1 /*************************************************************************
2 * myri10ge.c: Myricom Myri-10G Ethernet driver.
4 * Copyright (C) 2005 - 2009 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 COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
23 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
43 #include <linux/tcp.h>
44 #include <linux/netdevice.h>
45 #include <linux/skbuff.h>
46 #include <linux/string.h>
47 #include <linux/module.h>
48 #include <linux/pci.h>
49 #include <linux/dma-mapping.h>
50 #include <linux/etherdevice.h>
51 #include <linux/if_ether.h>
52 #include <linux/if_vlan.h>
53 #include <linux/inet_lro.h>
54 #include <linux/dca.h>
56 #include <linux/inet.h>
58 #include <linux/ethtool.h>
59 #include <linux/firmware.h>
60 #include <linux/delay.h>
61 #include <linux/timer.h>
62 #include <linux/vmalloc.h>
63 #include <linux/crc32.h>
64 #include <linux/moduleparam.h>
66 #include <linux/log2.h>
67 #include <linux/slab.h>
68 #include <net/checksum.h>
71 #include <asm/byteorder.h>
73 #include <asm/processor.h>
78 #include "myri10ge_mcp.h"
79 #include "myri10ge_mcp_gen_header.h"
81 #define MYRI10GE_VERSION_STR "1.5.2-1.459"
83 MODULE_DESCRIPTION("Myricom 10G driver (10GbE)");
84 MODULE_AUTHOR("Maintainer: help@myri.com");
85 MODULE_VERSION(MYRI10GE_VERSION_STR
);
86 MODULE_LICENSE("Dual BSD/GPL");
88 #define MYRI10GE_MAX_ETHER_MTU 9014
90 #define MYRI10GE_ETH_STOPPED 0
91 #define MYRI10GE_ETH_STOPPING 1
92 #define MYRI10GE_ETH_STARTING 2
93 #define MYRI10GE_ETH_RUNNING 3
94 #define MYRI10GE_ETH_OPEN_FAILED 4
96 #define MYRI10GE_EEPROM_STRINGS_SIZE 256
97 #define MYRI10GE_MAX_SEND_DESC_TSO ((65536 / 2048) * 2)
98 #define MYRI10GE_MAX_LRO_DESCRIPTORS 8
99 #define MYRI10GE_LRO_MAX_PKTS 64
101 #define MYRI10GE_NO_CONFIRM_DATA htonl(0xffffffff)
102 #define MYRI10GE_NO_RESPONSE_RESULT 0xffffffff
104 #define MYRI10GE_ALLOC_ORDER 0
105 #define MYRI10GE_ALLOC_SIZE ((1 << MYRI10GE_ALLOC_ORDER) * PAGE_SIZE)
106 #define MYRI10GE_MAX_FRAGS_PER_FRAME (MYRI10GE_MAX_ETHER_MTU/MYRI10GE_ALLOC_SIZE + 1)
108 #define MYRI10GE_MAX_SLICES 32
110 struct myri10ge_rx_buffer_state
{
113 DEFINE_DMA_UNMAP_ADDR(bus
);
114 DEFINE_DMA_UNMAP_LEN(len
);
117 struct myri10ge_tx_buffer_state
{
120 DEFINE_DMA_UNMAP_ADDR(bus
);
121 DEFINE_DMA_UNMAP_LEN(len
);
124 struct myri10ge_cmd
{
130 struct myri10ge_rx_buf
{
131 struct mcp_kreq_ether_recv __iomem
*lanai
; /* lanai ptr for recv ring */
132 struct mcp_kreq_ether_recv
*shadow
; /* host shadow of recv ring */
133 struct myri10ge_rx_buffer_state
*info
;
140 int mask
; /* number of rx slots -1 */
144 struct myri10ge_tx_buf
{
145 struct mcp_kreq_ether_send __iomem
*lanai
; /* lanai ptr for sendq */
146 __be32 __iomem
*send_go
; /* "go" doorbell ptr */
147 __be32 __iomem
*send_stop
; /* "stop" doorbell ptr */
148 struct mcp_kreq_ether_send
*req_list
; /* host shadow of sendq */
150 struct myri10ge_tx_buffer_state
*info
;
151 int mask
; /* number of transmit slots -1 */
152 int req ____cacheline_aligned
; /* transmit slots submitted */
153 int pkt_start
; /* packets started */
156 int done ____cacheline_aligned
; /* transmit slots completed */
157 int pkt_done
; /* packets completed */
162 struct myri10ge_rx_done
{
163 struct mcp_slot
*entry
;
167 struct net_lro_mgr lro_mgr
;
168 struct net_lro_desc lro_desc
[MYRI10GE_MAX_LRO_DESCRIPTORS
];
171 struct myri10ge_slice_netstats
{
172 unsigned long rx_packets
;
173 unsigned long tx_packets
;
174 unsigned long rx_bytes
;
175 unsigned long tx_bytes
;
176 unsigned long rx_dropped
;
177 unsigned long tx_dropped
;
180 struct myri10ge_slice_state
{
181 struct myri10ge_tx_buf tx
; /* transmit ring */
182 struct myri10ge_rx_buf rx_small
;
183 struct myri10ge_rx_buf rx_big
;
184 struct myri10ge_rx_done rx_done
;
185 struct net_device
*dev
;
186 struct napi_struct napi
;
187 struct myri10ge_priv
*mgp
;
188 struct myri10ge_slice_netstats stats
;
189 __be32 __iomem
*irq_claim
;
190 struct mcp_irq_data
*fw_stats
;
191 dma_addr_t fw_stats_bus
;
192 int watchdog_tx_done
;
194 int watchdog_rx_done
;
195 #ifdef CONFIG_MYRI10GE_DCA
198 __be32 __iomem
*dca_tag
;
203 struct myri10ge_priv
{
204 struct myri10ge_slice_state
*ss
;
205 int tx_boundary
; /* boundary transmits cannot cross */
207 int running
; /* running? */
208 int csum_flag
; /* rx_csums? */
212 struct net_device
*dev
;
213 spinlock_t stats_lock
;
216 unsigned long board_span
;
217 unsigned long iomem_base
;
218 __be32 __iomem
*irq_deassert
;
219 char *mac_addr_string
;
220 struct mcp_cmd_response
*cmd
;
222 struct pci_dev
*pdev
;
225 struct msix_entry
*msix_vectors
;
226 #ifdef CONFIG_MYRI10GE_DCA
230 unsigned int rdma_tags_available
;
232 __be32 __iomem
*intr_coal_delay_ptr
;
236 wait_queue_head_t down_wq
;
237 struct work_struct watchdog_work
;
238 struct timer_list watchdog_timer
;
242 bool fw_name_allocated
;
244 char eeprom_strings
[MYRI10GE_EEPROM_STRINGS_SIZE
];
245 char *product_code_string
;
246 char fw_version
[128];
250 int adopted_rx_filter_bug
;
251 u8 mac_addr
[6]; /* eeprom mac address */
252 unsigned long serial_number
;
253 int vendor_specific_offset
;
254 int fw_multicast_support
;
255 unsigned long features
;
262 unsigned int board_number
;
266 static char *myri10ge_fw_unaligned
= "myri10ge_ethp_z8e.dat";
267 static char *myri10ge_fw_aligned
= "myri10ge_eth_z8e.dat";
268 static char *myri10ge_fw_rss_unaligned
= "myri10ge_rss_ethp_z8e.dat";
269 static char *myri10ge_fw_rss_aligned
= "myri10ge_rss_eth_z8e.dat";
270 MODULE_FIRMWARE("myri10ge_ethp_z8e.dat");
271 MODULE_FIRMWARE("myri10ge_eth_z8e.dat");
272 MODULE_FIRMWARE("myri10ge_rss_ethp_z8e.dat");
273 MODULE_FIRMWARE("myri10ge_rss_eth_z8e.dat");
275 /* Careful: must be accessed under kparam_block_sysfs_write */
276 static char *myri10ge_fw_name
= NULL
;
277 module_param(myri10ge_fw_name
, charp
, S_IRUGO
| S_IWUSR
);
278 MODULE_PARM_DESC(myri10ge_fw_name
, "Firmware image name");
280 #define MYRI10GE_MAX_BOARDS 8
281 static char *myri10ge_fw_names
[MYRI10GE_MAX_BOARDS
] =
282 {[0 ... (MYRI10GE_MAX_BOARDS
- 1)] = NULL
};
283 module_param_array_named(myri10ge_fw_names
, myri10ge_fw_names
, charp
, NULL
,
285 MODULE_PARM_DESC(myri10ge_fw_name
, "Firmware image names per board");
287 static int myri10ge_ecrc_enable
= 1;
288 module_param(myri10ge_ecrc_enable
, int, S_IRUGO
);
289 MODULE_PARM_DESC(myri10ge_ecrc_enable
, "Enable Extended CRC on PCI-E");
291 static int myri10ge_small_bytes
= -1; /* -1 == auto */
292 module_param(myri10ge_small_bytes
, int, S_IRUGO
| S_IWUSR
);
293 MODULE_PARM_DESC(myri10ge_small_bytes
, "Threshold of small packets");
295 static int myri10ge_msi
= 1; /* enable msi by default */
296 module_param(myri10ge_msi
, int, S_IRUGO
| S_IWUSR
);
297 MODULE_PARM_DESC(myri10ge_msi
, "Enable Message Signalled Interrupts");
299 static int myri10ge_intr_coal_delay
= 75;
300 module_param(myri10ge_intr_coal_delay
, int, S_IRUGO
);
301 MODULE_PARM_DESC(myri10ge_intr_coal_delay
, "Interrupt coalescing delay");
303 static int myri10ge_flow_control
= 1;
304 module_param(myri10ge_flow_control
, int, S_IRUGO
);
305 MODULE_PARM_DESC(myri10ge_flow_control
, "Pause parameter");
307 static int myri10ge_deassert_wait
= 1;
308 module_param(myri10ge_deassert_wait
, int, S_IRUGO
| S_IWUSR
);
309 MODULE_PARM_DESC(myri10ge_deassert_wait
,
310 "Wait when deasserting legacy interrupts");
312 static int myri10ge_force_firmware
= 0;
313 module_param(myri10ge_force_firmware
, int, S_IRUGO
);
314 MODULE_PARM_DESC(myri10ge_force_firmware
,
315 "Force firmware to assume aligned completions");
317 static int myri10ge_initial_mtu
= MYRI10GE_MAX_ETHER_MTU
- ETH_HLEN
;
318 module_param(myri10ge_initial_mtu
, int, S_IRUGO
);
319 MODULE_PARM_DESC(myri10ge_initial_mtu
, "Initial MTU");
321 static int myri10ge_napi_weight
= 64;
322 module_param(myri10ge_napi_weight
, int, S_IRUGO
);
323 MODULE_PARM_DESC(myri10ge_napi_weight
, "Set NAPI weight");
325 static int myri10ge_watchdog_timeout
= 1;
326 module_param(myri10ge_watchdog_timeout
, int, S_IRUGO
);
327 MODULE_PARM_DESC(myri10ge_watchdog_timeout
, "Set watchdog timeout");
329 static int myri10ge_max_irq_loops
= 1048576;
330 module_param(myri10ge_max_irq_loops
, int, S_IRUGO
);
331 MODULE_PARM_DESC(myri10ge_max_irq_loops
,
332 "Set stuck legacy IRQ detection threshold");
334 #define MYRI10GE_MSG_DEFAULT NETIF_MSG_LINK
336 static int myri10ge_debug
= -1; /* defaults above */
337 module_param(myri10ge_debug
, int, 0);
338 MODULE_PARM_DESC(myri10ge_debug
, "Debug level (0=none,...,16=all)");
340 static int myri10ge_lro_max_pkts
= MYRI10GE_LRO_MAX_PKTS
;
341 module_param(myri10ge_lro_max_pkts
, int, S_IRUGO
);
342 MODULE_PARM_DESC(myri10ge_lro_max_pkts
,
343 "Number of LRO packets to be aggregated");
345 static int myri10ge_fill_thresh
= 256;
346 module_param(myri10ge_fill_thresh
, int, S_IRUGO
| S_IWUSR
);
347 MODULE_PARM_DESC(myri10ge_fill_thresh
, "Number of empty rx slots allowed");
349 static int myri10ge_reset_recover
= 1;
351 static int myri10ge_max_slices
= 1;
352 module_param(myri10ge_max_slices
, int, S_IRUGO
);
353 MODULE_PARM_DESC(myri10ge_max_slices
, "Max tx/rx queues");
355 static int myri10ge_rss_hash
= MXGEFW_RSS_HASH_TYPE_SRC_DST_PORT
;
356 module_param(myri10ge_rss_hash
, int, S_IRUGO
);
357 MODULE_PARM_DESC(myri10ge_rss_hash
, "Type of RSS hashing to do");
359 static int myri10ge_dca
= 1;
360 module_param(myri10ge_dca
, int, S_IRUGO
);
361 MODULE_PARM_DESC(myri10ge_dca
, "Enable DCA if possible");
363 #define MYRI10GE_FW_OFFSET 1024*1024
364 #define MYRI10GE_HIGHPART_TO_U32(X) \
365 (sizeof (X) == 8) ? ((u32)((u64)(X) >> 32)) : (0)
366 #define MYRI10GE_LOWPART_TO_U32(X) ((u32)(X))
368 #define myri10ge_pio_copy(to,from,size) __iowrite64_copy(to,from,size/8)
370 static void myri10ge_set_multicast_list(struct net_device
*dev
);
371 static netdev_tx_t
myri10ge_sw_tso(struct sk_buff
*skb
,
372 struct net_device
*dev
);
374 static inline void put_be32(__be32 val
, __be32 __iomem
* p
)
376 __raw_writel((__force __u32
) val
, (__force
void __iomem
*)p
);
379 static struct net_device_stats
*myri10ge_get_stats(struct net_device
*dev
);
381 static void set_fw_name(struct myri10ge_priv
*mgp
, char *name
, bool allocated
)
383 if (mgp
->fw_name_allocated
)
386 mgp
->fw_name_allocated
= allocated
;
390 myri10ge_send_cmd(struct myri10ge_priv
*mgp
, u32 cmd
,
391 struct myri10ge_cmd
*data
, int atomic
)
394 char buf_bytes
[sizeof(*buf
) + 8];
395 struct mcp_cmd_response
*response
= mgp
->cmd
;
396 char __iomem
*cmd_addr
= mgp
->sram
+ MXGEFW_ETH_CMD
;
397 u32 dma_low
, dma_high
, result
, value
;
400 /* ensure buf is aligned to 8 bytes */
401 buf
= (struct mcp_cmd
*)ALIGN((unsigned long)buf_bytes
, 8);
403 buf
->data0
= htonl(data
->data0
);
404 buf
->data1
= htonl(data
->data1
);
405 buf
->data2
= htonl(data
->data2
);
406 buf
->cmd
= htonl(cmd
);
407 dma_low
= MYRI10GE_LOWPART_TO_U32(mgp
->cmd_bus
);
408 dma_high
= MYRI10GE_HIGHPART_TO_U32(mgp
->cmd_bus
);
410 buf
->response_addr
.low
= htonl(dma_low
);
411 buf
->response_addr
.high
= htonl(dma_high
);
412 response
->result
= htonl(MYRI10GE_NO_RESPONSE_RESULT
);
414 myri10ge_pio_copy(cmd_addr
, buf
, sizeof(*buf
));
416 /* wait up to 15ms. Longest command is the DMA benchmark,
417 * which is capped at 5ms, but runs from a timeout handler
418 * that runs every 7.8ms. So a 15ms timeout leaves us with
422 /* if atomic is set, do not sleep,
423 * and try to get the completion quickly
424 * (1ms will be enough for those commands) */
425 for (sleep_total
= 0;
426 sleep_total
< 1000 &&
427 response
->result
== htonl(MYRI10GE_NO_RESPONSE_RESULT
);
433 /* use msleep for most command */
434 for (sleep_total
= 0;
436 response
->result
== htonl(MYRI10GE_NO_RESPONSE_RESULT
);
441 result
= ntohl(response
->result
);
442 value
= ntohl(response
->data
);
443 if (result
!= MYRI10GE_NO_RESPONSE_RESULT
) {
447 } else if (result
== MXGEFW_CMD_UNKNOWN
) {
449 } else if (result
== MXGEFW_CMD_ERROR_UNALIGNED
) {
451 } else if (result
== MXGEFW_CMD_ERROR_RANGE
&&
452 cmd
== MXGEFW_CMD_ENABLE_RSS_QUEUES
&&
454 data1
& MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES
) !=
458 dev_err(&mgp
->pdev
->dev
,
459 "command %d failed, result = %d\n",
465 dev_err(&mgp
->pdev
->dev
, "command %d timed out, result = %d\n",
471 * The eeprom strings on the lanaiX have the format
474 * PT:ddd mmm xx xx:xx:xx xx\0
475 * PV:ddd mmm xx xx:xx:xx xx\0
477 static int myri10ge_read_mac_addr(struct myri10ge_priv
*mgp
)
482 ptr
= mgp
->eeprom_strings
;
483 limit
= mgp
->eeprom_strings
