgp = netdev_priv(dev);
2199 return &mgp->stats;
2202 static void myri10ge_set_multicast_list(struct net_device *dev)
2204 struct myri10ge_cmd cmd;
2205 struct myri10ge_priv *mgp;
2206 struct dev_mc_list *mc_list;
2207 __be32 data[2] = {0, 0};
2208 int err;
2210 mgp = netdev_priv(dev);
2211 /* can be called from atomic contexts,
2212 * pass 1 to force atomicity in myri10ge_send_cmd() */
2213 myri10ge_change_promisc(mgp, dev->flags & IFF_PROMISC, 1);
2215 /* This firmware is known to not support multicast */
2216 if (!mgp->fw_multicast_support)
2217 return;
2219 /* Disable multicast filtering */
2221 err = myri10ge_send_cmd(mgp, MXGEFW_ENABLE_ALLMULTI, &cmd, 1);
2222 if (err != 0) {
2223 printk(KERN_ERR "myri10ge: %s: Failed MXGEFW_ENABLE_ALLMULTI,"
2224 " error status: %d\n", dev->name, err);
2225 goto abort;
2228 if (dev->flags & IFF_ALLMULTI) {
2229 /* request to disable multicast filtering, so quit here */
2230 return;
2233 /* Flush the filters */
2235 err = myri10ge_send_cmd(mgp, MXGEFW_LEAVE_ALL_MULTICAST_GROUPS,
2236 &cmd, 1);
2237 if (err != 0) {
2238 printk(KERN_ERR
2239 "myri10ge: %s: Failed MXGEFW_LEAVE_ALL_MULTICAST_GROUPS"
2240 ", error status: %d\n", dev->name, err);
2241 goto abort;
2244 /* Walk the multicast list, and add each address */
2245 for (mc_list = dev->mc_list; mc_list != NULL; mc_list = mc_list->next) {
2246 memcpy(data, &mc_list->dmi_addr, 6);
2247 cmd.data0 = ntohl(data[0]);
2248 cmd.data1 = ntohl(data[1]);
2249 err = myri10ge_send_cmd(mgp, MXGEFW_JOIN_MULTICAST_GROUP,
2250 &cmd, 1);
2252 if (err != 0) {
2253 printk(KERN_ERR "myri10ge: %s: Failed "
2254 "MXGEFW_JOIN_MULTICAST_GROUP, error status:"
2255 "%d\t", dev->name, err);
2256 printk(KERN_ERR "MAC %02x:%02x:%02x:%02x:%02x:%02x\n",
2257 ((unsigned char *)&mc_list->dmi_addr)[0],
2258 ((unsigned char *)&mc_list->dmi_addr)[1],
2259 ((unsigned char *)&mc_list->dmi_addr)[2],
2260 ((unsigned char *)&mc_list->dmi_addr)[3],
2261 ((unsigned char *)&mc_list->dmi_addr)[4],
2262 ((unsigned char *)&mc_list->dmi_addr)[5]
2264 goto abort;
2267 /* Enable multicast filtering */
2268 err = myri10ge_send_cmd(mgp, MXGEFW_DISABLE_ALLMULTI, &cmd, 1);
2269 if (err != 0) {
2270 printk(KERN_ERR "myri10ge: %s: Failed MXGEFW_DISABLE_ALLMULTI,"
2271 "error status: %d\n", dev->name, err);
2272 goto abort;
2275 return;
2277 abort:
2278 return;
2281 static int myri10ge_set_mac_address(struct net_device *dev, void *addr)
2283 struct sockaddr *sa = addr;
2284 struct myri10ge_priv *mgp = netdev_priv(dev);
2285 int status;
2287 if (!is_valid_ether_addr(sa->sa_data))
2288 return -EADDRNOTAVAIL;
2290 status = myri10ge_update_mac_address(mgp, sa->sa_data);
2291 if (status != 0) {
2292 printk(KERN_ERR
2293 "myri10ge: %s: changing mac address failed with %d\n",
2294 dev->name, status);
2295 return status;
2298 /* change the dev structure */
2299 memcpy(dev->dev_addr, sa->sa_data, 6);
2300 return 0;
2303 static int myri10ge_change_mtu(struct net_device *dev, int new_mtu)
2305 struct myri10ge_priv *mgp = netdev_priv(dev);
2306 int error = 0;
2308 if ((new_mtu < 68) || (ETH_HLEN + new_mtu > MYRI10GE_MAX_ETHER_MTU)) {
2309 printk(KERN_ERR "myri10ge: %s: new mtu (%d) is not valid\n",
2310 dev->name, new_mtu);
2311 return -EINVAL;
2313 printk(KERN_INFO "%s: changing mtu from %d to %d\n",
2314 dev->name, dev->mtu, new_mtu);
2315 if (mgp->running) {
2316 /* if we change the mtu on an active device, we must
2317 * reset the device so the firmware sees the change */
2318 myri10ge_close(dev);
2319 dev->mtu = new_mtu;
2320 myri10ge_open(dev);
2321 } else
2322 dev->mtu = new_mtu;
2324 return error;
2328 * Enable ECRC to align PCI-E Completion packets on an 8-byte boundary.
