1/*************************************************************************
2 * myri10ge.c: Myricom Myri-10G Ethernet driver.
3 *
4 * Copyright (C) 2005 - 2011 Myricom, Inc.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of Myricom, Inc. nor the names of its contributors
16 *    may be used to endorse or promote products derived from this software
17 *    without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
23 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 *
31 *
32 * If the eeprom on your board is not recent enough, you will need to get a
33 * newer firmware image at:
34 *   http://www.myri.com/scs/download-Myri10GE.html
35 *
36 * Contact Information:
37 *   <help@myri.com>
38 *   Myricom, Inc., 325N Santa Anita Avenue, Arcadia, CA 91006
39 *************************************************************************/
40
41#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
42
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/dca.h>
54#include <linux/ip.h>
55#include <linux/inet.h>
56#include <linux/in.h>
57#include <linux/ethtool.h>
58#include <linux/firmware.h>
59#include <linux/delay.h>
60#include <linux/timer.h>
61#include <linux/vmalloc.h>
62#include <linux/crc32.h>
63#include <linux/moduleparam.h>
64#include <linux/io.h>
65#include <linux/log2.h>
66#include <linux/slab.h>
67#include <linux/prefetch.h>
68#include <net/checksum.h>
69#include <net/gso.h>
70#include <net/ip.h>
71#include <net/tcp.h>
72#include <asm/byteorder.h>
73#include <asm/processor.h>
74
75#include "myri10ge_mcp.h"
76#include "myri10ge_mcp_gen_header.h"
77
78#define MYRI10GE_VERSION_STR "1.5.3-1.534"
79
80MODULE_DESCRIPTION("Myricom 10G driver (10GbE)");
81MODULE_AUTHOR("Maintainer: help@myri.com");
82MODULE_VERSION(MYRI10GE_VERSION_STR);
83MODULE_LICENSE("Dual BSD/GPL");
84
85#define MYRI10GE_MAX_ETHER_MTU 9014
86
87#define MYRI10GE_ETH_STOPPED 0
88#define MYRI10GE_ETH_STOPPING 1
89#define MYRI10GE_ETH_STARTING 2
90#define MYRI10GE_ETH_RUNNING 3
91#define MYRI10GE_ETH_OPEN_FAILED 4
92
93#define MYRI10GE_EEPROM_STRINGS_SIZE 256
94#define MYRI10GE_MAX_SEND_DESC_TSO ((65536 / 2048) * 2)
95
96#define MYRI10GE_NO_CONFIRM_DATA htonl(0xffffffff)
97#define MYRI10GE_NO_RESPONSE_RESULT 0xffffffff
98
99#define MYRI10GE_ALLOC_ORDER 0
100#define MYRI10GE_ALLOC_SIZE ((1 << MYRI10GE_ALLOC_ORDER) * PAGE_SIZE)
101#define MYRI10GE_MAX_FRAGS_PER_FRAME (MYRI10GE_MAX_ETHER_MTU/MYRI10GE_ALLOC_SIZE + 1)
102
103#define MYRI10GE_MAX_SLICES 32
104
105struct myri10ge_rx_buffer_state {
106	struct page *page;
107	int page_offset;
108	DEFINE_DMA_UNMAP_ADDR(bus);
109	DEFINE_DMA_UNMAP_LEN(len);
110};
111
112struct myri10ge_tx_buffer_state {
113	struct sk_buff *skb;
114	int last;
115	DEFINE_DMA_UNMAP_ADDR(bus);
116	DEFINE_DMA_UNMAP_LEN(len);
117};
118
119struct myri10ge_cmd {
120	u32 data0;
121	u32 data1;
122	u32 data2;
123};
124
125struct myri10ge_rx_buf {
126	struct mcp_kreq_ether_recv __iomem *lanai;	/* lanai ptr for recv ring */
127	struct mcp_kreq_ether_recv *shadow;	/* host shadow of recv ring */
128	struct myri10ge_rx_buffer_state *info;
129	struct page *page;
130	dma_addr_t bus;
131	int page_offset;
132	int cnt;
133	int fill_cnt;
134	int alloc_fail;
135	int mask;		/* number of rx slots -1 */
136	int watchdog_needed;
137};
138
139struct myri10ge_tx_buf {
140	struct mcp_kreq_ether_send __iomem *lanai;	/* lanai ptr for sendq */
141	__be32 __iomem *send_go;	/* "go" doorbell ptr */
142	__be32 __iomem *send_stop;	/* "stop" doorbell ptr */
143	struct mcp_kreq_ether_send *req_list;	/* host shadow of sendq */
144	char *req_bytes;
145	struct myri10ge_tx_buffer_state *info;
146	int mask;		/* number of transmit slots -1  */
147	int req ____cacheline_aligned;	/* transmit slots submitted     */
148	int pkt_start;		/* packets started */
149	int stop_queue;
150	int linearized;
151	int done ____cacheline_aligned;	/* transmit slots completed     */
152	int pkt_done;		/* packets completed */
153	int wake_queue;
154	int queue_active;
155};
156
157struct myri10ge_rx_done {
158	struct mcp_slot *entry;
159	dma_addr_t bus;
160	int cnt;
161	int idx;
162};
163
164struct myri10ge_slice_netstats {
165	unsigned long rx_packets;
166	unsigned long tx_packets;
167	unsigned long rx_bytes;
168	unsigned long tx_bytes;
169	unsigned long rx_dropped;
170	unsigned long tx_dropped;
171};
172
173struct myri10ge_slice_state {
174	struct myri10ge_tx_buf tx;	/* transmit ring        */
175	struct myri10ge_rx_buf rx_small;
176	struct myri10ge_rx_buf rx_big;
177	struct myri10ge_rx_done rx_done;
178	struct net_device *dev;
179	struct napi_struct napi;
180	struct myri10ge_priv *mgp;
181	struct myri10ge_slice_netstats stats;
182	__be32 __iomem *irq_claim;
183	struct mcp_irq_data *fw_stats;
184	dma_addr_t fw_stats_bus;
185	int watchdog_tx_done;
186	int watchdog_tx_req;
187	int watchdog_rx_done;
188	int stuck;
189#ifdef CONFIG_MYRI10GE_DCA
190	int cached_dca_tag;
191	int cpu;
192	__be32 __iomem *dca_tag;
193#endif
194	char irq_desc[32];
195};
196
197struct myri10ge_priv {
198	struct myri10ge_slice_state *ss;
199	int tx_boundary;	/* boundary transmits cannot cross */
200	int num_slices;
201	int running;		/* running?             */
202	int small_bytes;
203	int big_bytes;
204	int max_intr_slots;
205	struct net_device *dev;
206	u8 __iomem *sram;
207	int sram_size;
208	unsigned long board_span;
209	unsigned long iomem_base;
210	__be32 __iomem *irq_deassert;
211	char *mac_addr_string;
212	struct mcp_cmd_response *cmd;
213	dma_addr_t cmd_bus;
214	struct pci_dev *pdev;
215	int msi_enabled;
216	int msix_enabled;
217	struct msix_entry *msix_vectors;
218#ifdef CONFIG_MYRI10GE_DCA
219	int dca_enabled;
220	int relaxed_order;
221#endif
222	u32 link_state;
223	unsigned int rdma_tags_available;
224	int intr_coal_delay;
225	__be32 __iomem *intr_coal_delay_ptr;
226	int wc_cookie;
227	int down_cnt;
228	wait_queue_head_t down_wq;
229	struct work_struct watchdog_work;
230	struct timer_list watchdog_timer;
231	int watchdog_resets;
232	int watchdog_pause;
233	int pause;
234	bool fw_name_allocated;
235	char *fw_name;
236	char eeprom_strings[MYRI10GE_EEPROM_STRINGS_SIZE];
237	char *product_code_string;
238	char fw_version[128];
239	int fw_ver_major;
240	int fw_ver_minor;
241	int fw_ver_tiny;
242	int adopted_rx_filter_bug;
243	u8 mac_addr[ETH_ALEN];		/* eeprom mac address */
244	unsigned long serial_number;
245	int vendor_specific_offset;
246	int fw_multicast_support;
247	u32 features;
248	u32 max_tso6;
249	u32 read_dma;
250	u32 write_dma;
251	u32 read_write_dma;
252	u32 link_changes;
253	u32 msg_enable;
254	unsigned int board_number;
255	int rebooted;
256};
257
258static char *myri10ge_fw_unaligned = "myri10ge_ethp_z8e.dat";
259static char *myri10ge_fw_aligned = "myri10ge_eth_z8e.dat";
260static char *myri10ge_fw_rss_unaligned = "myri10ge_rss_ethp_z8e.dat";
261static char *myri10ge_fw_rss_aligned = "myri10ge_rss_eth_z8e.dat";
262MODULE_FIRMWARE("myri10ge_ethp_z8e.dat");
263MODULE_FIRMWARE("myri10ge_eth_z8e.dat");
264MODULE_FIRMWARE("myri10ge_rss_ethp_z8e.dat");
265MODULE_FIRMWARE("myri10ge_rss_eth_z8e.dat");
266
267/* Careful: must be accessed under kernel_param_lock() */
268static char *myri10ge_fw_name = NULL;
269module_param(myri10ge_fw_name, charp, 0644);
270MODULE_PARM_DESC(myri10ge_fw_name, "Firmware image name");
271
272#define MYRI10GE_MAX_BOARDS 8
273static char *myri10ge_fw_names[MYRI10GE_MAX_BOARDS] =
274    {[0 ... (MYRI10GE_MAX_BOARDS - 1)] = NULL };
275module_param_array_named(myri10ge_fw_names, myri10ge_fw_names, charp, NULL,
276			 0444);
277MODULE_PARM_DESC(myri10ge_fw_names, "Firmware image names per board");
278
279static int myri10ge_ecrc_enable = 1;
280module_param(myri10ge_ecrc_enable, int, 0444);
281MODULE_PARM_DESC(myri10ge_ecrc_enable, "Enable Extended CRC on PCI-E");
282
283static int myri10ge_small_bytes = -1;	/* -1 == auto */
284module_param(myri10ge_small_bytes, int, 0644);
285MODULE_PARM_DESC(myri10ge_small_bytes, "Threshold of small packets");
286
287static int myri10ge_msi = 1;	/* enable msi by default */
288module_param(myri10ge_msi, int, 0644);
289MODULE_PARM_DESC(myri10ge_msi, "Enable Message Signalled Interrupts");
290
291static int myri10ge_intr_coal_delay = 75;
292module_param(myri10ge_intr_coal_delay, int, 0444);
293MODULE_PARM_DESC(myri10ge_intr_coal_delay, "Interrupt coalescing delay");
294
295static int myri10ge_flow_control = 1;
296module_param(myri10ge_flow_control, int, 0444);
297MODULE_PARM_DESC(myri10ge_flow_control, "Pause parameter");
298
299static int myri10ge_deassert_wait = 1;
300module_param(myri10ge_deassert_wait, int, 0644);
301MODULE_PARM_DESC(myri10ge_deassert_wait,
302		 "Wait when deasserting legacy interrupts");
303
304static int myri10ge_force_firmware = 0;
305module_param(myri10ge_force_firmware, int, 0444);
306MODULE_PARM_DESC(myri10ge_force_firmware,
307		 "Force firmware to assume aligned completions");
308
309static int myri10ge_initial_mtu = MYRI10GE_MAX_ETHER_MTU - ETH_HLEN;
310module_param(myri10ge_initial_mtu, int, 0444);
311MODULE_PARM_DESC(myri10ge_initial_mtu, "Initial MTU");
312
313static int myri10ge_napi_weight = 64;
314module_param(myri10ge_napi_weight, int, 0444);
315MODULE_PARM_DESC(myri10ge_napi_weight, "Set NAPI weight");
316
317static int myri10ge_watchdog_timeout = 1;
318module_param(myri10ge_watchdog_timeout, int, 0444);
319MODULE_PARM_DESC(myri10ge_watchdog_timeout, "Set watchdog timeout");
320
321static int myri10ge_max_irq_loops = 1048576;
322module_param(myri10ge_max_irq_loops, int, 0444);
323MODULE_PARM_DESC(myri10ge_max_irq_loops,
324		 "Set stuck legacy IRQ detection threshold");
325
326#define MYRI10GE_MSG_DEFAULT NETIF_MSG_LINK
327
328static int myri10ge_debug = -1;	/* defaults above */
329module_param(myri10ge_debug, int, 0);
330MODULE_PARM_DESC(myri10ge_debug, "Debug level (0=none,...,16=all)");
331
332static int myri10ge_fill_thresh = 256;
333module_param(myri10ge_fill_thresh, int, 0644);
334MODULE_PARM_DESC(myri10ge_fill_thresh, "Number of empty rx slots allowed");
335
336static int myri10ge_reset_recover = 1;
337
338static int myri10ge_max_slices = 1;
339module_param(myri10ge_max_slices, int, 0444);
340MODULE_PARM_DESC(myri10ge_max_slices, "Max tx/rx queues");
341
342static int myri10ge_rss_hash = MXGEFW_RSS_HASH_TYPE_SRC_DST_PORT;
343module_param(myri10ge_rss_hash, int, 0444);
344MODULE_PARM_DESC(myri10ge_rss_hash, "Type of RSS hashing to do");
345
346static int myri10ge_dca = 1;
347module_param(myri10ge_dca, int, 0444);
348MODULE_PARM_DESC(myri10ge_dca, "Enable DCA if possible");
349
350#define MYRI10GE_FW_OFFSET 1024*1024
351#define MYRI10GE_HIGHPART_TO_U32(X) \
352(sizeof (X) == 8) ? ((u32)((u64)(X) >> 32)) : (0)
353#define MYRI10GE_LOWPART_TO_U32(X) ((u32)(X))
354
355#define myri10ge_pio_copy(to,from,size) __iowrite64_copy(to,from,size/8)
356
357static void myri10ge_set_multicast_list(struct net_device *dev);
358static netdev_tx_t myri10ge_sw_tso(struct sk_buff *skb,
359					 struct net_device *dev);
360
361static inline void put_be32(__be32 val, __be32 __iomem * p)
362{
363	__raw_writel((__force __u32) val, (__force void __iomem *)p);
364}
365
366static void myri10ge_get_stats(struct net_device *dev,
367			       struct rtnl_link_stats64 *stats);
368
369static void set_fw_name(struct myri10ge_priv *mgp, char *name, bool allocated)
370{
371	if (mgp->fw_name_allocated)
372		kfree(mgp->fw_name);
373	mgp->fw_name = name;
374	mgp->fw_name_allocated = allocated;
375}
376
377static int
378myri10ge_send_cmd(struct myri10ge_priv *mgp, u32 cmd,
379		  struct myri10ge_cmd *data, int atomic)
380{
381	struct mcp_cmd *buf;
382	char buf_bytes[sizeof(*buf) + 8];
383	struct mcp_cmd_response *response = mgp->cmd;
384	char __iomem *cmd_addr = mgp->sram + MXGEFW_ETH_CMD;
385	u32 dma_low, dma_high, result, value;
386	int sleep_total = 0;
387
388	/* ensure buf is aligned to 8 bytes */
389	buf = (struct mcp_cmd *)ALIGN((unsigned long)buf_bytes, 8);
390
391	buf->data0 = htonl(data->data0);
392	buf->data1 = htonl(data->data1);
393	buf->data2 = htonl(data->data2);
394	buf->cmd = htonl(cmd);
395	dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
396	dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
397
398	buf->response_addr.low = htonl(dma_low);
399	buf->response_addr.high = htonl(dma_high);
400	response->result = htonl(MYRI10GE_NO_RESPONSE_RESULT);
401	mb();
402	myri10ge_pio_copy(cmd_addr, buf, sizeof(*buf));
403
404	/* wait up to 15ms. Longest command is the DMA benchmark,
405	 * which is capped at 5ms, but runs from a timeout handler
406	 * that runs every 7.8ms. So a 15ms timeout leaves us with
407	 * a 2.2ms margin
408	 */
409	if (atomic) {
410		/* if atomic is set, do not sleep,
411		 * and try to get the completion quickly
412		 * (1ms will be enough for those commands) */
413		for (sleep_total = 0;
414		     sleep_total < 1000 &&
415		     response->result == htonl(MYRI10GE_NO_RESPONSE_RESULT);
416		     sleep_total += 10) {
417			udelay(10);
418			mb();
419		}
420	} else {
421		/* use msleep for most command */
422		for (sleep_total = 0;
423		     sleep_total < 15 &&
424		     response->result == htonl(MYRI10GE_NO_RESPONSE_RESULT);
425		     sleep_total++)
426			msleep(1);
427	}
428
429	result = ntohl(response->result);
430	value = ntohl(response->data);
431	if (result != MYRI10GE_NO_RESPONSE_RESULT) {
432		if (result == 0) {
433			data->data0 = value;
434			return 0;
435		} else if (result == MXGEFW_CMD_UNKNOWN) {
436			return -ENOSYS;
437		} else if (result == MXGEFW_CMD_ERROR_UNALIGNED) {
438			return -E2BIG;
439		} else if (result == MXGEFW_CMD_ERROR_RANGE &&
440			   cmd == MXGEFW_CMD_ENABLE_RSS_QUEUES &&
441			   (data->
442			    data1 & MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES) !=
443			   0) {
444			return -ERANGE;
445		} else {
446			dev_err(&mgp->pdev->dev,
447				"command %d failed, result = %d\n",
448				cmd, result);
449			return -ENXIO;
450		}
451	}
452
453	dev_err(&mgp->pdev->dev, "command %d timed out, result = %d\n",
454		cmd, result);
455	return -EAGAIN;
456}
457
458/*
459 * The eeprom strings on the lanaiX have the format
460 * SN=x\0
461 * MAC=x:x:x:x:x:x\0
462 * PT:ddd mmm xx xx:xx:xx xx\0
463 * PV:ddd mmm xx xx:xx:xx xx\0
464 */
465static int myri10ge_read_mac_addr(struct myri10ge_priv *mgp)
466{
467	char *ptr, *limit;
468	int i;
469
470	ptr = mgp->eeprom_strings;
471	limit = mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE;
472
473	while (*ptr != '\0' && ptr < limit) {
474		if (memcmp(ptr, "MAC=", 4) == 0) {
475			ptr += 4;
476			mgp->mac_addr_string = ptr;
477			for (i = 0; i < 6; i++) {
478				if ((ptr + 2) > limit)
479					goto abort;
480				mgp->mac_addr[i] =
481				    simple_strtoul(ptr, &ptr, 16);
482				ptr += 1;
483			}
484		}
485		if (memcmp(ptr, "PC=", 3) == 0) {
486			ptr += 3;
487			mgp->product_code_string = ptr;
488		}