+ MYRI10GE_EEPROM_STRINGS_SIZE
;
485 while (*ptr
!= '\0' && ptr
< limit
) {
486 if (memcmp(ptr
, "MAC=", 4) == 0) {
488 mgp
->mac_addr_string
= ptr
;
489 for (i
= 0; i
< 6; i
++) {
490 if ((ptr
+ 2) > limit
)
493 simple_strtoul(ptr
, &ptr
, 16);
497 if (memcmp(ptr
, "PC=", 3) == 0) {
499 mgp
->product_code_string
= ptr
;
501 if (memcmp((const void *)ptr
, "SN=", 3) == 0) {
503 mgp
->serial_number
= simple_strtoul(ptr
, &ptr
, 10);
505 while (ptr
< limit
&& *ptr
++) ;
511 dev_err(&mgp
->pdev
->dev
, "failed to parse eeprom_strings\n");
516 * Enable or disable periodic RDMAs from the host to make certain
517 * chipsets resend dropped PCIe messages
520 static void myri10ge_dummy_rdma(struct myri10ge_priv
*mgp
, int enable
)
522 char __iomem
*submit
;
523 __be32 buf
[16] __attribute__ ((__aligned__(8)));
524 u32 dma_low
, dma_high
;
527 /* clear confirmation addr */
531 /* send a rdma command to the PCIe engine, and wait for the
532 * response in the confirmation address. The firmware should
533 * write a -1 there to indicate it is alive and well
535 dma_low
= MYRI10GE_LOWPART_TO_U32(mgp
->cmd_bus
);
536 dma_high
= MYRI10GE_HIGHPART_TO_U32(mgp
->cmd_bus
);
538 buf
[0] = htonl(dma_high
); /* confirm addr MSW */
539 buf
[1] = htonl(dma_low
); /* confirm addr LSW */
540 buf
[2] = MYRI10GE_NO_CONFIRM_DATA
; /* confirm data */
541 buf
[3] = htonl(dma_high
); /* dummy addr MSW */
542 buf
[4] = htonl(dma_low
); /* dummy addr LSW */
543 buf
[5] = htonl(enable
); /* enable? */
545 submit
= mgp
->sram
+ MXGEFW_BOOT_DUMMY_RDMA
;
547 myri10ge_pio_copy(submit
, &buf
, sizeof(buf
));
548 for (i
= 0; mgp
->cmd
->data
!= MYRI10GE_NO_CONFIRM_DATA
&& i
< 20; i
++)
550 if (mgp
->cmd
->data
!= MYRI10GE_NO_CONFIRM_DATA
)
551 dev_err(&mgp
->pdev
->dev
, "dummy rdma %s failed\n",
552 (enable
? "enable" : "disable"));
556 myri10ge_validate_firmware(struct myri10ge_priv
*mgp
,
557 struct mcp_gen_header
*hdr
)
559 struct device
*dev
= &mgp
->pdev
->dev
;
561 /* check firmware type */
562 if (ntohl(hdr
->mcp_type
) != MCP_TYPE_ETH
) {
563 dev_err(dev
, "Bad firmware type: 0x%x\n", ntohl(hdr
->mcp_type
));
567 /* save firmware version for ethtool */
568 strncpy(mgp
->fw_version
, hdr
->version
, sizeof(mgp
->fw_version
));
570 sscanf(mgp
->fw_version
, "%d.%d.%d", &mgp
->fw_ver_major
,
571 &mgp
->fw_ver_minor
, &mgp
->fw_ver_tiny
);
573 if (!(mgp
->fw_ver_major
== MXGEFW_VERSION_MAJOR
&&
574 mgp
->fw_ver_minor
== MXGEFW_VERSION_MINOR
)) {
575 dev_err(dev
, "Found firmware version %s\n", mgp
->fw_version
);
576 dev_err(dev
, "Driver needs %d.%d\n", MXGEFW_VERSION_MAJOR
,
577 MXGEFW_VERSION_MINOR
);
583 static int myri10ge_load_hotplug_firmware(struct myri10ge_priv
*mgp
, u32
* size
)
585 unsigned crc
, reread_crc
;
586 const struct firmware
*fw
;
587 struct device
*dev
= &mgp
->pdev
->dev
;
588 unsigned char *fw_readback
;
589 struct mcp_gen_header
*hdr
;
594 if ((status
= request_firmware(&fw
, mgp
->fw_name
, dev
)) < 0) {
595 dev_err(dev
, "Unable to load %s firmware image via hotplug\n",
598 goto abort_with_nothing
;
603 if (fw
->size
>= mgp
->sram_size
- MYRI10GE_FW_OFFSET
||
604 fw
->size
< MCP_HEADER_PTR_OFFSET
+ 4) {
605 dev_err(dev
, "Firmware size invalid:%d\n", (int)fw
->size
);
611 hdr_offset
= ntohl(*(__be32
*) (fw
->data
+ MCP_HEADER_PTR_OFFSET
));
612 if ((hdr_offset
& 3) || hdr_offset
+ sizeof(*hdr
) > fw
->size
) {
613 dev_err(dev
, "Bad firmware file\n");
617 hdr
= (void *)(fw
->data
+ hdr_offset
);
619 status
= myri10ge_validate_firmware(mgp
, hdr
);
623 crc
= crc32(~0, fw
->data
, fw
->size
);
624 for (i
= 0; i
< fw
->size
; i
+= 256) {
625 myri10ge_pio_copy(mgp
->sram
+ MYRI10GE_FW_OFFSET
+ i
,
627 min(256U, (unsigned)(fw
->size
- i
)));
631 fw_readback
= vmalloc(fw
->size
);
636 /* corruption checking is good for parity recovery and buggy chipset */
637 memcpy_fromio(fw_readback
, mgp
->sram
+ MYRI10GE_FW_OFFSET
, fw
->size
);
638 reread_crc
= crc32(~0, fw_readback
, fw
->size
);
640 if (crc
!= reread_crc
) {
641 dev_err(dev
, "CRC failed(fw-len=%u), got 0x%x (expect 0x%x)\n",
642 (unsigned)fw
->size
, reread_crc
, crc
);
646 *size
= (u32
) fw
->size
;
649 release_firmware(fw
);
655 static int myri10ge_adopt_running_firmware(struct myri10ge_priv
*mgp
)
657 struct mcp_gen_header
*hdr
;
658 struct device
*dev
= &mgp
->pdev
->dev
;
659 const size_t bytes
= sizeof(struct mcp_gen_header
);
663 /* find running firmware header */
664 hdr_offset
= swab32(readl(mgp
->sram
+ MCP_HEADER_PTR_OFFSET
));
666 if ((hdr_offset
& 3) || hdr_offset
+ sizeof(*hdr
) > mgp
->sram_size
) {
667 dev_err(dev
, "Running firmware has bad header offset (%d)\n",
672 /* copy header of running firmware from SRAM to host memory to
673 * validate firmware */
674 hdr
= kmalloc(bytes
, GFP_KERNEL
);
676 dev_err(dev
, "could not malloc firmware hdr\n");
679 memcpy_fromio(hdr
, mgp
->sram
+ hdr_offset
, bytes
);
680 status
= myri10ge_validate_firmware(mgp
, hdr
);
683 /* check to see if adopted firmware has bug where adopting
684 * it will cause broadcasts to be filtered unless the NIC
685 * is kept in ALLMULTI mode */
686 if (mgp
->fw_ver_major
== 1 && mgp
->fw_ver_minor
== 4 &&
687 mgp
->fw_ver_tiny
>= 4 && mgp
->fw_ver_tiny
<= 11) {
688 mgp
->adopted_rx_filter_bug
= 1;
689 dev_warn(dev
, "Adopting fw %d.%d.%d: "
690 "working around rx filter bug\n",
691 mgp
->fw_ver_major
, mgp
->fw_ver_minor
,
697 static int myri10ge_get_firmware_capabilities(struct myri10ge_priv
*mgp
)
699 struct myri10ge_cmd cmd
;
702 /* probe for IPv6 TSO support */
703 mgp
->features
= NETIF_F_SG
| NETIF_F_HW_CSUM
| NETIF_F_TSO
;
704 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_MAX_TSO6_HDR_SIZE
,
707 mgp
->max_tso6
= cmd
.data0
;
708 mgp
->features
|= NETIF_F_TSO6
;
711 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_RX_RING_SIZE
, &cmd
, 0);
713 dev_err(&mgp
->pdev
->dev
,
714 "failed MXGEFW_CMD_GET_RX_RING_SIZE\n");
718 mgp
->max_intr_slots
= 2 * (cmd
.data0
/ sizeof(struct mcp_dma_addr
));
723 static int myri10ge_load_firmware(struct myri10ge_priv
*mgp
, int adopt
)
725 char __iomem
*submit
;
726 __be32 buf
[16] __attribute__ ((__aligned__(8)));
727 u32 dma_low
, dma_high
, size
;
731 status
= myri10ge_load_hotplug_firmware(mgp
, &size
);
735 dev_warn(&mgp
->pdev
->dev
, "hotplug firmware loading failed\n");
737 /* Do not attempt to adopt firmware if there
742 status
= myri10ge_adopt_running_firmware(mgp
);
744 dev_err(&mgp
->pdev
->dev
,
745 "failed to adopt running firmware\n");
748 dev_info(&mgp
->pdev
->dev
,
749 "Successfully adopted running firmware\n");
750 if (mgp
->tx_boundary
== 4096) {
751 dev_warn(&mgp
->pdev
->dev
,
752 "Using firmware currently running on NIC"
754 dev_warn(&mgp
->pdev
->dev
,
755 "performance consider loading optimized "
757 dev_warn(&mgp
->pdev
->dev
, "via hotplug\n");
760 set_fw_name(mgp
, "adopted", false);
761 mgp
->tx_boundary
= 2048;
762 myri10ge_dummy_rdma(mgp
, 1);
763 status
= myri10ge_get_firmware_capabilities(mgp
);
767 /* clear confirmation addr */
771 /* send a reload command to the bootstrap MCP, and wait for the
772 * response in the confirmation address. The firmware should
773 * write a -1 there to indicate it is alive and well
775 dma_low
= MYRI10GE_LOWPART_TO_U32(mgp
->cmd_bus
);
776 dma_high
= MYRI10GE_HIGHPART_TO_U32(mgp
->cmd_bus
);
778 buf
[0] = htonl(dma_high
); /* confirm addr MSW */
779 buf
[1] = htonl(dma_low
); /* confirm addr LSW */
780 buf
[2] = MYRI10GE_NO_CONFIRM_DATA
; /* confirm data */
782 /* FIX: All newest firmware should un-protect the bottom of
783 * the sram before handoff. However, the very first interfaces
784 * do not. Therefore the handoff copy must skip the first 8 bytes
786 buf
[3] = htonl(MYRI10GE_FW_OFFSET
+ 8); /* where the code starts */
787 buf
[4] = htonl(size
- 8); /* length of code */
788 buf
[5] = htonl(8); /* where to copy to */
789 buf
[6] = htonl(0); /* where to jump to */
791 submit
= mgp
->sram
+ MXGEFW_BOOT_HANDOFF
;
793 myri10ge_pio_copy(submit
, &buf
, sizeof(buf
));
798 while (mgp
->cmd
->data
!= MYRI10GE_NO_CONFIRM_DATA
&& i
< 9) {
802 if (mgp
->cmd
->data
!= MYRI10GE_NO_CONFIRM_DATA
) {
803 dev_err(&mgp
->pdev
->dev
, "handoff failed\n");
806 myri10ge_dummy_rdma(mgp
, 1);
807 status
= myri10ge_get_firmware_capabilities(mgp
);
812 static int myri10ge_update_mac_address(struct myri10ge_priv
*mgp
, u8
* addr
)
814 struct myri10ge_cmd cmd
;
817 cmd
.data0
= ((addr
[0] << 24) | (addr
[1] << 16)
818 | (addr
[2] << 8) | addr
[3]);
820 cmd
.data1
= ((addr
[4] << 8) | (addr
[5]));
822 status
= myri10ge_send_cmd(mgp
, MXGEFW_SET_MAC_ADDRESS
, &cmd
, 0);
826 static int myri10ge_change_pause(struct myri10ge_priv
*mgp
, int pause
)
828 struct myri10ge_cmd cmd
;
831 ctl
= pause
? MXGEFW_ENABLE_FLOW_CONTROL
: MXGEFW_DISABLE_FLOW_CONTROL
;
832 status
= myri10ge_send_cmd(mgp
, ctl
, &cmd
, 0);
835 netdev_err(mgp
->dev
, "Failed to set flow control mode\n");
843 myri10ge_change_promisc(struct myri10ge_priv
*mgp
, int promisc
, int atomic
)
845 struct myri10ge_cmd cmd
;
848 ctl
= promisc
? MXGEFW_ENABLE_PROMISC
: MXGEFW_DISABLE_PROMISC
;
849 status
= myri10ge_send_cmd(mgp
, ctl
, &cmd
, atomic
);
851 netdev_err(mgp
->dev
, "Failed to set promisc mode\n");
854 static int myri10ge_dma_test(struct myri10ge_priv
*mgp
, int test_type
)
856 struct myri10ge_cmd cmd
;
859 struct page
*dmatest_page
;
860 dma_addr_t dmatest_bus
;
863 dmatest_page
= alloc_page(GFP_KERNEL
);
866 dmatest_bus
= pci_map_page(mgp
->pdev
, dmatest_page
, 0, PAGE_SIZE
,
869 /* Run a small DMA test.
870 * The magic multipliers to the length tell the firmware
871 * to do DMA read, write, or read+write tests. The
872 * results are returned in cmd.data0. The upper 16
873 * bits or the return is the number of transfers completed.
874 * The lower 16 bits is the time in 0.5us ticks that the
875 * transfers took to complete.
878 len
= mgp
->tx_boundary
;
880 cmd
.data0
= MYRI10GE_LOWPART_TO_U32(dmatest_bus
);
881 cmd
.data1
= MYRI10GE_HIGHPART_TO_U32(dmatest_bus
);
882 cmd
.data2
= len
* 0x10000;
883 status
= myri10ge_send_cmd(mgp
, test_type
, &cmd
, 0);
888 mgp
->read_dma
= ((cmd
.data0
>> 16) * len
* 2) / (cmd
.data0
& 0xffff);
889 cmd
.data0
= MYRI10GE_LOWPART_TO_U32(dmatest_bus
);
890 cmd
.data1
= MYRI10GE_HIGHPART_TO_U32(dmatest_bus
);
891 cmd
.data2
= len
* 0x1;
892 status
= myri10ge_send_cmd(mgp
, test_type
, &cmd
, 0);
897 mgp
->write_dma
= ((cmd
.data0
>> 16) * len
* 2) / (cmd
.data0
& 0xffff);
899 cmd
.data0
= MYRI10GE_LOWPART_TO_U32(dmatest_bus
);
900 cmd
.data1
= MYRI10GE_HIGHPART_TO_U32(dmatest_bus
);
901 cmd
.data2
= len
* 0x10001;
902 status
= myri10ge_send_cmd(mgp
, test_type
, &cmd
, 0);
907 mgp
->read_write_dma
= ((cmd
.data0
>> 16) * len
* 2 * 2) /
908 (cmd
.data0
& 0xffff);
911 pci_unmap_page(mgp
->pdev
, dmatest_bus
, PAGE_SIZE
, DMA_BIDIRECTIONAL
);
912 put_page(dmatest_page
);
914 if (status
!= 0 && test_type
!= MXGEFW_CMD_UNALIGNED_TEST
)
915 dev_warn(&mgp
->pdev
->dev
, "DMA %s benchmark failed: %d\n",
921 static int myri10ge_reset(struct myri10ge_priv
*mgp
)
923 struct myri10ge_cmd cmd
;
924 struct myri10ge_slice_state
*ss
;
927 #ifdef CONFIG_MYRI10GE_DCA
928 unsigned long dca_tag_off
;
931 /* try to send a reset command to the card to see if it
933 memset(&cmd
, 0, sizeof(cmd
));
934 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_RESET
, &cmd
, 0);
936 dev_err(&mgp
->pdev
->dev
, "failed reset\n");
940 (void)myri10ge_dma_test(mgp
, MXGEFW_DMA_TEST
);
942 * Use non-ndis mcp_slot (eg, 4 bytes total,
943 * no toeplitz hash value returned. Older firmware will
944 * not understand this command, but will use the correct
945 * sized mcp_slot, so we ignore error returns
947 cmd
.data0
= MXGEFW_RSS_MCP_SLOT_TYPE_MIN
;
948 (void)myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_RSS_MCP_SLOT_TYPE
, &cmd
, 0);
950 /* Now exchange information about interrupts */
952 bytes
= mgp
->max_intr_slots
* sizeof(*mgp
->ss
[0].rx_done
.entry
);
953 cmd
.data0
= (u32
) bytes
;
954 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_INTRQ_SIZE
, &cmd
, 0);
957 * Even though we already know how many slices are supported
958 * via myri10ge_probe_slices() MXGEFW_CMD_GET_MAX_RSS_QUEUES
959 * has magic side effects, and must be called after a reset.
960 * It must be called prior to calling any RSS related cmds,
961 * including assigning an interrupt queue for anything but
962 * slice 0. It must also be called *after*
963 * MXGEFW_CMD_SET_INTRQ_SIZE, since the intrq size is used by
964 * the firmware to compute offsets.
