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/init.h>
37 #include <linux/pci.h>
38 #include <linux/dma-mapping.h>
39 #include <linux/netdevice.h>
40 #include <linux/etherdevice.h>
41 #include <linux/if_vlan.h>
42 #include <linux/mdio.h>
43 #include <linux/sockios.h>
44 #include <linux/workqueue.h>
45 #include <linux/proc_fs.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/firmware.h>
48 #include <linux/log2.h>
49 #include <linux/stringify.h>
50 #include <linux/sched.h>
51 #include <linux/slab.h>
52 #include <linux/uaccess.h>
53 #include <linux/nospec.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_DEVICE_TABLE(pci, cxgb3_pci_tbl);
109 
110 static int dflt_msg_enable = DFLT_MSG_ENABLE;
111 
112 module_param(dflt_msg_enable, int, 0644);
113 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T3 default message enable bitmap");
114 
115 /*
116  * The driver uses the best interrupt scheme available on a platform in the
117  * order MSI-X, MSI, legacy pin interrupts.  This parameter determines which
118  * of these schemes the driver may consider as follows:
119  *
120  * msi = 2: choose from among all three options
121  * msi = 1: only consider MSI and pin interrupts
122  * msi = 0: force pin interrupts
123  */
124 static int msi = 2;
125 
126 module_param(msi, int, 0644);
127 MODULE_PARM_DESC(msi, "whether to use MSI or MSI-X");
128 
129 /*
130  * The driver enables offload as a default.
131  * To disable it, use ofld_disable = 1.
132  */
133 
134 static int ofld_disable = 0;
135 
136 module_param(ofld_disable, int, 0644);
137 MODULE_PARM_DESC(ofld_disable, "whether to enable offload at init time or not");
138 
139 /*
140  * We have work elements that we need to cancel when an interface is taken
141  * down.  Normally the work elements would be executed by keventd but that
142  * can deadlock because of linkwatch.  If our close method takes the rtnl
143  * lock and linkwatch is ahead of our work elements in keventd, linkwatch
144  * will block keventd as it needs the rtnl lock, and we'll deadlock waiting
145  * for our work to complete.  Get our own work queue to solve this.
146  */
147 struct workqueue_struct *cxgb3_wq;
148 
149 /**
150  *	link_report - show link status and link speed/duplex
151  *	@dev: the port whose settings are to be reported
152  *
153  *	Shows the link status, speed, and duplex of a port.
154  */
155 static void link_report(struct net_device *dev)
156 {
157 	if (!netif_carrier_ok(dev))
158 		netdev_info(dev, "link down\n");
159 	else {
160 		const char *s = "10Mbps";
161 		const struct port_info *p = netdev_priv(dev);
162 
163 		switch (p->link_config.speed) {
164 		case SPEED_10000:
165 			s = "10Gbps";
166 			break;
167 		case SPEED_1000:
168 			s = "1000Mbps";
169 			break;
170 		case SPEED_100:
171 			s = "100Mbps";
172 			break;
173 		}
174 
175 		netdev_info(dev, "link up, %s, %s-duplex\n",
176 			    s, p->link_config.duplex == DUPLEX_FULL
177 			    ? "full" : "half");
178 	}
179 }
180 
181 static void enable_tx_fifo_drain(struct adapter *adapter,
182 				 struct port_info *pi)
183 {
184 	t3_set_reg_field(adapter, A_XGM_TXFIFO_CFG + pi->mac.offset, 0,
185 			 F_ENDROPPKT);
186 	t3_write_reg(adapter, A_XGM_RX_CTRL + pi->mac.offset, 0);
187 	t3_write_reg(adapter, A_XGM_TX_CTRL + pi->mac.offset, F_TXEN);
188 	t3_write_reg(adapter, A_XGM_RX_CTRL + pi->mac.offset, F_RXEN);
189 }
190 
191 static void disable_tx_fifo_drain(struct adapter *adapter,
192 				  struct port_info *pi)
193 {
194 	t3_set_reg_field(adapter, A_XGM_TXFIFO_CFG + pi->mac.offset,
195 			 F_ENDROPPKT, 0);
196 }
197 
198 void t3_os_link_fault(struct adapter *adap, int port_id, int state)
199 {
200 	struct net_device *dev = adap->port[port_id];
201 	struct port_info *pi = netdev_priv(dev);
202 
203 	if (state == netif_carrier_ok(dev))
204 		return;
205 
206 	if (state) {
207 		struct cmac *mac = &pi->mac;
208 
209 		netif_carrier_on(dev);
210 
211 		disable_tx_fifo_drain(adap, pi);
212 
213 		/* Clear local faults */
214 		t3_xgm_intr_disable(adap, pi->port_id);
215 		t3_read_reg(adap, A_XGM_INT_STATUS +
216 				    pi->mac.offset);
217 		t3_write_reg(adap,
218 			     A_XGM_INT_CAUSE + pi->mac.offset,
219 			     F_XGM_INT);
220 
221 		t3_set_reg_field(adap,
222 				 A_XGM_INT_ENABLE +
223 				 pi->mac.offset,
224 				 F_XGM_INT, F_XGM_INT);
225 		t3_xgm_intr_enable(adap, pi->port_id);
226 
227 		t3_mac_enable(mac, MAC_DIRECTION_TX);
228 	} else {
229 		netif_carrier_off(dev);
230 
231 		/* Flush TX FIFO */
232 		enable_tx_fifo_drain(adap, pi);
233 	}
234 	link_report(dev);
235 }
236 
237 /**
238  *	t3_os_link_changed - handle link status changes
239  *	@adapter: the adapter associated with the link change
240  *	@port_id: the port index whose limk status has changed
241  *	@link_stat: the new status of the link
242  *	@speed: the new speed setting
243  *	@duplex: the new duplex setting
244  *	@pause: the new flow-control setting
245  *
246  *	This is the OS-dependent handler for link status changes.  The OS
247  *	neutral handler takes care of most of the processing for these events,
248  *	then calls this handler for any OS-specific processing.
249  */
250 void t3_os_link_changed(struct adapter *adapter, int port_id, int link_stat,
251 			int speed, int duplex, int pause)
252 {
253 	struct net_device *dev = adapter->port[port_id];
254 	struct port_info *pi = netdev_priv(dev);
255 	struct cmac *mac = &pi->mac;
256 
257 	/* Skip changes from disabled ports. */
258 	if (!netif_running(dev))
259 		return;
260 
261 	if (link_stat != netif_carrier_ok(dev)) {
262 		if (link_stat) {
263 			disable_tx_fifo_drain(adapter, pi);
264 
265 			t3_mac_enable(mac, MAC_DIRECTION_RX);
266 
267 			/* Clear local faults */
268 			t3_xgm_intr_disable(adapter, pi->port_id);
269 			t3_read_reg(adapter, A_XGM_INT_STATUS +
270 				    pi->mac.offset);
271 			t3_write_reg(adapter,
272 				     A_XGM_INT_CAUSE + pi->mac.offset,
273 				     F_XGM_INT);
274 
275 			t3_set_reg_field(adapter,
276 					 A_XGM_INT_ENABLE + pi->mac.offset,
277 					 F_XGM_INT, F_XGM_INT);
278 			t3_xgm_intr_enable(adapter, pi->port_id);
279 
280 			netif_carrier_on(dev);
281 		} else {
282 			netif_carrier_off(dev);
283 
284 			t3_xgm_intr_disable(adapter, pi->port_id);
285 			t3_read_reg(adapter, A_XGM_INT_STATUS + pi->mac.offset);
286 			t3_set_reg_field(adapter,
287 					 A_XGM_INT_ENABLE + pi->mac.offset,
288 					 F_XGM_INT, 0);
289 
290 			if (is_10G(adapter))
291 				pi->phy.ops->power_down(&pi->phy, 1);
292 
293 			t3_read_reg(adapter, A_XGM_INT_STATUS + pi->mac.offset);
294 			t3_mac_disable(mac, MAC_DIRECTION_RX);
295 			t3_link_start(&pi->phy, mac, &pi->link_config);
296 
297 			/* Flush TX FIFO */
298 			enable_tx_fifo_drain(adapter, pi);
299 		}
300 
301 		link_report(dev);
302 	}
303 }
304 
305 /**
306  *	t3_os_phymod_changed - handle PHY module changes
307  *	@adap: the adapter associated with the link change
308  *	@port_id: the port index whose limk status has changed
309  *
310  *	This is the OS-dependent handler for PHY module changes.  It is
311  *	invoked when a PHY module is removed or inserted for any OS-specific
312  *	processing.
313  */
314 void t3_os_phymod_changed(struct adapter *adap, int port_id)
315 {
316 	static const char *mod_str[] = {
317 		NULL, "SR", "LR", "LRM", "TWINAX", "TWINAX", "unknown"
318 	};
319 
320 	const struct net_device *dev = adap->port[port_id];
321 	const struct port_info *pi = netdev_priv(dev);
322 
323 	if (pi->phy.modtype == phy_modtype_none)
324 		netdev_info(dev, "PHY module unplugged\n");
325 	else
326 		netdev_info(dev, "%s PHY module inserted\n",
327 			    mod_str[pi->phy.modtype]);
328 }
329 
330 static void cxgb_set_rxmode(struct net_device *dev)
331 {
332 	struct port_info *pi = netdev_priv(dev);
333 
334 	t3_mac_set_rx_mode(&pi->mac, dev);
335 }
336 
337 /**
338  *	link_start - enable a port
339  *	@dev: the device to enable
340  *
341  *	Performs the MAC and PHY actions needed to enable a port.
342  */
343 static void link_start(struct net_device *dev)
344 {
345 	struct port_info *pi = netdev_priv(dev);
346 	struct cmac *mac = &pi->mac;
347 
348 	t3_mac_reset(mac);
349 	t3_mac_set_num_ucast(mac, MAX_MAC_IDX);
350 	t3_mac_set_mtu(mac, dev->mtu);
351 	t3_mac_set_address(mac, LAN_MAC_IDX, dev->dev_addr);
352 	t3_mac_set_address(mac, SAN_MAC_IDX, pi->iscsic.mac_addr);
353 	t3_mac_set_rx_mode(mac, dev);
354 	t3_link_start(&pi->phy, mac, &pi->link_config);
355 	t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
356 }
357 
358 static inline void cxgb_disable_msi(struct adapter *adapter)
359 {
360 	if (adapter->flags & USING_MSIX) {
361 		pci_disable_msix(adapter->pdev);
362 		adapter->flags &= ~USING_MSIX;
363 	} else if (adapter->flags & USING_MSI) {
364 		pci_disable_msi(adapter->pdev);
365 		adapter->flags &= ~USING_MSI;
366 	}
367 }
368 
369 /*
370  * Interrupt handler for asynchronous events used with MSI-X.
371  */
372 static irqreturn_t t3_async_intr_handler(int irq, void *cookie)
373 {
374 	t3_slow_intr_handler(cookie);
375 	return IRQ_HANDLED;
376 }
377 
378 /*
379  * Name the MSI-X interrupts.
380  */
381 static void name_msix_vecs(struct adapter *adap)
382 {
383 	int i, j, msi_idx = 1, n = sizeof(adap->msix_info[0].desc) - 1;
384 
385 	snprintf(adap->msix_info[0].desc, n, "%s", adap->name);
386 	adap->msix_info[0].desc[n] = 0;
387 
388 	for_each_port(adap, j) {
389 		struct net_device *d = adap->port[j];
390 		const struct port_info *pi = netdev_priv(d);
391 
392 		for (i = 0; i < pi->nqsets; i++, msi_idx++) {
393 			snprintf(adap->msix_info[msi_idx].desc, n,
394 				 "%s-%d", d->name, pi->first_qset + i);
395 			adap->msix_info[msi_idx].desc[n] = 0;
396 		}
397 	}
398 }
399 
400 static int request_msix_data_irqs(struct adapter *adap)
401 {
402 	int i, j, err, qidx = 0;
403 
404 	for_each_port(adap, i) {
405 		int nqsets = adap2pinfo(adap, i)->nqsets;
406 
407 		for (j = 0; j < nqsets; ++j) {
408 			err = request_irq(adap->msix_info[qidx + 1].vec,
409 					  t3_intr_handler(adap,
410 							  adap->sge.qs[qidx].
411 							  rspq.polling), 0,
412 					  adap->msix_info[qidx + 1].desc,
413 					  &adap->sge.qs[qidx]);
414 			if (err) {
415 				while (--qidx >= 0)
416 					free_irq(adap->msix_info[qidx + 1].vec,
417 						 &adap->sge.qs[qidx]);
418 				return err;
419 			}
420 			qidx++;
421 		}
422 	}
423 	return 0;
424 }
425 
426 static void free_irq_resources(struct adapter *adapter)
427 {
428 	if (adapter->flags & USING_MSIX) {
429 		int i, n = 0;
430 
431 		free_irq(adapter->msix_info[0].vec, adapter);
432 		for_each_port(adapter, i)
433 			n += adap2pinfo(adapter, i)->nqsets;
434 
435 		for (i = 0; i < n; ++i)
436 			free_irq(adapter->msix_info[i + 1].vec,
437 				 &adapter->sge.qs[i]);
438 	} else
439 		free_irq(adapter->pdev->irq, adapter);
440 }
441 
442 static int await_mgmt_replies(struct adapter *adap, unsigned long init_cnt,
443 			      unsigned long n)
444 {
445 	int attempts = 10;
446 
447 	while (adap->sge.qs[0].rspq.offload_pkts < init_cnt + n) {
448 		if (!--attempts)
449 			return -ETIMEDOUT;
450 		msleep(10);
451 	}
452 	return 0;
453 }
454 
455 static int init_tp_parity(struct adapter *adap)
456 {
457 	int i;
458 	struct sk_buff *skb;
459 	struct cpl_set_tcb_field *greq;
460 	unsigned long cnt = adap->sge.qs[0].rspq.offload_pkts;
461 
462 	t3_tp_set_offload_mode(adap, 1);
463 
464 	for (i = 0; i < 16; i++) {
465 		struct cpl_smt_write_req *req;
466 
467 		skb = alloc_skb(sizeof(*req), GFP_KERNEL);
468 		if (!skb)
469 			skb = adap->nofail_skb;
470 		if (!skb)
471 			goto alloc_skb_fail;
472 
473 		req = __skb_put_zero(skb, sizeof(*req));
474 		req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
475 		OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, i));
476 		req->mtu_idx = NMTUS - 1;
477 		req->iff = i;
478 		t3_mgmt_tx(adap, skb);
479 		if (skb == adap->nofail_skb) {
480 			await_mgmt_replies(adap, cnt, i + 1);
481 			adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
482 			if (!adap->nofail_skb)
483 				goto alloc_skb_fail;
484 		}
485 	}
486 
487 	for (i = 0; i < 2048; i++) {
488 		struct cpl_l2t_write_req *req;
489 
490 		skb = alloc_skb(sizeof(*req), GFP_KERNEL);
491 		if (!skb)
492 			skb = adap->nofail_skb;
493 		if (!skb)
494 			goto alloc_skb_fail;
495 
496 		req = __skb_put_zero(skb, sizeof(*req));
497 		req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
498 		OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, i));
499 		req->params = htonl(V_L2T_W_IDX(i));
500 		t3_mgmt_tx(adap, skb);
501 		if (skb == adap->nofail_skb) {
502 			await_mgmt_replies(adap, cnt, 16 + i + 1);
503 			adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
504 			if (!adap->nofail_skb)
505 				goto alloc_skb_fail;
506 		}
507 	}
508 
509 	for (i = 0; i < 2048; i++) {
510 		struct cpl_rte_write_req *req;
511 
512 		skb = alloc_skb(sizeof(*req), GFP_KERNEL);
513 		if (!skb)
514 			skb = adap->nofail_skb;
515 		if (!skb)
516 			goto alloc_skb_fail;
517 
518 		req = __skb_put_zero(skb, sizeof(*req));
519 		req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
520 		OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RTE_WRITE_REQ, i));
521 		req->l2t_idx = htonl(V_L2T_W_IDX(i));
522 		t3_mgmt_tx(adap, skb);
523 		if (skb == adap->nofail_skb) {
524 			await_mgmt_replies(adap, cnt, 16 + 2048 + i + 1);
525 			adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
526 			if (!adap->nofail_skb)
527 				goto alloc_skb_fail;
528 		}
529 	}
530 
531 	skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
532 	if (!skb)
533 		skb = adap->nofail_skb;
534 	if (!skb)
535 		goto alloc_skb_fail;
536 
537 	greq = __skb_put_zero(skb, sizeof(*greq));
538 	greq->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
539 	OPCODE_TID(greq) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, 0));
540 	greq->mask = cpu_to_be64(1);
541 	t3_mgmt_tx(adap, skb);
542 
543 	i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1);
544 	if (skb == adap->nofail_skb) {
545 		i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1);
546 		adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
547 	}
548 
549 	t3_tp_set_offload_mode(adap, 0);
550 	return i;
551 
552 alloc_skb_fail:
553 	t3_tp_set_offload_mode(adap, 0);
554 	return -ENOMEM;
555 }
556 
557 /**
558  *	setup_rss - configure RSS
559  *	@adap: the adapter
560  *
561  *	Sets up RSS to distribute packets to multiple receive queues.  We
562  *	configure the RSS CPU lookup table to distribute to the number of HW
563  *	receive queues, and the response queue lookup table to narrow that
564  *	down to the response queues actually configured for each port.