489		if (memcmp((const void *)ptr, "SN=", 3) == 0) {
490			ptr += 3;
491			mgp->serial_number = simple_strtoul(ptr, &ptr, 10);
492		}
493		while (ptr < limit && *ptr++) ;
494	}
495
496	return 0;
497
498abort:
499	dev_err(&mgp->pdev->dev, "failed to parse eeprom_strings\n");
500	return -ENXIO;
501}
502
503/*
504 * Enable or disable periodic RDMAs from the host to make certain
505 * chipsets resend dropped PCIe messages
506 */
507
508static void myri10ge_dummy_rdma(struct myri10ge_priv *mgp, int enable)
509{
510	char __iomem *submit;
511	__be32 buf[16] __attribute__ ((__aligned__(8)));
512	u32 dma_low, dma_high;
513	int i;
514
515	/* clear confirmation addr */
516	mgp->cmd->data = 0;
517	mb();
518
519	/* send a rdma command to the PCIe engine, and wait for the
520	 * response in the confirmation address.  The firmware should
521	 * write a -1 there to indicate it is alive and well
522	 */
523	dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
524	dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
525
526	buf[0] = htonl(dma_high);	/* confirm addr MSW */
527	buf[1] = htonl(dma_low);	/* confirm addr LSW */
528	buf[2] = MYRI10GE_NO_CONFIRM_DATA;	/* confirm data */
529	buf[3] = htonl(dma_high);	/* dummy addr MSW */
530	buf[4] = htonl(dma_low);	/* dummy addr LSW */
531	buf[5] = htonl(enable);	/* enable? */
532
533	submit = mgp->sram + MXGEFW_BOOT_DUMMY_RDMA;
534
535	myri10ge_pio_copy(submit, &buf, sizeof(buf));
536	for (i = 0; mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 20; i++)
537		msleep(1);
538	if (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA)
539		dev_err(&mgp->pdev->dev, "dummy rdma %s failed\n",
540			(enable ? "enable" : "disable"));
541}
542
543static int
544myri10ge_validate_firmware(struct myri10ge_priv *mgp,
545			   struct mcp_gen_header *hdr)
546{
547	struct device *dev = &mgp->pdev->dev;
548
549	/* check firmware type */
550	if (ntohl(hdr->mcp_type) != MCP_TYPE_ETH) {
551		dev_err(dev, "Bad firmware type: 0x%x\n", ntohl(hdr->mcp_type));
552		return -EINVAL;
553	}
554
555	/* save firmware version for ethtool */
556	strscpy(mgp->fw_version, hdr->version, sizeof(mgp->fw_version));
557
558	sscanf(mgp->fw_version, "%d.%d.%d", &mgp->fw_ver_major,
559	       &mgp->fw_ver_minor, &mgp->fw_ver_tiny);
560
561	if (!(mgp->fw_ver_major == MXGEFW_VERSION_MAJOR &&
562	      mgp->fw_ver_minor == MXGEFW_VERSION_MINOR)) {
563		dev_err(dev, "Found firmware version %s\n", mgp->fw_version);
564		dev_err(dev, "Driver needs %d.%d\n", MXGEFW_VERSION_MAJOR,
565			MXGEFW_VERSION_MINOR);
566		return -EINVAL;
567	}
568	return 0;
569}
570
571static int myri10ge_load_hotplug_firmware(struct myri10ge_priv *mgp, u32 * size)
572{
573	unsigned crc, reread_crc;
574	const struct firmware *fw;
575	struct device *dev = &mgp->pdev->dev;
576	unsigned char *fw_readback;
577	struct mcp_gen_header *hdr;
578	size_t hdr_offset;
579	int status;
580	unsigned i;
581
582	if (request_firmware(&fw, mgp->fw_name, dev) < 0) {
583		dev_err(dev, "Unable to load %s firmware image via hotplug\n",
584			mgp->fw_name);
585		status = -EINVAL;
586		goto abort_with_nothing;
587	}
588
589	/* check size */
590
591	if (fw->size >= mgp->sram_size - MYRI10GE_FW_OFFSET ||
592	    fw->size < MCP_HEADER_PTR_OFFSET + 4) {
593		dev_err(dev, "Firmware size invalid:%d\n", (int)fw->size);
594		status = -EINVAL;
595		goto abort_with_fw;
596	}
597
598	/* check id */
599	hdr_offset = ntohl(*(__be32 *) (fw->data + MCP_HEADER_PTR_OFFSET));
600	if ((hdr_offset & 3) || hdr_offset + sizeof(*hdr) > fw->size) {
601		dev_err(dev, "Bad firmware file\n");
602		status = -EINVAL;
603		goto abort_with_fw;
604	}
605	hdr = (void *)(fw->data + hdr_offset);
606
607	status = myri10ge_validate_firmware(mgp, hdr);
608	if (status != 0)
609		goto abort_with_fw;
610
611	crc = crc32(~0, fw->data, fw->size);
612	for (i = 0; i < fw->size; i += 256) {
613		myri10ge_pio_copy(mgp->sram + MYRI10GE_FW_OFFSET + i,
614				  fw->data + i,
615				  min(256U, (unsigned)(fw->size - i)));
616		mb();
617		readb(mgp->sram);
618	}
619	fw_readback = vmalloc(fw->size);
620	if (!fw_readback) {
621		status = -ENOMEM;
622		goto abort_with_fw;
623	}
624	/* corruption checking is good for parity recovery and buggy chipset */
625	memcpy_fromio(fw_readback, mgp->sram + MYRI10GE_FW_OFFSET, fw->size);
626	reread_crc = crc32(~0, fw_readback, fw->size);
627	vfree(fw_readback);
628	if (crc != reread_crc) {
629		dev_err(dev, "CRC failed(fw-len=%u), got 0x%x (expect 0x%x)\n",
630			(unsigned)fw->size, reread_crc, crc);
631		status = -EIO;
632		goto abort_with_fw;
633	}
634	*size = (u32) fw->size;
635
636abort_with_fw:
637	release_firmware(fw);
638
639abort_with_nothing:
640	return status;
641}
642
643static int myri10ge_adopt_running_firmware(struct myri10ge_priv *mgp)
644{
645	struct mcp_gen_header *hdr;
646	struct device *dev = &mgp->pdev->dev;
647	const size_t bytes = sizeof(struct mcp_gen_header);
648	size_t hdr_offset;
649	int status;
650
651	/* find running firmware header */
652	hdr_offset = swab32(readl(mgp->sram + MCP_HEADER_PTR_OFFSET));
653
654	if ((hdr_offset & 3) || hdr_offset + sizeof(*hdr) > mgp->sram_size) {
655		dev_err(dev, "Running firmware has bad header offset (%d)\n",
656			(int)hdr_offset);
657		return -EIO;
658	}
659
660	/* copy header of running firmware from SRAM to host memory to
661	 * validate firmware */
662	hdr = kmalloc(bytes, GFP_KERNEL);
663	if (hdr == NULL)
664		return -ENOMEM;
665
666	memcpy_fromio(hdr, mgp->sram + hdr_offset, bytes);
667	status = myri10ge_validate_firmware(mgp, hdr);
668	kfree(hdr);
669
670	/* check to see if adopted firmware has bug where adopting
671	 * it will cause broadcasts to be filtered unless the NIC
672	 * is kept in ALLMULTI mode */
673	if (mgp->fw_ver_major == 1 && mgp->fw_ver_minor == 4 &&
674	    mgp->fw_ver_tiny >= 4 && mgp->fw_ver_tiny <= 11) {
675		mgp->adopted_rx_filter_bug = 1;
676		dev_warn(dev, "Adopting fw %d.%d.%d: "
677			 "working around rx filter bug\n",
678			 mgp->fw_ver_major, mgp->fw_ver_minor,
679			 mgp->fw_ver_tiny);
680	}
681	return status;
682}
683
684static int myri10ge_get_firmware_capabilities(struct myri10ge_priv *mgp)
685{
686	struct myri10ge_cmd cmd;
687	int status;
688
689	/* probe for IPv6 TSO support */
690	mgp->features = NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_TSO;
691	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_MAX_TSO6_HDR_SIZE,
692				   &cmd, 0);
693	if (status == 0) {
694		mgp->max_tso6 = cmd.data0;
695		mgp->features |= NETIF_F_TSO6;
696	}
697
698	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_RX_RING_SIZE, &cmd, 0);
699	if (status != 0) {
700		dev_err(&mgp->pdev->dev,
701			"failed MXGEFW_CMD_GET_RX_RING_SIZE\n");
702		return -ENXIO;
703	}
704
705	mgp->max_intr_slots = 2 * (cmd.data0 / sizeof(struct mcp_dma_addr));
706
707	return 0;
708}
709
710static int myri10ge_load_firmware(struct myri10ge_priv *mgp, int adopt)
711{
712	char __iomem *submit;
713	__be32 buf[16] __attribute__ ((__aligned__(8)));
714	u32 dma_low, dma_high, size;
715	int status, i;
716
717	size = 0;
718	status = myri10ge_load_hotplug_firmware(mgp, &size);
719	if (status) {
720		if (!adopt)
721			return status;
722		dev_warn(&mgp->pdev->dev, "hotplug firmware loading failed\n");
723
724		/* Do not attempt to adopt firmware if there
725		 * was a bad crc */
726		if (status == -EIO)
727			return status;
728
729		status = myri10ge_adopt_running_firmware(mgp);
730		if (status != 0) {
731			dev_err(&mgp->pdev->dev,
732				"failed to adopt running firmware\n");
733			return status;
734		}
735		dev_info(&mgp->pdev->dev,
736			 "Successfully adopted running firmware\n");
737		if (mgp->tx_boundary == 4096) {
738			dev_warn(&mgp->pdev->dev,
739				 "Using firmware currently running on NIC"
740				 ".  For optimal\n");
741			dev_warn(&mgp->pdev->dev,
742				 "performance consider loading optimized "
743				 "firmware\n");
744			dev_warn(&mgp->pdev->dev, "via hotplug\n");
745		}
746
747		set_fw_name(mgp, "adopted", false);
748		mgp->tx_boundary = 2048;
749		myri10ge_dummy_rdma(mgp, 1);
750		status = myri10ge_get_firmware_capabilities(mgp);
751		return status;
752	}
753
754	/* clear confirmation addr */
755	mgp->cmd->data = 0;
756	mb();
757
758	/* send a reload command to the bootstrap MCP, and wait for the
759	 *  response in the confirmation address.  The firmware should
760	 * write a -1 there to indicate it is alive and well
761	 */
762	dma_low = MYRI10GE_LOWPART_TO_U32(mgp->cmd_bus);
763	dma_high = MYRI10GE_HIGHPART_TO_U32(mgp->cmd_bus);
764
765	buf[0] = htonl(dma_high);	/* confirm addr MSW */
766	buf[1] = htonl(dma_low);	/* confirm addr LSW */
767	buf[2] = MYRI10GE_NO_CONFIRM_DATA;	/* confirm data */
768
769	/* FIX: All newest firmware should un-protect the bottom of
770	 * the sram before handoff. However, the very first interfaces
771	 * do not. Therefore the handoff copy must skip the first 8 bytes
772	 */
773	buf[3] = htonl(MYRI10GE_FW_OFFSET + 8);	/* where the code starts */
774	buf[4] = htonl(size - 8);	/* length of code */
775	buf[5] = htonl(8);	/* where to copy to */
776	buf[6] = htonl(0);	/* where to jump to */
777
778	submit = mgp->sram + MXGEFW_BOOT_HANDOFF;
779
780	myri10ge_pio_copy(submit, &buf, sizeof(buf));
781	mb();
782	msleep(1);
783	mb();
784	i = 0;
785	while (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA && i < 9) {
786		msleep(1 << i);
787		i++;
788	}
789	if (mgp->cmd->data != MYRI10GE_NO_CONFIRM_DATA) {
790		dev_err(&mgp->pdev->dev, "handoff failed\n");
791		return -ENXIO;
792	}
793	myri10ge_dummy_rdma(mgp, 1);
794	status = myri10ge_get_firmware_capabilities(mgp);
795
796	return status;
797}
798
799static int myri10ge_update_mac_address(struct myri10ge_priv *mgp,
800				       const u8 * addr)
801{
802	struct myri10ge_cmd cmd;
803	int status;
804
805	cmd.data0 = ((addr[0] << 24) | (addr[1] << 16)
806		     | (addr[2] << 8) | addr[3]);
807
808	cmd.data1 = ((addr[4] << 8) | (addr[5]));
809
810	status = myri10ge_send_cmd(mgp, MXGEFW_SET_MAC_ADDRESS, &cmd, 0);
811	return status;
812}
813
814static int myri10ge_change_pause(struct myri10ge_priv *mgp, int pause)
815{
816	struct myri10ge_cmd cmd;
817	int status, ctl;
818
819	ctl = pause ? MXGEFW_ENABLE_FLOW_CONTROL : MXGEFW_DISABLE_FLOW_CONTROL;
820	status = myri10ge_send_cmd(mgp, ctl, &cmd, 0);
821
822	if (status) {
823		netdev_err(mgp->dev, "Failed to set flow control mode\n");
824		return status;
825	}
826	mgp->pause = pause;
827	return 0;
828}
829
830static void
831myri10ge_change_promisc(struct myri10ge_priv *mgp, int promisc, int atomic)
832{
833	struct myri10ge_cmd cmd;
834	int status, ctl;
835
836	ctl = promisc ? MXGEFW_ENABLE_PROMISC : MXGEFW_DISABLE_PROMISC;
837	status = myri10ge_send_cmd(mgp, ctl, &cmd, atomic);
838	if (status)
839		netdev_err(mgp->dev, "Failed to set promisc mode\n");
840}
841
842static int myri10ge_dma_test(struct myri10ge_priv *mgp, int test_type)
843{
844	struct myri10ge_cmd cmd;
845	int status;
846	u32 len;
847	struct page *dmatest_page;
848	dma_addr_t dmatest_bus;
849	char *test = " ";
850
851	dmatest_page = alloc_page(GFP_KERNEL);
852	if (!dmatest_page)
853		return -ENOMEM;
854	dmatest_bus = dma_map_page(&mgp->pdev->dev, dmatest_page, 0,
855				   PAGE_SIZE, DMA_BIDIRECTIONAL);
856	if (unlikely(dma_mapping_error(&mgp->pdev->dev, dmatest_bus))) {
857		__free_page(dmatest_page);
858		return -ENOMEM;
859	}
860
861	/* Run a small DMA test.
862	 * The magic multipliers to the length tell the firmware
863	 * to do DMA read, write, or read+write tests.  The
864	 * results are returned in cmd.data0.  The upper 16
865	 * bits or the return is the number of transfers completed.
866	 * The lower 16 bits is the time in 0.5us ticks that the
867	 * transfers took to complete.
868	 */
869
870	len = mgp->tx_boundary;
871
872	cmd.data0 = MYRI10GE_LOWPART_TO_U32(dmatest_bus);
873	cmd.data1 = MYRI10GE_HIGHPART_TO_U32(dmatest_bus);
874	cmd.data2 = len * 0x10000;
875	status = myri10ge_send_cmd(mgp, test_type, &cmd, 0);
876	if (status != 0) {
877		test = "read";
878		goto abort;
879	}
880	mgp->read_dma = ((cmd.data0 >> 16) * len * 2) / (cmd.data0 & 0xffff);
881	cmd.data0 = MYRI10GE_LOWPART_TO_U32(dmatest_bus);
882	cmd.data1 = MYRI10GE_HIGHPART_TO_U32(dmatest_bus);
883	cmd.data2 = len * 0x1;
884	status = myri10ge_send_cmd(mgp, test_type, &cmd, 0);
885	if (status != 0) {
886		test = "write";
887		goto abort;
888	}
889	mgp->write_dma = ((cmd.data0 >> 16) * len * 2) / (cmd.data0 & 0xffff);
890
891	cmd.data0 = MYRI10GE_LOWPART_TO_U32(dmatest_bus);
892	cmd.data1 = MYRI10GE_HIGHPART_TO_U32(dmatest_bus);
893	cmd.data2 = len * 0x10001;
894	status = myri10ge_send_cmd(mgp, test_type, &cmd, 0);
895	if (status != 0) {
896		test = "read/write";
897		goto abort;
898	}
899	mgp->read_write_dma = ((cmd.data0 >> 16) * len * 2 * 2) /
900	    (cmd.data0 & 0xffff);
901
902abort:
903	dma_unmap_page(&mgp->pdev->dev, dmatest_bus, PAGE_SIZE,
904		       DMA_BIDIRECTIONAL);
905	put_page(dmatest_page);
906
907	if (status != 0 && test_type != MXGEFW_CMD_UNALIGNED_TEST)
908		dev_warn(&mgp->pdev->dev, "DMA %s benchmark failed: %d\n",
909			 test, status);
910
911	return status;
912}
913
914static int myri10ge_reset(struct myri10ge_priv *mgp)
915{
916	struct myri10ge_cmd cmd;
917	struct myri10ge_slice_state *ss;
918	int i, status;
919	size_t bytes;
920#ifdef CONFIG_MYRI10GE_DCA
921	unsigned long dca_tag_off;
922#endif
923
924	/* try to send a reset command to the card to see if it
925	 * is alive */
926	memset(&cmd, 0, sizeof(cmd));
927	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_RESET, &cmd, 0);
928	if (status != 0) {
929		dev_err(&mgp->pdev->dev, "failed reset\n");
930		return -ENXIO;
931	}
932
933	(void)myri10ge_dma_test(mgp, MXGEFW_DMA_TEST);
934	/*
935	 * Use non-ndis mcp_slot (eg, 4 bytes total,
936	 * no toeplitz hash value returned.  Older firmware will
937	 * not understand this command, but will use the correct
938	 * sized mcp_slot, so we ignore error returns
939	 */
940	cmd.data0 = MXGEFW_RSS_MCP_SLOT_TYPE_MIN;
941	(void)myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_RSS_MCP_SLOT_TYPE, &cmd, 0);
942
943	/* Now exchange information about interrupts  */
944
945	bytes = mgp->max_intr_slots * sizeof(*mgp->ss[0].rx_done.entry);
946	cmd.data0 = (u32) bytes;
947	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_SIZE, &cmd, 0);
948
949	/*
950	 * Even though we already know how many slices are supported
951	 * via myri10ge_probe_slices() MXGEFW_CMD_GET_MAX_RSS_QUEUES
952	 * has magic side effects, and must be called after a reset.
953	 * It must be called prior to calling any RSS related cmds,
954	 * including assigning an interrupt queue for anything but
955	 * slice 0.  It must also be called *after*
956	 * MXGEFW_CMD_SET_INTRQ_SIZE, since the intrq size is used by
957	 * the firmware to compute offsets.