2329 * Only do it if the bridge is a root port since we don't want to disturb
2330 * any other device, except if forced with myri10ge_ecrc_enable > 1.
2333 static void myri10ge_enable_ecrc(struct myri10ge_priv *mgp)
2335 struct pci_dev *bridge = mgp->pdev->bus->self;
2336 struct device *dev = &mgp->pdev->dev;
2337 unsigned cap;
2338 unsigned err_cap;
2339 u16 val;
2340 u8 ext_type;
2341 int ret;
2343 if (!myri10ge_ecrc_enable || !bridge)
2344 return;
2346 /* check that the bridge is a root port */
2347 cap = pci_find_capability(bridge, PCI_CAP_ID_EXP);
2348 pci_read_config_word(bridge, cap + PCI_CAP_FLAGS, &val);
2349 ext_type = (val & PCI_EXP_FLAGS_TYPE) >> 4;
2350 if (ext_type != PCI_EXP_TYPE_ROOT_PORT) {
2351 if (myri10ge_ecrc_enable > 1) {
2352 struct pci_dev *old_bridge = bridge;
2354 /* Walk the hierarchy up to the root port
2355 * where ECRC has to be enabled */
2356 do {
2357 bridge = bridge->bus->self;
2358 if (!bridge) {
2359 dev_err(dev,
2360 "Failed to find root port"
2361 " to force ECRC\n");
2362 return;
2364 cap =
2365 pci_find_capability(bridge, PCI_CAP_ID_EXP);
2366 pci_read_config_word(bridge,
2367 cap + PCI_CAP_FLAGS, &val);
2368 ext_type = (val & PCI_EXP_FLAGS_TYPE) >> 4;
2369 } while (ext_type != PCI_EXP_TYPE_ROOT_PORT);
2371 dev_info(dev,
2372 "Forcing ECRC on non-root port %s"
2373 " (enabling on root port %s)\n",
2374 pci_name(old_bridge), pci_name(bridge));
2375 } else {
2376 dev_err(dev,
2377 "Not enabling ECRC on non-root port %s\n",
2378 pci_name(bridge));
2379 return;
2383 cap = pci_find_ext_capability(bridge, PCI_EXT_CAP_ID_ERR);
2384 if (!cap)
2385 return;
2387 ret = pci_read_config_dword(bridge, cap + PCI_ERR_CAP, &err_cap);
2388 if (ret) {
2389 dev_err(dev, "failed reading ext-conf-space of %s\n",
2390 pci_name(bridge));
2391 dev_err(dev, "\t pci=nommconf in use? "
2392 "or buggy/incomplete/absent ACPI MCFG attr?\n");
2393 return;
2395 if (!(err_cap & PCI_ERR_CAP_ECRC_GENC))
2396 return;
2398 err_cap |= PCI_ERR_CAP_ECRC_GENE;
2399 pci_write_config_dword(bridge, cap + PCI_ERR_CAP, err_cap);
2400 dev_info(dev, "Enabled ECRC on upstream bridge %s\n", pci_name(bridge));
2401 mgp->tx.boundary = 4096;
2402 mgp->fw_name = myri10ge_fw_aligned;
2406 * The Lanai Z8E PCI-E interface achieves higher Read-DMA throughput
2407 * when the PCI-E Completion packets are aligned on an 8-byte
2408 * boundary. Some PCI-E chip sets always align Completion packets; on
2409 * the ones that do not, the alignment can be enforced by enabling
2410 * ECRC generation (if supported).