565  *	We always configure the RSS mapping for two ports since the mapping
566  *	table has plenty of entries.
567  */
568 static void setup_rss(struct adapter *adap)
569 {
570 	int i;
571 	unsigned int nq0 = adap2pinfo(adap, 0)->nqsets;
572 	unsigned int nq1 = adap->port[1] ? adap2pinfo(adap, 1)->nqsets : 1;
573 	u8 cpus[SGE_QSETS + 1];
574 	u16 rspq_map[RSS_TABLE_SIZE + 1];
575 
576 	for (i = 0; i < SGE_QSETS; ++i)
577 		cpus[i] = i;
578 	cpus[SGE_QSETS] = 0xff;	/* terminator */
579 
580 	for (i = 0; i < RSS_TABLE_SIZE / 2; ++i) {
581 		rspq_map[i] = i % nq0;
582 		rspq_map[i + RSS_TABLE_SIZE / 2] = (i % nq1) + nq0;
583 	}
584 	rspq_map[RSS_TABLE_SIZE] = 0xffff; /* terminator */
585 
586 	t3_config_rss(adap, F_RQFEEDBACKENABLE | F_TNLLKPEN | F_TNLMAPEN |
587 		      F_TNLPRTEN | F_TNL2TUPEN | F_TNL4TUPEN |
588 		      V_RRCPLCPUSIZE(6) | F_HASHTOEPLITZ, cpus, rspq_map);
589 }
590 
591 static void ring_dbs(struct adapter *adap)
592 {
593 	int i, j;
594 
595 	for (i = 0; i < SGE_QSETS; i++) {
596 		struct sge_qset *qs = &adap->sge.qs[i];
597 
598 		if (qs->adap)
599 			for (j = 0; j < SGE_TXQ_PER_SET; j++)
600 				t3_write_reg(adap, A_SG_KDOORBELL, F_SELEGRCNTX | V_EGRCNTX(qs->txq[j].cntxt_id));
601 	}
602 }
603 
604 static void init_napi(struct adapter *adap)
605 {
606 	int i;
607 
608 	for (i = 0; i < SGE_QSETS; i++) {
609 		struct sge_qset *qs = &adap->sge.qs[i];
610 
611 		if (qs->adap)
612 			netif_napi_add(qs->netdev, &qs->napi, qs->napi.poll,
613 				       64);
614 	}
615 
616 	/*
617 	 * netif_napi_add() can be called only once per napi_struct because it
618 	 * adds each new napi_struct to a list.  Be careful not to call it a
619 	 * second time, e.g., during EEH recovery, by making a note of it.
620 	 */
621 	adap->flags |= NAPI_INIT;
622 }
623 
624 /*
625  * Wait until all NAPI handlers are descheduled.  This includes the handlers of
626  * both netdevices representing interfaces and the dummy ones for the extra
627  * queues.
628  */
629 static void quiesce_rx(struct adapter *adap)
630 {
631 	int i;
632 
633 	for (i = 0; i < SGE_QSETS; i++)
634 		if (adap->sge.qs[i].adap)
635 			napi_disable(&adap->sge.qs[i].napi);
636 }
637 
638 static void enable_all_napi(struct adapter *adap)
639 {
640 	int i;
641 	for (i = 0; i < SGE_QSETS; i++)
642 		if (adap->sge.qs[i].adap)
643 			napi_enable(&adap->sge.qs[i].napi);
644 }
645 
646 /**
647  *	setup_sge_qsets - configure SGE Tx/Rx/response queues
648  *	@adap: the adapter
649  *
650  *	Determines how many sets of SGE queues to use and initializes them.
651  *	We support multiple queue sets per port if we have MSI-X, otherwise
652  *	just one queue set per port.
653  */
654 static int setup_sge_qsets(struct adapter *adap)
655 {
656 	int i, j, err, irq_idx = 0, qset_idx = 0;
657 	unsigned int ntxq = SGE_TXQ_PER_SET;
658 
659 	if (adap->params.rev > 0 && !(adap->flags & USING_MSI))
660 		irq_idx = -1;
661 
662 	for_each_port(adap, i) {
663 		struct net_device *dev = adap->port[i];
664 		struct port_info *pi = netdev_priv(dev);
665 
666 		pi->qs = &adap->sge.qs[pi->first_qset];
667 		for (j = 0; j < pi->nqsets; ++j, ++qset_idx) {
668 			err = t3_sge_alloc_qset(adap, qset_idx, 1,
669 				(adap->flags & USING_MSIX) ? qset_idx + 1 :
670 							     irq_idx,
671 				&adap->params.sge.qset[qset_idx], ntxq, dev,
672 				netdev_get_tx_queue(dev, j));
673 			if (err) {
674 				t3_free_sge_resources(adap);
675 				return err;
676 			}
677 		}
678 	}
679 
680 	return 0;
681 }
682 
683 static ssize_t attr_show(struct device *d, char *buf,
684 			 ssize_t(*format) (struct net_device *, char *))
685 {
686 	ssize_t len;
687 
688 	/* Synchronize with ioctls that may shut down the device */
689 	rtnl_lock();
690 	len = (*format) (to_net_dev(d), buf);
691 	rtnl_unlock();
692 	return len;
693 }
694 
695 static ssize_t attr_store(struct device *d,
696 			  const char *buf, size_t len,
697 			  ssize_t(*set) (struct net_device *, unsigned int),
698 			  unsigned int min_val, unsigned int max_val)
699 {
700 	ssize_t ret;
701 	unsigned int val;
702 
703 	if (!capable(CAP_NET_ADMIN))
704 		return -EPERM;
705 
706 	ret = kstrtouint(buf, 0, &val);
707 	if (ret)
708 		return ret;
709 	if (val < min_val || val > max_val)
710 		return -EINVAL;
711 
712 	rtnl_lock();
713 	ret = (*set) (to_net_dev(d), val);
714 	if (!ret)
715 		ret = len;
716 	rtnl_unlock();
717 	return ret;
718 }
719 
720 #define CXGB3_SHOW(name, val_expr) \
721 static ssize_t format_##name(struct net_device *dev, char *buf) \
722 { \
723 	struct port_info *pi = netdev_priv(dev); \
724 	struct adapter *adap = pi->adapter; \
725 	return sprintf(buf, "%u\n", val_expr); \
726 } \
727 static ssize_t show_##name(struct device *d, struct device_attribute *attr, \
728 			   char *buf) \
729 { \
730 	return attr_show(d, buf, format_##name); \
731 }
732 
733 static ssize_t set_nfilters(struct net_device *dev, unsigned int val)
734 {
735 	struct port_info *pi = netdev_priv(dev);
736 	struct adapter *adap = pi->adapter;
737 	int min_tids = is_offload(adap) ? MC5_MIN_TIDS : 0;
738 
739 	if (adap->flags & FULL_INIT_DONE)
740 		return -EBUSY;
741 	if (val && adap->params.rev == 0)
742 		return -EINVAL;
743 	if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nservers -
744 	    min_tids)
745 		return -EINVAL;
746 	adap->params.mc5.nfilters = val;
747 	return 0;
748 }
749 
750 static ssize_t store_nfilters(struct device *d, struct device_attribute *attr,
751 			      const char *buf, size_t len)
752 {
753 	return attr_store(d, buf, len, set_nfilters, 0, ~0);
754 }
755 
756 static ssize_t set_nservers(struct net_device *dev, unsigned int val)
757 {
758 	struct port_info *pi = netdev_priv(dev);
759 	struct adapter *adap = pi->adapter;
760 
761 	if (adap->flags & FULL_INIT_DONE)
762 		return -EBUSY;
763 	if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nfilters -
764 	    MC5_MIN_TIDS)
765 		return -EINVAL;
766 	adap->params.mc5.nservers = val;
767 	return 0;
768 }
769 
770 static ssize_t store_nservers(struct device *d, struct device_attribute *attr,
771 			      const char *buf, size_t len)
772 {
773 	return attr_store(d, buf, len, set_nservers, 0, ~0);
774 }
775 
776 #define CXGB3_ATTR_R(name, val_expr) \
777 CXGB3_SHOW(name, val_expr) \
778 static DEVICE_ATTR(name, 0444, show_##name, NULL)
779 
780 #define CXGB3_ATTR_RW(name, val_expr, store_method) \
781 CXGB3_SHOW(name, val_expr) \
782 static DEVICE_ATTR(name, 0644, show_##name, store_method)
783 
784 CXGB3_ATTR_R(cam_size, t3_mc5_size(&adap->mc5));
785 CXGB3_ATTR_RW(nfilters, adap->params.mc5.nfilters, store_nfilters);
786 CXGB3_ATTR_RW(nservers, adap->params.mc5.nservers, store_nservers);
787 
788 static struct attribute *cxgb3_attrs[] = {
789 	&dev_attr_cam_size.attr,
790 	&dev_attr_nfilters.attr,
791 	&dev_attr_nservers.attr,
792 	NULL
793 };
794 
795 static const struct attribute_group cxgb3_attr_group = {
796 	.attrs = cxgb3_attrs,
797 };
798 
799 static ssize_t tm_attr_show(struct device *d,
800 			    char *buf, int sched)
801 {
802 	struct port_info *pi = netdev_priv(to_net_dev(d));
803 	struct adapter *adap = pi->adapter;
804 	unsigned int v, addr, bpt, cpt;
805 	ssize_t len;
806 
807 	addr = A_TP_TX_MOD_Q1_Q0_RATE_LIMIT - sched / 2;
808 	rtnl_lock();
809 	t3_write_reg(adap, A_TP_TM_PIO_ADDR, addr);
810 	v = t3_read_reg(adap, A_TP_TM_PIO_DATA);
811 	if (sched & 1)
812 		v >>= 16;
813 	bpt = (v >> 8) & 0xff;
814 	cpt = v & 0xff;
815 	if (!cpt)
816 		len = sprintf(buf, "disabled\n");
817 	else {
818 		v = (adap->params.vpd.cclk * 1000) / cpt;
819 		len = sprintf(buf, "%u Kbps\n", (v * bpt) / 125);
820 	}
821 	rtnl_unlock();
822 	return len;
823 }
824 
825 static ssize_t tm_attr_store(struct device *d,
826 			     const char *buf, size_t len, int sched)
827 {
828 	struct port_info *pi = netdev_priv(to_net_dev(d));
829 	struct adapter *adap = pi->adapter;
830 	unsigned int val;
831 	ssize_t ret;
832 
833 	if (!capable(CAP_NET_ADMIN))
834 		return -EPERM;
835 
836 	ret = kstrtouint(buf, 0, &val);
837 	if (ret)
838 		return ret;
839 	if (val > 10000000)
840 		return -EINVAL;
841 
842 	rtnl_lock();
843 	ret = t3_config_sched(adap, val, sched);
844 	if (!ret)
845 		ret = len;
846 	rtnl_unlock();
847 	return ret;
848 }
849 
850 #define TM_ATTR(name, sched) \
851 static ssize_t show_##name(struct device *d, struct device_attribute *attr, \
852 			   char *buf) \
853 { \
854 	return tm_attr_show(d, buf, sched); \
855 } \
856 static ssize_t store_##name(struct device *d, struct device_attribute *attr, \
857 			    const char *buf, size_t len) \
858 { \
859 	return tm_attr_store(d, buf, len, sched); \
860 } \
861 static DEVICE_ATTR(name, 0644, show_##name, store_##name)
862 
863 TM_ATTR(sched0, 0);
864 TM_ATTR(sched1, 1);
865 TM_ATTR(sched2, 2);
866 TM_ATTR(sched3, 3);
867 TM_ATTR(sched4, 4);
868 TM_ATTR(sched5, 5);
869 TM_ATTR(sched6, 6);
870 TM_ATTR(sched7, 7);
871 
872 static struct attribute *offload_attrs[] = {
873 	&dev_attr_sched0.attr,
874 	&dev_attr_sched1.attr,
875 	&dev_attr_sched2.attr,
876 	&dev_attr_sched3.attr,
877 	&dev_attr_sched4.attr,
878 	&dev_attr_sched5.attr,
879 	&dev_attr_sched6.attr,
880 	&dev_attr_sched7.attr,
881 	NULL
882 };
883 
884 static const struct attribute_group offload_attr_group = {
885 	.attrs = offload_attrs,
886 };
887 
888 /*
889  * Sends an sk_buff to an offload queue driver
890  * after dealing with any active network taps.
