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 			  struct kernel_ethtool_ringparam *kernel_e,
1953 			  struct netlink_ext_ack *extack)
1954 {
1955 	struct port_info *pi = netdev_priv(dev);
1956 	struct adapter *adapter = pi->adapter;
1957 	const struct qset_params *q = &adapter->params.sge.qset[pi->first_qset];
1958 
1959 	e->rx_max_pending = MAX_RX_BUFFERS;
1960 	e->rx_jumbo_max_pending = MAX_RX_JUMBO_BUFFERS;
1961 	e->tx_max_pending = MAX_TXQ_ENTRIES;
1962 
1963 	e->rx_pending = q->fl_size;
1964 	e->rx_mini_pending = q->rspq_size;
1965 	e->rx_jumbo_pending = q->jumbo_size;
1966 	e->tx_pending = q->txq_size[0];
1967 }
1968 
1969 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e,
1970 			 struct kernel_ethtool_ringparam *kernel_e,
1971 			 struct netlink_ext_ack *extack)
1972 {
1973 	struct port_info *pi = netdev_priv(dev);
1974 	struct adapter *adapter = pi->adapter;
1975 	struct qset_params *q;
1976 	int i;
1977 
1978 	if (e->rx_pending > MAX_RX_BUFFERS ||
1979 	    e->rx_jumbo_pending > MAX_RX_JUMBO_BUFFERS ||
1980 	    e->tx_pending > MAX_TXQ_ENTRIES ||
1981 	    e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1982 	    e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1983 	    e->rx_pending < MIN_FL_ENTRIES ||
1984 	    e->rx_jumbo_pending < MIN_FL_ENTRIES ||
1985 	    e->tx_pending < adapter->params.nports * MIN_TXQ_ENTRIES)
1986 		return -EINVAL;
1987 
1988 	if (adapter->flags & FULL_INIT_DONE)
1989 		return -EBUSY;
1990 
1991 	q = &adapter->params.sge.qset[pi->first_qset];
1992 	for (i = 0; i < pi->nqsets; ++i, ++q) {
1993 		q->rspq_size = e->rx_mini_pending;
1994 		q->fl_size = e->rx_pending;
1995 		q->jumbo_size = e->rx_jumbo_pending;
1996 		q->txq_size[0] = e->tx_pending;
1997 		q->txq_size[1] = e->tx_pending;
1998 		q->txq_size[2] = e->tx_pending;
1999 	}
2000 	return 0;
2001 }
2002 
2003 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c,
2004 			struct kernel_ethtool_coalesce *kernel_coal,
2005 			struct netlink_ext_ack *extack)
2006 {
2007 	struct port_info *pi = netdev_priv(dev);
2008 	struct adapter *adapter = pi->adapter;
2009 	struct qset_params *qsp;
2010 	struct sge_qset *qs;
2011 	int i;
2012 
2013 	if (c->rx_coalesce_usecs * 10 > M_NEWTIMER)
2014 		return -EINVAL;
2015 
2016 	for (i = 0; i < pi->nqsets; i++) {
2017 		qsp = &adapter->params.sge.qset[i];
2018 		qs = &adapter->sge.qs[i];
2019 		qsp->coalesce_usecs = c->rx_coalesce_usecs;
2020 		t3_update_qset_coalesce(qs, qsp);
2021 	}
2022 
2023 	return 0;
2024 }
2025 
2026 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c,
2027 			struct kernel_ethtool_coalesce *kernel_coal,
2028 			struct netlink_ext_ack *extack)
2029 {
2030 	struct port_info *pi = netdev_priv(dev);
2031 	struct adapter *adapter = pi->adapter;
2032 	struct qset_params *q = adapter->params.sge.qset;
2033 
2034 	c->rx_coalesce_usecs = q->coalesce_usecs;
2035 	return 0;
2036 }
2037 
2038 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
2039 		      u8 * data)
2040 {
2041 	struct port_info *pi = netdev_priv(dev);
2042 	struct adapter *adapter = pi->adapter;
2043 	int cnt;
2044 
2045 	e->magic = EEPROM_MAGIC;
2046 	cnt = pci_read_vpd(adapter->pdev, e->offset, e->len, data);
2047 	if (cnt < 0)
2048 		return cnt;
2049 
2050 	e->len = cnt;
2051 
2052 	return 0;
2053 }
2054 
2055 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
2056 		      u8 * data)
2057 {
2058 	struct port_info *pi = netdev_priv(dev);
2059 	struct adapter *adapter = pi->adapter;
2060 	u32 aligned_offset, aligned_len;
2061 	u8 *buf;
2062 	int err;
2063 
2064 	if (eeprom->magic != EEPROM_MAGIC)
2065 		return -EINVAL;
2066 
2067 	aligned_offset = eeprom->offset & ~3;
2068 	aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
2069 
2070 	if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
2071 		buf = kmalloc(aligned_len, GFP_KERNEL);
2072 		if (!buf)
2073 			return -ENOMEM;
2074 		err = pci_read_vpd(adapter->pdev, aligned_offset, aligned_len,
2075 				   buf);
2076 		if (err < 0)
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 	err = pci_write_vpd(adapter->pdev, aligned_offset, aligned_len, buf);
2087 	if (err >= 0)
2088 		err = t3_seeprom_wp(adapter, 1);
2089 out:
2090 	if (buf != data)
2091 		kfree(buf);
2092 	return err < 0 ? err : 0;
2093 }
2094 
2095 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2096 {
2097 	wol->supported = 0;
2098 	wol->wolopts = 0;
2099 	memset(&wol->sopass, 0, sizeof(wol->sopass));
2100 }
2101 
2102 static const struct ethtool_ops cxgb_ethtool_ops = {
2103 	.supported_coalesce_params = ETHTOOL_COALESCE_RX_USECS,
2104 	.get_drvinfo = get_drvinfo,
2105 	.get_msglevel = get_msglevel,
2106 	.set_msglevel = set_msglevel,
2107 	.get_ringparam = get_sge_param,
2108 	.set_ringparam = set_sge_param,
2109 	.get_coalesce = get_coalesce,
2110 	.set_coalesce = set_coalesce,
2111 	.get_eeprom_len = get_eeprom_len,
2112 	.get_eeprom = get_eeprom,
2113 	.set_eeprom = set_eeprom,
2114 	.get_pauseparam = get_pauseparam,
2115 	.set_pauseparam = set_pauseparam,
2116 	.get_link = ethtool_op_get_link,
2117 	.get_strings = get_strings,
2118 	.set_phys_id = set_phys_id,
2119 	.nway_reset = restart_autoneg,
2120 	.get_sset_count = get_sset_count,
2121 	.get_ethtool_stats = get_stats,
2122 	.get_regs_len = get_regs_len,
2123 	.get_regs = get_regs,
2124 	.get_wol = get_wol,
2125 	.get_link_ksettings = get_link_ksettings,