2412 * When PCI-E Completion packets are not aligned, it is actually more
2413 * efficient to limit Read-DMA transactions to 2KB, rather than 4KB.
2415 * If the driver can neither enable ECRC nor verify that it has
2416 * already been enabled, then it must use a firmware image which works
2417 * around unaligned completion packets (myri10ge_ethp_z8e.dat), and it
2418 * should also ensure that it never gives the device a Read-DMA which is
2419 * larger than 2KB by setting the tx.boundary to 2KB. If ECRC is
2420 * enabled, then the driver should use the aligned (myri10ge_eth_z8e.dat)
2421 * firmware image, and set tx.boundary to 4KB.
2424 #define PCI_DEVICE_ID_INTEL_E5000_PCIE23 0x25f7
2425 #define PCI_DEVICE_ID_INTEL_E5000_PCIE47 0x25fa
2427 static void myri10ge_select_firmware(struct myri10ge_priv *mgp)
2429 struct pci_dev *bridge = mgp->pdev->bus->self;
2431 mgp->tx.boundary = 2048;
2432 mgp->fw_name = myri10ge_fw_unaligned;
2434 if (myri10ge_force_firmware == 0) {
2435 int link_width, exp_cap;
2436 u16 lnk;
2438 exp_cap = pci_find_capability(mgp->pdev, PCI_CAP_ID_EXP);
2439 pci_read_config_word(mgp->pdev, exp_cap + PCI_EXP_LNKSTA, &lnk);
2440 link_width = (lnk >> 4) & 0x3f;
2442 myri10ge_enable_ecrc(mgp);
2444 /* Check to see if Link is less than 8 or if the
2445 * upstream bridge is known to provide aligned
2446 * completions */
2447 if (link_width < 8) {
2448 dev_info(&mgp->pdev->dev, "PCIE x%d Link\n",
2449 link_width);
2450 mgp->tx.boundary = 4096;
2451 mgp->fw_name = myri10ge_fw_aligned;
2452 } else if (bridge &&
2453 /* ServerWorks HT2000/HT1000 */
2454 ((bridge->vendor == PCI_VENDOR_ID_SERVERWORKS
2455 && bridge->device ==
2456 PCI_DEVICE_ID_SERVERWORKS_HT2000_PCIE)
2457 /* All Intel E5000 PCIE ports */
2458 || (bridge->vendor == PCI_VENDOR_ID_INTEL
2459 && bridge->device >=
2460 PCI_DEVICE_ID_INTEL_E5000_PCIE23
2461 && bridge->device <=
2462 PCI_DEVICE_ID_INTEL_E5000_PCIE47))) {
2463 dev_info(&mgp->pdev->dev,
2464 "Assuming aligned completions (0x%x:0x%x)\n",
2465 bridge->vendor, bridge->device);
2466 mgp->tx.boundary = 4096;
2467 mgp->fw_name = myri10ge_fw_aligned;
2469 } else {
2470 if (myri10ge_force_firmware == 1) {
2471 dev_info(&mgp->pdev->dev,
2472 "Assuming aligned completions (forced)\n");
2473 mgp->tx.boundary = 4096;
2474 mgp->fw_name = myri10ge_fw_aligned;
2475 } else {
2476 dev_info(&mgp->pdev->dev,
2477 "Assuming unaligned completions (forced)\n");
2478 mgp->tx.boundary = 2048;
2479 mgp->fw_name = myri10ge_fw_unaligned;
2482 if (myri10ge_fw_name != NULL) {
2483 dev_info(&mgp->pdev->dev, "overriding firmware to %s\n",
2484 myri10ge_fw_name);
2485 mgp->fw_name = myri10ge_fw_name;
2489 static void myri10ge_save_state(struct myri10ge_priv *mgp)
2491 struct pci_dev *pdev = mgp->pdev;
2492 int cap;
2494 pci_save_state(pdev);
2495 /* now save PCIe and MSI state that Linux will not
2496 * save for us */
2497 cap = pci_find_capability(pdev, PCI_CAP_ID_EXP);
2498 pci_read_config_dword(pdev, cap + PCI_EXP_DEVCTL, &mgp->devctl);
2499 cap = pci_find_capability(pdev, PCI_CAP_ID_MSI);
2500 pci_read_config_word(pdev, cap + PCI_MSI_FLAGS, &mgp->msi_flags);
