1 /*
2  * Copyright (c) 2003-2008 Chelsio, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34 
35 #include <linux/module.h>
36 #include <linux/moduleparam.h>
37 #include <linux/init.h>
38 #include <linux/pci.h>
39 #include <linux/dma-mapping.h>
40 #include <linux/netdevice.h>
41 #include <linux/etherdevice.h>
42 #include <linux/if_vlan.h>
43 #include <linux/mdio.h>
44 #include <linux/sockios.h>
45 #include <linux/workqueue.h>
46 #include <linux/proc_fs.h>
47 #include <linux/rtnetlink.h>
48 #include <linux/firmware.h>
49 #include <linux/log2.h>
50 #include <linux/stringify.h>
51 #include <linux/sched.h>
52 #include <linux/slab.h>
53 #include <linux/uaccess.h>
54 
55 #include "common.h"
56 #include "cxgb3_ioctl.h"
57 #include "regs.h"
58 #include "cxgb3_offload.h"
59 #include "version.h"
60 
61 #include "cxgb3_ctl_defs.h"
62 #include "t3_cpl.h"
63 #include "firmware_exports.h"
64 
65 enum {
66 	MAX_TXQ_ENTRIES = 16384,
67 	MAX_CTRL_TXQ_ENTRIES = 1024,
68 	MAX_RSPQ_ENTRIES = 16384,
69 	MAX_RX_BUFFERS = 16384,
70 	MAX_RX_JUMBO_BUFFERS = 16384,
71 	MIN_TXQ_ENTRIES = 4,
72 	MIN_CTRL_TXQ_ENTRIES = 4,
73 	MIN_RSPQ_ENTRIES = 32,
74 	MIN_FL_ENTRIES = 32
75 };
76 
77 #define PORT_MASK ((1 << MAX_NPORTS) - 1)
78 
79 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
80 			 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
81 			 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
82 
83 #define EEPROM_MAGIC 0x38E2F10C
84 
85 #define CH_DEVICE(devid, idx) \
86 	{ PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, PCI_ANY_ID, 0, 0, idx }
87 
88 static const struct pci_device_id cxgb3_pci_tbl[] = {
89 	CH_DEVICE(0x20, 0),	/* PE9000 */
90 	CH_DEVICE(0x21, 1),	/* T302E */
91 	CH_DEVICE(0x22, 2),	/* T310E */
92 	CH_DEVICE(0x23, 3),	/* T320X */
93 	CH_DEVICE(0x24, 1),	/* T302X */
94 	CH_DEVICE(0x25, 3),	/* T320E */
95 	CH_DEVICE(0x26, 2),	/* T310X */
96 	CH_DEVICE(0x30, 2),	/* T3B10 */
97 	CH_DEVICE(0x31, 3),	/* T3B20 */
98 	CH_DEVICE(0x32, 1),	/* T3B02 */
99 	CH_DEVICE(0x35, 6),	/* T3C20-derived T3C10 */
100 	CH_DEVICE(0x36, 3),	/* S320E-CR */
101 	CH_DEVICE(0x37, 7),	/* N320E-G2 */
102 	{0,}
103 };
104 
105 MODULE_DESCRIPTION(DRV_DESC);
106 MODULE_AUTHOR("Chelsio Communications");
107 MODULE_LICENSE("Dual BSD/GPL");
108 MODULE_VERSION(DRV_VERSION);
109 MODULE_DEVICE_TABLE(pci, cxgb3_pci_tbl);
110 
111 static int dflt_msg_enable = DFLT_MSG_ENABLE;
112 
113 module_param(dflt_msg_enable, int, 0644);
114 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T3 default message enable bitmap");
115 
116 /*
117  * The driver uses the best interrupt scheme available on a platform in the
118  * order MSI-X, MSI, legacy pin interrupts.  This parameter determines which
119  * of these schemes the driver may consider as follows:
120  *
121  * msi = 2: choose from among all three options
122  * msi = 1: only consider MSI and pin interrupts
123  * msi = 0: force pin interrupts
124  */
125 static int msi = 2;
126 
127 module_param(msi, int, 0644);
128 MODULE_PARM_DESC(msi, "whether to use MSI or MSI-X");
129 
130 /*
131  * The driver enables offload as a default.
132  * To disable it, use ofld_disable = 1.
133  */
134 
135 static int ofld_disable = 0;
136 
137 module_param(ofld_disable, int, 0644);
138 MODULE_PARM_DESC(ofld_disable, "whether to enable offload at init time or not");
139 
140 /*
141  * We have work elements that we need to cancel when an interface is taken
142  * down.  Normally the work elements would be executed by keventd but that
143  * can deadlock because of linkwatch.  If our close method takes the rtnl
144  * lock and linkwatch is ahead of our work elements in keventd, linkwatch
145  * will block keventd as it needs the rtnl lock, and we'll deadlock waiting
146  * for our work to complete.  Get our own work queue to solve this.
147  */
148 struct workqueue_struct *cxgb3_wq;
149 
150 /**
151  *	link_report - show link status and link speed/duplex
152  *	@p: the port whose settings are to be reported
153  *
154  *	Shows the link status, speed, and duplex of a port.
155  */
156 static void link_report(struct net_device *dev)
157 {
158 	if (!netif_carrier_ok(dev))
159 		netdev_info(dev, "link down\n");
160 	else {
161 		const char *s = "10Mbps";
162 		const struct port_info *p = netdev_priv(dev);
163 
164 		switch (p->link_config.speed) {
165 		case SPEED_10000:
166 			s = "10Gbps";
167 			break;
168 		case SPEED_1000:
169 			s = "1000Mbps";
170 			break;
171 		case SPEED_100:
172 			s = "100Mbps";
173 			break;
174 		}
175 
176 		netdev_info(dev, "link up, %s, %s-duplex\n",
177 			    s, p->link_config.duplex == DUPLEX_FULL
178 			    ? "full" : "half");
179 	}
180 }
181 
182 static void enable_tx_fifo_drain(struct adapter *adapter,
183 				 struct port_info *pi)
184 {
185 	t3_set_reg_field(adapter, A_XGM_TXFIFO_CFG + pi->mac.offset, 0,
186 			 F_ENDROPPKT);
187 	t3_write_reg(adapter, A_XGM_RX_CTRL + pi->mac.offset, 0);
188 	t3_write_reg(adapter, A_XGM_TX_CTRL + pi->mac.offset, F_TXEN);
189 	t3_write_reg(adapter, A_XGM_RX_CTRL + pi->mac.offset, F_RXEN);
190 }
191 
192 static void disable_tx_fifo_drain(struct adapter *adapter,
193 				  struct port_info *pi)
194 {
195 	t3_set_reg_field(adapter, A_XGM_TXFIFO_CFG + pi->mac.offset,
196 			 F_ENDROPPKT, 0);
197 }
198 
199 void t3_os_link_fault(struct adapter *adap, int port_id, int state)
200 {
201 	struct net_device *dev = adap->port[port_id];
202 	struct port_info *pi = netdev_priv(dev);
203 
204 	if (state == netif_carrier_ok(dev))
205 		return;
206 
207 	if (state) {
208 		struct cmac *mac = &pi->mac;
209 
210 		netif_carrier_on(dev);
211 
212 		disable_tx_fifo_drain(adap, pi);
213 
214 		/* Clear local faults */
215 		t3_xgm_intr_disable(adap, pi->port_id);
216 		t3_read_reg(adap, A_XGM_INT_STATUS +
217 				    pi->mac.offset);
218 		t3_write_reg(adap,
219 			     A_XGM_INT_CAUSE + pi->mac.offset,
220 			     F_XGM_INT);
221 
222 		t3_set_reg_field(adap,
223 				 A_XGM_INT_ENABLE +
224 				 pi->mac.offset,
225 				 F_XGM_INT, F_XGM_INT);
226 		t3_xgm_intr_enable(adap, pi->port_id);
227 
228 		t3_mac_enable(mac, MAC_DIRECTION_TX);
229 	} else {
230 		netif_carrier_off(dev);
231 
232 		/* Flush TX FIFO */
233 		enable_tx_fifo_drain(adap, pi);
234 	}
235 	link_report(dev);
236 }
237 
238 /**
239  *	t3_os_link_changed - handle link status changes
240  *	@adapter: the adapter associated with the link change
241  *	@port_id: the port index whose limk status has changed
242  *	@link_stat: the new status of the link
243  *	@speed: the new speed setting
244  *	@duplex: the new duplex setting
245  *	@pause: the new flow-control setting
246  *
247  *	This is the OS-dependent handler for link status changes.  The OS
248  *	neutral handler takes care of most of the processing for these events,
249  *	then calls this handler for any OS-specific processing.
250  */
251 void t3_os_link_changed(struct adapter *adapter, int port_id, int link_stat,
252 			int speed, int duplex, int pause)
253 {
254 	struct net_device *dev = adapter->port[port_id];
255 	struct port_info *pi = netdev_priv(dev);
256 	struct cmac *mac = &pi->mac;
257 
258 	/* Skip changes from disabled ports. */
259 	if (!netif_running(dev))
260 		return;
261 
262 	if (link_stat != netif_carrier_ok(dev)) {
263 		if (link_stat) {
264 			disable_tx_fifo_drain(adapter, pi);
265 
266 			t3_mac_enable(mac, MAC_DIRECTION_RX);
267 
268 			/* Clear local faults */
269 			t3_xgm_intr_disable(adapter, pi->port_id);
270 			t3_read_reg(adapter, A_XGM_INT_STATUS +
271 				    pi->mac.offset);
272 			t3_write_reg(adapter,
273 				     A_XGM_INT_CAUSE + pi->mac.offset,
274 				     F_XGM_INT);
275 
276 			t3_set_reg_field(adapter,
277 					 A_XGM_INT_ENABLE + pi->mac.offset,
278 					 F_XGM_INT, F_XGM_INT);
279 			t3_xgm_intr_enable(adapter, pi->port_id);
280 
281 			netif_carrier_on(dev);
282 		} else {
283 			netif_carrier_off(dev);
284 
285 			t3_xgm_intr_disable(adapter, pi->port_id);
286 			t3_read_reg(adapter, A_XGM_INT_STATUS + pi->mac.offset);
287 			t3_set_reg_field(adapter,
288 					 A_XGM_INT_ENABLE + pi->mac.offset,
289 					 F_XGM_INT, 0);
290 
291 			if (is_10G(adapter))
292 				pi->phy.ops->power_down(&pi->phy, 1);
293 
294 			t3_read_reg(adapter, A_XGM_INT_STATUS + pi->mac.offset);
295 			t3_mac_disable(mac, MAC_DIRECTION_RX);
296 			t3_link_start(&pi->phy, mac, &pi->link_config);
297 
298 			/* Flush TX FIFO */
299 			enable_tx_fifo_drain(adapter, pi);
300 		}
301 
302 		link_report(dev);
303 	}
304 }
305 
306 /**
307  *	t3_os_phymod_changed - handle PHY module changes
308  *	@phy: the PHY reporting the module change
309  *	@mod_type: new module type
310  *
311  *	This is the OS-dependent handler for PHY module changes.  It is
312  *	invoked when a PHY module is removed or inserted for any OS-specific
313  *	processing.
314  */
315 void t3_os_phymod_changed(struct adapter *adap, int port_id)
316 {
317 	static const char *mod_str[] = {
318 		NULL, "SR", "LR", "LRM", "TWINAX", "TWINAX", "unknown"
319 	};
320 
321 	const struct net_device *dev = adap->port[port_id];
322 	const struct port_info *pi = netdev_priv(dev);
323 
324 	if (pi->phy.modtype == phy_modtype_none)
325 		netdev_info(dev, "PHY module unplugged\n");
326 	else
327 		netdev_info(dev, "%s PHY module inserted\n",
328 			    mod_str[pi->phy.modtype]);
329 }
330 
331 static void cxgb_set_rxmode(struct net_device *dev)
332 {
333 	struct port_info *pi = netdev_priv(dev);
334 
335 	t3_mac_set_rx_mode(&pi->mac, dev);
336 }
337 
338 /**
339  *	link_start - enable a port
340  *	@dev: the device to enable
341  *
342  *	Performs the MAC and PHY actions needed to enable a port.
343  */
344 static void link_start(struct net_device *dev)
345 {
346 	struct port_info *pi = netdev_priv(dev);
347 	struct cmac *mac = &pi->mac;
348 
349 	t3_mac_reset(mac);
350 	t3_mac_set_num_ucast(mac, MAX_MAC_IDX);
351 	t3_mac_set_mtu(mac, dev->mtu);
352 	t3_mac_set_address(mac, LAN_MAC_IDX, dev->dev_addr);
353 	t3_mac_set_address(mac, SAN_MAC_IDX, pi->iscsic.mac_addr);
354 	t3_mac_set_rx_mode(mac, dev);
355 	t3_link_start(&pi->phy, mac, &pi->link_config);
356 	t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
357 }
358 
359 static inline void cxgb_disable_msi(struct adapter *adapter)
360 {
361 	if (adapter->flags & USING_MSIX) {
362 		pci_disable_msix(adapter->pdev);
363 		adapter->flags &= ~USING_MSIX;
364 	} else if (adapter->flags & USING_MSI) {
365 		pci_disable_msi(adapter->pdev);
366 		adapter->flags &= ~USING_MSI;
367 	}
368 }
369 
370 /*
371  * Interrupt handler for asynchronous events used with MSI-X.
372  */
373 static irqreturn_t t3_async_intr_handler(int irq, void *cookie)
374 {
375 	t3_slow_intr_handler(cookie);
376 	return IRQ_HANDLED;
377 }
378 
379 /*
380  * Name the MSI-X interrupts.
381  */
382 static void name_msix_vecs(struct adapter *adap)
383 {
384 	int i, j, msi_idx = 1, n = sizeof(adap->msix_info[0].desc) - 1;
385 
386 	snprintf(adap->msix_info[0].desc, n, "%s", adap->name);
387 	adap->msix_info[0].desc[n] = 0;
388 
389 	for_each_port(adap, j) {
390 		struct net_device *d = adap->port[j];
391 		const struct port_info *pi = netdev_priv(d);
392 
393 		for (i = 0; i < pi->nqsets; i++, msi_idx++) {
394 			snprintf(adap->msix_info[msi_idx].desc, n,
395 				 "%s-%d", d->name, pi->first_qset + i);
396 			adap->msix_info[msi_idx].desc[n] = 0;
397 		}
398 	}
399 }
400 
401 static int request_msix_data_irqs(struct adapter *adap)
402 {
403 	int i, j, err, qidx = 0;
404 
405 	for_each_port(adap, i) {
406 		int nqsets = adap2pinfo(adap, i)->nqsets;
407 
408 		for (j = 0; j < nqsets; ++j) {
409 			err = request_irq(adap->msix_info[qidx + 1].vec,
410 					  t3_intr_handler(adap,
411 							  adap->sge.qs[qidx].