958	 */
959
960	if (mgp->num_slices > 1) {
961
962		/* ask the maximum number of slices it supports */
963		status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_MAX_RSS_QUEUES,
964					   &cmd, 0);
965		if (status != 0) {
966			dev_err(&mgp->pdev->dev,
967				"failed to get number of slices\n");
968		}
969
970		/*
971		 * MXGEFW_CMD_ENABLE_RSS_QUEUES must be called prior
972		 * to setting up the interrupt queue DMA
973		 */
974
975		cmd.data0 = mgp->num_slices;
976		cmd.data1 = MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE;
977		if (mgp->dev->real_num_tx_queues > 1)
978			cmd.data1 |= MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES;
979		status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ENABLE_RSS_QUEUES,
980					   &cmd, 0);
981
982		/* Firmware older than 1.4.32 only supports multiple
983		 * RX queues, so if we get an error, first retry using a
984		 * single TX queue before giving up */
985		if (status != 0 && mgp->dev->real_num_tx_queues > 1) {
986			netif_set_real_num_tx_queues(mgp->dev, 1);
987			cmd.data0 = mgp->num_slices;
988			cmd.data1 = MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE;
989			status = myri10ge_send_cmd(mgp,
990						   MXGEFW_CMD_ENABLE_RSS_QUEUES,
991						   &cmd, 0);
992		}
993
994		if (status != 0) {
995			dev_err(&mgp->pdev->dev,
996				"failed to set number of slices\n");
997
998			return status;
999		}
1000	}
1001	for (i = 0; i < mgp->num_slices; i++) {
1002		ss = &mgp->ss[i];
1003		cmd.data0 = MYRI10GE_LOWPART_TO_U32(ss->rx_done.bus);
1004		cmd.data1 = MYRI10GE_HIGHPART_TO_U32(ss->rx_done.bus);
1005		cmd.data2 = i;
1006		status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_DMA,
1007					    &cmd, 0);
1008	}
1009
1010	status |=
1011	    myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_IRQ_ACK_OFFSET, &cmd, 0);
1012	for (i = 0; i < mgp->num_slices; i++) {
1013		ss = &mgp->ss[i];
1014		ss->irq_claim =
1015		    (__iomem __be32 *) (mgp->sram + cmd.data0 + 8 * i);
1016	}
1017	status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_IRQ_DEASSERT_OFFSET,
1018				    &cmd, 0);
1019	mgp->irq_deassert = (__iomem __be32 *) (mgp->sram + cmd.data0);
1020
1021	status |= myri10ge_send_cmd
1022	    (mgp, MXGEFW_CMD_GET_INTR_COAL_DELAY_OFFSET, &cmd, 0);
1023	mgp->intr_coal_delay_ptr = (__iomem __be32 *) (mgp->sram + cmd.data0);
1024	if (status != 0) {
1025		dev_err(&mgp->pdev->dev, "failed set interrupt parameters\n");
1026		return status;
1027	}
1028	put_be32(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
1029
1030#ifdef CONFIG_MYRI10GE_DCA
1031	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_DCA_OFFSET, &cmd, 0);
1032	dca_tag_off = cmd.data0;
1033	for (i = 0; i < mgp->num_slices; i++) {
1034		ss = &mgp->ss[i];
1035		if (status == 0) {
1036			ss->dca_tag = (__iomem __be32 *)
1037			    (mgp->sram + dca_tag_off + 4 * i);
1038		} else {
1039			ss->dca_tag = NULL;
1040		}
1041	}
1042#endif				/* CONFIG_MYRI10GE_DCA */
1043
1044	/* reset mcp/driver shared state back to 0 */
1045
1046	mgp->link_changes = 0;
1047	for (i = 0; i < mgp->num_slices; i++) {
1048		ss = &mgp->ss[i];
1049
1050		memset(ss->rx_done.entry, 0, bytes);
1051		ss->tx.req = 0;
1052		ss->tx.done = 0;
1053		ss->tx.pkt_start = 0;
1054		ss->tx.pkt_done = 0;
1055		ss->rx_big.cnt = 0;
1056		ss->rx_small.cnt = 0;
1057		ss->rx_done.idx = 0;
1058		ss->rx_done.cnt = 0;
1059		ss->tx.wake_queue = 0;
1060		ss->tx.stop_queue = 0;
1061	}
1062
1063	status = myri10ge_update_mac_address(mgp, mgp->dev->dev_addr);
1064	myri10ge_change_pause(mgp, mgp->pause);
1065	myri10ge_set_multicast_list(mgp->dev);
1066	return status;
1067}
1068
1069#ifdef CONFIG_MYRI10GE_DCA
1070static int myri10ge_toggle_relaxed(struct pci_dev *pdev, int on)
1071{
1072	int ret;
1073	u16 ctl;
1074
1075	pcie_capability_read_word(pdev, PCI_EXP_DEVCTL, &ctl);
1076
1077	ret = (ctl & PCI_EXP_DEVCTL_RELAX_EN) >> 4;
1078	if (ret != on) {
1079		ctl &= ~PCI_EXP_DEVCTL_RELAX_EN;
1080		ctl |= (on << 4);
1081		pcie_capability_write_word(pdev, PCI_EXP_DEVCTL, ctl);
1082	}
1083	return ret;
1084}
1085
1086static void
1087myri10ge_write_dca(struct myri10ge_slice_state *ss, int cpu, int tag)
1088{
1089	ss->cached_dca_tag = tag;
1090	put_be32(htonl(tag), ss->dca_tag);
1091}
1092
1093static inline void myri10ge_update_dca(struct myri10ge_slice_state *ss)
1094{
1095	int cpu = get_cpu();
1096	int tag;
1097
1098	if (cpu != ss->cpu) {
1099		tag = dca3_get_tag(&ss->mgp->pdev->dev, cpu);
1100		if (ss->cached_dca_tag != tag)
1101			myri10ge_write_dca(ss, cpu, tag);
1102		ss->cpu = cpu;
1103	}
1104	put_cpu();
1105}
1106
1107static void myri10ge_setup_dca(struct myri10ge_priv *mgp)
1108{
1109	int err, i;
1110	struct pci_dev *pdev = mgp->pdev;
1111
1112	if (mgp->ss[0].dca_tag == NULL || mgp->dca_enabled)
1113		return;
1114	if (!myri10ge_dca) {
1115		dev_err(&pdev->dev, "dca disabled by administrator\n");
1116		return;
1117	}
1118	err = dca_add_requester(&pdev->dev);
1119	if (err) {
1120		if (err != -ENODEV)
1121			dev_err(&pdev->dev,
1122				"dca_add_requester() failed, err=%d\n", err);
1123		return;
1124	}
1125	mgp->relaxed_order = myri10ge_toggle_relaxed(pdev, 0);
1126	mgp->dca_enabled = 1;
1127	for (i = 0; i < mgp->num_slices; i++) {
1128		mgp->ss[i].cpu = -1;
1129		mgp->ss[i].cached_dca_tag = -1;
1130		myri10ge_update_dca(&mgp->ss[i]);
1131	}
1132}
1133
1134static void myri10ge_teardown_dca(struct myri10ge_priv *mgp)
1135{
1136	struct pci_dev *pdev = mgp->pdev;
1137
1138	if (!mgp->dca_enabled)
1139		return;
1140	mgp->dca_enabled = 0;
1141	if (mgp->relaxed_order)
1142		myri10ge_toggle_relaxed(pdev, 1);
1143	dca_remove_requester(&pdev->dev);
1144}
1145
1146static int myri10ge_notify_dca_device(struct device *dev, void *data)
1147{
1148	struct myri10ge_priv *mgp;
1149	unsigned long event;
1150
1151	mgp = dev_get_drvdata(dev);
1152	event = *(unsigned long *)data;
1153
1154	if (event == DCA_PROVIDER_ADD)
1155		myri10ge_setup_dca(mgp);
1156	else if (event == DCA_PROVIDER_REMOVE)
1157		myri10ge_teardown_dca(mgp);
1158	return 0;
1159}
1160#endif				/* CONFIG_MYRI10GE_DCA */
1161
1162static inline void
1163myri10ge_submit_8rx(struct mcp_kreq_ether_recv __iomem * dst,
1164		    struct mcp_kreq_ether_recv *src)
1165{
1166	__be32 low;
1167
1168	low = src->addr_low;
1169	src->addr_low = htonl(DMA_BIT_MASK(32));
1170	myri10ge_pio_copy(dst, src, 4 * sizeof(*src));
1171	mb();
1172	myri10ge_pio_copy(dst + 4, src + 4, 4 * sizeof(*src));
1173	mb();
1174	src->addr_low = low;
1175	put_be32(low, &dst->addr_low);
1176	mb();
1177}
1178
1179static void
1180myri10ge_alloc_rx_pages(struct myri10ge_priv *mgp, struct myri10ge_rx_buf *rx,
1181			int bytes, int watchdog)
1182{
1183	struct page *page;
1184	dma_addr_t bus;
1185	int idx;
1186#if MYRI10GE_ALLOC_SIZE > 4096
1187	int end_offset;
1188#endif
1189
1190	if (unlikely(rx->watchdog_needed && !watchdog))
1191		return;
1192
1193	/* try to refill entire ring */
1194	while (rx->fill_cnt != (rx->cnt + rx->mask + 1)) {
1195		idx = rx->fill_cnt & rx->mask;
1196		if (rx->page_offset + bytes <= MYRI10GE_ALLOC_SIZE) {
1197			/* we can use part of previous page */
1198			get_page(rx->page);
1199		} else {
1200			/* we need a new page */
1201			page =
1202			    alloc_pages(GFP_ATOMIC | __GFP_COMP,
1203					MYRI10GE_ALLOC_ORDER);
1204			if (unlikely(page == NULL)) {
1205				if (rx->fill_cnt - rx->cnt < 16)
1206					rx->watchdog_needed = 1;
1207				return;
1208			}
1209
1210			bus = dma_map_page(&mgp->pdev->dev, page, 0,
1211					   MYRI10GE_ALLOC_SIZE,
1212					   DMA_FROM_DEVICE);
1213			if (unlikely(dma_mapping_error(&mgp->pdev->dev, bus))) {
1214				__free_pages(page, MYRI10GE_ALLOC_ORDER);
1215				if (rx->fill_cnt - rx->cnt < 16)
1216					rx->watchdog_needed = 1;
1217				return;
1218			}
1219
1220			rx->page = page;
1221			rx->page_offset = 0;
1222			rx->bus = bus;
1223
1224		}
1225		rx->info[idx].page = rx->page;
1226		rx->info[idx].page_offset = rx->page_offset;
1227		/* note that this is the address of the start of the
1228		 * page */
1229		dma_unmap_addr_set(&rx->info[idx], bus, rx->bus);
1230		rx->shadow[idx].addr_low =
1231		    htonl(MYRI10GE_LOWPART_TO_U32(rx->bus) + rx->page_offset);
1232		rx->shadow[idx].addr_high =
1233		    htonl(MYRI10GE_HIGHPART_TO_U32(rx->bus));
1234
1235		/* start next packet on a cacheline boundary */
1236		rx->page_offset += SKB_DATA_ALIGN(bytes);
1237
1238#if MYRI10GE_ALLOC_SIZE > 4096
1239		/* don't cross a 4KB boundary */
1240		end_offset = rx->page_offset + bytes - 1;
1241		if ((unsigned)(rx->page_offset ^ end_offset) > 4095)
1242			rx->page_offset = end_offset & ~4095;
1243#endif
1244		rx->fill_cnt++;
1245
1246		/* copy 8 descriptors to the firmware at a time */
1247		if ((idx & 7) == 7) {
1248			myri10ge_submit_8rx(&rx->lanai[idx - 7],
1249					    &rx->shadow[idx - 7]);
1250		}
1251	}
1252}
1253
1254static inline void
1255myri10ge_unmap_rx_page(struct pci_dev *pdev,
1256		       struct myri10ge_rx_buffer_state *info, int bytes)
1257{
1258	/* unmap the recvd page if we're the only or last user of it */
1259	if (bytes >= MYRI10GE_ALLOC_SIZE / 2 ||
1260	    (info->page_offset + 2 * bytes) > MYRI10GE_ALLOC_SIZE) {
1261		dma_unmap_page(&pdev->dev, (dma_unmap_addr(info, bus)
1262					    & ~(MYRI10GE_ALLOC_SIZE - 1)),
1263			       MYRI10GE_ALLOC_SIZE, DMA_FROM_DEVICE);
1264	}
1265}
1266
1267/*
1268 * GRO does not support acceleration of tagged vlan frames, and
1269 * this NIC does not support vlan tag offload, so we must pop
1270 * the tag ourselves to be able to achieve GRO performance that
1271 * is comparable to LRO.
1272 */
1273
1274static inline void
1275myri10ge_vlan_rx(struct net_device *dev, void *addr, struct sk_buff *skb)
1276{
1277	u8 *va;
1278	struct vlan_ethhdr *veh;
1279	skb_frag_t *frag;
1280	__wsum vsum;
1281
1282	va = addr;
1283	va += MXGEFW_PAD;
1284	veh = (struct vlan_ethhdr *)va;
1285	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) ==
1286	    NETIF_F_HW_VLAN_CTAG_RX &&
1287	    veh->h_vlan_proto == htons(ETH_P_8021Q)) {
1288		/* fixup csum if needed */
1289		if (skb->ip_summed == CHECKSUM_COMPLETE) {
1290			vsum = csum_partial(va + ETH_HLEN, VLAN_HLEN, 0);
1291			skb->csum = csum_sub(skb->csum, vsum);
1292		}
1293		/* pop tag */
1294		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(veh->h_vlan_TCI));
1295		memmove(va + VLAN_HLEN, va, 2 * ETH_ALEN);
1296		skb->len -= VLAN_HLEN;
1297		skb->data_len -= VLAN_HLEN;
1298		frag = skb_shinfo(skb)->frags;
1299		skb_frag_off_add(frag, VLAN_HLEN);
1300		skb_frag_size_sub(frag, VLAN_HLEN);
1301	}
1302}
1303
1304#define MYRI10GE_HLEN 64 /* Bytes to copy from page to skb linear memory */
1305
1306static inline int
1307myri10ge_rx_done(struct myri10ge_slice_state *ss, int len, __wsum csum)
1308{
1309	struct myri10ge_priv *mgp = ss->mgp;
1310	struct sk_buff *skb;
1311	skb_frag_t *rx_frags;
1312	struct myri10ge_rx_buf *rx;
1313	int i, idx, remainder, bytes;
1314	struct pci_dev *pdev = mgp->pdev;
1315	struct net_device *dev = mgp->dev;
1316	u8 *va;
1317
1318	if (len <= mgp->small_bytes) {
1319		rx = &ss->rx_small;
1320		bytes = mgp->small_bytes;
1321	} else {
1322		rx = &ss->rx_big;
1323		bytes = mgp->big_bytes;
1324	}
1325
1326	len += MXGEFW_PAD;
1327	idx = rx->cnt & rx->mask;
1328	va = page_address(rx->info[idx].page) + rx->info[idx].page_offset;
1329	prefetch(va);
1330
1331	skb = napi_get_frags(&ss->napi);
1332	if (unlikely(skb == NULL)) {
1333		ss->stats.rx_dropped++;
1334		for (i = 0, remainder = len; remainder > 0; i++) {
1335			myri10ge_unmap_rx_page(pdev, &rx->info[idx], bytes);
1336			put_page(rx->info[idx].page);
1337			rx->cnt++;
1338			idx = rx->cnt & rx->mask;
1339			remainder -= MYRI10GE_ALLOC_SIZE;
1340		}
1341		return 0;
1342	}
1343	rx_frags = skb_shinfo(skb)->frags;
1344	/* Fill skb_frag_t(s) with data from our receive */
1345	for (i = 0, remainder = len; remainder > 0; i++) {
1346		myri10ge_unmap_rx_page(pdev, &rx->info[idx], bytes);
1347		skb_fill_page_desc(skb, i, rx->info[idx].page,
1348				   rx->info[idx].page_offset,
1349				   remainder < MYRI10GE_ALLOC_SIZE ?
1350				   remainder : MYRI10GE_ALLOC_SIZE);
1351		rx->cnt++;
1352		idx = rx->cnt & rx->mask;
1353		remainder -= MYRI10GE_ALLOC_SIZE;
1354	}
1355
1356	/* remove padding */
1357	skb_frag_off_add(&rx_frags[0], MXGEFW_PAD);
1358	skb_frag_size_sub(&rx_frags[0], MXGEFW_PAD);
1359	len -= MXGEFW_PAD;
1360
1361	skb->len = len;
1362	skb->data_len = len;
1363	skb->truesize += len;
1364	if (dev->features & NETIF_F_RXCSUM) {
1365		skb->ip_summed = CHECKSUM_COMPLETE;
1366		skb->csum = csum;
1367	}
1368	myri10ge_vlan_rx(mgp->dev, va, skb);
1369	skb_record_rx_queue(skb, ss - &mgp->ss[0]);
1370
1371	napi_gro_frags(&ss->napi);
1372
1373	return 1;
1374}
1375
1376static inline void
1377myri10ge_tx_done(struct myri10ge_slice_state *ss, int mcp_index)
1378{
1379	struct pci_dev *pdev = ss->mgp->pdev;
1380	struct myri10ge_tx_buf *tx = &ss->tx;
1381	struct netdev_queue *dev_queue;
1382	struct sk_buff *skb;
1383	int idx, len;
1384
1385	while (tx->pkt_done != mcp_index) {
1386		idx = tx->done & tx->mask;
1387		skb = tx->info[idx].skb;
1388
1389		/* Mark as free */
1390		tx->info[idx].skb = NULL;
1391		if (tx->info[idx].last) {
1392			tx->pkt_done++;
1393			tx->info[idx].last = 0;
1394		}
1395		tx->done++;
1396		len = dma_unmap_len(&tx->info[idx], len);
1397		dma_unmap_len_set(&tx->info[idx], len, 0);
1398		if (skb) {
1399			ss->stats.tx_bytes += skb->len;
1400			ss->stats.tx_packets++;
1401			dev_consume_skb_irq(skb);
1402			if (len)
1403				dma_unmap_single(&pdev->dev,
1404						 dma_unmap_addr(&tx->info[idx],
1405								bus), len,
1406						 DMA_TO_DEVICE);
1407		} else {
1408			if (len)
1409				dma_unmap_page(&pdev->dev,
1410					       dma_unmap_addr(&tx->info[idx],
1411							      bus), len,
1412					       DMA_TO_DEVICE);
1413		}
1414	}
1415
1416	dev_queue = netdev_get_tx_queue(ss->dev, ss - ss->mgp->ss);
1417	/*
1418	 * Make a minimal effort to prevent the NIC from polling an
1419	 * idle tx queue.  If we can't get the lock we leave the queue
1420	 * active. In this case, either a thread was about to start
1421	 * using the queue anyway, or we lost a race and the NIC will
1422	 * waste some of its resources polling an inactive queue for a
1423	 * while.
1424	 */
1425
1426	if ((ss->mgp->dev->real_num_tx_queues > 1) &&
1427	    __netif_tx_trylock(dev_queue)) {
1428		if (tx->req == tx->done) {
1429			tx->queue_active = 0;
1430			put_be32(htonl(1), tx->send_stop);
1431			mb();
1432		}
1433		__netif_tx_unlock(dev_queue);
1434	}
1435
1436	/* start the queue if we've stopped it */
1437	if (netif_tx_queue_stopped(dev_queue) &&
1438	    tx->req - tx->done < (tx->mask >> 1) &&
1439	    ss->mgp->running == MYRI10GE_ETH_RUNNING) {
1440		tx->wake_queue++;
1441		netif_tx_wake_queue(dev_queue);
1442	}
1443}
1444
1445static inline int
1446myri10ge_clean_rx_done(struct myri10ge_slice_state *ss, int budget)
1447{
1448	struct myri10ge_rx_done *rx_done = &ss->rx_done;
1449	struct myri10ge_priv *mgp = ss->mgp;
1450	unsigned long rx_bytes = 0;
1451	unsigned long rx_packets = 0;
1452	unsigned long rx_ok;
1453	int idx = rx_done->idx;
1454	int cnt = rx_done->cnt;
1455	int work_done = 0;
1456	u16 length;
1457	__wsum checksum;
1458
1459	while (rx_done->entry[idx].length != 0 && work_done < budget) {
1460		length = ntohs(rx_done->entry[idx].length);
1461		rx_done->entry[idx].length = 0;
1462		checksum = csum_unfold(rx_done->entry[idx].checksum);
1463		rx_ok = myri10ge_rx_done(ss, length, checksum);
1464		rx_packets += rx_ok;
1465		rx_bytes += rx_ok * (unsigned long)length;
1466		cnt++;
1467		idx = cnt & (mgp->max_intr_slots - 1);
1468		work_done++;
1469	}
1470	rx_done->idx = idx;
1471	rx_done->cnt = cnt;
1472	ss->stats.rx_packets += rx_packets;
1473	ss->stats.rx_bytes += rx_bytes;
1474
1475	/* restock receive rings if needed */
1476	if (ss->rx_small.fill_cnt - ss->rx_small.cnt < myri10ge_fill_thresh)
1477		myri10ge_alloc_rx_pages(mgp, &ss->rx_small,
1478					mgp->small_bytes + MXGEFW_PAD, 0);
1479	if (ss->rx_big.fill_cnt - ss->rx_big.cnt < myri10ge_fill_thresh)
1480		myri10ge_alloc_rx_pages(mgp, &ss->rx_big, mgp->big_bytes, 0);
1481
1482	return work_done;
1483}
1484
1485static inline void myri10ge_check_statblock(struct myri10ge_priv *mgp)
1486{
1487	struct mcp_irq_data *stats = mgp->ss[0].fw_stats;
1488
1489	if (unlikely(stats->stats_updated)) {
1490		unsigned link_up = ntohl(stats->link_up);
1491		if (mgp->link_state != link_up) {
1492			mgp->link_state = link_up;
1493
1494			if (mgp->link_state == MXGEFW_LINK_UP) {
1495				netif_info(mgp, link, mgp->dev, "link up\n");
1496				netif_carrier_on(mgp->dev);
1497				mgp->link_changes++;
1498			} else {
1499				netif_info(mgp, link, mgp->dev, "link %s\n",
1500					   (link_up == MXGEFW_LINK_MYRINET ?