2503 static void myri10ge_restore_state(struct myri10ge_priv *mgp)
2505 struct pci_dev *pdev = mgp->pdev;
2506 int cap;
2508 /* restore PCIe and MSI state that linux will not */
2509 cap = pci_find_capability(pdev, PCI_CAP_ID_EXP);
2510 pci_write_config_dword(pdev, cap + PCI_CAP_ID_EXP, mgp->devctl);
2511 cap = pci_find_capability(pdev, PCI_CAP_ID_MSI);
2512 pci_write_config_word(pdev, cap + PCI_MSI_FLAGS, mgp->msi_flags);
2514 pci_restore_state(pdev);
2517 #ifdef CONFIG_PM
2519 static int myri10ge_suspend(struct pci_dev *pdev, pm_message_t state)
2521 struct myri10ge_priv *mgp;
2522 struct net_device *netdev;
2524 mgp = pci_get_drvdata(pdev);
2525 if (mgp == NULL)
2526 return -EINVAL;
2527 netdev = mgp->dev;
2529 netif_device_detach(netdev);
2530 if (netif_running(netdev)) {
2531 printk(KERN_INFO "myri10ge: closing %s\n", netdev->name);
2532 rtnl_lock();
2533 myri10ge_close(netdev);
2534 rtnl_unlock();
2536 myri10ge_dummy_rdma(mgp, 0);
2537 free_irq(pdev->irq, mgp);
2538 myri10ge_save_state(mgp);
2539 pci_disable_device(pdev);
2540 pci_set_power_state(pdev, pci_choose_state(pdev, state));
2541 return 0;
2544 static int myri10ge_resume(struct pci_dev *pdev)
2546 struct myri10ge_priv *mgp;
2547 struct net_device *netdev;
2548 int status;
2549 u16 vendor;
2551 mgp = pci_get_drvdata(pdev);
2552 if (mgp == NULL)
2553 return -EINVAL;
2554 netdev = mgp->dev;
2555 pci_set_power_state(pdev, 0); /* zeros conf space as a side effect */
2556 msleep(5); /* give card time to respond */
2557 pci_read_config_word(mgp->pdev, PCI_VENDOR_ID, &vendor);
2558 if (vendor == 0xffff) {
2559 printk(KERN_ERR "myri10ge: %s: device disappeared!\n",
2560 mgp->dev->name);
2561 return -EIO;
2563 myri10ge_restore_state(mgp);
2565 status = pci_enable_device(pdev);
2566 if (status < 0) {
2567 dev_err(&pdev->dev, "failed to enable device\n");
2568 return -EIO;
2571 pci_set_master(pdev);
2573 status = request_irq(pdev->irq, myri10ge_intr, IRQF_SHARED,
2574 netdev->name, mgp);
2575 if (status != 0) {
2576 dev_err(&pdev->dev, "failed to allocate IRQ\n");
2577 goto abort_with_enabled;
2580 myri10ge_reset(mgp);
2581 myri10ge_dummy_rdma(mgp, 1);
2583 /* Save configuration space to be restored if the
2584 * nic resets due to a parity error */
2585 myri10ge_save_state(mgp);
2587 if (netif_running(netdev)) {
2588 rtnl_lock();
2589 myri10ge_open(netdev);
2590 rtnl_unlock();
2592 netif_device_attach(netdev);
2594 return 0;
2596 abort_with_enabled:
2597 pci_disable_device(pdev);
2598 return -EIO;
2602 #endif /* CONFIG_PM */
2604 static u32 myri10ge_read_reboot(struct myri10ge_priv *mgp)
2606 struct pci_dev *pdev = mgp->pdev;
2607 int vs = mgp->vendor_specific_offset;
2608 u32 reboot;
2610 /*enter read32 mode */
2611 pci_write_config_byte(pdev, vs + 0x10, 0x3);
2613 /*read REBOOT_STATUS (0xfffffff0) */
2614 pci_write_config_dword(pdev, vs + 0x18, 0xfffffff0);
2615 pci_read_config_dword(pdev, vs + 0x14, &reboot);
2616 return reboot;
2620 * This watchdog is used to check whether the board has suffered
2621 * from a parity error and needs to be recovered.