412 							  rspq.polling), 0,
413 					  adap->msix_info[qidx + 1].desc,
414 					  &adap->sge.qs[qidx]);
415 			if (err) {
416 				while (--qidx >= 0)
417 					free_irq(adap->msix_info[qidx + 1].vec,
418 						 &adap->sge.qs[qidx]);
419 				return err;
420 			}
421 			qidx++;
422 		}
423 	}
424 	return 0;
425 }
426 
427 static void free_irq_resources(struct adapter *adapter)
428 {
429 	if (adapter->flags & USING_MSIX) {
430 		int i, n = 0;
431 
432 		free_irq(adapter->msix_info[0].vec, adapter);
433 		for_each_port(adapter, i)
434 			n += adap2pinfo(adapter, i)->nqsets;
435 
436 		for (i = 0; i < n; ++i)
437 			free_irq(adapter->msix_info[i + 1].vec,
438 				 &adapter->sge.qs[i]);
439 	} else
440 		free_irq(adapter->pdev->irq, adapter);
441 }
442 
443 static int await_mgmt_replies(struct adapter *adap, unsigned long init_cnt,
444 			      unsigned long n)
445 {
446 	int attempts = 10;
447 
448 	while (adap->sge.qs[0].rspq.offload_pkts < init_cnt + n) {
449 		if (!--attempts)
450 			return -ETIMEDOUT;
451 		msleep(10);
452 	}
453 	return 0;
454 }
455 
456 static int init_tp_parity(struct adapter *adap)
457 {
458 	int i;
459 	struct sk_buff *skb;
460 	struct cpl_set_tcb_field *greq;
461 	unsigned long cnt = adap->sge.qs[0].rspq.offload_pkts;
462 
463 	t3_tp_set_offload_mode(adap, 1);
464 
465 	for (i = 0; i < 16; i++) {
466 		struct cpl_smt_write_req *req;
467 
468 		skb = alloc_skb(sizeof(*req), GFP_KERNEL);
469 		if (!skb)
470 			skb = adap->nofail_skb;
471 		if (!skb)
472 			goto alloc_skb_fail;
473 
474 		req = (struct cpl_smt_write_req *)__skb_put(skb, sizeof(*req));
475 		memset(req, 0, sizeof(*req));
476 		req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
477 		OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, i));
478 		req->mtu_idx = NMTUS - 1;
479 		req->iff = i;
480 		t3_mgmt_tx(adap, skb);
481 		if (skb == adap->nofail_skb) {
482 			await_mgmt_replies(adap, cnt, i + 1);
483 			adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
484 			if (!adap->nofail_skb)
485 				goto alloc_skb_fail;
486 		}
487 	}
488 
489 	for (i = 0; i < 2048; i++) {
490 		struct cpl_l2t_write_req *req;
491 
492 		skb = alloc_skb(sizeof(*req), GFP_KERNEL);
493 		if (!skb)
494 			skb = adap->nofail_skb;
495 		if (!skb)
496 			goto alloc_skb_fail;
497 
498 		req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req));
499 		memset(req, 0, sizeof(*req));
500 		req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
501 		OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, i));
502 		req->params = htonl(V_L2T_W_IDX(i));
503 		t3_mgmt_tx(adap, skb);
504 		if (skb == adap->nofail_skb) {
505 			await_mgmt_replies(adap, cnt, 16 + i + 1);
506 			adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
507 			if (!adap->nofail_skb)
508 				goto alloc_skb_fail;
509 		}
510 	}
511 
512 	for (i = 0; i < 2048; i++) {
513 		struct cpl_rte_write_req *req;
514 
515 		skb = alloc_skb(sizeof(*req), GFP_KERNEL);
516 		if (!skb)
517 			skb = adap->nofail_skb;
518 		if (!skb)
519 			goto alloc_skb_fail;
520 
521 		req = (struct cpl_rte_write_req *)__skb_put(skb, sizeof(*req));
522 		memset(req, 0, sizeof(*req));
523 		req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
524 		OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RTE_WRITE_REQ, i));
525 		req->l2t_idx = htonl(V_L2T_W_IDX(i));
526 		t3_mgmt_tx(adap, skb);
527 		if (skb == adap->nofail_skb) {
528 			await_mgmt_replies(adap, cnt, 16 + 2048 + i + 1);
529 			adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
530 			if (!adap->nofail_skb)
531 				goto alloc_skb_fail;
532 		}
533 	}
534 
535 	skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
536 	if (!skb)
537 		skb = adap->nofail_skb;
538 	if (!skb)
539 		goto alloc_skb_fail;
540 
541 	greq = (struct cpl_set_tcb_field *)__skb_put(skb, sizeof(*greq));
542 	memset(greq, 0, sizeof(*greq));
543 	greq->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
544 	OPCODE_TID(greq) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, 0));
545 	greq->mask = cpu_to_be64(1);
546 	t3_mgmt_tx(adap, skb);
547 
548 	i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1);
549 	if (skb == adap->nofail_skb) {
550 		i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1);
551 		adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
552 	}
553 
554 	t3_tp_set_offload_mode(adap, 0);
555 	return i;
556 
557 alloc_skb_fail:
558 	t3_tp_set_offload_mode(adap, 0);
559 	return -ENOMEM;
560 }
561 
562 /**
563  *	setup_rss - configure RSS
564  *	@adap: the adapter
565  *
566  *	Sets up RSS to distribute packets to multiple receive queues.  We
567  *	configure the RSS CPU lookup table to distribute to the number of HW
568  *	receive queues, and the response queue lookup table to narrow that
569  *	down to the response queues actually configured for each port.
570  *	We always configure the RSS mapping for two ports since the mapping
571  *	table has plenty of entries.
572  */
573 static void setup_rss(struct adapter *adap)
574 {
575 	int i;
576 	unsigned int nq0 = adap2pinfo(adap, 0)->nqsets;
577 	unsigned int nq1 = adap->port[1] ? adap2pinfo(adap, 1)->nqsets : 1;
578 	u8 cpus[SGE_QSETS + 1];
579 	u16 rspq_map[RSS_TABLE_SIZE + 1];
580 
581 	for (i = 0; i < SGE_QSETS; ++i)
582 		cpus[i] = i;
583 	cpus[SGE_QSETS] = 0xff;	/* terminator */
584 
585 	for (i = 0; i < RSS_TABLE_SIZE / 2; ++i) {
586 		rspq_map[i] = i % nq0;
587 		rspq_map[i + RSS_TABLE_SIZE / 2] = (i % nq1) + nq0;
588 	}
589 	rspq_map[RSS_TABLE_SIZE] = 0xffff; /* terminator */
590 
591 	t3_config_rss(adap, F_RQFEEDBACKENABLE | F_TNLLKPEN | F_TNLMAPEN |
592 		      F_TNLPRTEN | F_TNL2TUPEN | F_TNL4TUPEN |
593 		      V_RRCPLCPUSIZE(6) | F_HASHTOEPLITZ, cpus, rspq_map);
594 }
595 
596 static void ring_dbs(struct adapter *adap)
597 {
598 	int i, j;
599 
600 	for (i = 0; i < SGE_QSETS; i++) {
601 		struct sge_qset *qs = &adap->sge.qs[i];
602 
603 		if (qs->adap)
604 			for (j = 0; j < SGE_TXQ_PER_SET; j++)
605 				t3_write_reg(adap, A_SG_KDOORBELL, F_SELEGRCNTX | V_EGRCNTX(qs->txq[j].cntxt_id));
606 	}
607 }
608 
609 static void init_napi(struct adapter *adap)
610 {
611 	int i;
612 
613 	for (i = 0; i < SGE_QSETS; i++) {
614 		struct sge_qset *qs = &adap->sge.qs[i];
615 
616 		if (qs->adap)
617 			netif_napi_add(qs->netdev, &qs->napi, qs->napi.poll,
618 				       64);
619 	}
620 
621 	/*
622 	 * netif_napi_add() can be called only once per napi_struct because it
623 	 * adds each new napi_struct to a list.  Be careful not to call it a
624 	 * second time, e.g., during EEH recovery, by making a note of it.
625 	 */
626 	adap->flags |= NAPI_INIT;
627 }
628 
629 /*
630  * Wait until all NAPI handlers are descheduled.  This includes the handlers of
631  * both netdevices representing interfaces and the dummy ones for the extra
632  * queues.
633  */
634 static void quiesce_rx(struct adapter *adap)
635 {
636 	int i;
637 
638 	for (i = 0; i < SGE_QSETS; i++)
639 		if (adap->sge.qs[i].adap)
640 			napi_disable(&adap->sge.qs[i].napi);
641 }
642 
643 static void enable_all_napi(struct adapter *adap)
644 {
645 	int i;
646 	for (i = 0; i < SGE_QSETS; i++)
647 		if (adap->sge.qs[i].adap)
648 			napi_enable(&adap->sge.qs[i].napi);
649 }
650 
651 /**
652  *	setup_sge_qsets - configure SGE Tx/Rx/response queues
653  *	@adap: the adapter
654  *
655  *	Determines how many sets of SGE queues to use and initializes them.
656  *	We support multiple queue sets per port if we have MSI-X, otherwise
657  *	just one queue set per port.
658  */
659 static int setup_sge_qsets(struct adapter *adap)
660 {
661 	int i, j, err, irq_idx = 0, qset_idx = 0;
662 	unsigned int ntxq = SGE_TXQ_PER_SET;
663 
664 	if (adap->params.rev > 0 && !(adap->flags & USING_MSI))
665 		irq_idx = -1;
666 
667 	for_each_port(adap, i) {
668 		struct net_device *dev = adap->port[i];
669 		struct port_info *pi = netdev_priv(dev);
670 
671 		pi->qs = &adap->sge.qs[pi->first_qset];
672 		for (j = 0; j < pi->nqsets; ++j, ++qset_idx) {
673 			err = t3_sge_alloc_qset(adap, qset_idx, 1,
674 				(adap->flags & USING_MSIX) ? qset_idx + 1 :
675 							     irq_idx,
676 				&adap->params.sge.qset[qset_idx], ntxq, dev,
677 				netdev_get_tx_queue(dev, j));
678 			if (err) {
679 				t3_free_sge_resources(adap);
680 				return err;
681 			}
682 		}
683 	}
684 
685 	return 0;
686 }
687 
688 static ssize_t attr_show(struct device *d, char *buf,
689 			 ssize_t(*format) (struct net_device *, char *))
690 {
691 	ssize_t len;
692 
693 	/* Synchronize with ioctls that may shut down the device */
694 	rtnl_lock();
695 	len = (*format) (to_net_dev(d), buf);
696 	rtnl_unlock();
697 	return len;
698 }
699 
700 static ssize_t attr_store(struct device *d,
701 			  const char *buf, size_t len,
702 			  ssize_t(*set) (struct net_device *, unsigned int),
703 			  unsigned int min_val, unsigned int max_val)
704 {
705 	ssize_t ret;
706 	unsigned int val;
707 
708 	if (!capable(CAP_NET_ADMIN))
709 		return -EPERM;
710 
711 	ret = kstrtouint(buf, 0, &val);
712 	if (ret)
713 		return ret;
714 	if (val < min_val || val > max_val)
715 		return -EINVAL;
716 
717 	rtnl_lock();
718 	ret = (*set) (to_net_dev(d), val);
719 	if (!ret)
720 		ret = len;
721 	rtnl_unlock();
722 	return ret;
723 }
724 
725 #define CXGB3_SHOW(name, val_expr) \
726 static ssize_t format_##name(struct net_device *dev, char *buf) \
727 { \
728 	struct port_info *pi = netdev_priv(dev); \
729 	struct adapter *adap = pi->adapter; \
730 	return sprintf(buf, "%u\n", val_expr); \
731 } \
732 static ssize_t show_##name(struct device *d, struct device_attribute *attr, \
733 			   char *buf) \
734 { \
735 	return attr_show(d, buf, format_##name); \
736 }
737 
738 static ssize_t set_nfilters(struct net_device *dev, unsigned int val)
739 {
740 	struct port_info *pi = netdev_priv(dev);
741 	struct adapter *adap = pi->adapter;
742 	int min_tids = is_offload(adap) ? MC5_MIN_TIDS : 0;
743 
744 	if (adap->flags & FULL_INIT_DONE)
745 		return -EBUSY;
746 	if (val && adap->params.rev == 0)
747 		return -EINVAL;
748 	if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nservers -
749 	    min_tids)
750 		return -EINVAL;
751 	adap->params.mc5.nfilters = val;
752 	return 0;
753 }
754 
755 static ssize_t store_nfilters(struct device *d, struct device_attribute *attr,
756 			      const char *buf, size_t len)
757 {
758 	return attr_store(d, buf, len, set_nfilters, 0, ~0);
759 }
760 
761 static ssize_t set_nservers(struct net_device *dev, unsigned int val)
762 {
763 	struct port_info *pi = netdev_priv(dev);
764 	struct adapter *adap = pi->adapter;
765 
766 	if (adap->flags & FULL_INIT_DONE)
767 		return -EBUSY;
768 	if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nfilters -
769 	    MC5_MIN_TIDS)
770 		return -EINVAL;
771 	adap->params.mc5.nservers = val;
772 	return 0;
773 }
774 
775 static ssize_t store_nservers(struct device *d, struct device_attribute *attr,
776 			      const char *buf, size_t len)
777 {
778 	return attr_store(d, buf, len, set_nservers, 0, ~0);
779 }
780 
781 #define CXGB3_ATTR_R(name, val_expr) \
782 CXGB3_SHOW(name, val_expr) \
783 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
784 
785 #define CXGB3_ATTR_RW(name, val_expr, store_method) \
786 CXGB3_SHOW(name, val_expr) \
787 static DEVICE_ATTR(name, S_IRUGO | S_IWUSR, show_##name, store_method)
788 
789 CXGB3_ATTR_R(cam_size, t3_mc5_size(&adap->mc5));
790 CXGB3_ATTR_RW(nfilters, adap->params.mc5.nfilters, store_nfilters);
791 CXGB3_ATTR_RW(nservers, adap->params.mc5.nservers, store_nservers);
792 
793 static struct attribute *cxgb3_attrs[] = {
794 	&dev_attr_cam_size.attr,
795 	&dev_attr_nfilters.attr,
796 	&dev_attr_nservers.attr,
797 	NULL
798 };
799 
800 static struct attribute_group cxgb3_attr_group = {.attrs = cxgb3_attrs };
801 
802 static ssize_t tm_attr_show(struct device *d,
803 			    char *buf, int sched)
804 {
805 	struct port_info *pi = netdev_priv(to_net_dev(d));
806 	struct adapter *adap = pi->adapter;
807 	unsigned int v, addr, bpt, cpt;
808 	ssize_t len;
809 
810 	addr = A_TP_TX_MOD_Q1_Q0_RATE_LIMIT - sched / 2;
811 	rtnl_lock();
812 	t3_write_reg(adap, A_TP_TM_PIO_ADDR, addr);
813 	v = t3_read_reg(adap, A_TP_TM_PIO_DATA);
814 	if (sched & 1)
815 		v >>= 16;
816 	bpt = (v >> 8) & 0xff;
817 	cpt = v & 0xff;
818 	if (!cpt)
819 		len = sprintf(buf, "disabled\n");
820 	else {
821 		v = (adap->params.vpd.cclk * 1000) / cpt;
822 		len = sprintf(buf, "%u Kbps\n", (v * bpt) / 125);
823 	}
824 	rtnl_unlock();
825 	return len;
826 }
827 
828 static ssize_t tm_attr_store(struct device *d,
829 			     const char *buf, size_t len, int sched)
830 {
831 	struct port_info *pi = netdev_priv(to_net_dev(d));
832 	struct adapter *adap = pi->adapter;
833 	unsigned int val;
834 	ssize_t ret;
835 
836 	if (!capable(CAP_NET_ADMIN))
837 		return -EPERM;
838 
839 	ret = kstrtouint(buf, 0, &val);
840 	if (ret)
841 		return ret;
842 	if (val > 10000000)
843 		return -EINVAL;
844 
845 	rtnl_lock();
846 	ret = t3_config_sched(adap, val, sched);
847 	if (!ret)
848 		ret = len;
849 	rtnl_unlock();
850 	return ret;
851 }
852 
853 #define TM_ATTR(name, sched) \
854 static ssize_t show_##name(struct device *d, struct device_attribute *attr, \
855 			   char *buf) \
856 { \
857 	return tm_attr_show(d, buf, sched); \
858 } \
859 static ssize_t store_##name(struct device *d, struct device_attribute *attr, \
860 			    const char *buf, size_t len) \
861 { \
862 	return tm_attr_store(d, buf, len, sched); \
863 } \
864 static DEVICE_ATTR(name, S_IRUGO | S_IWUSR, show_##name, store_##name)
865 
866 TM_ATTR(sched0, 0);
867 TM_ATTR(sched1, 1);
868 TM_ATTR(sched2, 2);
869 TM_ATTR(sched3, 3);
870 TM_ATTR(sched4, 4);
871 TM_ATTR(sched5, 5);
872 TM_ATTR(sched6, 6);
873 TM_ATTR(sched7, 7);
874 
875 static struct attribute *offload_attrs[] = {
876 	&dev_attr_sched0.attr,
877 	&dev_attr_sched1.attr,
878 	&dev_attr_sched2.attr,
879 	&dev_attr_sched3.attr,
880 	&dev_attr_sched4.attr,
881 	&dev_attr_sched5.attr,
882 	&dev_attr_sched6.attr,
883 	&dev_attr_sched7.attr,
884 	NULL
885 };
886 
887 static struct attribute_group offload_attr_group = {.attrs = offload_attrs };
888 
889 /*
890  * Sends an sk_buff to an offload queue driver
891  * after dealing with any active network taps.
