1 /*************************************************************************
2 * myri10ge.c: Myricom Myri-10G Ethernet driver.
4 * Copyright (C) 2005, 2006 Myricom, Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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:
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>
52 #include <linux/inet.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>
63 #include <net/checksum.h>
64 #include <asm/byteorder.h>
66 #include <asm/processor.h>
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 0xffffffff
93 #define MYRI10GE_NO_RESPONSE_RESULT 0xffffffff
95 struct myri10ge_rx_buffer_state
{
97 DECLARE_PCI_UNMAP_ADDR(bus
)
98 DECLARE_PCI_UNMAP_LEN(len
)
101 struct myri10ge_tx_buffer_state
{
104 DECLARE_PCI_UNMAP_ADDR(bus
)
105 DECLARE_PCI_UNMAP_LEN(len
)
108 struct myri10ge_cmd
{
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
;
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 */
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
;
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
;
153 struct net_device
*dev
;
154 struct net_device_stats stats
;
157 unsigned long board_span
;
158 unsigned long iomem_base
;
159 u32 __iomem
*irq_claim
;
160 u32 __iomem
*irq_deassert
;
161 char *mac_addr_string
;
162 struct mcp_cmd_response
*cmd
;
164 struct mcp_irq_data
*fw_stats
;
165 dma_addr_t fw_stats_bus
;
166 struct pci_dev
*pdev
;
168 unsigned int link_state
;
169 unsigned int rdma_tags_available
;
171 u32 __iomem
*intr_coal_delay_ptr
;
176 wait_queue_head_t down_wq
;
177 struct work_struct watchdog_work
;
178 struct timer_list watchdog_timer
;
179 int watchdog_tx_done
;
184 char eeprom_strings
[MYRI10GE_EEPROM_STRINGS_SIZE
];
185 char fw_version
[128];
186 u8 mac_addr
[6]; /* eeprom mac address */
187 unsigned long serial_number
;
188 int vendor_specific_offset
;
196 static char *myri10ge_fw_unaligned
= "myri10ge_ethp_z8e.dat";
197 static char *myri10ge_fw_aligned
= "myri10ge_eth_z8e.dat";
199 static char *myri10ge_fw_name
= NULL
;
200 module_param(myri10ge_fw_name
, charp
, S_IRUGO
| S_IWUSR
);
201 MODULE_PARM_DESC(myri10ge_fw_name
, "Firmware image name\n");
203 static int myri10ge_ecrc_enable
= 1;
204 module_param(myri10ge_ecrc_enable
, int, S_IRUGO
);
205 MODULE_PARM_DESC(myri10ge_ecrc_enable
, "Enable Extended CRC on PCI-E\n");
207 static int myri10ge_max_intr_slots
= 1024;
208 module_param(myri10ge_max_intr_slots
, int, S_IRUGO
);
209 MODULE_PARM_DESC(myri10ge_max_intr_slots
, "Interrupt queue slots\n");
211 static int myri10ge_small_bytes
= -1; /* -1 == auto */
212 module_param(myri10ge_small_bytes
, int, S_IRUGO
| S_IWUSR
);
213 MODULE_PARM_DESC(myri10ge_small_bytes
, "Threshold of small packets\n");
215 static int myri10ge_msi
= 1; /* enable msi by default */
216 module_param(myri10ge_msi
, int, S_IRUGO
);
217 MODULE_PARM_DESC(myri10ge_msi
, "Enable Message Signalled Interrupts\n");
219 static int myri10ge_intr_coal_delay
= 25;
220 module_param(myri10ge_intr_coal_delay
, int, S_IRUGO
);
221 MODULE_PARM_DESC(myri10ge_intr_coal_delay
, "Interrupt coalescing delay\n");
223 static int myri10ge_flow_control
= 1;
224 module_param(myri10ge_flow_control
, int, S_IRUGO
);
225 MODULE_PARM_DESC(myri10ge_flow_control
, "Pause parameter\n");
227 static int myri10ge_deassert_wait
= 1;
228 module_param(myri10ge_deassert_wait
, int, S_IRUGO
| S_IWUSR
);
229 MODULE_PARM_DESC(myri10ge_deassert_wait
,
230 "Wait when deasserting legacy interrupts\n");
232 static int myri10ge_force_firmware
= 0;
233 module_param(myri10ge_force_firmware
, int, S_IRUGO
);
234 MODULE_PARM_DESC(myri10ge_force_firmware
,
235 "Force firmware to assume aligned completions\n");
237 static int myri10ge_skb_cross_4k
= 0;
238 module_param(myri10ge_skb_cross_4k
, int, S_IRUGO
| S_IWUSR
);
239 MODULE_PARM_DESC(myri10ge_skb_cross_4k
,
240 "Can a small skb cross a 4KB boundary?\n");
242 static int myri10ge_initial_mtu
= MYRI10GE_MAX_ETHER_MTU
- ETH_HLEN
;
243 module_param(myri10ge_initial_mtu
, int, S_IRUGO
);
244 MODULE_PARM_DESC(myri10ge_initial_mtu
, "Initial MTU\n");
246 static int myri10ge_napi_weight
= 64;
247 module_param(myri10ge_napi_weight
, int, S_IRUGO
);
248 MODULE_PARM_DESC(myri10ge_napi_weight
, "Set NAPI weight\n");
250 static int myri10ge_watchdog_timeout
= 1;
251 module_param(myri10ge_watchdog_timeout
, int, S_IRUGO
);
252 MODULE_PARM_DESC(myri10ge_watchdog_timeout
, "Set watchdog timeout\n");
254 static int myri10ge_max_irq_loops
= 1048576;
255 module_param(myri10ge_max_irq_loops
, int, S_IRUGO
);
256 MODULE_PARM_DESC(myri10ge_max_irq_loops
,
257 "Set stuck legacy IRQ detection threshold\n");
259 #define MYRI10GE_FW_OFFSET 1024*1024
260 #define MYRI10GE_HIGHPART_TO_U32(X) \
261 (sizeof (X) == 8) ? ((u32)((u64)(X) >> 32)) : (0)
262 #define MYRI10GE_LOWPART_TO_U32(X) ((u32)(X))
264 #define myri10ge_pio_copy(to,from,size) __iowrite64_copy(to,from,size/8)
267 myri10ge_send_cmd(struct myri10ge_priv
*mgp
, u32 cmd
,
268 struct myri10ge_cmd
*data
, int atomic
)
271 char buf_bytes
[sizeof(*buf
) + 8];
272 struct mcp_cmd_response
*response
= mgp
->cmd
;
273 char __iomem
*cmd_addr
= mgp
->sram
+ MXGEFW_CMD_OFFSET
;
274 u32 dma_low
, dma_high
, result
, value
;
277 /* ensure buf is aligned to 8 bytes */
278 buf
= (struct mcp_cmd
*)ALIGN((unsigned long)buf_bytes
, 8);
280 buf
->data0
= htonl(data
->data0
);
281 buf
->data1
= htonl(data
->data1
);
282 buf
->data2
= htonl(data
->data2
);
283 buf
->cmd
= htonl(cmd
);
284 dma_low
= MYRI10GE_LOWPART_TO_U32(mgp
->cmd_bus
);
285 dma_high
= MYRI10GE_HIGHPART_TO_U32(mgp
->cmd_bus
);
287 buf
->response_addr
.low
= htonl(dma_low
);
288 buf
->response_addr
.high
= htonl(dma_high
);
289 response
->result
= MYRI10GE_NO_RESPONSE_RESULT
;
291 myri10ge_pio_copy(cmd_addr
, buf
, sizeof(*buf
));
293 /* wait up to 15ms. Longest command is the DMA benchmark,
294 * which is capped at 5ms, but runs from a timeout handler
295 * that runs every 7.8ms. So a 15ms timeout leaves us with
299 /* if atomic is set, do not sleep,
300 * and try to get the completion quickly
301 * (1ms will be enough for those commands) */
302 for (sleep_total
= 0;
304 && response
->result
== MYRI10GE_NO_RESPONSE_RESULT
;
308 /* use msleep for most command */
309 for (sleep_total
= 0;
311 && response
->result
== MYRI10GE_NO_RESPONSE_RESULT
;
316 result
= ntohl(response
->result
);
317 value
= ntohl(response
->data
);
318 if (result
!= MYRI10GE_NO_RESPONSE_RESULT
) {
323 dev_err(&mgp
->pdev
->dev
,
324 "command %d failed, result = %d\n",
330 dev_err(&mgp
->pdev
->dev
, "command %d timed out, result = %d\n",
336 * The eeprom strings on the lanaiX have the format
339 * PT:ddd mmm xx xx:xx:xx xx\0
340 * PV:ddd mmm xx xx:xx:xx xx\0
342 static int myri10ge_read_mac_addr(struct myri10ge_priv
*mgp
)
347 ptr
= mgp
->eeprom_strings
;
348 limit
= mgp
->eeprom_strings
+ MYRI10GE_EEPROM_STRINGS_SIZE
;
350 while (*ptr
!= '\0' && ptr
< limit
) {
351 if (memcmp(ptr
, "MAC=", 4) == 0) {
353 mgp
->mac_addr_string
= ptr
;
354 for (i
= 0; i
< 6; i
++) {
355 if ((ptr
+ 2) > limit
)
358 simple_strtoul(ptr
, &ptr
, 16);
362 if (memcmp((const void *)ptr
, "SN=", 3) == 0) {
364 mgp
->serial_number
= simple_strtoul(ptr
, &ptr
, 10);
366 while (ptr
< limit
&& *ptr
++) ;
372 dev_err(&mgp
->pdev
->dev
, "failed to parse eeprom_strings\n");
377 * Enable or disable periodic RDMAs from the host to make certain
378 * chipsets resend dropped PCIe messages
381 static void myri10ge_dummy_rdma(struct myri10ge_priv
*mgp
, int enable
)
383 char __iomem
*submit
;
385 u32 dma_low
, dma_high
;
388 /* clear confirmation addr */
392 /* send a rdma command to the PCIe engine, and wait for the
393 * response in the confirmation address. The firmware should
394 * write a -1 there to indicate it is alive and well
396 dma_low
= MYRI10GE_LOWPART_TO_U32(mgp
->cmd_bus
);
397 dma_high
= MYRI10GE_HIGHPART_TO_U32(mgp
->cmd_bus
);
399 buf
[0] = htonl(dma_high
); /* confirm addr MSW */
400 buf
[1] = htonl(dma_low
); /* confirm addr LSW */
401 buf
[2] = htonl(MYRI10GE_NO_CONFIRM_DATA
); /* confirm data */
402 buf
[3] = htonl(dma_high
); /* dummy addr MSW */
403 buf
[4] = htonl(dma_low
); /* dummy addr LSW */
404 buf
[5] = htonl(enable
); /* enable? */
406 submit
= mgp
->sram
+ 0xfc01c0;
408 myri10ge_pio_copy(submit
, &buf
, sizeof(buf
));
409 for (i
= 0; mgp
->cmd
->data
!= MYRI10GE_NO_CONFIRM_DATA
&& i
< 20; i
++)
411 if (mgp
->cmd
->data
!= MYRI10GE_NO_CONFIRM_DATA
)
412 dev_err(&mgp
->pdev
->dev
, "dummy rdma %s failed\n",
413 (enable
? "enable" : "disable"));
417 myri10ge_validate_firmware(struct myri10ge_priv
*mgp
,
418 struct mcp_gen_header
*hdr
)
420 struct device
*dev
= &mgp
->pdev
->dev
;
423 /* check firmware type */
424 if (ntohl(hdr
->mcp_type
) != MCP_TYPE_ETH
) {
425 dev_err(dev
, "Bad firmware type: 0x%x\n", ntohl(hdr
->mcp_type
));
429 /* save firmware version for ethtool */
430 strncpy(mgp
->fw_version
, hdr
->version
, sizeof(mgp
->fw_version
));
432 sscanf(mgp
->fw_version
, "%d.%d", &major
, &minor
);
434 if (!(major
== MXGEFW_VERSION_MAJOR
&& minor
== MXGEFW_VERSION_MINOR
)) {
435 dev_err(dev
, "Found firmware version %s\n", mgp
->fw_version
);
436 dev_err(dev
, "Driver needs %d.%d\n", MXGEFW_VERSION_MAJOR
,
437 MXGEFW_VERSION_MINOR
);
443 static int myri10ge_load_hotplug_firmware(struct myri10ge_priv
*mgp
, u32
* size
)
445 unsigned crc
, reread_crc
;
446 const struct firmware
*fw
;
447 struct device