967 if (mgp
->num_slices
> 1) {
969 /* ask the maximum number of slices it supports */
970 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_MAX_RSS_QUEUES
,
973 dev_err(&mgp
->pdev
->dev
,
974 "failed to get number of slices\n");
978 * MXGEFW_CMD_ENABLE_RSS_QUEUES must be called prior
979 * to setting up the interrupt queue DMA
982 cmd
.data0
= mgp
->num_slices
;
983 cmd
.data1
= MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE
;
984 if (mgp
->dev
->real_num_tx_queues
> 1)
985 cmd
.data1
|= MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES
;
986 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_ENABLE_RSS_QUEUES
,
989 /* Firmware older than 1.4.32 only supports multiple
990 * RX queues, so if we get an error, first retry using a
991 * single TX queue before giving up */
992 if (status
!= 0 && mgp
->dev
->real_num_tx_queues
> 1) {
993 mgp
->dev
->real_num_tx_queues
= 1;
994 cmd
.data0
= mgp
->num_slices
;
995 cmd
.data1
= MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE
;
996 status
= myri10ge_send_cmd(mgp
,
997 MXGEFW_CMD_ENABLE_RSS_QUEUES
,
1002 dev_err(&mgp
->pdev
->dev
,
1003 "failed to set number of slices\n");
1008 for (i
= 0; i
< mgp
->num_slices
; i
++) {
1010 cmd
.data0
= MYRI10GE_LOWPART_TO_U32(ss
->rx_done
.bus
);
1011 cmd
.data1
= MYRI10GE_HIGHPART_TO_U32(ss
->rx_done
.bus
);
1013 status
|= myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_INTRQ_DMA
,
1018 myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_IRQ_ACK_OFFSET
, &cmd
, 0);
1019 for (i
= 0; i
< mgp
->num_slices
; i
++) {
1022 (__iomem __be32
*) (mgp
->sram
+ cmd
.data0
+ 8 * i
);
1024 status
|= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_IRQ_DEASSERT_OFFSET
,
1026 mgp
->irq_deassert
= (__iomem __be32
*) (mgp
->sram
+ cmd
.data0
);
1028 status
|= myri10ge_send_cmd
1029 (mgp
, MXGEFW_CMD_GET_INTR_COAL_DELAY_OFFSET
, &cmd
, 0);
1030 mgp
->intr_coal_delay_ptr
= (__iomem __be32
*) (mgp
->sram
+ cmd
.data0
);
1032 dev_err(&mgp
->pdev
->dev
, "failed set interrupt parameters\n");
1035 put_be32(htonl(mgp
->intr_coal_delay
), mgp
->intr_coal_delay_ptr
);
1037 #ifdef CONFIG_MYRI10GE_DCA
1038 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_DCA_OFFSET
, &cmd
, 0);
1039 dca_tag_off
= cmd
.data0
;
1040 for (i
= 0; i
< mgp
->num_slices
; i
++) {
1043 ss
->dca_tag
= (__iomem __be32
*)
1044 (mgp
->sram
+ dca_tag_off
+ 4 * i
);
1049 #endif /* CONFIG_MYRI10GE_DCA */
1051 /* reset mcp/driver shared state back to 0 */
1053 mgp
->link_changes
= 0;
1054 for (i
= 0; i
< mgp
->num_slices
; i
++) {
1057 memset(ss
->rx_done
.entry
, 0, bytes
);
1060 ss
->tx
.pkt_start
= 0;
1061 ss
->tx
.pkt_done
= 0;
1063 ss
->rx_small
.cnt
= 0;
1064 ss
->rx_done
.idx
= 0;
1065 ss
->rx_done
.cnt
= 0;
1066 ss
->tx
.wake_queue
= 0;
1067 ss
->tx
.stop_queue
= 0;
1070 status
= myri10ge_update_mac_address(mgp
, mgp
->dev
->dev_addr
);
1071 myri10ge_change_pause(mgp
, mgp
->pause
);
1072 myri10ge_set_multicast_list(mgp
->dev
);
1076 #ifdef CONFIG_MYRI10GE_DCA
1078 myri10ge_write_dca(struct myri10ge_slice_state
*ss
, int cpu
, int tag
)
1081 ss
->cached_dca_tag
= tag
;
1082 put_be32(htonl(tag
), ss
->dca_tag
);
1085 static inline void myri10ge_update_dca(struct myri10ge_slice_state
*ss
)
1087 int cpu
= get_cpu();
1090 if (cpu
!= ss
->cpu
) {
1091 tag
= dca_get_tag(cpu
);
1092 if (ss
->cached_dca_tag
!= tag
)
1093 myri10ge_write_dca(ss
, cpu
, tag
);
1098 static void myri10ge_setup_dca(struct myri10ge_priv
*mgp
)
1101 struct pci_dev
*pdev
= mgp
->pdev
;
1103 if (mgp
->ss
[0].dca_tag
== NULL
|| mgp
->dca_enabled
)
1105 if (!myri10ge_dca
) {
1106 dev_err(&pdev
->dev
, "dca disabled by administrator\n");
1109 err
= dca_add_requester(&pdev
->dev
);
1113 "dca_add_requester() failed, err=%d\n", err
);
1116 mgp
->dca_enabled
= 1;
1117 for (i
= 0; i
< mgp
->num_slices
; i
++)
1118 myri10ge_write_dca(&mgp
->ss
[i
], -1, 0);
1121 static void myri10ge_teardown_dca(struct myri10ge_priv
*mgp
)
1123 struct pci_dev
*pdev
= mgp
->pdev
;
1126 if (!mgp
->dca_enabled
)
1128 mgp
->dca_enabled
= 0;
1129 err
= dca_remove_requester(&pdev
->dev
);
1132 static int myri10ge_notify_dca_device(struct device
*dev
, void *data
)
1134 struct myri10ge_priv
*mgp
;
1135 unsigned long event
;
1137 mgp
= dev_get_drvdata(dev
);
1138 event
= *(unsigned long *)data
;
1140 if (event
== DCA_PROVIDER_ADD
)
1141 myri10ge_setup_dca(mgp
);
1142 else if (event
== DCA_PROVIDER_REMOVE
)
1143 myri10ge_teardown_dca(mgp
);
1146 #endif /* CONFIG_MYRI10GE_DCA */
1149 myri10ge_submit_8rx(struct mcp_kreq_ether_recv __iomem
* dst
,
1150 struct mcp_kreq_ether_recv
*src
)
1154 low
= src
->addr_low
;
1155 src
->addr_low
= htonl(DMA_BIT_MASK(32));
1156 myri10ge_pio_copy(dst
, src
, 4 * sizeof(*src
));
1158 myri10ge_pio_copy(dst
+ 4, src
+ 4, 4 * sizeof(*src
));
1160 src
->addr_low
= low
;
1161 put_be32(low
, &dst
->addr_low
);
1165 static inline void myri10ge_vlan_ip_csum(struct sk_buff
*skb
, __wsum hw_csum
)
1167 struct vlan_hdr
*vh
= (struct vlan_hdr
*)(skb
->data
);
1169 if ((skb
->protocol
== htons(ETH_P_8021Q
)) &&
1170 (vh
->h_vlan_encapsulated_proto
== htons(ETH_P_IP
) ||
1171 vh
->h_vlan_encapsulated_proto
== htons(ETH_P_IPV6
))) {
1172 skb
->csum
= hw_csum
;
1173 skb
->ip_summed
= CHECKSUM_COMPLETE
;
1178 myri10ge_rx_skb_build(struct sk_buff
*skb
, u8
* va
,
1179 struct skb_frag_struct
*rx_frags
, int len
, int hlen
)
1181 struct skb_frag_struct
*skb_frags
;
1183 skb
->len
= skb
->data_len
= len
;
1184 skb
->truesize
= len
+ sizeof(struct sk_buff
);
1185 /* attach the page(s) */
1187 skb_frags
= skb_shinfo(skb
)->frags
;
1189 memcpy(skb_frags
, rx_frags
, sizeof(*skb_frags
));
1190 len
-= rx_frags
->size
;
1193 skb_shinfo(skb
)->nr_frags
++;
1196 /* pskb_may_pull is not available in irq context, but
1197 * skb_pull() (for ether_pad and eth_type_trans()) requires
1198 * the beginning of the packet in skb_headlen(), move it
1200 skb_copy_to_linear_data(skb
, va
, hlen
);
1201 skb_shinfo(skb
)->frags
[0].page_offset
+= hlen
;
1202 skb_shinfo(skb
)->frags
[0].size
-= hlen
;
1203 skb
->data_len
-= hlen
;
1205 skb_pull(skb
, MXGEFW_PAD
);
1209 myri10ge_alloc_rx_pages(struct myri10ge_priv
*mgp
, struct myri10ge_rx_buf
*rx
,
1210 int bytes
, int watchdog
)
1214 #if MYRI10GE_ALLOC_SIZE > 4096
1218 if (unlikely(rx
->watchdog_needed
&& !watchdog
))
1221 /* try to refill entire ring */
1222 while (rx
->fill_cnt
!= (rx
->cnt
+ rx
->mask
+ 1)) {
1223 idx
= rx
->fill_cnt
& rx
->mask
;
1224 if (rx
->page_offset
+ bytes
<= MYRI10GE_ALLOC_SIZE
) {
1225 /* we can use part of previous page */
1228 /* we need a new page */
1230 alloc_pages(GFP_ATOMIC
| __GFP_COMP
,
1231 MYRI10GE_ALLOC_ORDER
);
1232 if (unlikely(page
== NULL
)) {
1233 if (rx
->fill_cnt
- rx
->cnt
< 16)
1234 rx
->watchdog_needed
= 1;
1238 rx
->page_offset
= 0;
1239 rx
->bus
= pci_map_page(mgp
->pdev
, page
, 0,
1240 MYRI10GE_ALLOC_SIZE
,
1241 PCI_DMA_FROMDEVICE
);
1243 rx
->info
[idx
].page
= rx
->page
;
1244 rx
->info
[idx
].page_offset
= rx
->page_offset
;
1245 /* note that this is the address of the start of the
1247 dma_unmap_addr_set(&rx
->info
[idx
], bus
, rx
->bus
);
1248 rx
->shadow
[idx
].addr_low
=
1249 htonl(MYRI10GE_LOWPART_TO_U32(rx
->bus
) + rx
->page_offset
);
1250 rx
->shadow
[idx
].addr_high
=
1251 htonl(MYRI10GE_HIGHPART_TO_U32(rx
->bus
));
1253 /* start next packet on a cacheline boundary */
1254 rx
->page_offset
+= SKB_DATA_ALIGN(bytes
);
1256 #if MYRI10GE_ALLOC_SIZE > 4096
1257 /* don't cross a 4KB boundary */
1258 end_offset
= rx
->page_offset
+ bytes
- 1;
1259 if ((unsigned)(rx
->page_offset
^ end_offset
) > 4095)
1260 rx
->page_offset
= end_offset
& ~4095;
1264 /* copy 8 descriptors to the firmware at a time */
1265 if ((idx
& 7) == 7) {
1266 myri10ge_submit_8rx(&rx
->lanai
[idx
- 7],
1267 &rx
->shadow
[idx
- 7]);
1273 myri10ge_unmap_rx_page(struct pci_dev
*pdev
,
1274 struct myri10ge_rx_buffer_state
*info
, int bytes
)
1276 /* unmap the recvd page if we're the only or last user of it */
1277 if (bytes
>= MYRI10GE_ALLOC_SIZE
/ 2 ||
1278 (info
->page_offset
+ 2 * bytes
) > MYRI10GE_ALLOC_SIZE
) {
1279 pci_unmap_page(pdev
, (dma_unmap_addr(info
, bus
)
1280 & ~(MYRI10GE_ALLOC_SIZE
- 1)),
1281 MYRI10GE_ALLOC_SIZE
, PCI_DMA_FROMDEVICE
);
1285 #define MYRI10GE_HLEN 64 /* The number of bytes to copy from a
1286 * page into an skb */
1289 myri10ge_rx_done(struct myri10ge_slice_state
*ss
, struct myri10ge_rx_buf
*rx
,
1290 int bytes
, int len
, __wsum csum
)
1292 struct myri10ge_priv
*mgp
= ss
->mgp
;
1293 struct sk_buff
*skb
;
1294 struct skb_frag_struct rx_frags
[MYRI10GE_MAX_FRAGS_PER_FRAME
];
1295 int i
, idx
, hlen
, remainder
;
1296 struct pci_dev
*pdev
= mgp
->pdev
;
1297 struct net_device
*dev
= mgp
->dev
;
1301 idx
= rx
->cnt
& rx
->mask
;
1302 va
= page_address(rx
->info
[idx
].page
) + rx
->info
[idx
].page_offset
;
1304 /* Fill skb_frag_struct(s) with data from our receive */
1305 for (i
= 0, remainder
= len
; remainder
> 0; i
++) {
1306 myri10ge_unmap_rx_page(pdev
, &rx
->info
[idx
], bytes
);
1307 rx_frags
[i
].page
= rx
->info
[idx
].page
;
1308 rx_frags
[i
].page_offset
= rx
->info
[idx
].page_offset
;
1309 if (remainder
< MYRI10GE_ALLOC_SIZE
)
1310 rx_frags
[i
].size
= remainder
;
1312 rx_frags
[i
].size
= MYRI10GE_ALLOC_SIZE
;
1314 idx
= rx
->cnt
& rx
->mask
;
1315 remainder
-= MYRI10GE_ALLOC_SIZE
;
1318 if (dev
->features
& NETIF_F_LRO
) {
1319 rx_frags
[0].page_offset
+= MXGEFW_PAD
;
1320 rx_frags
[0].size
-= MXGEFW_PAD
;
1322 lro_receive_frags(&ss
->rx_done
.lro_mgr
, rx_frags
,
1323 /* opaque, will come back in get_frag_header */
1325 (void *)(__force
unsigned long)csum
, csum
);
1330 hlen
= MYRI10GE_HLEN
> len
? len
: MYRI10GE_HLEN
;
1332 /* allocate an skb to attach the page(s) to. This is done
1333 * after trying LRO, so as to avoid skb allocation overheads */
1335 skb
= netdev_alloc_skb(dev
, MYRI10GE_HLEN
+ 16);
1336 if (unlikely(skb
== NULL
)) {
1337 ss
->stats
.rx_dropped
++;
1340 put_page(rx_frags
[i
].page
);
1345 /* Attach the pages to the skb, and trim off any padding */
1346 myri10ge_rx_skb_build(skb
, va
, rx_frags
, len
, hlen
);
1347 if (skb_shinfo(skb
)->frags
[0].size
<= 0) {
1348 put_page(skb_shinfo(skb
)->frags
[0].page
);
1349 skb_shinfo(skb
)->nr_frags
= 0;
1351 skb
->protocol
= eth_type_trans(skb
, dev
);
1352 skb_record_rx_queue(skb
, ss
- &mgp
->ss
[0]);
1354 if (mgp
->csum_flag
) {
1355 if ((skb
->protocol
== htons(ETH_P_IP
)) ||
1356 (skb
->protocol
== htons(ETH_P_IPV6
))) {
1358 skb
->ip_summed
= CHECKSUM_COMPLETE
;
1360 myri10ge_vlan_ip_csum(skb
, csum
);
1362 netif_receive_skb(skb
);
1367 myri10ge_tx_done(struct myri10ge_slice_state
*ss
, int mcp_index
)
1369 struct pci_dev
*pdev
= ss
->mgp
->pdev
;
1370 struct myri10ge_tx_buf
*tx
= &ss
->tx
;
1371 struct netdev_queue
*dev_queue
;
1372 struct sk_buff
*skb
;
1375 while (tx
->pkt_done
!= mcp_index
) {
1376 idx
= tx
->done
& tx
->mask
;
1377 skb
= tx
->info
[idx
].skb
;
1380 tx
->info
[idx
].skb
= NULL
;
1381 if (tx
->info
[idx
].last
) {
1383 tx
->info
[idx
].last
= 0;
1386 len
= dma_unmap_len(&tx
->info
[idx
], len
);
1387 dma_unmap_len_set(&tx
->info
[idx
], len
, 0);
1389 ss
->stats
.tx_bytes
+= skb
->len
;
1390 ss
->stats
.tx_packets
++;
1391 dev_kfree_skb_irq(skb
);
1393 pci_unmap_single(pdev
,
1394 dma_unmap_addr(&tx
->info
[idx
],
1399 pci_unmap_page(pdev
,
1400 dma_unmap_addr(&tx
->info
[idx
],
1406 dev_queue
= netdev_get_tx_queue(ss
->dev
, ss
- ss
->mgp
->ss
);
1408 * Make a minimal effort to prevent the NIC from polling an
1409 * idle tx queue. If we can't get the lock we leave the queue
1410 * active. In this case, either a thread was about to start
1411 * using the queue anyway, or we lost a race and the NIC will
1412 * waste some of its resources polling an inactive queue for a
1416 if ((ss
->mgp
->dev
->real_num_tx_queues
> 1) &&
1417 __netif_tx_trylock(dev_queue
)) {
1418 if (tx
->req
== tx
->done
) {
1419 tx
->queue_active
= 0;
1420 put_be32(htonl(1), tx
->send_stop
);
1424 __netif_tx_unlock(dev_queue
);
1427 /* start the queue if we've stopped it */
1428 if (netif_tx_queue_stopped(dev_queue
) &&
1429 tx
->req
- tx
->done
< (tx
->mask
>> 1)) {
1431 netif_tx_wake_queue(dev_queue
);
1436 myri10ge_clean_rx_done(struct myri10ge_slice_state
*ss
, int budget
)
1438 struct myri10ge_rx_done
*rx_done
= &ss
->rx_done
;
1439 struct myri10ge_priv
*mgp
= ss
->mgp
;
1440 struct net_device
*netdev
= mgp
->dev
;
1441 unsigned long rx_bytes
= 0;
1442 unsigned long rx_packets
= 0;
1443 unsigned long rx_ok
;
1445 int idx
= rx_done
->idx
;
1446 int cnt
= rx_done
->cnt
;
1451 while (rx_done
->entry
[idx
].length
!= 0 && work_done
< budget
) {
1452 length
= ntohs(rx_done
->entry
[idx
].length
);
1453 rx_done
->entry
[idx
].length
= 0;
1454 checksum
= csum_unfold(rx_done
->entry
[idx
].checksum
);
1455 if (length
<= mgp
->small_bytes
)
1456 rx_ok
= myri10ge_rx_done(ss
, &ss
->rx_small
,
1460 rx_ok
= myri10ge_rx_done(ss
, &ss
->rx_big
,
1463 rx_packets
+= rx_ok
;
1464 rx_bytes
+= rx_ok
* (unsigned long)length
;
1466 idx
= cnt
& (mgp
->max_intr_slots
- 1);
1471 ss
->stats
.rx_packets
+= rx_packets
;
1472 ss
->stats
.rx_bytes
+= rx_bytes
;
1474 if (netdev
->features
& NETIF_F_LRO
)
1475 lro_flush_all(&rx_done
->lro_mgr
);
1477 /* restock receive rings if needed */
1478 if (ss
->rx_small
.fill_cnt
- ss
->rx_small
.cnt
< myri10ge_fill_thresh
)
1479 myri10ge_alloc_rx_pages(mgp
, &ss
->rx_small
,
1480 mgp
->small_bytes
+ MXGEFW_PAD
, 0);
1481 if (ss
->rx_big
.fill_cnt
- ss
->rx_big
.cnt
< myri10ge_fill_thresh
)
1482 myri10ge_alloc_rx_pages(mgp
, &ss
->rx_big
, mgp
->big_bytes
, 0);
1487 static inline void myri10ge_check_statblock(struct myri10ge_priv
*mgp
)
1489 struct mcp_irq_data
*stats
= mgp
->ss
[0].fw_stats
;
1491 if (unlikely(stats
->stats_updated
)) {
1492 unsigned link_up
= ntohl(stats
->link_up
);
1493 if (mgp
->link_state
!= link_up
) {
1494 mgp
->link_state
= link_up
;
1496 if (mgp
->link_state
== MXGEFW_LINK_UP
) {
1497 if (netif_msg_link(mgp
))
1498 netdev_info(mgp
->dev
, "link up\n");
1499 netif_carrier_on(mgp
->dev
);
1500 mgp
->link_changes
++;
1502 if (netif_msg_link(mgp
))
1503 netdev_info(mgp
->dev
, "link %s\n",
1504 link_up
== MXGEFW_LINK_MYRINET
?
1505 "mismatch (Myrinet detected)" :
1507 netif_carrier_off(mgp
->dev
);
1508 mgp
->link_changes
++;
1511 if (mgp
->rdma_tags_available
!=
1512 ntohl(stats
->rdma_tags_available
)) {
1513 mgp
->rdma_tags_available
=
1514 ntohl(stats
->rdma_tags_available
);
1515 netdev_warn(mgp
->dev
, "RDMA timed out! %d tags left\n",
1516 mgp
->rdma_tags_available
);
1518 mgp
->down_cnt
+= stats
->link_down
;
1519 if (stats
->link_down
)
1520 wake_up(&mgp
->down_wq
);
1524 static int myri10ge_poll(struct napi_struct
*napi
, int budget
)
1526 struct myri10ge_slice_state
*ss
=
1527 container_of(napi
, struct myri10ge_slice_state
, napi
);
1530 #ifdef CONFIG_MYRI10GE_DCA
1531 if (ss
->mgp
->dca_enabled
)
1532 myri10ge_update_dca(ss
);
1535 /* process as many rx events as NAPI will allow */
1536 work_done
= myri10ge_clean_rx_done(ss
, budget
);
1538 if (work_done
< budget
) {
1539 napi_complete(napi
);
1540 put_be32(htonl(3), ss
->irq_claim
);
1545 static irqreturn_t
myri10ge_intr(int irq
, void *arg
)
1547 struct myri10ge_slice_state
*ss
= arg
;
1548 struct myri10ge_priv
*mgp
= ss
->mgp
;
1549 struct mcp_irq_data
*stats
= ss
->fw_stats
;
1550 struct myri10ge_tx_buf
*tx
= &ss
->tx
;
1551 u32 send_done_count
;
1554 /* an interrupt on a non-zero receive-only slice is implicitly
1555 * valid since MSI-X irqs are not shared */
1556 if ((mgp
->dev
->real_num_tx_queues
== 1) && (ss
!= mgp
->ss
)) {
1557 napi_schedule(&ss
->napi
);
1558 return (IRQ_HANDLED
);
1561 /* make sure it is our IRQ, and that the DMA has finished */
1562 if (unlikely(!stats
->valid
))
1565 /* low bit indicates receives are present, so schedule
1566 * napi poll handler */
1567 if (stats
->valid
& 1)
1568 napi_schedule(&ss
->napi
);
1570 if (!mgp
->msi_enabled
&& !mgp
->msix_enabled
) {
1571 put_be32(0, mgp
->irq_deassert
);
1572 if (!myri10ge_deassert_wait
)
1578 /* Wait for IRQ line to go low, if using INTx */
1582 /* check for transmit completes and receives */
1583 send_done_count
= ntohl(stats
->send_done_count
);
1584 if (send_done_count
!= tx
->pkt_done
)
1585 myri10ge_tx_done(ss
, (int)send_done_count
);
1586 if (unlikely(i
> myri10ge_max_irq_loops
)) {
1587 netdev_err(mgp
->dev
, "irq stuck?\n");
1589 schedule_work(&mgp
->watchdog_work
);
1591 if (likely(stats
->valid
== 0))
1597 /* Only slice 0 updates stats */
1599 myri10ge_check_statblock(mgp
);
1601 put_be32(htonl(3), ss
->irq_claim
+ 1);
1602 return (IRQ_HANDLED
);
1606 myri10ge_get_settings(struct net_device
*netdev
, struct ethtool_cmd
*cmd
)
1608 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1612 cmd
->autoneg
= AUTONEG_DISABLE
;
1613 cmd
->speed
= SPEED_10000
;
1614 cmd
->duplex
= DUPLEX_FULL
;
1617 * parse the product code to deterimine the interface type
1618 * (CX4, XFP, Quad Ribbon Fiber) by looking at the character
1619 * after the 3rd dash in the driver's cached copy of the
1620 * EEPROM's product code string.