892  */
893 static inline int offload_tx(struct t3cdev *tdev, struct sk_buff *skb)
894 {
895 	int ret;
896 
897 	local_bh_disable();
898 	ret = t3_offload_tx(tdev, skb);
899 	local_bh_enable();
900 	return ret;
901 }
902 
903 static int write_smt_entry(struct adapter *adapter, int idx)
904 {
905 	struct cpl_smt_write_req *req;
906 	struct port_info *pi = netdev_priv(adapter->port[idx]);
907 	struct sk_buff *skb = alloc_skb(sizeof(*req), GFP_KERNEL);
908 
909 	if (!skb)
910 		return -ENOMEM;
911 
912 	req = (struct cpl_smt_write_req *)__skb_put(skb, sizeof(*req));
913 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
914 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, idx));
915 	req->mtu_idx = NMTUS - 1;	/* should be 0 but there's a T3 bug */
916 	req->iff = idx;
917 	memcpy(req->src_mac0, adapter->port[idx]->dev_addr, ETH_ALEN);
918 	memcpy(req->src_mac1, pi->iscsic.mac_addr, ETH_ALEN);
919 	skb->priority = 1;
920 	offload_tx(&adapter->tdev, skb);
921 	return 0;
922 }
923 
924 static int init_smt(struct adapter *adapter)
925 {
926 	int i;
927 
928 	for_each_port(adapter, i)
929 	    write_smt_entry(adapter, i);
930 	return 0;
931 }
932 
933 static void init_port_mtus(struct adapter *adapter)
934 {
935 	unsigned int mtus = adapter->port[0]->mtu;
936 
937 	if (adapter->port[1])
938 		mtus |= adapter->port[1]->mtu << 16;
939 	t3_write_reg(adapter, A_TP_MTU_PORT_TABLE, mtus);
940 }
941 
942 static int send_pktsched_cmd(struct adapter *adap, int sched, int qidx, int lo,
943 			      int hi, int port)
944 {
945 	struct sk_buff *skb;
946 	struct mngt_pktsched_wr *req;
947 	int ret;
948 
949 	skb = alloc_skb(sizeof(*req), GFP_KERNEL);
950 	if (!skb)
951 		skb = adap->nofail_skb;
952 	if (!skb)
953 		return -ENOMEM;
954 
955 	req = (struct mngt_pktsched_wr *)skb_put(skb, sizeof(*req));
956 	req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_MNGT));
957 	req->mngt_opcode = FW_MNGTOPCODE_PKTSCHED_SET;
958 	req->sched = sched;
959 	req->idx = qidx;
960 	req->min = lo;
961 	req->max = hi;
962 	req->binding = port;
963 	ret = t3_mgmt_tx(adap, skb);
964 	if (skb == adap->nofail_skb) {
965 		adap->nofail_skb = alloc_skb(sizeof(struct cpl_set_tcb_field),
966 					     GFP_KERNEL);
967 		if (!adap->nofail_skb)
968 			ret = -ENOMEM;
969 	}
970 
971 	return ret;
972 }
973 
974 static int bind_qsets(struct adapter *adap)
975 {
976 	int i, j, err = 0;
977 
978 	for_each_port(adap, i) {
979 		const struct port_info *pi = adap2pinfo(adap, i);
980 
981 		for (j = 0; j < pi->nqsets; ++j) {
982 			int ret = send_pktsched_cmd(adap, 1,
983 						    pi->first_qset + j, -1,
984 						    -1, i);
985 			if (ret)
986 				err = ret;
987 		}
988 	}
989 
990 	return err;
991 }
992 
993 #define FW_VERSION __stringify(FW_VERSION_MAJOR) "."			\
994 	__stringify(FW_VERSION_MINOR) "." __stringify(FW_VERSION_MICRO)
995 #define FW_FNAME "cxgb3/t3fw-" FW_VERSION ".bin"
996 #define TPSRAM_VERSION __stringify(TP_VERSION_MAJOR) "."		\
997 	__stringify(TP_VERSION_MINOR) "." __stringify(TP_VERSION_MICRO)
998 #define TPSRAM_NAME "cxgb3/t3%c_psram-" TPSRAM_VERSION ".bin"
999 #define AEL2005_OPT_EDC_NAME "cxgb3/ael2005_opt_edc.bin"
1000 #define AEL2005_TWX_EDC_NAME "cxgb3/ael2005_twx_edc.bin"
1001 #define AEL2020_TWX_EDC_NAME "cxgb3/ael2020_twx_edc.bin"
1002 MODULE_FIRMWARE(FW_FNAME);
1003 MODULE_FIRMWARE("cxgb3/t3b_psram-" TPSRAM_VERSION ".bin");
1004 MODULE_FIRMWARE("cxgb3/t3c_psram-" TPSRAM_VERSION ".bin");
1005 MODULE_FIRMWARE(AEL2005_OPT_EDC_NAME);
1006 MODULE_FIRMWARE(AEL2005_TWX_EDC_NAME);
1007 MODULE_FIRMWARE(AEL2020_TWX_EDC_NAME);
1008 
1009 static inline const char *get_edc_fw_name(int edc_idx)
1010 {
1011 	const char *fw_name = NULL;
1012 
1013 	switch (edc_idx) {
1014 	case EDC_OPT_AEL2005:
1015 		fw_name = AEL2005_OPT_EDC_NAME;
1016 		break;
1017 	case EDC_TWX_AEL2005:
1018 		fw_name = AEL2005_TWX_EDC_NAME;
1019 		break;
1020 	case EDC_TWX_AEL2020:
1021 		fw_name = AEL2020_TWX_EDC_NAME;
1022 		break;
1023 	}
1024 	return fw_name;
1025 }
1026 
1027 int t3_get_edc_fw(struct cphy *phy, int edc_idx, int size)
1028 {
1029 	struct adapter *adapter = phy->adapter;
1030 	const struct firmware *fw;
1031 	const char *fw_name;
1032 	u32 csum;
1033 	const __be32 *p;
1034 	u16 *cache = phy->phy_cache;
1035 	int i, ret = -EINVAL;
1036 
1037 	fw_name = get_edc_fw_name(edc_idx);
1038 	if (fw_name)
1039 		ret = request_firmware(&fw, fw_name, &adapter->pdev->dev);
1040 	if (ret < 0) {
1041 		dev_err(&adapter->pdev->dev,
1042 			"could not upgrade firmware: unable to load %s\n",
1043 			fw_name);
1044 		return ret;
1045 	}
1046 
1047 	/* check size, take checksum in account */
1048 	if (fw->size > size + 4) {
1049 		CH_ERR(adapter, "firmware image too large %u, expected %d\n",
1050 		       (unsigned int)fw->size, size + 4);
1051 		ret = -EINVAL;
1052 	}
1053 
1054 	/* compute checksum */
1055 	p = (const __be32 *)fw->data;
1056 	for (csum = 0, i = 0; i < fw->size / sizeof(csum); i++)
1057 		csum += ntohl(p[i]);
1058 
1059 	if (csum != 0xffffffff) {
1060 		CH_ERR(adapter, "corrupted firmware image, checksum %u\n",
1061 		       csum);
1062 		ret = -EINVAL;
1063 	}
1064 
1065 	for (i = 0; i < size / 4 ; i++) {
1066 		*cache++ = (be32_to_cpu(p[i]) & 0xffff0000) >> 16;
1067 		*cache++ = be32_to_cpu(p[i]) & 0xffff;
1068 	}
1069 
1070 	release_firmware(fw);
1071 
1072 	return ret;
1073 }
1074 
1075 static int upgrade_fw(struct adapter *adap)
1076 {
1077 	int ret;
1078 	const struct firmware *fw;
1079 	struct device *dev = &adap->pdev->dev;
1080 
1081 	ret = request_firmware(&fw, FW_FNAME, dev);
1082 	if (ret < 0) {
1083 		dev_err(dev, "could not upgrade firmware: unable to load %s\n",
1084 			FW_FNAME);
1085 		return ret;
1086 	}
1087 	ret = t3_load_fw(adap, fw->data, fw->size);
1088 	release_firmware(fw);
1089 
1090 	if (ret == 0)
1091 		dev_info(dev, "successful upgrade to firmware %d.%d.%d\n",
1092 			 FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
1093 	else
1094 		dev_err(dev, "failed to upgrade to firmware %d.%d.%d\n",
1095 			FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
1096 
1097 	return ret;
1098 }
1099 
1100 static inline char t3rev2char(struct adapter *adapter)
1101 {
1102 	char rev = 0;
1103 
1104 	switch(adapter->params.rev) {
1105 	case T3_REV_B:
1106 	case T3_REV_B2:
1107 		rev = 'b';
1108 		break;
1109 	case T3_REV_C:
1110 		rev = 'c';
1111 		break;
1112 	}
1113 	return rev;
1114 }
1115 
1116 static int update_tpsram(struct adapter *adap)
1117 {
1118 	const struct firmware *tpsram;
1119 	char buf[64];
1120 	struct device *dev = &adap->pdev->dev;
1121 	int ret;
1122 	char rev;
1123 
1124 	rev = t3rev2char(adap);
1125 	if (!rev)
1126 		return 0;
1127 
1128 	snprintf(buf, sizeof(buf), TPSRAM_NAME, rev);
1129 
1130 	ret = request_firmware(&tpsram, buf, dev);
1131 	if (ret < 0) {
1132 		dev_err(dev, "could not load TP SRAM: unable to load %s\n",
1133 			buf);
1134 		return ret;
1135 	}
1136 
1137 	ret = t3_check_tpsram(adap, tpsram->data, tpsram->size);
1138 	if (ret)
1139 		goto release_tpsram;
1140 
1141 	ret = t3_set_proto_sram(adap, tpsram->data);
1142 	if (ret == 0)
1143 		dev_info(dev,
1144 			 "successful update of protocol engine "
1145 			 "to %d.%d.%d\n",
1146 			 TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
1147 	else
1148 		dev_err(dev, "failed to update of protocol engine %d.%d.%d\n",
1149 			TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
1150 	if (ret)
1151 		dev_err(dev, "loading protocol SRAM failed\n");
1152 
1153 release_tpsram:
1154 	release_firmware(tpsram);
1155 
1156 	return ret;
1157 }
1158 
1159 /**
1160  * t3_synchronize_rx - wait for current Rx processing on a port to complete
1161  * @adap: the adapter
1162  * @p: the port
1163  *
1164  * Ensures that current Rx processing on any of the queues associated with
1165  * the given port completes before returning.  We do this by acquiring and
1166  * releasing the locks of the response queues associated with the port.
1167  */
1168 static void t3_synchronize_rx(struct adapter *adap, const struct port_info *p)
1169 {
1170 	int i;
1171 
1172 	for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
1173 		struct sge_rspq *q = &adap->sge.qs[i].rspq;
1174 
1175 		spin_lock_irq(&q->lock);
1176 		spin_unlock_irq(&q->lock);
1177 	}
1178 }
1179 
1180 static void cxgb_vlan_mode(struct net_device *dev, netdev_features_t features)
1181 {
1182 	struct port_info *pi = netdev_priv(dev);
1183 	struct adapter *adapter = pi->adapter;
1184 
1185 	if (adapter->params.rev > 0) {
1186 		t3_set_vlan_accel(adapter, 1 << pi->port_id,
1187 				  features & NETIF_F_HW_VLAN_CTAG_RX);
1188 	} else {
1189 		/* single control for all ports */
1190 		unsigned int i, have_vlans = features & NETIF_F_HW_VLAN_CTAG_RX;
1191 
1192 		for_each_port(adapter, i)
1193 			have_vlans |=
1194 				adapter->port[i]->features &
1195 				NETIF_F_HW_VLAN_CTAG_RX;
1196 
1197 		t3_set_vlan_accel(adapter, 1, have_vlans);
1198 	}
1199 	t3_synchronize_rx(adapter, pi);
1200 }
1201 
1202 /**
1203  *	cxgb_up - enable the adapter
1204  *	@adapter: adapter being enabled
1205  *
1206  *	Called when the first port is enabled, this function performs the
1207  *	actions necessary to make an adapter operational, such as completing
1208  *	the initialization of HW modules, and enabling interrupts.
1209  *
1210  *	Must be called with the rtnl lock held.
1211  */
1212 static int cxgb_up(struct adapter *adap)
1213 {
1214 	int i, err;
1215 
1216 	if (!(adap->flags & FULL_INIT_DONE)) {
1217 		err = t3_check_fw_version(adap);
1218 		if (err == -EINVAL) {
1219 			err = upgrade_fw(adap);
1220 			CH_WARN(adap, "FW upgrade to %d.%d.%d %s\n",
1221 				FW_VERSION_MAJOR, FW_VERSION_MINOR,
1222 				FW_VERSION_MICRO, err ? "failed" : "succeeded");
1223 		}
1224 
1225 		err = t3_check_tpsram_version(adap);
1226 		if (err == -EINVAL) {
1227 			err = update_tpsram(adap);
1228 			CH_WARN(adap, "TP upgrade to %d.%d.%d %s\n",
1229 				TP_VERSION_MAJOR, TP_VERSION_MINOR,
1230 				TP_VERSION_MICRO, err ? "failed" : "succeeded");
1231 		}
1232 
1233 		/*
1234 		 * Clear interrupts now to catch errors if t3_init_hw fails.
1235 		 * We clear them again later as initialization may trigger
1236 		 * conditions that can interrupt.
1237 		 */
1238 		t3_intr_clear(adap);
1239 
1240 		err = t3_init_hw(adap, 0);
1241 		if (err)
1242 			goto out;
1243 
1244 		t3_set_reg_field(adap, A_TP_PARA_REG5, 0, F_RXDDPOFFINIT);
1245 		t3_write_reg(adap, A_ULPRX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12));
1246 
1247 		err = setup_sge_qsets(adap);
1248 		if (err)
1249 			goto out;
1250 
1251 		for_each_port(adap, i)
1252 			cxgb_vlan_mode(adap->port[i], adap->port[i]->features);
1253 
1254 		setup_rss(adap);
1255 		if (!(adap->flags & NAPI_INIT))
1256 			init_napi(adap);
1257 
1258 		t3_start_sge_timers(adap);
1259 		adap->flags |= FULL_INIT_DONE;
1260 	}
1261 
1262 	t3_intr_clear(adap);
1263 
1264 	if (adap->flags & USING_MSIX) {
1265 		name_msix_vecs(adap);
1266 		err = request_irq(adap->msix_info[0].vec,
1267 				  t3_async_intr_handler, 0,
1268 				  adap->msix_info[0].desc, adap);
1269 		if (err)
1270 			goto irq_err;
1271 
1272 		err = request_msix_data_irqs(adap);
1273 		if (err) {
1274 			free_irq(adap->msix_info[0].vec, adap);
1275 			goto irq_err;
1276 		}
1277 	} else if ((err = request_irq(adap->pdev->irq,
1278 				      t3_intr_handler(adap,
1279 						      adap->sge.qs[0].rspq.
1280 						      polling),
1281 				      (adap->flags & USING_MSI) ?
1282 				       0 : IRQF_SHARED,
1283 				      adap->name, adap)))
1284 		goto irq_err;
1285 
1286 	enable_all_napi(adap);
1287 	t3_sge_start(adap);
1288 	t3_intr_enable(adap);
1289 
1290 	if (adap->params.rev >= T3_REV_C && !(adap->flags & TP_PARITY_INIT) &&
1291 	    is_offload(adap) && init_tp_parity(adap) == 0)
1292 		adap->flags |= TP_PARITY_INIT;
1293 
1294 	if (adap->flags & TP_PARITY_INIT) {
1295 		t3_write_reg(adap, A_TP_INT_CAUSE,
1296 			     F_CMCACHEPERR | F_ARPLUTPERR);
1297 		t3_write_reg(adap, A_TP_INT_ENABLE, 0x7fbfffff);
1298 	}
1299 
1300 	if (!(adap->flags & QUEUES_BOUND)) {
1301 		int ret = bind_qsets(adap);
1302 
1303 		if (ret < 0) {
1304 			CH_ERR(adap, "failed to bind qsets, err %d\n", ret);
1305 			t3_intr_disable(adap);
1306 			free_irq_resources(adap);
1307 			err = ret;
1308 			goto out;
1309 		}
1310 		adap->flags |= QUEUES_BOUND;
1311 	}
1312 
1313 out:
1314 	return err;
1315 irq_err:
1316 	CH_ERR(adap, "request_irq failed, err %d\n", err);
1317 	goto out;
1318 }
1319 
1320 /*
1321  * Release resources when all the ports and offloading have been stopped.
1322  */
1323 static void cxgb_down(struct adapter *adapter, int on_wq)
1324 {
1325 	t3_sge_stop(adapter);
1326 	spin_lock_irq(&adapter->work_lock);	/* sync with PHY intr task */
1327 	t3_intr_disable(adapter);
1328 	spin_unlock_irq(&adapter->work_lock);
1329 
1330 	free_irq_resources(adapter);
1331 	quiesce_rx(adapter);
1332 	t3_sge_stop(adapter);
1333 	if (!on_wq)
1334 		flush_workqueue(cxgb3_wq);/* wait for external IRQ handler */
1335 }
1336 
1337 static void schedule_chk_task(struct adapter *adap)
1338 {
1339 	unsigned int timeo;
1340 
1341 	timeo = adap->params.linkpoll_period ?