2623 static void myri10ge_watchdog(void *arg)
2625 struct myri10ge_priv *mgp = arg;
2626 u32 reboot;
2627 int status;
2628 u16 cmd, vendor;
2630 mgp->watchdog_resets++;
2631 pci_read_config_word(mgp->pdev, PCI_COMMAND, &cmd);
2632 if ((cmd & PCI_COMMAND_MASTER) == 0) {
2633 /* Bus master DMA disabled? Check to see
2634 * if the card rebooted due to a parity error
2635 * For now, just report it */
2636 reboot = myri10ge_read_reboot(mgp);
2637 printk(KERN_ERR
2638 "myri10ge: %s: NIC rebooted (0x%x), resetting\n",
2639 mgp->dev->name, reboot);
2641 * A rebooted nic will come back with config space as
2642 * it was after power was applied to PCIe bus.
2643 * Attempt to restore config space which was saved
2644 * when the driver was loaded, or the last time the
2645 * nic was resumed from power saving mode.
2647 myri10ge_restore_state(mgp);
2648 } else {
2649 /* if we get back -1's from our slot, perhaps somebody
2650 * powered off our card. Don't try to reset it in
2651 * this case */
2652 if (cmd == 0xffff) {
2653 pci_read_config_word(mgp->pdev, PCI_VENDOR_ID, &vendor);
2654 if (vendor == 0xffff) {
2655 printk(KERN_ERR
2656 "myri10ge: %s: device disappeared!\n",
2657 mgp->dev->name);
2658 return;
2661 /* Perhaps it is a software error. Try to reset */
2663 printk(KERN_ERR "myri10ge: %s: device timeout, resetting\n",
2664 mgp->dev->name);
2665 printk(KERN_INFO "myri10ge: %s: %d %d %d %d %d\n",
2666 mgp->dev->name, mgp->tx.req, mgp->tx.done,
2667 mgp->tx.pkt_start, mgp->tx.pkt_done,
2668 (int)ntohl(mgp->fw_stats->send_done_count));
2669 msleep(2000);
2670 printk(KERN_INFO "myri10ge: %s: %d %d %d %d %d\n",
2671 mgp->dev->name, mgp->tx.req, mgp->tx.done,
2672 mgp->tx.pkt_start, mgp->tx.pkt_done,
2673 (int)ntohl(mgp->fw_stats->send_done_count));
2675 rtnl_lock();
2676 myri10ge_close(mgp->dev);
2677 status = myri10ge_load_firmware(mgp);
2678 if (status != 0)
2679 printk(KERN_ERR "myri10ge: %s: failed to load firmware\n",
2680 mgp->dev->name);
2681 else
2682 myri10ge_open(mgp->dev);
2683 rtnl_unlock();
2687 * We use our own timer routine rather than relying upon
2688 * netdev->tx_timeout because we have a very large hardware transmit
2689 * queue. Due to the large queue, the netdev->tx_timeout function
2690 * cannot detect a NIC with a parity error in a timely fashion if the
2691 * NIC is lightly loaded.