*dev
= &mgp
->pdev
->dev
;
448 struct mcp_gen_header
*hdr
;
453 if ((status
= request_firmware(&fw
, mgp
->fw_name
, dev
)) < 0) {
454 dev_err(dev
, "Unable to load %s firmware image via hotplug\n",
457 goto abort_with_nothing
;
462 if (fw
->size
>= mgp
->sram_size
- MYRI10GE_FW_OFFSET
||
463 fw
->size
< MCP_HEADER_PTR_OFFSET
+ 4) {
464 dev_err(dev
, "Firmware size invalid:%d\n", (int)fw
->size
);
470 hdr_offset
= ntohl(*(u32
*) (fw
->data
+ MCP_HEADER_PTR_OFFSET
));
471 if ((hdr_offset
& 3) || hdr_offset
+ sizeof(*hdr
) > fw
->size
) {
472 dev_err(dev
, "Bad firmware file\n");
476 hdr
= (void *)(fw
->data
+ hdr_offset
);
478 status
= myri10ge_validate_firmware(mgp
, hdr
);
482 crc
= crc32(~0, fw
->data
, fw
->size
);
483 for (i
= 0; i
< fw
->size
; i
+= 256) {
484 myri10ge_pio_copy(mgp
->sram
+ MYRI10GE_FW_OFFSET
+ i
,
486 min(256U, (unsigned)(fw
->size
- i
)));
490 /* corruption checking is good for parity recovery and buggy chipset */
491 memcpy_fromio(fw
->data
, mgp
->sram
+ MYRI10GE_FW_OFFSET
, fw
->size
);
492 reread_crc
= crc32(~0, fw
->data
, fw
->size
);
493 if (crc
!= reread_crc
) {
494 dev_err(dev
, "CRC failed(fw-len=%u), got 0x%x (expect 0x%x)\n",
495 (unsigned)fw
->size
, reread_crc
, crc
);
499 *size
= (u32
) fw
->size
;
502 release_firmware(fw
);
508 static int myri10ge_adopt_running_firmware(struct myri10ge_priv
*mgp
)
510 struct mcp_gen_header
*hdr
;
511 struct device
*dev
= &mgp
->pdev
->dev
;
512 const size_t bytes
= sizeof(struct mcp_gen_header
);
516 /* find running firmware header */
517 hdr_offset
= ntohl(__raw_readl(mgp
->sram
+ MCP_HEADER_PTR_OFFSET
));
519 if ((hdr_offset
& 3) || hdr_offset
+ sizeof(*hdr
) > mgp
->sram_size
) {
520 dev_err(dev
, "Running firmware has bad header offset (%d)\n",
525 /* copy header of running firmware from SRAM to host memory to
526 * validate firmware */
527 hdr
= kmalloc(bytes
, GFP_KERNEL
);
529 dev_err(dev
, "could not malloc firmware hdr\n");
532 memcpy_fromio(hdr
, mgp
->sram
+ hdr_offset
, bytes
);
533 status
= myri10ge_validate_firmware(mgp
, hdr
);
538 static int myri10ge_load_firmware(struct myri10ge_priv
*mgp
)
540 char __iomem
*submit
;
542 u32 dma_low
, dma_high
, size
;
546 status
= myri10ge_load_hotplug_firmware(mgp
, &size
);
548 dev_warn(&mgp
->pdev
->dev
, "hotplug firmware loading failed\n");
550 /* Do not attempt to adopt firmware if there
555 status
= myri10ge_adopt_running_firmware(mgp
);
557 dev_err(&mgp
->pdev
->dev
,
558 "failed to adopt running firmware\n");
561 dev_info(&mgp
->pdev
->dev
,
562 "Successfully adopted running firmware\n");
563 if (mgp
->tx
.boundary
== 4096) {
564 dev_warn(&mgp
->pdev
->dev
,
565 "Using firmware currently running on NIC"
567 dev_warn(&mgp
->pdev
->dev
,
568 "performance consider loading optimized "
570 dev_warn(&mgp
->pdev
->dev
, "via hotplug\n");
573 mgp
->fw_name
= "adopted";
574 mgp
->tx
.boundary
= 2048;
578 /* clear confirmation addr */
582 /* send a reload command to the bootstrap MCP, and wait for the
583 * response in the confirmation address. The firmware should
584 * write a -1 there to indicate it is alive and well
586 dma_low
= MYRI10GE_LOWPART_TO_U32(mgp
->cmd_bus
);
587 dma_high
= MYRI10GE_HIGHPART_TO_U32(mgp
->cmd_bus
);
589 buf
[0] = htonl(dma_high
); /* confirm addr MSW */
590 buf
[1] = htonl(dma_low
); /* confirm addr LSW */
591 buf
[2] = htonl(MYRI10GE_NO_CONFIRM_DATA
); /* confirm data */
593 /* FIX: All newest firmware should un-protect the bottom of
594 * the sram before handoff. However, the very first interfaces
595 * do not. Therefore the handoff copy must skip the first 8 bytes
597 buf
[3] = htonl(MYRI10GE_FW_OFFSET
+ 8); /* where the code starts */
598 buf
[4] = htonl(size
- 8); /* length of code */
599 buf
[5] = htonl(8); /* where to copy to */
600 buf
[6] = htonl(0); /* where to jump to */
602 submit
= mgp
->sram
+ 0xfc0000;
604 myri10ge_pio_copy(submit
, &buf
, sizeof(buf
));
609 while (mgp
->cmd
->data
!= MYRI10GE_NO_CONFIRM_DATA
&& i
< 20) {
613 if (mgp
->cmd
->data
!= MYRI10GE_NO_CONFIRM_DATA
) {
614 dev_err(&mgp
->pdev
->dev
, "handoff failed\n");
617 dev_info(&mgp
->pdev
->dev
, "handoff confirmed\n");
618 myri10ge_dummy_rdma(mgp
, 1);
623 static int myri10ge_update_mac_address(struct myri10ge_priv
*mgp
, u8
* addr
)
625 struct myri10ge_cmd cmd
;
628 cmd
.data0
= ((addr
[0] << 24) | (addr
[1] << 16)
629 | (addr
[2] << 8) | addr
[3]);
631 cmd
.data1
= ((addr
[4] << 8) | (addr
[5]));
633 status
= myri10ge_send_cmd(mgp
, MXGEFW_SET_MAC_ADDRESS
, &cmd
, 0);
637 static int myri10ge_change_pause(struct myri10ge_priv
*mgp
, int pause
)
639 struct myri10ge_cmd cmd
;
642 ctl
= pause
? MXGEFW_ENABLE_FLOW_CONTROL
: MXGEFW_DISABLE_FLOW_CONTROL
;
643 status
= myri10ge_send_cmd(mgp
, ctl
, &cmd
, 0);
647 "myri10ge: %s: Failed to set flow control mode\n",
656 myri10ge_change_promisc(struct myri10ge_priv
*mgp
, int promisc
, int atomic
)
658 struct myri10ge_cmd cmd
;
661 ctl
= promisc
? MXGEFW_ENABLE_PROMISC
: MXGEFW_DISABLE_PROMISC
;
662 status
= myri10ge_send_cmd(mgp
, ctl
, &cmd
, atomic
);
664 printk(KERN_ERR
"myri10ge: %s: Failed to set promisc mode\n",
668 static int myri10ge_reset(struct myri10ge_priv
*mgp
)
670 struct myri10ge_cmd cmd
;
675 /* try to send a reset command to the card to see if it
677 memset(&cmd
, 0, sizeof(cmd
));
678 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_RESET
, &cmd
, 0);
680 dev_err(&mgp
->pdev
->dev
, "failed reset\n");
684 /* Now exchange information about interrupts */
686 bytes
= myri10ge_max_intr_slots
* sizeof(*mgp
->rx_done
.entry
);
687 memset(mgp
->rx_done
.entry
, 0, bytes
);
688 cmd
.data0
= (u32
) bytes
;
689 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_INTRQ_SIZE
, &cmd
, 0);
690 cmd
.data0
= MYRI10GE_LOWPART_TO_U32(mgp
->rx_done
.bus
);
691 cmd
.data1
= MYRI10GE_HIGHPART_TO_U32(mgp
->rx_done
.bus
);
692 status
|= myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_INTRQ_DMA
, &cmd
, 0);
695 myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_IRQ_ACK_OFFSET
, &cmd
, 0);
696 mgp
->irq_claim
= (__iomem u32
*) (mgp
->sram
+ cmd
.data0
);
697 if (!mgp
->msi_enabled
) {
698 status
|= myri10ge_send_cmd
699 (mgp
, MXGEFW_CMD_GET_IRQ_DEASSERT_OFFSET
, &cmd
, 0);
700 mgp
->irq_deassert
= (__iomem u32
*) (mgp
->sram
+ cmd
.data0
);
703 status
|= myri10ge_send_cmd
704 (mgp
, MXGEFW_CMD_GET_INTR_COAL_DELAY_OFFSET
, &cmd
, 0);
705 mgp
->intr_coal_delay_ptr
= (__iomem u32
*) (mgp
->sram
+ cmd
.data0
);
707 dev_err(&mgp
->pdev
->dev
, "failed set interrupt parameters\n");
710 __raw_writel(htonl(mgp
->intr_coal_delay
), mgp
->intr_coal_delay_ptr
);
712 /* Run a small DMA test.
713 * The magic multipliers to the length tell the firmware
714 * to do DMA read, write, or read+write tests. The
715 * results are returned in cmd.data0. The upper 16
716 * bits or the return is the number of transfers completed.
717 * The lower 16 bits is the time in 0.5us ticks that the
718 * transfers took to complete.
721 len
= mgp
->tx
.boundary
;
723 cmd
.data0
= MYRI10GE_LOWPART_TO_U32(mgp
->rx_done
.bus
);
724 cmd
.data1
= MYRI10GE_HIGHPART_TO_U32(mgp
->rx_done
.bus
);
725 cmd
.data2
= len
* 0x10000;
726 status
= myri10ge_send_cmd(mgp
, MXGEFW_DMA_TEST
, &cmd
, 0);
728 mgp
->read_dma
= ((cmd
.data0
>> 16) * len
* 2) /
729 (cmd
.data0
& 0xffff);
731 dev_warn(&mgp
->pdev
->dev
, "DMA read benchmark failed: %d\n",
733 cmd
.data0
= MYRI10GE_LOWPART_TO_U32(mgp
->rx_done
.bus
);
734 cmd
.data1
= MYRI10GE_HIGHPART_TO_U32(mgp
->rx_done
.bus
);
735 cmd
.data2
= len
* 0x1;
736 status
= myri10ge_send_cmd(mgp
, MXGEFW_DMA_TEST
, &cmd
, 0);
738 mgp
->write_dma
= ((cmd
.data0
>> 16) * len
* 2) /
739 (cmd
.data0
& 0xffff);
741 dev_warn(&mgp
->pdev
->dev
, "DMA write benchmark failed: %d\n",
744 cmd
.data0
= MYRI10GE_LOWPART_TO_U32(mgp
->rx_done
.bus
);
745 cmd
.data1
= MYRI10GE_HIGHPART_TO_U32(mgp
->rx_done
.bus
);
746 cmd
.data2
= len
* 0x10001;
747 status
= myri10ge_send_cmd(mgp
, MXGEFW_DMA_TEST
, &cmd
, 0);
749 mgp
->read_write_dma
= ((cmd
.data0
>> 16) * len
* 2 * 2) /
750 (cmd
.data0
& 0xffff);
752 dev_warn(&mgp
->pdev
->dev
,
753 "DMA read/write benchmark failed: %d\n", status
);
755 memset(mgp
->rx_done
.entry
, 0, bytes
);
757 /* reset mcp/driver shared state back to 0 */
760 mgp
->tx
.pkt_start
= 0;
761 mgp
->tx
.pkt_done
= 0;
763 mgp
->rx_small
.cnt
= 0;
764 mgp
->rx_done
.idx
= 0;
765 mgp
->rx_done
.cnt
= 0;
766 status
= myri10ge_update_mac_address(mgp
, mgp
->dev
->dev_addr
);
767 myri10ge_change_promisc(mgp
, 0, 0);
768 myri10ge_change_pause(mgp
, mgp
->pause
);
773 myri10ge_submit_8rx(struct mcp_kreq_ether_recv __iomem
* dst
,
774 struct mcp_kreq_ether_recv
*src
)
779 src
->addr_low
= DMA_32BIT_MASK
;
780 myri10ge_pio_copy(dst
, src
, 8 * sizeof(*src
));
783 __raw_writel(low
, &dst
->addr_low
);
788 * Set of routines to get a new receive buffer. Any buffer which
789 * crosses a 4KB boundary must start on a 4KB boundary due to PCIe
790 * wdma restrictions. We also try to align any smaller allocation to
791 * at least a 16 byte boundary for efficiency. We assume the linux
792 * memory allocator works by powers of 2, and will not return memory
793 * smaller than 2KB which crosses a 4KB boundary. If it does, we fall
794 * back to allocating 2x as much space as required.
796 * We intend to replace large (>4KB) skb allocations by using
797 * pages directly and building a fraglist in the near future.