1342 	    (HZ * adap->params.linkpoll_period) / 10 :
1343 	    adap->params.stats_update_period * HZ;
1344 	if (timeo)
1345 		queue_delayed_work(cxgb3_wq, &adap->adap_check_task, timeo);
1346 }
1347 
1348 static int offload_open(struct net_device *dev)
1349 {
1350 	struct port_info *pi = netdev_priv(dev);
1351 	struct adapter *adapter = pi->adapter;
1352 	struct t3cdev *tdev = dev2t3cdev(dev);
1353 	int adap_up = adapter->open_device_map & PORT_MASK;
1354 	int err;
1355 
1356 	if (test_and_set_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
1357 		return 0;
1358 
1359 	if (!adap_up && (err = cxgb_up(adapter)) < 0)
1360 		goto out;
1361 
1362 	t3_tp_set_offload_mode(adapter, 1);
1363 	tdev->lldev = adapter->port[0];
1364 	err = cxgb3_offload_activate(adapter);
1365 	if (err)
1366 		goto out;
1367 
1368 	init_port_mtus(adapter);
1369 	t3_load_mtus(adapter, adapter->params.mtus, adapter->params.a_wnd,
1370 		     adapter->params.b_wnd,
1371 		     adapter->params.rev == 0 ?
1372 		     adapter->port[0]->mtu : 0xffff);
1373 	init_smt(adapter);
1374 
1375 	if (sysfs_create_group(&tdev->lldev->dev.kobj, &offload_attr_group))
1376 		dev_dbg(&dev->dev, "cannot create sysfs group\n");
1377 
1378 	/* Call back all registered clients */
1379 	cxgb3_add_clients(tdev);
1380 
1381 out:
1382 	/* restore them in case the offload module has changed them */
1383 	if (err) {
1384 		t3_tp_set_offload_mode(adapter, 0);
1385 		clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
1386 		cxgb3_set_dummy_ops(tdev);
1387 	}
1388 	return err;
1389 }
1390 
1391 static int offload_close(struct t3cdev *tdev)
1392 {
1393 	struct adapter *adapter = tdev2adap(tdev);
1394 	struct t3c_data *td = T3C_DATA(tdev);
1395 
1396 	if (!test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
1397 		return 0;
1398 
1399 	/* Call back all registered clients */
1400 	cxgb3_remove_clients(tdev);
1401 
1402 	sysfs_remove_group(&tdev->lldev->dev.kobj, &offload_attr_group);
1403 
1404 	/* Flush work scheduled while releasing TIDs */
1405 	flush_work(&td->tid_release_task);
1406 
1407 	tdev->lldev = NULL;
1408 	cxgb3_set_dummy_ops(tdev);
1409 	t3_tp_set_offload_mode(adapter, 0);
1410 	clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
1411 
1412 	if (!adapter->open_device_map)
1413 		cxgb_down(adapter, 0);
1414 
1415 	cxgb3_offload_deactivate(adapter);
1416 	return 0;
1417 }
1418 
1419 static int cxgb_open(struct net_device *dev)
1420 {
1421 	struct port_info *pi = netdev_priv(dev);
1422 	struct adapter *adapter = pi->adapter;
1423 	int other_ports = adapter->open_device_map & PORT_MASK;
1424 	int err;
1425 
1426 	if (!adapter->open_device_map && (err = cxgb_up(adapter)) < 0)
1427 		return err;
1428 
1429 	set_bit(pi->port_id, &adapter->open_device_map);
1430 	if (is_offload(adapter) && !ofld_disable) {
1431 		err = offload_open(dev);
1432 		if (err)
1433 			pr_warn("Could not initialize offload capabilities\n");
1434 	}
1435 
1436 	netif_set_real_num_tx_queues(dev, pi->nqsets);
1437 	err = netif_set_real_num_rx_queues(dev, pi->nqsets);
1438 	if (err)
1439 		return err;
1440 	link_start(dev);
1441 	t3_port_intr_enable(adapter, pi->port_id);
1442 	netif_tx_start_all_queues(dev);
1443 	if (!other_ports)
1444 		schedule_chk_task(adapter);
1445 
1446 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_PORT_UP, pi->port_id);
1447 	return 0;
1448 }
1449 
1450 static int __cxgb_close(struct net_device *dev, int on_wq)
1451 {
1452 	struct port_info *pi = netdev_priv(dev);
1453 	struct adapter *adapter = pi->adapter;
1454 
1455 
1456 	if (!adapter->open_device_map)
1457 		return 0;
1458 
1459 	/* Stop link fault interrupts */
1460 	t3_xgm_intr_disable(adapter, pi->port_id);
1461 	t3_read_reg(adapter, A_XGM_INT_STATUS + pi->mac.offset);
1462 
1463 	t3_port_intr_disable(adapter, pi->port_id);
1464 	netif_tx_stop_all_queues(dev);
1465 	pi->phy.ops->power_down(&pi->phy, 1);
1466 	netif_carrier_off(dev);
1467 	t3_mac_disable(&pi->mac, MAC_DIRECTION_TX | MAC_DIRECTION_RX);
1468 
1469 	spin_lock_irq(&adapter->work_lock);	/* sync with update task */
1470 	clear_bit(pi->port_id, &adapter->open_device_map);
1471 	spin_unlock_irq(&adapter->work_lock);
1472 
1473 	if (!(adapter->open_device_map & PORT_MASK))
1474 		cancel_delayed_work_sync(&adapter->adap_check_task);
1475 
1476 	if (!adapter->open_device_map)
1477 		cxgb_down(adapter, on_wq);
1478 
1479 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_PORT_DOWN, pi->port_id);
1480 	return 0;
1481 }
1482 
1483 static int cxgb_close(struct net_device *dev)
1484 {
1485 	return __cxgb_close(dev, 0);
1486 }
1487 
1488 static struct net_device_stats *cxgb_get_stats(struct net_device *dev)
1489 {
1490 	struct port_info *pi = netdev_priv(dev);
1491 	struct adapter *adapter = pi->adapter;
1492 	struct net_device_stats *ns = &pi->netstats;
1493 	const struct mac_stats *pstats;
1494 
1495 	spin_lock(&adapter->stats_lock);
1496 	pstats = t3_mac_update_stats(&pi->mac);
1497 	spin_unlock(&adapter->stats_lock);
1498 
1499 	ns->tx_bytes = pstats->tx_octets;
1500 	ns->tx_packets = pstats->tx_frames;
1501 	ns->rx_bytes = pstats->rx_octets;
1502 	ns->rx_packets = pstats->rx_frames;
1503 	ns->multicast = pstats->rx_mcast_frames;
1504 
1505 	ns->tx_errors = pstats->tx_underrun;
1506 	ns->rx_errors = pstats->rx_symbol_errs + pstats->rx_fcs_errs +
1507 	    pstats->rx_too_long + pstats->rx_jabber + pstats->rx_short +
1508 	    pstats->rx_fifo_ovfl;
1509 
1510 	/* detailed rx_errors */
1511 	ns->rx_length_errors = pstats->rx_jabber + pstats->rx_too_long;
1512 	ns->rx_over_errors = 0;
1513 	ns->rx_crc_errors = pstats->rx_fcs_errs;
1514 	ns->rx_frame_errors = pstats->rx_symbol_errs;
1515 	ns->rx_fifo_errors = pstats->rx_fifo_ovfl;
1516 	ns->rx_missed_errors = pstats->rx_cong_drops;
1517 
1518 	/* detailed tx_errors */
1519 	ns->tx_aborted_errors = 0;
1520 	ns->tx_carrier_errors = 0;
1521 	ns->tx_fifo_errors = pstats->tx_underrun;
1522 	ns->tx_heartbeat_errors = 0;
1523 	ns->tx_window_errors = 0;
1524 	return ns;
1525 }
1526 
1527 static u32 get_msglevel(struct net_device *dev)
1528 {
1529 	struct port_info *pi = netdev_priv(dev);
1530 	struct adapter *adapter = pi->adapter;
1531 
1532 	return adapter->msg_enable;
1533 }
1534 
1535 static void set_msglevel(struct net_device *dev, u32 val)
1536 {
1537 	struct port_info *pi = netdev_priv(dev);
1538 	struct adapter *adapter = pi->adapter;
1539 
1540 	adapter->msg_enable = val;
1541 }
1542 
1543 static const char stats_strings[][ETH_GSTRING_LEN] = {
1544 	"TxOctetsOK         ",
1545 	"TxFramesOK         ",
1546 	"TxMulticastFramesOK",
1547 	"TxBroadcastFramesOK",
1548 	"TxPauseFrames      ",
1549 	"TxUnderrun         ",
1550 	"TxExtUnderrun      ",
1551 
1552 	"TxFrames64         ",
1553 	"TxFrames65To127    ",
1554 	"TxFrames128To255   ",
1555 	"TxFrames256To511   ",
1556 	"TxFrames512To1023  ",
1557 	"TxFrames1024To1518 ",
1558 	"TxFrames1519ToMax  ",
1559 
1560 	"RxOctetsOK         ",
1561 	"RxFramesOK         ",
1562 	"RxMulticastFramesOK",
1563 	"RxBroadcastFramesOK",
1564 	"RxPauseFrames      ",
1565 	"RxFCSErrors        ",
1566 	"RxSymbolErrors     ",
1567 	"RxShortErrors      ",
1568 	"RxJabberErrors     ",
1569 	"RxLengthErrors     ",
1570 	"RxFIFOoverflow     ",
1571 
1572 	"RxFrames64         ",
1573 	"RxFrames65To127    ",
1574 	"RxFrames128To255   ",
1575 	"RxFrames256To511   ",
1576 	"RxFrames512To1023  ",
1577 	"RxFrames1024To1518 ",
1578 	"RxFrames1519ToMax  ",
1579 
1580 	"PhyFIFOErrors      ",
1581 	"TSO                ",
1582 	"VLANextractions    ",
1583 	"VLANinsertions     ",
1584 	"TxCsumOffload      ",
1585 	"RxCsumGood         ",
1586 	"LroAggregated      ",
1587 	"LroFlushed         ",
1588 	"LroNoDesc          ",
1589 	"RxDrops            ",
1590 
1591 	"CheckTXEnToggled   ",
1592 	"CheckResets        ",
1593 
1594 	"LinkFaults         ",
1595 };
1596 
1597 static int get_sset_count(struct net_device *dev, int sset)
1598 {
1599 	switch (sset) {
1600 	case ETH_SS_STATS:
1601 		return ARRAY_SIZE(stats_strings);
1602 	default:
1603 		return -EOPNOTSUPP;
1604 	}
1605 }
1606 
1607 #define T3_REGMAP_SIZE (3 * 1024)
1608 
1609 static int get_regs_len(struct net_device *dev)
1610 {
1611 	return T3_REGMAP_SIZE;
1612 }
1613 
1614 static int get_eeprom_len(struct net_device *dev)
1615 {
1616 	return EEPROMSIZE;
1617 }
1618 
1619 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1620 {
1621 	struct port_info *pi = netdev_priv(dev);
1622 	struct adapter *adapter = pi->adapter;
1623 	u32 fw_vers = 0;
1624 	u32 tp_vers = 0;
1625 
1626 	spin_lock(&adapter->stats_lock);
1627 	t3_get_fw_version(adapter, &fw_vers);
1628 	t3_get_tp_version(adapter, &tp_vers);
1629 	spin_unlock(&adapter->stats_lock);
1630 
1631 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1632 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1633 	strlcpy(info->bus_info, pci_name(adapter->pdev),
1634 		sizeof(info->bus_info));
1635 	if (fw_vers)
1636 		snprintf(info->fw_version, sizeof(info->fw_version),
1637 			 "%s %u.%u.%u TP %u.%u.%u",
1638 			 G_FW_VERSION_TYPE(fw_vers) ? "T" : "N",
1639 			 G_FW_VERSION_MAJOR(fw_vers),
1640 			 G_FW_VERSION_MINOR(fw_vers),
1641 			 G_FW_VERSION_MICRO(fw_vers),
1642 			 G_TP_VERSION_MAJOR(tp_vers),
1643 			 G_TP_VERSION_MINOR(tp_vers),
1644 			 G_TP_VERSION_MICRO(tp_vers));
1645 }
1646 
1647 static void get_strings(struct net_device *dev, u32 stringset, u8 * data)
1648 {
1649 	if (stringset == ETH_SS_STATS)
1650 		memcpy(data, stats_strings, sizeof(stats_strings));
1651 }
1652 
1653 static unsigned long collect_sge_port_stats(struct adapter *adapter,
1654 					    struct port_info *p, int idx)
1655 {
1656 	int i;
1657 	unsigned long tot = 0;
1658 
1659 	for (i = p->first_qset; i < p->first_qset + p->nqsets; ++i)
1660 		tot += adapter->sge.qs[i].port_stats[idx];
1661 	return tot;
1662 }
1663 
1664 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1665 		      u64 *data)
1666 {
1667 	struct port_info *pi = netdev_priv(dev);
1668 	struct adapter *adapter = pi->adapter;
1669 	const struct mac_stats *s;
1670 
1671 	spin_lock(&adapter->stats_lock);
1672 	s = t3_mac_update_stats(&pi->mac);
1673 	spin_unlock(&adapter->stats_lock);
1674 
1675 	*data++ = s->tx_octets;
1676 	*data++ = s->tx_frames;
1677 	*data++ = s->tx_mcast_frames;
1678 	*data++ = s->tx_bcast_frames;
1679 	*data++ = s->tx_pause;
1680 	*data++ = s->tx_underrun;
1681 	*data++ = s->tx_fifo_urun;
1682 
1683 	*data++ = s->tx_frames_64;
1684 	*data++ = s->tx_frames_65_127;
1685 	*data++ = s->tx_frames_128_255;
1686 	*data++ = s->tx_frames_256_511;
1687 	*data++ = s->tx_frames_512_1023;
1688 	*data++ = s->tx_frames_1024_1518;
1689 	*data++ = s->tx_frames_1519_max;
1690 
1691 	*data++ = s->rx_octets;
1692 	*data++ = s->rx_frames;
1693 	*data++ = s->rx_mcast_frames;
1694 	*data++ = s->rx_bcast_frames;
1695 	*data++ = s->rx_pause;
1696 	*data++ = s->rx_fcs_errs;
1697 	*data++ = s->rx_symbol_errs;
1698 	*data++ = s->rx_short;
1699 	*data++ = s->rx_jabber;
1700 	*data++ = s->rx_too_long;
1701 	*data++ = s->rx_fifo_ovfl;
1702 
1703 	*data++ = s->rx_frames_64;
1704 	*data++ = s->rx_frames_65_127;
1705 	*data++ = s->rx_frames_128_255;
1706 	*data++ = s->rx_frames_256_511;
1707 	*data++ = s->rx_frames_512_1023;
1708 	*data++ = s->rx_frames_1024_1518;
1709 	*data++ = s->rx_frames_1519_max;
1710 
1711 	*data++ = pi->phy.fifo_errors;
1712 
1713 	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TSO);
1714 	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANEX);
1715 	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANINS);
1716 	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TX_CSUM);
1717 	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_RX_CSUM_GOOD);
1718 	*data++ = 0;
1719 	*data++ = 0;
1720 	*data++ = 0;
1721 	*data++ = s->rx_cong_drops;
1722 
1723 	*data++ = s->num_toggled;
1724 	*data++ = s->num_resets;
1725 
1726 	*data++ = s->link_faults;
1727 }
1728 
1729 static inline void reg_block_dump(struct adapter *ap, void *buf,
1730 				  unsigned int start, unsigned int end)
1731 {
1732 	u32 *p = buf + start;
1733 
1734 	for (; start <= end; start += sizeof(u32))
1735 		*p++ = t3_read_reg(ap, start);
1736 }
1737 
1738 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1739 		     void *buf)
1740 {
1741 	struct port_info *pi = netdev_priv(dev);
1742 	struct adapter *ap = pi->adapter;
1743 
1744 	/*
1745 	 * Version scheme:
1746 	 * bits 0..9: chip version
1747 	 * bits 10..15: chip revision
1748 	 * bit 31: set for PCIe cards
1749 	 */
1750 	regs->version = 3 | (ap->params.rev << 10) | (is_pcie(ap) << 31);
1751 
1752 	/*
1753 	 * We skip the MAC statistics registers because they are clear-on-read.
1754 	 * Also reading multi-register stats would need to synchronize with the
1755 	 * periodic mac stats accumulation.  Hard to justify the complexity.