2693 static void myri10ge_watchdog_timer(unsigned long arg)
2695 struct myri10ge_priv *mgp;
2697 mgp = (struct myri10ge_priv *)arg;
2698 if (mgp->tx.req != mgp->tx.done &&
2699 mgp->tx.done == mgp->watchdog_tx_done &&
2700 mgp->watchdog_tx_req != mgp->watchdog_tx_done)
2701 /* nic seems like it might be stuck.. */
2702 schedule_work(&mgp->watchdog_work);
2703 else
2704 /* rearm timer */
2705 mod_timer(&mgp->watchdog_timer,
2706 jiffies + myri10ge_watchdog_timeout * HZ);
2708 mgp->watchdog_tx_done = mgp->tx.done;
2709 mgp->watchdog_tx_req = mgp->tx.req;
2712 static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2714 struct net_device *netdev;
2715 struct myri10ge_priv *mgp;
2716 struct device *dev = &pdev->dev;
2717 size_t bytes;
2718 int i;
2719 int status = -ENXIO;
2720 int cap;
2721 int dac_enabled;
2722 u16 val;
2724 netdev = alloc_etherdev(sizeof(*mgp));
2725 if (netdev == NULL) {
2726 dev_err(dev, "Could not allocate ethernet device\n");
2727 return -ENOMEM;
2730 mgp = netdev_priv(netdev);
2731 memset(mgp, 0, sizeof(*mgp));
2732 mgp->dev = netdev;
2733 mgp->pdev = pdev;
2734 mgp->csum_flag = MXGEFW_FLAGS_CKSUM;
2735 mgp->pause = myri10ge_flow_control;
2736 mgp->intr_coal_delay = myri10ge_intr_coal_delay;
2737 mgp->msg_enable = netif_msg_init(myri10ge_debug, MYRI10GE_MSG_DEFAULT);
2738 init_waitqueue_head(&mgp->down_wq);
2740 if (pci_enable_device(pdev)) {
2741 dev_err(&pdev->dev, "pci_enable_device call failed\n");
2742 status = -ENODEV;
2743 goto abort_with_netdev;
2745 myri10ge_select_firmware(mgp);
2747 /* Find the vendor-specific cap so we can check
2748 * the reboot register later on */
2749 mgp->vendor_specific_offset
2750 = pci_find_capability(pdev, PCI_CAP_ID_VNDR);
2752 /* Set our max read request to 4KB */
2753 cap = pci_find_capability(pdev, PCI_CAP_ID_EXP);
2754 if (cap < 64) {
2755 dev_err(&pdev->dev, "Bad PCI_CAP_ID_EXP location %d\n", cap);
2756 goto abort_with_netdev;
2758 status = pci_read_config_word(pdev, cap + PCI_EXP_DEVCTL, &val);
2759 if (status != 0) {
2760 dev_err(&pdev->dev, "Error %d reading PCI_EXP_DEVCTL\n",
2761 status);
2762 goto abort_with_netdev;
2764 val = (val & ~PCI_EXP_DEVCTL_READRQ) | (5 << 12);
2765 status = pci_write_config_word(pdev, cap + PCI_EXP_DEVCTL, val);
2766 if (status != 0) {
2767 dev_err(&pdev->dev, "Error %d writing PCI_EXP_DEVCTL\n",
2768 status);
2769 goto abort_with_netdev;
2772 pci_set_master(pdev);
2773 dac_enabled = 1;
2774 status = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
2775 if (status != 0) {
2776 dac_enabled = 0;
2777 dev_err(&pdev->dev,
2778 "64-bit pci address mask was refused, trying 32-bit");
2779 status = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
2781 if (status != 0) {
2782 dev_err(&pdev->dev, "Error %d setting DMA mask\n", status);
2783 goto abort_with_netdev;
2785 mgp->cmd = dma_alloc_coherent(&pdev->dev, sizeof(*mgp->cmd),
2786 &mgp->cmd_bus, GFP_KERNEL);
2787 if (mgp->cmd == NULL)
2788 goto abort_with_netdev;
2790 mgp->fw_stats = dma_alloc_coherent(&pdev->dev, sizeof(*mgp->fw_stats),
2791 &mgp->fw_stats_bus, GFP_KERNEL);
2792 if (mgp->fw_stats == NULL)
2793 goto abort_with_cmd;
2795 mgp->board_span = pci_resource_len(pdev, 0);
2796 mgp->iomem_base = pci_resource_start(pdev, 0);
2797 mgp->mtrr = -1;