891  */
892 static inline int offload_tx(struct t3cdev *tdev, struct sk_buff *skb)
893 {
894 	int ret;
895 
896 	local_bh_disable();
897 	ret = t3_offload_tx(tdev, skb);
898 	local_bh_enable();
899 	return ret;
900 }
901 
902 static int write_smt_entry(struct adapter *adapter, int idx)
903 {
904 	struct cpl_smt_write_req *req;
905 	struct port_info *pi = netdev_priv(adapter->port[idx]);
906 	struct sk_buff *skb = alloc_skb(sizeof(*req), GFP_KERNEL);
907 
908 	if (!skb)
909 		return -ENOMEM;
910 
911 	req = __skb_put(skb, sizeof(*req));
912 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
913 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, idx));
914 	req->mtu_idx = NMTUS - 1;	/* should be 0 but there's a T3 bug */
915 	req->iff = idx;
916 	memcpy(req->src_mac0, adapter->port[idx]->dev_addr, ETH_ALEN);
917 	memcpy(req->src_mac1, pi->iscsic.mac_addr, ETH_ALEN);
918 	skb->priority = 1;
919 	offload_tx(&adapter->tdev, skb);
920 	return 0;
921 }
922 
923 static int init_smt(struct adapter *adapter)
924 {
925 	int i;
926 
927 	for_each_port(adapter, i)
928 	    write_smt_entry(adapter, i);
929 	return 0;
930 }
931 
932 static void init_port_mtus(struct adapter *adapter)
933 {
934 	unsigned int mtus = adapter->port[0]->mtu;
935 
936 	if (adapter->port[1])
937 		mtus |= adapter->port[1]->mtu << 16;
938 	t3_write_reg(adapter, A_TP_MTU_PORT_TABLE, mtus);
939 }
940 
941 static int send_pktsched_cmd(struct adapter *adap, int sched, int qidx, int lo,
942 			      int hi, int port)
943 {
944 	struct sk_buff *skb;
945 	struct mngt_pktsched_wr *req;
946 	int ret;
947 
948 	skb = alloc_skb(sizeof(*req), GFP_KERNEL);
949 	if (!skb)
950 		skb = adap->nofail_skb;
951 	if (!skb)
952 		return -ENOMEM;
953 
954 	req = skb_put(skb, sizeof(*req));
955 	req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_MNGT));
956 	req->mngt_opcode = FW_MNGTOPCODE_PKTSCHED_SET;
957 	req->sched = sched;
958 	req->idx = qidx;
959 	req->min = lo;
960 	req->max = hi;
961 	req->binding = port;
962 	ret = t3_mgmt_tx(adap, skb);
963 	if (skb == adap->nofail_skb) {
964 		adap->nofail_skb = alloc_skb(sizeof(struct cpl_set_tcb_field),
965 					     GFP_KERNEL);
966 		if (!adap->nofail_skb)
967 			ret = -ENOMEM;
968 	}
969 
970 	return ret;
971 }
972 
973 static int bind_qsets(struct adapter *adap)
974 {
975 	int i, j, err = 0;
976 
977 	for_each_port(adap, i) {
978 		const struct port_info *pi = adap2pinfo(adap, i);
979 
980 		for (j = 0; j < pi->nqsets; ++j) {
981 			int ret = send_pktsched_cmd(adap, 1,
982 						    pi->first_qset + j, -1,
983 						    -1, i);
984 			if (ret)
985 				err = ret;
986 		}
987 	}
988 
989 	return err;
990 }
991 
992 #define FW_VERSION __stringify(FW_VERSION_MAJOR) "."			\
993 	__stringify(FW_VERSION_MINOR) "." __stringify(FW_VERSION_MICRO)
994 #define FW_FNAME "cxgb3/t3fw-" FW_VERSION ".bin"
995 #define TPSRAM_VERSION __stringify(TP_VERSION_MAJOR) "."		\
996 	__stringify(TP_VERSION_MINOR) "." __stringify(TP_VERSION_MICRO)
997 #define TPSRAM_NAME "cxgb3/t3%c_psram-" TPSRAM_VERSION ".bin"
998 #define AEL2005_OPT_EDC_NAME "cxgb3/ael2005_opt_edc.bin"
999 #define AEL2005_TWX_EDC_NAME "cxgb3/ael2005_twx_edc.bin"
1000 #define AEL2020_TWX_EDC_NAME "cxgb3/ael2020_twx_edc.bin"
1001 MODULE_FIRMWARE(FW_FNAME);
1002 MODULE_FIRMWARE("cxgb3/t3b_psram-" TPSRAM_VERSION ".bin");
1003 MODULE_FIRMWARE("cxgb3/t3c_psram-" TPSRAM_VERSION ".bin");
1004 MODULE_FIRMWARE(AEL2005_OPT_EDC_NAME);
1005 MODULE_FIRMWARE(AEL2005_TWX_EDC_NAME);
1006 MODULE_FIRMWARE(AEL2020_TWX_EDC_NAME);
1007 
1008 static inline const char *get_edc_fw_name(int edc_idx)
1009 {
1010 	const char *fw_name = NULL;
1011 
1012 	switch (edc_idx) {
1013 	case EDC_OPT_AEL2005:
1014 		fw_name = AEL2005_OPT_EDC_NAME;
1015 		break;
1016 	case EDC_TWX_AEL2005:
1017 		fw_name = AEL2005_TWX_EDC_NAME;
1018 		break;
1019 	case EDC_TWX_AEL2020:
1020 		fw_name = AEL2020_TWX_EDC_NAME;
1021 		break;
1022 	}
1023 	return fw_name;
1024 }
1025 
1026 int t3_get_edc_fw(struct cphy *phy, int edc_idx, int size)
1027 {
1028 	struct adapter *adapter = phy->adapter;
1029 	const struct firmware *fw;
1030 	const char *fw_name;
1031 	u32 csum;
1032 	const __be32 *p;
1033 	u16 *cache = phy->phy_cache;
1034 	int i, ret = -EINVAL;
1035 
1036 	fw_name = get_edc_fw_name(edc_idx);
1037 	if (fw_name)
1038 		ret = request_firmware(&fw, fw_name, &adapter->pdev->dev);
1039 	if (ret < 0) {
1040 		dev_err(&adapter->pdev->dev,
1041 			"could not upgrade firmware: unable to load %s\n",
1042 			fw_name);
1043 		return ret;
1044 	}
1045 
1046 	/* check size, take checksum in account */
1047 	if (fw->size > size + 4) {
1048 		CH_ERR(adapter, "firmware image too large %u, expected %d\n",
1049 		       (unsigned int)fw->size, size + 4);
1050 		ret = -EINVAL;
1051 	}
1052 
1053 	/* compute checksum */
1054 	p = (const __be32 *)fw->data;
1055 	for (csum = 0, i = 0; i < fw->size / sizeof(csum); i++)
1056 		csum += ntohl(p[i]);
1057 
1058 	if (csum != 0xffffffff) {
1059 		CH_ERR(adapter, "corrupted firmware image, checksum %u\n",
1060 		       csum);
1061 		ret = -EINVAL;
1062 	}
1063 
1064 	for (i = 0; i < size / 4 ; i++) {
1065 		*cache++ = (be32_to_cpu(p[i]) & 0xffff0000) >> 16;
1066 		*cache++ = be32_to_cpu(p[i]) & 0xffff;
1067 	}
1068 
1069 	release_firmware(fw);
1070 
1071 	return ret;
1072 }
1073 
1074 static int upgrade_fw(struct adapter *adap)
1075 {
1076 	int ret;
1077 	const struct firmware *fw;
1078 	struct device *dev = &adap->pdev->dev;
1079 
1080 	ret = request_firmware(&fw, FW_FNAME, dev);
1081 	if (ret < 0) {
1082 		dev_err(dev, "could not upgrade firmware: unable to load %s\n",
1083 			FW_FNAME);
1084 		return ret;
1085 	}
1086 	ret = t3_load_fw(adap, fw->data, fw->size);
1087 	release_firmware(fw);
1088 
1089 	if (ret == 0)
1090 		dev_info(dev, "successful upgrade to firmware %d.%d.%d\n",
1091 			 FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
1092 	else
1093 		dev_err(dev, "failed to upgrade to firmware %d.%d.%d\n",
1094 			FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
1095 
1096 	return ret;
1097 }
1098 
1099 static inline char t3rev2char(struct adapter *adapter)
1100 {
1101 	char rev = 0;
1102 
1103 	switch(adapter->params.rev) {
1104 	case T3_REV_B:
1105 	case T3_REV_B2:
1106 		rev = 'b';
1107 		break;
1108 	case T3_REV_C:
1109 		rev = 'c';
1110 		break;
1111 	}
1112 	return rev;
1113 }
1114 
1115 static int update_tpsram(struct adapter *adap)
1116 {
1117 	const struct firmware *tpsram;
1118 	char buf[64];
1119 	struct device *dev = &adap->pdev->dev;
1120 	int ret;
1121 	char rev;
1122 
1123 	rev = t3rev2char(adap);
1124 	if (!rev)
1125 		return 0;
1126 
1127 	snprintf(buf, sizeof(buf), TPSRAM_NAME, rev);
1128 
1129 	ret = request_firmware(&tpsram, buf, dev);
1130 	if (ret < 0) {
1131 		dev_err(dev, "could not load TP SRAM: unable to load %s\n",
1132 			buf);
1133 		return ret;
1134 	}
1135 
1136 	ret = t3_check_tpsram(adap, tpsram->data, tpsram->size);
1137 	if (ret)
1138 		goto release_tpsram;
1139 
1140 	ret = t3_set_proto_sram(adap, tpsram->data);
1141 	if (ret == 0)
1142 		dev_info(dev,
1143 			 "successful update of protocol engine "
1144 			 "to %d.%d.%d\n",
1145 			 TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
1146 	else
1147 		dev_err(dev, "failed to update of protocol engine %d.%d.%d\n",
1148 			TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
1149 	if (ret)
1150 		dev_err(dev, "loading protocol SRAM failed\n");
1151 
1152 release_tpsram:
1153 	release_firmware(tpsram);
1154 
1155 	return ret;
1156 }
1157 
1158 /**
1159  * t3_synchronize_rx - wait for current Rx processing on a port to complete
1160  * @adap: the adapter
1161  * @p: the port
1162  *
1163  * Ensures that current Rx processing on any of the queues associated with
1164  * the given port completes before returning.  We do this by acquiring and
1165  * releasing the locks of the response queues associated with the port.
1166  */
1167 static void t3_synchronize_rx(struct adapter *adap, const struct port_info *p)
1168 {
1169 	int i;
1170 
1171 	for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
1172 		struct sge_rspq *q = &adap->sge.qs[i].rspq;
1173 
1174 		spin_lock_irq(&q->lock);
1175 		spin_unlock_irq(&q->lock);
1176 	}
1177 }
1178 
1179 static void cxgb_vlan_mode(struct net_device *dev, netdev_features_t features)
1180 {
1181 	struct port_info *pi = netdev_priv(dev);
1182 	struct adapter *adapter = pi->adapter;
1183 
1184 	if (adapter->params.rev > 0) {
1185 		t3_set_vlan_accel(adapter, 1 << pi->port_id,
1186 				  features & NETIF_F_HW_VLAN_CTAG_RX);
1187 	} else {
1188 		/* single control for all ports */
1189 		unsigned int i, have_vlans = features & NETIF_F_HW_VLAN_CTAG_RX;
1190 
1191 		for_each_port(adapter, i)
1192 			have_vlans |=
1193 				adapter->port[i]->features &
1194 				NETIF_F_HW_VLAN_CTAG_RX;
1195 
1196 		t3_set_vlan_accel(adapter, 1, have_vlans);
1197 	}
1198 	t3_synchronize_rx(adapter, pi);
1199 }
1200 
1201 /**
1202  *	cxgb_up - enable the adapter
1203  *	@adap: adapter being enabled
1204  *
1205  *	Called when the first port is enabled, this function performs the
1206  *	actions necessary to make an adapter operational, such as completing
1207  *	the initialization of HW modules, and enabling interrupts.
1208  *
1209  *	Must be called with the rtnl lock held.
1210  */
1211 static int cxgb_up(struct adapter *adap)
1212 {
1213 	int i, err;
1214 
1215 	if (!(adap->flags & FULL_INIT_DONE)) {
1216 		err = t3_check_fw_version(adap);
1217 		if (err == -EINVAL) {
1218 			err = upgrade_fw(adap);
1219 			CH_WARN(adap, "FW upgrade to %d.%d.%d %s\n",
1220 				FW_VERSION_MAJOR, FW_VERSION_MINOR,
1221 				FW_VERSION_MICRO, err ? "failed" : "succeeded");
1222 		}
1223 
1224 		err = t3_check_tpsram_version(adap);
1225 		if (err == -EINVAL) {
1226 			err = update_tpsram(adap);
1227 			CH_WARN(adap, "TP upgrade to %d.%d.%d %s\n",
1228 				TP_VERSION_MAJOR, TP_VERSION_MINOR,
1229 				TP_VERSION_MICRO, err ? "failed" : "succeeded");
1230 		}
1231 
1232 		/*
1233 		 * Clear interrupts now to catch errors if t3_init_hw fails.
1234 		 * We clear them again later as initialization may trigger
1235 		 * conditions that can interrupt.
1236 		 */
1237 		t3_intr_clear(adap);
1238 
1239 		err = t3_init_hw(adap, 0);
1240 		if (err)
1241 			goto out;
1242 
1243 		t3_set_reg_field(adap, A_TP_PARA_REG5, 0, F_RXDDPOFFINIT);
1244 		t3_write_reg(adap, A_ULPRX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12));
1245 
1246 		err = setup_sge_qsets(adap);
1247 		if (err)
1248 			goto out;
1249 
1250 		for_each_port(adap, i)
1251 			cxgb_vlan_mode(adap->port[i], adap->port[i]->features);
1252 
1253 		setup_rss(adap);
1254 		if (!(adap->flags & NAPI_INIT))
1255 			init_napi(adap);
1256 
1257 		t3_start_sge_timers(adap);
1258 		adap->flags |= FULL_INIT_DONE;
1259 	}
1260 
1261 	t3_intr_clear(adap);
1262 
1263 	if (adap->flags & USING_MSIX) {
1264 		name_msix_vecs(adap);
1265 		err = request_irq(adap->msix_info[0].vec,
1266 				  t3_async_intr_handler, 0,
1267 				  adap->msix_info[0].desc, adap);
1268 		if (err)
1269 			goto irq_err;
1270 
1271 		err = request_msix_data_irqs(adap);
1272 		if (err) {
1273 			free_irq(adap->msix_info[0].vec, adap);
1274 			goto irq_err;
1275 		}
1276 	} else {
1277 		err = request_irq(adap->pdev->irq,
1278 				  t3_intr_handler(adap, adap->sge.qs[0].rspq.polling),
1279 				  (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
1280 				  adap->name, adap);
1281 		if (err)
1282 			goto irq_err;
1283 	}
1284 
1285 	enable_all_napi(adap);
1286 	t3_sge_start(adap);
1287 	t3_intr_enable(adap);
1288 
1289 	if (adap->params.rev >= T3_REV_C && !(adap->flags & TP_PARITY_INIT) &&
1290 	    is_offload(adap) && init_tp_parity(adap) == 0)
1291 		adap->flags |= TP_PARITY_INIT;
1292 
1293 	if (adap->flags & TP_PARITY_INIT) {
1294 		t3_write_reg(adap, A_TP_INT_CAUSE,
1295 			     F_CMCACHEPERR | F_ARPLUTPERR);
1296 		t3_write_reg(adap, A_TP_INT_ENABLE, 0x7fbfffff);
1297 	}
1298 
1299 	if (!(adap->flags & QUEUES_BOUND)) {
1300 		int ret = bind_qsets(adap);
1301 
1302 		if (ret < 0) {
1303 			CH_ERR(adap, "failed to bind qsets, err %d\n", ret);
1304 			t3_intr_disable(adap);
1305 			free_irq_resources(adap);
1306 			err = ret;
1307 			goto out;
1308 		}
1309 		adap->flags |= QUEUES_BOUND;
1310 	}
1311 
1312 out:
1313 	return err;
1314 irq_err:
1315 	CH_ERR(adap, "request_irq failed, err %d\n", err);
1316 	goto out;
1317 }
1318 
1319 /*
1320  * Release resources when all the ports and offloading have been stopped.
1321  */
1322 static void cxgb_down(struct adapter *adapter, int on_wq)
1323 {
1324 	t3_sge_stop(adapter);
1325 	spin_lock_irq(&adapter->work_lock);	/* sync with PHY intr task */
1326 	t3_intr_disable(adapter);
1327 	spin_unlock_irq(&adapter->work_lock);
1328 
1329 	free_irq_resources(adapter);
1330 	quiesce_rx(adapter);
1331 	t3_sge_stop(adapter);
1332 	if (!on_wq)
1333 		flush_workqueue(cxgb3_wq);/* wait for external IRQ handler */
1334 }
1335 
1336 static void schedule_chk_task(struct adapter *adap)
1337 {
1338 	unsigned int timeo;
1339 
1340 	timeo = adap->params.linkpoll_period ?
1341 	    (HZ * adap->params.linkpoll_period) / 10 :
1342 	    adap->params.stats_update_period * HZ;
1343 	if (timeo)
1344 		queue_delayed_work(cxgb3_wq, &adap->adap_check_task, timeo);
1345 }
1346 
1347 static int offload_open(struct net_device *dev)
1348 {
1349 	struct port_info *pi = netdev_priv(dev);
1350 	struct adapter *adapter = pi->adapter;
1351 	struct t3cdev *tdev = dev2t3cdev(dev);
1352 	int adap_up = adapter->open_device_map & PORT_MASK;
1353 	int err;
1354 
1355 	if (test_and_set_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
1356 		return 0;
1357 
1358 	if (!adap_up && (err = cxgb_up(adapter)) < 0)
1359 		goto out;
1360 
1361 	t3_tp_set_offload_mode(adapter, 1);
1362 	tdev->lldev = adapter->port[0];
1363 	err = cxgb3_offload_activate(adapter);
1364 	if (err)
1365 		goto out;
1366 
1367 	init_port_mtus(adapter);
1368 	t3_load_mtus(adapter, adapter->params.mtus, adapter->params.a_wnd,
1369 		     adapter->params.b_wnd,
1370 		     adapter->params.rev == 0 ?