1756 	 */
1757 	memset(buf, 0, T3_REGMAP_SIZE);
1758 	reg_block_dump(ap, buf, 0, A_SG_RSPQ_CREDIT_RETURN);
1759 	reg_block_dump(ap, buf, A_SG_HI_DRB_HI_THRSH, A_ULPRX_PBL_ULIMIT);
1760 	reg_block_dump(ap, buf, A_ULPTX_CONFIG, A_MPS_INT_CAUSE);
1761 	reg_block_dump(ap, buf, A_CPL_SWITCH_CNTRL, A_CPL_MAP_TBL_DATA);
1762 	reg_block_dump(ap, buf, A_SMB_GLOBAL_TIME_CFG, A_XGM_SERDES_STAT3);
1763 	reg_block_dump(ap, buf, A_XGM_SERDES_STATUS0,
1764 		       XGM_REG(A_XGM_SERDES_STAT3, 1));
1765 	reg_block_dump(ap, buf, XGM_REG(A_XGM_SERDES_STATUS0, 1),
1766 		       XGM_REG(A_XGM_RX_SPI4_SOP_EOP_CNT, 1));
1767 }
1768 
1769 static int restart_autoneg(struct net_device *dev)
1770 {
1771 	struct port_info *p = netdev_priv(dev);
1772 
1773 	if (!netif_running(dev))
1774 		return -EAGAIN;
1775 	if (p->link_config.autoneg != AUTONEG_ENABLE)
1776 		return -EINVAL;
1777 	p->phy.ops->autoneg_restart(&p->phy);
1778 	return 0;
1779 }
1780 
1781 static int set_phys_id(struct net_device *dev,
1782 		       enum ethtool_phys_id_state state)
1783 {
1784 	struct port_info *pi = netdev_priv(dev);
1785 	struct adapter *adapter = pi->adapter;
1786 
1787 	switch (state) {
1788 	case ETHTOOL_ID_ACTIVE:
1789 		return 1;	/* cycle on/off once per second */
1790 
1791 	case ETHTOOL_ID_OFF:
1792 		t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL, 0);
1793 		break;
1794 
1795 	case ETHTOOL_ID_ON:
1796 	case ETHTOOL_ID_INACTIVE:
1797 		t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
1798 			 F_GPIO0_OUT_VAL);
1799 	}
1800 
1801 	return 0;
1802 }
1803 
1804 static int get_link_ksettings(struct net_device *dev,
1805 			      struct ethtool_link_ksettings *cmd)
1806 {
1807 	struct port_info *p = netdev_priv(dev);
1808 	u32 supported;
1809 
1810 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
1811 						p->link_config.supported);
1812 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
1813 						p->link_config.advertising);
1814 
1815 	if (netif_carrier_ok(dev)) {
1816 		cmd->base.speed = p->link_config.speed;
1817 		cmd->base.duplex = p->link_config.duplex;
1818 	} else {
1819 		cmd->base.speed = SPEED_UNKNOWN;
1820 		cmd->base.duplex = DUPLEX_UNKNOWN;
1821 	}
1822 
1823 	ethtool_convert_link_mode_to_legacy_u32(&supported,
1824 						cmd->link_modes.supported);
1825 
1826 	cmd->base.port = (supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1827 	cmd->base.phy_address = p->phy.mdio.prtad;
1828 	cmd->base.autoneg = p->link_config.autoneg;
1829 	return 0;
1830 }
1831 
1832 static int speed_duplex_to_caps(int speed, int duplex)
1833 {
1834 	int cap = 0;
1835 
1836 	switch (speed) {
1837 	case SPEED_10:
1838 		if (duplex == DUPLEX_FULL)
1839 			cap = SUPPORTED_10baseT_Full;
1840 		else
1841 			cap = SUPPORTED_10baseT_Half;
1842 		break;
1843 	case SPEED_100:
1844 		if (duplex == DUPLEX_FULL)
1845 			cap = SUPPORTED_100baseT_Full;
1846 		else
1847 			cap = SUPPORTED_100baseT_Half;
1848 		break;
1849 	case SPEED_1000:
1850 		if (duplex == DUPLEX_FULL)
1851 			cap = SUPPORTED_1000baseT_Full;
1852 		else
1853 			cap = SUPPORTED_1000baseT_Half;
1854 		break;
1855 	case SPEED_10000:
1856 		if (duplex == DUPLEX_FULL)
1857 			cap = SUPPORTED_10000baseT_Full;
1858 	}
1859 	return cap;
1860 }
1861 
1862 #define ADVERTISED_MASK (ADVERTISED_10baseT_Half | ADVERTISED_10baseT_Full | \
1863 		      ADVERTISED_100baseT_Half | ADVERTISED_100baseT_Full | \
1864 		      ADVERTISED_1000baseT_Half | ADVERTISED_1000baseT_Full | \
1865 		      ADVERTISED_10000baseT_Full)
1866 
1867 static int set_link_ksettings(struct net_device *dev,
1868 			      const struct ethtool_link_ksettings *cmd)
1869 {
1870 	struct port_info *p = netdev_priv(dev);
1871 	struct link_config *lc = &p->link_config;
1872 	u32 advertising;
1873 
1874 	ethtool_convert_link_mode_to_legacy_u32(&advertising,
1875 						cmd->link_modes.advertising);
1876 
1877 	if (!(lc->supported & SUPPORTED_Autoneg)) {
1878 		/*
1879 		 * PHY offers a single speed/duplex.  See if that's what's
1880 		 * being requested.
1881 		 */
1882 		if (cmd->base.autoneg == AUTONEG_DISABLE) {
1883 			u32 speed = cmd->base.speed;
1884 			int cap = speed_duplex_to_caps(speed, cmd->base.duplex);
1885 			if (lc->supported & cap)
1886 				return 0;
1887 		}
1888 		return -EINVAL;
1889 	}
1890 
1891 	if (cmd->base.autoneg == AUTONEG_DISABLE) {
1892 		u32 speed = cmd->base.speed;
1893 		int cap = speed_duplex_to_caps(speed, cmd->base.duplex);
1894 
1895 		if (!(lc->supported & cap) || (speed == SPEED_1000))
1896 			return -EINVAL;
1897 		lc->requested_speed = speed;
1898 		lc->requested_duplex = cmd->base.duplex;
1899 		lc->advertising = 0;
1900 	} else {
1901 		advertising &= ADVERTISED_MASK;
1902 		advertising &= lc->supported;
1903 		if (!advertising)
1904 			return -EINVAL;
1905 		lc->requested_speed = SPEED_INVALID;
1906 		lc->requested_duplex = DUPLEX_INVALID;
1907 		lc->advertising = advertising | ADVERTISED_Autoneg;
1908 	}
1909 	lc->autoneg = cmd->base.autoneg;
1910 	if (netif_running(dev))
1911 		t3_link_start(&p->phy, &p->mac, lc);
1912 	return 0;
1913 }
1914 
1915 static void get_pauseparam(struct net_device *dev,
1916 			   struct ethtool_pauseparam *epause)
1917 {
1918 	struct port_info *p = netdev_priv(dev);
1919 
1920 	epause->autoneg = (p->link_config.requested_fc & PAUSE_AUTONEG) != 0;
1921 	epause->rx_pause = (p->link_config.fc & PAUSE_RX) != 0;
1922 	epause->tx_pause = (p->link_config.fc & PAUSE_TX) != 0;
1923 }
1924 
1925 static int set_pauseparam(struct net_device *dev,
1926 			  struct ethtool_pauseparam *epause)
1927 {
1928 	struct port_info *p = netdev_priv(dev);
1929 	struct link_config *lc = &p->link_config;
1930 
1931 	if (epause->autoneg == AUTONEG_DISABLE)
1932 		lc->requested_fc = 0;
1933 	else if (lc->supported & SUPPORTED_Autoneg)
1934 		lc->requested_fc = PAUSE_AUTONEG;
1935 	else
1936 		return -EINVAL;
1937 
1938 	if (epause->rx_pause)
1939 		lc->requested_fc |= PAUSE_RX;
1940 	if (epause->tx_pause)
1941 		lc->requested_fc |= PAUSE_TX;
1942 	if (lc->autoneg == AUTONEG_ENABLE) {
1943 		if (netif_running(dev))
1944 			t3_link_start(&p->phy, &p->mac, lc);
1945 	} else {
1946 		lc->fc = lc->requested_fc & (PAUSE_RX | PAUSE_TX);
1947 		if (netif_running(dev))
1948 			t3_mac_set_speed_duplex_fc(&p->mac, -1, -1, lc->fc);
1949 	}
1950 	return 0;
1951 }
1952 
1953 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1954 {
1955 	struct port_info *pi = netdev_priv(dev);
1956 	struct adapter *adapter = pi->adapter;
1957 	const struct qset_params *q = &adapter->params.sge.qset[pi->first_qset];
1958 
1959 	e->rx_max_pending = MAX_RX_BUFFERS;
1960 	e->rx_jumbo_max_pending = MAX_RX_JUMBO_BUFFERS;
1961 	e->tx_max_pending = MAX_TXQ_ENTRIES;
1962 
1963 	e->rx_pending = q->fl_size;
1964 	e->rx_mini_pending = q->rspq_size;
1965 	e->rx_jumbo_pending = q->jumbo_size;
1966 	e->tx_pending = q->txq_size[0];
1967 }
1968 
1969 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1970 {
1971 	struct port_info *pi = netdev_priv(dev);
1972 	struct adapter *adapter = pi->adapter;
1973 	struct qset_params *q;
1974 	int i;
1975 
1976 	if (e->rx_pending > MAX_RX_BUFFERS ||
1977 	    e->rx_jumbo_pending > MAX_RX_JUMBO_BUFFERS ||
1978 	    e->tx_pending > MAX_TXQ_ENTRIES ||
1979 	    e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1980 	    e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1981 	    e->rx_pending < MIN_FL_ENTRIES ||
1982 	    e->rx_jumbo_pending < MIN_FL_ENTRIES ||
1983 	    e->tx_pending < adapter->params.nports * MIN_TXQ_ENTRIES)
1984 		return -EINVAL;
1985 
1986 	if (adapter->flags & FULL_INIT_DONE)
1987 		return -EBUSY;
1988 
1989 	q = &adapter->params.sge.qset[pi->first_qset];
1990 	for (i = 0; i < pi->nqsets; ++i, ++q) {
1991 		q->rspq_size = e->rx_mini_pending;
1992 		q->fl_size = e->rx_pending;
1993 		q->jumbo_size = e->rx_jumbo_pending;
1994 		q->txq_size[0] = e->tx_pending;
1995 		q->txq_size[1] = e->tx_pending;
1996 		q->txq_size[2] = e->tx_pending;
1997 	}
1998 	return 0;
1999 }
2000 
2001 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2002 {
2003 	struct port_info *pi = netdev_priv(dev);
2004 	struct adapter *adapter = pi->adapter;
2005 	struct qset_params *qsp;
2006 	struct sge_qset *qs;
2007 	int i;
2008 
2009 	if (c->rx_coalesce_usecs * 10 > M_NEWTIMER)
2010 		return -EINVAL;
2011 
2012 	for (i = 0; i < pi->nqsets; i++) {
2013 		qsp = &adapter->params.sge.qset[i];
2014 		qs = &adapter->sge.qs[i];
2015 		qsp->coalesce_usecs = c->rx_coalesce_usecs;
2016 		t3_update_qset_coalesce(qs, qsp);
2017 	}
2018 
2019 	return 0;
2020 }
2021 
2022 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2023 {
2024 	struct port_info *pi = netdev_priv(dev);
2025 	struct adapter *adapter = pi->adapter;
2026 	struct qset_params *q = adapter->params.sge.qset;
2027 
2028 	c->rx_coalesce_usecs = q->coalesce_usecs;
2029 	return 0;
2030 }
2031 
2032 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
2033 		      u8 * data)
2034 {
2035 	struct port_info *pi = netdev_priv(dev);
2036 	struct adapter *adapter = pi->adapter;
2037 	int i, err = 0;
2038 
2039 	u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
2040 	if (!buf)
2041 		return -ENOMEM;
2042 
2043 	e->magic = EEPROM_MAGIC;
2044 	for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
2045 		err = t3_seeprom_read(adapter, i, (__le32 *) & buf[i]);
2046 
2047 	if (!err)
2048 		memcpy(data, buf + e->offset, e->len);
2049 	kfree(buf);
2050 	return err;
2051 }
2052 
2053 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
2054 		      u8 * data)
2055 {
2056 	struct port_info *pi = netdev_priv(dev);
2057 	struct adapter *adapter = pi->adapter;
2058 	u32 aligned_offset, aligned_len;
2059 	__le32 *p;
2060 	u8 *buf;
2061 	int err;
2062 
2063 	if (eeprom->magic != EEPROM_MAGIC)
2064 		return -EINVAL;
2065 
2066 	aligned_offset = eeprom->offset & ~3;
2067 	aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
2068 
2069 	if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
2070 		buf = kmalloc(aligned_len, GFP_KERNEL);
2071 		if (!buf)
2072 			return -ENOMEM;
2073 		err = t3_seeprom_read(adapter, aligned_offset, (__le32 *) buf);
2074 		if (!err && aligned_len > 4)
2075 			err = t3_seeprom_read(adapter,
2076 					      aligned_offset + aligned_len - 4,
2077 					      (__le32 *) & buf[aligned_len - 4]);
2078 		if (err)
2079 			goto out;
2080 		memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
2081 	} else
2082 		buf = data;
2083 
2084 	err = t3_seeprom_wp(adapter, 0);
2085 	if (err)
2086 		goto out;
2087 
2088 	for (p = (__le32 *) buf; !err && aligned_len; aligned_len -= 4, p++) {
2089 		err = t3_seeprom_write(adapter, aligned_offset, *p);
2090 		aligned_offset += 4;
2091 	}
2092 
2093 	if (!err)
2094 		err = t3_seeprom_wp(adapter, 1);
2095 out:
2096 	if (buf != data)
2097 		kfree(buf);
2098 	return err;
2099 }
2100 
2101 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2102 {
2103 	wol->supported = 0;
2104 	wol->wolopts = 0;
2105 	memset(&wol->sopass, 0, sizeof(wol->sopass));
2106 }
2107 
2108 static const struct ethtool_ops cxgb_ethtool_ops = {
2109 	.get_drvinfo = get_drvinfo,
2110 	.get_msglevel = get_msglevel,
2111 	.set_msglevel = set_msglevel,
2112 	.get_ringparam = get_sge_param,
2113 	.set_ringparam = set_sge_param,
2114 	.get_coalesce = get_coalesce,
2115 	.set_coalesce = set_coalesce,
2116 	.get_eeprom_len = get_eeprom_len,
2117 	.get_eeprom = get_eeprom,
2118 	.set_eeprom = set_eeprom,
2119 	.get_pauseparam = get_pauseparam,
2120 	.set_pauseparam = set_pauseparam,
2121 	.get_link = ethtool_op_get_link,
2122 	.get_strings = get_strings,
2123 	.set_phys_id = set_phys_id,
2124 	.nway_reset = restart_autoneg,
2125 	.get_sset_count = get_sset_count,
2126 	.get_ethtool_stats = get_stats,
2127 	.get_regs_len = get_regs_len,
2128 	.get_regs = get_regs,
2129 	.get_wol = get_wol,
2130 	.get_link_ksettings = get_link_ksettings,
2131 	.set_link_ksettings = set_link_ksettings,
2132 };
2133 
2134 static int in_range(int val, int lo, int hi)
2135 {
2136 	return val < 0 || (val <= hi && val >= lo);
2137 }
2138 
2139 static int cxgb_extension_ioctl(struct net_device *dev, void __user *useraddr)
2140 {
2141 	struct port_info *pi = netdev_priv(dev);
2142 	struct adapter *adapter = pi->adapter;
2143 	u32 cmd;
2144 	int ret;
2145 
2146 	if (copy_from_user(&cmd, useraddr, sizeof(cmd)))
2147 		return -EFAULT;
2148 
2149 	switch (cmd) {
2150 	case CHELSIO_SET_QSET_PARAMS:{
2151 		int i;
2152 		struct qset_params *q;
2153 		struct ch_qset_params t;
2154 		int q1 = pi->first_qset;
2155 		int nqsets = pi->nqsets;
2156 
2157 		if (!capable(CAP_NET_ADMIN))
2158 			return -EPERM;
2159 		if (copy_from_user(&t, useraddr, sizeof(t)))
2160 			return -EFAULT;
2161 		if (t.qset_idx >= SGE_QSETS)
2162 			return -EINVAL;
2163 		if (!in_range(t.intr_lat, 0, M_NEWTIMER) ||
2164 		    !in_range(t.cong_thres, 0, 255) ||
2165 		    !in_range(t.txq_size[0], MIN_TXQ_ENTRIES,
2166 			      MAX_TXQ_ENTRIES) ||
2167 		    !in_range(t.txq_size[1], MIN_TXQ_ENTRIES,
2168 			      MAX_TXQ_ENTRIES) ||
2169 		    !in_range(t.txq_size[2], MIN_CTRL_TXQ_ENTRIES,
2170 			      MAX_CTRL_TXQ_ENTRIES) ||
2171 		    !in_range(t.fl_size[0], MIN_FL_ENTRIES,
2172 			      MAX_RX_BUFFERS) ||
2173 		    !in_range(t.fl_size[1], MIN_FL_ENTRIES,
2174 			      MAX_RX_JUMBO_BUFFERS) ||
2175 		    !in_range(t.rspq_size, MIN_RSPQ_ENTRIES,
2176 			      MAX_RSPQ_ENTRIES))
2177 			return -EINVAL;
2178 
2179 		if ((adapter->flags & FULL_INIT_DONE) &&
2180 			(t.rspq_size >= 0 || t.fl_size[0] >= 0 ||
2181 			t.fl_size[1] >= 0 || t.txq_size[0] >= 0 ||
2182 			t.txq_size[1] >= 0 || t.txq_size[2] >= 0 ||
2183 			t.polling >= 0 || t.cong_thres >= 0))
2184 			return -EBUSY;
2185 
2186 		/* Allow setting of any available qset when offload enabled */
2187 		if (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2188 			q1 = 0;
2189 			for_each_port(adapter, i) {
2190 				pi = adap2pinfo(adapter, i);
2191 				nqsets += pi->first_qset + pi->nqsets;
2192 			}
2193 		}
2194 
2195 		if (t.qset_idx < q1)
2196 			return -EINVAL;
2197 		if (t.qset_idx > q1 + nqsets - 1)
2198 			return -EINVAL;
2199 
2200 		q = &adapter->params.sge.qset[t.qset_idx];
2201 
2202 		if (t.rspq_size >= 0)
2203 			q->rspq_size = t.rspq_size;
2204 		if (t.fl_size[0] >= 0)
2205 			q->fl_size = t.fl_size[0];
2206 		if (t.fl_size[1] >= 0)
2207 			q->jumbo_size = t.fl_size[1];
2208 		if (t.txq_size[0] >= 0)
2209 			q->txq_size[0] = t.txq_size[0];
2210 		if (t.txq_size[1] >= 0)
2211 			q->txq_size[1] = t.txq_size[1];
2212 		if (t.txq_size[2] >= 0)
2213 			q->txq_size[2] = t.txq_size[2];
2214 		if (t.cong_thres >= 0)
2215 			q->cong_thres = t.cong_thres;
2216 		if (t.intr_lat >= 0) {
2217 			struct sge_qset *qs =
2218 				&adapter->sge.qs[t.qset_idx];
2219 
2220 			q->coalesce_usecs = t.intr_lat;
2221 			t3_update_qset_coalesce(qs, q);
2222 		}
2223 		if (t.polling >= 0) {
2224 			if (adapter->flags & USING_MSIX)
2225 				q->polling = t.polling;
2226 			else {
2227 				/* No polling with INTx for T3A */
2228 				if (adapter->params.rev == 0 &&
2229 					!(adapter->flags & USING_MSI))
2230 					t.polling = 0;
2231 
2232 				for (i = 0; i < SGE_QSETS; i++) {
2233 					q = &adapter->params.sge.