2798 #ifdef CONFIG_MTRR
2799 mgp->mtrr = mtrr_add(mgp->iomem_base, mgp->board_span,
2800 MTRR_TYPE_WRCOMB, 1);
2801 #endif
2802 /* Hack. need to get rid of these magic numbers */
2803 mgp->sram_size =
2804 2 * 1024 * 1024 - (2 * (48 * 1024) + (32 * 1024)) - 0x100;
2805 if (mgp->sram_size > mgp->board_span) {
2806 dev_err(&pdev->dev, "board span %ld bytes too small\n",
2807 mgp->board_span);
2808 goto abort_with_wc;
2810 mgp->sram = ioremap(mgp->iomem_base, mgp->board_span);
2811 if (mgp->sram == NULL) {
2812 dev_err(&pdev->dev, "ioremap failed for %ld bytes at 0x%lx\n",
2813 mgp->board_span, mgp->iomem_base);
2814 status = -ENXIO;
2815 goto abort_with_wc;
2817 memcpy_fromio(mgp->eeprom_strings,
2818 mgp->sram + mgp->sram_size - MYRI10GE_EEPROM_STRINGS_SIZE,
2819 MYRI10GE_EEPROM_STRINGS_SIZE);
2820 memset(mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE - 2, 0, 2);
2821 status = myri10ge_read_mac_addr(mgp);
2822 if (status)
2823 goto abort_with_ioremap;
2825 for (i = 0; i < ETH_ALEN; i++)
2826 netdev->dev_addr[i] = mgp->mac_addr[i];
2828 /* allocate rx done ring */
2829 bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
2830 mgp->rx_done.entry = dma_alloc_coherent(&pdev->dev, bytes,
2831 &mgp->rx_done.bus, GFP_KERNEL);
2832 if (mgp->rx_done.entry == NULL)
2833 goto abort_with_ioremap;
2834 memset(mgp->rx_done.entry, 0, bytes);
2836 status = myri10ge_load_firmware(mgp);
2837 if (status != 0) {
2838 dev_err(&pdev->dev, "failed to load firmware\n");
2839 goto abort_with_rx_done;
2842 status = myri10ge_reset(mgp);
2843 if (status != 0) {
2844 dev_err(&pdev->dev, "failed reset\n");
2845 goto abort_with_firmware;
2848 if (myri10ge_msi) {
2849 status = pci_enable_msi(pdev);
2850 if (status != 0)
2851 dev_err(&pdev->dev,
2852 "Error %d setting up MSI; falling back to xPIC\n",
2853 status);
2854 else
2855 mgp->msi_enabled = 1;
2858 status = request_irq(pdev->irq, myri10ge_intr, IRQF_SHARED,
2859 netdev->name, mgp);
2860 if (status != 0) {
2861 dev_err(&pdev->dev, "failed to allocate IRQ\n");
2862 goto abort_with_firmware;
2865 pci_set_drvdata(pdev, mgp);
2866 if ((myri10ge_initial_mtu + ETH_HLEN) > MYRI10GE_MAX_ETHER_MTU)
2867 myri10ge_initial_mtu = MYRI10GE_MAX_ETHER_MTU - ETH_HLEN;
2868 if ((myri10ge_initial_mtu + ETH_HLEN) < 68)
2869 myri10ge_initial_mtu = 68;
2870 netdev->mtu = myri10ge_initial_mtu;
2871 netdev->open = myri10ge_open;
2872 netdev->stop = myri10ge_close;
2873 netdev->hard_start_xmit = myri10ge_xmit;
2874 netdev->get_stats = myri10ge_get_stats;
2875 netdev->base_addr = mgp->iomem_base;
2876 netdev->irq = pdev->irq;
2877 netdev->change_mtu = myri10ge_change_mtu;
2878 netdev->set_multicast_list = myri10ge_set_multicast_list;
2879 netdev->set_mac_address = myri10ge_set_mac_address;
2880 netdev->features = NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_TSO;
2881 if (dac_enabled)
2882 netdev->features |= NETIF_F_HIGHDMA;
2883 netdev->poll = myri10ge_poll;
2884 netdev->weight = myri10ge_napi_weight;
2886 /* Save configuration space to be restored if the
2887 * nic resets due to a parity error */
2888 myri10ge_save_state(mgp);
2890 /* Setup the watchdog timer */
2891 setup_timer(&mgp->watchdog_timer, myri10ge_watchdog_timer,
2892 (unsigned long)mgp);
2894 SET_ETHTOOL_OPS(netdev, &myri10ge_ethtool_ops);