2126 	.set_link_ksettings = set_link_ksettings,
2127 };
2128 
2129 static int in_range(int val, int lo, int hi)
2130 {
2131 	return val < 0 || (val <= hi && val >= lo);
2132 }
2133 
2134 static int cxgb_siocdevprivate(struct net_device *dev,
2135 			       struct ifreq *ifreq,
2136 			       void __user *useraddr,
2137 			       int cmd)
2138 {
2139 	struct port_info *pi = netdev_priv(dev);
2140 	struct adapter *adapter = pi->adapter;
2141 	int ret;
2142 
2143 	if (cmd != SIOCCHIOCTL)
2144 		return -EOPNOTSUPP;
2145 
2146 	if (copy_from_user(&cmd, useraddr, sizeof(cmd)))
2147 		return -EFAULT;
2148 
2149 	switch (cmd) {
2150 	case CHELSIO_SET_QSET_PARAMS:{
2151 		int i;
2152 		struct qset_params *q;
2153 		struct ch_qset_params t;
2154 		int q1 = pi->first_qset;
2155 		int nqsets = pi->nqsets;
2156 
2157 		if (!capable(CAP_NET_ADMIN))
2158 			return -EPERM;
2159 		if (copy_from_user(&t, useraddr, sizeof(t)))
2160 			return -EFAULT;
2161 		if (t.cmd != CHELSIO_SET_QSET_PARAMS)
2162 			return -EINVAL;
2163 		if (t.qset_idx >= SGE_QSETS)
2164 			return -EINVAL;
2165 		if (!in_range(t.intr_lat, 0, M_NEWTIMER) ||
2166 		    !in_range(t.cong_thres, 0, 255) ||
2167 		    !in_range(t.txq_size[0], MIN_TXQ_ENTRIES,
2168 			      MAX_TXQ_ENTRIES) ||
2169 		    !in_range(t.txq_size[1], MIN_TXQ_ENTRIES,
2170 			      MAX_TXQ_ENTRIES) ||
2171 		    !in_range(t.txq_size[2], MIN_CTRL_TXQ_ENTRIES,
2172 			      MAX_CTRL_TXQ_ENTRIES) ||
2173 		    !in_range(t.fl_size[0], MIN_FL_ENTRIES,
2174 			      MAX_RX_BUFFERS) ||
2175 		    !in_range(t.fl_size[1], MIN_FL_ENTRIES,
2176 			      MAX_RX_JUMBO_BUFFERS) ||
2177 		    !in_range(t.rspq_size, MIN_RSPQ_ENTRIES,
2178 			      MAX_RSPQ_ENTRIES))
2179 			return -EINVAL;
2180 
2181 		if ((adapter->flags & FULL_INIT_DONE) &&
2182 			(t.rspq_size >= 0 || t.fl_size[0] >= 0 ||
2183 			t.fl_size[1] >= 0 || t.txq_size[0] >= 0 ||
2184 			t.txq_size[1] >= 0 || t.txq_size[2] >= 0 ||
2185 			t.polling >= 0 || t.cong_thres >= 0))
2186 			return -EBUSY;
2187 
2188 		/* Allow setting of any available qset when offload enabled */
2189 		if (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2190 			q1 = 0;
2191 			for_each_port(adapter, i) {
2192 				pi = adap2pinfo(adapter, i);
2193 				nqsets += pi->first_qset + pi->nqsets;
2194 			}
2195 		}
2196 
2197 		if (t.qset_idx < q1)
2198 			return -EINVAL;
2199 		if (t.qset_idx > q1 + nqsets - 1)
2200 			return -EINVAL;
2201 
2202 		q = &adapter->params.sge.qset[t.qset_idx];
2203 
2204 		if (t.rspq_size >= 0)
2205 			q->rspq_size = t.rspq_size;
2206 		if (t.fl_size[0] >= 0)
2207 			q->fl_size = t.fl_size[0];
2208 		if (t.fl_size[1] >= 0)
2209 			q->jumbo_size = t.fl_size[1];
2210 		if (t.txq_size[0] >= 0)
2211 			q->txq_size[0] = t.txq_size[0];
2212 		if (t.txq_size[1] >= 0)
2213 			q->txq_size[1] = t.txq_size[1];
2214 		if (t.txq_size[2] >= 0)
2215 			q->txq_size[2] = t.txq_size[2];
2216 		if (t.cong_thres >= 0)
2217 			q->cong_thres = t.cong_thres;
2218 		if (t.intr_lat >= 0) {
2219 			struct sge_qset *qs =
2220 				&adapter->sge.qs[t.qset_idx];
2221 
2222 			q->coalesce_usecs = t.intr_lat;
2223 			t3_update_qset_coalesce(qs, q);
2224 		}
2225 		if (t.polling >= 0) {
2226 			if (adapter->flags & USING_MSIX)
2227 				q->polling = t.polling;
2228 			else {
2229 				/* No polling with INTx for T3A */
2230 				if (adapter->params.rev == 0 &&
2231 					!(adapter->flags & USING_MSI))
2232 					t.polling = 0;
2233 
2234 				for (i = 0; i < SGE_QSETS; i++) {
2235 					q = &adapter->params.sge.
2236 						qset[i];
2237 					q->polling = t.polling;
2238 				}
2239 			}
2240 		}
2241 
2242 		if (t.lro >= 0) {
2243 			if (t.lro)
2244 				dev->wanted_features |= NETIF_F_GRO;
2245 			else
2246 				dev->wanted_features &= ~NETIF_F_GRO;
2247 			netdev_update_features(dev);
2248 		}
2249 
2250 		break;
2251 	}
2252 	case CHELSIO_GET_QSET_PARAMS:{
2253 		struct qset_params *q;
2254 		struct ch_qset_params t;
2255 		int q1 = pi->first_qset;
2256 		int nqsets = pi->nqsets;
2257 		int i;
2258 
2259 		if (copy_from_user(&t, useraddr, sizeof(t)))
2260 			return -EFAULT;
2261 
2262 		if (t.cmd != CHELSIO_GET_QSET_PARAMS)
2263 			return -EINVAL;
2264 
2265 		/* Display qsets for all ports when offload enabled */
2266 		if (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2267 			q1 = 0;
2268 			for_each_port(adapter, i) {
2269 				pi = adap2pinfo(adapter, i);
2270 				nqsets = pi->first_qset + pi->nqsets;
2271 			}
2272 		}
2273 
2274 		if (t.qset_idx >= nqsets)
2275 			return -EINVAL;
2276 		t.qset_idx = array_index_nospec(t.qset_idx, nqsets);
2277 
2278 		q = &adapter->params.sge.qset[q1 + t.qset_idx];
2279 		t.rspq_size = q->rspq_size;
2280 		t.txq_size[0] = q->txq_size[0];
2281 		t.txq_size[1] = q->txq_size[1];
2282 		t.txq_size[2] = q->txq_size[2];
2283 		t.fl_size[0] = q->fl_size;
2284 		t.fl_size[1] = q->jumbo_size;
2285 		t.polling = q->polling;
2286 		t.lro = !!(dev->features & NETIF_F_GRO);
2287 		t.intr_lat = q->coalesce_usecs;
2288 		t.cong_thres = q->cong_thres;
2289 		t.qnum = q1;
2290 
2291 		if (adapter->flags & USING_MSIX)
2292 			t.vector = adapter->msix_info[q1 + t.qset_idx + 1].vec;
2293 		else
2294 			t.vector = adapter->pdev->irq;
2295 
2296 		if (copy_to_user(useraddr, &t, sizeof(t)))
2297 			return -EFAULT;
2298 		break;
2299 	}
2300 	case CHELSIO_SET_QSET_NUM:{
2301 		struct ch_reg edata;
2302 		unsigned int i, first_qset = 0, other_qsets = 0;