2234 						qset[i];
2235 					q->polling = t.polling;
2236 				}
2237 			}
2238 		}
2239 
2240 		if (t.lro >= 0) {
2241 			if (t.lro)
2242 				dev->wanted_features |= NETIF_F_GRO;
2243 			else
2244 				dev->wanted_features &= ~NETIF_F_GRO;
2245 			netdev_update_features(dev);
2246 		}
2247 
2248 		break;
2249 	}
2250 	case CHELSIO_GET_QSET_PARAMS:{
2251 		struct qset_params *q;
2252 		struct ch_qset_params t;
2253 		int q1 = pi->first_qset;
2254 		int nqsets = pi->nqsets;
2255 		int i;
2256 
2257 		if (copy_from_user(&t, useraddr, sizeof(t)))
2258 			return -EFAULT;
2259 
2260 		/* Display qsets for all ports when offload enabled */
2261 		if (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2262 			q1 = 0;
2263 			for_each_port(adapter, i) {
2264 				pi = adap2pinfo(adapter, i);
2265 				nqsets = pi->first_qset + pi->nqsets;
2266 			}
2267 		}
2268 
2269 		if (t.qset_idx >= nqsets)
2270 			return -EINVAL;
2271 
2272 		q = &adapter->params.sge.qset[q1 + t.qset_idx];
2273 		t.rspq_size = q->rspq_size;
2274 		t.txq_size[0] = q->txq_size[0];
2275 		t.txq_size[1] = q->txq_size[1];
2276 		t.txq_size[2] = q->txq_size[2];
2277 		t.fl_size[0] = q->fl_size;
2278 		t.fl_size[1] = q->jumbo_size;
2279 		t.polling = q->polling;
2280 		t.lro = !!(dev->features & NETIF_F_GRO);
2281 		t.intr_lat = q->coalesce_usecs;
2282 		t.cong_thres = q->cong_thres;
2283 		t.qnum = q1;
2284 
2285 		if (adapter->flags & USING_MSIX)
2286 			t.vector = adapter->msix_info[q1 + t.qset_idx + 1].vec;
2287 		else
2288 			t.vector = adapter->pdev->irq;
2289 
2290 		if (copy_to_user(useraddr, &t, sizeof(t)))
2291 			return -EFAULT;
2292 		break;
2293 	}
2294 	case CHELSIO_SET_QSET_NUM:{
2295 		struct ch_reg edata;
2296 		unsigned int i, first_qset = 0, other_qsets = 0;
2297 
2298 		if (!capable(CAP_NET_ADMIN))
2299 			return -EPERM;
2300 		if (adapter->flags & FULL_INIT_DONE)
2301 			return -EBUSY;
2302 		if (copy_from_user(&edata, useraddr, sizeof(edata)))
2303 			return -EFAULT;
2304 		if (edata.val < 1 ||
2305 			(edata.val > 1 && !(adapter->flags & USING_MSIX)))
2306 			return -EINVAL;
2307 
2308 		for_each_port(adapter, i)
2309 			if (adapter->port[i] && adapter->port[i] != dev)
2310 				other_qsets += adap2pinfo(adapter, i)->nqsets;
2311 
2312 		if (edata.val + other_qsets > SGE_QSETS)
2313 			return -EINVAL;
2314 
2315 		pi->nqsets = edata.val;
2316 
2317 		for_each_port(adapter, i)
2318 			if (adapter->port[i]) {
2319 				pi = adap2pinfo(adapter, i);
2320 				pi->first_qset = first_qset;
2321 				first_qset += pi->nqsets;
2322 			}
2323 		break;
2324 	}
2325 	case CHELSIO_GET_QSET_NUM:{
2326 		struct ch_reg edata;
2327 
2328 		memset(&edata, 0, sizeof(struct ch_reg));
2329 
2330 		edata.cmd = CHELSIO_GET_QSET_NUM;
2331 		edata.val = pi->nqsets;
2332 		if (copy_to_user(useraddr, &edata, sizeof(edata)))
2333 			return -EFAULT;
2334 		break;
2335 	}
2336 	case CHELSIO_LOAD_FW:{
2337 		u8 *fw_data;
2338 		struct ch_mem_range t;
2339 
2340 		if (!capable(CAP_SYS_RAWIO))
2341 			return -EPERM;
2342 		if (copy_from_user(&t, useraddr, sizeof(t)))
2343 			return -EFAULT;
2344 		/* Check t.len sanity ? */
2345 		fw_data = memdup_user(useraddr + sizeof(t), t.len);
2346 		if (IS_ERR(fw_data))
2347 			return PTR_ERR(fw_data);
2348 
2349 		ret = t3_load_fw(adapter, fw_data, t.len);
2350 		kfree(fw_data);
2351 		if (ret)
2352 			return ret;
2353 		break;
2354 	}
2355 	case CHELSIO_SETMTUTAB:{
2356 		struct ch_mtus m;
2357 		int i;
2358 
2359 		if (!is_offload(adapter))
2360 			return -EOPNOTSUPP;
2361 		if (!capable(CAP_NET_ADMIN))
2362 			return -EPERM;
2363 		if (offload_running(adapter))
2364 			return -EBUSY;
2365 		if (copy_from_user(&m, useraddr, sizeof(m)))
2366 			return -EFAULT;
2367 		if (m.nmtus != NMTUS)
2368 			return -EINVAL;
2369 		if (m.mtus[0] < 81)	/* accommodate SACK */
2370 			return -EINVAL;
2371 
2372 		/* MTUs must be in ascending order */
2373 		for (i = 1; i < NMTUS; ++i)
2374 			if (m.mtus[i] < m.mtus[i - 1])
2375 				return -EINVAL;
2376 
2377 		memcpy(adapter->params.mtus, m.mtus,
2378 			sizeof(adapter->params.mtus));
2379 		break;
2380 	}
2381 	case CHELSIO_GET_PM:{
2382 		struct tp_params *p = &adapter->params.tp;
2383 		struct ch_pm m = {.cmd = CHELSIO_GET_PM };
2384 
2385 		if (!is_offload(adapter))
2386 			return -EOPNOTSUPP;
2387 		m.tx_pg_sz = p->tx_pg_size;
2388 		m.tx_num_pg = p->tx_num_pgs;
2389 		m.rx_pg_sz = p->rx_pg_size;
2390 		m.rx_num_pg = p->rx_num_pgs;
2391 		m.pm_total = p->pmtx_size + p->chan_rx_size * p->nchan;
2392 		if (copy_to_user(useraddr, &m, sizeof(m)))
2393 			return -EFAULT;
2394 		break;
2395 	}
2396 	case CHELSIO_SET_PM:{
2397 		struct ch_pm m;
2398 		struct tp_params *p = &adapter->params.tp;
2399 
2400 		if (!is_offload(adapter))
2401 			return -EOPNOTSUPP;
2402 		if (!capable(CAP_NET_ADMIN))
2403 			return -EPERM;
2404 		if (adapter->flags & FULL_INIT_DONE)
2405 			return -EBUSY;
2406 		if (copy_from_user(&m, useraddr, sizeof(m)))
2407 			return -EFAULT;
2408 		if (!is_power_of_2(m.rx_pg_sz) ||
2409 			!is_power_of_2(m.tx_pg_sz))
2410 			return -EINVAL;	/* not power of 2 */
2411 		if (!(m.rx_pg_sz & 0x14000))
2412 			return -EINVAL;	/* not 16KB or 64KB */
2413 		if (!(m.tx_pg_sz & 0x1554000))
2414 			return -EINVAL;
2415 		if (m.tx_num_pg == -1)
2416 			m.tx_num_pg = p->tx_num_pgs;
2417 		if (m.rx_num_pg == -1)
2418 			m.rx_num_pg = p->rx_num_pgs;
2419 		if (m.tx_num_pg % 24 || m.rx_num_pg % 24)
2420 			return -EINVAL;
2421 		if (m.rx_num_pg * m.rx_pg_sz > p->chan_rx_size ||
2422 			m.tx_num_pg * m.tx_pg_sz > p->chan_tx_size)
2423 			return -EINVAL;
2424 		p->rx_pg_size = m.rx_pg_sz;
2425 		p->tx_pg_size = m.tx_pg_sz;
2426 		p->rx_num_pgs = m.rx_num_pg;
2427 		p->tx_num_pgs = m.tx_num_pg;
2428 		break;
2429 	}
2430 	case CHELSIO_GET_MEM:{
2431 		struct ch_mem_range t;
2432 		struct mc7 *mem;
2433 		u64 buf[32];
2434 
2435 		if (!is_offload(adapter))
2436 			return -EOPNOTSUPP;
2437 		if (!(adapter->flags & FULL_INIT_DONE))
2438 			return -EIO;	/* need the memory controllers */
2439 		if (copy_from_user(&t, useraddr, sizeof(t)))
2440 			return -EFAULT;
2441 		if ((t.addr & 7) || (t.len & 7))
2442 			return -EINVAL;
2443 		if (t.mem_id == MEM_CM)
2444 			mem = &adapter->cm;
2445 		else if (t.mem_id == MEM_PMRX)
2446 			mem = &adapter->pmrx;
2447 		else if (t.mem_id == MEM_PMTX)
2448 			mem = &adapter->pmtx;
2449 		else
2450 			return -EINVAL;
2451 
2452 		/*
2453 		 * Version scheme:
2454 		 * bits 0..9: chip version
2455 		 * bits 10..15: chip revision
2456 		 */
2457 		t.version = 3 | (adapter->params.rev << 10);
2458 		if (copy_to_user(useraddr, &t, sizeof(t)))
2459 			return -EFAULT;
2460 
2461 		/*
2462 		 * Read 256 bytes at a time as len can be large and we don't
2463 		 * want to use huge intermediate buffers.
2464 		 */
2465 		useraddr += sizeof(t);	/* advance to start of buffer */
2466 		while (t.len) {
2467 			unsigned int chunk =
2468 				min_t(unsigned int, t.len, sizeof(buf));
2469 
2470 			ret =
2471 				t3_mc7_bd_read(mem, t.addr / 8, chunk / 8,
2472 						buf);
2473 			if (ret)
2474 				return ret;
2475 			if (copy_to_user(useraddr, buf, chunk))
2476 				return -EFAULT;
2477 			useraddr += chunk;
2478 			t.addr += chunk;
2479 			t.len -= chunk;
2480 		}
2481 		break;
2482 	}
2483 	case CHELSIO_SET_TRACE_FILTER:{
2484 		struct ch_trace t;
2485 		const struct trace_params *tp;
2486 
2487 		if (!capable(CAP_NET_ADMIN))
2488 			return -EPERM;
2489 		if (!offload_running(adapter))
2490 			return -EAGAIN;
2491 		if (copy_from_user(&t, useraddr, sizeof(t)))
2492 			return -EFAULT;
2493 
2494 		tp = (const struct trace_params *)&t.sip;
2495 		if (t.config_tx)
2496 			t3_config_trace_filter(adapter, tp, 0,
2497 						t.invert_match,
2498 						t.trace_tx);
2499 		if (t.config_rx)
2500 			t3_config_trace_filter(adapter, tp, 1,
2501 						t.invert_match,
2502 						t.trace_rx);
2503 		break;
2504 	}
2505 	default:
2506 		return -EOPNOTSUPP;
2507 	}
2508 	return 0;
2509 }
2510 
2511 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2512 {
2513 	struct mii_ioctl_data *data = if_mii(req);
2514 	struct port_info *pi = netdev_priv(dev);
2515 	struct adapter *adapter = pi->adapter;
2516 
2517 	switch (cmd) {
2518 	case SIOCGMIIREG:
2519 	case SIOCSMIIREG:
2520 		/* Convert phy_id from older PRTAD/DEVAD format */
2521 		if (is_10G(adapter) &&
2522 		    !mdio_phy_id_is_c45(data->phy_id) &&
2523 		    (data->phy_id & 0x1f00) &&
2524 		    !(data->phy_id & 0xe0e0))
2525 			data->phy_id = mdio_phy_id_c45(data->phy_id >> 8,
2526 						       data->phy_id & 0x1f);
2527 		/* FALLTHRU */
2528 	case SIOCGMIIPHY:
2529 		return mdio_mii_ioctl(&pi->phy.mdio, data, cmd);
2530 	case SIOCCHIOCTL:
2531 		return cxgb_extension_ioctl(dev, req->ifr_data);
2532 	default:
2533 		return -EOPNOTSUPP;
2534 	}
2535 }
2536 
2537 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2538 {
2539 	struct port_info *pi = netdev_priv(dev);
2540 	struct adapter *adapter = pi->adapter;
2541 	int ret;
2542 
2543 	if ((ret = t3_mac_set_mtu(&pi->mac, new_mtu)))
2544 		return ret;
2545 	dev->mtu = new_mtu;
2546 	init_port_mtus(adapter);
2547 	if (adapter->params.rev == 0 && offload_running(adapter))
2548 		t3_load_mtus(adapter, adapter->params.mtus,
2549 			     adapter->params.a_wnd, adapter->params.b_wnd,
2550 			     adapter->port[0]->mtu);
2551 	return 0;
2552 }
2553 
2554 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2555 {
2556 	struct port_info *pi = netdev_priv(dev);
2557 	struct adapter *adapter = pi->adapter;
2558 	struct sockaddr *addr = p;
2559 
2560 	if (!is_valid_ether_addr(addr->sa_data))
2561 		return -EADDRNOTAVAIL;
2562 
2563 	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2564 	t3_mac_set_address(&pi->mac, LAN_MAC_IDX, dev->dev_addr);
2565 	if (offload_running(adapter))
2566 		write_smt_entry(adapter, pi->port_id);
2567 	return 0;
2568 }
2569 
2570 static netdev_features_t cxgb_fix_features(struct net_device *dev,
2571 	netdev_features_t features)
2572 {
2573 	/*
2574 	 * Since there is no support for separate rx/tx vlan accel
2575 	 * enable/disable make sure tx flag is always in same state as rx.