1622 ptr
= mgp
->product_code_string
;
1624 netdev_err(netdev
, "Missing product code\n");
1627 for (i
= 0; i
< 3; i
++, ptr
++) {
1628 ptr
= strchr(ptr
, '-');
1630 netdev_err(netdev
, "Invalid product code %s\n",
1631 mgp
->product_code_string
);
1637 if (*ptr
== 'R' || *ptr
== 'Q' || *ptr
== 'S') {
1638 /* We've found either an XFP, quad ribbon fiber, or SFP+ */
1639 cmd
->port
= PORT_FIBRE
;
1640 cmd
->supported
|= SUPPORTED_FIBRE
;
1641 cmd
->advertising
|= ADVERTISED_FIBRE
;
1643 cmd
->port
= PORT_OTHER
;
1645 if (*ptr
== 'R' || *ptr
== 'S')
1646 cmd
->transceiver
= XCVR_EXTERNAL
;
1648 cmd
->transceiver
= XCVR_INTERNAL
;
1654 myri10ge_get_drvinfo(struct net_device
*netdev
, struct ethtool_drvinfo
*info
)
1656 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1658 strlcpy(info
->driver
, "myri10ge", sizeof(info
->driver
));
1659 strlcpy(info
->version
, MYRI10GE_VERSION_STR
, sizeof(info
->version
));
1660 strlcpy(info
->fw_version
, mgp
->fw_version
, sizeof(info
->fw_version
));
1661 strlcpy(info
->bus_info
, pci_name(mgp
->pdev
), sizeof(info
->bus_info
));
1665 myri10ge_get_coalesce(struct net_device
*netdev
, struct ethtool_coalesce
*coal
)
1667 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1669 coal
->rx_coalesce_usecs
= mgp
->intr_coal_delay
;
1674 myri10ge_set_coalesce(struct net_device
*netdev
, struct ethtool_coalesce
*coal
)
1676 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1678 mgp
->intr_coal_delay
= coal
->rx_coalesce_usecs
;
1679 put_be32(htonl(mgp
->intr_coal_delay
), mgp
->intr_coal_delay_ptr
);
1684 myri10ge_get_pauseparam(struct net_device
*netdev
,
1685 struct ethtool_pauseparam
*pause
)
1687 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1690 pause
->rx_pause
= mgp
->pause
;
1691 pause
->tx_pause
= mgp
->pause
;
1695 myri10ge_set_pauseparam(struct net_device
*netdev
,
1696 struct ethtool_pauseparam
*pause
)
1698 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1700 if (pause
->tx_pause
!= mgp
->pause
)
1701 return myri10ge_change_pause(mgp
, pause
->tx_pause
);
1702 if (pause
->rx_pause
!= mgp
->pause
)
1703 return myri10ge_change_pause(mgp
, pause
->rx_pause
);
1704 if (pause
->autoneg
!= 0)
1710 myri10ge_get_ringparam(struct net_device
*netdev
,
1711 struct ethtool_ringparam
*ring
)
1713 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1715 ring
->rx_mini_max_pending
= mgp
->ss
[0].rx_small
.mask
+ 1;
1716 ring
->rx_max_pending
= mgp
->ss
[0].rx_big
.mask
+ 1;
1717 ring
->rx_jumbo_max_pending
= 0;
1718 ring
->tx_max_pending
= mgp
->ss
[0].tx
.mask
+ 1;
1719 ring
->rx_mini_pending
= ring
->rx_mini_max_pending
;
1720 ring
->rx_pending
= ring
->rx_max_pending
;
1721 ring
->rx_jumbo_pending
= ring
->rx_jumbo_max_pending
;
1722 ring
->tx_pending
= ring
->tx_max_pending
;
1725 static u32
myri10ge_get_rx_csum(struct net_device
*netdev
)
1727 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1735 static int myri10ge_set_rx_csum(struct net_device
*netdev
, u32 csum_enabled
)
1737 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1741 mgp
->csum_flag
= MXGEFW_FLAGS_CKSUM
;
1743 netdev
->features
&= ~NETIF_F_LRO
;
1750 static int myri10ge_set_tso(struct net_device
*netdev
, u32 tso_enabled
)
1752 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1753 unsigned long flags
= mgp
->features
& (NETIF_F_TSO6
| NETIF_F_TSO
);
1756 netdev
->features
|= flags
;
1758 netdev
->features
&= ~flags
;
1762 static const char myri10ge_gstrings_main_stats
[][ETH_GSTRING_LEN
] = {
1763 "rx_packets", "tx_packets", "rx_bytes", "tx_bytes", "rx_errors",
1764 "tx_errors", "rx_dropped", "tx_dropped", "multicast", "collisions",
1765 "rx_length_errors", "rx_over_errors", "rx_crc_errors",
1766 "rx_frame_errors", "rx_fifo_errors", "rx_missed_errors",
1767 "tx_aborted_errors", "tx_carrier_errors", "tx_fifo_errors",
1768 "tx_heartbeat_errors", "tx_window_errors",
1769 /* device-specific stats */
1770 "tx_boundary", "WC", "irq", "MSI", "MSIX",
1771 "read_dma_bw_MBs", "write_dma_bw_MBs", "read_write_dma_bw_MBs",
1772 "serial_number", "watchdog_resets",
1773 #ifdef CONFIG_MYRI10GE_DCA
1774 "dca_capable_firmware", "dca_device_present",
1776 "link_changes", "link_up", "dropped_link_overflow",
1777 "dropped_link_error_or_filtered",
1778 "dropped_pause", "dropped_bad_phy", "dropped_bad_crc32",
1779 "dropped_unicast_filtered", "dropped_multicast_filtered",
1780 "dropped_runt", "dropped_overrun", "dropped_no_small_buffer",
1781 "dropped_no_big_buffer"
1784 static const char myri10ge_gstrings_slice_stats
[][ETH_GSTRING_LEN
] = {
1785 "----------- slice ---------",
1786 "tx_pkt_start", "tx_pkt_done", "tx_req", "tx_done",
1787 "rx_small_cnt", "rx_big_cnt",
1788 "wake_queue", "stop_queue", "tx_linearized", "LRO aggregated",
1790 "LRO avg aggr", "LRO no_desc"
1793 #define MYRI10GE_NET_STATS_LEN 21
1794 #define MYRI10GE_MAIN_STATS_LEN ARRAY_SIZE(myri10ge_gstrings_main_stats)
1795 #define MYRI10GE_SLICE_STATS_LEN ARRAY_SIZE(myri10ge_gstrings_slice_stats)
1798 myri10ge_get_strings(struct net_device
*netdev
, u32 stringset
, u8
* data
)
1800 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1803 switch (stringset
) {
1805 memcpy(data
, *myri10ge_gstrings_main_stats
,
1806 sizeof(myri10ge_gstrings_main_stats
));
1807 data
+= sizeof(myri10ge_gstrings_main_stats
);
1808 for (i
= 0; i
< mgp
->num_slices
; i
++) {
1809 memcpy(data
, *myri10ge_gstrings_slice_stats
,
1810 sizeof(myri10ge_gstrings_slice_stats
));
1811 data
+= sizeof(myri10ge_gstrings_slice_stats
);
1817 static int myri10ge_get_sset_count(struct net_device
*netdev
, int sset
)
1819 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1823 return MYRI10GE_MAIN_STATS_LEN
+
1824 mgp
->num_slices
* MYRI10GE_SLICE_STATS_LEN
;
1831 myri10ge_get_ethtool_stats(struct net_device
*netdev
,
1832 struct ethtool_stats
*stats
, u64
* data
)
1834 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1835 struct myri10ge_slice_state
*ss
;
1839 /* force stats update */
1840 (void)myri10ge_get_stats(netdev
);
1841 for (i
= 0; i
< MYRI10GE_NET_STATS_LEN
; i
++)
1842 data
[i
] = ((unsigned long *)&netdev
->stats
)[i
];
1844 data
[i
++] = (unsigned int)mgp
->tx_boundary
;
1845 data
[i
++] = (unsigned int)mgp
->wc_enabled
;
1846 data
[i
++] = (unsigned int)mgp
->pdev
->irq
;
1847 data
[i
++] = (unsigned int)mgp
->msi_enabled
;
1848 data
[i
++] = (unsigned int)mgp
->msix_enabled
;
1849 data
[i
++] = (unsigned int)mgp
->read_dma
;
1850 data
[i
++] = (unsigned int)mgp
->write_dma
;
1851 data
[i
++] = (unsigned int)mgp
->read_write_dma
;
1852 data
[i
++] = (unsigned int)mgp
->serial_number
;
1853 data
[i
++] = (unsigned int)mgp
->watchdog_resets
;
1854 #ifdef CONFIG_MYRI10GE_DCA
1855 data
[i
++] = (unsigned int)(mgp
->ss
[0].dca_tag
!= NULL
);
1856 data
[i
++] = (unsigned int)(mgp
->dca_enabled
);
1858 data
[i
++] = (unsigned int)mgp
->link_changes
;
1860 /* firmware stats are useful only in the first slice */
1862 data
[i
++] = (unsigned int)ntohl(ss
->fw_stats
->link_up
);
1863 data
[i
++] = (unsigned int)ntohl(ss
->fw_stats
->dropped_link_overflow
);
1865 (unsigned int)ntohl(ss
->fw_stats
->dropped_link_error_or_filtered
);
1866 data
[i
++] = (unsigned int)ntohl(ss
->fw_stats
->dropped_pause
);
1867 data
[i
++] = (unsigned int)ntohl(ss
->fw_stats
->dropped_bad_phy
);
1868 data
[i
++] = (unsigned int)ntohl(ss
->fw_stats
->dropped_bad_crc32
);
1869 data
[i
++] = (unsigned int)ntohl(ss
->fw_stats
->dropped_unicast_filtered
);
1871 (unsigned int)ntohl(ss
->fw_stats
->dropped_multicast_filtered
);
1872 data
[i
++] = (unsigned int)ntohl(ss
->fw_stats
->dropped_runt
);
1873 data
[i
++] = (unsigned int)ntohl(ss
->fw_stats
->dropped_overrun
);
1874 data
[i
++] = (unsigned int)ntohl(ss
->fw_stats
->dropped_no_small_buffer
);
1875 data
[i
++] = (unsigned int)ntohl(ss
->fw_stats
->dropped_no_big_buffer
);
1877 for (slice
= 0; slice
< mgp
->num_slices
; slice
++) {
1878 ss
= &mgp
->ss
[slice
];
1880 data
[i
++] = (unsigned int)ss
->tx
.pkt_start
;
1881 data
[i
++] = (unsigned int)ss
->tx
.pkt_done
;
1882 data
[i
++] = (unsigned int)ss
->tx
.req
;
1883 data
[i
++] = (unsigned int)ss
->tx
.done
;
1884 data
[i
++] = (unsigned int)ss
->rx_small
.cnt
;
1885 data
[i
++] = (unsigned int)ss
->rx_big
.cnt
;
1886 data
[i
++] = (unsigned int)ss
->tx
.wake_queue
;
1887 data
[i
++] = (unsigned int)ss
->tx
.stop_queue
;
1888 data
[i
++] = (unsigned int)ss
->tx
.linearized
;
1889 data
[i
++] = ss
->rx_done
.lro_mgr
.stats
.aggregated
;
1890 data
[i
++] = ss
->rx_done
.lro_mgr
.stats
.flushed
;
1891 if (ss
->rx_done
.lro_mgr
.stats
.flushed
)
1892 data
[i
++] = ss
->rx_done
.lro_mgr
.stats
.aggregated
/
1893 ss
->rx_done
.lro_mgr
.stats
.flushed
;
1896 data
[i
++] = ss
->rx_done
.lro_mgr
.stats
.no_desc
;
1900 static void myri10ge_set_msglevel(struct net_device
*netdev
, u32 value
)
1902 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1903 mgp
->msg_enable
= value
;
1906 static u32
myri10ge_get_msglevel(struct net_device
*netdev
)
1908 struct myri10ge_priv
*mgp
= netdev_priv(netdev
);
1909 return mgp
->msg_enable
;
1912 static int myri10ge_set_flags(struct net_device
*netdev
, u32 value
)
1914 return ethtool_op_set_flags(netdev
, value
, ETH_FLAG_LRO
);
1917 static const struct ethtool_ops myri10ge_ethtool_ops
= {
1918 .get_settings
= myri10ge_get_settings
,
1919 .get_drvinfo
= myri10ge_get_drvinfo
,
1920 .get_coalesce
= myri10ge_get_coalesce
,
1921 .set_coalesce
= myri10ge_set_coalesce
,
1922 .get_pauseparam
= myri10ge_get_pauseparam
,
1923 .set_pauseparam
= myri10ge_set_pauseparam
,
1924 .get_ringparam
= myri10ge_get_ringparam
,
1925 .get_rx_csum
= myri10ge_get_rx_csum
,
1926 .set_rx_csum
= myri10ge_set_rx_csum
,
1927 .set_tx_csum
= ethtool_op_set_tx_hw_csum
,
1928 .set_sg
= ethtool_op_set_sg
,
1929 .set_tso
= myri10ge_set_tso
,
1930 .get_link
= ethtool_op_get_link
,
1931 .get_strings
= myri10ge_get_strings
,
1932 .get_sset_count
= myri10ge_get_sset_count
,
1933 .get_ethtool_stats
= myri10ge_get_ethtool_stats
,
1934 .set_msglevel
= myri10ge_set_msglevel
,
1935 .get_msglevel
= myri10ge_get_msglevel
,
1936 .get_flags
= ethtool_op_get_flags
,
1937 .set_flags
= myri10ge_set_flags
1940 static int myri10ge_allocate_rings(struct myri10ge_slice_state
*ss
)
1942 struct myri10ge_priv
*mgp
= ss
->mgp
;
1943 struct myri10ge_cmd cmd
;
1944 struct net_device
*dev
= mgp
->dev
;
1945 int tx_ring_size
, rx_ring_size
;
1946 int tx_ring_entries
, rx_ring_entries
;
1947 int i
, slice
, status
;
1950 /* get ring sizes */
1951 slice
= ss
- mgp
->ss
;
1953 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_SEND_RING_SIZE
, &cmd
, 0);
1954 tx_ring_size
= cmd
.data0
;
1956 status
|= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_RX_RING_SIZE
, &cmd
, 0);
1959 rx_ring_size
= cmd
.data0
;
1961 tx_ring_entries
= tx_ring_size
/ sizeof(struct mcp_kreq_ether_send
);
1962 rx_ring_entries
= rx_ring_size
/ sizeof(struct mcp_dma_addr
);
1963 ss
->tx
.mask
= tx_ring_entries
- 1;
1964 ss
->rx_small
.mask
= ss
->rx_big
.mask
= rx_ring_entries
- 1;
1968 /* allocate the host shadow rings */
1970 bytes
= 8 + (MYRI10GE_MAX_SEND_DESC_TSO
+ 4)
1971 * sizeof(*ss
->tx
.req_list
);
1972 ss
->tx
.req_bytes
= kzalloc(bytes
, GFP_KERNEL
);
1973 if (ss
->tx
.req_bytes
== NULL
)
1974 goto abort_with_nothing
;
1976 /* ensure req_list entries are aligned to 8 bytes */
1977 ss
->tx
.req_list
= (struct mcp_kreq_ether_send
*)
1978 ALIGN((unsigned long)ss
->tx
.req_bytes
, 8);
1979 ss
->tx
.queue_active
= 0;
1981 bytes
= rx_ring_entries
* sizeof(*ss
->rx_small
.shadow
);
1982 ss
->rx_small
.shadow
= kzalloc(bytes
, GFP_KERNEL
);
1983 if (ss
->rx_small
.shadow
== NULL
)
1984 goto abort_with_tx_req_bytes
;
1986 bytes
= rx_ring_entries
* sizeof(*ss
->rx_big
.shadow
);
1987 ss
->rx_big
.shadow
= kzalloc(bytes
, GFP_KERNEL
);
1988 if (ss
->rx_big
.shadow
== NULL
)
1989 goto abort_with_rx_small_shadow
;
1991 /* allocate the host info rings */
1993 bytes
= tx_ring_entries
* sizeof(*ss
->tx
.info
);
1994 ss
->tx
.info
= kzalloc(bytes
, GFP_KERNEL
);
1995 if (ss
->tx
.info
== NULL
)
1996 goto abort_with_rx_big_shadow
;
1998 bytes
= rx_ring_entries
* sizeof(*ss
->rx_small
.info
);
1999 ss
->rx_small
.info
= kzalloc(bytes
, GFP_KERNEL
);
2000 if (ss
->rx_small
.info
== NULL
)
2001 goto abort_with_tx_info
;
2003 bytes
= rx_ring_entries
* sizeof(*ss
->rx_big
.info
);
2004 ss
->rx_big
.info
= kzalloc(bytes
, GFP_KERNEL
);
2005 if (ss
->rx_big
.info
== NULL
)
2006 goto abort_with_rx_small_info
;
2008 /* Fill the receive rings */
2010 ss
->rx_small
.cnt
= 0;
2011 ss
->rx_big
.fill_cnt
= 0;
2012 ss
->rx_small
.fill_cnt
= 0;
2013 ss
->rx_small
.page_offset
= MYRI10GE_ALLOC_SIZE
;
2014 ss
->rx_big
.page_offset
= MYRI10GE_ALLOC_SIZE
;
2015 ss
->rx_small
.watchdog_needed
= 0;
2016 ss
->rx_big
.watchdog_needed
= 0;
2017 myri10ge_alloc_rx_pages(mgp
, &ss
->rx_small
,
2018 mgp
->small_bytes
+ MXGEFW_PAD
, 0);
2020 if (ss
->rx_small
.fill_cnt
< ss
->rx_small
.mask
+ 1) {
2021 netdev_err(dev
, "slice-%d: alloced only %d small bufs\n",
2022 slice
, ss
->rx_small
.fill_cnt
);
2023 goto abort_with_rx_small_ring
;
2026 myri10ge_alloc_rx_pages(mgp
, &ss
->rx_big
, mgp
->big_bytes
, 0);
2027 if (ss
->rx_big
.fill_cnt
< ss
->rx_big
.mask
+ 1) {
2028 netdev_err(dev
, "slice-%d: alloced only %d big bufs\n",
2029 slice
, ss
->rx_big
.fill_cnt
);
2030 goto abort_with_rx_big_ring
;
2035 abort_with_rx_big_ring
:
2036 for (i
= ss
->rx_big
.cnt
; i
< ss
->rx_big
.fill_cnt
; i
++) {
2037 int idx
= i
& ss
->rx_big
.mask
;
2038 myri10ge_unmap_rx_page(mgp
->pdev
, &ss
->rx_big
.info
[idx
],
2040 put_page(ss
->rx_big
.info
[idx
].page
);
2043 abort_with_rx_small_ring
:
2044 for (i
= ss
->rx_small
.cnt
; i
< ss
->rx_small
.fill_cnt
; i
++) {
2045 int idx
= i
& ss
->rx_small
.mask
;
2046 myri10ge_unmap_rx_page(mgp
->pdev
, &ss
->rx_small
.info
[idx
],
2047 mgp
->small_bytes
+ MXGEFW_PAD
);
2048 put_page(ss
->rx_small
.info
[idx
].page
);
2051 kfree(ss
->rx_big
.info
);
2053 abort_with_rx_small_info
:
2054 kfree(ss
->rx_small
.info
);
2059 abort_with_rx_big_shadow
:
2060 kfree(ss
->rx_big
.shadow
);