2303 
2304 		if (!capable(CAP_NET_ADMIN))
2305 			return -EPERM;
2306 		if (adapter->flags & FULL_INIT_DONE)
2307 			return -EBUSY;
2308 		if (copy_from_user(&edata, useraddr, sizeof(edata)))
2309 			return -EFAULT;
2310 		if (edata.cmd != CHELSIO_SET_QSET_NUM)
2311 			return -EINVAL;
2312 		if (edata.val < 1 ||
2313 			(edata.val > 1 && !(adapter->flags & USING_MSIX)))
2314 			return -EINVAL;
2315 
2316 		for_each_port(adapter, i)
2317 			if (adapter->port[i] && adapter->port[i] != dev)
2318 				other_qsets += adap2pinfo(adapter, i)->nqsets;
2319 
2320 		if (edata.val + other_qsets > SGE_QSETS)
2321 			return -EINVAL;
2322 
2323 		pi->nqsets = edata.val;
2324 
2325 		for_each_port(adapter, i)
2326 			if (adapter->port[i]) {
2327 				pi = adap2pinfo(adapter, i);
2328 				pi->first_qset = first_qset;
2329 				first_qset += pi->nqsets;
2330 			}
2331 		break;
2332 	}
2333 	case CHELSIO_GET_QSET_NUM:{
2334 		struct ch_reg edata;
2335 
2336 		memset(&edata, 0, sizeof(struct ch_reg));
2337 
2338 		edata.cmd = CHELSIO_GET_QSET_NUM;
2339 		edata.val = pi->nqsets;
2340 		if (copy_to_user(useraddr, &edata, sizeof(edata)))
2341 			return -EFAULT;
2342 		break;
2343 	}
2344 	case CHELSIO_LOAD_FW:{
2345 		u8 *fw_data;
2346 		struct ch_mem_range t;
2347 
2348 		if (!capable(CAP_SYS_RAWIO))
2349 			return -EPERM;
2350 		if (copy_from_user(&t, useraddr, sizeof(t)))
2351 			return -EFAULT;
2352 		if (t.cmd != CHELSIO_LOAD_FW)
2353 			return -EINVAL;
2354 		/* Check t.len sanity ? */
2355 		fw_data = memdup_user(useraddr + sizeof(t), t.len);
2356 		if (IS_ERR(fw_data))
2357 			return PTR_ERR(fw_data);
2358 
2359 		ret = t3_load_fw(adapter, fw_data, t.len);
2360 		kfree(fw_data);
2361 		if (ret)
2362 			return ret;
2363 		break;
2364 	}
2365 	case CHELSIO_SETMTUTAB:{
2366 		struct ch_mtus m;
2367 		int i;
2368 
2369 		if (!is_offload(adapter))
2370 			return -EOPNOTSUPP;
2371 		if (!capable(CAP_NET_ADMIN))
2372 			return -EPERM;
2373 		if (offload_running(adapter))
2374 			return -EBUSY;
2375 		if (copy_from_user(&m, useraddr, sizeof(m)))
2376 			return -EFAULT;
2377 		if (m.cmd != CHELSIO_SETMTUTAB)
2378 			return -EINVAL;
2379 		if (m.nmtus != NMTUS)
2380 			return -EINVAL;
2381 		if (m.mtus[0] < 81)	/* accommodate SACK */
2382 			return -EINVAL;
2383 
2384 		/* MTUs must be in ascending order */
2385 		for (i = 1; i < NMTUS; ++i)
2386 			if (m.mtus[i] < m.mtus[i - 1])
2387 				return -EINVAL;
2388 
2389 		memcpy(adapter->params.mtus, m.mtus,
2390 			sizeof(adapter->params.mtus));
2391 		break;
2392 	}
2393 	case CHELSIO_GET_PM:{
2394 		struct tp_params *p = &adapter->params.tp;
2395 		struct ch_pm m = {.cmd = CHELSIO_GET_PM };
2396 
2397 		if (!is_offload(adapter))
2398 			return -EOPNOTSUPP;
2399 		m.tx_pg_sz = p->tx_pg_size;
2400 		m.tx_num_pg = p->tx_num_pgs;
2401 		m.rx_pg_sz = p->rx_pg_size;
2402 		m.rx_num_pg = p->rx_num_pgs;
2403 		m.pm_total = p->pmtx_size + p->chan_rx_size * p->nchan;
2404 		if (copy_to_user(useraddr, &m, sizeof(m)))
2405 			return -EFAULT;
2406 		break;
2407 	}
2408 	case CHELSIO_SET_PM:{
2409 		struct ch_pm m;
2410 		struct tp_params *p = &adapter->params.tp;
2411 
2412 		if (!is_offload(adapter))
2413 			return -EOPNOTSUPP;
2414 		if (!capable(CAP_NET_ADMIN))
2415 			return -EPERM;
2416 		if (adapter->flags & FULL_INIT_DONE)
2417 			return -EBUSY;
2418 		if (copy_from_user(&m, useraddr, sizeof(m)))
2419 			return -EFAULT;
2420 		if (m.cmd != CHELSIO_SET_PM)
2421 			return -EINVAL;
2422 		if (!is_power_of_2(m.rx_pg_sz) ||
2423 			!is_power_of_2(m.tx_pg_sz))
2424 			return -EINVAL;	/* not power of 2 */
2425 		if (!(m.rx_pg_sz & 0x14000))
2426 			return -EINVAL;	/* not 16KB or 64KB */
2427 		if (!(m.tx_pg_sz & 0x1554000))
2428 			return -EINVAL;
2429 		if (m.tx_num_pg == -1)
2430 			m.tx_num_pg = p->tx_num_pgs;
2431 		if (m.rx_num_pg == -1)
2432 			m.rx_num_pg = p->rx_num_pgs;
2433 		if (m.tx_num_pg % 24 || m.rx_num_pg % 24)
2434 			return -EINVAL;
2435 		if (m.rx_num_pg * m.rx_pg_sz > p->chan_rx_size ||
2436 			m.tx_num_pg * m.tx_pg_sz > p->chan_tx_size)
2437 			return -EINVAL;
2438 		p->rx_pg_size = m.rx_pg_sz;
2439 		p->tx_pg_size = m.tx_pg_sz;
2440 		p->rx_num_pgs = m.rx_num_pg;
2441 		p->tx_num_pgs = m.tx_num_pg;
2442 		break;
2443 	}
2444 	case CHELSIO_GET_MEM:{
2445 		struct ch_mem_range t;
2446 		struct mc7 *mem;
2447 		u64 buf[32];
2448 
2449 		if (!is_offload(adapter))
2450 			return -EOPNOTSUPP;
2451 		if (!capable(CAP_NET_ADMIN))
2452 			return -EPERM;
2453 		if (!(adapter->flags & FULL_INIT_DONE))
2454 			return -EIO;	/* need the memory controllers */
2455 		if (copy_from_user(&t, useraddr, sizeof(t)))
2456 			return -EFAULT;
2457 		if (t.cmd != CHELSIO_GET_MEM)
2458 			return -EINVAL;
2459 		if ((t.addr & 7) || (t.len & 7))
2460 			return -EINVAL;
2461 		if (t.mem_id == MEM_CM)
2462 			mem = &adapter->cm;
2463 		else if (t.mem_id == MEM_PMRX)
2464 			mem = &adapter->pmrx;
2465 		else if (t.mem_id == MEM_PMTX)
2466 			mem = &adapter->pmtx;
2467 		else
2468 			return -EINVAL;
2469 
2470 		/*
2471 		 * Version scheme:
2472 		 * bits 0..9: chip version
2473 		 * bits 10..15: chip revision
2474 		 */
2475 		t.version = 3 | (adapter->params.rev << 10);
2476 		if (copy_to_user(useraddr, &t, sizeof(t)))
2477 			return -EFAULT;
2478 
2479 		/*
2480 		 * Read 256 bytes at a time as len can be large and we don't
2481 		 * want to use huge intermediate buffers.