2576 	 */
2577 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
2578 		features |= NETIF_F_HW_VLAN_CTAG_TX;
2579 	else
2580 		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2581 
2582 	return features;
2583 }
2584 
2585 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
2586 {
2587 	netdev_features_t changed = dev->features ^ features;
2588 
2589 	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2590 		cxgb_vlan_mode(dev, features);
2591 
2592 	return 0;
2593 }
2594 
2595 #ifdef CONFIG_NET_POLL_CONTROLLER
2596 static void cxgb_netpoll(struct net_device *dev)
2597 {
2598 	struct port_info *pi = netdev_priv(dev);
2599 	struct adapter *adapter = pi->adapter;
2600 	int qidx;
2601 
2602 	for (qidx = pi->first_qset; qidx < pi->first_qset + pi->nqsets; qidx++) {
2603 		struct sge_qset *qs = &adapter->sge.qs[qidx];
2604 		void *source;
2605 
2606 		if (adapter->flags & USING_MSIX)
2607 			source = qs;
2608 		else
2609 			source = adapter;
2610 
2611 		t3_intr_handler(adapter, qs->rspq.polling) (0, source);
2612 	}
2613 }
2614 #endif
2615 
2616 /*
2617  * Periodic accumulation of MAC statistics.
2618  */
2619 static void mac_stats_update(struct adapter *adapter)
2620 {
2621 	int i;
2622 
2623 	for_each_port(adapter, i) {
2624 		struct net_device *dev = adapter->port[i];
2625 		struct port_info *p = netdev_priv(dev);
2626 
2627 		if (netif_running(dev)) {
2628 			spin_lock(&adapter->stats_lock);
2629 			t3_mac_update_stats(&p->mac);
2630 			spin_unlock(&adapter->stats_lock);
2631 		}
2632 	}
2633 }
2634 
2635 static void check_link_status(struct adapter *adapter)
2636 {
2637 	int i;
2638 
2639 	for_each_port(adapter, i) {
2640 		struct net_device *dev = adapter->port[i];
2641 		struct port_info *p = netdev_priv(dev);
2642 		int link_fault;
2643 
2644 		spin_lock_irq(&adapter->work_lock);
2645 		link_fault = p->link_fault;
2646 		spin_unlock_irq(&adapter->work_lock);
2647 
2648 		if (link_fault) {
2649 			t3_link_fault(adapter, i);
2650 			continue;
2651 		}
2652 
2653 		if (!(p->phy.caps & SUPPORTED_IRQ) && netif_running(dev)) {
2654 			t3_xgm_intr_disable(adapter, i);
2655 			t3_read_reg(adapter, A_XGM_INT_STATUS + p->mac.offset);
2656 
2657 			t3_link_changed(adapter, i);
2658 			t3_xgm_intr_enable(adapter, i);
2659 		}
2660 	}
2661 }
2662 
2663 static void check_t3b2_mac(struct adapter *adapter)
2664 {
2665 	int i;
2666 
2667 	if (!rtnl_trylock())	/* synchronize with ifdown */
2668 		return;
2669 
2670 	for_each_port(adapter, i) {
2671 		struct net_device *dev = adapter->port[i];
2672 		struct port_info *p = netdev_priv(dev);
2673 		int status;
2674 
2675 		if (!netif_running(dev))
2676 			continue;
2677 
2678 		status = 0;
2679 		if (netif_running(dev) && netif_carrier_ok(dev))
2680 			status = t3b2_mac_watchdog_task(&p->mac);
2681 		if (status == 1)
2682 			p->mac.stats.num_toggled++;
2683 		else if (status == 2) {
2684 			struct cmac *mac = &p->mac;
2685 
2686 			t3_mac_set_mtu(mac, dev->mtu);
2687 			t3_mac_set_address(mac, LAN_MAC_IDX, dev->dev_addr);
2688 			cxgb_set_rxmode(dev);
2689 			t3_link_start(&p->phy, mac, &p->link_config);
2690 			t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
2691 			t3_port_intr_enable(adapter, p->port_id);
2692 			p->mac.stats.num_resets++;
2693 		}
2694 	}
2695 	rtnl_unlock();
2696 }
2697 
2698 
2699 static void t3_adap_check_task(struct work_struct *work)
2700 {
2701 	struct adapter *adapter = container_of(work, struct adapter,
2702 					       adap_check_task.work);
2703 	const struct adapter_params *p = &adapter->params;
2704 	int port;
2705 	unsigned int v, status, reset;
2706 
2707 	adapter->check_task_cnt++;
2708 
2709 	check_link_status(adapter);
2710 
2711 	/* Accumulate MAC stats if needed */
2712 	if (!p->linkpoll_period ||
2713 	    (adapter->check_task_cnt * p->linkpoll_period) / 10 >=
2714 	    p->stats_update_period) {
2715 		mac_stats_update(adapter);
2716 		adapter->check_task_cnt = 0;
2717 	}
2718 
2719 	if (p->rev == T3_REV_B2)
2720 		check_t3b2_mac(adapter);
2721 
2722 	/*
2723 	 * Scan the XGMAC's to check for various conditions which we want to
2724 	 * monitor in a periodic polling manner rather than via an interrupt
2725 	 * condition.  This is used for conditions which would otherwise flood
2726 	 * the system with interrupts and we only really need to know that the
2727 	 * conditions are "happening" ...  For each condition we count the
2728 	 * detection of the condition and reset it for the next polling loop.
2729 	 */
2730 	for_each_port(adapter, port) {
2731 		struct cmac *mac =  &adap2pinfo(adapter, port)->mac;
2732 		u32 cause;
2733 
2734 		cause = t3_read_reg(adapter, A_XGM_INT_CAUSE + mac->offset);
2735 		reset = 0;
2736 		if (cause & F_RXFIFO_OVERFLOW) {
2737 			mac->stats.rx_fifo_ovfl++;
2738 			reset |= F_RXFIFO_OVERFLOW;
2739 		}
2740 
2741 		t3_write_reg(adapter, A_XGM_INT_CAUSE + mac->offset, reset);
2742 	}
2743 
2744 	/*
2745 	 * We do the same as above for FL_EMPTY interrupts.
2746 	 */
2747 	status = t3_read_reg(adapter, A_SG_INT_CAUSE);
2748 	reset = 0;
2749 
2750 	if (status & F_FLEMPTY) {
2751 		struct sge_qset *qs = &adapter->sge.qs[0];
2752 		int i = 0;
2753 
2754 		reset |= F_FLEMPTY;
2755 
2756 		v = (t3_read_reg(adapter, A_SG_RSPQ_FL_STATUS) >> S_FL0EMPTY) &
2757 		    0xffff;
2758 
2759 		while (v) {
2760 			qs->fl[i].empty += (v & 1);
2761 			if (i)
2762 				qs++;
2763 			i ^= 1;
2764 			v >>= 1;
2765 		}
2766 	}
2767 
2768 	t3_write_reg(adapter, A_SG_INT_CAUSE, reset);
2769 
2770 	/* Schedule the next check update if any port is active. */
2771 	spin_lock_irq(&adapter->work_lock);
2772 	if (adapter->open_device_map & PORT_MASK)
2773 		schedule_chk_task(adapter);
2774 	spin_unlock_irq(&adapter->work_lock);
2775 }
2776 
2777 static void db_full_task(struct work_struct *work)
2778 {
2779 	struct adapter *adapter = container_of(work, struct adapter,
2780 					       db_full_task);
2781 
2782 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_FULL, 0);
2783 }
2784 
2785 static void db_empty_task(struct work_struct *work)
2786 {
2787 	struct adapter *adapter = container_of(work, struct adapter,
2788 					       db_empty_task);
2789 
2790 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_EMPTY, 0);
2791 }
2792 
2793 static void db_drop_task(struct work_struct *work)
2794 {
2795 	struct adapter *adapter = container_of(work, struct adapter,
2796 					       db_drop_task);
2797 	unsigned long delay = 1000;
2798 	unsigned short r;
2799 
2800 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_DROP, 0);
2801 
2802 	/*
2803 	 * Sleep a while before ringing the driver qset dbs.
2804 	 * The delay is between 1000-2023 usecs.
2805 	 */
2806 	get_random_bytes(&r, 2);
2807 	delay += r & 1023;
2808 	set_current_state(TASK_UNINTERRUPTIBLE);
2809 	schedule_timeout(usecs_to_jiffies(delay));
2810 	ring_dbs(adapter);
2811 }
2812 
2813 /*
2814  * Processes external (PHY) interrupts in process context.
2815  */
2816 static void ext_intr_task(struct work_struct *work)
2817 {
2818 	struct adapter *adapter = container_of(work, struct adapter,
2819 					       ext_intr_handler_task);
2820 	int i;
2821 
2822 	/* Disable link fault interrupts */
2823 	for_each_port(adapter, i) {
2824 		struct net_device *dev = adapter->port[i];
2825 		struct port_info *p = netdev_priv(dev);
2826 
2827 		t3_xgm_intr_disable(adapter, i);
2828 		t3_read_reg(adapter, A_XGM_INT_STATUS + p->mac.offset);
2829 	}
2830 
2831 	/* Re-enable link fault interrupts */
2832 	t3_phy_intr_handler(adapter);
2833 
2834 	for_each_port(adapter, i)
2835 		t3_xgm_intr_enable(adapter, i);
2836 
2837 	/* Now reenable external interrupts */
2838 	spin_lock_irq(&adapter->work_lock);
2839 	if (adapter->slow_intr_mask) {
2840 		adapter->slow_intr_mask |= F_T3DBG;
2841 		t3_write_reg(adapter, A_PL_INT_CAUSE0, F_T3DBG);
2842 		t3_write_reg(adapter, A_PL_INT_ENABLE0,
2843 			     adapter->slow_intr_mask);
2844 	}
2845 	spin_unlock_irq(&adapter->work_lock);
2846 }
2847 
2848 /*
2849  * Interrupt-context handler for external (PHY) interrupts.
2850  */
2851 void t3_os_ext_intr_handler(struct adapter *adapter)
2852 {
2853 	/*
2854 	 * Schedule a task to handle external interrupts as they may be slow
2855 	 * and we use a mutex to protect MDIO registers.  We disable PHY
2856 	 * interrupts in the meantime and let the task reenable them when
2857 	 * it's done.
2858 	 */
2859 	spin_lock(&adapter->work_lock);
2860 	if (adapter->slow_intr_mask) {
2861 		adapter->slow_intr_mask &= ~F_T3DBG;
2862 		t3_write_reg(adapter, A_PL_INT_ENABLE0,
2863 			     adapter->slow_intr_mask);
2864 		queue_work(cxgb3_wq, &adapter->ext_intr_handler_task);
2865 	}
2866 	spin_unlock(&adapter->work_lock);
2867 }
2868 
2869 void t3_os_link_fault_handler(struct adapter *adapter, int port_id)
2870 {
2871 	struct net_device *netdev = adapter->port[port_id];
2872 	struct port_info *pi = netdev_priv(netdev);
2873 
2874 	spin_lock(&adapter->work_lock);
2875 	pi->link_fault = 1;
2876 	spin_unlock(&adapter->work_lock);
2877 }
2878 
2879 static int t3_adapter_error(struct adapter *adapter, int reset, int on_wq)
2880 {
2881 	int i, ret = 0;
2882 
2883 	if (is_offload(adapter) &&
2884 	    test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2885 		cxgb3_event_notify(&adapter->tdev, OFFLOAD_STATUS_DOWN, 0);
2886 		offload_close(&adapter->tdev);
2887 	}
2888 
2889 	/* Stop all ports */
2890 	for_each_port(adapter, i) {
2891 		struct net_device *netdev = adapter->port[i];
2892 
2893 		if (netif_running(netdev))
2894 			__cxgb_close(netdev, on_wq);
2895 	}
2896 
2897 	/* Stop SGE timers */
2898 	t3_stop_sge_timers(adapter);
2899 
2900 	adapter->flags &= ~FULL_INIT_DONE;
2901 
2902 	if (reset)
2903 		ret = t3_reset_adapter(adapter);
2904 
2905 	pci_disable_device(adapter->pdev);
2906 
2907 	return ret;
2908 }
2909 
2910 static int t3_reenable_adapter(struct adapter *adapter)
2911 {
2912 	if (pci_enable_device(adapter->pdev)) {
2913 		dev_err(&adapter->pdev->dev,
2914 			"Cannot re-enable PCI device after reset.\n");
2915 		goto err;
2916 	}
2917 	pci_set_master(adapter->pdev);
2918 	pci_restore_state(adapter->pdev);
2919 	pci_save_state(adapter->pdev);
2920 
2921 	/* Free sge resources */
2922 	t3_free_sge_resources(adapter);
2923 
2924 	if (t3_replay_prep_adapter(adapter))
2925 		goto err;
2926 
2927 	return 0;
2928 err:
2929 	return -1;
2930 }
2931 
2932 static void t3_resume_ports(struct adapter *adapter)
2933 {
2934 	int i;
2935 
2936 	/* Restart the ports */
2937 	for_each_port(adapter, i) {
2938 		struct net_device *netdev = adapter->port[i];
2939 
2940 		if (netif_running(netdev)) {
2941 			if (cxgb_open(netdev)) {
2942 				dev_err(&adapter->pdev->dev,
2943 					"can't bring device back up"
2944 					" after reset\n");
2945 				continue;
2946 			}
2947 		}
2948 	}
2949 
2950 	if (is_offload(adapter) && !ofld_disable)
2951 		cxgb3_event_notify(&adapter->tdev, OFFLOAD_STATUS_UP, 0);
2952 }
2953 
2954 /*
2955  * processes a fatal error.
2956  * Bring the ports down, reset the chip, bring the ports back up.
2957  */
2958 static void fatal_error_task(struct work_struct *work)
2959 {
2960 	struct adapter *adapter = container_of(work, struct adapter,
2961 					       fatal_error_handler_task);
2962 	int err = 0;
2963 
2964 	rtnl_lock();
2965 	err = t3_adapter_error(adapter, 1, 1);
2966 	if (!err)
2967 		err = t3_reenable_adapter(adapter);
2968 	if (!err)
2969 		t3_resume_ports(adapter);
2970 
2971 	CH_ALERT(adapter, "adapter reset %s\n", err ? "failed" : "succeeded");
2972 	rtnl_unlock();
2973 }
2974 
2975 void t3_fatal_err(struct adapter *adapter)
2976 {
2977 	unsigned int fw_status[4];
2978 
2979 	if (adapter->flags & FULL_INIT_DONE) {
2980 		t3_sge_stop(adapter);
2981 		t3_write_reg(adapter, A_XGM_TX_CTRL, 0);
2982 		t3_write_reg(adapter, A_XGM_RX_CTRL, 0);
2983 		t3_write_reg(adapter, XGM_REG(A_XGM_TX_CTRL, 1), 0);
2984 		t3_write_reg(adapter, XGM_REG(A_XGM_RX_CTRL, 1), 0);
2985 
2986 		spin_lock(&adapter->work_lock);
2987 		t3_intr_disable(adapter);
2988 		queue_work(cxgb3_wq, &adapter->fatal_error_handler_task);
2989 		spin_unlock(&adapter->work_lock);
2990 	}
2991 	CH_ALERT(adapter, "encountered fatal error, operation suspended\n");
2992 	if (!t3_cim_ctl_blk_read(adapter, 0xa0, 4, fw_status))
2993 		CH_ALERT(adapter, "FW status: 0x%x, 0x%x, 0x%x, 0x%x\n",
2994 			 fw_status[0], fw_status[1],
2995 			 fw_status[2], fw_status[3]);
2996 }
2997 
2998 /**
2999  * t3_io_error_detected - called when PCI error is detected
3000  * @pdev: Pointer to PCI device
3001  * @state: The current pci connection state
3002  *
3003  * This function is called after a PCI bus error affecting
3004  * this device has been detected.
3005  */
3006 static pci_ers_result_t t3_io_error_detected(struct pci_dev *pdev,
3007 					     pci_channel_state_t state)
3008 {
3009 	struct adapter *adapter = pci_get_drvdata(pdev);
3010 
3011 	if (state == pci_channel_io_perm_failure)
3012 		return PCI_ERS_RESULT_DISCONNECT;
3013 
3014 	t3_adapter_error(adapter, 0, 0);
3015 
3016 	/* Request a slot reset. */
3017 	return PCI_ERS_RESULT_NEED_RESET;
3018 }
3019 
3020 /**
3021  * t3_io_slot_reset - called after the pci bus has been reset.
3022  * @pdev: Pointer to PCI device
3023  *
3024  * Restart the card from scratch, as if from a cold-boot.
3025  */
3026 static pci_ers_result_t t3_io_slot_reset(struct pci_dev *pdev)
3027 {
3028 	struct adapter *adapter = pci_get_drvdata(pdev);
3029 
3030 	if (!t3_reenable_adapter(adapter))
3031 		return PCI_ERS_RESULT_RECOVERED;
3032 
3033 	return PCI_ERS_RESULT_DISCONNECT;
3034 }
3035 
3036 /**
3037  * t3_io_resume - called when traffic can start flowing again.