800 static inline struct sk_buff
*myri10ge_alloc_big(int bytes
)
803 unsigned long data
, roundup
;
805 skb
= dev_alloc_skb(bytes
+ 4096 + MXGEFW_PAD
);
809 /* Correct skb->truesize so that socket buffer
810 * accounting is not confused the rounding we must
811 * do to satisfy alignment constraints.
813 skb
->truesize
-= 4096;
815 data
= (unsigned long)(skb
->data
);
816 roundup
= (-data
) & (4095);
817 skb_reserve(skb
, roundup
);
821 /* Allocate 2x as much space as required and use whichever portion
822 * does not cross a 4KB boundary */
823 static inline struct sk_buff
*myri10ge_alloc_small_safe(unsigned int bytes
)
826 unsigned long data
, boundary
;
828 skb
= dev_alloc_skb(2 * (bytes
+ MXGEFW_PAD
) - 1);
829 if (unlikely(skb
== NULL
))
832 /* Correct skb->truesize so that socket buffer
833 * accounting is not confused the rounding we must
834 * do to satisfy alignment constraints.
836 skb
->truesize
-= bytes
+ MXGEFW_PAD
;
838 data
= (unsigned long)(skb
->data
);
839 boundary
= (data
+ 4095UL) & ~4095UL;
840 if ((boundary
- data
) >= (bytes
+ MXGEFW_PAD
))
843 skb_reserve(skb
, boundary
- data
);
847 /* Allocate just enough space, and verify that the allocated
848 * space does not cross a 4KB boundary */
849 static inline struct sk_buff
*myri10ge_alloc_small(int bytes
)
852 unsigned long roundup
, data
, end
;
854 skb
= dev_alloc_skb(bytes
+ 16 + MXGEFW_PAD
);
855 if (unlikely(skb
== NULL
))
858 /* Round allocated buffer to 16 byte boundary */
859 data
= (unsigned long)(skb
->data
);
860 roundup
= (-data
) & 15UL;
861 skb_reserve(skb
, roundup
);
862 /* Verify that the data buffer does not cross a page boundary */
863 data
= (unsigned long)(skb
->data
);
864 end
= data
+ bytes
+ MXGEFW_PAD
- 1;
865 if (unlikely(((end
>> 12) != (data
>> 12)) && (data
& 4095UL))) {
867 "myri10ge_alloc_small: small skb crossed 4KB boundary\n");
868 myri10ge_skb_cross_4k
= 1;
869 dev_kfree_skb_any(skb
);
870 skb
= myri10ge_alloc_small_safe(bytes
);
876 myri10ge_getbuf(struct myri10ge_rx_buf
*rx
, struct pci_dev
*pdev
, int bytes
,
883 bytes
+= VLAN_HLEN
; /* account for 802.1q vlan tag */
885 if ((bytes
+ MXGEFW_PAD
) > (4096 - 16) /* linux overhead */ )
886 skb
= myri10ge_alloc_big(bytes
);
887 else if (myri10ge_skb_cross_4k
)
888 skb
= myri10ge_alloc_small_safe(bytes
);
890 skb
= myri10ge_alloc_small(bytes
);
892 if (unlikely(skb
== NULL
)) {
898 /* set len so that it only covers the area we
899 * need mapped for DMA */
900 len
= bytes
+ MXGEFW_PAD
;
902 bus
= pci_map_single(pdev
, skb
->data
, len
, PCI_DMA_FROMDEVICE
);
903 rx
->info
[idx
].skb
= skb
;
904 pci_unmap_addr_set(&rx
->info
[idx
], bus
, bus
);
905 pci_unmap_len_set(&rx
->info
[idx
], len
, len
);
906 rx
->shadow
[idx
].addr_low
= htonl(MYRI10GE_LOWPART_TO_U32(bus
));
907 rx
->shadow
[idx
].addr_high
= htonl(MYRI10GE_HIGHPART_TO_U32(bus
));
910 /* copy 8 descriptors (64-bytes) to the mcp at a time */
911 if ((idx
& 7) == 7) {
912 if (rx
->wc_fifo
== NULL
)
913 myri10ge_submit_8rx(&rx
->lanai
[idx
- 7],
914 &rx
->shadow
[idx
- 7]);
917 myri10ge_pio_copy(rx
->wc_fifo
,
918 &rx
->shadow
[idx
- 7], 64);
924 static inline void myri10ge_vlan_ip_csum(struct sk_buff
*skb
, u16 hw_csum
)
926 struct vlan_hdr
*vh
= (struct vlan_hdr
*)(skb
->data
);
928 if ((skb
->protocol
== ntohs(ETH_P_8021Q
)) &&
929 (vh
->h_vlan_encapsulated_proto
== htons(ETH_P_IP
) ||
930 vh
->h_vlan_encapsulated_proto
== htons(ETH_P_IPV6
))) {
932 skb
->ip_summed
= CHECKSUM_HW
;
936 static inline unsigned long
937 myri10ge_rx_done(struct myri10ge_priv
*mgp
, struct myri10ge_rx_buf
*rx
,
938 int bytes
, int len
, int csum
)
944 idx
= rx
->cnt
& rx
->mask
;
947 /* save a pointer to the received skb */
948 skb
= rx
->info
[idx
].skb
;
949 bus
= pci_unmap_addr(&rx
->info
[idx
], bus
);
950 unmap_len
= pci_unmap_len(&rx
->info
[idx
], len
);
952 /* try to replace the received skb */
953 if (myri10ge_getbuf(rx
, mgp
->pdev
, bytes
, idx
)) {
954 /* drop the frame -- the old skbuf is re-cycled */
955 mgp
->stats
.rx_dropped
+= 1;
959 /* unmap the recvd skb */
960 pci_unmap_single(mgp
->pdev
, bus
, unmap_len
, PCI_DMA_FROMDEVICE
);
962 /* mcp implicitly skips 1st bytes so that packet is properly
964 skb_reserve(skb
, MXGEFW_PAD
);
966 /* set the length of the frame */
969 skb
->protocol
= eth_type_trans(skb
, mgp
->dev
);
971 if (mgp
->csum_flag
) {
972 if ((skb
->protocol
== ntohs(ETH_P_IP
)) ||
973 (skb
->protocol
== ntohs(ETH_P_IPV6
))) {
974 skb
->csum
= ntohs((u16
) csum
);
975 skb
->ip_summed
= CHECKSUM_HW
;
977 myri10ge_vlan_ip_csum(skb
, ntohs((u16
) csum
));
980 netif_receive_skb(skb
);
981 mgp
->dev
->last_rx
= jiffies
;
985 static inline void myri10ge_tx_done(struct myri10ge_priv
*mgp
, int mcp_index
)
987 struct pci_dev
*pdev
= mgp
->pdev
;
988 struct myri10ge_tx_buf
*tx
= &mgp
->tx
;
993 while (tx
->pkt_done
!= mcp_index
) {
994 idx
= tx
->done
& tx
->mask
;
995 skb
= tx
->info
[idx
].skb
;
998 tx
->info
[idx
].skb
= NULL
;
999 if (tx
->info
[idx
].last
) {
1001 tx
->info
[idx
].last
= 0;
1004 len
= pci_unmap_len(&tx
->info
[idx
], len
);
1005 pci_unmap_len_set(&tx
->info
[idx
], len
, 0);
1007 mgp
->stats
.tx_bytes
+= skb
->len
;
1008 mgp
->stats
.tx_packets
++;
1009 dev_kfree_skb_irq(skb
);
1011 pci_unmap_single(pdev
,
1012 pci_unmap_addr(&tx
->info
[idx
],
1017 pci_unmap_page(pdev
,
1018 pci_unmap_addr(&tx
->info
[idx
],
1023 /* limit potential for livelock by only handling
1024 * 2 full tx rings per call */
1025 if (unlikely(++limit
> 2 * tx
->mask
))
1028 /* start the queue if we've stopped it */
1029 if (netif_queue_stopped(mgp
->dev
)
1030 && tx
->req
- tx
->done
< (tx
->mask
>> 1)) {
1032 netif_wake_queue(mgp
->dev
);
1036 static inline void myri10ge_clean_rx_done(struct myri10ge_priv
*mgp
, int *limit
)
1038 struct myri10ge_rx_done
*rx_done
= &mgp
->rx_done
;
1039 unsigned long rx_bytes
= 0;
1040 unsigned long rx_packets
= 0;
1041 unsigned long rx_ok
;
1043 int idx
= rx_done
->idx
;
1044 int cnt
= rx_done
->cnt
;
1048 while (rx_done
->entry
[idx
].length
!= 0 && *limit
!= 0) {
1049 length
= ntohs(rx_done
->entry
[idx
].length
);
1050 rx_done
->entry
[idx
].length
= 0;
1051 checksum
= ntohs(rx_done
->entry
[idx
].checksum
);
1052 if (length
<= mgp
->small_bytes
)
1053 rx_ok
= myri10ge_rx_done(mgp
, &mgp
->rx_small
,
1057 rx_ok
= myri10ge_rx_done(mgp
, &mgp
->rx_big
,
1058 mgp
->dev
->mtu
+ ETH_HLEN
,
1060 rx_packets
+= rx_ok
;
1061 rx_bytes
+= rx_ok
* (unsigned long)length
;
1063 idx
= cnt
& (myri10ge_max_intr_slots
- 1);
1065 /* limit potential for livelock by only handling a
1066 * limited number of frames. */
1071 mgp
->stats
.rx_packets
+= rx_packets
;
1072 mgp
->stats
.rx_bytes
+= rx_bytes
;
1075 static inline void myri10ge_check_statblock(struct myri10ge_priv
*mgp
)
1077 struct mcp_irq_data
*stats
= mgp
->fw_stats
;
1079 if (unlikely(stats
->stats_updated
)) {
1080 if (mgp
->link_state
!= stats
->link_up
) {
1081 mgp
->link_state
= stats
->link_up
;
1082 if (mgp
->link_state
) {
1083 printk(KERN_INFO
"myri10ge: %s: link up\n",
1085 netif_carrier_on(mgp
->dev
);
1087 printk(KERN_INFO
"myri10ge: %s: link down\n",
1089 netif_carrier_off(mgp
->dev
);
1092 if (mgp
->rdma_tags_available
!=
1093 ntohl(mgp
->fw_stats
->rdma_tags_available
)) {
1094 mgp
->rdma_tags_available
=
1095 ntohl(mgp
->fw_stats
->rdma_tags_available
);
1096 printk(KERN_WARNING
"myri10ge: %s: RDMA timed out! "
1097 "%d tags left\n", mgp
->dev
->name
,
1098 mgp
->rdma_tags_available
);
1100 mgp
->down_cnt
+= stats
->link_down
;
1101 if (stats
->link_down
)
1102 wake_up(&mgp
->down_wq
);
1106 static int myri10ge_poll(struct net_device
*netdev
, int *budget
)
1108 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1109 struct myri10ge_rx_done
*rx_done
= &mgp
->rx_done
;
1110 int limit
, orig_limit
, work_done
;
1112 /* process as many rx events as NAPI will allow */
1113 limit
= min(*budget
, netdev
->quota
);
1115 myri10ge_clean_rx_done(mgp
, &limit
);
1116 work_done
= orig_limit
- limit
;
1117 *budget
-= work_done
;
1118 netdev
->quota
-= work_done
;
1120 if (rx_done
->entry
[rx_done
->idx
].length
== 0 || !netif_running(netdev
)) {
1121 netif_rx_complete(netdev
);
1122 __raw_writel(htonl(3), mgp
->irq_claim
);
1128 static irqreturn_t
myri10ge_intr(int irq
, void *arg
, struct pt_regs
*regs
)
1130 struct myri10ge_priv
*mgp
= arg
;
1131 struct mcp_irq_data
*stats
= mgp
->fw_stats
;
1132 struct myri10ge_tx_buf
*tx
= &mgp
->tx
;
1133 u32 send_done_count
;
1136 /* make sure it is our IRQ, and that the DMA has finished */
1137 if (unlikely(!stats
->valid
))
1140 /* low bit indicates receives are present, so schedule
1141 * napi poll handler */
1142 if (stats
->valid
& 1)
1143 netif_rx_schedule(mgp
->dev
);
1145 if (!mgp
->msi_enabled
) {
1146 __raw_writel(0, mgp
->irq_deassert
);
1147 if (!myri10ge_deassert_wait
)
1153 /* Wait for IRQ line to go low, if using INTx */
1157 /* check for transmit completes and receives */
1158 send_done_count
= ntohl(stats
->send_done_count
);
1159 if (send_done_count
!= tx
->pkt_done
)
1160 myri10ge_tx_done(mgp
, (int)send_done_count
);