2482 		 */
2483 		useraddr += sizeof(t);	/* advance to start of buffer */
2484 		while (t.len) {
2485 			unsigned int chunk =
2486 				min_t(unsigned int, t.len, sizeof(buf));
2487 
2488 			ret =
2489 				t3_mc7_bd_read(mem, t.addr / 8, chunk / 8,
2490 						buf);
2491 			if (ret)
2492 				return ret;
2493 			if (copy_to_user(useraddr, buf, chunk))
2494 				return -EFAULT;
2495 			useraddr += chunk;
2496 			t.addr += chunk;
2497 			t.len -= chunk;
2498 		}
2499 		break;
2500 	}
2501 	case CHELSIO_SET_TRACE_FILTER:{
2502 		struct ch_trace t;
2503 		const struct trace_params *tp;
2504 
2505 		if (!capable(CAP_NET_ADMIN))
2506 			return -EPERM;
2507 		if (!offload_running(adapter))
2508 			return -EAGAIN;
2509 		if (copy_from_user(&t, useraddr, sizeof(t)))
2510 			return -EFAULT;
2511 		if (t.cmd != CHELSIO_SET_TRACE_FILTER)
2512 			return -EINVAL;
2513 
2514 		tp = (const struct trace_params *)&t.sip;
2515 		if (t.config_tx)
2516 			t3_config_trace_filter(adapter, tp, 0,
2517 						t.invert_match,
2518 						t.trace_tx);
2519 		if (t.config_rx)
2520 			t3_config_trace_filter(adapter, tp, 1,
2521 						t.invert_match,
2522 						t.trace_rx);
2523 		break;
2524 	}
2525 	default:
2526 		return -EOPNOTSUPP;
2527 	}
2528 	return 0;
2529 }
2530 
2531 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2532 {
2533 	struct mii_ioctl_data *data = if_mii(req);
2534 	struct port_info *pi = netdev_priv(dev);
2535 	struct adapter *adapter = pi->adapter;
2536 
2537 	switch (cmd) {
2538 	case SIOCGMIIREG:
2539 	case SIOCSMIIREG:
2540 		/* Convert phy_id from older PRTAD/DEVAD format */
2541 		if (is_10G(adapter) &&
2542 		    !mdio_phy_id_is_c45(data->phy_id) &&
2543 		    (data->phy_id & 0x1f00) &&
2544 		    !(data->phy_id & 0xe0e0))
2545 			data->phy_id = mdio_phy_id_c45(data->phy_id >> 8,
2546 						       data->phy_id & 0x1f);
2547 		fallthrough;
2548 	case SIOCGMIIPHY:
2549 		return mdio_mii_ioctl(&pi->phy.mdio, data, cmd);
2550 	default:
2551 		return -EOPNOTSUPP;
2552 	}
2553 }
2554 
2555 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2556 {
2557 	struct port_info *pi = netdev_priv(dev);
2558 	struct adapter *adapter = pi->adapter;
2559 	int ret;
2560 
2561 	if ((ret = t3_mac_set_mtu(&pi->mac, new_mtu)))
2562 		return ret;
2563 	dev->mtu = new_mtu;
2564 	init_port_mtus(adapter);
2565 	if (adapter->params.rev == 0 && offload_running(adapter))
2566 		t3_load_mtus(adapter, adapter->params.mtus,
2567 			     adapter->params.a_wnd, adapter->params.b_wnd,
2568 			     adapter->port[0]->mtu);
2569 	return 0;
2570 }
2571 
2572 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2573 {
2574 	struct port_info *pi = netdev_priv(dev);
2575 	struct adapter *adapter = pi->adapter;
2576 	struct sockaddr *addr = p;
2577 
2578 	if (!is_valid_ether_addr(addr->sa_data))
2579 		return -EADDRNOTAVAIL;
2580 
2581 	eth_hw_addr_set(dev, addr->sa_data);
2582 	t3_mac_set_address(&pi->mac, LAN_MAC_IDX, dev->dev_addr);
2583 	if (offload_running(adapter))
2584 		write_smt_entry(adapter, pi->port_id);
2585 	return 0;
2586 }
2587 
2588 static netdev_features_t cxgb_fix_features(struct net_device *dev,
2589 	netdev_features_t features)
2590 {
2591 	/*
2592 	 * Since there is no support for separate rx/tx vlan accel
2593 	 * enable/disable make sure tx flag is always in same state as rx.
2594 	 */
2595 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
2596 		features |= NETIF_F_HW_VLAN_CTAG_TX;
2597 	else
2598 		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2599 
2600 	return features;
2601 }
2602 
2603 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
2604 {
2605 	netdev_features_t changed = dev->features ^ features;
2606 
2607 	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2608 		cxgb_vlan_mode(dev, features);
2609 
2610 	return 0;
2611 }
2612 
2613 #ifdef CONFIG_NET_POLL_CONTROLLER
2614 static void cxgb_netpoll(struct net_device *dev)
2615 {
2616 	struct port_info *pi = netdev_priv(dev);
2617 	struct adapter *adapter = pi->adapter;
2618 	int qidx;
2619 
2620 	for (qidx = pi->first_qset; qidx < pi->first_qset + pi->nqsets; qidx++) {
2621 		struct sge_qset *qs = &adapter->sge.qs[qidx];
2622 		void *source;
2623 
2624 		if (adapter->flags & USING_MSIX)
2625 			source = qs;
2626 		else
2627 			source = adapter;
2628 
2629 		t3_intr_handler(adapter, qs->rspq.polling) (0, source);
2630 	}
2631 }
2632 #endif
2633 
2634 /*
2635  * Periodic accumulation of MAC statistics.