2062 abort_with_rx_small_shadow
:
2063 kfree(ss
->rx_small
.shadow
);
2065 abort_with_tx_req_bytes
:
2066 kfree(ss
->tx
.req_bytes
);
2067 ss
->tx
.req_bytes
= NULL
;
2068 ss
->tx
.req_list
= NULL
;
2074 static void myri10ge_free_rings(struct myri10ge_slice_state
*ss
)
2076 struct myri10ge_priv
*mgp
= ss
->mgp
;
2077 struct sk_buff
*skb
;
2078 struct myri10ge_tx_buf
*tx
;
2081 /* If not allocated, skip it */
2082 if (ss
->tx
.req_list
== NULL
)
2085 for (i
= ss
->rx_big
.cnt
; i
< ss
->rx_big
.fill_cnt
; i
++) {
2086 idx
= i
& ss
->rx_big
.mask
;
2087 if (i
== ss
->rx_big
.fill_cnt
- 1)
2088 ss
->rx_big
.info
[idx
].page_offset
= MYRI10GE_ALLOC_SIZE
;
2089 myri10ge_unmap_rx_page(mgp
->pdev
, &ss
->rx_big
.info
[idx
],
2091 put_page(ss
->rx_big
.info
[idx
].page
);
2094 for (i
= ss
->rx_small
.cnt
; i
< ss
->rx_small
.fill_cnt
; i
++) {
2095 idx
= i
& ss
->rx_small
.mask
;
2096 if (i
== ss
->rx_small
.fill_cnt
- 1)
2097 ss
->rx_small
.info
[idx
].page_offset
=
2098 MYRI10GE_ALLOC_SIZE
;
2099 myri10ge_unmap_rx_page(mgp
->pdev
, &ss
->rx_small
.info
[idx
],
2100 mgp
->small_bytes
+ MXGEFW_PAD
);
2101 put_page(ss
->rx_small
.info
[idx
].page
);
2104 while (tx
->done
!= tx
->req
) {
2105 idx
= tx
->done
& tx
->mask
;
2106 skb
= tx
->info
[idx
].skb
;
2109 tx
->info
[idx
].skb
= NULL
;
2111 len
= dma_unmap_len(&tx
->info
[idx
], len
);
2112 dma_unmap_len_set(&tx
->info
[idx
], len
, 0);
2114 ss
->stats
.tx_dropped
++;
2115 dev_kfree_skb_any(skb
);
2117 pci_unmap_single(mgp
->pdev
,
2118 dma_unmap_addr(&tx
->info
[idx
],
2123 pci_unmap_page(mgp
->pdev
,
2124 dma_unmap_addr(&tx
->info
[idx
],
2129 kfree(ss
->rx_big
.info
);
2131 kfree(ss
->rx_small
.info
);
2135 kfree(ss
->rx_big
.shadow
);
2137 kfree(ss
->rx_small
.shadow
);
2139 kfree(ss
->tx
.req_bytes
);
2140 ss
->tx
.req_bytes
= NULL
;
2141 ss
->tx
.req_list
= NULL
;
2144 static int myri10ge_request_irq(struct myri10ge_priv
*mgp
)
2146 struct pci_dev
*pdev
= mgp
->pdev
;
2147 struct myri10ge_slice_state
*ss
;
2148 struct net_device
*netdev
= mgp
->dev
;
2152 mgp
->msi_enabled
= 0;
2153 mgp
->msix_enabled
= 0;
2156 if (mgp
->num_slices
> 1) {
2158 pci_enable_msix(pdev
, mgp
->msix_vectors
,
2161 mgp
->msix_enabled
= 1;
2164 "Error %d setting up MSI-X\n", status
);
2168 if (mgp
->msix_enabled
== 0) {
2169 status
= pci_enable_msi(pdev
);
2172 "Error %d setting up MSI; falling back to xPIC\n",
2175 mgp
->msi_enabled
= 1;
2179 if (mgp
->msix_enabled
) {
2180 for (i
= 0; i
< mgp
->num_slices
; i
++) {
2182 snprintf(ss
->irq_desc
, sizeof(ss
->irq_desc
),
2183 "%s:slice-%d", netdev
->name
, i
);
2184 status
= request_irq(mgp
->msix_vectors
[i
].vector
,
2185 myri10ge_intr
, 0, ss
->irq_desc
,
2189 "slice %d failed to allocate IRQ\n", i
);
2192 free_irq(mgp
->msix_vectors
[i
].vector
,
2196 pci_disable_msix(pdev
);
2201 status
= request_irq(pdev
->irq
, myri10ge_intr
, IRQF_SHARED
,
2202 mgp
->dev
->name
, &mgp
->ss
[0]);
2204 dev_err(&pdev
->dev
, "failed to allocate IRQ\n");
2205 if (mgp
->msi_enabled
)
2206 pci_disable_msi(pdev
);
2212 static void myri10ge_free_irq(struct myri10ge_priv
*mgp
)
2214 struct pci_dev
*pdev
= mgp
->pdev
;
2217 if (mgp
->msix_enabled
) {
2218 for (i
= 0; i
< mgp
->num_slices
; i
++)
2219 free_irq(mgp
->msix_vectors
[i
].vector
, &mgp
->ss
[i
]);
2221 free_irq(pdev
->irq
, &mgp
->ss
[0]);
2223 if (mgp
->msi_enabled
)
2224 pci_disable_msi(pdev
);
2225 if (mgp
->msix_enabled
)
2226 pci_disable_msix(pdev
);
2230 myri10ge_get_frag_header(struct skb_frag_struct
*frag
, void **mac_hdr
,
2231 void **ip_hdr
, void **tcpudp_hdr
,
2232 u64
* hdr_flags
, void *priv
)
2235 struct vlan_ethhdr
*veh
;
2237 u8
*va
= page_address(frag
->page
) + frag
->page_offset
;
2238 unsigned long ll_hlen
;
2239 /* passed opaque through lro_receive_frags() */
2240 __wsum csum
= (__force __wsum
) (unsigned long)priv
;
2242 /* find the mac header, aborting if not IPv4 */
2244 eh
= (struct ethhdr
*)va
;
2247 if (eh
->h_proto
!= htons(ETH_P_IP
)) {
2248 if (eh
->h_proto
== htons(ETH_P_8021Q
)) {
2249 veh
= (struct vlan_ethhdr
*)va
;
2250 if (veh
->h_vlan_encapsulated_proto
!= htons(ETH_P_IP
))
2253 ll_hlen
+= VLAN_HLEN
;
2256 * HW checksum starts ETH_HLEN bytes into
2257 * frame, so we must subtract off the VLAN
2258 * header's checksum before csum can be used
2260 csum
= csum_sub(csum
, csum_partial(va
+ ETH_HLEN
,
2266 *hdr_flags
= LRO_IPV4
;
2268 iph
= (struct iphdr
*)(va
+ ll_hlen
);
2270 if (iph
->protocol
!= IPPROTO_TCP
)
2272 if (iph
->frag_off
& htons(IP_MF
| IP_OFFSET
))
2274 *hdr_flags
|= LRO_TCP
;
2275 *tcpudp_hdr
= (u8
*) (*ip_hdr
) + (iph
->ihl
<< 2);
2277 /* verify the IP checksum */
2278 if (unlikely(ip_fast_csum((u8
*) iph
, iph
->ihl
)))
2281 /* verify the checksum */
2282 if (unlikely(csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
2283 ntohs(iph
->tot_len
) - (iph
->ihl
<< 2),
2284 IPPROTO_TCP
, csum
)))
2290 static int myri10ge_get_txrx(struct myri10ge_priv
*mgp
, int slice
)
2292 struct myri10ge_cmd cmd
;
2293 struct myri10ge_slice_state
*ss
;
2296 ss
= &mgp
->ss
[slice
];
2298 if (slice
== 0 || (mgp
->dev
->real_num_tx_queues
> 1)) {
2300 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_SEND_OFFSET
,
2302 ss
->tx
.lanai
= (struct mcp_kreq_ether_send __iomem
*)
2303 (mgp
->sram
+ cmd
.data0
);
2306 status
|= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_SMALL_RX_OFFSET
,
2308 ss
->rx_small
.lanai
= (struct mcp_kreq_ether_recv __iomem
*)
2309 (mgp
->sram
+ cmd
.data0
);
2312 status
|= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_BIG_RX_OFFSET
, &cmd
, 0);
2313 ss
->rx_big
.lanai
= (struct mcp_kreq_ether_recv __iomem
*)
2314 (mgp
->sram
+ cmd
.data0
);
2316 ss
->tx
.send_go
= (__iomem __be32
*)
2317 (mgp
->sram
+ MXGEFW_ETH_SEND_GO
+ 64 * slice
);
2318 ss
->tx
.send_stop
= (__iomem __be32
*)
2319 (mgp
->sram
+ MXGEFW_ETH_SEND_STOP
+ 64 * slice
);
2324 static int myri10ge_set_stats(struct myri10ge_priv
*mgp
, int slice
)
2326 struct myri10ge_cmd cmd
;
2327 struct myri10ge_slice_state
*ss
;
2330 ss
= &mgp
->ss
[slice
];
2331 cmd
.data0
= MYRI10GE_LOWPART_TO_U32(ss
->fw_stats_bus
);
2332 cmd
.data1
= MYRI10GE_HIGHPART_TO_U32(ss
->fw_stats_bus
);
2333 cmd
.data2
= sizeof(struct mcp_irq_data
) | (slice
<< 16);
2334 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_STATS_DMA_V2
, &cmd
, 0);
2335 if (status
== -ENOSYS
) {
2336 dma_addr_t bus
= ss
->fw_stats_bus
;
2339 bus
+= offsetof(struct mcp_irq_data
, send_done_count
);
2340 cmd
.data0
= MYRI10GE_LOWPART_TO_U32(bus
);
2341 cmd
.data1
= MYRI10GE_HIGHPART_TO_U32(bus
);
2342 status
= myri10ge_send_cmd(mgp
,
2343 MXGEFW_CMD_SET_STATS_DMA_OBSOLETE
,
2345 /* Firmware cannot support multicast without STATS_DMA_V2 */
2346 mgp
->fw_multicast_support
= 0;
2348 mgp
->fw_multicast_support
= 1;
2353 static int myri10ge_open(struct net_device
*dev
)
2355 struct myri10ge_slice_state
*ss
;
2356 struct myri10ge_priv
*mgp
= netdev_priv(dev
);
2357 struct myri10ge_cmd cmd
;
2358 int i
, status
, big_pow2
, slice
;
2360 struct net_lro_mgr
*lro_mgr
;
2362 if (mgp
->running
!= MYRI10GE_ETH_STOPPED
)
2365 mgp
->running
= MYRI10GE_ETH_STARTING
;
2366 status
= myri10ge_reset(mgp
);
2368 netdev_err(dev
, "failed reset\n");
2369 goto abort_with_nothing
;
2372 if (mgp
->num_slices
> 1) {
2373 cmd
.data0
= mgp
->num_slices
;
2374 cmd
.data1
= MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE
;
2375 if (mgp
->dev
->real_num_tx_queues
> 1)
2376 cmd
.data1
|= MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES
;
2377 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_ENABLE_RSS_QUEUES
,
2380 netdev_err(dev
, "failed to set number of slices\n");
2381 goto abort_with_nothing
;
2383 /* setup the indirection table */
2384 cmd
.data0
= mgp
->num_slices
;
2385 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_RSS_TABLE_SIZE
,
2388 status
|= myri10ge_send_cmd(mgp
,
2389 MXGEFW_CMD_GET_RSS_TABLE_OFFSET
,
2392 netdev_err(dev
, "failed to setup rss tables\n");
2393 goto abort_with_nothing
;
2396 /* just enable an identity mapping */
2397 itable
= mgp
->sram
+ cmd
.data0
;
2398 for (i
= 0; i
< mgp
->num_slices
; i
++)
2399 __raw_writeb(i
, &itable
[i
]);
2402 cmd
.data1
= myri10ge_rss_hash
;
2403 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_RSS_ENABLE
,
2406 netdev_err(dev
, "failed to enable slices\n");
2407 goto abort_with_nothing
;
2411 status
= myri10ge_request_irq(mgp
);
2413 goto abort_with_nothing
;
2415 /* decide what small buffer size to use. For good TCP rx
2416 * performance, it is important to not receive 1514 byte
2417 * frames into jumbo buffers, as it confuses the socket buffer
2418 * accounting code, leading to drops and erratic performance.
2421 if (dev
->mtu
<= ETH_DATA_LEN
)
2422 /* enough for a TCP header */
2423 mgp
->small_bytes
= (128 > SMP_CACHE_BYTES
)
2424 ? (128 - MXGEFW_PAD
)
2425 : (SMP_CACHE_BYTES
- MXGEFW_PAD
);
2427 /* enough for a vlan encapsulated ETH_DATA_LEN frame */
2428 mgp
->small_bytes
= VLAN_ETH_FRAME_LEN
;
2430 /* Override the small buffer size? */
2431 if (myri10ge_small_bytes
> 0)
2432 mgp
->small_bytes
= myri10ge_small_bytes
;
2434 /* Firmware needs the big buff size as a power of 2. Lie and
2435 * tell him the buffer is larger, because we only use 1
2436 * buffer/pkt, and the mtu will prevent overruns.
2438 big_pow2
= dev
->mtu
+ ETH_HLEN
+ VLAN_HLEN
+ MXGEFW_PAD
;
2439 if (big_pow2
< MYRI10GE_ALLOC_SIZE
/ 2) {
2440 while (!is_power_of_2(big_pow2
))
2442 mgp
->big_bytes
= dev
->mtu
+ ETH_HLEN
+ VLAN_HLEN
+ MXGEFW_PAD
;
2444 big_pow2
= MYRI10GE_ALLOC_SIZE
;
2445 mgp
->big_bytes
= big_pow2
;
2448 /* setup the per-slice data structures */
2449 for (slice
= 0; slice
< mgp
->num_slices
; slice
++) {
2450 ss
= &mgp
->ss
[slice
];
2452 status
= myri10ge_get_txrx(mgp
, slice
);
2454 netdev_err(dev
, "failed to get ring sizes or locations\n");
2455 goto abort_with_rings
;
2457 status
= myri10ge_allocate_rings(ss
);
2459 goto abort_with_rings
;
2461 /* only firmware which supports multiple TX queues
2462 * supports setting up the tx stats on non-zero
2464 if (slice
== 0 || mgp
->dev
->real_num_tx_queues
> 1)
2465 status
= myri10ge_set_stats(mgp
, slice
);
2467 netdev_err(dev
, "Couldn't set stats DMA\n");
2468 goto abort_with_rings
;
2471 lro_mgr
= &ss
->rx_done
.lro_mgr
;
2473 lro_mgr
->features
= LRO_F_NAPI
;
2474 lro_mgr
->ip_summed
= CHECKSUM_COMPLETE
;
2475 lro_mgr
->ip_summed_aggr
= CHECKSUM_UNNECESSARY
;
2476 lro_mgr
->max_desc
= MYRI10GE_MAX_LRO_DESCRIPTORS
;
2477 lro_mgr
->lro_arr
= ss
->rx_done
.lro_desc
;
2478 lro_mgr
->get_frag_header
= myri10ge_get_frag_header
;
2479 lro_mgr
->max_aggr
= myri10ge_lro_max_pkts
;
2480 lro_mgr
->frag_align_pad
= 2;
2481 if (lro_mgr
->max_aggr
> MAX_SKB_FRAGS
)
2482 lro_mgr
->max_aggr
= MAX_SKB_FRAGS
;
2484 /* must happen prior to any irq */
2485 napi_enable(&(ss
)->napi
);
2488 /* now give firmware buffers sizes, and MTU */
2489 cmd
.data0
= dev
->mtu
+ ETH_HLEN
+ VLAN_HLEN
;
2490 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_MTU
, &cmd
, 0);
2491 cmd
.data0
= mgp
->small_bytes
;
2493 myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_SMALL_BUFFER_SIZE
, &cmd
, 0);
2494 cmd
.data0
= big_pow2
;
2496 myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_BIG_BUFFER_SIZE
, &cmd
, 0);
2498 netdev_err(dev
, "Couldn't set buffer sizes\n");
2499 goto abort_with_rings
;
2503 * Set Linux style TSO mode; this is needed only on newer
2504 * firmware versions. Older versions default to Linux
2508 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_TSO_MODE
, &cmd
, 0);
2509 if (status
&& status
!= -ENOSYS
) {
2510 netdev_err(dev
, "Couldn't set TSO mode\n");
2511 goto abort_with_rings
;
2514 mgp
->link_state
= ~0U;
2515 mgp
->rdma_tags_available
= 15;
2517 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_ETHERNET_UP
, &cmd
, 0);
2519 netdev_err(dev
, "Couldn't bring up link\n");
2520 goto abort_with_rings
;
2523 mgp
->running
= MYRI10GE_ETH_RUNNING
;
2524 mgp
->watchdog_timer
.expires
= jiffies
+ myri10ge_watchdog_timeout
* HZ
;
2525 add_timer(&mgp
->watchdog_timer
);
2526 netif_tx_wake_all_queues(dev
);
2533 napi_disable(&mgp
->ss
[slice
].napi
);
2535 for (i
= 0; i
< mgp
->num_slices
; i
++)
2536 myri10ge_free_rings(&mgp
->ss
[i
]);
2538 myri10ge_free_irq(mgp
);
2541 mgp
->running
= MYRI10GE_ETH_STOPPED
;
2545 static int myri10ge_close(struct net_device
*dev
)
2547 struct myri10ge_priv
*mgp
= netdev_priv(dev
);
2548 struct myri10ge_cmd cmd
;
2549 int status
, old_down_cnt
;
2552 if (mgp
->running
!= MYRI10GE_ETH_RUNNING
)
2555 if (mgp
->ss
[0].tx
.req_bytes
== NULL
)
2558 del_timer_sync(&mgp
->watchdog_timer
);
2559 mgp
->running
= MYRI10GE_ETH_STOPPING
;
2560 for (i
= 0; i
< mgp
->num_slices
; i
++) {
2561 napi_disable(&mgp
->ss
[i
].napi
);
2563 netif_carrier_off(dev
);
2565 netif_tx_stop_all_queues(dev
);
2566 if (mgp
->rebooted
== 0) {
2567 old_down_cnt
= mgp
->down_cnt
;
2570 myri10ge_send_cmd(mgp
, MXGEFW_CMD_ETHERNET_DOWN
, &cmd
, 0);
2572 netdev_err(dev
, "Couldn't bring down link\n");
2574 wait_event_timeout(mgp
->down_wq
, old_down_cnt
!= mgp
->down_cnt
,
2576 if (old_down_cnt
== mgp
->down_cnt
)
2577 netdev_err(dev
, "never got down irq\n");
2579 netif_tx_disable(dev
);
2580 myri10ge_free_irq(mgp
);
2581 for (i
= 0; i
< mgp
->num_slices
; i
++)
2582 myri10ge_free_rings(&mgp
->ss
[i
]);
2584 mgp
->running
= MYRI10GE_ETH_STOPPED
;
2588 /* copy an array of struct mcp_kreq_ether_send's to the mcp. Copy
2589 * backwards one at a time and handle ring wraps */
2592 myri10ge_submit_req_backwards(struct myri10ge_tx_buf
*tx
,
2593 struct mcp_kreq_ether_send
*src
, int cnt
)
2595 int idx
, starting_slot
;
2596 starting_slot
= tx
->req
;
2599 idx
= (starting_slot
+ cnt
) & tx
->mask
;
2600 myri10ge_pio_copy(&tx
->lanai
[idx
], &src
[cnt
], sizeof(*src
));
2606 * copy an array of struct mcp_kreq_ether_send's to the mcp. Copy
2607 * at most 32 bytes at a time, so as to avoid involving the software
2608 * pio handler in the nic. We re-write the first segment's flags
2609 * to mark them valid only after writing the entire chain.