1161 if (unlikely(i
> myri10ge_max_irq_loops
)) {
1162 printk(KERN_WARNING
"myri10ge: %s: irq stuck?\n",
1165 schedule_work(&mgp
->watchdog_work
);
1167 if (likely(stats
->valid
== 0))
1173 myri10ge_check_statblock(mgp
);
1175 __raw_writel(htonl(3), mgp
->irq_claim
+ 1);
1176 return (IRQ_HANDLED
);
1180 myri10ge_get_settings(struct net_device
*netdev
, struct ethtool_cmd
*cmd
)
1182 cmd
->autoneg
= AUTONEG_DISABLE
;
1183 cmd
->speed
= SPEED_10000
;
1184 cmd
->duplex
= DUPLEX_FULL
;
1189 myri10ge_get_drvinfo(struct net_device
*netdev
, struct ethtool_drvinfo
*info
)
1191 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1193 strlcpy(info
->driver
, "myri10ge", sizeof(info
->driver
));
1194 strlcpy(info
->version
, MYRI10GE_VERSION_STR
, sizeof(info
->version
));
1195 strlcpy(info
->fw_version
, mgp
->fw_version
, sizeof(info
->fw_version
));
1196 strlcpy(info
->bus_info
, pci_name(mgp
->pdev
), sizeof(info
->bus_info
));
1200 myri10ge_get_coalesce(struct net_device
*netdev
, struct ethtool_coalesce
*coal
)
1202 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1203 coal
->rx_coalesce_usecs
= mgp
->intr_coal_delay
;
1208 myri10ge_set_coalesce(struct net_device
*netdev
, struct ethtool_coalesce
*coal
)
1210 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1212 mgp
->intr_coal_delay
= coal
->rx_coalesce_usecs
;
1213 __raw_writel(htonl(mgp
->intr_coal_delay
), mgp
->intr_coal_delay_ptr
);
1218 myri10ge_get_pauseparam(struct net_device
*netdev
,
1219 struct ethtool_pauseparam
*pause
)
1221 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1224 pause
->rx_pause
= mgp
->pause
;
1225 pause
->tx_pause
= mgp
->pause
;
1229 myri10ge_set_pauseparam(struct net_device
*netdev
,
1230 struct ethtool_pauseparam
*pause
)
1232 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1234 if (pause
->tx_pause
!= mgp
->pause
)
1235 return myri10ge_change_pause(mgp
, pause
->tx_pause
);
1236 if (pause
->rx_pause
!= mgp
->pause
)
1237 return myri10ge_change_pause(mgp
, pause
->tx_pause
);
1238 if (pause
->autoneg
!= 0)
1244 myri10ge_get_ringparam(struct net_device
*netdev
,
1245 struct ethtool_ringparam
*ring
)
1247 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1249 ring
->rx_mini_max_pending
= mgp
->rx_small
.mask
+ 1;
1250 ring
->rx_max_pending
= mgp
->rx_big
.mask
+ 1;
1251 ring
->rx_jumbo_max_pending
= 0;
1252 ring
->tx_max_pending
= mgp
->rx_small
.mask
+ 1;
1253 ring
->rx_mini_pending
= ring
->rx_mini_max_pending
;
1254 ring
->rx_pending
= ring
->rx_max_pending
;
1255 ring
->rx_jumbo_pending
= ring
->rx_jumbo_max_pending
;
1256 ring
->tx_pending
= ring
->tx_max_pending
;
1259 static u32
myri10ge_get_rx_csum(struct net_device
*netdev
)
1261 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1268 static int myri10ge_set_rx_csum(struct net_device
*netdev
, u32 csum_enabled
)
1270 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1272 mgp
->csum_flag
= MXGEFW_FLAGS_CKSUM
;
1278 static const char myri10ge_gstrings_stats
[][ETH_GSTRING_LEN
] = {
1279 "rx_packets", "tx_packets", "rx_bytes", "tx_bytes", "rx_errors",
1280 "tx_errors", "rx_dropped", "tx_dropped", "multicast", "collisions",
1281 "rx_length_errors", "rx_over_errors", "rx_crc_errors",
1282 "rx_frame_errors", "rx_fifo_errors", "rx_missed_errors",
1283 "tx_aborted_errors", "tx_carrier_errors", "tx_fifo_errors",
1284 "tx_heartbeat_errors", "tx_window_errors",
1285 /* device-specific stats */
1286 "tx_boundary", "WC", "irq", "MSI",
1287 "read_dma_bw_MBs", "write_dma_bw_MBs", "read_write_dma_bw_MBs",
1288 "serial_number", "tx_pkt_start", "tx_pkt_done",
1289 "tx_req", "tx_done", "rx_small_cnt", "rx_big_cnt",
1290 "wake_queue", "stop_queue", "watchdog_resets", "tx_linearized",
1291 "link_up", "dropped_link_overflow", "dropped_link_error_or_filtered",
1292 "dropped_runt", "dropped_overrun", "dropped_no_small_buffer",
1293 "dropped_no_big_buffer"
1296 #define MYRI10GE_NET_STATS_LEN 21
1297 #define MYRI10GE_STATS_LEN sizeof(myri10ge_gstrings_stats) / ETH_GSTRING_LEN
1300 myri10ge_get_strings(struct net_device
*netdev
, u32 stringset
, u8
* data
)
1302 switch (stringset
) {
1304 memcpy(data
, *myri10ge_gstrings_stats
,
1305 sizeof(myri10ge_gstrings_stats
));
1310 static int myri10ge_get_stats_count(struct net_device
*netdev
)
1312 return MYRI10GE_STATS_LEN
;
1316 myri10ge_get_ethtool_stats(struct net_device
*netdev
,
1317 struct ethtool_stats
*stats
, u64
* data
)
1319 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1322 for (i
= 0; i
< MYRI10GE_NET_STATS_LEN
; i
++)
1323 data
[i
] = ((unsigned long *)&mgp
->stats
)[i
];
1325 data
[i
++] = (unsigned int)mgp
->tx
.boundary
;
1326 data
[i
++] = (unsigned int)(mgp
->mtrr
>= 0);
1327 data
[i
++] = (unsigned int)mgp
->pdev
->irq
;
1328 data
[i
++] = (unsigned int)mgp
->msi_enabled
;
1329 data
[i
++] = (unsigned int)mgp
->read_dma
;
1330 data
[i
++] = (unsigned int)mgp
->write_dma
;
1331 data
[i
++] = (unsigned int)mgp
->read_write_dma
;
1332 data
[i
++] = (unsigned int)mgp
->serial_number
;
1333 data
[i
++] = (unsigned int)mgp
->tx
.pkt_start
;
1334 data
[i
++] = (unsigned int)mgp
->tx
.pkt_done
;
1335 data
[i
++] = (unsigned int)mgp
->tx
.req
;
1336 data
[i
++] = (unsigned int)mgp
->tx
.done
;
1337 data
[i
++] = (unsigned int)mgp
->rx_small
.cnt
;
1338 data
[i
++] = (unsigned int)mgp
->rx_big
.cnt
;
1339 data
[i
++] = (unsigned int)mgp
->wake_queue
;
1340 data
[i
++] = (unsigned int)mgp
->stop_queue
;
1341 data
[i
++] = (unsigned int)mgp
->watchdog_resets
;
1342 data
[i
++] = (unsigned int)mgp
->tx_linearized
;
1343 data
[i
++] = (unsigned int)ntohl(mgp
->fw_stats
->link_up
);
1344 data
[i
++] = (unsigned int)ntohl(mgp
->fw_stats
->dropped_link_overflow
);
1346 (unsigned int)ntohl(mgp
->fw_stats
->dropped_link_error_or_filtered
);
1347 data
[i
++] = (unsigned int)ntohl(mgp
->fw_stats
->dropped_runt
);
1348 data
[i
++] = (unsigned int)ntohl(mgp
->fw_stats
->dropped_overrun
);
1349 data
[i
++] = (unsigned int)ntohl(mgp
->fw_stats
->dropped_no_small_buffer
);
1350 data
[i
++] = (unsigned int)ntohl(mgp
->fw_stats
->dropped_no_big_buffer
);
1353 static struct ethtool_ops myri10ge_ethtool_ops
= {
1354 .get_settings
= myri10ge_get_settings
,
1355 .get_drvinfo
= myri10ge_get_drvinfo
,
1356 .get_coalesce
= myri10ge_get_coalesce
,
1357 .set_coalesce
= myri10ge_set_coalesce
,
1358 .get_pauseparam
= myri10ge_get_pauseparam
,
1359 .set_pauseparam
= myri10ge_set_pauseparam
,
1360 .get_ringparam
= myri10ge_get_ringparam
,
1361 .get_rx_csum
= myri10ge_get_rx_csum
,
1362 .set_rx_csum
= myri10ge_set_rx_csum
,
1363 .get_tx_csum
= ethtool_op_get_tx_csum
,
1364 .set_tx_csum
= ethtool_op_set_tx_hw_csum
,
1365 .get_sg
= ethtool_op_get_sg
,
1366 .set_sg
= ethtool_op_set_sg
,
1368 .get_tso
= ethtool_op_get_tso
,
1369 .set_tso
= ethtool_op_set_tso
,
1371 .get_strings
= myri10ge_get_strings
,
1372 .get_stats_count
= myri10ge_get_stats_count
,
1373 .get_ethtool_stats
= myri10ge_get_ethtool_stats
1376 static int myri10ge_allocate_rings(struct net_device
*dev
)
1378 struct myri10ge_priv
*mgp
;
1379 struct myri10ge_cmd cmd
;
1380 int tx_ring_size
, rx_ring_size
;
1381 int tx_ring_entries
, rx_ring_entries
;
1385 mgp
= netdev_priv(dev
);
1387 /* get ring sizes */
1389 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_SEND_RING_SIZE
, &cmd
, 0);
1390 tx_ring_size
= cmd
.data0
;
1391 status
|= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_RX_RING_SIZE
, &cmd
, 0);
1392 rx_ring_size
= cmd
.data0
;
1394 tx_ring_entries
= tx_ring_size
/ sizeof(struct mcp_kreq_ether_send
);
1395 rx_ring_entries
= rx_ring_size
/ sizeof(struct mcp_dma_addr
);
1396 mgp
->tx
.mask
= tx_ring_entries
- 1;
1397 mgp
->rx_small
.mask
= mgp
->rx_big
.mask
= rx_ring_entries
- 1;
1399 /* allocate the host shadow rings */
1401 bytes
= 8 + (MYRI10GE_MAX_SEND_DESC_TSO
+ 4)
1402 * sizeof(*mgp
->tx
.req_list
);
1403 mgp
->tx
.req_bytes
= kzalloc(bytes
, GFP_KERNEL
);
1404 if (mgp
->tx
.req_bytes
== NULL
)
1405 goto abort_with_nothing
;
1407 /* ensure req_list entries are aligned to 8 bytes */
1408 mgp
->tx
.req_list
= (struct mcp_kreq_ether_send
*)
1409 ALIGN((unsigned long)mgp
->tx
.req_bytes
, 8);
1411 bytes
= rx_ring_entries
* sizeof(*mgp
->rx_small
.shadow
);
1412 mgp
->rx_small
.shadow
= kzalloc(bytes
, GFP_KERNEL
);
1413 if (mgp
->rx_small
.shadow
== NULL
)
1414 goto abort_with_tx_req_bytes
;
1416 bytes
= rx_ring_entries
* sizeof(*mgp
->rx_big
.shadow
);
1417 mgp
->rx_big
.shadow
= kzalloc(bytes
, GFP_KERNEL
);
1418 if (mgp
->rx_big
.shadow
== NULL
)
1419 goto abort_with_rx_small_shadow
;
1421 /* allocate the host info rings */
1423 bytes
= tx_ring_entries
* sizeof(*mgp
->tx
.info
);
1424 mgp
->tx
.info
= kzalloc(bytes
, GFP_KERNEL
);
1425 if (mgp
->tx
.info
== NULL
)
1426 goto abort_with_rx_big_shadow
;
1428 bytes
= rx_ring_entries
* sizeof(*mgp
->rx_small
.info
);
1429 mgp
->rx_small
.info
= kzalloc(bytes
, GFP_KERNEL
);
1430 if (mgp
->rx_small
.info
== NULL
)
1431 goto abort_with_tx_info
;
1433 bytes
= rx_ring_entries
* sizeof(*mgp
->rx_big
.info
);
1434 mgp
->rx_big
.info
= kzalloc(bytes
, GFP_KERNEL
);
1435 if (mgp