2636  */
2637 static void mac_stats_update(struct adapter *adapter)
2638 {
2639 	int i;
2640 
2641 	for_each_port(adapter, i) {
2642 		struct net_device *dev = adapter->port[i];
2643 		struct port_info *p = netdev_priv(dev);
2644 
2645 		if (netif_running(dev)) {
2646 			spin_lock(&adapter->stats_lock);
2647 			t3_mac_update_stats(&p->mac);
2648 			spin_unlock(&adapter->stats_lock);
2649 		}
2650 	}
2651 }
2652 
2653 static void check_link_status(struct adapter *adapter)
2654 {
2655 	int i;
2656 
2657 	for_each_port(adapter, i) {
2658 		struct net_device *dev = adapter->port[i];
2659 		struct port_info *p = netdev_priv(dev);
2660 		int link_fault;
2661 
2662 		spin_lock_irq(&adapter->work_lock);
2663 		link_fault = p->link_fault;
2664 		spin_unlock_irq(&adapter->work_lock);
2665 
2666 		if (link_fault) {
2667 			t3_link_fault(adapter, i);
2668 			continue;
2669 		}
2670 
2671 		if (!(p->phy.caps & SUPPORTED_IRQ) && netif_running(dev)) {
2672 			t3_xgm_intr_disable(adapter, i);
2673 			t3_read_reg(adapter, A_XGM_INT_STATUS + p->mac.offset);
2674 
2675 			t3_link_changed(adapter, i);
2676 			t3_xgm_intr_enable(adapter, i);
2677 		}
2678 	}
2679 }
2680 
2681 static void check_t3b2_mac(struct adapter *adapter)
2682 {
2683 	int i;
2684 
2685 	if (!rtnl_trylock())	/* synchronize with ifdown */
2686 		return;
2687 
2688 	for_each_port(adapter, i) {
2689 		struct net_device *dev = adapter->port[i];
2690 		struct port_info *p = netdev_priv(dev);
2691 		int status;
2692 
2693 		if (!netif_running(dev))
2694 			continue;
2695 
2696 		status = 0;
2697 		if (netif_running(dev) && netif_carrier_ok(dev))
2698 			status = t3b2_mac_watchdog_task(&p->mac);
2699 		if (status == 1)
2700 			p->mac.stats.num_toggled++;
2701 		else if (status == 2) {
2702 			struct cmac *mac = &p->mac;
2703 
2704 			t3_mac_set_mtu(mac, dev->mtu);
2705 			t3_mac_set_address(mac, LAN_MAC_IDX, dev->dev_addr);
2706 			cxgb_set_rxmode(dev);
2707 			t3_link_start(&p->phy, mac, &p->link_config);
2708 			t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
2709 			t3_port_intr_enable(adapter, p->port_id);
2710 			p->mac.stats.num_resets++;
2711 		}
2712 	}
2713 	rtnl_unlock();
2714 }
2715 
2716 
2717 static void t3_adap_check_task(struct work_struct *work)
2718 {
2719 	struct adapter *adapter = container_of(work, struct adapter,
2720 					       adap_check_task.work);
2721 	const struct adapter_params *p = &adapter->params;
2722 	int port;
2723 	unsigned int v, status, reset;
2724 
2725 	adapter->check_task_cnt++;
2726 
2727 	check_link_status(adapter);
2728 
2729 	/* Accumulate MAC stats if needed */
2730 	if (!p->linkpoll_period ||
2731 	    (adapter->check_task_cnt * p->linkpoll_period) / 10 >=
2732 	    p->stats_update_period) {
2733 		mac_stats_update(adapter);
2734 		adapter->check_task_cnt = 0;
2735 	}
2736 
2737 	if (p->rev == T3_REV_B2)
2738 		check_t3b2_mac(adapter);
2739 
2740 	/*
2741 	 * Scan the XGMAC's to check for various conditions which we want to
2742 	 * monitor in a periodic polling manner rather than via an interrupt
2743 	 * condition.  This is used for conditions which would otherwise flood
2744 	 * the system with interrupts and we only really need to know that the
2745 	 * conditions are "happening" ...  For each condition we count the
2746 	 * detection of the condition and reset it for the next polling loop.
2747 	 */
2748 	for_each_port(adapter, port) {
2749 		struct cmac *mac =  &adap2pinfo(adapter, port)->mac;
2750 		u32 cause;
2751 
2752 		cause = t3_read_reg(adapter, A_XGM_INT_CAUSE + mac->offset);
2753 		reset = 0;
2754 		if (cause & F_RXFIFO_OVERFLOW) {
2755 			mac->stats.rx_fifo_ovfl++;
2756 			reset |= F_RXFIFO_OVERFLOW;
2757 		}
2758 
2759 		t3_write_reg(adapter, A_XGM_INT_CAUSE + mac->offset, reset);
2760 	}
2761 
2762 	/*
2763 	 * We do the same as above for FL_EMPTY interrupts.
2764 	 */
2765 	status = t3_read_reg(adapter, A_SG_INT_CAUSE);
2766 	reset = 0;
2767 
2768 	if (status & F_FLEMPTY) {
2769 		struct sge_qset *qs = &adapter->sge.qs[0];
2770 		int i = 0;
2771 
2772 		reset |= F_FLEMPTY;
2773 
2774 		v = (t3_read_reg(adapter, A_SG_RSPQ_FL_STATUS) >> S_FL0EMPTY) &
2775 		    0xffff;
2776 
2777 		while (v) {
2778 			qs->fl[i].empty += (v & 1);
2779 			if (i)
2780 				qs++;
2781 			i ^= 1;
2782 			v >>= 1;
2783 		}
2784 	}
2785 
2786 	t3_write_reg(adapter, A_SG_INT_CAUSE, reset);
2787 
2788 	/* Schedule the next check update if any port is active. */
2789 	spin_lock_irq(&adapter->work_lock);
2790 	if (adapter->open_device_map & PORT_MASK)
2791 		schedule_chk_task(adapter);
2792 	spin_unlock_irq(&adapter->work_lock);
2793 }
2794 
2795 static void db_full_task(struct work_struct *work)
2796 {
2797 	struct adapter *adapter = container_of(work, struct adapter,
2798 					       db_full_task);
2799 
2800 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_FULL, 0);
2801 }
2802 
2803 static void db_empty_task(struct work_struct *work)
2804 {
2805 	struct adapter *adapter = container_of(work, struct adapter,
2806 					       db_empty_task);
2807 
2808 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_EMPTY, 0);
2809 }
2810 
2811 static void db_drop_task(struct work_struct *work)
2812 {
2813 	struct adapter *adapter = container_of(work, struct adapter,
2814 					       db_drop_task);
2815 	unsigned long delay = 1000;
2816 	unsigned short r;
2817 
2818 	cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_DROP, 0);
2819 
2820 	/*
2821 	 * Sleep a while before ringing the driver qset dbs.
2822 	 * The delay is between 1000-2023 usecs.
2823 	 */
2824 	get_random_bytes(&r, 2);
2825 	delay += r & 1023;
2826 	set_current_state(TASK_UNINTERRUPTIBLE);
2827 	schedule_timeout(usecs_to_jiffies(delay));
2828 	ring_dbs(adapter);
2829 }
2830 
2831 /*
2832  * Processes external (PHY) interrupts in process context.
2833  */
2834 static void ext_intr_task(struct work_struct *work)
2835 {
2836 	struct adapter *adapter = container_of(work, struct adapter,
2837 					       ext_intr_handler_task);
2838 	int i;
2839 
2840 	/* Disable link fault interrupts */
2841 	for_each_port(adapter, i) {
2842 		struct net_device *dev = adapter->port[i];
2843 		struct port_info *p = netdev_priv(dev);
2844 
2845 		t3_xgm_intr_disable(adapter, i);
2846 		t3_read_reg(adapter, A_XGM_INT_STATUS + p->mac.offset);
2847 	}
2848 
2849 	/* Re-enable link fault interrupts */
2850 	t3_phy_intr_handler(adapter);
2851 
2852 	for_each_port(adapter, i)
2853 		t3_xgm_intr_enable(adapter, i);
2854 
2855 	/* Now reenable external interrupts */
2856 	spin_lock_irq(&adapter->work_lock);
2857 	if (adapter->slow_intr_mask) {
2858 		adapter->slow_intr_mask |= F_T3DBG;
2859 		t3_write_reg(adapter, A_PL_INT_CAUSE0, F_T3DBG);
2860 		t3_write_reg(adapter, A_PL_INT_ENABLE0,
2861 			     adapter->slow_intr_mask);
2862 	}
2863 	spin_unlock_irq(&adapter->work_lock);
2864 }
2865 
2866 /*
2867  * Interrupt-context handler for external (PHY) interrupts.