1371 		     adapter->port[0]->mtu : 0xffff);
1372 	init_smt(adapter);
1373 
1374 	if (sysfs_create_group(&tdev->lldev->dev.kobj, &offload_attr_group))
1375 		dev_dbg(&dev->dev, "cannot create sysfs group\n");
1376 
1377 	/* Call back all registered clients */
1378 	cxgb3_add_clients(tdev);
1379 
1380 out:
1381 	/* restore them in case the offload module has changed them */
1382 	if (err) {
1383 		t3_tp_set_offload_mode(adapter, 0);
1384 		clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
1385 		cxgb3_set_dummy_ops(tdev);
1386 	}
1387 	return err;
1388 }
1389 
1390 static int offload_close(struct t3cdev *tdev)
1391 {
1392 	struct adapter *adapter = tdev2adap(tdev);
1393 	struct t3c_data *td = T3C_DATA(tdev);
1394 
1395 	if (!test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
1396 		return 0;
1397 
1398 	/* Call back all registered clients */
1399 	cxgb3_remove_clients(tdev);
1400 
1401 	sysfs_remove_group(&tdev->lldev->dev.kobj, &offload_attr_group);
1402 
1403 	/* Flush work scheduled while releasing TIDs */
1404 	flush_work(&td->tid_release_task);
1405 
1406 	tdev->lldev = NULL;
1407 	cxgb3_set_dummy_ops(tdev);
1408 	t3_tp_set_offload_mode(adapter, 0);
1409 	clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
1410 
1411 	if (!adapter->open_device_map)
1412 		cxgb_down(adapter, 0);
1413 
1414 	cxgb3_offload_deactivate(adapter);
1415 	return 0;
1416 }
1417 
1418 static int cxgb_open(struct net_device *dev)
1419 {
1420 	struct port_info *pi = netdev_priv(dev);
1421 	struct adapter *adapter = pi->adapter;
1422 	int other_ports = adapter->open_device_map & PORT_MASK;
1423 	int err;
1424 
1425 	if (!adapter->open_device_map && (err = cxgb_up(adapter)) < 0)
1426 		return err;
1427 
1428 	set_bit(pi->port_id, &adapter->open_device_map);
1429 	if (is_offload(adapter) && !ofld_disable) {
1430 		err = offload_open(dev);
1431 		if (err)
1432 			pr_warn("Could not initialize offload capabilities\n");
1433 	}
1434 
1435 	netif_set_real_num_tx_queues(dev, pi->nqsets);
1436 	err = netif_set_real_num_rx_queues(dev, pi->nqsets);
1437 	if (err)
1438 		return err;
1439 	link_start(dev);
1440 	t3_port_intr_enable(adapter, pi->port_id);
1441 	netif_tx_start_all_queues(dev);
1442 	if (!other_ports)
1443 		schedule_chk_task(adapter);
1444 
1445 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_PORT_UP, pi->port_id);
1446 	return 0;
1447 }
1448 
1449 static int __cxgb_close(struct net_device *dev, int on_wq)
1450 {
1451 	struct port_info *pi = netdev_priv(dev);
1452 	struct adapter *adapter = pi->adapter;
1453 
1454 
1455 	if (!adapter->open_device_map)
1456 		return 0;
1457 
1458 	/* Stop link fault interrupts */
1459 	t3_xgm_intr_disable(adapter, pi->port_id);
1460 	t3_read_reg(adapter, A_XGM_INT_STATUS + pi->mac.offset);
1461 
1462 	t3_port_intr_disable(adapter, pi->port_id);
1463 	netif_tx_stop_all_queues(dev);
1464 	pi->phy.ops->power_down(&pi->phy, 1);
1465 	netif_carrier_off(dev);
1466 	t3_mac_disable(&pi->mac, MAC_DIRECTION_TX | MAC_DIRECTION_RX);
1467 
1468 	spin_lock_irq(&adapter->work_lock);	/* sync with update task */
1469 	clear_bit(pi->port_id, &adapter->open_device_map);
1470 	spin_unlock_irq(&adapter->work_lock);
1471 
1472 	if (!(adapter->open_device_map & PORT_MASK))
1473 		cancel_delayed_work_sync(&adapter->adap_check_task);
1474 
1475 	if (!adapter->open_device_map)
1476 		cxgb_down(adapter, on_wq);
1477 
1478 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_PORT_DOWN, pi->port_id);
1479 	return 0;
1480 }
1481 
1482 static int cxgb_close(struct net_device *dev)
1483 {
1484 	return __cxgb_close(dev, 0);
1485 }
1486 
1487 static struct net_device_stats *cxgb_get_stats(struct net_device *dev)
1488 {
1489 	struct port_info *pi = netdev_priv(dev);
1490 	struct adapter *adapter = pi->adapter;
1491 	struct net_device_stats *ns = &dev->stats;
1492 	const struct mac_stats *pstats;
1493 
1494 	spin_lock(&adapter->stats_lock);
1495 	pstats = t3_mac_update_stats(&pi->mac);
1496 	spin_unlock(&adapter->stats_lock);
1497 
1498 	ns->tx_bytes = pstats->tx_octets;
1499 	ns->tx_packets = pstats->tx_frames;
1500 	ns->rx_bytes = pstats->rx_octets;
1501 	ns->rx_packets = pstats->rx_frames;
1502 	ns->multicast = pstats->rx_mcast_frames;
1503 
1504 	ns->tx_errors = pstats->tx_underrun;
1505 	ns->rx_errors = pstats->rx_symbol_errs + pstats->rx_fcs_errs +
1506 	    pstats->rx_too_long + pstats->rx_jabber + pstats->rx_short +
1507 	    pstats->rx_fifo_ovfl;
1508 
1509 	/* detailed rx_errors */
1510 	ns->rx_length_errors = pstats->rx_jabber + pstats->rx_too_long;
1511 	ns->rx_over_errors = 0;
1512 	ns->rx_crc_errors = pstats->rx_fcs_errs;
1513 	ns->rx_frame_errors = pstats->rx_symbol_errs;
1514 	ns->rx_fifo_errors = pstats->rx_fifo_ovfl;
1515 	ns->rx_missed_errors = pstats->rx_cong_drops;
1516 
1517 	/* detailed tx_errors */
1518 	ns->tx_aborted_errors = 0;
1519 	ns->tx_carrier_errors = 0;
1520 	ns->tx_fifo_errors = pstats->tx_underrun;
1521 	ns->tx_heartbeat_errors = 0;
1522 	ns->tx_window_errors = 0;
1523 	return ns;
1524 }
1525 
1526 static u32 get_msglevel(struct net_device *dev)
1527 {
1528 	struct port_info *pi = netdev_priv(dev);
1529 	struct adapter *adapter = pi->adapter;
1530 
1531 	return adapter->msg_enable;
1532 }
1533 
1534 static void set_msglevel(struct net_device *dev, u32 val)
1535 {
1536 	struct port_info *pi = netdev_priv(dev);
1537 	struct adapter *adapter = pi->adapter;
1538 
1539 	adapter->msg_enable = val;
1540 }
1541 
1542 static const char stats_strings[][ETH_GSTRING_LEN] = {
1543 	"TxOctetsOK         ",
1544 	"TxFramesOK         ",
1545 	"TxMulticastFramesOK",
1546 	"TxBroadcastFramesOK",
1547 	"TxPauseFrames      ",
1548 	"TxUnderrun         ",
1549 	"TxExtUnderrun      ",
1550 
1551 	"TxFrames64         ",
1552 	"TxFrames65To127    ",
1553 	"TxFrames128To255   ",
1554 	"TxFrames256To511   ",
1555 	"TxFrames512To1023  ",
1556 	"TxFrames1024To1518 ",
1557 	"TxFrames1519ToMax  ",
1558 
1559 	"RxOctetsOK         ",
1560 	"RxFramesOK         ",
1561 	"RxMulticastFramesOK",
1562 	"RxBroadcastFramesOK",
1563 	"RxPauseFrames      ",
1564 	"RxFCSErrors        ",
1565 	"RxSymbolErrors     ",
1566 	"RxShortErrors      ",
1567 	"RxJabberErrors     ",
1568 	"RxLengthErrors     ",
1569 	"RxFIFOoverflow     ",
1570 
1571 	"RxFrames64         ",
1572 	"RxFrames65To127    ",
1573 	"RxFrames128To255   ",
1574 	"RxFrames256To511   ",
1575 	"RxFrames512To1023  ",
1576 	"RxFrames1024To1518 ",
1577 	"RxFrames1519ToMax  ",
1578 
1579 	"PhyFIFOErrors      ",
1580 	"TSO                ",
1581 	"VLANextractions    ",
1582 	"VLANinsertions     ",
1583 	"TxCsumOffload      ",
1584 	"RxCsumGood         ",
1585 	"LroAggregated      ",
1586 	"LroFlushed         ",
1587 	"LroNoDesc          ",
1588 	"RxDrops            ",
1589 
1590 	"CheckTXEnToggled   ",
1591 	"CheckResets        ",
1592 
1593 	"LinkFaults         ",
1594 };
1595 
1596 static int get_sset_count(struct net_device *dev, int sset)
1597 {
1598 	switch (sset) {
1599 	case ETH_SS_STATS:
1600 		return ARRAY_SIZE(stats_strings);
1601 	default:
1602 		return -EOPNOTSUPP;
1603 	}
1604 }
1605 
1606 #define T3_REGMAP_SIZE (3 * 1024)
1607 
1608 static int get_regs_len(struct net_device *dev)
1609 {
1610 	return T3_REGMAP_SIZE;
1611 }
1612 
1613 static int get_eeprom_len(struct net_device *dev)
1614 {
1615 	return EEPROMSIZE;
1616 }
1617 
1618 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1619 {
1620 	struct port_info *pi = netdev_priv(dev);
1621 	struct adapter *adapter = pi->adapter;
1622 	u32 fw_vers = 0;
1623 	u32 tp_vers = 0;
1624 
1625 	spin_lock(&adapter->stats_lock);
1626 	t3_get_fw_version(adapter, &fw_vers);
1627 	t3_get_tp_version(adapter, &tp_vers);
1628 	spin_unlock(&adapter->stats_lock);
1629 
1630 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1631 	strlcpy(info->bus_info, pci_name(adapter->pdev),
1632 		sizeof(info->bus_info));
1633 	if (fw_vers)
1634 		snprintf(info->fw_version, sizeof(info->fw_version),
1635 			 "%s %u.%u.%u TP %u.%u.%u",
1636 			 G_FW_VERSION_TYPE(fw_vers) ? "T" : "N",
1637 			 G_FW_VERSION_MAJOR(fw_vers),
1638 			 G_FW_VERSION_MINOR(fw_vers),
1639 			 G_FW_VERSION_MICRO(fw_vers),
1640 			 G_TP_VERSION_MAJOR(tp_vers),
1641 			 G_TP_VERSION_MINOR(tp_vers),
1642 			 G_TP_VERSION_MICRO(tp_vers));
1643 }
1644 
1645 static void get_strings(struct net_device *dev, u32 stringset, u8 * data)
1646 {
1647 	if (stringset == ETH_SS_STATS)
1648 		memcpy(data, stats_strings, sizeof(stats_strings));
1649 }
1650 
1651 static unsigned long collect_sge_port_stats(struct adapter *adapter,
1652 					    struct port_info *p, int idx)
1653 {
1654 	int i;
1655 	unsigned long tot = 0;
1656 
1657 	for (i = p->first_qset; i < p->first_qset + p->nqsets; ++i)
1658 		tot += adapter->sge.qs[i].port_stats[idx];
1659 	return tot;
1660 }
1661 
1662 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1663 		      u64 *data)
1664 {
1665 	struct port_info *pi = netdev_priv(dev);
1666 	struct adapter *adapter = pi->adapter;
1667 	const struct mac_stats *s;
1668 
1669 	spin_lock(&adapter->stats_lock);
1670 	s = t3_mac_update_stats(&pi->mac);
1671 	spin_unlock(&adapter->stats_lock);
1672 
1673 	*data++ = s->tx_octets;
1674 	*data++ = s->tx_frames;
1675 	*data++ = s->tx_mcast_frames;
1676 	*data++ = s->tx_bcast_frames;
1677 	*data++ = s->tx_pause;
1678 	*data++ = s->tx_underrun;
1679 	*data++ = s->tx_fifo_urun;
1680 
1681 	*data++ = s->tx_frames_64;
1682 	*data++ = s->tx_frames_65_127;
1683 	*data++ = s->tx_frames_128_255;
1684 	*data++ = s->tx_frames_256_511;
1685 	*data++ = s->tx_frames_512_1023;
1686 	*data++ = s->tx_frames_1024_1518;
1687 	*data++ = s->tx_frames_1519_max;
1688 
1689 	*data++ = s->rx_octets;
1690 	*data++ = s->rx_frames;
1691 	*data++ = s->rx_mcast_frames;
1692 	*data++ = s->rx_bcast_frames;
1693 	*data++ = s->rx_pause;
1694 	*data++ = s->rx_fcs_errs;
1695 	*data++ = s->rx_symbol_errs;
1696 	*data++ = s->rx_short;
1697 	*data++ = s->rx_jabber;
1698 	*data++ = s->rx_too_long;
1699 	*data++ = s->rx_fifo_ovfl;
1700 
1701 	*data++ = s->rx_frames_64;
1702 	*data++ = s->rx_frames_65_127;
1703 	*data++ = s->rx_frames_128_255;
1704 	*data++ = s->rx_frames_256_511;
1705 	*data++ = s->rx_frames_512_1023;
1706 	*data++ = s->rx_frames_1024_1518;
1707 	*data++ = s->rx_frames_1519_max;
1708 
1709 	*data++ = pi->phy.fifo_errors;
1710 
1711 	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TSO);
1712 	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANEX);
1713 	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANINS);
1714 	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TX_CSUM);
1715 	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_RX_CSUM_GOOD);
1716 	*data++ = 0;
1717 	*data++ = 0;
1718 	*data++ = 0;
1719 	*data++ = s->rx_cong_drops;
1720 
1721 	*data++ = s->num_toggled;
1722 	*data++ = s->num_resets;
1723 
1724 	*data++ = s->link_faults;
1725 }
1726 
1727 static inline void reg_block_dump(struct adapter *ap, void *buf,
1728 				  unsigned int start, unsigned int end)
1729 {
1730 	u32 *p = buf + start;
1731 
1732 	for (; start <= end; start += sizeof(u32))
1733 		*p++ = t3_read_reg(ap, start);
1734 }
1735 
1736 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1737 		     void *buf)
1738 {
1739 	struct port_info *pi = netdev_priv(dev);
1740 	struct adapter *ap = pi->adapter;
1741 
1742 	/*
1743 	 * Version scheme:
1744 	 * bits 0..9: chip version
1745 	 * bits 10..15: chip revision
1746 	 * bit 31: set for PCIe cards
1747 	 */
1748 	regs->version = 3 | (ap->params.rev << 10) | (is_pcie(ap) << 31);
1749 
1750 	/*
1751 	 * We skip the MAC statistics registers because they are clear-on-read.
1752 	 * Also reading multi-register stats would need to synchronize with the
1753 	 * periodic mac stats accumulation.  Hard to justify the complexity.
1754 	 */
1755 	memset(buf, 0, T3_REGMAP_SIZE);
1756 	reg_block_dump(ap, buf, 0, A_SG_RSPQ_CREDIT_RETURN);
1757 	reg_block_dump(ap, buf, A_SG_HI_DRB_HI_THRSH, A_ULPRX_PBL_ULIMIT);
1758 	reg_block_dump(ap, buf, A_ULPTX_CONFIG, A_MPS_INT_CAUSE);
1759 	reg_block_dump(ap, buf, A_CPL_SWITCH_CNTRL, A_CPL_MAP_TBL_DATA);
1760 	reg_block_dump(ap, buf, A_SMB_GLOBAL_TIME_CFG, A_XGM_SERDES_STAT3);
1761 	reg_block_dump(ap, buf, A_XGM_SERDES_STATUS0,
1762 		       XGM_REG(A_XGM_SERDES_STAT3, 1));
1763 	reg_block_dump(ap, buf, XGM_REG(A_XGM_SERDES_STATUS0, 1),
1764 		       XGM_REG(A_XGM_RX_SPI4_SOP_EOP_CNT, 1));
1765 }
1766 
1767 static int restart_autoneg(struct net_device *dev)
1768 {
1769 	struct port_info *p = netdev_priv(dev);
1770 
1771 	if (!netif_running(dev))
1772 		return -EAGAIN;
1773 	if (p->link_config.autoneg != AUTONEG_ENABLE)
1774 		return -EINVAL;
1775 	p->phy.ops->autoneg_restart(&p->phy);
1776 	return 0;
1777 }
1778 
1779 static int set_phys_id(struct net_device *dev,
1780 		       enum ethtool_phys_id_state state)
1781 {
1782 	struct port_info *pi = netdev_priv(dev);
1783 	struct adapter *adapter = pi->adapter;
1784 
1785 	switch (state) {
1786 	case ETHTOOL_ID_ACTIVE:
1787 		return 1;	/* cycle on/off once per second */
1788 
1789 	case ETHTOOL_ID_OFF:
1790 		t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL, 0);
1791 		break;
1792 
1793 	case ETHTOOL_ID_ON:
1794 	case ETHTOOL_ID_INACTIVE:
1795 		t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
1796 			 F_GPIO0_OUT_VAL);
1797 	}
1798 
1799 	return 0;
1800 }
1801 
1802 static int get_link_ksettings(struct net_device *dev,
1803 			      struct ethtool_link_ksettings *cmd)
1804 {
1805 	struct port_info *p = netdev_priv(dev);
1806 	u32 supported;
1807 
1808 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
1809 						p->link_config.supported);
1810 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
1811 						p->link_config.advertising);
1812 
1813 	if (netif_carrier_ok(dev)) {
1814 		cmd->base.speed = p->link_config.speed;
1815 		cmd->base.duplex = p->link_config.duplex;
1816 	} else {
1817 		cmd->base.speed = SPEED_UNKNOWN;
1818 		cmd->base.duplex = DUPLEX_UNKNOWN;
1819 	}
1820 
1821 	ethtool_convert_link_mode_to_legacy_u32(&supported,
1822 						cmd->link_modes.supported);
1823 
1824 	cmd->base.port = (supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1825 	cmd->base.phy_address = p->phy.mdio.prtad;
1826 	cmd->base.autoneg = p->link_config.autoneg;
1827 	return 0;
1828 }
1829 
1830 static int speed_duplex_to_caps(int speed, int duplex)
1831 {
1832 	int cap = 0;
1833 
1834 	switch (speed) {
1835 	case SPEED_10:
1836 		if (duplex == DUPLEX_FULL)
1837 			cap = SUPPORTED_10baseT_Full;
1838 		else
1839 			cap = SUPPORTED_10baseT_Half;
1840 		break;
1841 	case SPEED_100:
1842 		if (duplex == DUPLEX_FULL)
1843 			cap = SUPPORTED_100baseT_Full;
1844 		else
1845 			cap = SUPPORTED_100baseT_Half;
1846 		break;
1847 	case SPEED_1000:
1848 		if (duplex == DUPLEX_FULL)
1849 			cap = SUPPORTED_1000baseT_Full;
1850 		else
1851 			cap = SUPPORTED_1000baseT_Half;
1852 		break;
1853 	case SPEED_10000:
1854 		if (duplex == DUPLEX_FULL)
1855 			cap = SUPPORTED_10000baseT_Full;
1856 	}
1857 	return cap;
1858 }
1859 
1860 #define ADVERTISED_MASK (ADVERTISED_10baseT_Half | ADVERTISED_10baseT_Full | \
1861 		      ADVERTISED_100baseT_Half | ADVERTISED_100baseT_Full | \
1862 		      ADVERTISED_1000baseT_Half | ADVERTISED_1000baseT_Full | \
1863 		      ADVERTISED_10000baseT_Full)
1864 
1865 static int set_link_ksettings(struct net_device *dev,
1866 			      const struct ethtool_link_ksettings *cmd)
1867 {
1868 	struct port_info *p = netdev_priv(dev);
1869 	struct link_config *lc = &p->link_config;
1870 	u32 advertising;
1871 
1872 	ethtool_convert_link_mode_to_legacy_u32(&advertising,
1873 						cmd->link_modes.advertising);
1874 
1875 	if (!(lc->supported & SUPPORTED_Autoneg)) {
1876 		/*
1877 		 * PHY offers a single speed/duplex.  See if that's what's
1878 		 * being requested.