->rx_big
.info
== NULL
)
1436 goto abort_with_rx_small_info
;
1438 /* Fill the receive rings */
1440 for (i
= 0; i
<= mgp
->rx_small
.mask
; i
++) {
1441 status
= myri10ge_getbuf(&mgp
->rx_small
, mgp
->pdev
,
1442 mgp
->small_bytes
, i
);
1445 "myri10ge: %s: alloced only %d small bufs\n",
1447 goto abort_with_rx_small_ring
;
1451 for (i
= 0; i
<= mgp
->rx_big
.mask
; i
++) {
1453 myri10ge_getbuf(&mgp
->rx_big
, mgp
->pdev
,
1454 dev
->mtu
+ ETH_HLEN
, i
);
1457 "myri10ge: %s: alloced only %d big bufs\n",
1459 goto abort_with_rx_big_ring
;
1465 abort_with_rx_big_ring
:
1466 for (i
= 0; i
<= mgp
->rx_big
.mask
; i
++) {
1467 if (mgp
->rx_big
.info
[i
].skb
!= NULL
)
1468 dev_kfree_skb_any(mgp
->rx_big
.info
[i
].skb
);
1469 if (pci_unmap_len(&mgp
->rx_big
.info
[i
], len
))
1470 pci_unmap_single(mgp
->pdev
,
1471 pci_unmap_addr(&mgp
->rx_big
.info
[i
],
1473 pci_unmap_len(&mgp
->rx_big
.info
[i
],
1475 PCI_DMA_FROMDEVICE
);
1478 abort_with_rx_small_ring
:
1479 for (i
= 0; i
<= mgp
->rx_small
.mask
; i
++) {
1480 if (mgp
->rx_small
.info
[i
].skb
!= NULL
)
1481 dev_kfree_skb_any(mgp
->rx_small
.info
[i
].skb
);
1482 if (pci_unmap_len(&mgp
->rx_small
.info
[i
], len
))
1483 pci_unmap_single(mgp
->pdev
,
1484 pci_unmap_addr(&mgp
->rx_small
.info
[i
],
1486 pci_unmap_len(&mgp
->rx_small
.info
[i
],
1488 PCI_DMA_FROMDEVICE
);
1490 kfree(mgp
->rx_big
.info
);
1492 abort_with_rx_small_info
:
1493 kfree(mgp
->rx_small
.info
);
1496 kfree(mgp
->tx
.info
);
1498 abort_with_rx_big_shadow
:
1499 kfree(mgp
->rx_big
.shadow
);
1501 abort_with_rx_small_shadow
:
1502 kfree(mgp
->rx_small
.shadow
);
1504 abort_with_tx_req_bytes
:
1505 kfree(mgp
->tx
.req_bytes
);
1506 mgp
->tx
.req_bytes
= NULL
;
1507 mgp
->tx
.req_list
= NULL
;
1513 static void myri10ge_free_rings(struct net_device
*dev
)
1515 struct myri10ge_priv
*mgp
;
1516 struct sk_buff
*skb
;
1517 struct myri10ge_tx_buf
*tx
;
1520 mgp
= netdev_priv(dev
);
1522 for (i
= 0; i
<= mgp
->rx_big
.mask
; i
++) {
1523 if (mgp
->rx_big
.info
[i
].skb
!= NULL
)
1524 dev_kfree_skb_any(mgp
->rx_big
.info
[i
].skb
);
1525 if (pci_unmap_len(&mgp
->rx_big
.info
[i
], len
))
1526 pci_unmap_single(mgp
->pdev
,
1527 pci_unmap_addr(&mgp
->rx_big
.info
[i
],
1529 pci_unmap_len(&mgp
->rx_big
.info
[i
],
1531 PCI_DMA_FROMDEVICE
);
1534 for (i
= 0; i
<= mgp
->rx_small
.mask
; i
++) {
1535 if (mgp
->rx_small
.info
[i
].skb
!= NULL
)
1536 dev_kfree_skb_any(mgp
->rx_small
.info
[i
].skb
);
1537 if (pci_unmap_len(&mgp
->rx_small
.info
[i
], len
))
1538 pci_unmap_single(mgp
->pdev
,
1539 pci_unmap_addr(&mgp
->rx_small
.info
[i
],
1541 pci_unmap_len(&mgp
->rx_small
.info
[i
],
1543 PCI_DMA_FROMDEVICE
);
1547 while (tx
->done
!= tx
->req
) {
1548 idx
= tx
->done
& tx
->mask
;
1549 skb
= tx
->info
[idx
].skb
;
1552 tx
->info
[idx
].skb
= NULL
;
1554 len
= pci_unmap_len(&tx
->info
[idx
], len
);
1555 pci_unmap_len_set(&tx
->info
[idx
], len
, 0);
1557 mgp
->stats
.tx_dropped
++;
1558 dev_kfree_skb_any(skb
);
1560 pci_unmap_single(mgp
->pdev
,
1561 pci_unmap_addr(&tx
->info
[idx
],
1566 pci_unmap_page(mgp
->pdev
,
1567 pci_unmap_addr(&tx
->info
[idx
],
1572 kfree(mgp
->rx_big
.info
);
1574 kfree(mgp
->rx_small
.info
);
1576 kfree(mgp
->tx
.info
);
1578 kfree(mgp
->rx_big
.shadow
);
1580 kfree(mgp
->rx_small
.shadow
);
1582 kfree(mgp
->tx
.req_bytes
);
1583 mgp
->tx
.req_bytes
= NULL
;
1584 mgp
->tx
.req_list
= NULL
;
1587 static int myri10ge_open(struct net_device
*dev
)
1589 struct myri10ge_priv
*mgp
;
1590 struct myri10ge_cmd cmd
;
1591 int status
, big_pow2
;
1593 mgp
= netdev_priv(dev
);
1595 if (mgp
->running
!= MYRI10GE_ETH_STOPPED
)
1598 mgp
->running
= MYRI10GE_ETH_STARTING
;
1599 status
= myri10ge_reset(mgp
);
1601 printk(KERN_ERR
"myri10ge: %s: failed reset\n", dev
->name
);
1602 mgp
->running
= MYRI10GE_ETH_STOPPED
;
1606 /* decide what small buffer size to use. For good TCP rx
1607 * performance, it is important to not receive 1514 byte
1608 * frames into jumbo buffers, as it confuses the socket buffer
1609 * accounting code, leading to drops and erratic performance.
1612 if (dev
->mtu
<= ETH_DATA_LEN
)
1613 mgp
->small_bytes
= 128; /* enough for a TCP header */
1615 mgp
->small_bytes
= ETH_FRAME_LEN
; /* enough for an ETH_DATA_LEN frame */
1617 /* Override the small buffer size? */
1618 if (myri10ge_small_bytes
> 0)
1619 mgp
->small_bytes
= myri10ge_small_bytes
;
1621 /* If the user sets an obscenely small MTU, adjust the small
1622 * bytes down to nearly nothing */
1623 if (mgp
->small_bytes
>= (dev
->mtu
+ ETH_HLEN
))
1624 mgp
->small_bytes
= 64;
1626 /* get the lanai pointers to the send and receive rings */
1628 status
|= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_SEND_OFFSET
, &cmd
, 0);
1630 (struct mcp_kreq_ether_send __iomem
*)(mgp
->sram
+ cmd
.data0
);
1633 myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_SMALL_RX_OFFSET
, &cmd
, 0);
1634 mgp
->rx_small
.lanai
=
1635 (struct mcp_kreq_ether_recv __iomem
*)(mgp
->sram
+ cmd
.data0
);
1637 status
|= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_BIG_RX_OFFSET
, &cmd
, 0);
1639 (struct mcp_kreq_ether_recv __iomem
*)(mgp
->sram
+ cmd
.data0
);
1643 "myri10ge: %s: failed to get ring sizes or locations\n",
1645 mgp
->running
= MYRI10GE_ETH_STOPPED
;
1649 if (mgp
->mtrr
>= 0) {
1650 mgp
->tx
.wc_fifo
= (u8 __iomem
*) mgp
->sram
+ 0x200000;
1651 mgp
->rx_small
.wc_fifo
= (u8 __iomem
*) mgp
->sram
+ 0x300000;
1652 mgp
->rx_big
.wc_fifo
= (u8 __iomem
*) mgp
->sram
+ 0x340000;
1654 mgp
->tx
.wc_fifo
= NULL
;
1655 mgp
->rx_small
.wc_fifo
= NULL
;
1656 mgp
->rx_big
.wc_fifo
= NULL
;
1659 status
= myri10ge_allocate_rings(dev
);
1661 goto abort_with_nothing
;
1663 /* Firmware needs the big buff size as a power of 2. Lie and
1664 * tell him the buffer is larger, because we only use 1
1665 * buffer/pkt, and the mtu will prevent overruns.
1667 big_pow2
= dev
->mtu
+ ETH_HLEN
+ MXGEFW_PAD
;
1668 while ((big_pow2
& (big_pow2
- 1)) != 0)
1671 /* now give firmware buffers sizes, and MTU */
1672 cmd
.data0
= dev
->mtu
+ ETH_HLEN
+ VLAN_HLEN
;
1673 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_MTU
, &cmd
, 0);
1674 cmd
.data0
= mgp
->small_bytes
;
1676 myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_SMALL_BUFFER_SIZE
, &cmd
, 0);
1677 cmd
.data0
= big_pow2
;
1679 myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_BIG_BUFFER_SIZE
, &cmd
, 0);
1681 printk(KERN_ERR
"myri10ge: %s: Couldn't set buffer sizes\n",
1683 goto abort_with_rings
;
1686 cmd
.data0
= MYRI10GE_LOWPART_TO_U32(mgp
->fw_stats_bus
);
1687 cmd
.data1
= MYRI10GE_HIGHPART_TO_U32(mgp
->fw_stats_bus
);
1688 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_STATS_DMA
, &cmd
, 0);
1690 printk(KERN_ERR
"myri10ge: %s: Couldn't set stats DMA\n",
1692 goto abort_with_rings
;
1695 mgp
->link_state
= -1;
1696 mgp
->rdma_tags_available
= 15;
1698 netif_poll_enable(mgp
->dev
); /* must happen prior to any irq */
1700 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_ETHERNET_UP
, &cmd
, 0);
1702 printk(KERN_ERR
"myri10ge: %s: Couldn't bring up link\n",
1704 goto abort_with_rings
;
1707 mgp
->wake_queue
= 0;
1708 mgp
->stop_queue
= 0;
1709 mgp
->running
= MYRI10GE_ETH_RUNNING
;
1710 mgp
->watchdog_timer
.expires
= jiffies
+ myri10ge_watchdog_timeout
* HZ
;
1711 add_timer(&mgp
->watchdog_timer
);
1712 netif_wake_queue(dev
);
1716 myri10ge_free_rings(dev
);
1719 mgp
->running
= MYRI10GE_ETH_STOPPED
;
1723 static int myri10ge_close(struct net_device
*dev
)
1725 struct myri10ge_priv
*mgp
;
1726 struct myri10ge_cmd cmd
;
1727 int status
, old_down_cnt
;
1729 mgp
= netdev_priv(dev
);
1731 if (mgp
->running
!= MYRI10GE_ETH_RUNNING
)
1734 if (mgp
->tx
.req_bytes
== NULL
)
1737 del_timer_sync(&mgp
->watchdog_timer
);
1738 mgp
->running
= MYRI10GE_ETH_STOPPING
;
1739 netif_poll_disable(mgp
->dev
);
1740 netif_carrier_off(dev
);
1741 netif_stop_queue(dev
);
1742 old_down_cnt
= mgp
->down_cnt
;
1744 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_ETHERNET_DOWN
, &cmd
, 0);
1746 printk(KERN_ERR
"myri10ge: %s: Couldn't bring down link\n",
1749 wait_event_timeout(mgp
->down_wq
, old_down_cnt
!= mgp
->down_cnt
, HZ
);
1750 if (old_down_cnt
== mgp
->down_cnt
)
1751 printk(KERN_ERR
"myri10ge: %s never got down irq\n", dev
->name
);
1753 netif_tx_disable(dev
);
1755 myri10ge_free_rings(dev
);
1757 mgp
->running
= MYRI10GE_ETH_STOPPED
;
1761 /* copy an array of struct mcp_kreq_ether_send's to the mcp. Copy
1762 * backwards one at a time and handle ring wraps */
1765 myri10ge_submit_req_backwards(struct myri10ge_tx_buf
*tx
,
1766 struct mcp_kreq_ether_send
*src
, int cnt
)
1768 int idx
, starting_slot
;
1769 starting_slot
= tx
->req
;
1772 idx
= (starting_slot
+ cnt
) & tx
->mask
;
1773 myri10ge_pio_copy(&tx
->lanai
[idx
], &src
[cnt
], sizeof(*src
));
1779 * copy an array of struct mcp_kreq_ether_send's to the mcp. Copy
1780 * at most 32 bytes at a time, so as to avoid involving the software
1781 * pio handler in the nic. We re-write the first segment's flags
1782 * to mark them valid only after writing the entire chain.