2613 myri10ge_submit_req(struct myri10ge_tx_buf
*tx
, struct mcp_kreq_ether_send
*src
,
2617 struct mcp_kreq_ether_send __iomem
*dstp
, *dst
;
2618 struct mcp_kreq_ether_send
*srcp
;
2621 idx
= tx
->req
& tx
->mask
;
2623 last_flags
= src
->flags
;
2626 dst
= dstp
= &tx
->lanai
[idx
];
2629 if ((idx
+ cnt
) < tx
->mask
) {
2630 for (i
= 0; i
< (cnt
- 1); i
+= 2) {
2631 myri10ge_pio_copy(dstp
, srcp
, 2 * sizeof(*src
));
2632 mb(); /* force write every 32 bytes */
2637 /* submit all but the first request, and ensure
2638 * that it is submitted below */
2639 myri10ge_submit_req_backwards(tx
, src
, cnt
);
2643 /* submit the first request */
2644 myri10ge_pio_copy(dstp
, srcp
, sizeof(*src
));
2645 mb(); /* barrier before setting valid flag */
2648 /* re-write the last 32-bits with the valid flags */
2649 src
->flags
= last_flags
;
2650 put_be32(*((__be32
*) src
+ 3), (__be32 __iomem
*) dst
+ 3);
2656 * Transmit a packet. We need to split the packet so that a single
2657 * segment does not cross myri10ge->tx_boundary, so this makes segment
2658 * counting tricky. So rather than try to count segments up front, we
2659 * just give up if there are too few segments to hold a reasonably
2660 * fragmented packet currently available. If we run
2661 * out of segments while preparing a packet for DMA, we just linearize
2665 static netdev_tx_t
myri10ge_xmit(struct sk_buff
*skb
,
2666 struct net_device
*dev
)
2668 struct myri10ge_priv
*mgp
= netdev_priv(dev
);
2669 struct myri10ge_slice_state
*ss
;
2670 struct mcp_kreq_ether_send
*req
;
2671 struct myri10ge_tx_buf
*tx
;
2672 struct skb_frag_struct
*frag
;
2673 struct netdev_queue
*netdev_queue
;
2676 __be32 high_swapped
;
2678 int idx
, last_idx
, avail
, frag_cnt
, frag_idx
, count
, mss
, max_segments
;
2679 u16 pseudo_hdr_offset
, cksum_offset
, queue
;
2680 int cum_len
, seglen
, boundary
, rdma_count
;
2683 queue
= skb_get_queue_mapping(skb
);
2684 ss
= &mgp
->ss
[queue
];
2685 netdev_queue
= netdev_get_tx_queue(mgp
->dev
, queue
);
2690 avail
= tx
->mask
- 1 - (tx
->req
- tx
->done
);
2693 max_segments
= MXGEFW_MAX_SEND_DESC
;
2695 if (skb_is_gso(skb
)) {
2696 mss
= skb_shinfo(skb
)->gso_size
;
2697 max_segments
= MYRI10GE_MAX_SEND_DESC_TSO
;
2700 if ((unlikely(avail
< max_segments
))) {
2701 /* we are out of transmit resources */
2703 netif_tx_stop_queue(netdev_queue
);
2704 return NETDEV_TX_BUSY
;
2707 /* Setup checksum offloading, if needed */
2709 pseudo_hdr_offset
= 0;
2711 flags
= (MXGEFW_FLAGS_NO_TSO
| MXGEFW_FLAGS_FIRST
);
2712 if (likely(skb
->ip_summed
== CHECKSUM_PARTIAL
)) {
2713 cksum_offset
= skb_transport_offset(skb
);
2714 pseudo_hdr_offset
= cksum_offset
+ skb
->csum_offset
;
2715 /* If the headers are excessively large, then we must
2716 * fall back to a software checksum */
2717 if (unlikely(!mss
&& (cksum_offset
> 255 ||
2718 pseudo_hdr_offset
> 127))) {
2719 if (skb_checksum_help(skb
))
2722 pseudo_hdr_offset
= 0;
2724 odd_flag
= MXGEFW_FLAGS_ALIGN_ODD
;
2725 flags
|= MXGEFW_FLAGS_CKSUM
;
2731 if (mss
) { /* TSO */
2732 /* this removes any CKSUM flag from before */
2733 flags
= (MXGEFW_FLAGS_TSO_HDR
| MXGEFW_FLAGS_FIRST
);
2735 /* negative cum_len signifies to the
2736 * send loop that we are still in the
2737 * header portion of the TSO packet.
2738 * TSO header can be at most 1KB long */
2739 cum_len
= -(skb_transport_offset(skb
) + tcp_hdrlen(skb
));
2741 /* for IPv6 TSO, the checksum offset stores the
2742 * TCP header length, to save the firmware from
2743 * the need to parse the headers */
2744 if (skb_is_gso_v6(skb
)) {
2745 cksum_offset
= tcp_hdrlen(skb
);
2746 /* Can only handle headers <= max_tso6 long */
2747 if (unlikely(-cum_len
> mgp
->max_tso6
))
2748 return myri10ge_sw_tso(skb
, dev
);
2750 /* for TSO, pseudo_hdr_offset holds mss.
2751 * The firmware figures out where to put
2752 * the checksum by parsing the header. */
2753 pseudo_hdr_offset
= mss
;
2755 /* Mark small packets, and pad out tiny packets */
2756 if (skb
->len
<= MXGEFW_SEND_SMALL_SIZE
) {
2757 flags
|= MXGEFW_FLAGS_SMALL
;
2759 /* pad frames to at least ETH_ZLEN bytes */
2760 if (unlikely(skb
->len
< ETH_ZLEN
)) {
2761 if (skb_padto(skb
, ETH_ZLEN
)) {
2762 /* The packet is gone, so we must
2764 ss
->stats
.tx_dropped
+= 1;
2765 return NETDEV_TX_OK
;
2767 /* adjust the len to account for the zero pad
2768 * so that the nic can know how long it is */
2769 skb
->len
= ETH_ZLEN
;
2773 /* map the skb for DMA */
2774 len
= skb_headlen(skb
);
2775 idx
= tx
->req
& tx
->mask
;
2776 tx
->info
[idx
].skb
= skb
;
2777 bus
= pci_map_single(mgp
->pdev
, skb
->data
, len
, PCI_DMA_TODEVICE
);
2778 dma_unmap_addr_set(&tx
->info
[idx
], bus
, bus
);
2779 dma_unmap_len_set(&tx
->info
[idx
], len
, len
);
2781 frag_cnt
= skb_shinfo(skb
)->nr_frags
;
2786 /* "rdma_count" is the number of RDMAs belonging to the
2787 * current packet BEFORE the current send request. For
2788 * non-TSO packets, this is equal to "count".
2789 * For TSO packets, rdma_count needs to be reset
2790 * to 0 after a segment cut.
2792 * The rdma_count field of the send request is
2793 * the number of RDMAs of the packet starting at
2794 * that request. For TSO send requests with one ore more cuts
2795 * in the middle, this is the number of RDMAs starting
2796 * after the last cut in the request. All previous
2797 * segments before the last cut implicitly have 1 RDMA.
2799 * Since the number of RDMAs is not known beforehand,
2800 * it must be filled-in retroactively - after each
2801 * segmentation cut or at the end of the entire packet.
2805 /* Break the SKB or Fragment up into pieces which
2806 * do not cross mgp->tx_boundary */
2807 low
= MYRI10GE_LOWPART_TO_U32(bus
);
2808 high_swapped
= htonl(MYRI10GE_HIGHPART_TO_U32(bus
));
2813 if (unlikely(count
== max_segments
))
2814 goto abort_linearize
;
2817 (low
+ mgp
->tx_boundary
) & ~(mgp
->tx_boundary
- 1);
2818 seglen
= boundary
- low
;
2821 flags_next
= flags
& ~MXGEFW_FLAGS_FIRST
;
2822 cum_len_next
= cum_len
+ seglen
;
2823 if (mss
) { /* TSO */
2824 (req
- rdma_count
)->rdma_count
= rdma_count
+ 1;
2826 if (likely(cum_len
>= 0)) { /* payload */
2827 int next_is_first
, chop
;
2829 chop
= (cum_len_next
> mss
);
2830 cum_len_next
= cum_len_next
% mss
;
2831 next_is_first
= (cum_len_next
== 0);
2832 flags
|= chop
* MXGEFW_FLAGS_TSO_CHOP
;
2833 flags_next
|= next_is_first
*
2835 rdma_count
|= -(chop
| next_is_first
);
2836 rdma_count
+= chop
& !next_is_first
;
2837 } else if (likely(cum_len_next
>= 0)) { /* header ends */
2843 small
= (mss
<= MXGEFW_SEND_SMALL_SIZE
);
2844 flags_next
= MXGEFW_FLAGS_TSO_PLD
|
2845 MXGEFW_FLAGS_FIRST
|
2846 (small
* MXGEFW_FLAGS_SMALL
);
2849 req
->addr_high
= high_swapped
;
2850 req
->addr_low
= htonl(low
);
2851 req
->pseudo_hdr_offset
= htons(pseudo_hdr_offset
);
2852 req
->pad
= 0; /* complete solid 16-byte block; does this matter? */
2853 req
->rdma_count
= 1;
2854 req
->length
= htons(seglen
);
2855 req
->cksum_offset
= cksum_offset
;
2856 req
->flags
= flags
| ((cum_len
& 1) * odd_flag
);
2860 cum_len
= cum_len_next
;
2865 if (cksum_offset
!= 0 && !(mss
&& skb_is_gso_v6(skb
))) {
2866 if (unlikely(cksum_offset
> seglen
))
2867 cksum_offset
-= seglen
;
2872 if (frag_idx
== frag_cnt
)
2875 /* map next fragment for DMA */
2876 idx
= (count
+ tx
->req
) & tx
->mask
;
2877 frag
= &skb_shinfo(skb
)->frags
[frag_idx
];
2880 bus
= pci_map_page(mgp
->pdev
, frag
->page
, frag
->page_offset
,
2881 len
, PCI_DMA_TODEVICE
);
2882 dma_unmap_addr_set(&tx
->info
[idx
], bus
, bus
);
2883 dma_unmap_len_set(&tx
->info
[idx
], len
, len
);
2886 (req
- rdma_count
)->rdma_count
= rdma_count
;
2890 req
->flags
|= MXGEFW_FLAGS_TSO_LAST
;
2891 } while (!(req
->flags
& (MXGEFW_FLAGS_TSO_CHOP
|
2892 MXGEFW_FLAGS_FIRST
)));
2893 idx
= ((count
- 1) + tx
->req
) & tx
->mask
;
2894 tx
->info
[idx
].last
= 1;
2895 myri10ge_submit_req(tx
, tx
->req_list
, count
);
2896 /* if using multiple tx queues, make sure NIC polls the
2898 if ((mgp
->dev
->real_num_tx_queues
> 1) && tx
->queue_active
== 0) {
2899 tx
->queue_active
= 1;
2900 put_be32(htonl(1), tx
->send_go
);
2905 if ((avail
- count
) < MXGEFW_MAX_SEND_DESC
) {
2907 netif_tx_stop_queue(netdev_queue
);
2909 return NETDEV_TX_OK
;
2912 /* Free any DMA resources we've alloced and clear out the skb
2913 * slot so as to not trip up assertions, and to avoid a
2914 * double-free if linearizing fails */
2916 last_idx
= (idx
+ 1) & tx
->mask
;
2917 idx
= tx
->req
& tx
->mask
;
2918 tx
->info
[idx
].skb
= NULL
;
2920 len
= dma_unmap_len(&tx
->info
[idx
], len
);
2922 if (tx
->info
[idx
].skb
!= NULL
)
2923 pci_unmap_single(mgp
->pdev
,
2924 dma_unmap_addr(&tx
->info
[idx
],
2928 pci_unmap_page(mgp
->pdev
,
2929 dma_unmap_addr(&tx
->info
[idx
],
2932 dma_unmap_len_set(&tx
->info
[idx
], len
, 0);
2933 tx
->info
[idx
].skb
= NULL
;
2935 idx
= (idx
+ 1) & tx
->mask
;
2936 } while (idx
!= last_idx
);
2937 if (skb_is_gso(skb
)) {
2938 netdev_err(mgp
->dev
, "TSO but wanted to linearize?!?!?\n");
2942 if (skb_linearize(skb
))
2949 dev_kfree_skb_any(skb
);
2950 ss
->stats
.tx_dropped
+= 1;
2951 return NETDEV_TX_OK
;
2955 static netdev_tx_t
myri10ge_sw_tso(struct sk_buff
*skb
,
2956 struct net_device
*dev
)
2958 struct sk_buff
*segs
, *curr
;
2959 struct myri10ge_priv
*mgp
= netdev_priv(dev
);
2960 struct myri10ge_slice_state
*ss
;
2963 segs
= skb_gso_segment(skb
, dev
->features
& ~NETIF_F_TSO6
);
2971 status
= myri10ge_xmit(curr
, dev
);
2973 dev_kfree_skb_any(curr
);
2978 dev_kfree_skb_any(segs
);
2983 dev_kfree_skb_any(skb
);
2984 return NETDEV_TX_OK
;
2987 ss
= &mgp
->ss
[skb_get_queue_mapping(skb
)];
2988 dev_kfree_skb_any(skb
);
2989 ss
->stats
.tx_dropped
+= 1;
2990 return NETDEV_TX_OK
;
2993 static struct net_device_stats
*myri10ge_get_stats(struct net_device
*dev
)
2995 struct myri10ge_priv
*mgp
= netdev_priv(dev
);
2996 struct myri10ge_slice_netstats
*slice_stats
;
2997 struct net_device_stats
*stats
= &dev
->stats
;
3000 spin_lock(&mgp
->stats_lock
);
3001 memset(stats
, 0, sizeof(*stats
));
3002 for (i
= 0; i
< mgp
->num_slices
; i
++) {
3003 slice_stats
= &mgp
->ss
[i
].stats
;
3004 stats
->rx_packets
+= slice_stats
->rx_packets
;
3005 stats
->tx_packets
+= slice_stats
->tx_packets
;
3006 stats
->rx_bytes
+= slice_stats
->rx_bytes
;
3007 stats
->tx_bytes
+= slice_stats
->tx_bytes
;
3008 stats
->rx_dropped
+= slice_stats
->rx_dropped
;
3009 stats
->tx_dropped
+= slice_stats
->tx_dropped
;
3011 spin_unlock(&mgp
->stats_lock
);
3015 static void myri10ge_set_multicast_list(struct net_device
*dev
)
3017 struct myri10ge_priv
*mgp
= netdev_priv(dev
);
3018 struct myri10ge_cmd cmd
;
3019 struct netdev_hw_addr
*ha
;
3020 __be32 data
[2] = { 0, 0 };
3023 /* can be called from atomic contexts,
3024 * pass 1 to force atomicity in myri10ge_send_cmd() */
3025 myri10ge_change_promisc(mgp
, dev
->flags
& IFF_PROMISC
, 1);
3027 /* This firmware is known to not support multicast */
3028 if (!mgp
->fw_multicast_support
)
3031 /* Disable multicast filtering */
3033 err
= myri10ge_send_cmd(mgp
, MXGEFW_ENABLE_ALLMULTI
, &cmd
, 1);
3035 netdev_err(dev
, "Failed MXGEFW_ENABLE_ALLMULTI, error status: %d\n",
3040 if ((dev
->flags
& IFF_ALLMULTI
) || mgp
->adopted_rx_filter_bug
) {
3041 /* request to disable multicast filtering, so quit here */
3045 /* Flush the filters */
3047 err
= myri10ge_send_cmd(mgp
, MXGEFW_LEAVE_ALL_MULTICAST_GROUPS
,
3050 netdev_err(dev
, "Failed MXGEFW_LEAVE_ALL_MULTICAST_GROUPS, error status: %d\n",
3055 /* Walk the multicast list, and add each address */
3056 netdev_for_each_mc_addr(ha
, dev
) {
3057 memcpy(data
, &ha
->addr
, 6);
3058 cmd
.data0
= ntohl(data
[0]);
3059 cmd
.data1
= ntohl(data
[1]);
3060 err
= myri10ge_send_cmd(mgp
, MXGEFW_JOIN_MULTICAST_GROUP
,
3064 netdev_err(dev
, "Failed MXGEFW_JOIN_MULTICAST_GROUP, error status:%d %pM\n",
3069 /* Enable multicast filtering */
3070 err
= myri10ge_send_cmd(mgp
, MXGEFW_DISABLE_ALLMULTI
, &cmd
, 1);
3072 netdev_err(dev
, "Failed MXGEFW_DISABLE_ALLMULTI, error status: %d\n",
3083 static int myri10ge_set_mac_address(struct net_device
*dev
, void *addr
)
3085 struct sockaddr
*sa
= addr
;
3086 struct myri10ge_priv
*mgp
= netdev_priv(dev
);
3089 if (!is_valid_ether_addr(sa
->sa_data
))
3090 return -EADDRNOTAVAIL
;
3092 status
= myri10ge_update_mac_address(mgp
, sa
->sa_data
);
3094 netdev_err(dev
, "changing mac address failed with %d\n",
3099 /* change the dev structure */
3100 memcpy(dev
->dev_addr
, sa
->sa_data
, 6);
3104 static int myri10ge_change_mtu(struct net_device
*dev
, int new_mtu
)
3106 struct myri10ge_priv
*mgp
= netdev_priv(dev
);
3109 if ((new_mtu
< 68) || (ETH_HLEN
+ new_mtu
> MYRI10GE_MAX_ETHER_MTU
)) {
3110 netdev_err(dev
, "new mtu (%d) is not valid\n", new_mtu
);
3113 netdev_info(dev
, "changing mtu from %d to %d\n", dev
->mtu
, new_mtu
);
3115 /* if we change the mtu on an active device, we must
3116 * reset the device so the firmware sees the change */
3117 myri10ge_close(dev
);
3127 * Enable ECRC to align PCI-E Completion packets on an 8-byte boundary.