1879 		 */
1880 		if (cmd->base.autoneg == AUTONEG_DISABLE) {
1881 			u32 speed = cmd->base.speed;
1882 			int cap = speed_duplex_to_caps(speed, cmd->base.duplex);
1883 			if (lc->supported & cap)
1884 				return 0;
1885 		}
1886 		return -EINVAL;
1887 	}
1888 
1889 	if (cmd->base.autoneg == AUTONEG_DISABLE) {
1890 		u32 speed = cmd->base.speed;
1891 		int cap = speed_duplex_to_caps(speed, cmd->base.duplex);
1892 
1893 		if (!(lc->supported & cap) || (speed == SPEED_1000))
1894 			return -EINVAL;
1895 		lc->requested_speed = speed;
1896 		lc->requested_duplex = cmd->base.duplex;
1897 		lc->advertising = 0;
1898 	} else {
1899 		advertising &= ADVERTISED_MASK;
1900 		advertising &= lc->supported;
1901 		if (!advertising)
1902 			return -EINVAL;
1903 		lc->requested_speed = SPEED_INVALID;
1904 		lc->requested_duplex = DUPLEX_INVALID;
1905 		lc->advertising = advertising | ADVERTISED_Autoneg;
1906 	}
1907 	lc->autoneg = cmd->base.autoneg;
1908 	if (netif_running(dev))
1909 		t3_link_start(&p->phy, &p->mac, lc);
1910 	return 0;
1911 }
1912 
1913 static void get_pauseparam(struct net_device *dev,
1914 			   struct ethtool_pauseparam *epause)
1915 {
1916 	struct port_info *p = netdev_priv(dev);
1917 
1918 	epause->autoneg = (p->link_config.requested_fc & PAUSE_AUTONEG) != 0;
1919 	epause->rx_pause = (p->link_config.fc & PAUSE_RX) != 0;
1920 	epause->tx_pause = (p->link_config.fc & PAUSE_TX) != 0;
1921 }
1922 
1923 static int set_pauseparam(struct net_device *dev,
1924 			  struct ethtool_pauseparam *epause)
1925 {
1926 	struct port_info *p = netdev_priv(dev);
1927 	struct link_config *lc = &p->link_config;
1928 
1929 	if (epause->autoneg == AUTONEG_DISABLE)
1930 		lc->requested_fc = 0;
1931 	else if (lc->supported & SUPPORTED_Autoneg)
1932 		lc->requested_fc = PAUSE_AUTONEG;
1933 	else
1934 		return -EINVAL;
1935 
1936 	if (epause->rx_pause)
1937 		lc->requested_fc |= PAUSE_RX;
1938 	if (epause->tx_pause)
1939 		lc->requested_fc |= PAUSE_TX;
1940 	if (lc->autoneg == AUTONEG_ENABLE) {
1941 		if (netif_running(dev))
1942 			t3_link_start(&p->phy, &p->mac, lc);
1943 	} else {
1944 		lc->fc = lc->requested_fc & (PAUSE_RX | PAUSE_TX);
1945 		if (netif_running(dev))
1946 			t3_mac_set_speed_duplex_fc(&p->mac, -1, -1, lc->fc);
1947 	}
1948 	return 0;
1949 }
1950 
1951 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1952 {
1953 	struct port_info *pi = netdev_priv(dev);
1954 	struct adapter *adapter = pi->adapter;
1955 	const struct qset_params *q = &adapter->params.sge.qset[pi->first_qset];
1956 
1957 	e->rx_max_pending = MAX_RX_BUFFERS;
1958 	e->rx_jumbo_max_pending = MAX_RX_JUMBO_BUFFERS;
1959 	e->tx_max_pending = MAX_TXQ_ENTRIES;
1960 
1961 	e->rx_pending = q->fl_size;
1962 	e->rx_mini_pending = q->rspq_size;
1963 	e->rx_jumbo_pending = q->jumbo_size;
1964 	e->tx_pending = q->txq_size[0];
1965 }
1966 
1967 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1968 {
1969 	struct port_info *pi = netdev_priv(dev);
1970 	struct adapter *adapter = pi->adapter;
1971 	struct qset_params *q;
1972 	int i;
1973 
1974 	if (e->rx_pending > MAX_RX_BUFFERS ||
1975 	    e->rx_jumbo_pending > MAX_RX_JUMBO_BUFFERS ||
1976 	    e->tx_pending > MAX_TXQ_ENTRIES ||
1977 	    e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1978 	    e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1979 	    e->rx_pending < MIN_FL_ENTRIES ||
1980 	    e->rx_jumbo_pending < MIN_FL_ENTRIES ||
1981 	    e->tx_pending < adapter->params.nports * MIN_TXQ_ENTRIES)
1982 		return -EINVAL;
1983 
1984 	if (adapter->flags & FULL_INIT_DONE)
1985 		return -EBUSY;
1986 
1987 	q = &adapter->params.sge.qset[pi->first_qset];
1988 	for (i = 0; i < pi->nqsets; ++i, ++q) {
1989 		q->rspq_size = e->rx_mini_pending;
1990 		q->fl_size = e->rx_pending;
1991 		q->jumbo_size = e->rx_jumbo_pending;
1992 		q->txq_size[0] = e->tx_pending;
1993 		q->txq_size[1] = e->tx_pending;
1994 		q->txq_size[2] = e->tx_pending;
1995 	}
1996 	return 0;
1997 }
1998 
1999 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2000 {
2001 	struct port_info *pi = netdev_priv(dev);
2002 	struct adapter *adapter = pi->adapter;
2003 	struct qset_params *qsp;
2004 	struct sge_qset *qs;
2005 	int i;
2006 
2007 	if (c->rx_coalesce_usecs * 10 > M_NEWTIMER)
2008 		return -EINVAL;
2009 
2010 	for (i = 0; i < pi->nqsets; i++) {
2011 		qsp = &adapter->params.sge.qset[i];
2012 		qs = &adapter->sge.qs[i];
2013 		qsp->coalesce_usecs = c->rx_coalesce_usecs;
2014 		t3_update_qset_coalesce(qs, qsp);
2015 	}
2016 
2017 	return 0;
2018 }
2019 
2020 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2021 {
2022 	struct port_info *pi = netdev_priv(dev);
2023 	struct adapter *adapter = pi->adapter;
2024 	struct qset_params *q = adapter->params.sge.qset;
2025 
2026 	c->rx_coalesce_usecs = q->coalesce_usecs;
2027 	return 0;
2028 }
2029 
2030 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
2031 		      u8 * data)
2032 {
2033 	struct port_info *pi = netdev_priv(dev);
2034 	struct adapter *adapter = pi->adapter;
2035 	int i, err = 0;
2036 
2037 	u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
2038 	if (!buf)
2039 		return -ENOMEM;
2040 
2041 	e->magic = EEPROM_MAGIC;
2042 	for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
2043 		err = t3_seeprom_read(adapter, i, (__le32 *) & buf[i]);
2044 
2045 	if (!err)
2046 		memcpy(data, buf + e->offset, e->len);
2047 	kfree(buf);
2048 	return err;
2049 }
2050 
2051 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
2052 		      u8 * data)
2053 {
2054 	struct port_info *pi = netdev_priv(dev);
2055 	struct adapter *adapter = pi->adapter;
2056 	u32 aligned_offset, aligned_len;
2057 	__le32 *p;
2058 	u8 *buf;
2059 	int err;
2060 
2061 	if (eeprom->magic != EEPROM_MAGIC)
2062 		return -EINVAL;
2063 
2064 	aligned_offset = eeprom->offset & ~3;
2065 	aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
2066 
2067 	if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
2068 		buf = kmalloc(aligned_len, GFP_KERNEL);
2069 		if (!buf)
2070 			return -ENOMEM;
2071 		err = t3_seeprom_read(adapter, aligned_offset, (__le32 *) buf);
2072 		if (!err && aligned_len > 4)
2073 			err = t3_seeprom_read(adapter,
2074 					      aligned_offset + aligned_len - 4,
2075 					      (__le32 *) & buf[aligned_len - 4]);
2076 		if (err)
2077 			goto out;
2078 		memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
2079 	} else
2080 		buf = data;
2081 
2082 	err = t3_seeprom_wp(adapter, 0);
2083 	if (err)
2084 		goto out;
2085 
2086 	for (p = (__le32 *) buf; !err && aligned_len; aligned_len -= 4, p++) {
2087 		err = t3_seeprom_write(adapter, aligned_offset, *p);
2088 		aligned_offset += 4;
2089 	}
2090 
2091 	if (!err)
2092 		err = t3_seeprom_wp(adapter, 1);
2093 out:
2094 	if (buf != data)
2095 		kfree(buf);
2096 	return err;
2097 }
2098 
2099 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2100 {
2101 	wol->supported = 0;
2102 	wol->wolopts = 0;
2103 	memset(&wol->sopass, 0, sizeof(wol->sopass));
2104 }
2105 
2106 static const struct ethtool_ops cxgb_ethtool_ops = {
2107 	.supported_coalesce_params = ETHTOOL_COALESCE_RX_USECS,
2108 	.get_drvinfo = get_drvinfo,
2109 	.get_msglevel = get_msglevel,
2110 	.set_msglevel = set_msglevel,
2111 	.get_ringparam = get_sge_param,
2112 	.set_ringparam = set_sge_param,
2113 	.get_coalesce = get_coalesce,
2114 	.set_coalesce = set_coalesce,
2115 	.get_eeprom_len = get_eeprom_len,
2116 	.get_eeprom = get_eeprom,
2117 	.set_eeprom = set_eeprom,
2118 	.get_pauseparam = get_pauseparam,
2119 	.set_pauseparam = set_pauseparam,
2120 	.get_link = ethtool_op_get_link,
2121 	.get_strings = get_strings,
2122 	.set_phys_id = set_phys_id,
2123 	.nway_reset = restart_autoneg,
2124 	.get_sset_count = get_sset_count,
2125 	.get_ethtool_stats = get_stats,
2126 	.get_regs_len = get_regs_len,
2127 	.get_regs = get_regs,
2128 	.get_wol = get_wol,
2129 	.get_link_ksettings = get_link_ksettings,
2130 	.set_link_ksettings = set_link_ksettings,
2131 };
2132 
2133 static int in_range(int val, int lo, int hi)
2134 {
2135 	return val < 0 || (val <= hi && val >= lo);
2136 }
2137 
2138 static int cxgb_extension_ioctl(struct net_device *dev, void __user *useraddr)
2139 {
2140 	struct port_info *pi = netdev_priv(dev);
2141 	struct adapter *adapter = pi->adapter;
2142 	u32 cmd;
2143 	int ret;
2144 
2145 	if (copy_from_user(&cmd, useraddr, sizeof(cmd)))
2146 		return -EFAULT;
2147 
2148 	switch (cmd) {
2149 	case CHELSIO_SET_QSET_PARAMS:{
2150 		int i;
2151 		struct qset_params *q;
2152 		struct ch_qset_params t;
2153 		int q1 = pi->first_qset;
2154 		int nqsets = pi->nqsets;
2155 
2156 		if (!capable(CAP_NET_ADMIN))
2157 			return -EPERM;
2158 		if (copy_from_user(&t, useraddr, sizeof(t)))
2159 			return -EFAULT;
2160 		if (t.cmd != CHELSIO_SET_QSET_PARAMS)
2161 			return -EINVAL;
2162 		if (t.qset_idx >= SGE_QSETS)
2163 			return -EINVAL;
2164 		if (!in_range(t.intr_lat, 0, M_NEWTIMER) ||
2165 		    !in_range(t.cong_thres, 0, 255) ||
2166 		    !in_range(t.txq_size[0], MIN_TXQ_ENTRIES,
2167 			      MAX_TXQ_ENTRIES) ||
2168 		    !in_range(t.txq_size[1], MIN_TXQ_ENTRIES,
2169 			      MAX_TXQ_ENTRIES) ||
2170 		    !in_range(t.txq_size[2], MIN_CTRL_TXQ_ENTRIES,
2171 			      MAX_CTRL_TXQ_ENTRIES) ||
2172 		    !in_range(t.fl_size[0], MIN_FL_ENTRIES,
2173 			      MAX_RX_BUFFERS) ||
2174 		    !in_range(t.fl_size[1], MIN_FL_ENTRIES,
2175 			      MAX_RX_JUMBO_BUFFERS) ||
2176 		    !in_range(t.rspq_size, MIN_RSPQ_ENTRIES,
2177 			      MAX_RSPQ_ENTRIES))
2178 			return -EINVAL;
2179 
2180 		if ((adapter->flags & FULL_INIT_DONE) &&
2181 			(t.rspq_size >= 0 || t.fl_size[0] >= 0 ||
2182 			t.fl_size[1] >= 0 || t.txq_size[0] >= 0 ||
2183 			t.txq_size[1] >= 0 || t.txq_size[2] >= 0 ||
2184 			t.polling >= 0 || t.cong_thres >= 0))
2185 			return -EBUSY;
2186 
2187 		/* Allow setting of any available qset when offload enabled */
2188 		if (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2189 			q1 = 0;
2190 			for_each_port(adapter, i) {
2191 				pi = adap2pinfo(adapter, i);
2192 				nqsets += pi->first_qset + pi->nqsets;
2193 			}
2194 		}
2195 
2196 		if (t.qset_idx < q1)
2197 			return -EINVAL;
2198 		if (t.qset_idx > q1 + nqsets - 1)
2199 			return -EINVAL;
2200 
2201 		q = &adapter->params.sge.qset[t.qset_idx];
2202 
2203 		if (t.rspq_size >= 0)
2204 			q->rspq_size = t.rspq_size;
2205 		if (t.fl_size[0] >= 0)
2206 			q->fl_size = t.fl_size[0];
2207 		if (t.fl_size[1] >= 0)
2208 			q->jumbo_size = t.fl_size[1];
2209 		if (t.txq_size[0] >= 0)
2210 			q->txq_size[0] = t.txq_size[0];
2211 		if (t.txq_size[1] >= 0)
2212 			q->txq_size[1] = t.txq_size[1];
2213 		if (t.txq_size[2] >= 0)
2214 			q->txq_size[2] = t.txq_size[2];
2215 		if (t.cong_thres >= 0)
2216 			q->cong_thres = t.cong_thres;
2217 		if (t.intr_lat >= 0) {
2218 			struct sge_qset *qs =
2219 				&adapter->sge.qs[t.qset_idx];
2220 
2221 			q->coalesce_usecs = t.intr_lat;
2222 			t3_update_qset_coalesce(qs, q);
2223 		}
2224 		if (t.polling >= 0) {
2225 			if (adapter->flags & USING_MSIX)
2226 				q->polling = t.polling;
2227 			else {
2228 				/* No polling with INTx for T3A */
2229 				if (adapter->params.rev == 0 &&
2230 					!(adapter->flags & USING_MSI))
2231 					t.polling = 0;
2232 
2233 				for (i = 0; i < SGE_QSETS; i++) {
2234 					q = &adapter->params.sge.