1786 myri10ge_submit_req(struct myri10ge_tx_buf
*tx
, struct mcp_kreq_ether_send
*src
,
1790 struct mcp_kreq_ether_send __iomem
*dstp
, *dst
;
1791 struct mcp_kreq_ether_send
*srcp
;
1794 idx
= tx
->req
& tx
->mask
;
1796 last_flags
= src
->flags
;
1799 dst
= dstp
= &tx
->lanai
[idx
];
1802 if ((idx
+ cnt
) < tx
->mask
) {
1803 for (i
= 0; i
< (cnt
- 1); i
+= 2) {
1804 myri10ge_pio_copy(dstp
, srcp
, 2 * sizeof(*src
));
1805 mb(); /* force write every 32 bytes */
1810 /* submit all but the first request, and ensure
1811 * that it is submitted below */
1812 myri10ge_submit_req_backwards(tx
, src
, cnt
);
1816 /* submit the first request */
1817 myri10ge_pio_copy(dstp
, srcp
, sizeof(*src
));
1818 mb(); /* barrier before setting valid flag */
1821 /* re-write the last 32-bits with the valid flags */
1822 src
->flags
= last_flags
;
1823 __raw_writel(*((u32
*) src
+ 3), (u32 __iomem
*) dst
+ 3);
1829 myri10ge_submit_req_wc(struct myri10ge_tx_buf
*tx
,
1830 struct mcp_kreq_ether_send
*src
, int cnt
)
1835 myri10ge_pio_copy(tx
->wc_fifo
, src
, 64);
1841 /* pad it to 64 bytes. The src is 64 bytes bigger than it
1842 * needs to be so that we don't overrun it */
1843 myri10ge_pio_copy(tx
->wc_fifo
+ (cnt
<< 18), src
, 64);
1849 * Transmit a packet. We need to split the packet so that a single
1850 * segment does not cross myri10ge->tx.boundary, so this makes segment
1851 * counting tricky. So rather than try to count segments up front, we
1852 * just give up if there are too few segments to hold a reasonably
1853 * fragmented packet currently available. If we run
1854 * out of segments while preparing a packet for DMA, we just linearize
1858 static int myri10ge_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1860 struct myri10ge_priv
*mgp
= netdev_priv(dev
);
1861 struct mcp_kreq_ether_send
*req
;
1862 struct myri10ge_tx_buf
*tx
= &mgp
->tx
;
1863 struct skb_frag_struct
*frag
;
1865 u32 low
, high_swapped
;
1867 int idx
, last_idx
, avail
, frag_cnt
, frag_idx
, count
, mss
, max_segments
;
1868 u16 pseudo_hdr_offset
, cksum_offset
;
1869 int cum_len
, seglen
, boundary
, rdma_count
;
1874 avail
= tx
->mask
- 1 - (tx
->req
- tx
->done
);
1877 max_segments
= MXGEFW_MAX_SEND_DESC
;
1880 if (skb
->len
> (dev
->mtu
+ ETH_HLEN
)) {
1881 mss
= skb_shinfo(skb
)->gso_size
;
1883 max_segments
= MYRI10GE_MAX_SEND_DESC_TSO
;
1885 #endif /*NETIF_F_TSO */
1887 if ((unlikely(avail
< max_segments
))) {
1888 /* we are out of transmit resources */
1890 netif_stop_queue(dev
);
1894 /* Setup checksum offloading, if needed */
1896 pseudo_hdr_offset
= 0;
1898 flags
= (MXGEFW_FLAGS_NO_TSO
| MXGEFW_FLAGS_FIRST
);
1899 if (likely(skb
->ip_summed
== CHECKSUM_HW
)) {
1900 cksum_offset
= (skb
->h
.raw
- skb
->data
);
1901 pseudo_hdr_offset
= (skb
->h
.raw
+ skb
->csum
) - skb
->data
;
1902 /* If the headers are excessively large, then we must
1903 * fall back to a software checksum */
1904 if (unlikely(cksum_offset
> 255 || pseudo_hdr_offset
> 127)) {
1905 if (skb_checksum_help(skb
, 0))
1908 pseudo_hdr_offset
= 0;
1910 pseudo_hdr_offset
= htons(pseudo_hdr_offset
);
1911 odd_flag
= MXGEFW_FLAGS_ALIGN_ODD
;
1912 flags
|= MXGEFW_FLAGS_CKSUM
;
1919 if (mss
) { /* TSO */
1920 /* this removes any CKSUM flag from before */
1921 flags
= (MXGEFW_FLAGS_TSO_HDR
| MXGEFW_FLAGS_FIRST
);
1923 /* negative cum_len signifies to the
1924 * send loop that we are still in the
1925 * header portion of the TSO packet.
1926 * TSO header must be at most 134 bytes long */
1927 cum_len
= -((skb
->h
.raw
- skb
->data
) + (skb
->h
.th
->doff
<< 2));
1929 /* for TSO, pseudo_hdr_offset holds mss.
1930 * The firmware figures out where to put
1931 * the checksum by parsing the header. */
1932 pseudo_hdr_offset
= htons(mss
);
1934 #endif /*NETIF_F_TSO */
1935 /* Mark small packets, and pad out tiny packets */
1936 if (skb
->len
<= MXGEFW_SEND_SMALL_SIZE
) {
1937 flags
|= MXGEFW_FLAGS_SMALL
;
1939 /* pad frames to at least ETH_ZLEN bytes */
1940 if (unlikely(skb
->len
< ETH_ZLEN
)) {
1941 if (skb_padto(skb
, ETH_ZLEN
)) {
1942 /* The packet is gone, so we must
1944 mgp
->stats
.tx_dropped
+= 1;
1947 /* adjust the len to account for the zero pad
1948 * so that the nic can know how long it is */
1949 skb
->len
= ETH_ZLEN
;
1953 /* map the skb for DMA */
1954 len
= skb
->len
- skb
->data_len
;
1955 idx
= tx
->req
& tx
->mask
;
1956 tx
->info
[idx
].skb
= skb
;
1957 bus
= pci_map_single(mgp
->pdev
, skb
->data
, len
, PCI_DMA_TODEVICE
);
1958 pci_unmap_addr_set(&tx
->info
[idx
], bus
, bus
);
1959 pci_unmap_len_set(&tx
->info
[idx
], len
, len
);
1961 frag_cnt
= skb_shinfo(skb
)->nr_frags
;
1966 /* "rdma_count" is the number of RDMAs belonging to the
1967 * current packet BEFORE the current send request. For
1968 * non-TSO packets, this is equal to "count".
1969 * For TSO packets, rdma_count needs to be reset
1970 * to 0 after a segment cut.
1972 * The rdma_count field of the send request is
1973 * the number of RDMAs of the packet starting at
1974 * that request. For TSO send requests with one ore more cuts
1975 * in the middle, this is the number of RDMAs starting
1976 * after the last cut in the request. All previous
1977 * segments before the last cut implicitly have 1 RDMA.
1979 * Since the number of RDMAs is not known beforehand,
1980 * it must be filled-in retroactively - after each
1981 * segmentation cut or at the end of the entire packet.
1985 /* Break the SKB or Fragment up into pieces which
1986 * do not cross mgp->tx.boundary */
1987 low
= MYRI10GE_LOWPART_TO_U32(bus
);
1988 high_swapped
= htonl(MYRI10GE_HIGHPART_TO_U32(bus
));
1993 if (unlikely(count
== max_segments
))
1994 goto abort_linearize
;
1996 boundary
= (low
+ tx
->boundary
) & ~(tx
->boundary
- 1);
1997 seglen
= boundary
- low
;
2000 flags_next
= flags
& ~MXGEFW_FLAGS_FIRST
;
2001 cum_len_next
= cum_len
+ seglen
;
2003 if (mss
) { /* TSO */
2004 (req
- rdma_count
)->rdma_count
= rdma_count
+ 1;
2006 if (likely(cum_len
>= 0)) { /* payload */
2007 int next_is_first
, chop
;
2009 chop
= (cum_len_next
> mss
);
2010 cum_len_next
= cum_len_next
% mss
;
2011 next_is_first
= (cum_len_next
== 0);
2012 flags
|= chop
* MXGEFW_FLAGS_TSO_CHOP
;
2013 flags_next
|= next_is_first
*
2015 rdma_count
|= -(chop
| next_is_first
);
2016 rdma_count
+= chop
& !next_is_first
;
2017 } else if (likely(cum_len_next
>= 0)) { /* header ends */
2023 small
= (mss
<= MXGEFW_SEND_SMALL_SIZE
);
2024 flags_next
= MXGEFW_FLAGS_TSO_PLD
|
2025 MXGEFW_FLAGS_FIRST
|
2026 (small
* MXGEFW_FLAGS_SMALL
);
2029 #endif /* NETIF_F_TSO */
2030 req
->addr_high
= high_swapped
;
2031 req
->addr_low
= htonl(low
);
2032 req
->pseudo_hdr_offset
= pseudo_hdr_offset
;
2033 req
->pad
= 0; /* complete solid 16-byte block; does this matter? */
2034 req
->rdma_count
= 1;
2035 req
->length
= htons(seglen
);
2036 req
->cksum_offset
= cksum_offset
;
2037 req
->flags
= flags
| ((cum_len
& 1) * odd_flag
);
2041 cum_len
= cum_len_next
;
2046 if (unlikely(cksum_offset
> seglen
))
2047 cksum_offset
-= seglen
;
2051 if (frag_idx
== frag_cnt
)
2054 /* map next fragment for DMA */
2055 idx
= (count
+ tx
->req
) & tx
->mask
;
2056 frag
= &skb_shinfo(skb
)->frags
[frag_idx
];
2059 bus
= pci_map_page(mgp
->pdev
, frag
->page
, frag
->page_offset
,
2060 len
, PCI_DMA_TODEVICE
);
2061 pci_unmap_addr_set(&tx
->info
[idx
], bus
, bus
);
2062 pci_unmap_len_set(&tx
->info
[idx
], len
, len
);
2065 (req
- rdma_count
)->rdma_count
= rdma_count
;
2070 req
->flags
|= MXGEFW_FLAGS_TSO_LAST
;
2071 } while (!(req
->flags
& (MXGEFW_FLAGS_TSO_CHOP
|
2072 MXGEFW_FLAGS_FIRST
)));
2074 idx
= ((count
- 1) + tx
->req
) & tx
->mask
;
2075 tx
->info
[idx
].last
= 1;
2076 if (tx
->wc_fifo
== NULL
)
2077 myri10ge_submit_req(tx
, tx
->req_list
, count
);
2079 myri10ge_submit_req_wc(tx
, tx
->req_list
, count
);
2081 if ((avail
- count
) < MXGEFW_MAX_SEND_DESC
) {
2083 netif_stop_queue(dev
);
2085 dev
->trans_start
= jiffies
;
2089 /* Free any DMA resources we've alloced and clear out the skb
2090 * slot so as to not trip up assertions, and to avoid a
2091 * double-free if linearizing fails */
2093 last_idx
= (idx
+ 1) & tx
->mask
;
2094 idx
= tx
->req
& tx
->mask
;
2095 tx
->info
[idx
].skb
= NULL
;
2097 len
= pci_unmap_len(&tx
->info
[idx
], len
);
2099 if (tx
->info
[idx
].skb
!= NULL
)
2100 pci_unmap_single(mgp
->pdev
,
2101 pci_unmap_addr(&tx
->info
[idx
],
2105 pci_unmap_page(mgp
->pdev
,
2106 pci_unmap_addr(&tx
->info
[idx
],
2109 pci_unmap_len_set(&tx
->info
[idx
], len
, 0);
2110 tx
->info
[idx
].skb
= NULL
;
2112 idx
= (idx
+ 1) & tx
->mask
;
2113 } while (idx
!= last_idx
);
2114 if (skb_is_gso(skb
)) {
2116 "myri10ge: %s: TSO but wanted to linearize?!?!?\n",
2121 if (skb_linearize(skb
))
2124 mgp
->tx_linearized
++;
2128 dev_kfree_skb_any(skb
);
2129 mgp
->stats
.tx_dropped
+= 1;
2134 static struct net_device_stats
*myri10ge_get_stats(struct net_device
*dev
)
2136 struct myri10ge_priv
*mgp
= netdev_priv(dev
);
2140 static void myri10ge_set_multicast_list(struct net_device
*dev
)
2142 /* can be called from atomic contexts,
2143 * pass 1 to force atomicity in myri10ge_send_cmd() */
2144 myri10ge_change_promisc(netdev_priv(dev
), dev
->flags
& IFF_PROMISC
, 1);
2147 static int myri10ge_set_mac_address(struct net_device
*dev
, void *addr
)
2149 struct sockaddr
*sa
= addr
;
2150 struct myri10ge_priv
*mgp
= netdev_priv(dev
);
2153 if (!is_valid_ether_addr(sa
->sa_data
))
2154 return -EADDRNOTAVAIL
;
2156 status
= myri10ge_update_mac_address(mgp
, sa
->sa_data
);
2159 "myri10ge: %s: changing mac address failed with %d\n",
2164 /* change the dev structure */
2165 memcpy(dev
->dev_addr
, sa
->sa_data
, 6);
2169 static int myri10ge_change_mtu(struct net_device
*dev
, int new_mtu
)
2171 struct myri10ge_priv
*mgp
= netdev_priv(dev
);
2174 if ((new_mtu
< 68) || (ETH_HLEN
+ new_mtu
> MYRI10GE_MAX_ETHER_MTU
)) {
2175 printk(KERN_ERR
"myri10ge: %s: new mtu (%d) is not valid\n",
2176 dev
->name
, new_mtu
);
2179 printk(KERN_INFO
"%s: changing mtu from %d to %d\n",
2180 dev
->name
, dev
->mtu
, new_mtu
);
2182 /* if we change the mtu on an active device, we must
2183 * reset the device so the firmware sees the change */
2184 myri10ge_close(dev
);
2194 * Enable ECRC to align PCI-E Completion packets on an 8-byte boundary.