1501					    "mismatch (Myrinet detected)" :
1502					    "down"));
1503				netif_carrier_off(mgp->dev);
1504				mgp->link_changes++;
1505			}
1506		}
1507		if (mgp->rdma_tags_available !=
1508		    ntohl(stats->rdma_tags_available)) {
1509			mgp->rdma_tags_available =
1510			    ntohl(stats->rdma_tags_available);
1511			netdev_warn(mgp->dev, "RDMA timed out! %d tags left\n",
1512				    mgp->rdma_tags_available);
1513		}
1514		mgp->down_cnt += stats->link_down;
1515		if (stats->link_down)
1516			wake_up(&mgp->down_wq);
1517	}
1518}
1519
1520static int myri10ge_poll(struct napi_struct *napi, int budget)
1521{
1522	struct myri10ge_slice_state *ss =
1523	    container_of(napi, struct myri10ge_slice_state, napi);
1524	int work_done;
1525
1526#ifdef CONFIG_MYRI10GE_DCA
1527	if (ss->mgp->dca_enabled)
1528		myri10ge_update_dca(ss);
1529#endif
1530	/* process as many rx events as NAPI will allow */
1531	work_done = myri10ge_clean_rx_done(ss, budget);
1532
1533	if (work_done < budget) {
1534		napi_complete_done(napi, work_done);
1535		put_be32(htonl(3), ss->irq_claim);
1536	}
1537	return work_done;
1538}
1539
1540static irqreturn_t myri10ge_intr(int irq, void *arg)
1541{
1542	struct myri10ge_slice_state *ss = arg;
1543	struct myri10ge_priv *mgp = ss->mgp;
1544	struct mcp_irq_data *stats = ss->fw_stats;
1545	struct myri10ge_tx_buf *tx = &ss->tx;
1546	u32 send_done_count;
1547	int i;
1548
1549	/* an interrupt on a non-zero receive-only slice is implicitly
1550	 * valid  since MSI-X irqs are not shared */
1551	if ((mgp->dev->real_num_tx_queues == 1) && (ss != mgp->ss)) {
1552		napi_schedule(&ss->napi);
1553		return IRQ_HANDLED;
1554	}
1555
1556	/* make sure it is our IRQ, and that the DMA has finished */
1557	if (unlikely(!stats->valid))
1558		return IRQ_NONE;
1559
1560	/* low bit indicates receives are present, so schedule
1561	 * napi poll handler */
1562	if (stats->valid & 1)
1563		napi_schedule(&ss->napi);
1564
1565	if (!mgp->msi_enabled && !mgp->msix_enabled) {
1566		put_be32(0, mgp->irq_deassert);
1567		if (!myri10ge_deassert_wait)
1568			stats->valid = 0;
1569		mb();
1570	} else
1571		stats->valid = 0;
1572
1573	/* Wait for IRQ line to go low, if using INTx */
1574	i = 0;
1575	while (1) {
1576		i++;
1577		/* check for transmit completes and receives */
1578		send_done_count = ntohl(stats->send_done_count);
1579		if (send_done_count != tx->pkt_done)
1580			myri10ge_tx_done(ss, (int)send_done_count);
1581		if (unlikely(i > myri10ge_max_irq_loops)) {
1582			netdev_warn(mgp->dev, "irq stuck?\n");
1583			stats->valid = 0;
1584			schedule_work(&mgp->watchdog_work);
1585		}
1586		if (likely(stats->valid == 0))
1587			break;
1588		cpu_relax();
1589		barrier();
1590	}
1591
1592	/* Only slice 0 updates stats */
1593	if (ss == mgp->ss)
1594		myri10ge_check_statblock(mgp);
1595
1596	put_be32(htonl(3), ss->irq_claim + 1);
1597	return IRQ_HANDLED;
1598}
1599
1600static int
1601myri10ge_get_link_ksettings(struct net_device *netdev,
1602			    struct ethtool_link_ksettings *cmd)
1603{
1604	struct myri10ge_priv *mgp = netdev_priv(netdev);
1605	char *ptr;
1606	int i;
1607
1608	cmd->base.autoneg = AUTONEG_DISABLE;
1609	cmd->base.speed = SPEED_10000;
1610	cmd->base.duplex = DUPLEX_FULL;
1611
1612	/*
1613	 * parse the product code to deterimine the interface type
1614	 * (CX4, XFP, Quad Ribbon Fiber) by looking at the character
1615	 * after the 3rd dash in the driver's cached copy of the
1616	 * EEPROM's product code string.
1617	 */
1618	ptr = mgp->product_code_string;
1619	if (ptr == NULL) {
1620		netdev_err(netdev, "Missing product code\n");
1621		return 0;
1622	}
1623	for (i = 0; i < 3; i++, ptr++) {
1624		ptr = strchr(ptr, '-');
1625		if (ptr == NULL) {
1626			netdev_err(netdev, "Invalid product code %s\n",
1627				   mgp->product_code_string);
1628			return 0;
1629		}
1630	}
1631	if (*ptr == '2')
1632		ptr++;
1633	if (*ptr == 'R' || *ptr == 'Q' || *ptr == 'S') {
1634		/* We've found either an XFP, quad ribbon fiber, or SFP+ */
1635		cmd->base.port = PORT_FIBRE;
1636		ethtool_link_ksettings_add_link_mode(cmd, supported, FIBRE);
1637		ethtool_link_ksettings_add_link_mode(cmd, advertising, FIBRE);
1638	} else {
1639		cmd->base.port = PORT_OTHER;
1640	}
1641
1642	return 0;
1643}
1644
1645static void
1646myri10ge_get_drvinfo(struct net_device *netdev, struct ethtool_drvinfo *info)
1647{
1648	struct myri10ge_priv *mgp = netdev_priv(netdev);
1649
1650	strscpy(info->driver, "myri10ge", sizeof(info->driver));
1651	strscpy(info->version, MYRI10GE_VERSION_STR, sizeof(info->version));
1652	strscpy(info->fw_version, mgp->fw_version, sizeof(info->fw_version));
1653	strscpy(info->bus_info, pci_name(mgp->pdev), sizeof(info->bus_info));
1654}
1655
1656static int myri10ge_get_coalesce(struct net_device *netdev,
1657				 struct ethtool_coalesce *coal,
1658				 struct kernel_ethtool_coalesce *kernel_coal,
1659				 struct netlink_ext_ack *extack)
1660{
1661	struct myri10ge_priv *mgp = netdev_priv(netdev);
1662
1663	coal->rx_coalesce_usecs = mgp->intr_coal_delay;
1664	return 0;
1665}
1666
1667static int myri10ge_set_coalesce(struct net_device *netdev,
1668				 struct ethtool_coalesce *coal,
1669				 struct kernel_ethtool_coalesce *kernel_coal,
1670				 struct netlink_ext_ack *extack)
1671{
1672	struct myri10ge_priv *mgp = netdev_priv(netdev);
1673
1674	mgp->intr_coal_delay = coal->rx_coalesce_usecs;
1675	put_be32(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
1676	return 0;
1677}
1678
1679static void
1680myri10ge_get_pauseparam(struct net_device *netdev,
1681			struct ethtool_pauseparam *pause)
1682{
1683	struct myri10ge_priv *mgp = netdev_priv(netdev);
1684
1685	pause->autoneg = 0;
1686	pause->rx_pause = mgp->pause;
1687	pause->tx_pause = mgp->pause;
1688}
1689
1690static int
1691myri10ge_set_pauseparam(struct net_device *netdev,
1692			struct ethtool_pauseparam *pause)
1693{
1694	struct myri10ge_priv *mgp = netdev_priv(netdev);
1695
1696	if (pause->tx_pause != mgp->pause)
1697		return myri10ge_change_pause(mgp, pause->tx_pause);
1698	if (pause->rx_pause != mgp->pause)
1699		return myri10ge_change_pause(mgp, pause->rx_pause);
1700	if (pause->autoneg != 0)
1701		return -EINVAL;
1702	return 0;
1703}
1704
1705static void
1706myri10ge_get_ringparam(struct net_device *netdev,
1707		       struct ethtool_ringparam *ring,
1708		       struct kernel_ethtool_ringparam *kernel_ring,
1709		       struct netlink_ext_ack *extack)
1710{
1711	struct myri10ge_priv *mgp = netdev_priv(netdev);
1712
1713	ring->rx_mini_max_pending = mgp->ss[0].rx_small.mask + 1;
1714	ring->rx_max_pending = mgp->ss[0].rx_big.mask + 1;
1715	ring->rx_jumbo_max_pending = 0;
1716	ring->tx_max_pending = mgp->ss[0].tx.mask + 1;
1717	ring->rx_mini_pending = ring->rx_mini_max_pending;
1718	ring->rx_pending = ring->rx_max_pending;
1719	ring->rx_jumbo_pending = ring->rx_jumbo_max_pending;
1720	ring->tx_pending = ring->tx_max_pending;
1721}
1722
1723static const char myri10ge_gstrings_main_stats[][ETH_GSTRING_LEN] = {
1724	"rx_packets", "tx_packets", "rx_bytes", "tx_bytes", "rx_errors",
1725	"tx_errors", "rx_dropped", "tx_dropped", "multicast", "collisions",
1726	"rx_length_errors", "rx_over_errors", "rx_crc_errors",
1727	"rx_frame_errors", "rx_fifo_errors", "rx_missed_errors",
1728	"tx_aborted_errors", "tx_carrier_errors", "tx_fifo_errors",
1729	"tx_heartbeat_errors", "tx_window_errors",
1730	/* device-specific stats */
1731	"tx_boundary", "irq", "MSI", "MSIX",
1732	"read_dma_bw_MBs", "write_dma_bw_MBs", "read_write_dma_bw_MBs",
1733	"serial_number", "watchdog_resets",
1734#ifdef CONFIG_MYRI10GE_DCA
1735	"dca_capable_firmware", "dca_device_present",
1736#endif
1737	"link_changes", "link_up", "dropped_link_overflow",
1738	"dropped_link_error_or_filtered",
1739	"dropped_pause", "dropped_bad_phy", "dropped_bad_crc32",
1740	"dropped_unicast_filtered", "dropped_multicast_filtered",
1741	"dropped_runt", "dropped_overrun", "dropped_no_small_buffer",
1742	"dropped_no_big_buffer"
1743};
1744
1745static const char myri10ge_gstrings_slice_stats[][ETH_GSTRING_LEN] = {
1746	"----------- slice ---------",
1747	"tx_pkt_start", "tx_pkt_done", "tx_req", "tx_done",
1748	"rx_small_cnt", "rx_big_cnt",
1749	"wake_queue", "stop_queue", "tx_linearized",
1750};
1751
1752#define MYRI10GE_NET_STATS_LEN      21
1753#define MYRI10GE_MAIN_STATS_LEN  ARRAY_SIZE(myri10ge_gstrings_main_stats)
1754#define MYRI10GE_SLICE_STATS_LEN  ARRAY_SIZE(myri10ge_gstrings_slice_stats)
1755
1756static void
1757myri10ge_get_strings(struct net_device *netdev, u32 stringset, u8 * data)
1758{
1759	struct myri10ge_priv *mgp = netdev_priv(netdev);
1760	int i;
1761
1762	switch (stringset) {
1763	case ETH_SS_STATS:
1764		memcpy(data, *myri10ge_gstrings_main_stats,
1765		       sizeof(myri10ge_gstrings_main_stats));
1766		data += sizeof(myri10ge_gstrings_main_stats);
1767		for (i = 0; i < mgp->num_slices; i++) {
1768			memcpy(data, *myri10ge_gstrings_slice_stats,
1769			       sizeof(myri10ge_gstrings_slice_stats));
1770			data += sizeof(myri10ge_gstrings_slice_stats);
1771		}
1772		break;
1773	}
1774}
1775
1776static int myri10ge_get_sset_count(struct net_device *netdev, int sset)
1777{
1778	struct myri10ge_priv *mgp = netdev_priv(netdev);
1779
1780	switch (sset) {
1781	case ETH_SS_STATS:
1782		return MYRI10GE_MAIN_STATS_LEN +
1783		    mgp->num_slices * MYRI10GE_SLICE_STATS_LEN;
1784	default:
1785		return -EOPNOTSUPP;
1786	}
1787}
1788
1789static void
1790myri10ge_get_ethtool_stats(struct net_device *netdev,
1791			   struct ethtool_stats *stats, u64 * data)
1792{
1793	struct myri10ge_priv *mgp = netdev_priv(netdev);
1794	struct myri10ge_slice_state *ss;
1795	struct rtnl_link_stats64 link_stats;
1796	int slice;
1797	int i;
1798
1799	/* force stats update */
1800	memset(&link_stats, 0, sizeof(link_stats));
1801	(void)myri10ge_get_stats(netdev, &link_stats);
1802	for (i = 0; i < MYRI10GE_NET_STATS_LEN; i++)
1803		data[i] = ((u64 *)&link_stats)[i];
1804
1805	data[i++] = (unsigned int)mgp->tx_boundary;
1806	data[i++] = (unsigned int)mgp->pdev->irq;
1807	data[i++] = (unsigned int)mgp->msi_enabled;
1808	data[i++] = (unsigned int)mgp->msix_enabled;
1809	data[i++] = (unsigned int)mgp->read_dma;
1810	data[i++] = (unsigned int)mgp->write_dma;
1811	data[i++] = (unsigned int)mgp->read_write_dma;
1812	data[i++] = (unsigned int)mgp->serial_number;
1813	data[i++] = (unsigned int)mgp->watchdog_resets;
1814#ifdef CONFIG_MYRI10GE_DCA
1815	data[i++] = (unsigned int)(mgp->ss[0].dca_tag != NULL);
1816	data[i++] = (unsigned int)(mgp->dca_enabled);
1817#endif
1818	data[i++] = (unsigned int)mgp->link_changes;
1819
1820	/* firmware stats are useful only in the first slice */
1821	ss = &mgp->ss[0];
1822	data[i++] = (unsigned int)ntohl(ss->fw_stats->link_up);
1823	data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_link_overflow);
1824	data[i++] =
1825	    (unsigned int)ntohl(ss->fw_stats->dropped_link_error_or_filtered);
1826	data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_pause);
1827	data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_bad_phy);
1828	data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_bad_crc32);
1829	data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_unicast_filtered);
1830	data[i++] =
1831	    (unsigned int)ntohl(ss->fw_stats->dropped_multicast_filtered);
1832	data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_runt);
1833	data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_overrun);
1834	data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_no_small_buffer);
1835	data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_no_big_buffer);
1836
1837	for (slice = 0; slice < mgp->num_slices; slice++) {
1838		ss = &mgp->ss[slice];
1839		data[i++] = slice;
1840		data[i++] = (unsigned int)ss->tx.pkt_start;
1841		data[i++] = (unsigned int)ss->tx.pkt_done;
1842		data[i++] = (unsigned int)ss->tx.req;
1843		data[i++] = (unsigned int)ss->tx.done;
1844		data[i++] = (unsigned int)ss->rx_small.cnt;
1845		data[i++] = (unsigned int)ss->rx_big.cnt;
1846		data[i++] = (unsigned int)ss->tx.wake_queue;
1847		data[i++] = (unsigned int)ss->tx.stop_queue;
1848		data[i++] = (unsigned int)ss->tx.linearized;
1849	}
1850}
1851
1852static void myri10ge_set_msglevel(struct net_device *netdev, u32 value)
1853{
1854	struct myri10ge_priv *mgp = netdev_priv(netdev);
1855	mgp->msg_enable = value;
1856}
1857
1858static u32 myri10ge_get_msglevel(struct net_device *netdev)
1859{
1860	struct myri10ge_priv *mgp = netdev_priv(netdev);
1861	return mgp->msg_enable;
1862}
1863
1864/*
1865 * Use a low-level command to change the LED behavior. Rather than
1866 * blinking (which is the normal case), when identify is used, the
1867 * yellow LED turns solid.
1868 */
1869static int myri10ge_led(struct myri10ge_priv *mgp, int on)
1870{
1871	struct mcp_gen_header *hdr;
1872	struct device *dev = &mgp->pdev->dev;
1873	size_t hdr_off, pattern_off, hdr_len;
1874	u32 pattern = 0xfffffffe;
1875
1876	/* find running firmware header */
1877	hdr_off = swab32(readl(mgp->sram + MCP_HEADER_PTR_OFFSET));
1878	if ((hdr_off & 3) || hdr_off + sizeof(*hdr) > mgp->sram_size) {
1879		dev_err(dev, "Running firmware has bad header offset (%d)\n",
1880			(int)hdr_off);
1881		return -EIO;
1882	}
1883	hdr_len = swab32(readl(mgp->sram + hdr_off +
1884			       offsetof(struct mcp_gen_header, header_length)));
1885	pattern_off = hdr_off + offsetof(struct mcp_gen_header, led_pattern);
1886	if (pattern_off >= (hdr_len + hdr_off)) {
1887		dev_info(dev, "Firmware does not support LED identification\n");
1888		return -EINVAL;
1889	}
1890	if (!on)
1891		pattern = swab32(readl(mgp->sram + pattern_off + 4));
1892	writel(swab32(pattern), mgp->sram + pattern_off);
1893	return 0;
1894}
1895
1896static int
1897myri10ge_phys_id(struct net_device *netdev, enum ethtool_phys_id_state state)
1898{
1899	struct myri10ge_priv *mgp = netdev_priv(netdev);
1900	int rc;
1901
1902	switch (state) {
1903	case ETHTOOL_ID_ACTIVE:
1904		rc = myri10ge_led(mgp, 1);
1905		break;
1906
1907	case ETHTOOL_ID_INACTIVE:
1908		rc =  myri10ge_led(mgp, 0);
1909		break;
1910
1911	default:
1912		rc = -EINVAL;
1913	}
1914
1915	return rc;
1916}
1917
1918static const struct ethtool_ops myri10ge_ethtool_ops = {
1919	.supported_coalesce_params = ETHTOOL_COALESCE_RX_USECS,
1920	.get_drvinfo = myri10ge_get_drvinfo,
1921	.get_coalesce = myri10ge_get_coalesce,
1922	.set_coalesce = myri10ge_set_coalesce,
1923	.get_pauseparam = myri10ge_get_pauseparam,
1924	.set_pauseparam = myri10ge_set_pauseparam,
1925	.get_ringparam = myri10ge_get_ringparam,
1926	.get_link = ethtool_op_get_link,
1927	.get_strings = myri10ge_get_strings,
1928	.get_sset_count = myri10ge_get_sset_count,
1929	.get_ethtool_stats = myri10ge_get_ethtool_stats,
1930	.set_msglevel = myri10ge_set_msglevel,
1931	.get_msglevel = myri10ge_get_msglevel,
1932	.set_phys_id = myri10ge_phys_id,
1933	.get_link_ksettings = myri10ge_get_link_ksettings,
1934};
1935
1936static int myri10ge_allocate_rings(struct myri10ge_slice_state *ss)
1937{
1938	struct myri10ge_priv *mgp = ss->mgp;
1939	struct myri10ge_cmd cmd;
1940	struct net_device *dev = mgp->dev;
1941	int tx_ring_size, rx_ring_size;
1942	int tx_ring_entries, rx_ring_entries;
1943	int i, slice, status;
1944	size_t bytes;
1945
1946	/* get ring sizes */
1947	slice = ss - mgp->ss;
1948	cmd.data0 = slice;
1949	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_RING_SIZE, &cmd, 0);
1950	tx_ring_size = cmd.data0;
1951	cmd.data0 = slice;
1952	status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_RX_RING_SIZE, &cmd, 0);
1953	if (status != 0)
1954		return status;
1955	rx_ring_size = cmd.data0;
1956
1957	tx_ring_entries = tx_ring_size / sizeof(struct mcp_kreq_ether_send);
1958	rx_ring_entries = rx_ring_size / sizeof(struct mcp_dma_addr);
1959	ss->tx.mask = tx_ring_entries - 1;
1960	ss->rx_small.mask = ss->rx_big.mask = rx_ring_entries - 1;
1961
1962	status = -ENOMEM;
1963
1964	/* allocate the host shadow rings */
1965
1966	bytes = 8 + (MYRI10GE_MAX_SEND_DESC_TSO + 4)
1967	    * sizeof(*ss->tx.req_list);
1968	ss->tx.req_bytes = kzalloc(bytes, GFP_KERNEL);
1969	if (ss->tx.req_bytes == NULL)
1970		goto abort_with_nothing;
1971
1972	/* ensure req_list entries are aligned to 8 bytes */
1973	ss->tx.req_list = (struct mcp_kreq_ether_send *)
1974	    ALIGN((unsigned long)ss->tx.req_bytes, 8);
1975	ss->tx.queue_active = 0;
1976
1977	bytes = rx_ring_entries * sizeof(*ss->rx_small.shadow);
1978	ss->rx_small.shadow = kzalloc(bytes, GFP_KERNEL);
1979	if (ss->rx_small.shadow == NULL)
1980		goto abort_with_tx_req_bytes;
1981
1982	bytes = rx_ring_entries * sizeof(*ss->rx_big.shadow);
1983	ss->rx_big.shadow = kzalloc(bytes, GFP_KERNEL);
1984	if (ss->rx_big.shadow == NULL)
1985		goto abort_with_rx_small_shadow;
1986