2235 						qset[i];
2236 					q->polling = t.polling;
2237 				}
2238 			}
2239 		}
2240 
2241 		if (t.lro >= 0) {
2242 			if (t.lro)
2243 				dev->wanted_features |= NETIF_F_GRO;
2244 			else
2245 				dev->wanted_features &= ~NETIF_F_GRO;
2246 			netdev_update_features(dev);
2247 		}
2248 
2249 		break;
2250 	}
2251 	case CHELSIO_GET_QSET_PARAMS:{
2252 		struct qset_params *q;
2253 		struct ch_qset_params t;
2254 		int q1 = pi->first_qset;
2255 		int nqsets = pi->nqsets;
2256 		int i;
2257 
2258 		if (copy_from_user(&t, useraddr, sizeof(t)))
2259 			return -EFAULT;
2260 
2261 		if (t.cmd != CHELSIO_GET_QSET_PARAMS)
2262 			return -EINVAL;
2263 
2264 		/* Display qsets for all ports when offload enabled */
2265 		if (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2266 			q1 = 0;
2267 			for_each_port(adapter, i) {
2268 				pi = adap2pinfo(adapter, i);
2269 				nqsets = pi->first_qset + pi->nqsets;
2270 			}
2271 		}
2272 
2273 		if (t.qset_idx >= nqsets)
2274 			return -EINVAL;
2275 		t.qset_idx = array_index_nospec(t.qset_idx, nqsets);
2276 
2277 		q = &adapter->params.sge.qset[q1 + t.qset_idx];
2278 		t.rspq_size = q->rspq_size;
2279 		t.txq_size[0] = q->txq_size[0];
2280 		t.txq_size[1] = q->txq_size[1];
2281 		t.txq_size[2] = q->txq_size[2];
2282 		t.fl_size[0] = q->fl_size;
2283 		t.fl_size[1] = q->jumbo_size;
2284 		t.polling = q->polling;
2285 		t.lro = !!(dev->features & NETIF_F_GRO);
2286 		t.intr_lat = q->coalesce_usecs;
2287 		t.cong_thres = q->cong_thres;
2288 		t.qnum = q1;
2289 
2290 		if (adapter->flags & USING_MSIX)
2291 			t.vector = adapter->msix_info[q1 + t.qset_idx + 1].vec;
2292 		else
2293 			t.vector = adapter->pdev->irq;
2294 
2295 		if (copy_to_user(useraddr, &t, sizeof(t)))
2296 			return -EFAULT;
2297 		break;
2298 	}
2299 	case CHELSIO_SET_QSET_NUM:{
2300 		struct ch_reg edata;
2301 		unsigned int i, first_qset = 0, other_qsets = 0;
2302 
2303 		if (!capable(CAP_NET_ADMIN))
2304 			return -EPERM;
2305 		if (adapter->flags & FULL_INIT_DONE)
2306 			return -EBUSY;
2307 		if (copy_from_user(&edata, useraddr, sizeof(edata)))
2308 			return -EFAULT;
2309 		if (edata.cmd != CHELSIO_SET_QSET_NUM)
2310 			return -EINVAL;
2311 		if (edata.val < 1 ||
2312 			(edata.val > 1 && !(adapter->flags & USING_MSIX)))
2313 			return -EINVAL;
2314 
2315 		for_each_port(adapter, i)
2316 			if (adapter->port[i] && adapter->port[i] != dev)
2317 				other_qsets += adap2pinfo(adapter, i)->nqsets;
2318 
2319 		if (edata.val + other_qsets > SGE_QSETS)
2320 			return -EINVAL;
2321 
2322 		pi->nqsets = edata.val;
2323 
2324 		for_each_port(adapter, i)
2325 			if (adapter->port[i]) {
2326 				pi = adap2pinfo(adapter, i);
2327 				pi->first_qset = first_qset;
2328 				first_qset += pi->nqsets;
2329 			}
2330 		break;
2331 	}
2332 	case CHELSIO_GET_QSET_NUM:{
2333 		struct ch_reg edata;
2334 
2335 		memset(&edata, 0, sizeof(struct ch_reg));
2336 
2337 		edata.cmd = CHELSIO_GET_QSET_NUM;
2338 		edata.val = pi->nqsets;
2339 		if (copy_to_user(useraddr, &edata, sizeof(edata)))
2340 			return -EFAULT;
2341 		break;
2342 	}
2343 	case CHELSIO_LOAD_FW:{
2344 		u8 *fw_data;
2345 		struct ch_mem_range t;
2346 
2347 		if (!capable(CAP_SYS_RAWIO))
2348 			return -EPERM;
2349 		if (copy_from_user(&t, useraddr, sizeof(t)))
2350 			return -EFAULT;
2351 		if (t.cmd != CHELSIO_LOAD_FW)
2352 			return -EINVAL;
2353 		/* Check t.len sanity ? */
2354 		fw_data = memdup_user(useraddr + sizeof(t), t.len);
2355 		if (IS_ERR(fw_data))
2356 			return PTR_ERR(fw_data);
2357 
2358 		ret = t3_load_fw(adapter, fw_data, t.len);
2359 		kfree(fw_data);
2360 		if (ret)
2361 			return ret;
2362 		break;
2363 	}
2364 	case CHELSIO_SETMTUTAB:{
2365 		struct ch_mtus m;
2366 		int i;
2367 
2368 		if (!is_offload(adapter))
2369 			return -EOPNOTSUPP;
2370 		if (!capable(CAP_NET_ADMIN))
2371 			return -EPERM;
2372 		if (offload_running(adapter))
2373 			return -EBUSY;
2374 		if (copy_from_user(&m, useraddr, sizeof(m)))
2375 			return -EFAULT;
2376 		if (m.cmd != CHELSIO_SETMTUTAB)
2377 			return -EINVAL;
2378 		if (m.nmtus != NMTUS)
2379 			return -EINVAL;
2380 		if (m.mtus[0] < 81)	/* accommodate SACK */
2381 			return -EINVAL;
2382 
2383 		/* MTUs must be in ascending order */
2384 		for (i = 1; i < NMTUS; ++i)
2385 			if (m.mtus[i] < m.mtus[i - 1])
2386 				return -EINVAL;
2387 
2388 		memcpy(adapter->params.mtus, m.mtus,
2389 			sizeof(adapter->params.mtus));
2390 		break;
2391 	}
2392 	case CHELSIO_GET_PM:{
2393 		struct tp_params *p = &adapter->params.tp;
2394 		struct ch_pm m = {.cmd = CHELSIO_GET_PM };
2395 
2396 		if (!is_offload(adapter))
2397 			return -EOPNOTSUPP;
2398 		m.tx_pg_sz = p->tx_pg_size;
2399 		m.tx_num_pg = p->tx_num_pgs;
2400 		m.rx_pg_sz = p->rx_pg_size;
2401 		m.rx_num_pg = p->rx_num_pgs;
2402 		m.pm_total = p->pmtx_size + p->chan_rx_size * p->nchan;
2403 		if (copy_to_user(useraddr, &m, sizeof(m)))
2404 			return -EFAULT;
2405 		break;
2406 	}
2407 	case CHELSIO_SET_PM:{
2408 		struct ch_pm m;
2409 		struct tp_params *p = &adapter->params.tp;
2410 
2411 		if (!is_offload(adapter))
2412 			return -EOPNOTSUPP;
2413 		if (!capable(CAP_NET_ADMIN))
2414 			return -EPERM;
2415 		if (adapter->flags & FULL_INIT_DONE)
2416 			return -EBUSY;
2417 		if (copy_from_user(&m, useraddr, sizeof(m)))
2418 			return -EFAULT;
2419 		if (m.cmd != CHELSIO_SET_PM)
2420 			return -EINVAL;
2421 		if (!is_power_of_2(m.rx_pg_sz) ||
2422 			!is_power_of_2(m.tx_pg_sz))
2423 			return -EINVAL;	/* not power of 2 */
2424 		if (!(m.rx_pg_sz & 0x14000))
2425 			return -EINVAL;	/* not 16KB or 64KB */
2426 		if (!(m.tx_pg_sz & 0x1554000))
2427 			return -EINVAL;
2428 		if (m.tx_num_pg == -1)
2429 			m.tx_num_pg = p->tx_num_pgs;
2430 		if (m.rx_num_pg == -1)
2431 			m.rx_num_pg = p->rx_num_pgs;
2432 		if (m.tx_num_pg % 24 || m.rx_num_pg % 24)
2433 			return -EINVAL;
2434 		if (m.rx_num_pg * m.rx_pg_sz > p->chan_rx_size ||
2435 			m.tx_num_pg * m.tx_pg_sz > p->chan_tx_size)
2436 			return -EINVAL;
2437 		p->rx_pg_size = m.rx_pg_sz;
2438 		p->tx_pg_size = m.tx_pg_sz;
2439 		p->rx_num_pgs = m.rx_num_pg;
2440 		p->tx_num_pgs = m.tx_num_pg;
2441 		break;
2442 	}
2443 	case CHELSIO_GET_MEM:{
2444 		struct ch_mem_range t;
2445 		struct mc7 *mem;
2446 		u64 buf[32];
2447 
2448 		if (!is_offload(adapter))
2449 			return -EOPNOTSUPP;
2450 		if (!capable(CAP_NET_ADMIN))
2451 			return -EPERM;
2452 		if (!(adapter->flags & FULL_INIT_DONE))
2453 			return -EIO;	/* need the memory controllers */
2454 		if (copy_from_user(&t, useraddr, sizeof(t)))
2455 			return -EFAULT;
2456 		if (t.cmd != CHELSIO_GET_MEM)
2457 			return -EINVAL;
2458 		if ((t.addr & 7) || (t.len & 7))
2459 			return -EINVAL;
2460 		if (t.mem_id == MEM_CM)
2461 			mem = &adapter->cm;
2462 		else if (t.mem_id == MEM_PMRX)
2463 			mem = &adapter->pmrx;
2464 		else if (t.mem_id == MEM_PMTX)
2465 			mem = &adapter->pmtx;
2466 		else
2467 			return -EINVAL;
2468 
2469 		/*
2470 		 * Version scheme:
2471 		 * bits 0..9: chip version
2472 		 * bits 10..15: chip revision
2473 		 */
2474 		t.version = 3 | (adapter->params.rev << 10);
2475 		if (copy_to_user(useraddr, &t, sizeof(t)))
2476 			return -EFAULT;
2477 
2478 		/*
2479 		 * Read 256 bytes at a time as len can be large and we don't
2480 		 * want to use huge intermediate buffers.
2481 		 */
2482 		useraddr += sizeof(t);	/* advance to start of buffer */
2483 		while (t.len) {
2484 			unsigned int chunk =
2485 				min_t(unsigned int, t.len, sizeof(buf));
2486 
2487 			ret =
2488 				t3_mc7_bd_read(mem, t.addr / 8, chunk / 8,
2489 						buf);
2490 			if (ret)
2491 				return ret;
2492 			if (copy_to_user(useraddr, buf, chunk))
2493 				return -EFAULT;
2494 			useraddr += chunk;
2495 			t.addr += chunk;
2496 			t.len -= chunk;
2497 		}
2498 		break;
2499 	}
2500 	case CHELSIO_SET_TRACE_FILTER:{
2501 		struct ch_trace t;
2502 		const struct trace_params *tp;
2503 
2504 		if (!capable(CAP_NET_ADMIN))
2505 			return -EPERM;
2506 		if (!offload_running(adapter))
2507 			return -EAGAIN;
2508 		if (copy_from_user(&t, useraddr, sizeof(t)))
2509 			return -EFAULT;
2510 		if (t.cmd != CHELSIO_SET_TRACE_FILTER)
2511 			return -EINVAL;
2512 
2513 		tp = (const struct trace_params *)&t.sip;
2514 		if (t.config_tx)
2515 			t3_config_trace_filter(adapter, tp, 0,
2516 						t.invert_match,
2517 						t.trace_tx);
2518 		if (t.config_rx)
2519 			t3_config_trace_filter(adapter, tp, 1,
2520 						t.invert_match,
2521 						t.trace_rx);
2522 		break;
2523 	}
2524 	default:
2525 		return -EOPNOTSUPP;
2526 	}
2527 	return 0;
2528 }
2529 
2530 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2531 {
2532 	struct mii_ioctl_data *data = if_mii(req);
2533 	struct port_info *pi = netdev_priv(dev);
2534 	struct adapter *adapter = pi->adapter;
2535 
2536 	switch (cmd) {
2537 	case SIOCGMIIREG:
2538 	case SIOCSMIIREG:
2539 		/* Convert phy_id from older PRTAD/DEVAD format */
2540 		if (is_10G(adapter) &&
2541 		    !mdio_phy_id_is_c45(data->phy_id) &&
2542 		    (data->phy_id & 0x1f00) &&
2543 		    !(data->phy_id & 0xe0e0))
2544 			data->phy_id = mdio_phy_id_c45(data->phy_id >> 8,
2545 						       data->phy_id & 0x1f);
2546 		fallthrough;
2547 	case SIOCGMIIPHY:
2548 		return mdio_mii_ioctl(&pi->phy.mdio, data, cmd);
2549 	case SIOCCHIOCTL:
2550 		return cxgb_extension_ioctl(dev, req->ifr_data);
2551 	default:
2552 		return -EOPNOTSUPP;
2553 	}
2554 }
2555 
2556 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2557 {
2558 	struct port_info *pi = netdev_priv(dev);
2559 	struct adapter *adapter = pi->adapter;
2560 	int ret;
2561 
2562 	if ((ret = t3_mac_set_mtu(&pi->mac, new_mtu)))
2563 		return ret;
2564 	dev->mtu = new_mtu;
2565 	init_port_mtus(adapter);
2566 	if (adapter->params.rev == 0 && offload_running(adapter))
2567 		t3_load_mtus(adapter, adapter->params.mtus,
2568 			     adapter->params.a_wnd, adapter->params.b_wnd,
2569 			     adapter->port[0]->mtu);
2570 	return 0;
2571 }
2572 
2573 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2574 {
2575 	struct port_info *pi = netdev_priv(dev);
2576 	struct adapter *adapter = pi->adapter;
2577 	struct sockaddr *addr = p;
2578 
2579 	if (!is_valid_ether_addr(addr->sa_data))
2580 		return -EADDRNOTAVAIL;
2581 
2582 	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2583 	t3_mac_set_address(&pi->mac, LAN_MAC_IDX, dev->dev_addr);
2584 	if (offload_running(adapter))
2585 		write_smt_entry(adapter, pi->port_id);
2586 	return 0;
2587 }
2588 
2589 static netdev_features_t cxgb_fix_features(struct net_device *dev,
2590 	netdev_features_t features)
2591 {
2592 	/*
2593 	 * Since there is no support for separate rx/tx vlan accel
2594 	 * enable/disable make sure tx flag is always in same state as rx.
2595 	 */
2596 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
2597 		features |= NETIF_F_HW_VLAN_CTAG_TX;
2598 	else
2599 		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2600 
2601 	return features;
2602 }
2603 
2604 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
2605 {
2606 	netdev_features_t changed = dev->features ^ features;
2607 
2608 	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2609 		cxgb_vlan_mode(dev, features);
2610 
2611 	return 0;
2612 }
2613 
2614 #ifdef CONFIG_NET_POLL_CONTROLLER
2615 static void cxgb_netpoll(struct net_device *dev)
2616 {
2617 	struct port_info *pi = netdev_priv(dev);
2618 	struct adapter *adapter = pi->adapter;
2619 	int qidx;
2620 
2621 	for (qidx = pi->first_qset; qidx < pi->first_qset + pi->nqsets; qidx++) {
2622 		struct sge_qset *qs = &adapter->sge.qs[qidx];
2623 		void *source;
2624 
2625 		if (adapter->flags & USING_MSIX)
2626 			source = qs;
2627 		else
2628 			source = adapter;
2629 
2630 		t3_intr_handler(adapter, qs->rspq.polling) (0, source);
2631 	}
2632 }
2633 #endif
2634 
2635 /*
2636  * Periodic accumulation of MAC statistics.