2195 * Only do it if the bridge is a root port since we don't want to disturb
2196 * any other device, except if forced with myri10ge_ecrc_enable > 1.
2199 static void myri10ge_enable_ecrc(struct myri10ge_priv
*mgp
)
2201 struct pci_dev
*bridge
= mgp
->pdev
->bus
->self
;
2202 struct device
*dev
= &mgp
->pdev
->dev
;
2209 if (!myri10ge_ecrc_enable
|| !bridge
)
2212 /* check that the bridge is a root port */
2213 cap
= pci_find_capability(bridge
, PCI_CAP_ID_EXP
);
2214 pci_read_config_word(bridge
, cap
+ PCI_CAP_FLAGS
, &val
);
2215 ext_type
= (val
& PCI_EXP_FLAGS_TYPE
) >> 4;
2216 if (ext_type
!= PCI_EXP_TYPE_ROOT_PORT
) {
2217 if (myri10ge_ecrc_enable
> 1) {
2218 struct pci_dev
*old_bridge
= bridge
;
2220 /* Walk the hierarchy up to the root port
2221 * where ECRC has to be enabled */
2223 bridge
= bridge
->bus
->self
;
2226 "Failed to find root port"
2227 " to force ECRC\n");
2231 pci_find_capability(bridge
, PCI_CAP_ID_EXP
);
2232 pci_read_config_word(bridge
,
2233 cap
+ PCI_CAP_FLAGS
, &val
);
2234 ext_type
= (val
& PCI_EXP_FLAGS_TYPE
) >> 4;
2235 } while (ext_type
!= PCI_EXP_TYPE_ROOT_PORT
);
2238 "Forcing ECRC on non-root port %s"
2239 " (enabling on root port %s)\n",
2240 pci_name(old_bridge
), pci_name(bridge
));
2243 "Not enabling ECRC on non-root port %s\n",
2249 cap
= pci_find_ext_capability(bridge
, PCI_EXT_CAP_ID_ERR
);
2253 ret
= pci_read_config_dword(bridge
, cap
+ PCI_ERR_CAP
, &err_cap
);
2255 dev_err(dev
, "failed reading ext-conf-space of %s\n",
2257 dev_err(dev
, "\t pci=nommconf in use? "
2258 "or buggy/incomplete/absent ACPI MCFG attr?\n");
2261 if (!(err_cap
& PCI_ERR_CAP_ECRC_GENC
))
2264 err_cap
|= PCI_ERR_CAP_ECRC_GENE
;
2265 pci_write_config_dword(bridge
, cap
+ PCI_ERR_CAP
, err_cap
);
2266 dev_info(dev
, "Enabled ECRC on upstream bridge %s\n", pci_name(bridge
));
2267 mgp
->tx
.boundary
= 4096;
2268 mgp
->fw_name
= myri10ge_fw_aligned
;
2272 * The Lanai Z8E PCI-E interface achieves higher Read-DMA throughput
2273 * when the PCI-E Completion packets are aligned on an 8-byte
2274 * boundary. Some PCI-E chip sets always align Completion packets; on
2275 * the ones that do not, the alignment can be enforced by enabling
2276 * ECRC generation (if supported).
2278 * When PCI-E Completion packets are not aligned, it is actually more
2279 * efficient to limit Read-DMA transactions to 2KB, rather than 4KB.
2281 * If the driver can neither enable ECRC nor verify that it has
2282 * already been enabled, then it must use a firmware image which works
2283 * around unaligned completion packets (myri10ge_ethp_z8e.dat), and it
2284 * should also ensure that it never gives the device a Read-DMA which is
2285 * larger than 2KB by setting the tx.boundary to 2KB. If ECRC is
2286 * enabled, then the driver should use the aligned (myri10ge_eth_z8e.dat)
2287 * firmware image, and set tx.boundary to 4KB.
2290 #define PCI_DEVICE_ID_SERVERWORKS_HT2000_PCIE 0x0132
2292 static void myri10ge_select_firmware(struct myri10ge_priv
*mgp
)
2294 struct pci_dev
*bridge
= mgp
->pdev
->bus
->self
;
2296 mgp
->tx
.boundary
= 2048;
2297 mgp
->fw_name
= myri10ge_fw_unaligned
;
2299 if (myri10ge_force_firmware
== 0) {
2300 myri10ge_enable_ecrc(mgp
);
2302 /* Check to see if the upstream bridge is known to
2303 * provide aligned completions */
2305 /* ServerWorks HT2000/HT1000 */
2306 && bridge
->vendor
== PCI_VENDOR_ID_SERVERWORKS
2307 && bridge
->device
==
2308 PCI_DEVICE_ID_SERVERWORKS_HT2000_PCIE
) {
2309 dev_info(&mgp
->pdev
->dev
,
2310 "Assuming aligned completions (0x%x:0x%x)\n",
2311 bridge
->vendor
, bridge
->device
);
2312 mgp
->tx
.boundary
= 4096;
2313 mgp
->fw_name
= myri10ge_fw_aligned
;
2316 if (myri10ge_force_firmware
== 1) {
2317 dev_info(&mgp
->pdev
->dev
,
2318 "Assuming aligned completions (forced)\n");
2319 mgp
->tx
.boundary
= 4096;
2320 mgp
->fw_name
= myri10ge_fw_aligned
;
2322 dev_info(&mgp
->pdev
->dev
,
2323 "Assuming unaligned completions (forced)\n");
2324 mgp
->tx
.boundary
= 2048;
2325 mgp
->fw_name
= myri10ge_fw_unaligned
;
2328 if (myri10ge_fw_name
!= NULL
) {
2329 dev_info(&mgp
->pdev
->dev
, "overriding firmware to %s\n",
2331 mgp
->fw_name
= myri10ge_fw_name
;
2335 static void myri10ge_save_state(struct myri10ge_priv
*mgp
)
2337 struct pci_dev
*pdev
= mgp
->pdev
;
2340 pci_save_state(pdev
);
2341 /* now save PCIe and MSI state that Linux will not
2343 cap
= pci_find_capability(pdev
, PCI_CAP_ID_EXP
);
2344 pci_read_config_dword(pdev
, cap
+ PCI_EXP_DEVCTL
, &mgp
->devctl
);
2345 cap
= pci_find_capability(pdev
, PCI_CAP_ID_MSI
);
2346 pci_read_config_word(pdev
, cap
+ PCI_MSI_FLAGS
, &mgp
->msi_flags
);
2349 static void myri10ge_restore_state(struct myri10ge_priv
*mgp
)
2351 struct pci_dev
*pdev
= mgp
->pdev
;
2354 /* restore PCIe and MSI state that linux will not */
2355 cap
= pci_find_capability(pdev
, PCI_CAP_ID_EXP
);
2356 pci_write_config_dword(pdev
, cap
+ PCI_CAP_ID_EXP
, mgp
->devctl
);
2357 cap
= pci_find_capability(pdev
, PCI_CAP_ID_MSI
);
2358 pci_write_config_word(pdev
, cap
+ PCI_MSI_FLAGS
, mgp
->msi_flags
);
2360 pci_restore_state(pdev
);
2365 static int myri10ge_suspend(struct pci_dev
*pdev
, pm_message_t state
)
2367 struct myri10ge_priv
*mgp
;
2368 struct net_device
*netdev
;
2370 mgp
= pci_get_drvdata(pdev
);
2375 netif_device_detach(netdev
);
2376 if (netif_running(netdev
)) {
2377 printk(KERN_INFO
"myri10ge: closing %s\n", netdev
->name
);
2379 myri10ge_close(netdev
);
2382 myri10ge_dummy_rdma(mgp
, 0);
2383 free_irq(pdev
->irq
, mgp
);
2384 myri10ge_save_state(mgp
);
2385 pci_disable_device(pdev
);
2386 pci_set_power_state(pdev
, pci_choose_state(pdev
, state
));
2390 static int myri10ge_resume(struct pci_dev
*pdev
)
2392 struct myri10ge_priv
*mgp
;
2393 struct net_device
*netdev
;
2397 mgp
= pci_get_drvdata(pdev
);
2401 pci_set_power_state(pdev
, 0); /* zeros conf space as a side effect */
2402 msleep(5); /* give card time to respond */
2403 pci_read_config_word(mgp
->pdev
, PCI_VENDOR_ID
, &vendor
);
2404 if (vendor
== 0xffff) {
2405 printk(KERN_ERR
"myri10ge: %s: device disappeared!\n",
2409 myri10ge_restore_state(mgp
);
2411 status
= pci_enable_device(pdev
);
2413 dev_err(&pdev
->dev
, "failed to enable device\n");
2417 pci_set_master(pdev
);
2419 status
= request_irq(pdev
->irq
, myri10ge_intr
, IRQF_SHARED
,
2422 dev_err(&pdev
->dev
, "failed to allocate IRQ\n");
2423 goto abort_with_enabled
;
2426 myri10ge_reset(mgp
);
2427 myri10ge_dummy_rdma(mgp
, mgp
->tx
.boundary
!= 4096);
2429 /* Save configuration space to be restored if the
2430 * nic resets due to a parity error */
2431 myri10ge_save_state(mgp
);
2433 if (netif_running(netdev
)) {
2435 myri10ge_open(netdev
);
2438 netif_device_attach(netdev
);
2443 pci_disable_device(pdev
);
2448 #endif /* CONFIG_PM */
2450 static u32
myri10ge_read_reboot(struct myri10ge_priv
*mgp
)
2452 struct pci_dev
*pdev
= mgp
->pdev
;
2453 int vs
= mgp
->vendor_specific_offset
;
2456 /*enter read32 mode */
2457 pci_write_config_byte(pdev
, vs
+ 0x10, 0x3);
2459 /*read REBOOT_STATUS (0xfffffff0) */
2460 pci_write_config_dword(pdev
, vs
+ 0x18, 0xfffffff0);
2461 pci_read_config_dword(pdev
, vs
+ 0x14, &reboot
);
2466 * This watchdog is used to check whether the board has suffered
2467 * from a parity error and needs to be recovered.
2469 static void myri10ge_watchdog(void *arg
)
2471 struct myri10ge_priv
*mgp
= arg
;
2476 mgp
->watchdog_resets
++;
2477 pci_read_config_word(mgp
->pdev
, PCI_COMMAND
, &cmd
);
2478 if ((cmd
& PCI_COMMAND_MASTER
) == 0) {
2479 /* Bus master DMA disabled? Check to see
2480 * if the card rebooted due to a parity error
2481 * For now, just report it */
2482 reboot
= myri10ge_read_reboot(mgp
);
2484 "myri10ge: %s: NIC rebooted (0x%x), resetting\n",
2485 mgp
->dev
->name
, reboot
);
2487 * A rebooted nic will come back with config space as
2488 * it was after power was applied to PCIe bus.
2489 * Attempt to restore config space which was saved
2490 * when the driver was loaded, or the last time the
2491 * nic was resumed from power saving mode.
2493 myri10ge_restore_state(mgp
);
2495 /* if we get back -1's from our slot, perhaps somebody
2496 * powered off our card. Don't try to reset it in
2498 if (cmd
== 0xffff) {
2499 pci_read_config_word(mgp
->pdev
, PCI_VENDOR_ID
, &vendor
);
2500 if (vendor
== 0xffff) {
2502 "myri10ge: %s: device disappeared!\n",
2507 /* Perhaps it is a software error. Try to reset */
2509 printk(KERN_ERR
"myri10ge: %s: device timeout, resetting\n",
2511 printk(KERN_INFO
"myri10ge: %s: %d %d %d %d %d\n",
2512 mgp
->dev
->name
, mgp
->tx
.req
, mgp
->tx
.done
,
2513 mgp
->tx
.pkt_start
, mgp
->tx
.pkt_done
,
2514 (int)ntohl(mgp
->fw_stats
->send_done_count
));
2516 printk(KERN_INFO
"myri10ge: %s: %d %d %d %d %d\n",
2517 mgp
->dev
->name
, mgp
->tx
.req
, mgp
->tx
.done
,
2518 mgp
->tx
.pkt_start
, mgp
->tx
.pkt_done
,
2519 (int)ntohl(mgp
->fw_stats
->send_done_count
));
2522 myri10ge_close(mgp
->dev
);
2523 status
= myri10ge_load_firmware(mgp
);
2525 printk(KERN_ERR
"myri10ge: %s: failed to load firmware\n",
2528 myri10ge_open(mgp
->dev
);
2533 * We use our own timer routine rather than relying upon
2534 * netdev->tx_timeout because we have a very large hardware transmit
2535 * queue. Due to the large queue, the netdev->tx_timeout function
2536 * cannot detect a NIC with a parity error in a timely fashion if the
2537 * NIC is lightly loaded.