1987	/* allocate the host info rings */
1988
1989	bytes = tx_ring_entries * sizeof(*ss->tx.info);
1990	ss->tx.info = kzalloc(bytes, GFP_KERNEL);
1991	if (ss->tx.info == NULL)
1992		goto abort_with_rx_big_shadow;
1993
1994	bytes = rx_ring_entries * sizeof(*ss->rx_small.info);
1995	ss->rx_small.info = kzalloc(bytes, GFP_KERNEL);
1996	if (ss->rx_small.info == NULL)
1997		goto abort_with_tx_info;
1998
1999	bytes = rx_ring_entries * sizeof(*ss->rx_big.info);
2000	ss->rx_big.info = kzalloc(bytes, GFP_KERNEL);
2001	if (ss->rx_big.info == NULL)
2002		goto abort_with_rx_small_info;
2003
2004	/* Fill the receive rings */
2005	ss->rx_big.cnt = 0;
2006	ss->rx_small.cnt = 0;
2007	ss->rx_big.fill_cnt = 0;
2008	ss->rx_small.fill_cnt = 0;
2009	ss->rx_small.page_offset = MYRI10GE_ALLOC_SIZE;
2010	ss->rx_big.page_offset = MYRI10GE_ALLOC_SIZE;
2011	ss->rx_small.watchdog_needed = 0;
2012	ss->rx_big.watchdog_needed = 0;
2013	if (mgp->small_bytes == 0) {
2014		ss->rx_small.fill_cnt = ss->rx_small.mask + 1;
2015	} else {
2016		myri10ge_alloc_rx_pages(mgp, &ss->rx_small,
2017					mgp->small_bytes + MXGEFW_PAD, 0);
2018	}
2019
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;
2024	}
2025
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;
2031	}
2032
2033	return 0;
2034
2035abort_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],
2039				       mgp->big_bytes);
2040		put_page(ss->rx_big.info[idx].page);
2041	}
2042
2043abort_with_rx_small_ring:
2044	if (mgp->small_bytes == 0)
2045		ss->rx_small.fill_cnt = ss->rx_small.cnt;
2046	for (i = ss->rx_small.cnt; i < ss->rx_small.fill_cnt; i++) {
2047		int idx = i & ss->rx_small.mask;
2048		myri10ge_unmap_rx_page(mgp->pdev, &ss->rx_small.info[idx],
2049				       mgp->small_bytes + MXGEFW_PAD);
2050		put_page(ss->rx_small.info[idx].page);
2051	}
2052
2053	kfree(ss->rx_big.info);
2054
2055abort_with_rx_small_info:
2056	kfree(ss->rx_small.info);
2057
2058abort_with_tx_info:
2059	kfree(ss->tx.info);
2060
2061abort_with_rx_big_shadow:
2062	kfree(ss->rx_big.shadow);
2063
2064abort_with_rx_small_shadow:
2065	kfree(ss->rx_small.shadow);
2066
2067abort_with_tx_req_bytes:
2068	kfree(ss->tx.req_bytes);
2069	ss->tx.req_bytes = NULL;
2070	ss->tx.req_list = NULL;
2071
2072abort_with_nothing:
2073	return status;
2074}
2075
2076static void myri10ge_free_rings(struct myri10ge_slice_state *ss)
2077{
2078	struct myri10ge_priv *mgp = ss->mgp;
2079	struct sk_buff *skb;
2080	struct myri10ge_tx_buf *tx;
2081	int i, len, idx;
2082
2083	/* If not allocated, skip it */
2084	if (ss->tx.req_list == NULL)
2085		return;
2086
2087	for (i = ss->rx_big.cnt; i < ss->rx_big.fill_cnt; i++) {
2088		idx = i & ss->rx_big.mask;
2089		if (i == ss->rx_big.fill_cnt - 1)
2090			ss->rx_big.info[idx].page_offset = MYRI10GE_ALLOC_SIZE;
2091		myri10ge_unmap_rx_page(mgp->pdev, &ss->rx_big.info[idx],
2092				       mgp->big_bytes);
2093		put_page(ss->rx_big.info[idx].page);
2094	}
2095
2096	if (mgp->small_bytes == 0)
2097		ss->rx_small.fill_cnt = ss->rx_small.cnt;
2098	for (i = ss->rx_small.cnt; i < ss->rx_small.fill_cnt; i++) {
2099		idx = i & ss->rx_small.mask;
2100		if (i == ss->rx_small.fill_cnt - 1)
2101			ss->rx_small.info[idx].page_offset =
2102			    MYRI10GE_ALLOC_SIZE;
2103		myri10ge_unmap_rx_page(mgp->pdev, &ss->rx_small.info[idx],
2104				       mgp->small_bytes + MXGEFW_PAD);
2105		put_page(ss->rx_small.info[idx].page);
2106	}
2107	tx = &ss->tx;
2108	while (tx->done != tx->req) {
2109		idx = tx->done & tx->mask;
2110		skb = tx->info[idx].skb;
2111
2112		/* Mark as free */
2113		tx->info[idx].skb = NULL;
2114		tx->done++;
2115		len = dma_unmap_len(&tx->info[idx], len);
2116		dma_unmap_len_set(&tx->info[idx], len, 0);
2117		if (skb) {
2118			ss->stats.tx_dropped++;
2119			dev_kfree_skb_any(skb);
2120			if (len)
2121				dma_unmap_single(&mgp->pdev->dev,
2122						 dma_unmap_addr(&tx->info[idx],
2123								bus), len,
2124						 DMA_TO_DEVICE);
2125		} else {
2126			if (len)
2127				dma_unmap_page(&mgp->pdev->dev,
2128					       dma_unmap_addr(&tx->info[idx],
2129							      bus), len,
2130					       DMA_TO_DEVICE);
2131		}
2132	}
2133	kfree(ss->rx_big.info);
2134
2135	kfree(ss->rx_small.info);
2136
2137	kfree(ss->tx.info);
2138
2139	kfree(ss->rx_big.shadow);
2140
2141	kfree(ss->rx_small.shadow);
2142
2143	kfree(ss->tx.req_bytes);
2144	ss->tx.req_bytes = NULL;
2145	ss->tx.req_list = NULL;
2146}
2147
2148static int myri10ge_request_irq(struct myri10ge_priv *mgp)
2149{
2150	struct pci_dev *pdev = mgp->pdev;
2151	struct myri10ge_slice_state *ss;
2152	struct net_device *netdev = mgp->dev;
2153	int i;
2154	int status;
2155
2156	mgp->msi_enabled = 0;
2157	mgp->msix_enabled = 0;
2158	status = 0;
2159	if (myri10ge_msi) {
2160		if (mgp->num_slices > 1) {
2161			status = pci_enable_msix_range(pdev, mgp->msix_vectors,
2162					mgp->num_slices, mgp->num_slices);
2163			if (status < 0) {
2164				dev_err(&pdev->dev,
2165					"Error %d setting up MSI-X\n", status);
2166				return status;
2167			}
2168			mgp->msix_enabled = 1;
2169		}
2170		if (mgp->msix_enabled == 0) {
2171			status = pci_enable_msi(pdev);
2172			if (status != 0) {
2173				dev_err(&pdev->dev,
2174					"Error %d setting up MSI; falling back to xPIC\n",
2175					status);
2176			} else {
2177				mgp->msi_enabled = 1;
2178			}
2179		}
2180	}
2181	if (mgp->msix_enabled) {
2182		for (i = 0; i < mgp->num_slices; i++) {
2183			ss = &mgp->ss[i];
2184			snprintf(ss->irq_desc, sizeof(ss->irq_desc),
2185				 "%s:slice-%d", netdev->name, i);
2186			status = request_irq(mgp->msix_vectors[i].vector,
2187					     myri10ge_intr, 0, ss->irq_desc,
2188					     ss);
2189			if (status != 0) {
2190				dev_err(&pdev->dev,
2191					"slice %d failed to allocate IRQ\n", i);
2192				i--;
2193				while (i >= 0) {
2194					free_irq(mgp->msix_vectors[i].vector,
2195						 &mgp->ss[i]);
2196					i--;
2197				}
2198				pci_disable_msix(pdev);
2199				return status;
2200			}
2201		}
2202	} else {
2203		status = request_irq(pdev->irq, myri10ge_intr, IRQF_SHARED,
2204				     mgp->dev->name, &mgp->ss[0]);
2205		if (status != 0) {
2206			dev_err(&pdev->dev, "failed to allocate IRQ\n");
2207			if (mgp->msi_enabled)
2208				pci_disable_msi(pdev);
2209		}
2210	}
2211	return status;
2212}
2213
2214static void myri10ge_free_irq(struct myri10ge_priv *mgp)
2215{
2216	struct pci_dev *pdev = mgp->pdev;
2217	int i;
2218
2219	if (mgp->msix_enabled) {
2220		for (i = 0; i < mgp->num_slices; i++)
2221			free_irq(mgp->msix_vectors[i].vector, &mgp->ss[i]);
2222	} else {
2223		free_irq(pdev->irq, &mgp->ss[0]);
2224	}
2225	if (mgp->msi_enabled)
2226		pci_disable_msi(pdev);
2227	if (mgp->msix_enabled)
2228		pci_disable_msix(pdev);
2229}
2230
2231static int myri10ge_get_txrx(struct myri10ge_priv *mgp, int slice)
2232{
2233	struct myri10ge_cmd cmd;
2234	struct myri10ge_slice_state *ss;
2235	int status;
2236
2237	ss = &mgp->ss[slice];
2238	status = 0;
2239	if (slice == 0 || (mgp->dev->real_num_tx_queues > 1)) {
2240		cmd.data0 = slice;
2241		status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_OFFSET,
2242					   &cmd, 0);
2243		ss->tx.lanai = (struct mcp_kreq_ether_send __iomem *)
2244		    (mgp->sram + cmd.data0);
2245	}
2246	cmd.data0 = slice;
2247	status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SMALL_RX_OFFSET,
2248				    &cmd, 0);
2249	ss->rx_small.lanai = (struct mcp_kreq_ether_recv __iomem *)
2250	    (mgp->sram + cmd.data0);
2251
2252	cmd.data0 = slice;
2253	status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_BIG_RX_OFFSET, &cmd, 0);
2254	ss->rx_big.lanai = (struct mcp_kreq_ether_recv __iomem *)
2255	    (mgp->sram + cmd.data0);
2256
2257	ss->tx.send_go = (__iomem __be32 *)
2258	    (mgp->sram + MXGEFW_ETH_SEND_GO + 64 * slice);
2259	ss->tx.send_stop = (__iomem __be32 *)
2260	    (mgp->sram + MXGEFW_ETH_SEND_STOP + 64 * slice);
2261	return status;
2262
2263}
2264
2265static int myri10ge_set_stats(struct myri10ge_priv *mgp, int slice)
2266{
2267	struct myri10ge_cmd cmd;
2268	struct myri10ge_slice_state *ss;
2269	int status;
2270
2271	ss = &mgp->ss[slice];
2272	cmd.data0 = MYRI10GE_LOWPART_TO_U32(ss->fw_stats_bus);
2273	cmd.data1 = MYRI10GE_HIGHPART_TO_U32(ss->fw_stats_bus);
2274	cmd.data2 = sizeof(struct mcp_irq_data) | (slice << 16);
2275	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_STATS_DMA_V2, &cmd, 0);
2276	if (status == -ENOSYS) {
2277		dma_addr_t bus = ss->fw_stats_bus;
2278		if (slice != 0)
2279			return -EINVAL;
2280		bus += offsetof(struct mcp_irq_data, send_done_count);
2281		cmd.data0 = MYRI10GE_LOWPART_TO_U32(bus);
2282		cmd.data1 = MYRI10GE_HIGHPART_TO_U32(bus);
2283		status = myri10ge_send_cmd(mgp,
2284					   MXGEFW_CMD_SET_STATS_DMA_OBSOLETE,
2285					   &cmd, 0);
2286		/* Firmware cannot support multicast without STATS_DMA_V2 */
2287		mgp->fw_multicast_support = 0;
2288	} else {
2289		mgp->fw_multicast_support = 1;
2290	}
2291	return 0;
2292}
2293
2294static int myri10ge_open(struct net_device *dev)
2295{
2296	struct myri10ge_slice_state *ss;
2297	struct myri10ge_priv *mgp = netdev_priv(dev);
2298	struct myri10ge_cmd cmd;
2299	int i, status, big_pow2, slice;
2300	u8 __iomem *itable;
2301
2302	if (mgp->running != MYRI10GE_ETH_STOPPED)
2303		return -EBUSY;
2304
2305	mgp->running = MYRI10GE_ETH_STARTING;
2306	status = myri10ge_reset(mgp);
2307	if (status != 0) {
2308		netdev_err(dev, "failed reset\n");
2309		goto abort_with_nothing;
2310	}
2311
2312	if (mgp->num_slices > 1) {
2313		cmd.data0 = mgp->num_slices;
2314		cmd.data1 = MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE;
2315		if (mgp->dev->real_num_tx_queues > 1)
2316			cmd.data1 |= MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES;
2317		status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ENABLE_RSS_QUEUES,
2318					   &cmd, 0);
2319		if (status != 0) {
2320			netdev_err(dev, "failed to set number of slices\n");
2321			goto abort_with_nothing;
2322		}
2323		/* setup the indirection table */
2324		cmd.data0 = mgp->num_slices;
2325		status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_RSS_TABLE_SIZE,
2326					   &cmd, 0);
2327
2328		status |= myri10ge_send_cmd(mgp,
2329					    MXGEFW_CMD_GET_RSS_TABLE_OFFSET,
2330					    &cmd, 0);
2331		if (status != 0) {
2332			netdev_err(dev, "failed to setup rss tables\n");
2333			goto abort_with_nothing;
2334		}
2335
2336		/* just enable an identity mapping */
2337		itable = mgp->sram + cmd.data0;
2338		for (i = 0; i < mgp->num_slices; i++)
2339			__raw_writeb(i, &itable[i]);
2340
2341		cmd.data0 = 1;
2342		cmd.data1 = myri10ge_rss_hash;
2343		status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_RSS_ENABLE,
2344					   &cmd, 0);
2345		if (status != 0) {
2346			netdev_err(dev, "failed to enable slices\n");
2347			goto abort_with_nothing;
2348		}
2349	}
2350
2351	status = myri10ge_request_irq(mgp);
2352	if (status != 0)
2353		goto abort_with_nothing;
2354
2355	/* decide what small buffer size to use.  For good TCP rx
2356	 * performance, it is important to not receive 1514 byte
2357	 * frames into jumbo buffers, as it confuses the socket buffer
2358	 * accounting code, leading to drops and erratic performance.
2359	 */
2360
2361	if (dev->mtu <= ETH_DATA_LEN)
2362		/* enough for a TCP header */
2363		mgp->small_bytes = (128 > SMP_CACHE_BYTES)
2364		    ? (128 - MXGEFW_PAD)
2365		    : (SMP_CACHE_BYTES - MXGEFW_PAD);
2366	else
2367		/* enough for a vlan encapsulated ETH_DATA_LEN frame */
2368		mgp->small_bytes = VLAN_ETH_FRAME_LEN;
2369
2370	/* Override the small buffer size? */
2371	if (myri10ge_small_bytes >= 0)
2372		mgp->small_bytes = myri10ge_small_bytes;
2373
2374	/* Firmware needs the big buff size as a power of 2.  Lie and
2375	 * tell him the buffer is larger, because we only use 1
2376	 * buffer/pkt, and the mtu will prevent overruns.
2377	 */
2378	big_pow2 = dev->mtu + ETH_HLEN + VLAN_HLEN + MXGEFW_PAD;
2379	if (big_pow2 < MYRI10GE_ALLOC_SIZE / 2) {
2380		while (!is_power_of_2(big_pow2))
2381			big_pow2++;
2382		mgp->big_bytes = dev->mtu + ETH_HLEN + VLAN_HLEN + MXGEFW_PAD;
2383	} else {
2384		big_pow2 = MYRI10GE_ALLOC_SIZE;
2385		mgp->big_bytes = big_pow2;
2386	}
2387
2388	/* setup the per-slice data structures */
2389	for (slice = 0; slice < mgp->num_slices; slice++) {
2390		ss = &mgp->ss[slice];
2391
2392		status = myri10ge_get_txrx(mgp, slice);
2393		if (status != 0) {
2394			netdev_err(dev, "failed to get ring sizes or locations\n");
2395			goto abort_with_rings;
2396		}
2397		status = myri10ge_allocate_rings(ss);
2398		if (status != 0)
2399			goto abort_with_rings;
2400
2401		/* only firmware which supports multiple TX queues
2402		 * supports setting up the tx stats on non-zero
2403		 * slices */
2404		if (slice == 0 || mgp->dev->real_num_tx_queues > 1)
2405			status = myri10ge_set_stats(mgp, slice);
2406		if (status) {
2407			netdev_err(dev, "Couldn't set stats DMA\n");
2408			goto abort_with_rings;
2409		}
2410
2411		/* must happen prior to any irq */
2412		napi_enable(&(ss)->napi);
2413	}
2414
2415	/* now give firmware buffers sizes, and MTU */
2416	cmd.data0 = dev->mtu + ETH_HLEN + VLAN_HLEN;
2417	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_MTU, &cmd, 0);
2418	cmd.data0 = mgp->small_bytes;
2419	status |=
2420	    myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_SMALL_BUFFER_SIZE, &cmd, 0);
2421	cmd.data0 = big_pow2;
2422	status |=
2423	    myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_BIG_BUFFER_SIZE, &cmd, 0);
2424	if (status) {
2425		netdev_err(dev, "Couldn't set buffer sizes\n");
2426		goto abort_with_rings;
2427	}
2428
2429	/*
2430	 * Set Linux style TSO mode; this is needed only on newer
2431	 *  firmware versions.  Older versions default to Linux
2432	 *  style TSO
2433	 */
2434	cmd.data0 = 0;
2435	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_TSO_MODE, &cmd, 0);
2436	if (status && status != -ENOSYS) {
2437		netdev_err(dev, "Couldn't set TSO mode\n");
2438		goto abort_with_rings;
2439	}
2440
2441	mgp->link_state = ~0U;
2442	mgp->rdma_tags_available = 15;
2443
2444	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_UP, &cmd, 0);
2445	if (status) {
2446		netdev_err(dev, "Couldn't bring up link\n");
2447		goto abort_with_rings;
2448	}
2449
2450	mgp->running = MYRI10GE_ETH_RUNNING;
2451	mgp->watchdog_timer.expires = jiffies + myri10ge_watchdog_timeout * HZ;
2452	add_timer(&mgp->watchdog_timer);
2453	netif_tx_wake_all_queues(dev);
2454
2455	return 0;
2456
2457abort_with_rings:
2458	while (slice) {
2459		slice--;
2460		napi_disable(&mgp->ss[slice].napi);
2461	}
2462	for (i = 0; i < mgp->num_slices; i++)
2463		myri10ge_free_rings(&mgp->ss[i]);
2464
2465	myri10ge_free_irq(mgp);
2466
2467abort_with_nothing:
2468	mgp->running = MYRI10GE_ETH_STOPPED;
2469	return -ENOMEM;
2470}
2471
2472static int myri10ge_close(struct net_device *dev)
2473{
2474	struct myri10ge_priv *mgp = netdev_priv(dev);
2475	struct myri10ge_cmd cmd;
2476	int status, old_down_cnt;
2477	int i;
2478
2479	if (mgp->running != MYRI10GE_ETH_RUNNING)
2480		return 0;
2481
2482	if (mgp->ss[0].tx.req_bytes == NULL)
2483		return 0;
2484
2485	del_timer_sync(&mgp->watchdog_timer);
2486	mgp->running = MYRI10GE_ETH_STOPPING;
2487	for (i = 0; i < mgp->num_slices; i++)
2488		napi_disable(&mgp->ss[i].napi);
2489
2490	netif_carrier_off(dev);
2491
2492	netif_tx_stop_all_queues(dev);
2493	if (mgp->rebooted == 0) {
2494		old_down_cnt = mgp->down_cnt;
2495		mb();
2496		status =
2497		    myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_DOWN, &cmd, 0);
2498		if (status)
2499			netdev_err(dev, "Couldn't bring down link\n");
2500
2501		wait_event_timeout(mgp->down_wq, old_down_cnt != mgp->down_cnt,
2502				   HZ);
2503		if (old_down_cnt == mgp->down_cnt)
2504			netdev_err(dev, "never got down irq\n");
2505	}
2506	netif_tx_disable(dev);
2507	myri10ge_free_irq(mgp);
2508	for (i = 0; i < mgp->num_slices; i++)
2509		myri10ge_free_rings(&mgp->ss[i]);
2510
2511	mgp->running = MYRI10GE_ETH_STOPPED;
2512	return 0;
2513}
2514
2515/* copy an array of struct mcp_kreq_ether_send's to the mcp.  Copy
2516 * backwards one at a time and handle ring wraps */
2517
2518static inline void
2519myri10ge_submit_req_backwards(struct myri10ge_tx_buf *tx,
2520			      struct mcp_kreq_ether_send *src, int cnt)
2521{
2522	int idx, starting_slot;
2523	starting_slot = tx->req;
2524	while (cnt > 1) {
2525		cnt--;
2526		idx = (starting_slot + cnt) & tx->mask;
2527		myri10ge_pio_copy(&tx->lanai[idx], &src[cnt], sizeof(*src));
2528		mb();
2529	}
2530}
2531
2532/*
2533 * copy an array of struct mcp_kreq_ether_send's to the mcp.  Copy
2534 * at most 32 bytes at a time, so as to avoid involving the software
2535 * pio handler in the nic.   We re-write the first segment's flags
2536 * to mark them valid only after writing the entire chain.