2868  */
2869 void t3_os_ext_intr_handler(struct adapter *adapter)
2870 {
2871 	/*
2872 	 * Schedule a task to handle external interrupts as they may be slow
2873 	 * and we use a mutex to protect MDIO registers.  We disable PHY
2874 	 * interrupts in the meantime and let the task reenable them when
2875 	 * it's done.
2876 	 */
2877 	spin_lock(&adapter->work_lock);
2878 	if (adapter->slow_intr_mask) {
2879 		adapter->slow_intr_mask &= ~F_T3DBG;
2880 		t3_write_reg(adapter, A_PL_INT_ENABLE0,
2881 			     adapter->slow_intr_mask);
2882 		queue_work(cxgb3_wq, &adapter->ext_intr_handler_task);
2883 	}
2884 	spin_unlock(&adapter->work_lock);
2885 }
2886 
2887 void t3_os_link_fault_handler(struct adapter *adapter, int port_id)
2888 {
2889 	struct net_device *netdev = adapter->port[port_id];
2890 	struct port_info *pi = netdev_priv(netdev);
2891 
2892 	spin_lock(&adapter->work_lock);
2893 	pi->link_fault = 1;
2894 	spin_unlock(&adapter->work_lock);
2895 }
2896 
2897 static int t3_adapter_error(struct adapter *adapter, int reset, int on_wq)
2898 {
2899 	int i, ret = 0;
2900 
2901 	if (is_offload(adapter) &&
2902 	    test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2903 		cxgb3_event_notify(&adapter->tdev, OFFLOAD_STATUS_DOWN, 0);
2904 		offload_close(&adapter->tdev);
2905 	}
2906 
2907 	/* Stop all ports */
2908 	for_each_port(adapter, i) {
2909 		struct net_device *netdev = adapter->port[i];
2910 
2911 		if (netif_running(netdev))
2912 			__cxgb_close(netdev, on_wq);
2913 	}
2914 
2915 	/* Stop SGE timers */
2916 	t3_stop_sge_timers(adapter);
2917 
2918 	adapter->flags &= ~FULL_INIT_DONE;
2919 
2920 	if (reset)
2921 		ret = t3_reset_adapter(adapter);
2922 
2923 	pci_disable_device(adapter->pdev);
2924 
2925 	return ret;
2926 }
2927 
2928 static int t3_reenable_adapter(struct adapter *adapter)
2929 {
2930 	if (pci_enable_device(adapter->pdev)) {
2931 		dev_err(&adapter->pdev->dev,
2932 			"Cannot re-enable PCI device after reset.\n");
2933 		goto err;
2934 	}
2935 	pci_set_master(adapter->pdev);
2936 	pci_restore_state(adapter->pdev);
2937 	pci_save_state(adapter->pdev);
2938 
2939 	/* Free sge resources */
2940 	t3_free_sge_resources(adapter);
2941 
2942 	if (t3_replay_prep_adapter(adapter))
2943 		goto err;
2944 
2945 	return 0;
2946 err:
2947 	return -1;
2948 }
2949 
2950 static void t3_resume_ports(struct adapter *adapter)
2951 {
2952 	int i;
2953 
2954 	/* Restart the ports */
2955 	for_each_port(adapter, i) {
2956 		struct net_device *netdev = adapter->port[i];
2957 
2958 		if (netif_running(netdev)) {
2959 			if (cxgb_open(netdev)) {
2960 				dev_err(&adapter->pdev->dev,
2961 					"can't bring device back up"
2962 					" after reset\n");
2963 				continue;
2964 			}
2965 		}
2966 	}
2967 
2968 	if (is_offload(adapter) && !ofld_disable)
2969 		cxgb3_event_notify(&adapter->tdev, OFFLOAD_STATUS_UP, 0);
2970 }
2971 
2972 /*
2973  * processes a fatal error.
2974  * Bring the ports down, reset the chip, bring the ports back up.
2975  */
2976 static void fatal_error_task(struct work_struct *work)
2977 {
2978 	struct adapter *adapter = container_of(work, struct adapter,
2979 					       fatal_error_handler_task);
2980 	int err = 0;
2981 
2982 	rtnl_lock();
2983 	err = t3_adapter_error(adapter, 1, 1);
2984 	if (!err)
2985 		err = t3_reenable_adapter(adapter);
2986 	if (!err)
2987 		t3_resume_ports(adapter);
2988 
2989 	CH_ALERT(adapter, "adapter reset %s\n", err ? "failed" : "succeeded");
2990 	rtnl_unlock();
2991 }
2992 
2993 void t3_fatal_err(struct adapter *adapter)
2994 {
2995 	unsigned int fw_status[4];
2996 
2997 	if (adapter->flags & FULL_INIT_DONE) {
2998 		t3_sge_stop_dma(adapter);
2999 		t3_write_reg(adapter, A_XGM_TX_CTRL, 0);
3000 		t3_write_reg(adapter, A_XGM_RX_CTRL, 0);
3001 		t3_write_reg(adapter, XGM_REG(A_XGM_TX_CTRL, 1), 0);
3002 		t3_write_reg(adapter, XGM_REG(A_XGM_RX_CTRL, 1), 0);
3003 
3004 		spin_lock(&adapter->work_lock);
3005 		t3_intr_disable(adapter);
3006 		queue_work(cxgb3_wq, &adapter->fatal_error_handler_task);
3007 		spin_unlock(&adapter->work_lock);
3008 	}
3009 	CH_ALERT(adapter, "encountered fatal error, operation suspended\n");
3010 	if (!t3_cim_ctl_blk_read(adapter, 0xa0, 4, fw_status))
3011 		CH_ALERT(adapter, "FW status: 0x%x, 0x%x, 0x%x, 0x%x\n",
3012 			 fw_status[0], fw_status[1],
3013 			 fw_status[2], fw_status[3]);
3014 }
3015 
3016 /**
3017  * t3_io_error_detected - called when PCI error is detected
3018  * @pdev: Pointer to PCI device
3019  * @state: The current pci connection state
3020  *
3021  * This function is called after a PCI bus error affecting
3022  * this device has been detected.
3023  */
3024 static pci_ers_result_t t3_io_error_detected(struct pci_dev *pdev,
3025 					     pci_channel_state_t state)
3026 {
3027 	struct adapter *adapter = pci_get_drvdata(pdev);
3028 
3029 	if (state == pci_channel_io_perm_failure)
3030 		return PCI_ERS_RESULT_DISCONNECT;
3031 
3032 	t3_adapter_error(adapter, 0, 0);
3033 
3034 	/* Request a slot reset. */
3035 	return PCI_ERS_RESULT_NEED_RESET;
3036 }
3037 
3038 /**
3039  * t3_io_slot_reset - called after the pci bus has been reset.
3040  * @pdev: Pointer to PCI device
3041  *
3042  * Restart the card from scratch, as if from a cold-boot.
3043  */
3044 static pci_ers_result_t t3_io_slot_reset(struct pci_dev *pdev)
3045 {
3046 	struct adapter *adapter = pci_get_drvdata(pdev);
3047 
3048 	if (!t3_reenable_adapter(adapter))
3049 		return PCI_ERS_RESULT_RECOVERED;
3050 
3051 	return PCI_ERS_RESULT_DISCONNECT;
3052 }
3053 
3054 /**
3055  * t3_io_resume - called when traffic can start flowing again.
3056  * @pdev: Pointer to PCI device
3057  *
3058  * This callback is called when the error recovery driver tells us that
3059  * its OK to resume normal operation.