2539 static void myri10ge_watchdog_timer(unsigned long arg
)
2541 struct myri10ge_priv
*mgp
;
2543 mgp
= (struct myri10ge_priv
*)arg
;
2544 if (mgp
->tx
.req
!= mgp
->tx
.done
&&
2545 mgp
->tx
.done
== mgp
->watchdog_tx_done
)
2546 /* nic seems like it might be stuck.. */
2547 schedule_work(&mgp
->watchdog_work
);
2550 mod_timer(&mgp
->watchdog_timer
,
2551 jiffies
+ myri10ge_watchdog_timeout
* HZ
);
2553 mgp
->watchdog_tx_done
= mgp
->tx
.done
;
2556 static int myri10ge_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
2558 struct net_device
*netdev
;
2559 struct myri10ge_priv
*mgp
;
2560 struct device
*dev
= &pdev
->dev
;
2563 int status
= -ENXIO
;
2568 netdev
= alloc_etherdev(sizeof(*mgp
));
2569 if (netdev
== NULL
) {
2570 dev_err(dev
, "Could not allocate ethernet device\n");
2574 mgp
= netdev_priv(netdev
);
2575 memset(mgp
, 0, sizeof(*mgp
));
2578 mgp
->csum_flag
= MXGEFW_FLAGS_CKSUM
;
2579 mgp
->pause
= myri10ge_flow_control
;
2580 mgp
->intr_coal_delay
= myri10ge_intr_coal_delay
;
2581 init_waitqueue_head(&mgp
->down_wq
);
2583 if (pci_enable_device(pdev
)) {
2584 dev_err(&pdev
->dev
, "pci_enable_device call failed\n");
2586 goto abort_with_netdev
;
2588 myri10ge_select_firmware(mgp
);
2590 /* Find the vendor-specific cap so we can check
2591 * the reboot register later on */
2592 mgp
->vendor_specific_offset
2593 = pci_find_capability(pdev
, PCI_CAP_ID_VNDR
);
2595 /* Set our max read request to 4KB */
2596 cap
= pci_find_capability(pdev
, PCI_CAP_ID_EXP
);
2598 dev_err(&pdev
->dev
, "Bad PCI_CAP_ID_EXP location %d\n", cap
);
2599 goto abort_with_netdev
;
2601 status
= pci_read_config_word(pdev
, cap
+ PCI_EXP_DEVCTL
, &val
);
2603 dev_err(&pdev
->dev
, "Error %d reading PCI_EXP_DEVCTL\n",
2605 goto abort_with_netdev
;
2607 val
= (val
& ~PCI_EXP_DEVCTL_READRQ
) | (5 << 12);
2608 status
= pci_write_config_word(pdev
, cap
+ PCI_EXP_DEVCTL
, val
);
2610 dev_err(&pdev
->dev
, "Error %d writing PCI_EXP_DEVCTL\n",
2612 goto abort_with_netdev
;
2615 pci_set_master(pdev
);
2617 status
= pci_set_dma_mask(pdev
, DMA_64BIT_MASK
);
2621 "64-bit pci address mask was refused, trying 32-bit");
2622 status
= pci_set_dma_mask(pdev
, DMA_32BIT_MASK
);
2625 dev_err(&pdev
->dev
, "Error %d setting DMA mask\n", status
);
2626 goto abort_with_netdev
;
2628 mgp
->cmd
= dma_alloc_coherent(&pdev
->dev
, sizeof(*mgp
->cmd
),
2629 &mgp
->cmd_bus
, GFP_KERNEL
);
2630 if (mgp
->cmd
== NULL
)
2631 goto abort_with_netdev
;
2633 mgp
->fw_stats
= dma_alloc_coherent(&pdev
->dev
, sizeof(*mgp
->fw_stats
),
2634 &mgp
->fw_stats_bus
, GFP_KERNEL
);
2635 if (mgp
->fw_stats
== NULL
)
2636 goto abort_with_cmd
;
2638 mgp
->board_span
= pci_resource_len(pdev
, 0);
2639 mgp
->iomem_base
= pci_resource_start(pdev
, 0);
2642 mgp
->mtrr
= mtrr_add(mgp
->iomem_base
, mgp
->board_span
,
2643 MTRR_TYPE_WRCOMB
, 1);
2645 /* Hack. need to get rid of these magic numbers */
2647 2 * 1024 * 1024 - (2 * (48 * 1024) + (32 * 1024)) - 0x100;
2648 if (mgp
->sram_size
> mgp
->board_span
) {
2649 dev_err(&pdev
->dev
, "board span %ld bytes too small\n",
2653 mgp
->sram
= ioremap(mgp
->iomem_base
, mgp
->board_span
);
2654 if (mgp
->sram
== NULL
) {
2655 dev_err(&pdev
->dev
, "ioremap failed for %ld bytes at 0x%lx\n",
2656 mgp
->board_span
, mgp
->iomem_base
);
2660 memcpy_fromio(mgp
->eeprom_strings
,
2661 mgp
->sram
+ mgp
->sram_size
- MYRI10GE_EEPROM_STRINGS_SIZE
,
2662 MYRI10GE_EEPROM_STRINGS_SIZE
);
2663 memset(mgp
->eeprom_strings
+ MYRI10GE_EEPROM_STRINGS_SIZE
- 2, 0, 2);
2664 status
= myri10ge_read_mac_addr(mgp
);
2666 goto abort_with_ioremap
;
2668 for (i
= 0; i
< ETH_ALEN
; i
++)
2669 netdev
->dev_addr
[i
] = mgp
->mac_addr
[i
];
2671 /* allocate rx done ring */
2672 bytes
= myri10ge_max_intr_slots
* sizeof(*mgp
->rx_done
.entry
);
2673 mgp
->rx_done
.entry
= dma_alloc_coherent(&pdev
->dev
, bytes
,
2674 &mgp
->rx_done
.bus
, GFP_KERNEL
);
2675 if (mgp
->rx_done
.entry
== NULL
)
2676 goto abort_with_ioremap
;
2677 memset(mgp
->rx_done
.entry
, 0, bytes
);
2679 status
= myri10ge_load_firmware(mgp
);
2681 dev_err(&pdev
->dev
, "failed to load firmware\n");
2682 goto abort_with_rx_done
;
2685 status
= myri10ge_reset(mgp
);
2687 dev_err(&pdev
->dev
, "failed reset\n");
2688 goto abort_with_firmware
;
2692 status
= pci_enable_msi(pdev
);
2695 "Error %d setting up MSI; falling back to xPIC\n",
2698 mgp
->msi_enabled
= 1;
2701 status
= request_irq(pdev
->irq
, myri10ge_intr
, IRQF_SHARED
,
2704 dev_err(&pdev
->dev
, "failed to allocate IRQ\n");
2705 goto abort_with_firmware
;
2708 pci_set_drvdata(pdev
, mgp
);
2709 if ((myri10ge_initial_mtu
+ ETH_HLEN
) > MYRI10GE_MAX_ETHER_MTU
)
2710 myri10ge_initial_mtu
= MYRI10GE_MAX_ETHER_MTU
- ETH_HLEN
;
2711 if ((myri10ge_initial_mtu
+ ETH_HLEN
) < 68)
2712 myri10ge_initial_mtu
= 68;
2713 netdev
->mtu
= myri10ge_initial_mtu
;
2714 netdev
->open
= myri10ge_open
;
2715 netdev
->stop
= myri10ge_close
;
2716 netdev
->hard_start_xmit
= myri10ge_xmit
;
2717 netdev
->get_stats
= myri10ge_get_stats
;
2718 netdev
->base_addr
= mgp
->iomem_base
;
2719 netdev
->irq
= pdev
->irq
;
2720 netdev
->change_mtu
= myri10ge_change_mtu
;
2721 netdev
->set_multicast_list
= myri10ge_set_multicast_list
;
2722 netdev
->set_mac_address
= myri10ge_set_mac_address
;
2723 netdev
->features
= NETIF_F_SG
| NETIF_F_HW_CSUM
| NETIF_F_TSO
;
2725 netdev
->features
|= NETIF_F_HIGHDMA
;
2726 netdev
->poll
= myri10ge_poll
;
2727 netdev
->weight
= myri10ge_napi_weight
;
2729 /* Save configuration space to be restored if the
2730 * nic resets due to a parity error */
2731 myri10ge_save_state(mgp
);
2733 /* Setup the watchdog timer */
2734 setup_timer(&mgp
->watchdog_timer
, myri10ge_watchdog_timer
,
2735 (unsigned long)mgp
);
2737 SET_ETHTOOL_OPS(netdev
, &myri10ge_ethtool_ops
);
2738 INIT_WORK(&mgp
->watchdog_work
, myri10ge_watchdog
, mgp
);
2739 status
= register_netdev(netdev
);
2741 dev_err(&pdev
->dev
, "register_netdev failed: %d\n", status
);
2742 goto abort_with_irq
;
2744 dev_info(dev
, "%s IRQ %d, tx bndry %d, fw %s, WC %s\n",
2745 (mgp
->msi_enabled
? "MSI" : "xPIC"),
2746 pdev
->irq
, mgp
->tx
.boundary
, mgp
->fw_name
,
2747 (mgp
->mtrr
>= 0 ? "Enabled" : "Disabled"));
2752 free_irq(pdev
->irq
, mgp
);
2753 if (mgp
->msi_enabled
)
2754 pci_disable_msi(pdev
);
2756 abort_with_firmware
:
2757 myri10ge_dummy_rdma(mgp
, 0);
2760 bytes
= myri10ge_max_intr_slots
* sizeof(*mgp
->rx_done
.entry
);
2761 dma_free_coherent(&pdev
->dev
, bytes
,
2762 mgp
->rx_done
.entry
, mgp
->rx_done
.bus
);
2770 mtrr_del(mgp
->mtrr
, mgp
->iomem_base
, mgp
->board_span
);
2772 dma_free_coherent(&pdev
->dev
, sizeof(*mgp
->fw_stats
),
2773 mgp
->fw_stats
, mgp
->fw_stats_bus
);
2776 dma_free_coherent(&pdev
->dev
, sizeof(*mgp
->cmd
),
2777 mgp
->cmd
, mgp
->cmd_bus
);
2781 free_netdev(netdev
);
2788 * Does what is necessary to shutdown one Myrinet device. Called
2789 * once for each Myrinet card by the kernel when a module is
2792 static void myri10ge_remove(struct pci_dev
*pdev
)
2794 struct myri10ge_priv
*mgp
;
2795 struct net_device
*netdev
;
2798 mgp
= pci_get_drvdata(pdev
);
2802 flush_scheduled_work();
2804 unregister_netdev(netdev
);
2805 free_irq(pdev
->irq
, mgp
);
2806 if (mgp
->msi_enabled
)
2807 pci_disable_msi(pdev
);
2809 myri10ge_dummy_rdma(mgp
, 0);
2811 bytes
= myri10ge_max_intr_slots
* sizeof(*mgp
->rx_done
.entry
);
2812 dma_free_coherent(&pdev
->dev
, bytes
,
2813 mgp
->rx_done
.entry
, mgp
->rx_done
.bus
);
2819 mtrr_del(mgp
->mtrr
, mgp
->iomem_base
, mgp
->board_span
);
2821 dma_free_coherent(&pdev
->dev
, sizeof(*mgp
->fw_stats
),
2822 mgp
->fw_stats
, mgp
->fw_stats_bus
);
2824 dma_free_coherent(&pdev
->dev
, sizeof(*mgp
->cmd
),
2825 mgp
->cmd
, mgp
->cmd_bus
);
2827 free_netdev(netdev
);
2828 pci_set_drvdata(pdev
, NULL
);
2831 #define PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E 0x0008
2833 static struct pci_device_id myri10ge_pci_tbl
[] = {
2834 {PCI_DEVICE(PCI_VENDOR_ID_MYRICOM
, PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E
)},
2838 static struct pci_driver myri10ge_driver
= {
2840 .probe
= myri10ge_probe
,
2841 .remove
= myri10ge_remove
,
2842 .id_table
= myri10ge_pci_tbl
,
2844 .suspend
= myri10ge_suspend
,
2845 .resume
= myri10ge_resume
,
2849 static __init
int myri10ge_init_module(void)
2851 printk(KERN_INFO
"%s: Version %s\n", myri10ge_driver
.name
,
2852 MYRI10GE_VERSION_STR
);
2853 return pci_register_driver(&myri10ge_driver
);
2856 module_init(myri10ge_init_module
);
2858 static __exit
void myri10ge_cleanup_module(void)
2860 pci_unregister_driver(&myri10ge_driver
);
2863 module_exit(myri10ge_cleanup_module
);