2537 */
2538
2539static inline void
2540myri10ge_submit_req(struct myri10ge_tx_buf *tx, struct mcp_kreq_ether_send *src,
2541		    int cnt)
2542{
2543	int idx, i;
2544	struct mcp_kreq_ether_send __iomem *dstp, *dst;
2545	struct mcp_kreq_ether_send *srcp;
2546	u8 last_flags;
2547
2548	idx = tx->req & tx->mask;
2549
2550	last_flags = src->flags;
2551	src->flags = 0;
2552	mb();
2553	dst = dstp = &tx->lanai[idx];
2554	srcp = src;
2555
2556	if ((idx + cnt) < tx->mask) {
2557		for (i = 0; i < (cnt - 1); i += 2) {
2558			myri10ge_pio_copy(dstp, srcp, 2 * sizeof(*src));
2559			mb();	/* force write every 32 bytes */
2560			srcp += 2;
2561			dstp += 2;
2562		}
2563	} else {
2564		/* submit all but the first request, and ensure
2565		 * that it is submitted below */
2566		myri10ge_submit_req_backwards(tx, src, cnt);
2567		i = 0;
2568	}
2569	if (i < cnt) {
2570		/* submit the first request */
2571		myri10ge_pio_copy(dstp, srcp, sizeof(*src));
2572		mb();		/* barrier before setting valid flag */
2573	}
2574
2575	/* re-write the last 32-bits with the valid flags */
2576	src->flags = last_flags;
2577	put_be32(*((__be32 *) src + 3), (__be32 __iomem *) dst + 3);
2578	tx->req += cnt;
2579	mb();
2580}
2581
2582static void myri10ge_unmap_tx_dma(struct myri10ge_priv *mgp,
2583				  struct myri10ge_tx_buf *tx, int idx)
2584{
2585	unsigned int len;
2586	int last_idx;
2587
2588	/* Free any DMA resources we've alloced and clear out the skb slot */
2589	last_idx = (idx + 1) & tx->mask;
2590	idx = tx->req & tx->mask;
2591	do {
2592		len = dma_unmap_len(&tx->info[idx], len);
2593		if (len) {
2594			if (tx->info[idx].skb != NULL)
2595				dma_unmap_single(&mgp->pdev->dev,
2596						 dma_unmap_addr(&tx->info[idx],
2597								bus), len,
2598						 DMA_TO_DEVICE);
2599			else
2600				dma_unmap_page(&mgp->pdev->dev,
2601					       dma_unmap_addr(&tx->info[idx],
2602							      bus), len,
2603					       DMA_TO_DEVICE);
2604			dma_unmap_len_set(&tx->info[idx], len, 0);
2605			tx->info[idx].skb = NULL;
2606		}
2607		idx = (idx + 1) & tx->mask;
2608	} while (idx != last_idx);
2609}
2610
2611/*
2612 * Transmit a packet.  We need to split the packet so that a single
2613 * segment does not cross myri10ge->tx_boundary, so this makes segment
2614 * counting tricky.  So rather than try to count segments up front, we
2615 * just give up if there are too few segments to hold a reasonably
2616 * fragmented packet currently available.  If we run
2617 * out of segments while preparing a packet for DMA, we just linearize
2618 * it and try again.
2619 */
2620
2621static netdev_tx_t myri10ge_xmit(struct sk_buff *skb,
2622				       struct net_device *dev)
2623{
2624	struct myri10ge_priv *mgp = netdev_priv(dev);
2625	struct myri10ge_slice_state *ss;
2626	struct mcp_kreq_ether_send *req;
2627	struct myri10ge_tx_buf *tx;
2628	skb_frag_t *frag;
2629	struct netdev_queue *netdev_queue;
2630	dma_addr_t bus;
2631	u32 low;
2632	__be32 high_swapped;
2633	unsigned int len;
2634	int idx, avail, frag_cnt, frag_idx, count, mss, max_segments;
2635	u16 pseudo_hdr_offset, cksum_offset, queue;
2636	int cum_len, seglen, boundary, rdma_count;
2637	u8 flags, odd_flag;
2638
2639	queue = skb_get_queue_mapping(skb);
2640	ss = &mgp->ss[queue];
2641	netdev_queue = netdev_get_tx_queue(mgp->dev, queue);
2642	tx = &ss->tx;
2643
2644again:
2645	req = tx->req_list;
2646	avail = tx->mask - 1 - (tx->req - tx->done);
2647
2648	mss = 0;
2649	max_segments = MXGEFW_MAX_SEND_DESC;
2650
2651	if (skb_is_gso(skb)) {
2652		mss = skb_shinfo(skb)->gso_size;
2653		max_segments = MYRI10GE_MAX_SEND_DESC_TSO;
2654	}
2655
2656	if ((unlikely(avail < max_segments))) {
2657		/* we are out of transmit resources */
2658		tx->stop_queue++;
2659		netif_tx_stop_queue(netdev_queue);
2660		return NETDEV_TX_BUSY;
2661	}
2662
2663	/* Setup checksum offloading, if needed */
2664	cksum_offset = 0;
2665	pseudo_hdr_offset = 0;
2666	odd_flag = 0;
2667	flags = (MXGEFW_FLAGS_NO_TSO | MXGEFW_FLAGS_FIRST);
2668	if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
2669		cksum_offset = skb_checksum_start_offset(skb);
2670		pseudo_hdr_offset = cksum_offset + skb->csum_offset;
2671		/* If the headers are excessively large, then we must
2672		 * fall back to a software checksum */
2673		if (unlikely(!mss && (cksum_offset > 255 ||
2674				      pseudo_hdr_offset > 127))) {
2675			if (skb_checksum_help(skb))
2676				goto drop;
2677			cksum_offset = 0;
2678			pseudo_hdr_offset = 0;
2679		} else {
2680			odd_flag = MXGEFW_FLAGS_ALIGN_ODD;
2681			flags |= MXGEFW_FLAGS_CKSUM;
2682		}
2683	}
2684
2685	cum_len = 0;
2686
2687	if (mss) {		/* TSO */
2688		/* this removes any CKSUM flag from before */
2689		flags = (MXGEFW_FLAGS_TSO_HDR | MXGEFW_FLAGS_FIRST);
2690
2691		/* negative cum_len signifies to the
2692		 * send loop that we are still in the
2693		 * header portion of the TSO packet.
2694		 * TSO header can be at most 1KB long */
2695		cum_len = -skb_tcp_all_headers(skb);
2696
2697		/* for IPv6 TSO, the checksum offset stores the
2698		 * TCP header length, to save the firmware from
2699		 * the need to parse the headers */
2700		if (skb_is_gso_v6(skb)) {
2701			cksum_offset = tcp_hdrlen(skb);
2702			/* Can only handle headers <= max_tso6 long */
2703			if (unlikely(-cum_len > mgp->max_tso6))
2704				return myri10ge_sw_tso(skb, dev);
2705		}
2706		/* for TSO, pseudo_hdr_offset holds mss.
2707		 * The firmware figures out where to put
2708		 * the checksum by parsing the header. */
2709		pseudo_hdr_offset = mss;
2710	} else
2711		/* Mark small packets, and pad out tiny packets */
2712	if (skb->len <= MXGEFW_SEND_SMALL_SIZE) {
2713		flags |= MXGEFW_FLAGS_SMALL;
2714
2715		/* pad frames to at least ETH_ZLEN bytes */
2716		if (eth_skb_pad(skb)) {
2717			/* The packet is gone, so we must
2718			 * return 0 */
2719			ss->stats.tx_dropped += 1;
2720			return NETDEV_TX_OK;
2721		}
2722	}
2723
2724	/* map the skb for DMA */
2725	len = skb_headlen(skb);
2726	bus = dma_map_single(&mgp->pdev->dev, skb->data, len, DMA_TO_DEVICE);
2727	if (unlikely(dma_mapping_error(&mgp->pdev->dev, bus)))
2728		goto drop;
2729
2730	idx = tx->req & tx->mask;
2731	tx->info[idx].skb = skb;
2732	dma_unmap_addr_set(&tx->info[idx], bus, bus);
2733	dma_unmap_len_set(&tx->info[idx], len, len);
2734
2735	frag_cnt = skb_shinfo(skb)->nr_frags;
2736	frag_idx = 0;
2737	count = 0;
2738	rdma_count = 0;
2739
2740	/* "rdma_count" is the number of RDMAs belonging to the
2741	 * current packet BEFORE the current send request. For
2742	 * non-TSO packets, this is equal to "count".
2743	 * For TSO packets, rdma_count needs to be reset
2744	 * to 0 after a segment cut.
2745	 *
2746	 * The rdma_count field of the send request is
2747	 * the number of RDMAs of the packet starting at
2748	 * that request. For TSO send requests with one ore more cuts
2749	 * in the middle, this is the number of RDMAs starting
2750	 * after the last cut in the request. All previous
2751	 * segments before the last cut implicitly have 1 RDMA.
2752	 *
2753	 * Since the number of RDMAs is not known beforehand,
2754	 * it must be filled-in retroactively - after each
2755	 * segmentation cut or at the end of the entire packet.
2756	 */
2757
2758	while (1) {
2759		/* Break the SKB or Fragment up into pieces which
2760		 * do not cross mgp->tx_boundary */
2761		low = MYRI10GE_LOWPART_TO_U32(bus);
2762		high_swapped = htonl(MYRI10GE_HIGHPART_TO_U32(bus));
2763		while (len) {
2764			u8 flags_next;
2765			int cum_len_next;
2766
2767			if (unlikely(count == max_segments))
2768				goto abort_linearize;
2769
2770			boundary =
2771			    (low + mgp->tx_boundary) & ~(mgp->tx_boundary - 1);
2772			seglen = boundary - low;
2773			if (seglen > len)
2774				seglen = len;
2775			flags_next = flags & ~MXGEFW_FLAGS_FIRST;
2776			cum_len_next = cum_len + seglen;
2777			if (mss) {	/* TSO */
2778				(req - rdma_count)->rdma_count = rdma_count + 1;
2779
2780				if (likely(cum_len >= 0)) {	/* payload */
2781					int next_is_first, chop;
2782
2783					chop = (cum_len_next > mss);
2784					cum_len_next = cum_len_next % mss;
2785					next_is_first = (cum_len_next == 0);
2786					flags |= chop * MXGEFW_FLAGS_TSO_CHOP;
2787					flags_next |= next_is_first *
2788					    MXGEFW_FLAGS_FIRST;
2789					rdma_count |= -(chop | next_is_first);
2790					rdma_count += chop & ~next_is_first;
2791				} else if (likely(cum_len_next >= 0)) {	/* header ends */
2792					int small;
2793
2794					rdma_count = -1;
2795					cum_len_next = 0;
2796					seglen = -cum_len;
2797					small = (mss <= MXGEFW_SEND_SMALL_SIZE);
2798					flags_next = MXGEFW_FLAGS_TSO_PLD |
2799					    MXGEFW_FLAGS_FIRST |
2800					    (small * MXGEFW_FLAGS_SMALL);
2801				}
2802			}
2803			req->addr_high = high_swapped;
2804			req->addr_low = htonl(low);
2805			req->pseudo_hdr_offset = htons(pseudo_hdr_offset);
2806			req->pad = 0;	/* complete solid 16-byte block; does this matter? */
2807			req->rdma_count = 1;
2808			req->length = htons(seglen);
2809			req->cksum_offset = cksum_offset;
2810			req->flags = flags | ((cum_len & 1) * odd_flag);
2811
2812			low += seglen;
2813			len -= seglen;
2814			cum_len = cum_len_next;
2815			flags = flags_next;
2816			req++;
2817			count++;
2818			rdma_count++;
2819			if (cksum_offset != 0 && !(mss && skb_is_gso_v6(skb))) {
2820				if (unlikely(cksum_offset > seglen))
2821					cksum_offset -= seglen;
2822				else
2823					cksum_offset = 0;
2824			}
2825		}
2826		if (frag_idx == frag_cnt)
2827			break;
2828
2829		/* map next fragment for DMA */
2830		frag = &skb_shinfo(skb)->frags[frag_idx];
2831		frag_idx++;
2832		len = skb_frag_size(frag);
2833		bus = skb_frag_dma_map(&mgp->pdev->dev, frag, 0, len,
2834				       DMA_TO_DEVICE);
2835		if (unlikely(dma_mapping_error(&mgp->pdev->dev, bus))) {
2836			myri10ge_unmap_tx_dma(mgp, tx, idx);
2837			goto drop;
2838		}
2839		idx = (count + tx->req) & tx->mask;
2840		dma_unmap_addr_set(&tx->info[idx], bus, bus);
2841		dma_unmap_len_set(&tx->info[idx], len, len);
2842	}
2843
2844	(req - rdma_count)->rdma_count = rdma_count;
2845	if (mss)
2846		do {
2847			req--;
2848			req->flags |= MXGEFW_FLAGS_TSO_LAST;
2849		} while (!(req->flags & (MXGEFW_FLAGS_TSO_CHOP |
2850					 MXGEFW_FLAGS_FIRST)));
2851	idx = ((count - 1) + tx->req) & tx->mask;
2852	tx->info[idx].last = 1;
2853	myri10ge_submit_req(tx, tx->req_list, count);
2854	/* if using multiple tx queues, make sure NIC polls the
2855	 * current slice */
2856	if ((mgp->dev->real_num_tx_queues > 1) && tx->queue_active == 0) {
2857		tx->queue_active = 1;
2858		put_be32(htonl(1), tx->send_go);
2859		mb();
2860	}
2861	tx->pkt_start++;
2862	if ((avail - count) < MXGEFW_MAX_SEND_DESC) {
2863		tx->stop_queue++;
2864		netif_tx_stop_queue(netdev_queue);
2865	}
2866	return NETDEV_TX_OK;
2867
2868abort_linearize:
2869	myri10ge_unmap_tx_dma(mgp, tx, idx);
2870
2871	if (skb_is_gso(skb)) {
2872		netdev_err(mgp->dev, "TSO but wanted to linearize?!?!?\n");
2873		goto drop;
2874	}
2875
2876	if (skb_linearize(skb))
2877		goto drop;
2878
2879	tx->linearized++;
2880	goto again;
2881
2882drop:
2883	dev_kfree_skb_any(skb);
2884	ss->stats.tx_dropped += 1;
2885	return NETDEV_TX_OK;
2886
2887}
2888
2889static netdev_tx_t myri10ge_sw_tso(struct sk_buff *skb,
2890					 struct net_device *dev)
2891{
2892	struct sk_buff *segs, *curr, *next;
2893	struct myri10ge_priv *mgp = netdev_priv(dev);
2894	struct myri10ge_slice_state *ss;
2895	netdev_tx_t status;
2896
2897	segs = skb_gso_segment(skb, dev->features & ~NETIF_F_TSO6);
2898	if (IS_ERR(segs))
2899		goto drop;
2900
2901	skb_list_walk_safe(segs, curr, next) {
2902		skb_mark_not_on_list(curr);
2903		status = myri10ge_xmit(curr, dev);
2904		if (status != 0) {
2905			dev_kfree_skb_any(curr);
2906			skb_list_walk_safe(next, curr, next) {
2907				curr->next = NULL;
2908				dev_kfree_skb_any(curr);
2909			}
2910			goto drop;
2911		}
2912	}
2913	dev_kfree_skb_any(skb);
2914	return NETDEV_TX_OK;
2915
2916drop:
2917	ss = &mgp->ss[skb_get_queue_mapping(skb)];
2918	dev_kfree_skb_any(skb);
2919	ss->stats.tx_dropped += 1;
2920	return NETDEV_TX_OK;
2921}
2922
2923static void myri10ge_get_stats(struct net_device *dev,
2924			       struct rtnl_link_stats64 *stats)
2925{
2926	const struct myri10ge_priv *mgp = netdev_priv(dev);
2927	const struct myri10ge_slice_netstats *slice_stats;
2928	int i;
2929
2930	for (i = 0; i < mgp->num_slices; i++) {
2931		slice_stats = &mgp->ss[i].stats;
2932		stats->rx_packets += slice_stats->rx_packets;
2933		stats->tx_packets += slice_stats->tx_packets;
2934		stats->rx_bytes += slice_stats->rx_bytes;
2935		stats->tx_bytes += slice_stats->tx_bytes;
2936		stats->rx_dropped += slice_stats->rx_dropped;
2937		stats->tx_dropped += slice_stats->tx_dropped;
2938	}
2939}
2940
2941static void myri10ge_set_multicast_list(struct net_device *dev)
2942{
2943	struct myri10ge_priv *mgp = netdev_priv(dev);
2944	struct myri10ge_cmd cmd;
2945	struct netdev_hw_addr *ha;
2946	__be32 data[2] = { 0, 0 };
2947	int err;
2948
2949	/* can be called from atomic contexts,
2950	 * pass 1 to force atomicity in myri10ge_send_cmd() */
2951	myri10ge_change_promisc(mgp, dev->flags & IFF_PROMISC, 1);
2952
2953	/* This firmware is known to not support multicast */
2954	if (!mgp->fw_multicast_support)
2955		return;
2956
2957	/* Disable multicast filtering */
2958
2959	err = myri10ge_send_cmd(mgp, MXGEFW_ENABLE_ALLMULTI, &cmd, 1);
2960	if (err != 0) {
2961		netdev_err(dev, "Failed MXGEFW_ENABLE_ALLMULTI, error status: %d\n",
2962			   err);
2963		goto abort;
2964	}
2965
2966	if ((dev->flags & IFF_ALLMULTI) || mgp->adopted_rx_filter_bug) {
2967		/* request to disable multicast filtering, so quit here */
2968		return;
2969	}
2970
2971	/* Flush the filters */
2972
2973	err = myri10ge_send_cmd(mgp, MXGEFW_LEAVE_ALL_MULTICAST_GROUPS,
2974				&cmd, 1);
2975	if (err != 0) {
2976		netdev_err(dev, "Failed MXGEFW_LEAVE_ALL_MULTICAST_GROUPS, error status: %d\n",
2977			   err);
2978		goto abort;
2979	}
2980
2981	/* Walk the multicast list, and add each address */
2982	netdev_for_each_mc_addr(ha, dev) {
2983		memcpy(data, &ha->addr, ETH_ALEN);
2984		cmd.data0 = ntohl(data[0]);
2985		cmd.data1 = ntohl(data[1]);
2986		err = myri10ge_send_cmd(mgp, MXGEFW_JOIN_MULTICAST_GROUP,
2987					&cmd, 1);
2988
2989		if (err != 0) {
2990			netdev_err(dev, "Failed MXGEFW_JOIN_MULTICAST_GROUP, error status:%d %pM\n",
2991				   err, ha->addr);
2992			goto abort;
2993		}
2994	}
2995	/* Enable multicast filtering */
2996	err = myri10ge_send_cmd(mgp, MXGEFW_DISABLE_ALLMULTI, &cmd, 1);
2997	if (err != 0) {
2998		netdev_err(dev, "Failed MXGEFW_DISABLE_ALLMULTI, error status: %d\n",
2999			   err);
3000		goto abort;
3001	}
3002
3003	return;
3004
3005abort:
3006	return;
3007}
3008
3009static int myri10ge_set_mac_address(struct net_device *dev, void *addr)
3010{
3011	struct sockaddr *sa = addr;
3012	struct myri10ge_priv *mgp = netdev_priv(dev);
3013	int status;
3014
3015	if (!is_valid_ether_addr(sa->sa_data))
3016		return -EADDRNOTAVAIL;
3017
3018	status = myri10ge_update_mac_address(mgp, sa->sa_data);
3019	if (status != 0) {
3020		netdev_err(dev, "changing mac address failed with %d\n",
3021			   status);
3022		return status;
3023	}
3024
3025	/* change the dev structure */
3026	eth_hw_addr_set(dev, sa->sa_data);
3027	return 0;
3028}
3029
3030static int myri10ge_change_mtu(struct net_device *dev, int new_mtu)
3031{
3032	struct myri10ge_priv *mgp = netdev_priv(dev);
3033
3034	netdev_info(dev, "changing mtu from %d to %d\n", dev->mtu, new_mtu);
3035	if (mgp->running) {
3036		/* if we change the mtu on an active device, we must
3037		 * reset the device so the firmware sees the change */
3038		myri10ge_close(dev);
3039		dev->mtu = new_mtu;
3040		myri10ge_open(dev);
3041	} else
3042		dev->mtu = new_mtu;
3043
3044	return 0;
3045}
3046
3047/*
3048 * Enable ECRC to align PCI-E Completion packets on an 8-byte boundary.