3128 * Only do it if the bridge is a root port since we don't want to disturb
3129 * any other device, except if forced with myri10ge_ecrc_enable > 1.
3132 static void myri10ge_enable_ecrc(struct myri10ge_priv
*mgp
)
3134 struct pci_dev
*bridge
= mgp
->pdev
->bus
->self
;
3135 struct device
*dev
= &mgp
->pdev
->dev
;
3142 if (!myri10ge_ecrc_enable
|| !bridge
)
3145 /* check that the bridge is a root port */
3146 cap
= pci_find_capability(bridge
, PCI_CAP_ID_EXP
);
3147 pci_read_config_word(bridge
, cap
+ PCI_CAP_FLAGS
, &val
);
3148 ext_type
= (val
& PCI_EXP_FLAGS_TYPE
) >> 4;
3149 if (ext_type
!= PCI_EXP_TYPE_ROOT_PORT
) {
3150 if (myri10ge_ecrc_enable
> 1) {
3151 struct pci_dev
*prev_bridge
, *old_bridge
= bridge
;
3153 /* Walk the hierarchy up to the root port
3154 * where ECRC has to be enabled */
3156 prev_bridge
= bridge
;
3157 bridge
= bridge
->bus
->self
;
3158 if (!bridge
|| prev_bridge
== bridge
) {
3160 "Failed to find root port"
3161 " to force ECRC\n");
3165 pci_find_capability(bridge
, PCI_CAP_ID_EXP
);
3166 pci_read_config_word(bridge
,
3167 cap
+ PCI_CAP_FLAGS
, &val
);
3168 ext_type
= (val
& PCI_EXP_FLAGS_TYPE
) >> 4;
3169 } while (ext_type
!= PCI_EXP_TYPE_ROOT_PORT
);
3172 "Forcing ECRC on non-root port %s"
3173 " (enabling on root port %s)\n",
3174 pci_name(old_bridge
), pci_name(bridge
));
3177 "Not enabling ECRC on non-root port %s\n",
3183 cap
= pci_find_ext_capability(bridge
, PCI_EXT_CAP_ID_ERR
);
3187 ret
= pci_read_config_dword(bridge
, cap
+ PCI_ERR_CAP
, &err_cap
);
3189 dev_err(dev
, "failed reading ext-conf-space of %s\n",
3191 dev_err(dev
, "\t pci=nommconf in use? "
3192 "or buggy/incomplete/absent ACPI MCFG attr?\n");
3195 if (!(err_cap
& PCI_ERR_CAP_ECRC_GENC
))
3198 err_cap
|= PCI_ERR_CAP_ECRC_GENE
;
3199 pci_write_config_dword(bridge
, cap
+ PCI_ERR_CAP
, err_cap
);
3200 dev_info(dev
, "Enabled ECRC on upstream bridge %s\n", pci_name(bridge
));
3204 * The Lanai Z8E PCI-E interface achieves higher Read-DMA throughput
3205 * when the PCI-E Completion packets are aligned on an 8-byte
3206 * boundary. Some PCI-E chip sets always align Completion packets; on
3207 * the ones that do not, the alignment can be enforced by enabling
3208 * ECRC generation (if supported).
3210 * When PCI-E Completion packets are not aligned, it is actually more
3211 * efficient to limit Read-DMA transactions to 2KB, rather than 4KB.
3213 * If the driver can neither enable ECRC nor verify that it has
3214 * already been enabled, then it must use a firmware image which works
3215 * around unaligned completion packets (myri10ge_rss_ethp_z8e.dat), and it
3216 * should also ensure that it never gives the device a Read-DMA which is
3217 * larger than 2KB by setting the tx_boundary to 2KB. If ECRC is
3218 * enabled, then the driver should use the aligned (myri10ge_rss_eth_z8e.dat)
3219 * firmware image, and set tx_boundary to 4KB.
3222 static void myri10ge_firmware_probe(struct myri10ge_priv
*mgp
)
3224 struct pci_dev
*pdev
= mgp
->pdev
;
3225 struct device
*dev
= &pdev
->dev
;
3228 mgp
->tx_boundary
= 4096;
3230 * Verify the max read request size was set to 4KB
3231 * before trying the test with 4KB.
3233 status
= pcie_get_readrq(pdev
);
3235 dev_err(dev
, "Couldn't read max read req size: %d\n", status
);
3238 if (status
!= 4096) {
3239 dev_warn(dev
, "Max Read Request size != 4096 (%d)\n", status
);
3240 mgp
->tx_boundary
= 2048;
3243 * load the optimized firmware (which assumes aligned PCIe
3244 * completions) in order to see if it works on this host.
3246 set_fw_name(mgp
, myri10ge_fw_aligned
, false);
3247 status
= myri10ge_load_firmware(mgp
, 1);
3253 * Enable ECRC if possible
3255 myri10ge_enable_ecrc(mgp
);
3258 * Run a DMA test which watches for unaligned completions and
3259 * aborts on the first one seen.
3262 status
= myri10ge_dma_test(mgp
, MXGEFW_CMD_UNALIGNED_TEST
);
3264 return; /* keep the aligned firmware */
3266 if (status
!= -E2BIG
)
3267 dev_warn(dev
, "DMA test failed: %d\n", status
);
3268 if (status
== -ENOSYS
)
3269 dev_warn(dev
, "Falling back to ethp! "
3270 "Please install up to date fw\n");
3272 /* fall back to using the unaligned firmware */
3273 mgp
->tx_boundary
= 2048;
3274 set_fw_name(mgp
, myri10ge_fw_unaligned
, false);
3278 static void myri10ge_select_firmware(struct myri10ge_priv
*mgp
)
3282 if (myri10ge_force_firmware
== 0) {
3283 int link_width
, exp_cap
;
3286 exp_cap
= pci_find_capability(mgp
->pdev
, PCI_CAP_ID_EXP
);
3287 pci_read_config_word(mgp
->pdev
, exp_cap
+ PCI_EXP_LNKSTA
, &lnk
);
3288 link_width
= (lnk
>> 4) & 0x3f;
3290 /* Check to see if Link is less than 8 or if the
3291 * upstream bridge is known to provide aligned
3293 if (link_width
< 8) {
3294 dev_info(&mgp
->pdev
->dev
, "PCIE x%d Link\n",
3296 mgp
->tx_boundary
= 4096;
3297 set_fw_name(mgp
, myri10ge_fw_aligned
, false);
3299 myri10ge_firmware_probe(mgp
);
3302 if (myri10ge_force_firmware
== 1) {
3303 dev_info(&mgp
->pdev
->dev
,
3304 "Assuming aligned completions (forced)\n");
3305 mgp
->tx_boundary
= 4096;
3306 set_fw_name(mgp
, myri10ge_fw_aligned
, false);
3308 dev_info(&mgp
->pdev
->dev
,
3309 "Assuming unaligned completions (forced)\n");
3310 mgp
->tx_boundary
= 2048;
3311 set_fw_name(mgp
, myri10ge_fw_unaligned
, false);
3315 kparam_block_sysfs_write(myri10ge_fw_name
);
3316 if (myri10ge_fw_name
!= NULL
) {
3317 char *fw_name
= kstrdup(myri10ge_fw_name
, GFP_KERNEL
);
3320 set_fw_name(mgp
, fw_name
, true);
3323 kparam_unblock_sysfs_write(myri10ge_fw_name
);
3325 if (mgp
->board_number
< MYRI10GE_MAX_BOARDS
&&
3326 myri10ge_fw_names
[mgp
->board_number
] != NULL
&&
3327 strlen(myri10ge_fw_names
[mgp
->board_number
])) {
3328 set_fw_name(mgp
, myri10ge_fw_names
[mgp
->board_number
], false);
3332 dev_info(&mgp
->pdev
->dev
, "overriding firmware to %s\n",
3337 static int myri10ge_suspend(struct pci_dev
*pdev
, pm_message_t state
)
3339 struct myri10ge_priv
*mgp
;
3340 struct net_device
*netdev
;
3342 mgp
= pci_get_drvdata(pdev
);
3347 netif_device_detach(netdev
);
3348 if (netif_running(netdev
)) {
3349 netdev_info(netdev
, "closing\n");
3351 myri10ge_close(netdev
);
3354 myri10ge_dummy_rdma(mgp
, 0);
3355 pci_save_state(pdev
);
3356 pci_disable_device(pdev
);
3358 return pci_set_power_state(pdev
, pci_choose_state(pdev
, state
));
3361 static int myri10ge_resume(struct pci_dev
*pdev
)
3363 struct myri10ge_priv
*mgp
;
3364 struct net_device
*netdev
;
3368 mgp
= pci_get_drvdata(pdev
);
3372 pci_set_power_state(pdev
, 0); /* zeros conf space as a side effect */
3373 msleep(5); /* give card time to respond */
3374 pci_read_config_word(mgp
->pdev
, PCI_VENDOR_ID
, &vendor
);
3375 if (vendor
== 0xffff) {
3376 netdev_err(mgp
->dev
, "device disappeared!\n");
3380 status
= pci_restore_state(pdev
);
3384 status
= pci_enable_device(pdev
);
3386 dev_err(&pdev
->dev
, "failed to enable device\n");
3390 pci_set_master(pdev
);
3392 myri10ge_reset(mgp
);
3393 myri10ge_dummy_rdma(mgp
, 1);
3395 /* Save configuration space to be restored if the
3396 * nic resets due to a parity error */
3397 pci_save_state(pdev
);
3399 if (netif_running(netdev
)) {
3401 status
= myri10ge_open(netdev
);
3404 goto abort_with_enabled
;
3407 netif_device_attach(netdev
);
3412 pci_disable_device(pdev
);
3416 #endif /* CONFIG_PM */
3418 static u32
myri10ge_read_reboot(struct myri10ge_priv
*mgp
)
3420 struct pci_dev
*pdev
= mgp
->pdev
;
3421 int vs
= mgp
->vendor_specific_offset
;
3424 /*enter read32 mode */
3425 pci_write_config_byte(pdev
, vs
+ 0x10, 0x3);
3427 /*read REBOOT_STATUS (0xfffffff0) */
3428 pci_write_config_dword(pdev
, vs
+ 0x18, 0xfffffff0);
3429 pci_read_config_dword(pdev
, vs
+ 0x14, &reboot
);
3434 * This watchdog is used to check whether the board has suffered
3435 * from a parity error and needs to be recovered.
3437 static void myri10ge_watchdog(struct work_struct
*work
)
3439 struct myri10ge_priv
*mgp
=
3440 container_of(work
, struct myri10ge_priv
, watchdog_work
);
3441 struct myri10ge_tx_buf
*tx
;
3443 int status
, rebooted
;
3447 mgp
->watchdog_resets
++;
3448 pci_read_config_word(mgp
->pdev
, PCI_COMMAND
, &cmd
);
3450 if ((cmd
& PCI_COMMAND_MASTER
) == 0) {
3451 /* Bus master DMA disabled? Check to see
3452 * if the card rebooted due to a parity error
3453 * For now, just report it */
3454 reboot
= myri10ge_read_reboot(mgp
);
3455 netdev_err(mgp
->dev
, "NIC rebooted (0x%x),%s resetting\n",
3457 myri10ge_reset_recover
? "" : " not");
3458 if (myri10ge_reset_recover
== 0)
3463 myri10ge_close(mgp
->dev
);
3464 myri10ge_reset_recover
--;
3467 * A rebooted nic will come back with config space as
3468 * it was after power was applied to PCIe bus.
3469 * Attempt to restore config space which was saved
3470 * when the driver was loaded, or the last time the
3471 * nic was resumed from power saving mode.
3473 pci_restore_state(mgp
->pdev
);
3475 /* save state again for accounting reasons */
3476 pci_save_state(mgp
->pdev
);
3479 /* if we get back -1's from our slot, perhaps somebody
3480 * powered off our card. Don't try to reset it in
3482 if (cmd
== 0xffff) {
3483 pci_read_config_word(mgp
->pdev
, PCI_VENDOR_ID
, &vendor
);
3484 if (vendor
== 0xffff) {
3485 netdev_err(mgp
->dev
, "device disappeared!\n");
3489 /* Perhaps it is a software error. Try to reset */
3491 netdev_err(mgp
->dev
, "device timeout, resetting\n");
3492 for (i
= 0; i
< mgp
->num_slices
; i
++) {
3493 tx
= &mgp
->ss
[i
].tx
;
3494 netdev_err(mgp
->dev
, "(%d): %d %d %d %d %d %d\n",
3495 i
, tx
->queue_active
, tx
->req
,
3496 tx
->done
, tx
->pkt_start
, tx
->pkt_done
,
3497 (int)ntohl(mgp
->ss
[i
].fw_stats
->
3500 netdev_info(mgp
->dev
, "(%d): %d %d %d %d %d %d\n",
3501 i
, tx
->queue_active
, tx
->req
,
3502 tx
->done
, tx
->pkt_start
, tx
->pkt_done
,
3503 (int)ntohl(mgp
->ss
[i
].fw_stats
->
3510 myri10ge_close(mgp
->dev
);
3512 status
= myri10ge_load_firmware(mgp
, 1);
3514 netdev_err(mgp
->dev
, "failed to load firmware\n");
3516 myri10ge_open(mgp
->dev
);
3521 * We use our own timer routine rather than relying upon
3522 * netdev->tx_timeout because we have a very large hardware transmit
3523 * queue. Due to the large queue, the netdev->tx_timeout function
3524 * cannot detect a NIC with a parity error in a timely fashion if the
3525 * NIC is lightly loaded.
3527 static void myri10ge_watchdog_timer(unsigned long arg
)
3529 struct myri10ge_priv
*mgp
;
3530 struct myri10ge_slice_state
*ss
;
3531 int i
, reset_needed
, busy_slice_cnt
;
3535 mgp
= (struct myri10ge_priv
*)arg
;
3537 rx_pause_cnt
= ntohl(mgp
->ss
[0].fw_stats
->dropped_pause
);
3539 for (i
= 0, reset_needed
= 0;
3540 i
< mgp
->num_slices
&& reset_needed
== 0; ++i
) {
3543 if (ss
->rx_small
.watchdog_needed
) {
3544 myri10ge_alloc_rx_pages(mgp
, &ss
->rx_small
,
3545 mgp
->small_bytes
+ MXGEFW_PAD
,
3547 if (ss
->rx_small
.fill_cnt
- ss
->rx_small
.cnt
>=
3548 myri10ge_fill_thresh
)
3549 ss
->rx_small
.watchdog_needed
= 0;
3551 if (ss
->rx_big
.watchdog_needed
) {
3552 myri10ge_alloc_rx_pages(mgp
, &ss
->rx_big
,
3554 if (ss
->rx_big
.fill_cnt
- ss
->rx_big
.cnt
>=
3555 myri10ge_fill_thresh
)
3556 ss
->rx_big
.watchdog_needed
= 0;
3559 if (ss
->tx
.req
!= ss
->tx
.done
&&
3560 ss
->tx
.done
== ss
->watchdog_tx_done
&&
3561 ss
->watchdog_tx_req
!= ss
->watchdog_tx_done
) {
3562 /* nic seems like it might be stuck.. */
3563 if (rx_pause_cnt
!= mgp
->watchdog_pause
) {
3564 if (net_ratelimit())
3565 netdev_err(mgp
->dev
, "slice %d: TX paused, check link partner\n",
3568 netdev_warn(mgp
->dev
, "slice %d stuck:", i
);
3572 if (ss
->watchdog_tx_done
!= ss
->tx
.done
||
3573 ss
->watchdog_rx_done
!= ss
->rx_done
.cnt
) {
3576 ss
->watchdog_tx_done
= ss
->tx
.done
;
3577 ss
->watchdog_tx_req
= ss
->tx
.req
;
3578 ss
->watchdog_rx_done
= ss
->rx_done
.cnt
;
3580 /* if we've sent or received no traffic, poll the NIC to
3581 * ensure it is still there. Otherwise, we risk not noticing
3582 * an error in a timely fashion */
3583 if (busy_slice_cnt
== 0) {
3584 pci_read_config_word(mgp
->pdev
, PCI_COMMAND
, &cmd
);
3585 if ((cmd
& PCI_COMMAND_MASTER
) == 0) {
3589 mgp
->watchdog_pause
= rx_pause_cnt
;
3592 schedule_work(&mgp
->watchdog_work
);
3595 mod_timer(&mgp
->watchdog_timer
,
3596 jiffies
+ myri10ge_watchdog_timeout
* HZ
);
3600 static void myri10ge_free_slices(struct myri10ge_priv
*mgp
)
3602 struct myri10ge_slice_state
*ss
;
3603 struct pci_dev
*pdev
= mgp
->pdev
;
3607 if (mgp
->ss
== NULL
)
3610 for (i
= 0; i
< mgp
->num_slices
; i
++) {
3612 if (ss
->rx_done
.entry
!= NULL
) {
3613 bytes
= mgp
->max_intr_slots
*
3614 sizeof(*ss
->rx_done
.entry
);
3615 dma_free_coherent(&pdev
->dev
, bytes
,
3616 ss
->rx_done
.entry
, ss
->rx_done
.bus
);
3617 ss
->rx_done
.entry
= NULL
;
3619 if (ss
->fw_stats
!= NULL
) {
3620 bytes
= sizeof(*ss
->fw_stats
);
3621 dma_free_coherent(&pdev
->dev
, bytes
,
3622 ss
->fw_stats
, ss
->fw_stats_bus
);
3623 ss
->fw_stats
= NULL
;
3630 static int myri10ge_alloc_slices(struct myri10ge_priv
*mgp
)
3632 struct myri10ge_slice_state
*ss
;
3633 struct pci_dev
*pdev
= mgp
->pdev
;
3637 bytes
= sizeof(*mgp
->ss
) * mgp
->num_slices
;
3638 mgp
->ss
= kzalloc(bytes
, GFP_KERNEL
);
3639 if (mgp
->ss
== NULL
) {
3643 for (i
= 0; i
< mgp
->num_slices
; i
++) {
3645 bytes
= mgp
->max_intr_slots
* sizeof(*ss
->rx_done
.entry
);
3646 ss
->rx_done
.entry
= dma_alloc_coherent(&pdev
->dev
, bytes
,
3649 if (ss
->rx_done
.entry
== NULL
)
3651 memset(ss
->rx_done
.entry
, 0, bytes
);
3652 bytes
= sizeof(*ss
->fw_stats
);
3653 ss
->fw_stats
= dma_alloc_coherent(&pdev
->dev
, bytes
,
3656 if (ss
->fw_stats
== NULL
)
3660 netif_napi_add(ss
->dev
, &ss
->napi
, myri10ge_poll
,
3661 myri10ge_napi_weight
);
3665 myri10ge_free_slices(mgp
);
3670 * This function determines the number of slices supported.