2637  */
2638 static void mac_stats_update(struct adapter *adapter)
2639 {
2640 	int i;
2641 
2642 	for_each_port(adapter, i) {
2643 		struct net_device *dev = adapter->port[i];
2644 		struct port_info *p = netdev_priv(dev);
2645 
2646 		if (netif_running(dev)) {
2647 			spin_lock(&adapter->stats_lock);
2648 			t3_mac_update_stats(&p->mac);
2649 			spin_unlock(&adapter->stats_lock);
2650 		}
2651 	}
2652 }
2653 
2654 static void check_link_status(struct adapter *adapter)
2655 {
2656 	int i;
2657 
2658 	for_each_port(adapter, i) {
2659 		struct net_device *dev = adapter->port[i];
2660 		struct port_info *p = netdev_priv(dev);
2661 		int link_fault;
2662 
2663 		spin_lock_irq(&adapter->work_lock);
2664 		link_fault = p->link_fault;
2665 		spin_unlock_irq(&adapter->work_lock);
2666 
2667 		if (link_fault) {
2668 			t3_link_fault(adapter, i);
2669 			continue;
2670 		}
2671 
2672 		if (!(p->phy.caps & SUPPORTED_IRQ) && netif_running(dev)) {
2673 			t3_xgm_intr_disable(adapter, i);
2674 			t3_read_reg(adapter, A_XGM_INT_STATUS + p->mac.offset);
2675 
2676 			t3_link_changed(adapter, i);
2677 			t3_xgm_intr_enable(adapter, i);
2678 		}
2679 	}
2680 }
2681 
2682 static void check_t3b2_mac(struct adapter *adapter)
2683 {
2684 	int i;
2685 
2686 	if (!rtnl_trylock())	/* synchronize with ifdown */
2687 		return;
2688 
2689 	for_each_port(adapter, i) {
2690 		struct net_device *dev = adapter->port[i];
2691 		struct port_info *p = netdev_priv(dev);
2692 		int status;
2693 
2694 		if (!netif_running(dev))
2695 			continue;
2696 
2697 		status = 0;
2698 		if (netif_running(dev) && netif_carrier_ok(dev))
2699 			status = t3b2_mac_watchdog_task(&p->mac);
2700 		if (status == 1)
2701 			p->mac.stats.num_toggled++;
2702 		else if (status == 2) {
2703 			struct cmac *mac = &p->mac;
2704 
2705 			t3_mac_set_mtu(mac, dev->mtu);
2706 			t3_mac_set_address(mac, LAN_MAC_IDX, dev->dev_addr);
2707 			cxgb_set_rxmode(dev);
2708 			t3_link_start(&p->phy, mac, &p->link_config);
2709 			t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
2710 			t3_port_intr_enable(adapter, p->port_id);
2711 			p->mac.stats.num_resets++;
2712 		}
2713 	}
2714 	rtnl_unlock();
2715 }
2716 
2717 
2718 static void t3_adap_check_task(struct work_struct *work)
2719 {
2720 	struct adapter *adapter = container_of(work, struct adapter,
2721 					       adap_check_task.work);
2722 	const struct adapter_params *p = &adapter->params;
2723 	int port;
2724 	unsigned int v, status, reset;
2725 
2726 	adapter->check_task_cnt++;
2727 
2728 	check_link_status(adapter);
2729 
2730 	/* Accumulate MAC stats if needed */
2731 	if (!p->linkpoll_period ||
2732 	    (adapter->check_task_cnt * p->linkpoll_period) / 10 >=
2733 	    p->stats_update_period) {
2734 		mac_stats_update(adapter);
2735 		adapter->check_task_cnt = 0;
2736 	}
2737 
2738 	if (p->rev == T3_REV_B2)
2739 		check_t3b2_mac(adapter);
2740 
2741 	/*
2742 	 * Scan the XGMAC's to check for various conditions which we want to
2743 	 * monitor in a periodic polling manner rather than via an interrupt
2744 	 * condition.  This is used for conditions which would otherwise flood
2745 	 * the system with interrupts and we only really need to know that the
2746 	 * conditions are "happening" ...  For each condition we count the
2747 	 * detection of the condition and reset it for the next polling loop.
2748 	 */
2749 	for_each_port(adapter, port) {
2750 		struct cmac *mac =  &adap2pinfo(adapter, port)->mac;
2751 		u32 cause;
2752 
2753 		cause = t3_read_reg(adapter, A_XGM_INT_CAUSE + mac->offset);
2754 		reset = 0;
2755 		if (cause & F_RXFIFO_OVERFLOW) {
2756 			mac->stats.rx_fifo_ovfl++;
2757 			reset |= F_RXFIFO_OVERFLOW;
2758 		}
2759 
2760 		t3_write_reg(adapter, A_XGM_INT_CAUSE + mac->offset, reset);
2761 	}
2762 
2763 	/*
2764 	 * We do the same as above for FL_EMPTY interrupts.
2765 	 */
2766 	status = t3_read_reg(adapter, A_SG_INT_CAUSE);
2767 	reset = 0;
2768 
2769 	if (status & F_FLEMPTY) {
2770 		struct sge_qset *qs = &adapter->sge.qs[0];
2771 		int i = 0;
2772 
2773 		reset |= F_FLEMPTY;
2774 
2775 		v = (t3_read_reg(adapter, A_SG_RSPQ_FL_STATUS) >> S_FL0EMPTY) &
2776 		    0xffff;
2777 
2778 		while (v) {
2779 			qs->fl[i].empty += (v & 1);
2780 			if (i)
2781 				qs++;
2782 			i ^= 1;
2783 			v >>= 1;
2784 		}
2785 	}
2786 
2787 	t3_write_reg(adapter, A_SG_INT_CAUSE, reset);
2788 
2789 	/* Schedule the next check update if any port is active. */
2790 	spin_lock_irq(&adapter->work_lock);
2791 	if (adapter->open_device_map & PORT_MASK)
2792 		schedule_chk_task(adapter);
2793 	spin_unlock_irq(&adapter->work_lock);
2794 }
2795 
2796 static void db_full_task(struct work_struct *work)
2797 {
2798 	struct adapter *adapter = container_of(work, struct adapter,
2799 					       db_full_task);
2800 
2801 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_FULL, 0);
2802 }
2803 
2804 static void db_empty_task(struct work_struct *work)
2805 {
2806 	struct adapter *adapter = container_of(work, struct adapter,
2807 					       db_empty_task);
2808 
2809 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_EMPTY, 0);
2810 }
2811 
2812 static void db_drop_task(struct work_struct *work)
2813 {
2814 	struct adapter *adapter = container_of(work, struct adapter,
2815 					       db_drop_task);
2816 	unsigned long delay = 1000;
2817 	unsigned short r;
2818 
2819 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_DROP, 0);
2820 
2821 	/*
2822 	 * Sleep a while before ringing the driver qset dbs.
2823 	 * The delay is between 1000-2023 usecs.
2824 	 */
2825 	get_random_bytes(&r, 2);
2826 	delay += r & 1023;
2827 	set_current_state(TASK_UNINTERRUPTIBLE);
2828 	schedule_timeout(usecs_to_jiffies(delay));
2829 	ring_dbs(adapter);
2830 }
2831 
2832 /*
2833  * Processes external (PHY) interrupts in process context.
2834  */
2835 static void ext_intr_task(struct work_struct *work)
2836 {
2837 	struct adapter *adapter = container_of(work, struct adapter,
2838 					       ext_intr_handler_task);
2839 	int i;
2840 
2841 	/* Disable link fault interrupts */
2842 	for_each_port(adapter, i) {
2843 		struct net_device *dev = adapter->port[i];
2844 		struct port_info *p = netdev_priv(dev);
2845 
2846 		t3_xgm_intr_disable(adapter, i);
2847 		t3_read_reg(adapter, A_XGM_INT_STATUS + p->mac.offset);
2848 	}
2849 
2850 	/* Re-enable link fault interrupts */
2851 	t3_phy_intr_handler(adapter);
2852 
2853 	for_each_port(adapter, i)
2854 		t3_xgm_intr_enable(adapter, i);
2855 
2856 	/* Now reenable external interrupts */
2857 	spin_lock_irq(&adapter->work_lock);
2858 	if (adapter->slow_intr_mask) {
2859 		adapter->slow_intr_mask |= F_T3DBG;
2860 		t3_write_reg(adapter, A_PL_INT_CAUSE0, F_T3DBG);
2861 		t3_write_reg(adapter, A_PL_INT_ENABLE0,
2862 			     adapter->slow_intr_mask);
2863 	}
2864 	spin_unlock_irq(&adapter->work_lock);
2865 }
2866 
2867 /*
2868  * Interrupt-context handler for external (PHY) interrupts.
2869  */
2870 void t3_os_ext_intr_handler(struct adapter *adapter)
2871 {
2872 	/*
2873 	 * Schedule a task to handle external interrupts as they may be slow
2874 	 * and we use a mutex to protect MDIO registers.  We disable PHY
2875 	 * interrupts in the meantime and let the task reenable them when
2876 	 * it's done.
2877 	 */
2878 	spin_lock(&adapter->work_lock);
2879 	if (adapter->slow_intr_mask) {
2880 		adapter->slow_intr_mask &= ~F_T3DBG;
2881 		t3_write_reg(adapter, A_PL_INT_ENABLE0,
2882 			     adapter->slow_intr_mask);
2883 		queue_work(cxgb3_wq, &adapter->ext_intr_handler_task);
2884 	}
2885 	spin_unlock(&adapter->work_lock);
2886 }
2887 
2888 void t3_os_link_fault_handler(struct adapter *adapter, int port_id)
2889 {
2890 	struct net_device *netdev = adapter->port[port_id];
2891 	struct port_info *pi = netdev_priv(netdev);
2892 
2893 	spin_lock(&adapter->work_lock);
2894 	pi->link_fault = 1;
2895 	spin_unlock(&adapter->work_lock);
2896 }
2897 
2898 static int t3_adapter_error(struct adapter *adapter, int reset, int on_wq)
2899 {
2900 	int i, ret = 0;
2901 
2902 	if (is_offload(adapter) &&
2903 	    test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2904 		cxgb3_event_notify(&adapter->tdev, OFFLOAD_STATUS_DOWN, 0);
2905 		offload_close(&adapter->tdev);
2906 	}
2907 
2908 	/* Stop all ports */
2909 	for_each_port(adapter, i) {
2910 		struct net_device *netdev = adapter->port[i];
2911 
2912 		if (netif_running(netdev))
2913 			__cxgb_close(netdev, on_wq);
2914 	}
2915 
2916 	/* Stop SGE timers */
2917 	t3_stop_sge_timers(adapter);
2918 
2919 	adapter->flags &= ~FULL_INIT_DONE;
2920 
2921 	if (reset)
2922 		ret = t3_reset_adapter(adapter);
2923 
2924 	pci_disable_device(adapter->pdev);
2925 
2926 	return ret;
2927 }
2928 
2929 static int t3_reenable_adapter(struct adapter *adapter)
2930 {
2931 	if (pci_enable_device(adapter->pdev)) {
2932 		dev_err(&adapter->pdev->dev,
2933 			"Cannot re-enable PCI device after reset.\n");
2934 		goto err;
2935 	}
2936 	pci_set_master(adapter->pdev);
2937 	pci_restore_state(adapter->pdev);
2938 	pci_save_state(adapter->pdev);
2939 
2940 	/* Free sge resources */
2941 	t3_free_sge_resources(adapter);
2942 
2943 	if (t3_replay_prep_adapter(adapter))
2944 		goto err;
2945 
2946 	return 0;
2947 err:
2948 	return -1;
2949 }
2950 
2951 static void t3_resume_ports(struct adapter *adapter)
2952 {
2953 	int i;
2954 
2955 	/* Restart the ports */
2956 	for_each_port(adapter, i) {
2957 		struct net_device *netdev = adapter->port[i];
2958 
2959 		if (netif_running(netdev)) {
2960 			if (cxgb_open(netdev)) {
2961 				dev_err(&adapter->pdev->dev,
2962 					"can't bring device back up"
2963 					" after reset\n");
2964 				continue;
2965 			}
2966 		}
2967 	}
2968 
2969 	if (is_offload(adapter) && !ofld_disable)
2970 		cxgb3_event_notify(&adapter->tdev, OFFLOAD_STATUS_UP, 0);
2971 }
2972 
2973 /*
2974  * processes a fatal error.
2975  * Bring the ports down, reset the chip, bring the ports back up.
2976  */
2977 static void fatal_error_task(struct work_struct *work)
2978 {
2979 	struct adapter *adapter = container_of(work, struct adapter,
2980 					       fatal_error_handler_task);
2981 	int err = 0;
2982 
2983 	rtnl_lock();
2984 	err = t3_adapter_error(adapter, 1, 1);
2985 	if (!err)
2986 		err = t3_reenable_adapter(adapter);
2987 	if (!err)
2988 		t3_resume_ports(adapter);
2989 
2990 	CH_ALERT(adapter, "adapter reset %s\n", err ? "failed" : "succeeded");
2991 	rtnl_unlock();
2992 }
2993 
2994 void t3_fatal_err(struct adapter *adapter)
2995 {
2996 	unsigned int fw_status[4];
2997 
2998 	if (adapter->flags & FULL_INIT_DONE) {
2999 		t3_sge_stop_dma(adapter);
3000 		t3_write_reg(adapter, A_XGM_TX_CTRL, 0);
3001 		t3_write_reg(adapter, A_XGM_RX_CTRL, 0);
3002 		t3_write_reg(adapter, XGM_REG(A_XGM_TX_CTRL, 1), 0);
3003 		t3_write_reg(adapter, XGM_REG(A_XGM_RX_CTRL, 1), 0);
3004 
3005 		spin_lock(&adapter->work_lock);
3006 		t3_intr_disable(adapter);
3007 		queue_work(cxgb3_wq, &adapter->fatal_error_handler_task);
3008 		spin_unlock(&adapter->work_lock);
3009 	}
3010 	CH_ALERT(adapter, "encountered fatal error, operation suspended\n");
3011 	if (!t3_cim_ctl_blk_read(adapter, 0xa0, 4, fw_status))
3012 		CH_ALERT(adapter, "FW status: 0x%x, 0x%x, 0x%x, 0x%x\n",
3013 			 fw_status[0], fw_status[1],
3014 			 fw_status[2], fw_status[3]);
3015 }
3016 
3017 /**
3018  * t3_io_error_detected - called when PCI error is detected
3019  * @pdev: Pointer to PCI device
3020  * @state: The current pci connection state
3021  *
3022  * This function is called after a PCI bus error affecting
3023  * this device has been detected.
3024  */
3025 static pci_ers_result_t t3_io_error_detected(struct pci_dev *pdev,
3026 					     pci_channel_state_t state)
3027 {
3028 	struct adapter *adapter = pci_get_drvdata(pdev);
3029 
3030 	if (state == pci_channel_io_perm_failure)
3031 		return PCI_ERS_RESULT_DISCONNECT;
3032 
3033 	t3_adapter_error(adapter, 0, 0);
3034 
3035 	/* Request a slot reset. */
3036 	return PCI_ERS_RESULT_NEED_RESET;
3037 }
3038 
3039 /**
3040  * t3_io_slot_reset - called after the pci bus has been reset.
3041  * @pdev: Pointer to PCI device
3042  *
3043  * Restart the card from scratch, as if from a cold-boot.
3044  */
3045 static pci_ers_result_t t3_io_slot_reset(struct pci_dev *pdev)
3046 {
3047 	struct adapter *adapter = pci_get_drvdata(pdev);
3048 
3049 	if (!t3_reenable_adapter(adapter))
3050 		return PCI_ERS_RESULT_RECOVERED;
3051 
3052 	return PCI_ERS_RESULT_DISCONNECT;
3053 }
3054 
3055 /**
3056  * t3_io_resume - called when traffic can start flowing again.
3057  * @pdev: Pointer to PCI device
3058  *
3059  * This callback is called when the error recovery driver tells us that
3060  * its OK to resume normal operation.