3049 * Only do it if the bridge is a root port since we don't want to disturb
3050 * any other device, except if forced with myri10ge_ecrc_enable > 1.
3051 */
3052
3053static void myri10ge_enable_ecrc(struct myri10ge_priv *mgp)
3054{
3055	struct pci_dev *bridge = mgp->pdev->bus->self;
3056	struct device *dev = &mgp->pdev->dev;
3057	int cap;
3058	unsigned err_cap;
3059	int ret;
3060
3061	if (!myri10ge_ecrc_enable || !bridge)
3062		return;
3063
3064	/* check that the bridge is a root port */
3065	if (pci_pcie_type(bridge) != PCI_EXP_TYPE_ROOT_PORT) {
3066		if (myri10ge_ecrc_enable > 1) {
3067			struct pci_dev *prev_bridge, *old_bridge = bridge;
3068
3069			/* Walk the hierarchy up to the root port
3070			 * where ECRC has to be enabled */
3071			do {
3072				prev_bridge = bridge;
3073				bridge = bridge->bus->self;
3074				if (!bridge || prev_bridge == bridge) {
3075					dev_err(dev,
3076						"Failed to find root port"
3077						" to force ECRC\n");
3078					return;
3079				}
3080			} while (pci_pcie_type(bridge) !=
3081				 PCI_EXP_TYPE_ROOT_PORT);
3082
3083			dev_info(dev,
3084				 "Forcing ECRC on non-root port %s"
3085				 " (enabling on root port %s)\n",
3086				 pci_name(old_bridge), pci_name(bridge));
3087		} else {
3088			dev_err(dev,
3089				"Not enabling ECRC on non-root port %s\n",
3090				pci_name(bridge));
3091			return;
3092		}
3093	}
3094
3095	cap = pci_find_ext_capability(bridge, PCI_EXT_CAP_ID_ERR);
3096	if (!cap)
3097		return;
3098
3099	ret = pci_read_config_dword(bridge, cap + PCI_ERR_CAP, &err_cap);
3100	if (ret) {
3101		dev_err(dev, "failed reading ext-conf-space of %s\n",
3102			pci_name(bridge));
3103		dev_err(dev, "\t pci=nommconf in use? "
3104			"or buggy/incomplete/absent ACPI MCFG attr?\n");
3105		return;
3106	}
3107	if (!(err_cap & PCI_ERR_CAP_ECRC_GENC))
3108		return;
3109
3110	err_cap |= PCI_ERR_CAP_ECRC_GENE;
3111	pci_write_config_dword(bridge, cap + PCI_ERR_CAP, err_cap);
3112	dev_info(dev, "Enabled ECRC on upstream bridge %s\n", pci_name(bridge));
3113}
3114
3115/*
3116 * The Lanai Z8E PCI-E interface achieves higher Read-DMA throughput
3117 * when the PCI-E Completion packets are aligned on an 8-byte
3118 * boundary.  Some PCI-E chip sets always align Completion packets; on
3119 * the ones that do not, the alignment can be enforced by enabling
3120 * ECRC generation (if supported).
3121 *
3122 * When PCI-E Completion packets are not aligned, it is actually more
3123 * efficient to limit Read-DMA transactions to 2KB, rather than 4KB.
3124 *
3125 * If the driver can neither enable ECRC nor verify that it has
3126 * already been enabled, then it must use a firmware image which works
3127 * around unaligned completion packets (myri10ge_rss_ethp_z8e.dat), and it
3128 * should also ensure that it never gives the device a Read-DMA which is
3129 * larger than 2KB by setting the tx_boundary to 2KB.  If ECRC is
3130 * enabled, then the driver should use the aligned (myri10ge_rss_eth_z8e.dat)
3131 * firmware image, and set tx_boundary to 4KB.
3132 */
3133
3134static void myri10ge_firmware_probe(struct myri10ge_priv *mgp)
3135{
3136	struct pci_dev *pdev = mgp->pdev;
3137	struct device *dev = &pdev->dev;
3138	int status;
3139
3140	mgp->tx_boundary = 4096;
3141	/*
3142	 * Verify the max read request size was set to 4KB
3143	 * before trying the test with 4KB.
3144	 */
3145	status = pcie_get_readrq(pdev);
3146	if (status < 0) {
3147		dev_err(dev, "Couldn't read max read req size: %d\n", status);
3148		goto abort;
3149	}
3150	if (status != 4096) {
3151		dev_warn(dev, "Max Read Request size != 4096 (%d)\n", status);
3152		mgp->tx_boundary = 2048;
3153	}
3154	/*
3155	 * load the optimized firmware (which assumes aligned PCIe
3156	 * completions) in order to see if it works on this host.
3157	 */
3158	set_fw_name(mgp, myri10ge_fw_aligned, false);
3159	status = myri10ge_load_firmware(mgp, 1);
3160	if (status != 0) {
3161		goto abort;
3162	}
3163
3164	/*
3165	 * Enable ECRC if possible
3166	 */
3167	myri10ge_enable_ecrc(mgp);
3168
3169	/*
3170	 * Run a DMA test which watches for unaligned completions and
3171	 * aborts on the first one seen.
3172	 */
3173
3174	status = myri10ge_dma_test(mgp, MXGEFW_CMD_UNALIGNED_TEST);
3175	if (status == 0)
3176		return;		/* keep the aligned firmware */
3177
3178	if (status != -E2BIG)
3179		dev_warn(dev, "DMA test failed: %d\n", status);
3180	if (status == -ENOSYS)
3181		dev_warn(dev, "Falling back to ethp! "
3182			 "Please install up to date fw\n");
3183abort:
3184	/* fall back to using the unaligned firmware */
3185	mgp->tx_boundary = 2048;
3186	set_fw_name(mgp, myri10ge_fw_unaligned, false);
3187}
3188
3189static void myri10ge_select_firmware(struct myri10ge_priv *mgp)
3190{
3191	int overridden = 0;
3192
3193	if (myri10ge_force_firmware == 0) {
3194		int link_width;
3195		u16 lnk;
3196
3197		pcie_capability_read_word(mgp->pdev, PCI_EXP_LNKSTA, &lnk);
3198		link_width = (lnk >> 4) & 0x3f;
3199
3200		/* Check to see if Link is less than 8 or if the
3201		 * upstream bridge is known to provide aligned
3202		 * completions */
3203		if (link_width < 8) {
3204			dev_info(&mgp->pdev->dev, "PCIE x%d Link\n",
3205				 link_width);
3206			mgp->tx_boundary = 4096;
3207			set_fw_name(mgp, myri10ge_fw_aligned, false);
3208		} else {
3209			myri10ge_firmware_probe(mgp);
3210		}
3211	} else {
3212		if (myri10ge_force_firmware == 1) {
3213			dev_info(&mgp->pdev->dev,
3214				 "Assuming aligned completions (forced)\n");
3215			mgp->tx_boundary = 4096;
3216			set_fw_name(mgp, myri10ge_fw_aligned, false);
3217		} else {
3218			dev_info(&mgp->pdev->dev,
3219				 "Assuming unaligned completions (forced)\n");
3220			mgp->tx_boundary = 2048;
3221			set_fw_name(mgp, myri10ge_fw_unaligned, false);
3222		}
3223	}
3224
3225	kernel_param_lock(THIS_MODULE);
3226	if (myri10ge_fw_name != NULL) {
3227		char *fw_name = kstrdup(myri10ge_fw_name, GFP_KERNEL);
3228		if (fw_name) {
3229			overridden = 1;
3230			set_fw_name(mgp, fw_name, true);
3231		}
3232	}
3233	kernel_param_unlock(THIS_MODULE);
3234
3235	if (mgp->board_number < MYRI10GE_MAX_BOARDS &&
3236	    myri10ge_fw_names[mgp->board_number] != NULL &&
3237	    strlen(myri10ge_fw_names[mgp->board_number])) {
3238		set_fw_name(mgp, myri10ge_fw_names[mgp->board_number], false);
3239		overridden = 1;
3240	}
3241	if (overridden)
3242		dev_info(&mgp->pdev->dev, "overriding firmware to %s\n",
3243			 mgp->fw_name);
3244}
3245
3246static void myri10ge_mask_surprise_down(struct pci_dev *pdev)
3247{
3248	struct pci_dev *bridge = pdev->bus->self;
3249	int cap;
3250	u32 mask;
3251
3252	if (bridge == NULL)
3253		return;
3254
3255	cap = pci_find_ext_capability(bridge, PCI_EXT_CAP_ID_ERR);
3256	if (cap) {
3257		/* a sram parity error can cause a surprise link
3258		 * down; since we expect and can recover from sram
3259		 * parity errors, mask surprise link down events */
3260		pci_read_config_dword(bridge, cap + PCI_ERR_UNCOR_MASK, &mask);
3261		mask |= 0x20;
3262		pci_write_config_dword(bridge, cap + PCI_ERR_UNCOR_MASK, mask);
3263	}
3264}
3265
3266static int __maybe_unused myri10ge_suspend(struct device *dev)
3267{
3268	struct myri10ge_priv *mgp;
3269	struct net_device *netdev;
3270
3271	mgp = dev_get_drvdata(dev);
3272	if (mgp == NULL)
3273		return -EINVAL;
3274	netdev = mgp->dev;
3275
3276	netif_device_detach(netdev);
3277	if (netif_running(netdev)) {
3278		netdev_info(netdev, "closing\n");
3279		rtnl_lock();
3280		myri10ge_close(netdev);
3281		rtnl_unlock();
3282	}
3283	myri10ge_dummy_rdma(mgp, 0);
3284
3285	return 0;
3286}
3287
3288static int __maybe_unused myri10ge_resume(struct device *dev)
3289{
3290	struct pci_dev *pdev = to_pci_dev(dev);
3291	struct myri10ge_priv *mgp;
3292	struct net_device *netdev;
3293	int status;
3294	u16 vendor;
3295
3296	mgp = pci_get_drvdata(pdev);
3297	if (mgp == NULL)
3298		return -EINVAL;
3299	netdev = mgp->dev;
3300	msleep(5);		/* give card time to respond */
3301	pci_read_config_word(mgp->pdev, PCI_VENDOR_ID, &vendor);
3302	if (vendor == 0xffff) {
3303		netdev_err(mgp->dev, "device disappeared!\n");
3304		return -EIO;
3305	}
3306
3307	myri10ge_reset(mgp);
3308	myri10ge_dummy_rdma(mgp, 1);
3309
3310	if (netif_running(netdev)) {
3311		rtnl_lock();
3312		status = myri10ge_open(netdev);
3313		rtnl_unlock();
3314		if (status != 0)
3315			goto abort_with_enabled;
3316
3317	}
3318	netif_device_attach(netdev);
3319
3320	return 0;
3321
3322abort_with_enabled:
3323	return -EIO;
3324}
3325
3326static u32 myri10ge_read_reboot(struct myri10ge_priv *mgp)
3327{
3328	struct pci_dev *pdev = mgp->pdev;
3329	int vs = mgp->vendor_specific_offset;
3330	u32 reboot;
3331
3332	/*enter read32 mode */
3333	pci_write_config_byte(pdev, vs + 0x10, 0x3);
3334
3335	/*read REBOOT_STATUS (0xfffffff0) */
3336	pci_write_config_dword(pdev, vs + 0x18, 0xfffffff0);
3337	pci_read_config_dword(pdev, vs + 0x14, &reboot);
3338	return reboot;
3339}
3340
3341static void
3342myri10ge_check_slice(struct myri10ge_slice_state *ss, int *reset_needed,
3343		     int *busy_slice_cnt, u32 rx_pause_cnt)
3344{
3345	struct myri10ge_priv *mgp = ss->mgp;
3346	int slice = ss - mgp->ss;
3347
3348	if (ss->tx.req != ss->tx.done &&
3349	    ss->tx.done == ss->watchdog_tx_done &&
3350	    ss->watchdog_tx_req != ss->watchdog_tx_done) {
3351		/* nic seems like it might be stuck.. */
3352		if (rx_pause_cnt != mgp->watchdog_pause) {
3353			if (net_ratelimit())
3354				netdev_warn(mgp->dev, "slice %d: TX paused, "
3355					    "check link partner\n", slice);
3356		} else {
3357			netdev_warn(mgp->dev,
3358				    "slice %d: TX stuck %d %d %d %d %d %d\n",
3359				    slice, ss->tx.queue_active, ss->tx.req,
3360				    ss->tx.done, ss->tx.pkt_start,
3361				    ss->tx.pkt_done,
3362				    (int)ntohl(mgp->ss[slice].fw_stats->
3363					       send_done_count));
3364			*reset_needed = 1;
3365			ss->stuck = 1;
3366		}
3367	}
3368	if (ss->watchdog_tx_done != ss->tx.done ||
3369	    ss->watchdog_rx_done != ss->rx_done.cnt) {
3370		*busy_slice_cnt += 1;
3371	}
3372	ss->watchdog_tx_done = ss->tx.done;
3373	ss->watchdog_tx_req = ss->tx.req;
3374	ss->watchdog_rx_done = ss->rx_done.cnt;
3375}
3376
3377/*
3378 * This watchdog is used to check whether the board has suffered
3379 * from a parity error and needs to be recovered.
3380 */
3381static void myri10ge_watchdog(struct work_struct *work)
3382{
3383	struct myri10ge_priv *mgp =
3384	    container_of(work, struct myri10ge_priv, watchdog_work);
3385	struct myri10ge_slice_state *ss;
3386	u32 reboot, rx_pause_cnt;
3387	int status, rebooted;
3388	int i;
3389	int reset_needed = 0;
3390	int busy_slice_cnt = 0;
3391	u16 cmd, vendor;
3392
3393	mgp->watchdog_resets++;
3394	pci_read_config_word(mgp->pdev, PCI_COMMAND, &cmd);
3395	rebooted = 0;
3396	if ((cmd & PCI_COMMAND_MASTER) == 0) {
3397		/* Bus master DMA disabled?  Check to see
3398		 * if the card rebooted due to a parity error
3399		 * For now, just report it */
3400		reboot = myri10ge_read_reboot(mgp);
3401		netdev_err(mgp->dev, "NIC rebooted (0x%x),%s resetting\n",
3402			   reboot, myri10ge_reset_recover ? "" : " not");
3403		if (myri10ge_reset_recover == 0)
3404			return;
3405		rtnl_lock();
3406		mgp->rebooted = 1;
3407		rebooted = 1;
3408		myri10ge_close(mgp->dev);
3409		myri10ge_reset_recover--;
3410		mgp->rebooted = 0;
3411		/*
3412		 * A rebooted nic will come back with config space as
3413		 * it was after power was applied to PCIe bus.
3414		 * Attempt to restore config space which was saved
3415		 * when the driver was loaded, or the last time the
3416		 * nic was resumed from power saving mode.
3417		 */
3418		pci_restore_state(mgp->pdev);
3419
3420		/* save state again for accounting reasons */
3421		pci_save_state(mgp->pdev);
3422
3423	} else {
3424		/* if we get back -1's from our slot, perhaps somebody
3425		 * powered off our card.  Don't try to reset it in
3426		 * this case */
3427		if (cmd == 0xffff) {
3428			pci_read_config_word(mgp->pdev, PCI_VENDOR_ID, &vendor);
3429			if (vendor == 0xffff) {
3430				netdev_err(mgp->dev, "device disappeared!\n");
3431				return;
3432			}
3433		}
3434		/* Perhaps it is a software error. See if stuck slice
3435		 * has recovered, reset if not */
3436		rx_pause_cnt = ntohl(mgp->ss[0].fw_stats->dropped_pause);
3437		for (i = 0; i < mgp->num_slices; i++) {
3438			ss = mgp->ss;
3439			if (ss->stuck) {
3440				myri10ge_check_slice(ss, &reset_needed,
3441						     &busy_slice_cnt,
3442						     rx_pause_cnt);
3443				ss->stuck = 0;
3444			}
3445		}
3446		if (!reset_needed) {
3447			netdev_dbg(mgp->dev, "not resetting\n");
3448			return;
3449		}
3450
3451		netdev_err(mgp->dev, "device timeout, resetting\n");
3452	}
3453
3454	if (!rebooted) {
3455		rtnl_lock();
3456		myri10ge_close(mgp->dev);
3457	}
3458	status = myri10ge_load_firmware(mgp, 1);
3459	if (status != 0)
3460		netdev_err(mgp->dev, "failed to load firmware\n");
3461	else
3462		myri10ge_open(mgp->dev);
3463	rtnl_unlock();
3464}
3465
3466/*
3467 * We use our own timer routine rather than relying upon
3468 * netdev->tx_timeout because we have a very large hardware transmit
3469 * queue.  Due to the large queue, the netdev->tx_timeout function
3470 * cannot detect a NIC with a parity error in a timely fashion if the
3471 * NIC is lightly loaded.
3472 */
3473static void myri10ge_watchdog_timer(struct timer_list *t)
3474{
3475	struct myri10ge_priv *mgp;
3476	struct myri10ge_slice_state *ss;
3477	int i, reset_needed, busy_slice_cnt;
3478	u32 rx_pause_cnt;
3479	u16 cmd;
3480
3481	mgp = from_timer(mgp, t, watchdog_timer);
3482
3483	rx_pause_cnt = ntohl(mgp->ss[0].fw_stats->dropped_pause);
3484	busy_slice_cnt = 0;
3485	for (i = 0, reset_needed = 0;
3486	     i < mgp->num_slices && reset_needed == 0; ++i) {
3487
3488		ss = &mgp->ss[i];
3489		if (ss->rx_small.watchdog_needed) {
3490			myri10ge_alloc_rx_pages(mgp, &ss->rx_small,
3491						mgp->small_bytes + MXGEFW_PAD,
3492						1);
3493			if (ss->rx_small.fill_cnt - ss->rx_small.cnt >=
3494			    myri10ge_fill_thresh)
3495				ss->rx_small.watchdog_needed = 0;
3496		}
3497		if (ss->rx_big.watchdog_needed) {
3498			myri10ge_alloc_rx_pages(mgp, &ss->rx_big,
3499						mgp->big_bytes, 1);
3500			if (ss->rx_big.fill_cnt - ss->rx_big.cnt >=
3501			    myri10ge_fill_thresh)
3502				ss->rx_big.watchdog_needed = 0;
3503		}
3504		myri10ge_check_slice(ss, &reset_needed, &busy_slice_cnt,
3505				     rx_pause_cnt);
3506	}
3507	/* if we've sent or received no traffic, poll the NIC to
3508	 * ensure it is still there.  Otherwise, we risk not noticing
3509	 * an error in a timely fashion */
3510	if (busy_slice_cnt == 0) {
3511		pci_read_config_word(mgp->pdev, PCI_COMMAND, &cmd);
3512		if ((cmd & PCI_COMMAND_MASTER) == 0) {
3513			reset_needed = 1;
3514		}
3515	}
3516	mgp->watchdog_pause = rx_pause_cnt;
3517
3518	if (reset_needed) {
3519		schedule_work(&mgp->watchdog_work);
3520	} else {
3521		/* rearm timer */
3522		mod_timer(&mgp->watchdog_timer,
3523			  jiffies + myri10ge_watchdog_timeout * HZ);
3524	}
3525}
3526
3527static void myri10ge_free_slices(struct myri10ge_priv *mgp)
3528{
3529	struct myri10ge_slice_state *ss;
3530	struct pci_dev *pdev = mgp->pdev;
3531	size_t bytes;
3532	int i;
3533
3534	if (mgp->ss == NULL)
3535		return;
3536
3537	for (i = 0; i < mgp->num_slices; i++) {
3538		ss = &mgp->ss[i];
3539		if (ss->rx_done.entry != NULL) {
3540			bytes = mgp->max_intr_slots *
3541			    sizeof(*ss->rx_done.entry);
3542			dma_free_coherent(&pdev->dev, bytes,
3543					  ss->rx_done.entry, ss->rx_done.bus);
3544			ss->rx_done.entry = NULL;
3545		}
3546		if (ss->fw_stats != NULL) {
3547			bytes = sizeof(*ss->fw_stats);
3548			dma_free_coherent(&pdev->dev, bytes,
3549					  ss->fw_stats, ss->fw_stats_bus);
3550			ss->fw_stats = NULL;
3551		}
3552		__netif_napi_del(&ss->napi);
3553	}
3554	/* Wait till napi structs are no longer used, and then free ss. */
3555	synchronize_net();
3556	kfree(mgp->ss);
3557	mgp->ss = NULL;
3558}
3559
3560static int myri10ge_alloc_slices(struct myri10ge_priv *mgp)
3561{
3562	struct myri10ge_slice_state *ss;
3563	struct pci_dev *pdev = mgp->pdev;
3564	size_t bytes;
3565	int i;
3566
3567	bytes = sizeof(*mgp->ss) * mgp->num_slices;
3568	mgp->ss = kzalloc(bytes, GFP_KERNEL);
3569	if (mgp->ss == NULL) {
3570		return -ENOMEM;
3571	}
3572
3573	for (i = 0; i < mgp->num_slices; i++) {
3574		ss = &mgp->ss[i];
3575		bytes = mgp->max_intr_slots * sizeof(*ss->rx_done.entry);
3576		ss->rx_done.entry = dma_alloc_coherent(&pdev->dev, bytes,
3577						       &ss->rx_done.bus,
3578						       GFP_KERNEL);
3579		if (ss->rx_done.entry == NULL)
3580			goto abort;
3581		bytes = sizeof(*ss->fw_stats);
3582		ss->fw_stats = dma_alloc_coherent(&pdev->dev, bytes,
3583						  &ss->fw_stats_bus,
3584						  GFP_KERNEL);
3585		if (ss->fw_stats == NULL)
3586			goto abort;
3587		ss->mgp = mgp;
3588		ss->dev = mgp->dev;
3589		netif_napi_add_weight(ss->dev, &ss->napi, myri10ge_poll,
3590				      myri10ge_napi_weight);
3591	}
3592	return 0;
3593abort:
3594	myri10ge_free_slices(mgp);
3595	return -ENOMEM;
3596}
3597
3598/*
3599 * This function determines the number of slices supported.