3060  */
3061 static void t3_io_resume(struct pci_dev *pdev)
3062 {
3063 	struct adapter *adapter = pci_get_drvdata(pdev);
3064 
3065 	CH_ALERT(adapter, "adapter recovering, PEX ERR 0x%x\n",
3066 		 t3_read_reg(adapter, A_PCIE_PEX_ERR));
3067 
3068 	rtnl_lock();
3069 	t3_resume_ports(adapter);
3070 	rtnl_unlock();
3071 }
3072 
3073 static const struct pci_error_handlers t3_err_handler = {
3074 	.error_detected = t3_io_error_detected,
3075 	.slot_reset = t3_io_slot_reset,
3076 	.resume = t3_io_resume,
3077 };
3078 
3079 /*
3080  * Set the number of qsets based on the number of CPUs and the number of ports,
3081  * not to exceed the number of available qsets, assuming there are enough qsets
3082  * per port in HW.
3083  */
3084 static void set_nqsets(struct adapter *adap)
3085 {
3086 	int i, j = 0;
3087 	int num_cpus = netif_get_num_default_rss_queues();
3088 	int hwports = adap->params.nports;
3089 	int nqsets = adap->msix_nvectors - 1;
3090 
3091 	if (adap->params.rev > 0 && adap->flags & USING_MSIX) {
3092 		if (hwports == 2 &&
3093 		    (hwports * nqsets > SGE_QSETS ||
3094 		     num_cpus >= nqsets / hwports))
3095 			nqsets /= hwports;
3096 		if (nqsets > num_cpus)
3097 			nqsets = num_cpus;
3098 		if (nqsets < 1 || hwports == 4)
3099 			nqsets = 1;
3100 	} else {
3101 		nqsets = 1;
3102 	}
3103 
3104 	for_each_port(adap, i) {
3105 		struct port_info *pi = adap2pinfo(adap, i);
3106 
3107 		pi->first_qset = j;
3108 		pi->nqsets = nqsets;
3109 		j = pi->first_qset + nqsets;
3110 
3111 		dev_info(&adap->pdev->dev,
3112 			 "Port %d using %d queue sets.\n", i, nqsets);
3113 	}
3114 }
3115 
3116 static int cxgb_enable_msix(struct adapter *adap)
3117 {
3118 	struct msix_entry entries[SGE_QSETS + 1];
3119 	int vectors;
3120 	int i;
3121 
3122 	vectors = ARRAY_SIZE(entries);
3123 	for (i = 0; i < vectors; ++i)
3124 		entries[i].entry = i;
3125 
3126 	vectors = pci_enable_msix_range(adap->pdev, entries,
3127 					adap->params.nports + 1, vectors);
3128 	if (vectors < 0)
3129 		return vectors;
3130 
3131 	for (i = 0; i < vectors; ++i)
3132 		adap->msix_info[i].vec = entries[i].vector;
3133 	adap->msix_nvectors = vectors;
3134 
3135 	return 0;
3136 }
3137 
3138 static void print_port_info(struct adapter *adap, const struct adapter_info *ai)
3139 {
3140 	static const char *pci_variant[] = {
3141 		"PCI", "PCI-X", "PCI-X ECC", "PCI-X 266", "PCI Express"
3142 	};
3143 
3144 	int i;
3145 	char buf[80];
3146 
3147 	if (is_pcie(adap))
3148 		snprintf(buf, sizeof(buf), "%s x%d",
3149 			 pci_variant[adap->params.pci.variant],
3150 			 adap->params.pci.width);
3151 	else
3152 		snprintf(buf, sizeof(buf), "%s %dMHz/%d-bit",
3153 			 pci_variant[adap->params.pci.variant],
3154 			 adap->params.pci.speed, adap->params.pci.width);
3155 
3156 	for_each_port(adap, i) {
3157 		struct net_device *dev = adap->port[i];
3158 		const struct port_info *pi = netdev_priv(dev);
3159 
3160 		if (!test_bit(i, &adap->registered_device_map))
3161 			continue;
3162 		netdev_info(dev, "%s %s %sNIC (rev %d) %s%s\n",
3163 			    ai->desc, pi->phy.desc,
3164 			    is_offload(adap) ? "R" : "", adap->params.rev, buf,
3165 			    (adap->flags & USING_MSIX) ? " MSI-X" :
3166 			    (adap->flags & USING_MSI) ? " MSI" : "");
3167 		if (adap->name == dev->name && adap->params.vpd.mclk)
3168 			pr_info("%s: %uMB CM, %uMB PMTX, %uMB PMRX, S/N: %s\n",
3169 			       adap->name, t3_mc7_size(&adap->cm) >> 20,
3170 			       t3_mc7_size(&adap->pmtx) >> 20,
3171 			       t3_mc7_size(&adap->pmrx) >> 20,
3172 			       adap->params.vpd.sn);
3173 	}
3174 }
3175 
3176 static const struct net_device_ops cxgb_netdev_ops = {
3177 	.ndo_open		= cxgb_open,
3178 	.ndo_stop		= cxgb_close,
3179 	.ndo_start_xmit		= t3_eth_xmit,
3180 	.ndo_get_stats		= cxgb_get_stats,
3181 	.ndo_validate_addr	= eth_validate_addr,
3182 	.ndo_set_rx_mode	= cxgb_set_rxmode,
3183 	.ndo_eth_ioctl		= cxgb_ioctl,
3184 	.ndo_siocdevprivate	= cxgb_siocdevprivate,
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;
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 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3235 	if (err) {
3236 		dev_err(&pdev->dev, "no usable DMA configuration\n");
3237 		goto out_release_regions;
3238 	}
3239 
3240 	pci_set_master(pdev);
3241 	pci_save_state(pdev);
3242 
3243 	mmio_start = pci_resource_start(pdev, 0);
3244 	mmio_len = pci_resource_len(pdev, 0);
3245 	ai = t3_get_adapter_info(ent->driver_data);
3246 
3247 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
3248 	if (!adapter) {
3249 		err = -ENOMEM;
3250 		goto out_release_regions;
3251 	}
3252 
3253 	adapter->nofail_skb =
3254 		alloc_skb(sizeof(struct cpl_set_tcb_field), GFP_KERNEL);
3255 	if (!adapter->nofail_skb) {
3256 		dev_err(&pdev->dev, "cannot allocate nofail buffer\n");
3257 		err = -ENOMEM;
3258 		goto out_free_adapter;
3259 	}
3260 
3261 	adapter->regs = ioremap(mmio_start, mmio_len);
3262 	if (!adapter->regs) {
3263 		dev_err(&pdev->dev, "cannot map device registers\n");
3264 		err = -ENOMEM;
3265 		goto out_free_adapter_nofail;
3266 	}
3267 
3268 	adapter->pdev = pdev;
3269 	adapter->name = pci_name(pdev);
3270 	adapter->msg_enable = dflt_msg_enable;
3271 	adapter->mmio_len = mmio_len;
3272 
3273 	mutex_init(&adapter->mdio_lock);
3274 	spin_lock_init(&adapter->work_lock);
3275 	spin_lock_init(&adapter->stats_lock);
3276 
3277 	INIT_LIST_HEAD(&adapter->adapter_list);
3278 	INIT_WORK(&adapter->ext_intr_handler_task, ext_intr_task);
3279 	INIT_WORK(&adapter->fatal_error_handler_task, fatal_error_task);
3280 
3281 	INIT_WORK(&adapter->db_full_task, db_full_task);