3061  */
3062 static void t3_io_resume(struct pci_dev *pdev)
3063 {
3064 	struct adapter *adapter = pci_get_drvdata(pdev);
3065 
3066 	CH_ALERT(adapter, "adapter recovering, PEX ERR 0x%x\n",
3067 		 t3_read_reg(adapter, A_PCIE_PEX_ERR));
3068 
3069 	rtnl_lock();
3070 	t3_resume_ports(adapter);
3071 	rtnl_unlock();
3072 }
3073 
3074 static const struct pci_error_handlers t3_err_handler = {
3075 	.error_detected = t3_io_error_detected,
3076 	.slot_reset = t3_io_slot_reset,
3077 	.resume = t3_io_resume,
3078 };
3079 
3080 /*
3081  * Set the number of qsets based on the number of CPUs and the number of ports,
3082  * not to exceed the number of available qsets, assuming there are enough qsets
3083  * per port in HW.
3084  */
3085 static void set_nqsets(struct adapter *adap)
3086 {
3087 	int i, j = 0;
3088 	int num_cpus = netif_get_num_default_rss_queues();
3089 	int hwports = adap->params.nports;
3090 	int nqsets = adap->msix_nvectors - 1;
3091 
3092 	if (adap->params.rev > 0 && adap->flags & USING_MSIX) {
3093 		if (hwports == 2 &&
3094 		    (hwports * nqsets > SGE_QSETS ||
3095 		     num_cpus >= nqsets / hwports))
3096 			nqsets /= hwports;
3097 		if (nqsets > num_cpus)
3098 			nqsets = num_cpus;
3099 		if (nqsets < 1 || hwports == 4)
3100 			nqsets = 1;
3101 	} else {
3102 		nqsets = 1;
3103 	}
3104 
3105 	for_each_port(adap, i) {
3106 		struct port_info *pi = adap2pinfo(adap, i);
3107 
3108 		pi->first_qset = j;
3109 		pi->nqsets = nqsets;
3110 		j = pi->first_qset + nqsets;
3111 
3112 		dev_info(&adap->pdev->dev,
3113 			 "Port %d using %d queue sets.\n", i, nqsets);
3114 	}
3115 }
3116 
3117 static int cxgb_enable_msix(struct adapter *adap)
3118 {
3119 	struct msix_entry entries[SGE_QSETS + 1];
3120 	int vectors;
3121 	int i;
3122 
3123 	vectors = ARRAY_SIZE(entries);
3124 	for (i = 0; i < vectors; ++i)
3125 		entries[i].entry = i;
3126 
3127 	vectors = pci_enable_msix_range(adap->pdev, entries,
3128 					adap->params.nports + 1, vectors);
3129 	if (vectors < 0)
3130 		return vectors;
3131 
3132 	for (i = 0; i < vectors; ++i)
3133 		adap->msix_info[i].vec = entries[i].vector;
3134 	adap->msix_nvectors = vectors;
3135 
3136 	return 0;
3137 }
3138 
3139 static void print_port_info(struct adapter *adap, const struct adapter_info *ai)
3140 {
3141 	static const char *pci_variant[] = {
3142 		"PCI", "PCI-X", "PCI-X ECC", "PCI-X 266", "PCI Express"
3143 	};
3144 
3145 	int i;
3146 	char buf[80];
3147 
3148 	if (is_pcie(adap))
3149 		snprintf(buf, sizeof(buf), "%s x%d",
3150 			 pci_variant[adap->params.pci.variant],
3151 			 adap->params.pci.width);
3152 	else
3153 		snprintf(buf, sizeof(buf), "%s %dMHz/%d-bit",
3154 			 pci_variant[adap->params.pci.variant],
3155 			 adap->params.pci.speed, adap->params.pci.width);
3156 
3157 	for_each_port(adap, i) {
3158 		struct net_device *dev = adap->port[i];
3159 		const struct port_info *pi = netdev_priv(dev);
3160 
3161 		if (!test_bit(i, &adap->registered_device_map))
3162 			continue;
3163 		netdev_info(dev, "%s %s %sNIC (rev %d) %s%s\n",
3164 			    ai->desc, pi->phy.desc,
3165 			    is_offload(adap) ? "R" : "", adap->params.rev, buf,
3166 			    (adap->flags & USING_MSIX) ? " MSI-X" :
3167 			    (adap->flags & USING_MSI) ? " MSI" : "");
3168 		if (adap->name == dev->name && adap->params.vpd.mclk)
3169 			pr_info("%s: %uMB CM, %uMB PMTX, %uMB PMRX, S/N: %s\n",
3170 			       adap->name, t3_mc7_size(&adap->cm) >> 20,
3171 			       t3_mc7_size(&adap->pmtx) >> 20,
3172 			       t3_mc7_size(&adap->pmrx) >> 20,
3173 			       adap->params.vpd.sn);
3174 	}
3175 }
3176 
3177 static const struct net_device_ops cxgb_netdev_ops = {
3178 	.ndo_open		= cxgb_open,
3179 	.ndo_stop		= cxgb_close,
3180 	.ndo_start_xmit		= t3_eth_xmit,
3181 	.ndo_get_stats		= cxgb_get_stats,
3182 	.ndo_validate_addr	= eth_validate_addr,
3183 	.ndo_set_rx_mode	= cxgb_set_rxmode,
3184 	.ndo_do_ioctl		= cxgb_ioctl,
3185 	.ndo_change_mtu		= cxgb_change_mtu,
3186 	.ndo_set_mac_address	= cxgb_set_mac_addr,
3187 	.ndo_fix_features	= cxgb_fix_features,
3188 	.ndo_set_features	= cxgb_set_features,
3189 #ifdef CONFIG_NET_POLL_CONTROLLER
3190 	.ndo_poll_controller	= cxgb_netpoll,
3191 #endif
3192 };
3193 
3194 static void cxgb3_init_iscsi_mac(struct net_device *dev)
3195 {
3196 	struct port_info *pi = netdev_priv(dev);
3197 
3198 	memcpy(pi->iscsic.mac_addr, dev->dev_addr, ETH_ALEN);
3199 	pi->iscsic.mac_addr[3] |= 0x80;
3200 }
3201 
3202 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
3203 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
3204 			NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
3205 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
3206 {
3207 	int i, err, pci_using_dac = 0;
3208 	resource_size_t mmio_start, mmio_len;
3209 	const struct adapter_info *ai;
3210 	struct adapter *adapter = NULL;
3211 	struct port_info *pi;
3212 
3213 	if (!cxgb3_wq) {
3214 		cxgb3_wq = create_singlethread_workqueue(DRV_NAME);
3215 		if (!cxgb3_wq) {
3216 			pr_err("cannot initialize work queue\n");
3217 			return -ENOMEM;
3218 		}
3219 	}
3220 
3221 	err = pci_enable_device(pdev);
3222 	if (err) {
3223 		dev_err(&pdev->dev, "cannot enable PCI device\n");
3224 		goto out;
3225 	}
3226 
3227 	err = pci_request_regions(pdev, DRV_NAME);
3228 	if (err) {
3229 		/* Just info, some other driver may have claimed the device. */
3230 		dev_info(&pdev->dev, "cannot obtain PCI resources\n");
3231 		goto out_disable_device;
3232 	}
3233 
3234 	if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
3235 		pci_using_dac = 1;
3236 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3237 		if (err) {
3238 			dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
3239 			       "coherent allocations\n");
3240 			goto out_release_regions;
3241 		}
3242 	} else if ((err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32))) != 0) {
3243 		dev_err(&pdev->dev, "no usable DMA configuration\n");
3244 		goto out_release_regions;
3245 	}
3246 
3247 	pci_set_master(pdev);
3248 	pci_save_state(pdev);
3249 
3250 	mmio_start = pci_resource_start(pdev, 0);
3251 	mmio_len = pci_resource_len(pdev, 0);
3252 	ai = t3_get_adapter_info(ent->driver_data);
3253 
3254 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
3255 	if (!adapter) {
3256 		err = -ENOMEM;
3257 		goto out_release_regions;
3258 	}
3259 
3260 	adapter->nofail_skb =
3261 		alloc_skb(sizeof(struct cpl_set_tcb_field), GFP_KERNEL);
3262 	if (!adapter->nofail_skb) {
3263 		dev_err(&pdev->dev, "cannot allocate nofail buffer\n");
3264 		err = -ENOMEM;
3265 		goto out_free_adapter;
3266 	}
3267 
3268 	adapter->regs = ioremap(mmio_start, mmio_len);
3269 	if (!adapter->regs) {
3270 		dev_err(&pdev->dev, "cannot map device registers\n");
3271 		err = -ENOMEM;
3272 		goto out_free_adapter_nofail;
3273 	}
3274 
3275 	adapter->pdev = pdev;
3276 	adapter->name = pci_name(pdev);
3277 	adapter->msg_enable = dflt_msg_enable;
3278 	adapter->mmio_len = mmio_len;
3279 
3280 	mutex_init(&adapter->mdio_lock);
3281 	spin_lock_init(&adapter->work_lock);
3282 	spin_lock_init(&adapter->stats_lock);
3283 
3284 	INIT_LIST_HEAD(&adapter->adapter_list);
3285 	INIT_WORK(&adapter->ext_intr_handler_task, ext_intr_task);
3286 	INIT_WORK(&adapter->fatal_error_handler_task, fatal_error_task);
3287 
3288 	INIT_WORK(&adapter->db_full_task, db_full_task);
3289 	INIT_WORK(&adapter->db_empty_task, db_empty_task);
3290 	INIT_WORK(&adapter->db_drop_task, db_drop_task);
3291 
3292 	INIT_DELAYED_WORK(&adapter->adap_check_task, t3_adap_check_task);
3293 
3294 	for (i = 0; i < ai->nports0 + ai->nports1; ++i) {
3295 		struct net_device *netdev;
3296 
3297 		netdev = alloc_etherdev_mq(sizeof(struct port_info), SGE_QSETS);
3298 		if (!netdev) {
3299 			err = -ENOMEM;
3300 			goto out_free_dev;
3301 		}
3302 
3303 		SET_NETDEV_DEV(netdev, &pdev->dev);
3304 
3305 		adapter->port[i] = netdev;
3306 		pi = netdev_priv(netdev);
3307 		pi->adapter = adapter;
3308 		pi->port_id = i;
3309 		netif_carrier_off(netdev);
3310 		netdev->irq = pdev->irq;
3311 		netdev->mem_start = mmio_start;
3312 		netdev->mem_end = mmio_start + mmio_len - 1;
3313 		netdev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM |
3314 			NETIF_F_TSO | NETIF_F_RXCSUM | NETIF_F_HW_VLAN_CTAG_RX;
3315 		netdev->features |= netdev->hw_features |
3316 				    NETIF_F_HW_VLAN_CTAG_TX;
3317 		netdev->vlan_features |= netdev->features & VLAN_FEAT;
3318 		if (pci_using_dac)
3319 			netdev->features |= NETIF_F_HIGHDMA;
3320 
3321 		netdev->netdev_ops = &cxgb_netdev_ops;
3322 		netdev->ethtool_ops = &cxgb_ethtool_ops;
3323 		netdev->min_mtu = 81;
3324 		netdev->max_mtu = ETH_MAX_MTU;
3325 		netdev->dev_port = pi->port_id;
3326 	}
3327 
3328 	pci_set_drvdata(pdev, adapter);
3329 	if (t3_prep_adapter(adapter, ai, 1) < 0) {
3330 		err = -ENODEV;
3331 		goto out_free_dev;
3332 	}
3333 
3334 	/*
3335 	 * The card is now ready to go.  If any errors occur during device
3336 	 * registration we do not fail the whole card but rather proceed only
3337 	 * with the ports we manage to register successfully.  However we must
3338 	 * register at least one net device.
3339 	 */
3340 	for_each_port(adapter, i) {
3341 		err = register_netdev(adapter->port[i]);
3342 		if (err)
3343 			dev_warn(&pdev->dev,
3344 				 "cannot register net device %s, skipping\n",
3345 				 adapter->port[i]->name);
3346 		else {
3347 			/*
3348 			 * Change the name we use for messages to the name of
3349 			 * the first successfully registered interface.
3350 			 */
3351 			if (!adapter->registered_device_map)
3352 				adapter->name = adapter->port[i]->name;
3353 
3354 			__set_bit(i, &adapter->registered_device_map);
3355 		}
3356 	}
3357 	if (!adapter->registered_device_map) {
3358 		dev_err(&pdev->dev, "could not register any net devices\n");
3359 		goto out_free_dev;
3360 	}
3361 
3362 	for_each_port(adapter, i)
3363 		cxgb3_init_iscsi_mac(adapter->port[i]);
3364 
3365 	/* Driver's ready. Reflect it on LEDs */
3366 	t3_led_ready(adapter);
3367 
3368 	if (is_offload(adapter)) {
3369 		__set_bit(OFFLOAD_DEVMAP_BIT, &adapter->registered_device_map);
3370 		cxgb3_adapter_ofld(adapter);
3371 	}
3372 
3373 	/* See what interrupts we'll be using */
3374 	if (msi > 1 && cxgb_enable_msix(adapter) == 0)
3375 		adapter->flags |= USING_MSIX;
3376 	else if (msi > 0 && pci_enable_msi(pdev) == 0)
3377 		adapter->flags |= USING_MSI;
3378 
3379 	set_nqsets(adapter);
3380 
3381 	err = sysfs_create_group(&adapter->port[0]->dev.kobj,
3382 				 &cxgb3_attr_group);
3383 	if (err) {
3384 		dev_err(&pdev->dev, "cannot create sysfs group\n");
3385 		goto out_close_led;
3386 	}
3387 
3388 	print_port_info(adapter, ai);
3389 	return 0;
3390 
3391 out_close_led:
3392 	t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL, 0);
3393 
3394 out_free_dev:
3395 	iounmap(adapter->regs);
3396 	for (i = ai->nports0 + ai->nports1 - 1; i >= 0; --i)
3397 		if (adapter->port[i])
3398 			free_netdev(adapter->port[i]);
3399 
3400 out_free_adapter_nofail:
3401 	kfree_skb(adapter->nofail_skb);
3402 
3403 out_free_adapter:
3404 	kfree(adapter);
3405 
3406 out_release_regions:
3407 	pci_release_regions(pdev);
3408 out_disable_device:
3409 	pci_disable_device(pdev);
3410 out:
3411 	return err;
3412 }
3413 
3414 static void remove_one(struct pci_dev *pdev)
3415 {
3416 	struct adapter *adapter = pci_get_drvdata(pdev);
3417 
3418 	if (adapter) {
3419 		int i;
3420 
3421 		t3_sge_stop(adapter);
3422 		sysfs_remove_group(&adapter->port[0]->dev.kobj,
3423 				   &cxgb3_attr_group);
3424 
3425 		if (is_offload(adapter)) {
3426 			cxgb3_adapter_unofld(adapter);
3427 			if (test_bit(OFFLOAD_DEVMAP_BIT,
3428 				     &adapter->open_device_map))
3429 				offload_close(&adapter->tdev);
3430 		}
3431 
3432 		for_each_port(adapter, i)
3433 		    if (test_bit(i, &adapter->registered_device_map))
3434 			unregister_netdev(adapter->port[i]);
3435 
3436 		t3_stop_sge_timers(adapter);
3437 		t3_free_sge_resources(adapter);
3438 		cxgb_disable_msi(adapter);
3439 
3440 		for_each_port(adapter, i)
3441 			if (adapter->port[i])
3442 				free_netdev(adapter->port[i]);
3443 
3444 		iounmap(adapter->regs);
3445 		kfree_skb(adapter->nofail_skb);
3446 		kfree(adapter);
3447 		pci_release_regions(pdev);
3448 		pci_disable_device(pdev);
3449 	}
3450 }
3451 
3452 static struct pci_driver driver = {
3453 	.name = DRV_NAME,
3454 	.id_table = cxgb3_pci_tbl,
3455 	.probe = init_one,
3456 	.remove = remove_one,
3457 	.err_handler = &t3_err_handler,
3458 };
3459 
3460 static int __init cxgb3_init_module(void)
3461 {
3462 	int ret;
3463 
3464 	cxgb3_offload_init();
3465 
3466 	ret = pci_register_driver(&driver);
3467 	return ret;
3468 }
3469 
3470 static void __exit cxgb3_cleanup_module(void)
3471 {
3472 	pci_unregister_driver(&driver);
3473 	if (cxgb3_wq)
3474 		destroy_workqueue(cxgb3_wq);
3475 }
3476 
3477 module_init(cxgb3_init_module);
3478 module_exit(cxgb3_cleanup_module);
3479