3600 * The number slices is the minimum of the number of CPUS,
3601 * the number of MSI-X irqs supported, the number of slices
3602 * supported by the firmware
3603 */
3604static void myri10ge_probe_slices(struct myri10ge_priv *mgp)
3605{
3606	struct myri10ge_cmd cmd;
3607	struct pci_dev *pdev = mgp->pdev;
3608	char *old_fw;
3609	bool old_allocated;
3610	int i, status, ncpus;
3611
3612	mgp->num_slices = 1;
3613	ncpus = netif_get_num_default_rss_queues();
3614
3615	if (myri10ge_max_slices == 1 || !pdev->msix_cap ||
3616	    (myri10ge_max_slices == -1 && ncpus < 2))
3617		return;
3618
3619	/* try to load the slice aware rss firmware */
3620	old_fw = mgp->fw_name;
3621	old_allocated = mgp->fw_name_allocated;
3622	/* don't free old_fw if we override it. */
3623	mgp->fw_name_allocated = false;
3624
3625	if (myri10ge_fw_name != NULL) {
3626		dev_info(&mgp->pdev->dev, "overriding rss firmware to %s\n",
3627			 myri10ge_fw_name);
3628		set_fw_name(mgp, myri10ge_fw_name, false);
3629	} else if (old_fw == myri10ge_fw_aligned)
3630		set_fw_name(mgp, myri10ge_fw_rss_aligned, false);
3631	else
3632		set_fw_name(mgp, myri10ge_fw_rss_unaligned, false);
3633	status = myri10ge_load_firmware(mgp, 0);
3634	if (status != 0) {
3635		dev_info(&pdev->dev, "Rss firmware not found\n");
3636		if (old_allocated)
3637			kfree(old_fw);
3638		return;
3639	}
3640
3641	/* hit the board with a reset to ensure it is alive */
3642	memset(&cmd, 0, sizeof(cmd));
3643	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_RESET, &cmd, 0);
3644	if (status != 0) {
3645		dev_err(&mgp->pdev->dev, "failed reset\n");
3646		goto abort_with_fw;
3647	}
3648
3649	mgp->max_intr_slots = cmd.data0 / sizeof(struct mcp_slot);
3650
3651	/* tell it the size of the interrupt queues */
3652	cmd.data0 = mgp->max_intr_slots * sizeof(struct mcp_slot);
3653	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_SIZE, &cmd, 0);
3654	if (status != 0) {
3655		dev_err(&mgp->pdev->dev, "failed MXGEFW_CMD_SET_INTRQ_SIZE\n");
3656		goto abort_with_fw;
3657	}
3658
3659	/* ask the maximum number of slices it supports */
3660	status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_MAX_RSS_QUEUES, &cmd, 0);
3661	if (status != 0)
3662		goto abort_with_fw;
3663	else
3664		mgp->num_slices = cmd.data0;
3665
3666	/* Only allow multiple slices if MSI-X is usable */
3667	if (!myri10ge_msi) {
3668		goto abort_with_fw;
3669	}
3670
3671	/* if the admin did not specify a limit to how many
3672	 * slices we should use, cap it automatically to the
3673	 * number of CPUs currently online */
3674	if (myri10ge_max_slices == -1)
3675		myri10ge_max_slices = ncpus;
3676
3677	if (mgp->num_slices > myri10ge_max_slices)
3678		mgp->num_slices = myri10ge_max_slices;
3679
3680	/* Now try to allocate as many MSI-X vectors as we have
3681	 * slices. We give up on MSI-X if we can only get a single
3682	 * vector. */
3683
3684	mgp->msix_vectors = kcalloc(mgp->num_slices, sizeof(*mgp->msix_vectors),
3685				    GFP_KERNEL);
3686	if (mgp->msix_vectors == NULL)
3687		goto no_msix;
3688	for (i = 0; i < mgp->num_slices; i++) {
3689		mgp->msix_vectors[i].entry = i;
3690	}
3691
3692	while (mgp->num_slices > 1) {
3693		mgp->num_slices = rounddown_pow_of_two(mgp->num_slices);
3694		if (mgp->num_slices == 1)
3695			goto no_msix;
3696		status = pci_enable_msix_range(pdev,
3697					       mgp->msix_vectors,
3698					       mgp->num_slices,
3699					       mgp->num_slices);
3700		if (status < 0)
3701			goto no_msix;
3702
3703		pci_disable_msix(pdev);
3704
3705		if (status == mgp->num_slices) {
3706			if (old_allocated)
3707				kfree(old_fw);
3708			return;
3709		} else {
3710			mgp->num_slices = status;
3711		}
3712	}
3713
3714no_msix:
3715	if (mgp->msix_vectors != NULL) {
3716		kfree(mgp->msix_vectors);
3717		mgp->msix_vectors = NULL;
3718	}
3719
3720abort_with_fw:
3721	mgp->num_slices = 1;
3722	set_fw_name(mgp, old_fw, old_allocated);
3723	myri10ge_load_firmware(mgp, 0);
3724}
3725
3726static const struct net_device_ops myri10ge_netdev_ops = {
3727	.ndo_open		= myri10ge_open,
3728	.ndo_stop		= myri10ge_close,
3729	.ndo_start_xmit		= myri10ge_xmit,
3730	.ndo_get_stats64	= myri10ge_get_stats,
3731	.ndo_validate_addr	= eth_validate_addr,
3732	.ndo_change_mtu		= myri10ge_change_mtu,
3733	.ndo_set_rx_mode	= myri10ge_set_multicast_list,
3734	.ndo_set_mac_address	= myri10ge_set_mac_address,
3735};
3736
3737static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3738{
3739	struct net_device *netdev;
3740	struct myri10ge_priv *mgp;
3741	struct device *dev = &pdev->dev;
3742	int status = -ENXIO;
3743	unsigned hdr_offset, ss_offset;
3744	static int board_number;
3745
3746	netdev = alloc_etherdev_mq(sizeof(*mgp), MYRI10GE_MAX_SLICES);
3747	if (netdev == NULL)
3748		return -ENOMEM;
3749
3750	SET_NETDEV_DEV(netdev, &pdev->dev);
3751
3752	mgp = netdev_priv(netdev);
3753	mgp->dev = netdev;
3754	mgp->pdev = pdev;
3755	mgp->pause = myri10ge_flow_control;
3756	mgp->intr_coal_delay = myri10ge_intr_coal_delay;
3757	mgp->msg_enable = netif_msg_init(myri10ge_debug, MYRI10GE_MSG_DEFAULT);
3758	mgp->board_number = board_number;
3759	init_waitqueue_head(&mgp->down_wq);
3760
3761	if (pci_enable_device(pdev)) {
3762		dev_err(&pdev->dev, "pci_enable_device call failed\n");
3763		status = -ENODEV;
3764		goto abort_with_netdev;
3765	}
3766
3767	/* Find the vendor-specific cap so we can check
3768	 * the reboot register later on */
3769	mgp->vendor_specific_offset
3770	    = pci_find_capability(pdev, PCI_CAP_ID_VNDR);
3771
3772	/* Set our max read request to 4KB */
3773	status = pcie_set_readrq(pdev, 4096);
3774	if (status != 0) {
3775		dev_err(&pdev->dev, "Error %d writing PCI_EXP_DEVCTL\n",
3776			status);
3777		goto abort_with_enabled;
3778	}
3779
3780	myri10ge_mask_surprise_down(pdev);
3781	pci_set_master(pdev);
3782	status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3783	if (status != 0) {
3784		dev_err(&pdev->dev, "Error %d setting DMA mask\n", status);
3785		goto abort_with_enabled;
3786	}
3787	mgp->cmd = dma_alloc_coherent(&pdev->dev, sizeof(*mgp->cmd),
3788				      &mgp->cmd_bus, GFP_KERNEL);
3789	if (!mgp->cmd) {
3790		status = -ENOMEM;
3791		goto abort_with_enabled;
3792	}
3793
3794	mgp->board_span = pci_resource_len(pdev, 0);
3795	mgp->iomem_base = pci_resource_start(pdev, 0);
3796	mgp->wc_cookie = arch_phys_wc_add(mgp->iomem_base, mgp->board_span);
3797	mgp->sram = ioremap_wc(mgp->iomem_base, mgp->board_span);
3798	if (mgp->sram == NULL) {
3799		dev_err(&pdev->dev, "ioremap failed for %ld bytes at 0x%lx\n",
3800			mgp->board_span, mgp->iomem_base);
3801		status = -ENXIO;
3802		goto abort_with_mtrr;
3803	}
3804	hdr_offset =
3805	    swab32(readl(mgp->sram + MCP_HEADER_PTR_OFFSET)) & 0xffffc;
3806	ss_offset = hdr_offset + offsetof(struct mcp_gen_header, string_specs);
3807	mgp->sram_size = swab32(readl(mgp->sram + ss_offset));
3808	if (mgp->sram_size > mgp->board_span ||
3809	    mgp->sram_size <= MYRI10GE_FW_OFFSET) {
3810		dev_err(&pdev->dev,
3811			"invalid sram_size %dB or board span %ldB\n",
3812			mgp->sram_size, mgp->board_span);
3813		status = -EINVAL;
3814		goto abort_with_ioremap;
3815	}
3816	memcpy_fromio(mgp->eeprom_strings,
3817		      mgp->sram + mgp->sram_size, MYRI10GE_EEPROM_STRINGS_SIZE);
3818	memset(mgp->eeprom_strings + MYRI10GE_EEPROM_STRINGS_SIZE - 2, 0, 2);
3819	status = myri10ge_read_mac_addr(mgp);
3820	if (status)
3821		goto abort_with_ioremap;
3822
3823	eth_hw_addr_set(netdev, mgp->mac_addr);
3824
3825	myri10ge_select_firmware(mgp);
3826
3827	status = myri10ge_load_firmware(mgp, 1);
3828	if (status != 0) {
3829		dev_err(&pdev->dev, "failed to load firmware\n");
3830		goto abort_with_ioremap;
3831	}
3832	myri10ge_probe_slices(mgp);
3833	status = myri10ge_alloc_slices(mgp);
3834	if (status != 0) {
3835		dev_err(&pdev->dev, "failed to alloc slice state\n");
3836		goto abort_with_firmware;
3837	}
3838	netif_set_real_num_tx_queues(netdev, mgp->num_slices);
3839	netif_set_real_num_rx_queues(netdev, mgp->num_slices);
3840	status = myri10ge_reset(mgp);
3841	if (status != 0) {
3842		dev_err(&pdev->dev, "failed reset\n");
3843		goto abort_with_slices;
3844	}
3845#ifdef CONFIG_MYRI10GE_DCA
3846	myri10ge_setup_dca(mgp);
3847#endif
3848	pci_set_drvdata(pdev, mgp);
3849
3850	/* MTU range: 68 - 9000 */
3851	netdev->min_mtu = ETH_MIN_MTU;
3852	netdev->max_mtu = MYRI10GE_MAX_ETHER_MTU - ETH_HLEN;
3853
3854	if (myri10ge_initial_mtu > netdev->max_mtu)
3855		myri10ge_initial_mtu = netdev->max_mtu;
3856	if (myri10ge_initial_mtu < netdev->min_mtu)
3857		myri10ge_initial_mtu = netdev->min_mtu;
3858
3859	netdev->mtu = myri10ge_initial_mtu;
3860
3861	netdev->netdev_ops = &myri10ge_netdev_ops;
3862	netdev->hw_features = mgp->features | NETIF_F_RXCSUM;
3863
3864	/* fake NETIF_F_HW_VLAN_CTAG_RX for good GRO performance */
3865	netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
3866
3867	netdev->features = netdev->hw_features | NETIF_F_HIGHDMA;
3868
3869	netdev->vlan_features |= mgp->features;
3870	if (mgp->fw_ver_tiny < 37)
3871		netdev->vlan_features &= ~NETIF_F_TSO6;
3872	if (mgp->fw_ver_tiny < 32)
3873		netdev->vlan_features &= ~NETIF_F_TSO;
3874
3875	/* make sure we can get an irq, and that MSI can be
3876	 * setup (if available). */
3877	status = myri10ge_request_irq(mgp);
3878	if (status != 0)
3879		goto abort_with_slices;
3880	myri10ge_free_irq(mgp);
3881
3882	/* Save configuration space to be restored if the
3883	 * nic resets due to a parity error */
3884	pci_save_state(pdev);
3885
3886	/* Setup the watchdog timer */
3887	timer_setup(&mgp->watchdog_timer, myri10ge_watchdog_timer, 0);
3888
3889	netdev->ethtool_ops = &myri10ge_ethtool_ops;
3890	INIT_WORK(&mgp->watchdog_work, myri10ge_watchdog);
3891	status = register_netdev(netdev);
3892	if (status != 0) {
3893		dev_err(&pdev->dev, "register_netdev failed: %d\n", status);
3894		goto abort_with_state;
3895	}
3896	if (mgp->msix_enabled)
3897		dev_info(dev, "%d MSI-X IRQs, tx bndry %d, fw %s, MTRR %s, WC Enabled\n",
3898			 mgp->num_slices, mgp->tx_boundary, mgp->fw_name,
3899			 (mgp->wc_cookie > 0 ? "Enabled" : "Disabled"));
3900	else
3901		dev_info(dev, "%s IRQ %d, tx bndry %d, fw %s, MTRR %s, WC Enabled\n",
3902			 mgp->msi_enabled ? "MSI" : "xPIC",
3903			 pdev->irq, mgp->tx_boundary, mgp->fw_name,
3904			 (mgp->wc_cookie > 0 ? "Enabled" : "Disabled"));
3905
3906	board_number++;
3907	return 0;
3908
3909abort_with_state:
3910	pci_restore_state(pdev);
3911
3912abort_with_slices:
3913	myri10ge_free_slices(mgp);
3914
3915abort_with_firmware:
3916	kfree(mgp->msix_vectors);
3917	myri10ge_dummy_rdma(mgp, 0);
3918
3919abort_with_ioremap:
3920	if (mgp->mac_addr_string != NULL)
3921		dev_err(&pdev->dev,
3922			"myri10ge_probe() failed: MAC=%s, SN=%ld\n",
3923			mgp->mac_addr_string, mgp->serial_number);
3924	iounmap(mgp->sram);
3925
3926abort_with_mtrr:
3927	arch_phys_wc_del(mgp->wc_cookie);
3928	dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
3929			  mgp->cmd, mgp->cmd_bus);
3930
3931abort_with_enabled:
3932	pci_disable_device(pdev);
3933
3934abort_with_netdev:
3935	set_fw_name(mgp, NULL, false);
3936	free_netdev(netdev);
3937	return status;
3938}
3939
3940/*
3941 * myri10ge_remove
3942 *
3943 * Does what is necessary to shutdown one Myrinet device. Called
3944 *   once for each Myrinet card by the kernel when a module is
3945 *   unloaded.
3946 */
3947static void myri10ge_remove(struct pci_dev *pdev)
3948{
3949	struct myri10ge_priv *mgp;
3950	struct net_device *netdev;
3951
3952	mgp = pci_get_drvdata(pdev);
3953	if (mgp == NULL)
3954		return;
3955
3956	cancel_work_sync(&mgp->watchdog_work);
3957	netdev = mgp->dev;
3958	unregister_netdev(netdev);
3959
3960#ifdef CONFIG_MYRI10GE_DCA
3961	myri10ge_teardown_dca(mgp);
3962#endif
3963	myri10ge_dummy_rdma(mgp, 0);
3964
3965	/* avoid a memory leak */
3966	pci_restore_state(pdev);
3967
3968	iounmap(mgp->sram);
3969	arch_phys_wc_del(mgp->wc_cookie);
3970	myri10ge_free_slices(mgp);
3971	kfree(mgp->msix_vectors);
3972	dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
3973			  mgp->cmd, mgp->cmd_bus);
3974
3975	set_fw_name(mgp, NULL, false);
3976	free_netdev(netdev);
3977	pci_disable_device(pdev);
3978}
3979
3980#define PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E 	0x0008
3981#define PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E_9	0x0009
3982
3983static const struct pci_device_id myri10ge_pci_tbl[] = {
3984	{PCI_DEVICE(PCI_VENDOR_ID_MYRICOM, PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E)},
3985	{PCI_DEVICE
3986	 (PCI_VENDOR_ID_MYRICOM, PCI_DEVICE_ID_MYRICOM_MYRI10GE_Z8E_9)},
3987	{0},
3988};
3989
3990MODULE_DEVICE_TABLE(pci, myri10ge_pci_tbl);
3991
3992static SIMPLE_DEV_PM_OPS(myri10ge_pm_ops, myri10ge_suspend, myri10ge_resume);
3993
3994static struct pci_driver myri10ge_driver = {
3995	.name = "myri10ge",
3996	.probe = myri10ge_probe,
3997	.remove = myri10ge_remove,
3998	.id_table = myri10ge_pci_tbl,
3999	.driver.pm = &myri10ge_pm_ops,
4000};
4001
4002#ifdef CONFIG_MYRI10GE_DCA
4003static int
4004myri10ge_notify_dca(struct notifier_block *nb, unsigned long event, void *p)
4005{
4006	int err = driver_for_each_device(&myri10ge_driver.driver,
4007					 NULL, &event,
4008					 myri10ge_notify_dca_device);
4009
4010	if (err)
4011		return NOTIFY_BAD;
4012	return NOTIFY_DONE;
4013}
4014
4015static struct notifier_block myri10ge_dca_notifier = {
4016	.notifier_call = myri10ge_notify_dca,
4017	.next = NULL,
4018	.priority = 0,
4019};
4020#endif				/* CONFIG_MYRI10GE_DCA */
4021
4022static __init int myri10ge_init_module(void)
4023{
4024	pr_info("Version %s\n", MYRI10GE_VERSION_STR);
4025
4026	if (myri10ge_rss_hash > MXGEFW_RSS_HASH_TYPE_MAX) {
4027		pr_err("Illegal rssh hash type %d, defaulting to source port\n",
4028		       myri10ge_rss_hash);
4029		myri10ge_rss_hash = MXGEFW_RSS_HASH_TYPE_SRC_PORT;
4030	}
4031#ifdef CONFIG_MYRI10GE_DCA
4032	dca_register_notify(&myri10ge_dca_notifier);
4033#endif
4034	if (myri10ge_max_slices > MYRI10GE_MAX_SLICES)
4035		myri10ge_max_slices = MYRI10GE_MAX_SLICES;
4036
4037	return pci_register_driver(&myri10ge_driver);
4038}
4039
4040module_init(myri10ge_init_module);
4041
4042static __exit void myri10ge_cleanup_module(void)
4043{
4044#ifdef CONFIG_MYRI10GE_DCA
4045	dca_unregister_notify(&myri10ge_dca_notifier);
4046#endif
4047	pci_unregister_driver(&myri10ge_driver);
4048}
4049
4050module_exit(myri10ge_cleanup_module);
4051