3671 * The number slices is the minumum of the number of CPUS,
3672 * the number of MSI-X irqs supported, the number of slices
3673 * supported by the firmware
3675 static void myri10ge_probe_slices(struct myri10ge_priv
*mgp
)
3677 struct myri10ge_cmd cmd
;
3678 struct pci_dev
*pdev
= mgp
->pdev
;
3681 int i
, status
, ncpus
, msix_cap
;
3683 mgp
->num_slices
= 1;
3684 msix_cap
= pci_find_capability(pdev
, PCI_CAP_ID_MSIX
);
3685 ncpus
= num_online_cpus();
3687 if (myri10ge_max_slices
== 1 || msix_cap
== 0 ||
3688 (myri10ge_max_slices
== -1 && ncpus
< 2))
3691 /* try to load the slice aware rss firmware */
3692 old_fw
= mgp
->fw_name
;
3693 old_allocated
= mgp
->fw_name_allocated
;
3694 /* don't free old_fw if we override it. */
3695 mgp
->fw_name_allocated
= false;
3697 if (myri10ge_fw_name
!= NULL
) {
3698 dev_info(&mgp
->pdev
->dev
, "overriding rss firmware to %s\n",
3700 set_fw_name(mgp
, myri10ge_fw_name
, false);
3701 } else if (old_fw
== myri10ge_fw_aligned
)
3702 set_fw_name(mgp
, myri10ge_fw_rss_aligned
, false);
3704 set_fw_name(mgp
, myri10ge_fw_rss_unaligned
, false);
3705 status
= myri10ge_load_firmware(mgp
, 0);
3707 dev_info(&pdev
->dev
, "Rss firmware not found\n");
3713 /* hit the board with a reset to ensure it is alive */
3714 memset(&cmd
, 0, sizeof(cmd
));
3715 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_RESET
, &cmd
, 0);
3717 dev_err(&mgp
->pdev
->dev
, "failed reset\n");
3721 mgp
->max_intr_slots
= cmd
.data0
/ sizeof(struct mcp_slot
);
3723 /* tell it the size of the interrupt queues */
3724 cmd
.data0
= mgp
->max_intr_slots
* sizeof(struct mcp_slot
);
3725 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_SET_INTRQ_SIZE
, &cmd
, 0);
3727 dev_err(&mgp
->pdev
->dev
, "failed MXGEFW_CMD_SET_INTRQ_SIZE\n");
3731 /* ask the maximum number of slices it supports */
3732 status
= myri10ge_send_cmd(mgp
, MXGEFW_CMD_GET_MAX_RSS_QUEUES
, &cmd
, 0);
3736 mgp
->num_slices
= cmd
.data0
;
3738 /* Only allow multiple slices if MSI-X is usable */
3739 if (!myri10ge_msi
) {
3743 /* if the admin did not specify a limit to how many
3744 * slices we should use, cap it automatically to the
3745 * number of CPUs currently online */
3746 if (myri10ge_max_slices
== -1)
3747 myri10ge_max_slices
= ncpus
;
3749 if (mgp
->num_slices
> myri10ge_max_slices
)
3750 mgp
->num_slices
= myri10ge_max_slices
;
3752 /* Now try to allocate as many MSI-X vectors as we have
3753 * slices. We give up on MSI-X if we can only get a single
3756 mgp
->msix_vectors
= kzalloc(mgp
->num_slices
*
3757 sizeof(*mgp
->msix_vectors
), GFP_KERNEL
);
3758 if (mgp
->msix_vectors
== NULL
)
3760 for (i
= 0; i
< mgp
->num_slices
; i
++) {
3761 mgp
->msix_vectors
[i
].entry
= i
;
3764 while (mgp
->num_slices
> 1) {
3765 /* make sure it is a power of two */
3766 while (!is_power_of_2(mgp
->num_slices
))
3768 if (mgp
->num_slices
== 1)
3770 status
= pci_enable_msix(pdev
, mgp
->msix_vectors
,
3773 pci_disable_msix(pdev
);
3779 mgp
->num_slices
= status
;
3785 if (mgp
->msix_vectors
!= NULL
) {
3786 kfree(mgp
->msix_vectors
);
3787 mgp
->msix_vectors
= NULL
;
3791 mgp
->num_slices
= 1;
3792 set_fw_name(mgp
, old_fw
, old_allocated
);
3793 myri10ge_load_firmware(mgp
, 0);
3796 static const struct net_device_ops myri10ge_netdev_ops
= {
3797 .ndo_open
= myri10ge_open
,
3798 .ndo_stop
= myri10ge_close
,
3799 .ndo_start_xmit
= myri10ge_xmit
,
3800 .ndo_get_stats
= myri10ge_get_stats
,
3801 .ndo_validate_addr
= eth_validate_addr
,
3802 .ndo_change_mtu
= myri10ge_change_mtu
,
3803 .ndo_set_multicast_list
= myri10ge_set_multicast_list
,
3804 .ndo_set_mac_address
= myri10ge_set_mac_address
,
3807 static int myri10ge_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
3809 struct net_device
*netdev
;
3810 struct myri10ge_priv
*mgp
;
3811 struct device
*dev
= &pdev
->dev
;
3813 int status
= -ENXIO
;
3815 unsigned hdr_offset
, ss_offset
;
3816 static int board_number
;
3818 netdev
= alloc_etherdev_mq(sizeof(*mgp
), MYRI10GE_MAX_SLICES
);
3819 if (netdev
== NULL
) {
3820 dev_err(dev
, "Could not allocate ethernet device\n");
3824 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
3826 mgp
= netdev_priv(netdev
);
3829 mgp
->csum_flag
= MXGEFW_FLAGS_CKSUM
;
3830 mgp
->pause
= myri10ge_flow_control
;
3831 mgp
->intr_coal_delay
= myri10ge_intr_coal_delay
;
3832 mgp
->msg_enable
= netif_msg_init(myri10ge_debug
, MYRI10GE_MSG_DEFAULT
);
3833 mgp
->board_number
= board_number
;
3834 init_waitqueue_head(&mgp
->down_wq
);
3836 if (pci_enable_device(pdev
)) {
3837 dev_err(&pdev
->dev
, "pci_enable_device call failed\n");
3839 goto abort_with_netdev
;
3842 /* Find the vendor-specific cap so we can check
3843 * the reboot register later on */
3844 mgp
->vendor_specific_offset
3845 = pci_find_capability(pdev
, PCI_CAP_ID_VNDR
);
3847 /* Set our max read request to 4KB */
3848 status
= pcie_set_readrq(pdev
, 4096);
3850 dev_err(&pdev
->dev
, "Error %d writing PCI_EXP_DEVCTL\n",
3852 goto abort_with_enabled
;
3855 pci_set_master(pdev
);
3857 status
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(64));
3861 "64-bit pci address mask was refused, "
3863 status
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(32));
3866 dev_err(&pdev
->dev
, "Error %d setting DMA mask\n", status
);
3867 goto abort_with_enabled
;
3869 (void)pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(64));
3870 mgp
->cmd
= dma_alloc_coherent(&pdev
->dev
, sizeof(*mgp
->cmd
),
3871 &mgp
->cmd_bus
, GFP_KERNEL
);
3872 if (mgp
->cmd
== NULL
)
3873 goto abort_with_enabled
;
3875 mgp
->board_span
= pci_resource_len(pdev
, 0);
3876 mgp
->iomem_base
= pci_resource_start(pdev
, 0);
3878 mgp
->wc_enabled
= 0;
3880 mgp
->mtrr
= mtrr_add(mgp
->iomem_base
, mgp
->board_span
,
3881 MTRR_TYPE_WRCOMB
, 1);
3883 mgp
->wc_enabled
= 1;
3885 mgp
->sram
= ioremap_wc(mgp
->iomem_base
, mgp
->board_span
);
3886 if (mgp
->sram
== NULL
) {
3887 dev_err(&pdev
->dev
, "ioremap failed for %ld bytes at 0x%lx\n",
3888 mgp
->board_span
, mgp
->iomem_base
);
3890 goto abort_with_mtrr
;
3893 ntohl(__raw_readl(mgp
->sram
+ MCP_HEADER_PTR_OFFSET
)) & 0xffffc;
3894 ss_offset
= hdr_offset
+ offsetof(struct mcp_gen_header
, string_specs
);
3895 mgp
->sram_size
= ntohl(__raw_readl(mgp
->sram
+ ss_offset
));
3896 if (mgp
->sram_size
> mgp
->board_span
||
3897 mgp
->sram_size
<= MYRI10GE_FW_OFFSET
) {
3899 "invalid sram_size %dB or board span %ldB\n",
3900 mgp
->sram_size
, mgp
->board_span
);
3901 goto abort_with_ioremap
;
3903 memcpy_fromio(mgp
->eeprom_strings
,
3904 mgp
->sram
+ mgp
->sram_size
, MYRI10GE_EEPROM_STRINGS_SIZE
);
3905 memset(mgp
->eeprom_strings
+ MYRI10GE_EEPROM_STRINGS_SIZE
- 2, 0, 2);
3906 status
= myri10ge_read_mac_addr(mgp
);
3908 goto abort_with_ioremap
;
3910 for (i
= 0; i
< ETH_ALEN
; i
++)
3911 netdev
->dev_addr
[i
] = mgp
->mac_addr
[i
];
3913 myri10ge_select_firmware(mgp
);
3915 status
= myri10ge_load_firmware(mgp
, 1);
3917 dev_err(&pdev
->dev
, "failed to load firmware\n");
3918 goto abort_with_ioremap
;
3920 myri10ge_probe_slices(mgp
);
3921 status
= myri10ge_alloc_slices(mgp
);
3923 dev_err(&pdev
->dev
, "failed to alloc slice state\n");
3924 goto abort_with_firmware
;
3926 netdev
->real_num_tx_queues
= mgp
->num_slices
;
3927 status
= myri10ge_reset(mgp
);
3929 dev_err(&pdev
->dev
, "failed reset\n");
3930 goto abort_with_slices
;
3932 #ifdef CONFIG_MYRI10GE_DCA
3933 myri10ge_setup_dca(mgp
);
3935 pci_set_drvdata(pdev
, mgp
);
3936 if ((myri10ge_initial_mtu
+ ETH_HLEN
) > MYRI10GE_MAX_ETHER_MTU
)
3937 myri10ge_initial_mtu
= MYRI10GE_MAX_ETHER_MTU
- ETH_HLEN
;
3938 if ((myri10ge_initial_mtu
+ ETH_HLEN
) < 68)
3939 myri10ge_initial_mtu
= 68;
3941 netdev
->netdev_ops
= &myri10ge_netdev_ops
;
3942 netdev
->mtu
= myri10ge_initial_mtu
;
3943 netdev
->base_addr
= mgp
->iomem_base
;
3944 netdev
->features
= mgp
->features
;
3947 netdev
->features
|= NETIF_F_HIGHDMA
;
3948 netdev
->features
|= NETIF_F_LRO
;
3950 netdev
->vlan_features
|= mgp
->features
;
3951 if (mgp
->fw_ver_tiny
< 37)
3952 netdev
->vlan_features
&= ~NETIF_F_TSO6
;
3953 if (mgp
->fw_ver_tiny
< 32)
3954 netdev
->vlan_features
&= ~NETIF_F_TSO
;
3956 /* make sure we can get an irq, and that MSI can be
3957 * setup (if available). Also ensure netdev->irq
3958 * is set to correct value if MSI is enabled */
3959 status
= myri10ge_request_irq(mgp
);
3961 goto abort_with_firmware
;
3962 netdev
->irq
= pdev
->irq
;
3963 myri10ge_free_irq(mgp
);
3965 /* Save configuration space to be restored if the
3966 * nic resets due to a parity error */
3967 pci_save_state(pdev
);
3969 /* Setup the watchdog timer */
3970 setup_timer(&mgp
->watchdog_timer
, myri10ge_watchdog_timer
,
3971 (unsigned long)mgp
);
3973 spin_lock_init(&mgp
->stats_lock
);
3974 SET_ETHTOOL_OPS(netdev
, &myri10ge_ethtool_ops
);
3975 INIT_WORK(&mgp
->watchdog_work
, myri10ge_watchdog
);
3976 status
= register_netdev(netdev
);
3978 dev_err(&pdev
->dev
, "register_netdev failed: %d\n", status
);
3979 goto abort_with_state
;
3981 if (mgp
->msix_enabled
)
3982 dev_info(dev
, "%d MSI-X IRQs, tx bndry %d, fw %s, WC %s\n",
3983 mgp
->num_slices
, mgp
->tx_boundary
, mgp
->fw_name
,
3984 (mgp
->wc_enabled
? "Enabled" : "Disabled"));
3986 dev_info(dev
, "%s IRQ %d, tx bndry %d, fw %s, WC %s\n",
3987 mgp
->msi_enabled
? "MSI" : "xPIC",
3988 netdev
->irq
, mgp
->tx_boundary
, mgp
->fw_name
,
3989 (mgp
->wc_enabled
? "Enabled" : "Disabled"));
3995 pci_restore_state(pdev
);
3998 myri10ge_free_slices(mgp
);
4000 abort_with_firmware
:
4001 myri10ge_dummy_rdma(mgp
, 0);
4004 if (mgp
->mac_addr_string
!= NULL
)
4006 "myri10ge_probe() failed: MAC=%s, SN=%ld\n",
4007 mgp
->mac_addr_string
, mgp
->serial_number
);
4013 mtrr_del(mgp
->mtrr
, mgp
->iomem_base
, mgp
->board_span
);
4015 dma_free_coherent(&pdev
->dev
, sizeof(*mgp
->cmd
),
4016 mgp
->cmd
, mgp
->cmd_bus
);
4019 pci_disable_device(pdev
);
4022 set_fw_name(mgp
, NULL
, false);
4023 free_netdev(netdev
);
4030 * Does what is necessary to shutdown one Myrinet device. Called
4031 * once for each Myrinet card by the kernel when a module is
4034 static void myri10ge_remove(struct pci_dev
*pdev
)
4036 struct myri10ge_priv
*mgp
;
4037 struct net_device
*netdev
;
4039 mgp
= pci_get_drvdata(pdev
);
4043 flush_scheduled_work();
4045 unregister_netdev(netdev
);
4047 #ifdef CONFIG_MYRI10GE_DCA
4048 myri10ge_teardown_dca(mgp
);
4050 myri10ge_dummy_rdma(mgp
, 0);
4052 /* avoid a memory leak */
4053 pci_restore_state(pdev
);
4059 mtrr_del(mgp
->mtrr
, mgp
->iomem_base
, mgp
->board_span
);
4061 myri10ge_free_slices(mgp
);
4062 if (mgp
->msix_vectors
!= NULL
)
4063 kfree(mgp
->msix_vectors
);
4064 dma_free_coherent(&pdev
->dev
, sizeof(*mgp
->cmd
),
4065 mgp
->cmd
, mgp
->cmd_bus
);
4067 set_fw_name(mgp
, NULL
, false);
4068 free_netdev(netdev
);
4069 pci_disable_device(pdev
);
4070 pci_set_drvdata(pdev
, NULL
);
4073 #define PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E 0x0008
4074 #define PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E_9 0x0009
4076 static DEFINE_PCI_DEVICE_TABLE(myri10ge_pci_tbl
) = {
4077 {PCI_DEVICE(PCI_VENDOR_ID_MYRICOM
, PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E
)},
4079 (PCI_VENDOR_ID_MYRICOM
, PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E_9
)},
4083 MODULE_DEVICE_TABLE(pci
, myri10ge_pci_tbl
);
4085 static struct pci_driver myri10ge_driver
= {
4087 .probe
= myri10ge_probe
,
4088 .remove
= myri10ge_remove
,
4089 .id_table
= myri10ge_pci_tbl
,
4091 .suspend
= myri10ge_suspend
,
4092 .resume
= myri10ge_resume
,
4096 #ifdef CONFIG_MYRI10GE_DCA
4098 myri10ge_notify_dca(struct notifier_block
*nb
, unsigned long event
, void *p
)
4100 int err
= driver_for_each_device(&myri10ge_driver
.driver
,
4102 myri10ge_notify_dca_device
);
4109 static struct notifier_block myri10ge_dca_notifier
= {
4110 .notifier_call
= myri10ge_notify_dca
,
4114 #endif /* CONFIG_MYRI10GE_DCA */
4116 static __init
int myri10ge_init_module(void)
4118 pr_info("Version %s\n", MYRI10GE_VERSION_STR
);
4120 if (myri10ge_rss_hash
> MXGEFW_RSS_HASH_TYPE_MAX
) {
4121 pr_err("Illegal rssh hash type %d, defaulting to source port\n",
4123 myri10ge_rss_hash
= MXGEFW_RSS_HASH_TYPE_SRC_PORT
;
4125 #ifdef CONFIG_MYRI10GE_DCA
4126 dca_register_notify(&myri10ge_dca_notifier
);
4128 if (myri10ge_max_slices
> MYRI10GE_MAX_SLICES
)
4129 myri10ge_max_slices
= MYRI10GE_MAX_SLICES
;
4131 return pci_register_driver(&myri10ge_driver
);
4134 module_init(myri10ge_init_module
);
4136 static __exit
void myri10ge_cleanup_module(void)
4138 #ifdef CONFIG_MYRI10GE_DCA
4139 dca_unregister_notify(&myri10ge_dca_notifier
);
4141 pci_unregister_driver(&myri10ge_driver
);
4144 module_exit(myri10ge_cleanup_module
);