3038  * @pdev: Pointer to PCI device
3039  *
3040  * This callback is called when the error recovery driver tells us that
3041  * its OK to resume normal operation.
3042  */
3043 static void t3_io_resume(struct pci_dev *pdev)
3044 {
3045 	struct adapter *adapter = pci_get_drvdata(pdev);
3046 
3047 	CH_ALERT(adapter, "adapter recovering, PEX ERR 0x%x\n",
3048 		 t3_read_reg(adapter, A_PCIE_PEX_ERR));
3049 
3050 	rtnl_lock();
3051 	t3_resume_ports(adapter);
3052 	rtnl_unlock();
3053 }
3054 
3055 static const struct pci_error_handlers t3_err_handler = {
3056 	.error_detected = t3_io_error_detected,
3057 	.slot_reset = t3_io_slot_reset,
3058 	.resume = t3_io_resume,
3059 };
3060 
3061 /*
3062  * Set the number of qsets based on the number of CPUs and the number of ports,
3063  * not to exceed the number of available qsets, assuming there are enough qsets
3064  * per port in HW.
3065  */
3066 static void set_nqsets(struct adapter *adap)
3067 {
3068 	int i, j = 0;
3069 	int num_cpus = netif_get_num_default_rss_queues();
3070 	int hwports = adap->params.nports;
3071 	int nqsets = adap->msix_nvectors - 1;
3072 
3073 	if (adap->params.rev > 0 && adap->flags & USING_MSIX) {
3074 		if (hwports == 2 &&
3075 		    (hwports * nqsets > SGE_QSETS ||
3076 		     num_cpus >= nqsets / hwports))
3077 			nqsets /= hwports;
3078 		if (nqsets > num_cpus)
3079 			nqsets = num_cpus;
3080 		if (nqsets < 1 || hwports == 4)
3081 			nqsets = 1;
3082 	} else
3083 		nqsets = 1;
3084 
3085 	for_each_port(adap, i) {
3086 		struct port_info *pi = adap2pinfo(adap, i);
3087 
3088 		pi->first_qset = j;
3089 		pi->nqsets = nqsets;
3090 		j = pi->first_qset + nqsets;
3091 
3092 		dev_info(&adap->pdev->dev,
3093 			 "Port %d using %d queue sets.\n", i, nqsets);
3094 	}
3095 }
3096 
3097 static int cxgb_enable_msix(struct adapter *adap)
3098 {
3099 	struct msix_entry entries[SGE_QSETS + 1];
3100 	int vectors;
3101 	int i;
3102 
3103 	vectors = ARRAY_SIZE(entries);
3104 	for (i = 0; i < vectors; ++i)
3105 		entries[i].entry = i;
3106 
3107 	vectors = pci_enable_msix_range(adap->pdev, entries,
3108 					adap->params.nports + 1, vectors);
3109 	if (vectors < 0)
3110 		return vectors;
3111 
3112 	for (i = 0; i < vectors; ++i)
3113 		adap->msix_info[i].vec = entries[i].vector;
3114 	adap->msix_nvectors = vectors;
3115 
3116 	return 0;
3117 }
3118 
3119 static void print_port_info(struct adapter *adap, const struct adapter_info *ai)
3120 {
3121 	static const char *pci_variant[] = {
3122 		"PCI", "PCI-X", "PCI-X ECC", "PCI-X 266", "PCI Express"
3123 	};
3124 
3125 	int i;
3126 	char buf[80];
3127 
3128 	if (is_pcie(adap))
3129 		snprintf(buf, sizeof(buf), "%s x%d",
3130 			 pci_variant[adap->params.pci.variant],
3131 			 adap->params.pci.width);
3132 	else
3133 		snprintf(buf, sizeof(buf), "%s %dMHz/%d-bit",
3134 			 pci_variant[adap->params.pci.variant],
3135 			 adap->params.pci.speed, adap->params.pci.width);
3136 
3137 	for_each_port(adap, i) {
3138 		struct net_device *dev = adap->port[i];
3139 		const struct port_info *pi = netdev_priv(dev);
3140 
3141 		if (!test_bit(i, &adap->registered_device_map))
3142 			continue;
3143 		netdev_info(dev, "%s %s %sNIC (rev %d) %s%s\n",
3144 			    ai->desc, pi->phy.desc,
3145 			    is_offload(adap) ? "R" : "", adap->params.rev, buf,
3146 			    (adap->flags & USING_MSIX) ? " MSI-X" :
3147 			    (adap->flags & USING_MSI) ? " MSI" : "");
3148 		if (adap->name == dev->name && adap->params.vpd.mclk)
3149 			pr_info("%s: %uMB CM, %uMB PMTX, %uMB PMRX, S/N: %s\n",
3150 			       adap->name, t3_mc7_size(&adap->cm) >> 20,
3151 			       t3_mc7_size(&adap->pmtx) >> 20,
3152 			       t3_mc7_size(&adap->pmrx) >> 20,
3153 			       adap->params.vpd.sn);
3154 	}
3155 }
3156 
3157 static const struct net_device_ops cxgb_netdev_ops = {
3158 	.ndo_open		= cxgb_open,
3159 	.ndo_stop		= cxgb_close,
3160 	.ndo_start_xmit		= t3_eth_xmit,
3161 	.ndo_get_stats		= cxgb_get_stats,
3162 	.ndo_validate_addr	= eth_validate_addr,
3163 	.ndo_set_rx_mode	= cxgb_set_rxmode,
3164 	.ndo_do_ioctl		= cxgb_ioctl,
3165 	.ndo_change_mtu		= cxgb_change_mtu,
3166 	.ndo_set_mac_address	= cxgb_set_mac_addr,
3167 	.ndo_fix_features	= cxgb_fix_features,
3168 	.ndo_set_features	= cxgb_set_features,
3169 #ifdef CONFIG_NET_POLL_CONTROLLER
3170 	.ndo_poll_controller	= cxgb_netpoll,
3171 #endif
3172 };
3173 
3174 static void cxgb3_init_iscsi_mac(struct net_device *dev)
3175 {
3176 	struct port_info *pi = netdev_priv(dev);
3177 
3178 	memcpy(pi->iscsic.mac_addr, dev->dev_addr, ETH_ALEN);
3179 	pi->iscsic.mac_addr[3] |= 0x80;
3180 }
3181 
3182 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
3183 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
3184 			NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
3185 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
3186 {
3187 	int i, err, pci_using_dac = 0;
3188 	resource_size_t mmio_start, mmio_len;
3189 	const struct adapter_info *ai;
3190 	struct adapter *adapter = NULL;
3191 	struct port_info *pi;
3192 
3193 	pr_info_once("%s - version %s\n", DRV_DESC, DRV_VERSION);
3194 
3195 	if (!cxgb3_wq) {
3196 		cxgb3_wq = create_singlethread_workqueue(DRV_NAME);
3197 		if (!cxgb3_wq) {
3198 			pr_err("cannot initialize work queue\n");
3199 			return -ENOMEM;
3200 		}
3201 	}
3202 
3203 	err = pci_enable_device(pdev);
3204 	if (err) {
3205 		dev_err(&pdev->dev, "cannot enable PCI device\n");
3206 		goto out;
3207 	}
3208 
3209 	err = pci_request_regions(pdev, DRV_NAME);
3210 	if (err) {
3211 		/* Just info, some other driver may have claimed the device. */
3212 		dev_info(&pdev->dev, "cannot obtain PCI resources\n");
3213 		goto out_disable_device;
3214 	}
3215 
3216 	if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
3217 		pci_using_dac = 1;
3218 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3219 		if (err) {
3220 			dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
3221 			       "coherent allocations\n");
3222 			goto out_release_regions;
3223 		}
3224 	} else if ((err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32))) != 0) {
3225 		dev_err(&pdev->dev, "no usable DMA configuration\n");
3226 		goto out_release_regions;
3227 	}
3228 
3229 	pci_set_master(pdev);
3230 	pci_save_state(pdev);
3231 
3232 	mmio_start = pci_resource_start(pdev, 0);
3233 	mmio_len = pci_resource_len(pdev, 0);
3234 	ai = t3_get_adapter_info(ent->driver_data);
3235 
3236 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
3237 	if (!adapter) {
3238 		err = -ENOMEM;
3239 		goto out_release_regions;
3240 	}
3241 
3242 	adapter->nofail_skb =
3243 		alloc_skb(sizeof(struct cpl_set_tcb_field), GFP_KERNEL);
3244 	if (!adapter->nofail_skb) {
3245 		dev_err(&pdev->dev, "cannot allocate nofail buffer\n");
3246 		err = -ENOMEM;
3247 		goto out_free_adapter;
3248 	}
3249 
3250 	adapter->regs = ioremap_nocache(mmio_start, mmio_len);
3251 	if (!adapter->regs) {
3252 		dev_err(&pdev->dev, "cannot map device registers\n");
3253 		err = -ENOMEM;
3254 		goto out_free_adapter;
3255 	}
3256 
3257 	adapter->pdev = pdev;
3258 	adapter->name = pci_name(pdev);
3259 	adapter->msg_enable = dflt_msg_enable;
3260 	adapter->mmio_len = mmio_len;
3261 
3262 	mutex_init(&adapter->mdio_lock);
3263 	spin_lock_init(&adapter->work_lock);
3264 	spin_lock_init(&adapter->stats_lock);
3265 
3266 	INIT_LIST_HEAD(&adapter->adapter_list);
3267 	INIT_WORK(&adapter->ext_intr_handler_task, ext_intr_task);
3268 	INIT_WORK(&adapter->fatal_error_handler_task, fatal_error_task);
3269 
3270 	INIT_WORK(&adapter->db_full_task, db_full_task);
3271 	INIT_WORK(&adapter->db_empty_task, db_empty_task);
3272 	INIT_WORK(&adapter->db_drop_task, db_drop_task);
3273 
3274 	INIT_DELAYED_WORK(&adapter->adap_check_task, t3_adap_check_task);
3275 
3276 	for (i = 0; i < ai->nports0 + ai->nports1; ++i) {
3277 		struct net_device *netdev;
3278 
3279 		netdev = alloc_etherdev_mq(sizeof(struct port_info), SGE_QSETS);
3280 		if (!netdev) {
3281 			err = -ENOMEM;
3282 			goto out_free_dev;
3283 		}
3284 
3285 		SET_NETDEV_DEV(netdev, &pdev->dev);
3286 
3287 		adapter->port[i] = netdev;
3288 		pi = netdev_priv(netdev);
3289 		pi->adapter = adapter;
3290 		pi->port_id = i;
3291 		netif_carrier_off(netdev);
3292 		netdev->irq = pdev->irq;
3293 		netdev->mem_start = mmio_start;
3294 		netdev->mem_end = mmio_start + mmio_len - 1;
3295 		netdev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM |
3296 			NETIF_F_TSO | NETIF_F_RXCSUM | NETIF_F_HW_VLAN_CTAG_RX;
3297 		netdev->features |= netdev->hw_features |
3298 				    NETIF_F_HW_VLAN_CTAG_TX;
3299 		netdev->vlan_features |= netdev->features & VLAN_FEAT;
3300 		if (pci_using_dac)
3301 			netdev->features |= NETIF_F_HIGHDMA;
3302 
3303 		netdev->netdev_ops = &cxgb_netdev_ops;
3304 		netdev->ethtool_ops = &cxgb_ethtool_ops;
3305 		netdev->min_mtu = 81;
3306 		netdev->max_mtu = ETH_MAX_MTU;
3307 	}
3308 
3309 	pci_set_drvdata(pdev, adapter);
3310 	if (t3_prep_adapter(adapter, ai, 1) < 0) {
3311 		err = -ENODEV;
3312 		goto out_free_dev;
3313 	}
3314 
3315 	/*
3316 	 * The card is now ready to go.  If any errors occur during device
3317 	 * registration we do not fail the whole card but rather proceed only
3318 	 * with the ports we manage to register successfully.  However we must
3319 	 * register at least one net device.
3320 	 */
3321 	for_each_port(adapter, i) {
3322 		err = register_netdev(adapter->port[i]);
3323 		if (err)
3324 			dev_warn(&pdev->dev,
3325 				 "cannot register net device %s, skipping\n",
3326 				 adapter->port[i]->name);
3327 		else {
3328 			/*
3329 			 * Change the name we use for messages to the name of
3330 			 * the first successfully registered interface.
3331 			 */
3332 			if (!adapter->registered_device_map)
3333 				adapter->name = adapter->port[i]->name;
3334 
3335 			__set_bit(i, &adapter->registered_device_map);
3336 		}
3337 	}
3338 	if (!adapter->registered_device_map) {
3339 		dev_err(&pdev->dev, "could not register any net devices\n");
3340 		goto out_free_dev;
3341 	}
3342 
3343 	for_each_port(adapter, i)
3344 		cxgb3_init_iscsi_mac(adapter->port[i]);
3345 
3346 	/* Driver's ready. Reflect it on LEDs */
3347 	t3_led_ready(adapter);
3348 
3349 	if (is_offload(adapter)) {
3350 		__set_bit(OFFLOAD_DEVMAP_BIT, &adapter->registered_device_map);
3351 		cxgb3_adapter_ofld(adapter);
3352 	}
3353 
3354 	/* See what interrupts we'll be using */
3355 	if (msi > 1 && cxgb_enable_msix(adapter) == 0)
3356 		adapter->flags |= USING_MSIX;
3357 	else if (msi > 0 && pci_enable_msi(pdev) == 0)
3358 		adapter->flags |= USING_MSI;
3359 
3360 	set_nqsets(adapter);
3361 
3362 	err = sysfs_create_group(&adapter->port[0]->dev.kobj,
3363 				 &cxgb3_attr_group);
3364 
3365 	print_port_info(adapter, ai);
3366 	return 0;
3367 
3368 out_free_dev:
3369 	iounmap(adapter->regs);
3370 	for (i = ai->nports0 + ai->nports1 - 1; i >= 0; --i)
3371 		if (adapter->port[i])
3372 			free_netdev(adapter->port[i]);
3373 
3374 out_free_adapter:
3375 	kfree(adapter);
3376 
3377 out_release_regions:
3378 	pci_release_regions(pdev);
3379 out_disable_device:
3380 	pci_disable_device(pdev);
3381 out:
3382 	return err;
3383 }
3384 
3385 static void remove_one(struct pci_dev *pdev)
3386 {
3387 	struct adapter *adapter = pci_get_drvdata(pdev);
3388 
3389 	if (adapter) {
3390 		int i;
3391 
3392 		t3_sge_stop(adapter);
3393 		sysfs_remove_group(&adapter->port[0]->dev.kobj,
3394 				   &cxgb3_attr_group);
3395 
3396 		if (is_offload(adapter)) {
3397 			cxgb3_adapter_unofld(adapter);
3398 			if (test_bit(OFFLOAD_DEVMAP_BIT,
3399 				     &adapter->open_device_map))
3400 				offload_close(&adapter->tdev);
3401 		}
3402 
3403 		for_each_port(adapter, i)
3404 		    if (test_bit(i, &adapter->registered_device_map))
3405 			unregister_netdev(adapter->port[i]);
3406 
3407 		t3_stop_sge_timers(adapter);
3408 		t3_free_sge_resources(adapter);
3409 		cxgb_disable_msi(adapter);
3410 
3411 		for_each_port(adapter, i)
3412 			if (adapter->port[i])
3413 				free_netdev(adapter->port[i]);
3414 
3415 		iounmap(adapter->regs);
3416 		if (adapter->nofail_skb)
3417 			kfree_skb(adapter->nofail_skb);
3418 		kfree(adapter);
3419 		pci_release_regions(pdev);
3420 		pci_disable_device(pdev);
3421 	}
3422 }
3423 
3424 static struct pci_driver driver = {
3425 	.name = DRV_NAME,
3426 	.id_table = cxgb3_pci_tbl,
3427 	.probe = init_one,
3428 	.remove = remove_one,
3429 	.err_handler = &t3_err_handler,
3430 };
3431 
3432 static int __init cxgb3_init_module(void)
3433 {
3434 	int ret;
3435 
3436 	cxgb3_offload_init();
3437 
3438 	ret = pci_register_driver(&driver);
3439 	return ret;
3440 }
3441 
3442 static void __exit cxgb3_cleanup_module(void)
3443 {
3444 	pci_unregister_driver(&driver);
3445 	if (cxgb3_wq)
3446 		destroy_workqueue(cxgb3_wq);
3447 }
3448 
3449 module_init(cxgb3_init_module);
3450 module_exit(cxgb3_cleanup_module);
3451