2895 INIT_WORK(&mgp->watchdog_work, myri10ge_watchdog, mgp);
2896 status = register_netdev(netdev);
2897 if (status != 0) {
2898 dev_err(&pdev->dev, "register_netdev failed: %d\n", status);
2899 goto abort_with_irq;
2901 dev_info(dev, "%s IRQ %d, tx bndry %d, fw %s, WC %s\n",
2902 (mgp->msi_enabled ? "MSI" : "xPIC"),
2903 pdev->irq, mgp->tx.boundary, mgp->fw_name,
2904 (mgp->mtrr >= 0 ? "Enabled" : "Disabled"));
2906 return 0;
2908 abort_with_irq:
2909 free_irq(pdev->irq, mgp);
2910 if (mgp->msi_enabled)
2911 pci_disable_msi(pdev);
2913 abort_with_firmware:
2914 myri10ge_dummy_rdma(mgp, 0);
2916 abort_with_rx_done:
2917 bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
2918 dma_free_coherent(&pdev->dev, bytes,
2919 mgp->rx_done.entry, mgp->rx_done.bus);
2921 abort_with_ioremap:
2922 iounmap(mgp->sram);
2924 abort_with_wc:
2925 #ifdef CONFIG_MTRR
2926 if (mgp->mtrr >= 0)
2927 mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span);
2928 #endif
2929 dma_free_coherent(&pdev->dev, sizeof(*mgp->fw_stats),
2930 mgp->fw_stats, mgp->fw_stats_bus);
2932 abort_with_cmd:
2933 dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
2934 mgp->cmd, mgp->cmd_bus);
2936 abort_with_netdev:
2938 free_netdev(netdev);
2939 return status;
2943 * myri10ge_remove
2945 * Does what is necessary to shutdown one Myrinet device. Called
2946 * once for each Myrinet card by the kernel when a module is
2947 * unloaded.
2949 static void myri10ge_remove(struct pci_dev *pdev)
2951 struct myri10ge_priv *mgp;
2952 struct net_device *netdev;
2953 size_t bytes;
2955 mgp = pci_get_drvdata(pdev);
2956 if (mgp == NULL)
2957 return;
2959 flush_scheduled_work();
2960 netdev = mgp->dev;
2961 unregister_netdev(netdev);
2962 free_irq(pdev->irq, mgp);
2963 if (mgp->msi_enabled)
2964 pci_disable_msi(pdev);
2966 myri10ge_dummy_rdma(mgp, 0);
2968 bytes = myri10ge_max_intr_slots * sizeof(*mgp->rx_done.entry);
2969 dma_free_coherent(&pdev->dev, bytes,
2970 mgp->rx_done.entry, mgp->rx_done.bus);
2972 iounmap(mgp->sram);
2974 #ifdef CONFIG_MTRR
2975 if (mgp->mtrr >= 0)
2976 mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span);
2977 #endif
2978 dma_free_coherent(&pdev->dev, sizeof(*mgp->fw_stats),
2979 mgp->fw_stats, mgp->fw_stats_bus);
2981 dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
2982 mgp->cmd, mgp->cmd_bus);
2984 free_netdev(netdev);
2985 pci_set_drvdata(pdev, NULL);
2988 #define PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E 0x0008
2990 static struct pci_device_id myri10ge_pci_tbl[] = {
2991 {PCI_DEVICE(PCI_VENDOR_ID_MYRICOM, PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E)},
2992 {0},
2995 static struct pci_driver myri10ge_driver = {
2996 .name = "myri10ge",
2997 .probe = myri10ge_probe,
2998 .remove = myri10ge_remove,
2999 .id_table = myri10ge_pci_tbl,
3000 #ifdef CONFIG_PM
3001 .suspend = myri10ge_suspend,
3002 .resume = myri10ge_resume,
3003 #endif
3006 static __init int myri10ge_init_module(void)
3008 printk(KERN_INFO "%s: Version %s\n", myri10ge_driver.name,
3009 MYRI10GE_VERSION_STR);
3010 return pci_register_driver(&myri10ge_driver);
3013 module_init(myri10ge_init_module);
3015 static __exit void myri10ge_cleanup_module(void)
3017 pci_unregister_driver(&myri10ge_driver);
3020 module_exit(myri10ge_cleanup_module);