3282 	INIT_WORK(&adapter->db_empty_task, db_empty_task);
3283 	INIT_WORK(&adapter->db_drop_task, db_drop_task);
3284 
3285 	INIT_DELAYED_WORK(&adapter->adap_check_task, t3_adap_check_task);
3286 
3287 	for (i = 0; i < ai->nports0 + ai->nports1; ++i) {
3288 		struct net_device *netdev;
3289 
3290 		netdev = alloc_etherdev_mq(sizeof(struct port_info), SGE_QSETS);
3291 		if (!netdev) {
3292 			err = -ENOMEM;
3293 			goto out_free_dev;
3294 		}
3295 
3296 		SET_NETDEV_DEV(netdev, &pdev->dev);
3297 
3298 		adapter->port[i] = netdev;
3299 		pi = netdev_priv(netdev);
3300 		pi->adapter = adapter;
3301 		pi->port_id = i;
3302 		netif_carrier_off(netdev);
3303 		netdev->irq = pdev->irq;
3304 		netdev->mem_start = mmio_start;
3305 		netdev->mem_end = mmio_start + mmio_len - 1;
3306 		netdev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM |
3307 			NETIF_F_TSO | NETIF_F_RXCSUM | NETIF_F_HW_VLAN_CTAG_RX;
3308 		netdev->features |= netdev->hw_features |
3309 				    NETIF_F_HW_VLAN_CTAG_TX;
3310 		netdev->vlan_features |= netdev->features & VLAN_FEAT;
3311 
3312 		netdev->features |= NETIF_F_HIGHDMA;
3313 
3314 		netdev->netdev_ops = &cxgb_netdev_ops;
3315 		netdev->ethtool_ops = &cxgb_ethtool_ops;
3316 		netdev->min_mtu = 81;
3317 		netdev->max_mtu = ETH_MAX_MTU;
3318 		netdev->dev_port = pi->port_id;
3319 	}
3320 
3321 	pci_set_drvdata(pdev, adapter);
3322 	if (t3_prep_adapter(adapter, ai, 1) < 0) {
3323 		err = -ENODEV;
3324 		goto out_free_dev;
3325 	}
3326 
3327 	/*
3328 	 * The card is now ready to go.  If any errors occur during device
3329 	 * registration we do not fail the whole card but rather proceed only
3330 	 * with the ports we manage to register successfully.  However we must
3331 	 * register at least one net device.
3332 	 */
3333 	for_each_port(adapter, i) {
3334 		err = register_netdev(adapter->port[i]);
3335 		if (err)
3336 			dev_warn(&pdev->dev,
3337 				 "cannot register net device %s, skipping\n",
3338 				 adapter->port[i]->name);
3339 		else {
3340 			/*
3341 			 * Change the name we use for messages to the name of
3342 			 * the first successfully registered interface.
3343 			 */
3344 			if (!adapter->registered_device_map)
3345 				adapter->name = adapter->port[i]->name;
3346 
3347 			__set_bit(i, &adapter->registered_device_map);
3348 		}
3349 	}
3350 	if (!adapter->registered_device_map) {
3351 		dev_err(&pdev->dev, "could not register any net devices\n");
3352 		goto out_free_dev;
3353 	}
3354 
3355 	for_each_port(adapter, i)
3356 		cxgb3_init_iscsi_mac(adapter->port[i]);
3357 
3358 	/* Driver's ready. Reflect it on LEDs */
3359 	t3_led_ready(adapter);
3360 
3361 	if (is_offload(adapter)) {
3362 		__set_bit(OFFLOAD_DEVMAP_BIT, &adapter->registered_device_map);
3363 		cxgb3_adapter_ofld(adapter);
3364 	}
3365 
3366 	/* See what interrupts we'll be using */
3367 	if (msi > 1 && cxgb_enable_msix(adapter) == 0)
3368 		adapter->flags |= USING_MSIX;
3369 	else if (msi > 0 && pci_enable_msi(pdev) == 0)
3370 		adapter->flags |= USING_MSI;
3371 
3372 	set_nqsets(adapter);
3373 
3374 	err = sysfs_create_group(&adapter->port[0]->dev.kobj,
3375 				 &cxgb3_attr_group);
3376 	if (err) {
3377 		dev_err(&pdev->dev, "cannot create sysfs group\n");
3378 		goto out_close_led;
3379 	}
3380 
3381 	print_port_info(adapter, ai);
3382 	return 0;
3383 
3384 out_close_led:
3385 	t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL, 0);
3386 
3387 out_free_dev:
3388 	iounmap(adapter->regs);
3389 	for (i = ai->nports0 + ai->nports1 - 1; i >= 0; --i)
3390 		if (adapter->port[i])
3391 			free_netdev(adapter->port[i]);
3392 
3393 out_free_adapter_nofail:
3394 	kfree_skb(adapter->nofail_skb);
3395 
3396 out_free_adapter:
3397 	kfree(adapter);
3398 
3399 out_release_regions:
3400 	pci_release_regions(pdev);
3401 out_disable_device:
3402 	pci_disable_device(pdev);
3403 out:
3404 	return err;
3405 }
3406 
3407 static void remove_one(struct pci_dev *pdev)
3408 {
3409 	struct adapter *adapter = pci_get_drvdata(pdev);
3410 
3411 	if (adapter) {
3412 		int i;
3413 
3414 		t3_sge_stop(adapter);
3415 		sysfs_remove_group(&adapter->port[0]->dev.kobj,
3416 				   &cxgb3_attr_group);
3417 
3418 		if (is_offload(adapter)) {
3419 			cxgb3_adapter_unofld(adapter);
3420 			if (test_bit(OFFLOAD_DEVMAP_BIT,
3421 				     &adapter->open_device_map))
3422 				offload_close(&adapter->tdev);
3423 		}
3424 
3425 		for_each_port(adapter, i)
3426 		    if (test_bit(i, &adapter->registered_device_map))
3427 			unregister_netdev(adapter->port[i]);
3428 
3429 		t3_stop_sge_timers(adapter);
3430 		t3_free_sge_resources(adapter);
3431 		cxgb_disable_msi(adapter);
3432 
3433 		for_each_port(adapter, i)
3434 			if (adapter->port[i])
3435 				free_netdev(adapter->port[i]);
3436 
3437 		iounmap(adapter->regs);
3438 		kfree_skb(adapter->nofail_skb);
3439 		kfree(adapter);
3440 		pci_release_regions(pdev);
3441 		pci_disable_device(pdev);
3442 	}
3443 }
3444 
3445 static struct pci_driver driver = {
3446 	.name = DRV_NAME,
3447 	.id_table = cxgb3_pci_tbl,
3448 	.probe = init_one,
3449 	.remove = remove_one,
3450 	.err_handler = &t3_err_handler,
3451 };
3452 
3453 static int __init cxgb3_init_module(void)
3454 {
3455 	int ret;
3456 
3457 	cxgb3_offload_init();
3458 
3459 	ret = pci_register_driver(&driver);
3460 	return ret;
3461 }
3462 
3463 static void __exit cxgb3_cleanup_module(void)
3464 {
3465 	pci_unregister_driver(&driver);
3466 	if (cxgb3_wq)
3467 		destroy_workqueue(cxgb3_wq);
3468 }
3469 
3470 module_init(cxgb3_init_module);
3471 module_exit(cxgb3_cleanup_module);
3472